Core Concepts in Economics: Fundamentals

This post serves as a primer for future posts related to topics in economics. While not necessarily core to a critical thinking curriculum, economic concepts are vital for understanding a modern policy environment. I cannot begin to describe the vast disconnect between basic economic literacy and the folk models held by large swaths of the general public. As an economist myself, perhaps I am biased in thinking many of our problems in society are in some way, caused by this illiteracy. However, I think there is a strong case to be made that obsolete, ideologically laden, underdeveloped, and empirically unsound economic mental models held by various groups, are contributing significantly to the deterioration of the nations welfare. Therefore, I think it's important to lay out how professional or academic economists come to their conclusions. The public broadly has no idea what economists do; how they reason, their models of the economy, what a model even is, their sources of information etc. This post introduces some of the most basic concepts implicit in most economic theory and regularly used by applied economists. Lets start with some basics. These might seem abstract initially, so I'll try to concretize them:
 

Table of Contents

Economy

Is defined as:
"a system of interactions among agents (households, firms, and governments) in which scarce resources are allocated through markets and institutions to produce, distribute, and consume goods and services."
This is actually a lot to unpack; what is a system? what is a market? what is an agent? what are goods? Each of these terms within this definition are just as abstract as the definition itself. Its also important to point out that different schools of economics will often define an economy differently, emphasizing certain aspects they find to be of crucial importance. For our purposes here, just note that this definition will likely be broadly accepted by most economists and is consistent with definitions within textbooks. 

After years if study, I've come to appreciate the conceptualization discussed by George E. Mobus in "Systems Science: Theory, Analysis, Modeling, and Design". It is defined here as an instance of the broader class of systems called Complex Adaptive Evolutionary Systems (CAES). The author conceives of a "generic economic system", that is usually embedded within a larger system (like an ecosystem), characterized by certain elements such as the way in which the system obtains resources in energy/materials, how they do internal work to grow/replicate/maintain themselves, and how they export waste back into their environment. In other words, an economy is a type pattern within a broader set of CAES. Specifically, he states:
"What all of these systems do internally is to use the energy and material imports in work processes that build and maintain essential internal structures. Waste products are inevitable in all such products in nature. The organization of the work processes and their ongoing management (which collectively we can call the governance of the system) is the pattern of which we spoke. Whether we are talking about the internal organization and dynamics of a living cell, of the multicellular organism of a Neolithic tribe, a single household, a complex organization, or a modern nation state, the patterns of organization and dynamics follow them. These patterns cluster under the title 'economy'."
This is an even more generic definition that invokes concepts from system science. It's analogous to a biological subsystem that performs that performs vital functions that ensure the survival of the supersystem in which it is embedded. Doyne Farmer, a complexity economist, has drawn an analogy between an economic system and a metabolic system. The economic system functions like a metabolic system, in that it consumes resources, processes energy & information, and adapts over time. It is always changing, growing, or decaying. It's primary function is the flow of energy and matter. Using this analogy, we can conceive of the economic metabolism; how economies transform inputs into outputs, similarly to how living organisms metabolize food into energy and biomass. 

I'll admit that this is definitely a heterodox approach but you can see the parallels with how mainstream economists define an economy. You might see in standard textbooks the etymology of the term "economy". It originally derives from "oikonomia", a Greek term meaning "management of a household". In ancient Greece, the "masters" of the household had to practice "economic activities", like raising sufficient food to support the household. Mobus states that "this meant managing processes so that the household recognizes an accumulation of wealth that would ultimately provide for the support of subsequent generations." In modern times, these households are embedded within a broader complex system and interact with institutions such as "markets" to achieve these ends. Mobus offers another abstraction:
"At an abstract level, an economy is a fabric of transactions, a network of sub-processes of resource acquisition , transformation (value added), production, and consumption, in which biological beings are sustained.... In the systems view, an economy is a way of managing the flows of high potential energy and of the transformation of high entropy materials into low entropy assets via work processes that use that high potential energy to do useful work. The low entropy assets support the existence and stability of the system."
The key thing to note with these definitions, is that economists are concerned with the flow and utilization of scarce resources, typically within some national boundary, but increasingly at a global scale. 

Models (Modeling)

So, what is a model? At its core, a model is a simplified representation of reality. It’s a tool we use to describe, explain, or predict how something works. Consider a map; it's not the territory, rather its a simplified version that leaves out irrelevant details, allowing you to navigate the terrain efficiently. A model is a map of a system of interest. Many different models can describe the same system. A single system like the economy, or climate, or a human body can be described in many different ways, depending on what you want to understand or predict, the data you have access to, and generally speaking the question you are asking. In economics, there are a plethora of models that describe the same economy. The choice of model depends on the question, context, and needed level of accuracy. The level of fidelity or granularity of the model also depends on these considerations. Low fidelity models are fast, easy to understand, captures big-picture behavior. They are good for teaching, quick decisions, or back-of-the-envelope estimates. Medium fidelity models capture more details; attempting to balance tractability and accuracy. High fidelity models are very detailed, often computational, can model complex interactions; they are used when precision really matters. Many economic models are "coarse grained models"; coarse-graining means reducing a system’s complexity by grouping together small-scale components and modeling only the aggregate behavior. Think of it like pixels on a screen; fine grained would be something like ultra HD while coarse grained would be something lower resolution, but you can still make out the general objects in the screen. 

Everyone has a model of reality in their mind. Our brains construct maps of reality, which we use to navigate the terrain. The deliberate act of modeling, allows us to interrogate, investigate, and revise these models. This is particularly useful for explicating our assumptions. Models are highly sensitive to their underlying assumptions. If we are not aware of these assumptions, our model will be inaccurate and useless. Every mathematical model starts with assumptions about what variables matter, how the variables relate, what is constant vs what can change, the environment in which the system is embedded, and assumptions about the fundamental unit of analysis (in economics, our "agents"). Even if our model is not mathematical, it still depends on these sorts of assumptions. The use of mathematics in the physical sciences and economics is for clarity. Language is incredibly ambiguous and vague; mathematization allows us to be incredibly precise with our definitions; leaving little room for ambiguity when we must interpret the model. Model assumptions define the world. Models don’t discover truth,  they operate within a truth you’ve defined. So if your assumptions are flawed, your model will produce misleading results, even if the math is perfect. Some models are relatively unaffected by slight changes in the underlying assumptions, other models are highly sensitive. In finance, models like the Black-Scholes equation assume no transaction costs and continuous trading. In the real world, these assumptions break down and so does the model’s accuracy. Our assumptions also guide the interpretation of the model results. Therefore, the economist (ideally) should be constantly asking “What’s this model assuming? And is that reasonable for the question I care about?”

I would argue that many economics students don't fully appreciate the rationale behind mathematical modeling in economics or more broadly across disciplines. The very essence of economics is to treat the economy as a system governed by universal laws. We invoke modeling notions such as equilibrium to describe how markets stabilize over time in response to external perturbations. We construct analogies like "market forces" which help explain movements in some quantity of interest, typically price. We use differential equations to model the dynamics of quantities like inflation or interest over time. Much of these techniques are inspired by physical sciences. In fact, many economists are simply mathematicians who apply this math to questions about economics. 

The problem I've recognized within economics departments, is that students pay less attention to the process of modeling, and focus more on the results of a particular economic model. It is easier to memorize the implications of a model, rather than to interrogate the structure and assumptions governing a model. I've simultaneously recognized an overemphasis on formalisms and equation solving within economics departments. What I mean is, students are evaluated by their abilities to solve a particular set of equations, rather than by the model thinking required to solve economic problems. Both of these miss the point of what were doing. The first set of students is primarily interested in the answers while the latter set of students is treating economics as if its mathematics. 

Educators really need to stress why we even model to begin with. I've explained briefly what a model is, while only alluding to the many reasons why we might want to model. The "why" we model is core to scientific inquiry. Much of the "Why" I'll be describing comes from Why Model by Joshua Epstein and Different Modeling Purposes by Bruce Edmonds. 

Epstein argues that modeling is an inherent part of human cognition. Whenever individuals make projections or imagine scenarios, such as the spread of an epidemic or the outcome of a war, they are effectively running mental models. These are implicit models with hidden assumptions, untested internal consistency, and unknown logical consequences. The distinction lies in making these models explicit, allowing for scrutiny, replication, and validation.​ Building explicit models involves clearly stating assumptions, which facilitates understanding their implications. Explicit models can be shared, tested against data, and refined. They enable sensitivity analysis, allowing researchers to explore how changes in parameters affect outcomes. This process is crucial for identifying uncertainties, robust regions, and critical thresholds, particularly in policy-making contexts.​ Epstein also describes the value of modeling beyond out of sample prediction. He lists 16 different uses:
  1. Explain (very distinct from predict)
  2. Guide data collection
  3. Illuminate core dynamics
  4. Suggest dynamical analogies
  5. Discover new questions
  6. Promote a scientific habit of mind
  7. Bound (bracket) outcomes to plausible ranges
  8. Illuminate core uncertainties.
  9. Offer crisis options in near-real time
  10. Demonstrate tradeoffs / suggest efficiencies
  11. Challenge the robustness of prevailing theory through perturbations
  12. Expose prevailing wisdom as incompatible with available data
  13. Train practitioners
  14. Discipline the policy dialogue
  15. Educate the general public
  16. Reveal the apparently simple (complex) to be complex (simple)
Explanation is distinct from prediction. When scientists use the word "predict", they do not necessarily mean "forecast". A quote attributed to Neils Bohr says "Prediction is hard, especially about the future!" This highlights the distinction between prediction "within sample" vs prediction "out of sample", the latter being something more akin to a forecast; an extrapolation based on historical trends. To predict in a scientific sense, means given a model of a system, some input (X) is expected to produce output (Y) within some margin of error. So for example, if I know the mass of a solid and its acceleration, then I can predict what the force will be; it is determined by the formalization of the system. I can then conduct an experiment to verify whether my description corresponds to observations. In economics, if i know the quantity demanded and the quantity supplied of a product, I can determine the market price. Sometimes prediction is very hard given the complexity of a system. This is fine, because the model can still be explanatory. Epstein gives the example of plate tectonics being explanatory of earthquakes, despite not predicting the time and location of any specific earthquake. 

Epstein critiques the naïve inductivist view of science, which assumes that researchers first gather data and then build models to explain it. This view is common among both non-modelers and some modelers, especially in the social sciences, where it's often believed that one should "collect lots of data and run regressions." While data-driven research can be valuable, Epstein argues that this is not how science typically works. In many significant scientific breakthroughs, theory came before data and actually guided what data should be sought. Models are not just tools for explaining existing data, they are invaluable for shaping and guiding data collection. Without models, researchers might not know what data is most relevant or worth collecting in the first place. 

Models don’t have to be precisely accurate to be incredibly useful. In fact, all the best models are technically "wrong", they are simplifications or idealizations of reality. But this wrongness doesn’t diminish their value. On the contrary, their simplicity and abstraction make them powerful tools for understanding the core dynamics of complex systems. Though models are idealizations and approximations, they nevertheless enable us to describe broad qualitative behavior of a system such as feedback loops, threshold effects, tipping points etc. The real question isn't whether a model is idealized (all models are), but whether the model is a fertile idealization, does it generate insight, understanding, and foundational intuition; hence George Box famously stating that  “All models are wrong, but some are useful.” When first introduced to modeling, it can seem absurd to deliberately simplify, ignoring potentially relevant details. The truth of the matter is that we simply have to ignore things, we do all the time, without realizing it. We simply cannot function in the world without ignoring most facts about the world. We have limited cognitive resources; we could not possibly consider the mountains of information bombarding our senses at any given moment. Formal modeling forces us to be very explicit with what we are considering; meaning that it's clearly communicated what's not being considered, thus providing deeper intellectual engagement with the question at hand. 

Paul Smaldino gives an thoughtful anecdote illustrating the precision of a formalism provided by modeling. Remember from earlier, verbal models suffer from ambiguity. Formalizing theories in terms of mathematical models helps us be precise with our concepts. Smaldino recounts an experience from his undergraduate days when he and a friend, while waiting in a theater basement, constructed a whimsical LEGO figure they dubbed a "Cubist chicken." Both agreed on its identity until a third friend questioned its features. Upon attempting to explain, they realized each had a different interpretation of which parts represented the chicken's head, body, and tail. This divergence highlighted that their shared understanding was more assumed than actual.​ The parable serves as a metaphor for the pitfalls of relying solely on verbal models in science. While verbal descriptions can seem comprehensive, they often harbor ambiguities that lead to misinterpretations. In contrast, formal models, despite being simplifications, require explicit definitions and assumptions, fostering clearer communication and understanding among researchers. Smaldino emphasizes that the "stupidity" or simplicity of models is a strength, as it forces clarity and facilitates the testing of specific hypotheses.​ Models force us to make sure we are all talking about the same thing. A corollary to precision is tractability. Formal models are logical engines that transform assumptions into conclusion. Stating assumptions precisely allows us to know what necessarily follows from those assumptions, which helps us find potential gaps in our explanations. 

Mainstream economics frequently gets criticized by its approach to modeling. This brings us to the distinction between as-is and as-if modeling. As-if models assume that agents behave "as if" they are behaving a certain way (optimizing) even if they don't do so in reality. For example, the classic rational actor model in economics assumes individuals maximize utility as if they were calculating costs and benefits precisely, even though real humans may not. Friedman famously defended this in "The Methodology of Positive Economics" (1953), arguing that the realism of assumptions is less important than the accuracy of predictions. Mainstream economics often uses as-if models for their analytical tractability and predictive usefulness. On the other hand, "as-is" models try to represent how agents actually behave based on empirical observation, often including heuristics, bounded rationality, learning, or psychological factors. These models are often less mathematically neat but aim to be more descriptive of real-world behavior. An example might be an agent based model that computationally encodes more complex decision rules an agent might adaptively utilize, within an evolving system. Economists like J. Doyne Farmer and others in the complexity economics tradition (at the Santa Fe Institute, inspired by Herbert Simon) are strong proponents of as-is modeling, especially using agent-based models and data-driven simulations. They reject the representative agent:, emphasizing heterogeneity and micro-level interaction. They emphasize emergence of macro-level phenomena from simple decentralized rules. They place a heavy emphasis on empirical validation; models must fit and explain data, not simply derive from axioms. From the mainstream economist perspective, the goal is to predict behavior, so highly unrealistic idealizations are fine if they aid in prediction. They deemphasize descriptive realism. 

Systems

When describing "the economy", we are describing an economic system. This was somewhat alluded to in the definition of economics. We are interested in studying aspects about a particular system of interest. Sometimes we are interested in studying the "global economic system". Other times, we are interested in studying a proper subsystem of the global economic system, like the financial system. But what exactly is a "system"? How do economists think about systems? I'll first start with a broad definition of "system" from systems science, before modifying it to encapsulate how economists broadly think of "systems". Economists are usually not this explicit when conceiving "systems"; in fact, in all my years of study I rarely find economists formally defining "system". Mainstream economists certainly do not formalize the notion. Heterodox approaches like complexity economists tend to be more descriptive. The following formalization is consistent with heterodox approaches taken by someone like Brian Arthur. But we will see broadly, economists are concerned with these elements.

Mobus and Kalton describe a system as a cohesive, organized whole composed of interrelated components that work together to produce behaviors or functions not reducible to those of the individual parts. A system exists within a boundary, interacts with its environment, and processes inputs and outputs in structured ways. It exhibits emergence, meaning the behavior or purpose of the whole cannot be understood merely by analyzing its parts in isolation. Systems can be open or closed, adaptive or rigid, and operate across multiple scales and timeframes. They maintain internal organization through feedback, control, and governance mechanisms, and are often hierarchically embedded in larger systems.

Formally, a system Si,lS_{i,l} at identity ii and hierarchical level ll is defined as the 9-tuple:

Si,l=(Ci,l,Ni,l,Srci,l,Snki,l,Gi,l,Bi,l,Ki,l,Hi,l,Δti,l)S_{i,l} = (C_{i,l}, N_{i,l}, Src_{i,l}, Snk_{i,l}, G_{i,l}, B_{i,l}, K_{i,l}, H_{i,l}, \Delta t_{i,l})

Each component of this tuple describes a key aspect of what constitutes a system in precise, structural terms.

1. Ci,lC_{i,l}: Components

This is the set of parts or elements that make up the system. Each component may itself be a system (a subsystem), reflecting the hierarchical nature of systems in general. Components are defined by their roles and characteristics within the system, and their inclusion in the system may not always be binary—it is sometimes useful to model their membership as “fuzzy,” meaning that components may participate in the system to varying degrees or under specific conditions. For instance, in an ecosystem, a migratory species might only be part of the system seasonally, making its membership conditional.

The components of an economy include a vast array of agents and institutions—households, firms, banks, government agencies, and markets. Each of these components has a defined role: households provide labor and consume goods, firms produce goods and services, banks manage financial intermediation, and governments set fiscal and monetary policy. These entities interact within and across sectors, forming the basic building blocks of the system. Some components, like multinational corporations or informal economies, may participate across boundaries, making them partially included or fuzzy in membership.

2. Ni,lN_{i,l}: Network of Relations

This element represents the structural or functional connections between components. It defines how parts of the system influence, support, or depend on each other. These relationships can be bidirectional or unidirectional, physical (like wires connecting electrical components) or abstract (like authority or influence in a social organization). The network structure determines the topology of the system and underlies many dynamic behaviors, such as cascades, feedback loops, and resilience to disruption.

In an economy, components are linked through a dense network of financial, legal, and social relationships. Firms are connected to consumers through market exchanges, to suppliers via supply chains, and to banks through credit and investment. Governments interact with all other components through taxation, regulation, and public spending. These relationships define flows of goods, services, money, labor, and influence. The structure of these connections—centralized or distributed, robust or fragile—has a strong impact on economic performance and resilience to shocks.

3. Srci,lSrc_{i,l}: Sources

Sources refer to the inputs that enter the system from its environment. These might include material resources, energy, or information—anything required for the system to function and persist over time. In a manufacturing system, for example, raw materials are sources. In a cognitive system, sensory stimuli are inputs. The nature and availability of these sources can have a profound impact on how the system behaves or whether it can sustain itself.

The sources of an economy are the external inputs that support its functioning. These include natural resources (like oil, minerals, or water), imported goods and services, foreign investments, immigrant labor, and technological innovation originating abroad. These sources flow into the system and are transformed, consumed, or circulated internally. For a closed economy, these sources would be limited, but in reality, most modern economies are highly open and reliant on continuous input from the global system.

4. Snki,lSnk_{i,l}: Sinks

Complementary to sources, sinks are the outputs of the system—where its products, wastes, or by-products go. Sinks represent how the system interacts with and impacts its external environment. In an ecological system, this could be the dispersal of nutrients or the release of waste products. In an economic system, sinks might be the markets that receive goods or the environment that absorbs pollution. A system’s outputs can affect not only its own stability but also the systems it interfaces with.

Sinks are where the economy’s outputs go, including exports, emissions, and waste. For example, manufactured goods might be sold to international markets, information products consumed globally, or pollutants expelled into the environment. These outputs affect external systems—ecological, social, and economic. Negative sinks, like environmental degradation, can feed back as costs to the system, making them crucial to sustainability and long-term modeling.

5. Gi,lG_{i,l}: Flow Graph

The flow graph represents the directed movement of resources—whether energy, matter, or information—between components within the system. It formalizes the system's internal dynamics and enables modeling of how internal processes operate, such as transport, transformation, or communication. Flow graphs are typically weighted and directional, capturing both the pathways and intensities of flows, and are critical for analyzing systemic phenomena like bottlenecks, delays, and accumulation.

The flow graph of an economy maps how resources circulate between components—money flows from consumers to producers, taxes from businesses to government, subsidies from government to agriculture, and so on. It includes supply chains, labor markets, investment flows, and trade routes. This graph allows us to trace bottlenecks, feedback loops (like inflationary spirals), and cyclical behaviors (like recessions). Monetary policy and interest rates, for instance, are interventions into specific flow patterns meant to influence broader systemic outcomes.

6. Bi,lB_{i,l}: Boundary

The boundary distinguishes what is part of the system from what is not. It may be physical (like the hull of a ship), functional (such as firewall rules in a network), or even conceptual (the defined scope of a scientific model). Boundaries determine the scope of analysis and define where inputs enter and outputs exit. They are essential for understanding how a system maintains its integrity, interfaces with its environment, and evolves over time. In some cases, boundaries may shift, blur, or be contested, especially in social or conceptual systems.

An economic system’s boundary defines what is considered part of the national or regional economy and what lies outside it. For example, the boundary of the U.S. economy would include all production and consumption activities within its jurisdiction, but also interactions with foreign economies through trade and capital flows. The boundary is often fuzzy—offshore accounts, black markets, or informal economies may blur the lines of inclusion. How we define the boundary impacts the scope of data collection (GDP, for instance) and policy-making.

7. Ki,lK_{i,l}: Knowledge

Knowledge is the information stored, encoded, or maintained within the system. This might be in the form of data, memory, genetic information, operating rules, or even learned behaviors. Knowledge enables the system to regulate itself, adapt, and evolve in response to internal or external conditions. In biological systems, this could be DNA; in human organizations, it could be culture or institutional memory. Knowledge may also define how the system models itself or anticipates its environment, enabling more sophisticated forms of control.

Knowledge in an economy includes institutional memory, laws, technologies, business practices, education, and even cultural norms. It resides in human capital, embedded in institutions, and codified in technologies and systems of production. This internal knowledge base enables innovation, guides decision-making, and supports coordination across vast distances and organizational layers. A highly developed economy typically has dense knowledge structures that promote adaptability and efficiency.

8. Hi,lH_{i,l}: Governance

Governance encompasses the mechanisms, rules, and feedback processes that guide the system’s behavior and maintain its stability. This may include control systems, management structures, or algorithms. It can be centralized or distributed, rigid or adaptive. Governance ensures that the system responds to changes, corrects errors, and aligns its operations with desired outcomes. In ecosystems, governance might take the form of natural feedback loops; in engineered systems, it might be a set of protocols or a software routine managing operations.

Economic governance is exercised through central banks, treasuries, regulatory bodies, legal frameworks, and international institutions like the IMF or WTO. It includes fiscal policy (spending and taxation), monetary policy (interest rates, money supply), and regulatory actions (banking laws, labor protections). These governance mechanisms manage inflation, employment, growth, and inequality. Effective governance keeps the system stable, resilient to shocks, and aligned with societal goals. Poor governance can lead to systemic crises.

9. Δti,l\Delta t_{i,l}: Time Interval

Finally, the time interval refers to the period over which the system is observed, modeled, or understood. It provides temporal context, distinguishing between fast, transient processes and long-term, evolutionary changes. Some systems operate over milliseconds (e.g., electronic circuits), others over centuries (e.g., climate systems). Time determines not only how the system behaves but also how we interpret causality, feedback, and system lifecycle stages such as growth, decay, and renewal.

The time interval for studying an economy could vary dramatically depending on the question—short-term intervals might focus on quarterly business cycles, while long-term intervals could examine industrial development, technological evolution, or demographic shifts over decades. Time plays a critical role in understanding lag effects, feedback delays, and compounding dynamics like debt accumulation or climate-related economic changes. Economies are dynamic systems that evolve, adapt, and sometimes collapse across different time horizons.

Given all of this, it's clear that economists are concerned with "economic systems". They are certainly concerned with governance, knowledge structures, network relations etc. However, they do not use formalisms like what's used in systems science. When referring to "systems", they will use designations like "capitalist system" or "mixed system"; highlighting different aspects of these systems that are present or absent. In mainstream economics, the concept of a “system” is only partially formalized, it’s often used metaphorically or descriptively (often ideologically laden), but elements of it are embedded in formal models without always being explicitly labeled as “systems.” For example, dynamic stochastic general equilibrium (DSGE) models, used in macroeconomics, formalize how an economy responds over time to shocks. They're system-like in structure, with state variablesfeedback, and evolution over time. In systems, equilibrium is a state of balance or homeostasis. In economics, equilibrium is central; markets “clear” when supply equals demand; economists typically assume there to be some homeostasis absent external perturbations. This is a central analogy used in economics to physical systems. Systems are defined by what’s inside (endogenous) and what’s outside (exogenous inputs); more on this later because these are absolutely critical concepts you must understand to understand the practice of economics. In economics, models clearly separate endogenous variables (explained within the model) and exogenous ones (shocks, policy, technology); this mirrors the "system boundary" concept. In many applications, like game theory or IO analysis, “system” means a set of interrelated equations that represent how different variables (like prices, consumption, wages) influence each other. It’s a structural or computational sense of a system, closer to engineering or control theory. Economic agents are modeled like controllers optimizing an objective (utility, profit) subject to constraints (budget, production function). Very similar to optimal control or operations research. Central banks have access to the systems control variables, which are the levers used to influence the systems behavior. Monetary policy is modeled with feedback rules (like the taylor rule), which is essentially a PID controller. DSGE models are analyzed for stability around a steady state, and then impulse response functions are used to study how these systems react to shocks (impulses). Every agent has an objective function they're seeking to optimize and follow simple rules for optimizing these function over time. When I was in graduate school, I learned quite a lot about state space models. The difference between a systems science approach, is that economists typically do not consider emergence, adaptation, agent heterogeneity, network interactions, path dependence, tipping points, learning, or evolution. It primary uses a rational agent framework (more on this later) which assumes away the possibility of these phenomenon. 

When economists are speaking more loosely about "systems", they are using the term in the broad metaphorical sense rather than the rigorous mathematical sense. They are essentially shorthand for institutional arrangement, decision making structures, and norms of organization. They refer to who owns resource, how decisions are made, and the role of the government. I personally think we can juxtapose the systems science formalism onto these terms. For example, "Capitalist system" would by an instance of a broader class of related systems that can be described using the 9-tuple above. I really wish the discipline went this way because there is so much confusion around these ideologically loaded terms; can't imagine how much of humanity is wasted speaking past each other because they have no conceptual foundations. "Systems" in this loose sense, refers to broad institutional configurations, typically involving the following:
  • Ownership of the means of production
  • Coordination mechanisms (think prices)
  • Incentive Structures (think the profit motive)
  • Role of the state (think of "laissez faire")
  • Allocation of Resources (budgeting, planning, markets)
  • Legal and Institutional Framework (property rights, taxation etc)
Samuelson uses "capitalist" and "market economy" pretty much interchangeably; emphasizing property rights, private enterprise, and markets. Other economists emphasize "price signals" as what determines the allocation of resources, based on decentralized choice of individual agents. Honestly, given these loose definitions, anything can be capitalist. Not going down that rabbit hole, but I think it depends on which aspects of the system we are primarily focused on. 

One last concept I think is important; system decomposition. I'll quote Smaldino from Modeling Social Behavior: 
"What are the parts of the system we are interested in? What are their properties? What are the relationships between the parts and their properties? How do those properties and relationships change? Decomposition consists of usable answers to these questions"
This is relevant to the modeling section above, because when you want to hypothesize something about your system, you first must articulate the parts of that system. Economists are primarily concerned with causal hypotheses, meaning the level of description must be sufficient to capture the parts of the system relevant to the question. There is no right level of decomposition; it fundamentally depends on the question, and the value of the model depends on how well it's decomposition answers the question. Oh and remember, assumptions are behind all of this. 

Optimization

Optimization is not only a cornerstone method in economics but also a conceptual lens to view systems and agents interacting within those systems. It is the engine of economic reasoning and is at the heart of every economic analysis. Whether it's a consumer choosing between products, a firm deciding how much to produce, or a government allocating a budget, economics assumes that agents make the best possible choices given their objectives and constraints. Consumers are modeled as maximizing utility given their income and the prices of goods. Firms aim to maximize profits by choosing how much labor and capital to employ. Policymakers often try to optimize welfare, balancing trade-offs like equity and efficiency. This framework allows economists to derive predictions and policy implications from first principles. It's not that everyone consciously solves equations, but the optimization lens provides a consistent method to analyze decisions in a world of scarce resources. Optimization connects economics with mathematics, especially calculus and linear programming, turning abstract choices into solvable problems from the economists perspective, to make predictions about real world scenarios. From marginal analysis to game theory, the logic of "best choice under constraints" drives much of economic reasoning. It frames the way economists see the world: as a series of trade-offs, decisions, and outcomes shaped by rational pursuit of goals within limits.

Most modern economic research begins with a simple, core idea: individuals and institutions make purposeful decisions to achieve their objectives as best they can, given constraints. This assumption allows economists to use optimization to build formal models of behavior (utility maximization, profit maximization, more on these later), derive testable implications from these models and simulate outcomes under different assumptions or scenarios. For example, a paper studying household consumption might assume that people maximize lifetime utility subject to a budget constraint. This provides a structured, solvable problem whose solution can be compared with real-world data. Optimization gives the model its internal logic and empirical tractability. By using optimization, these theories don't just describe behavior they provide a mechanism to explain why certain outcomes occur, and how changes in conditions (prices, income, technology) lead to changes in behavior. Because economic models are grounded in optimization, the policy advice that emerges from them is, by design, about improving outcomes—maximizing welfare, minimizing inefficiencies, or designing better incentives. For instance tax policy is analyzed through models where households optimize labor supply decisions, allowing economists to estimate optimal tax rates, monetary policy uses models where central banks optimize over inflation and output trade-offs and environmental regulation applies optimization to balance economic costs with ecological benefits. If the models are robust, the policy recommendations derived from them aim to move the system closer to an optimal state, given real-world constraints and trade-offs.

You can think of optimization as a way of thinking; a framework for formulating, analyzing, and solving decision problems across economics, operations research, and decision sciences. At its core, optimization thinking involves Decision variables (what can be chosen), Objective functions (what is to be maximized or minimized), and Constraints (rules that must be satisfied):
  1. Identify objectives
    • What is the agent (a person, a firm, a policymaker) trying to achieve?
    • Make the goals explicit
  2. Clarify constraints
    • What limits the available choices—budget, time, regulations, technology?
  3. Map feasible alternatives
    • What are the real options available?
  4. Evaluate trade-offs
    • How do changes in one choice affect the outcome or cost of another?
  5. Choose the best course of action
    • What’s the most effective or efficient choice given the above?
    • Evaluate the costs and benefits of alternatives.

Economists are not just thinking about these concepts devoid of practical application. Often they are hired by employers to engage in prescriptive modeling on behalf of the firm. They migh construct a model of business operations within the firm to ask what should be done to acheive an optimal outcome (usually maximizing profits). Within academic economics, we use optimization primarily for explanatory and predictive purposes to describe features of a whatever unit of analysis we are concerned with. But outside of academics, economists often work in a prescriptive decision making capacity under conditions of uncertainty, risk, or resource limits where they must identify optimal course of action in situations involving strategic interaction.

Most economists should have this generic formulation internalized:

Let \( x \in \mathbb{R}^n \) be the vector of decision variables. A standard constrained optimization problem is formulated as:

\[ \begin{aligned} \text{Minimize (or Maximize)} \quad & f(x) \\\\ \text{subject to} \quad & g_i(x) \leq b_i, \quad i = 1, \dots, m \\\\ & h_j(x) = c_j, \quad j = 1, \dots, p \\\\ & x \in X \end{aligned} \]

Where:

  • \( f(x) \) is the objective function to be minimized or maximized (e.g., cost, utility, profit).
  • \( x \) is the vector of decision variables (e.g., quantities to produce, allocate, consume).
  • \( g_i(x) \leq b_i \) are inequality constraints, representing resource limits or bounds.
  • \( h_j(x) = c_j \) are equality constraints, such as balance conditions or conservation laws.
  • \( X \) is the feasible set—the domain of allowable solutions (e.g., \( x \geq 0 \), or \( x \) must be integer-valued).

Whether you're modeling a consumer’s utility maximization, a firm’s cost minimization, or a government's resource allocation, almost all applied optimization problems in economics and decision science can be framed in this form.

Dynamic optimization is also very pervasive within economics. Dynamic optimization models capture decisions that unfold over time, where today’s choice affects tomorrow’s possibilities. These can be broken down into discrete time models or continuous time models; where the major difference is whether the time variable can be discretized. Economists use dynamic optimization to model:

  • How consumers smooth consumption across life (life-cycle hypothesis)
  • How firms invest in capital over time (investment theory)
  • How governments plan fiscal or monetary policy (dynamic programming)
  • How agents form expectations and adjust behavior (rational expectations models)
  • How does a shock today (e.g., a technology shock or policy change) propagate over time?
  • What’s the optimal investment plan when capital is costly to adjust?

The key ingredients are:

  1. A state variable that summarizes the relevant "current condition" (e.g., wealth, capital stock)
  2. A control variable (or decision variable) that the agent chooses (e.g., how much to consume or invest)
  3. A transition equation that describes how today’s decision shapes tomorrow’s state
  4. An objective function that evaluates the total value of decisions over time

Dynamic optimization is MUCH harder to explain succinctly and I struggled with it (and still do) when in graduate school. I am not going to provide the generic mathematical setup but i'll provide a very common problem macroeconomists should be familiar with. The Cake Eating Problem, a problem of intertemporal choice, asks: how much of X should I enjoy today and how much of X should I leave for the future (where X is cake)? This sounds really trivial at first until you realize it's an extremely general problem; a trade-off between current and future utility. Imagine you have a fixed-sized cake, say, one whole cake. You can eat some of it now, and save the rest for later. But once you eat a piece, it’s gone. The challenge is: How should you allocate consumption of the cake over time to maximize your total satisfaction (utility)? Is it optimal to eat it all now? Should you eat a small piece a day? If you eat more today, you'll have less tomorrow, and therefore lower future utility. The goal is to find an optimal policy (the best sequence of decisions), a specific plan for how much cake to eat at each period that balances current and future satisfaction. Economists use this idea to study decisions where you have a limited resource and you have to decide how to use it over time; like money, food, energy, or natural resources. The goal is to make choices today that don't ruin your happiness tomorrow. So here is the formal model:

  • Let \( W_t \) be the amount of cake (or wealth) at time \( t \).
  • You choose how much cake to eat: \( c_t \), with \( 0 \leq c_t \leq W_t \).

The remaining cake becomes:

\[ W_{t+1} = W_t - c_t \]

  • You derive utility from eating: \( u(c_t) \), where \( u(\cdot) \) is a utility function (e.g., \( u(c) = \ln(c) \)).
  • You discount future utility with a factor \( \beta \in (0, 1) \), meaning you value today’s consumption more than tomorrow’s.

Your goal is to choose the sequence \( \{c_t\}_{t=0}^{\infty} \) to maximize lifetime utility:

\[ \max_{\{c_t\}} \sum_{t=0}^{\infty} \beta^t u(c_t) \]

subject to:

\[ \begin{aligned} W_{t+1} &= W_t - c_t \\\\ W_0 &> 0, \quad c_t \in [0, W_t] \end{aligned} \]

This is typically solved by setting up the bellman equation and solving it with dynamic programming; using methods such as value iteration. You can see this resource for more methods.

A brief note on the discount factor; this is essentially a scaling factor that tells you how much you value future consumption compared to current consumption. This is such a core concept in economics and finance that I cannot possibly do it justice in this section, so I'll save it for later. But for now just see it as a scaling factor that reduces the weight of future utility compared to current utility. If it is less than zero, you eat more now. If it is closer to one, you save more for the future. It models the idea that people usually prefer good things sooner and bad things later, something very intuitive. This pops up literally everywhere in economics and is conceptually (almost identical) to the idea of an interest rate.

Rationality

Economists describe "agents" as "rational". I've made use of the term "agent" in earlier sections but haven't explicitly defined how economists use it. The "agent" is the fundamental unit of analysis shared by most economists. This is the basic building block or elementary unit in many economic models; the lowest level at which behavior is assumed or described. In disciplines like physics, the fundamental unit might be a particle. In Sociology, it might be a group or institution. It's important to note that, some people self described as economists, take the institution to be fundamental. The fundamental unit doesn't have to be the "smallest" per se, it's typically just the starting point for a specific line of inquiry. In economics, it is an "agent"; an individual decision making entity. We build models by specifying how these agents behave.

"Agents" in economics have ordered and complete preferences, respond to incentives, make choices given constraints, and are modeled to optimized an objective like utility (more on that in a bit) or profit. A "consumer" is an agent that maximizes utility from consumption of finished products, goods, services etc. A firm is an "agent" that maximizes profit or minimizes costs. A government is an agent that maximizes "welfare" (more on that later). Macroeconomists make use of the "representative agent", encapsulating the average behavior of a collection of homogenous agents. We often assume that agents are rational (the point of this section), forward looking, self interested, autonomous and atomistic. You may think you know what these terms refer to, but economists tend to use them in very specific ways. For example, self interest need not mean "selfishness". Atomistic is pretty much an assumption about how behaviors are influenced by peer groups. 

In behavioral economics, agents are conceptualized differently. For example, in the classical conceptualization, economists assume transitive preferences, meaning if I prefer A to B and B to C, then I prefer A to C. There is a well orderedness that follows classical laws of logic. Behavioral economists on the other hand, relax this assumption, frequently showing empirically that this assumption doesn't hold in reality. They also use a model of bounded rationality, in which cognitive resources are fundamentally constrained, implying we engage in suboptimal behavior utilizing simple decision rules like heuristics. These heuristics are influenced by framing, emotions, social norms. They can be inconsistent intertemporally and context dependent. They are subject to cognitive biases that may have been evolutionary inherited. Notice how this becomes extremely more complex to model. In agent based modeling, the "atomistic" and "homogenous" assumptions are relaxed. Agents are conceived as heterogenous and adaptive. They don't "optimize" anything, rather they follow simple rules of thumb and use learning algorithms (implicitly). Macroeconomic phenomenon are the emergent patterns arising from this underlying heterogeneity. In other words, there is no "representative agent". In game theory, agents are fundamentally strategic. They form Nth order beliefs about other peoples beliefs, updating these beliefs as games (situations of strategic integration) evolve. They project signals to competitors to influence their competitors beliefs. They can be both cooperative and competitive dependent on perceived payoffs. In social theory, agents are inherently inseparable from their social structures. Decisions aren't made based on constrained optimization and are shaped by culture, norms, and institutions. Preferences aren't a "given"; they are not taken for granted. 

Notice how assumptions about the fundamental unit of analysis literally determines the conclusions we draw about "the economy". The choice of how to model an agent, determines what behavior is explain, what can be predicted, and what policies are recommended. Mainstream economists use the classical conceptualization of an agent, reflecting methodological individualism. There is nothing inherently wrong with this approach, insofar as it is a useful model of agency. It is becoming increasable augmented and challenged by more pluralistic conceptions of agency.

The foundational view of rationality in economics is the idea that agents make decisions that are consistent with their preferences and goals, given the information they have access to and their constraints. This is a form of "instrumental rationality", a practical rationality that describes how people choose efficient means to achieve their goals. Economists assume preferences are consistent. This means they are complete, transitive, independent and monotone. Completeness refers to the idea that, given a complete set of choices, it is possible for the agent to rank all the options, according to some preference mapping (i.e. utility). Transitivity was mentioned earlier. Independence refers to the idea that preferences are formed in isolation and that a persons choice between two options are not influenced by the presence or absence or a third option. In economics, we call this "Independence of Irrelevant Alternatives", or IIA. This is a crucial assumption ensuring preferences are well defined and consistent; meaning that if it fails then we can't model agents as optimizers. The IIA assumption implies that choices can be predicted based solely on the properties of the options being considered, not on irrelevant circumstances or other options that might be available. Monotone preferences, or "Non-Satiation", means that "more is better". An agent will always choose a bundle of goods that contains more of every good, subject to their constraints. I cannot stress how foundational this assumption is. It assumes that consumers always prefer more of a good or service to less; and that there are no limits to their potential satisfaction of consuming more. Satisfaction might increase at a decreasing rate, this is called diminishing marginal utility, something I'll write about later. Monotone functions are functions that either always increase or decrease within their domain. This is what we assume about consumer utility with respect to consuming a product. An important implication of this is that preferences are convex . Convexity refers to a feature of a mathematical function. Why is this important in economics? Simply put, preferences must be convex, because if not then its pretty frickin hard to optimize.

It's so easy to take this for granted when studying economics. It's also why many people struggle with studying economics; they're unfamiliar with the foundational assumptions of the fundamental unit of analysis. Every economic theory, recommendation, policy proposal, analysis etc in some form or another, builds off this. Many economists literally take it for granted; they do not recognize these as simplifying assumptions for the model, but see them as eternal truths about human behavior. One of my main contentions with economics, is the economists who are incapable of critically analyzing these central assumptions. When I was in highschool, I remember taking an economics class where on one day, the teaching was so dogmatic that in hindsight, the best description of the situation was child abuse. For the record, the instructor was not an economist. He probably had no familiarization with mathematical modeling. But he was trying to indoctrinate us with the monotonicity assumption by insisting that it's an eternal truth of "homo economicus" that we have infinite wants, and that we would act on these infinite wants if it werent for scarcity (constraints). This was not a discussion, this was not interactive, it was taught as a central dogma; literally like a Sunday School session. It's always been interesting to me how a modeling formalization has literally become ideology. 

I have no problem with this rationality assumption in principle. It makes modeling choice mathematically tractable. It can offer some predictive power. It provides a normative benchmark, allowing us to make comparisons between different policies because it serves as a baseline for efficiency. However, I am a pluralist when it comes to economic methodology. I think this framework is a special instance of a broader set of possible descriptions of human behavior. It's unable to explain many real world phenomena like bubbles, crises, and persistent unemployment. It is inconsistnt with behavioral predictions like loss aversion, framing effects, and time inconsistency. It's not capable of predicting how agents will respond to policy and price changes universally. It's really contrived; I prefer as-is modeling to as-if modeling. Assuming perfect information and infinite computation is kind of absurd and obviously not empirically valid. Rationality is useful in that it provides coherence and rigor to models, but it abstracts away human psychology, institutional aspects, and computational limits (more on this later).

Nevertheless, like I mentioned, this is foundational in economics. It informs how the models are constructed, and the subsequent policy recommendations. For example, the supply and demand model imply rent control is economically inefficient. Since the framework is deductive in nature, the implications of the model would recommend removing this policy. If A then B, A therefore B. If rent control, then inefficiency, rent control therefore inefficiency. These deductions literally fall out of the assumptions about our fundamental unit of analysis and how agents aggregate. In macroeconomics, a representative agent stands in as the single agent who can represent the economy. Heterogeneity is very difficult to model because aggregation is impossible (well, technically its possible, but its not possible to have a stable single equilibrium). This agent is assumed to solve intertemporal optimization problems. They also have rational expectations about the future, this is the forward looking assumption. Specifically, it means that individuals gather all the available information, including past trends and economic data. They form expectations based on this historical data. These are statistical expectations about key macroeconomic measures such as inflation and interest rates. Therefore, central banks set interest rates based on forward looking inflation expectations falling directly out of the DSGE model assuming rational expectations. 

There are a variety of different conceptualizations of rationality, as I've touch on briefly above. Some of these highlight the importance of cognitive resources. I think this is absolutely crucial for understanding rationality. Dan Sperber and Deirdre Wilson present a concept of "Relevance" in their book "Relevance: Communication and Cognition" which I think implies a model of rationality that is not only realistic, but consistent with definition of rationality in computational fields, and therefore cross disciplinary agreed upon. To me, this is an important sign of a concepts usefulness; if other disciplines converge on something similar. The authors introduce the cognitive principle of relevance. Human cognition is geared towards the maximization of relevance. In other words, we tend to pay attention to information that yields the most cognitive effect (like new insights or changes in belief) for the least processing effort. This leads to the communicative principle of relevance; Every act of communication carries with it the presumption of its own relevance. When someone says something, the listener expects that it is worth the mental effort to understand — i.e., it will be relevant enough to justify the attention. Communication is not just about encoding and decoding messages (as traditional code models suggest), but about inferring intentions. Listeners use context and assumptions to figure out what the speaker meant, not just what they said. They define communication as a two part process: Ostension, in which a speaker signals they want to communicate, and inference, the lister inteprets the signal, guided by the assumption it will be relevant. Communication and thought are guided by the search for relevance, and this means acheiving the most meaningful impact with the least amount of cognitive effort. 

How is a theory of relevance, well, relevant to rationality? Well, their theory gives us a description about how cognitive faculties function. Rationality and decision making are obviously interconnected with that. Their theory implies something about rationality because it gives us a description of cognitive information processing, something fundamental to the concept of rationality. Rememeber, economists simply assume infinite processing power. How absurd is that? If you are in anyway familiar with computational complexity, supercomputing, or scientific computing, you'll be very aware of the fact that some of the fastest computers in the world still can only provide approximations to even relatively simply computations. And yet, economists think its safe to assume that, the human brain is capable of handling some of the most computationally complex problems. The theory of relevance highlights the fact that communication is an inferential process, which depends on cognitive resources. We cant possibly attend to all information, or even determine which information is relevant to a decision problem, without simplfying assumptions. This implies a model of bounded rationality. Humans are not perfectly logical agents. Instead of maximizing truth or utility in a strict sense, we use heuristics (mental shortcuts) to make satisficing decisions — ones that are good enough given our cognitive limitations. According to Sperber and Wilson, rationality is driven by the search for relevance. A person is rational if they pursue thoughts, beliefs, and interpretations that provide high cognitive effects (like useful inferences or knowledge) with minimal effort; this can be completely independent of utility. Their model of rationality is inferential, we interpret others not just by decoding language, but by inferring their intentions in context, guided by relevance. So communication is rational when it makes those inferences easy and rewarding.

Tom Griffiths, a cognitive scientist at Princeton, has a computational model of reality that is consistent with the rationality implied by Sperber and Wilsons communication model. Griffiths argues that to understand human rationality, we should think in terms of optimal use of limited resources; including time, information, and cognitive capacity. His work blends Bayesian inference, machine learning, and resource-bounded computation to model how people make decisions and draw conclusions. Tom Griffiths’ theory of rationality sees humans as computationally rational agents; not perfectly logical, but using clever approximations, heuristics, and probabilistic reasoning to solve problems efficiently under real-world constraints.

His approach is very interesting in my opinion because elements of it still incorporate the "as-if" modeling approach taken by economists while also incorporating insights from artificial intelligence, psychology, and computer science. For example, he assumes that at some level, humans reason as if they're doing bayesian inference. At some general level, they handle uncertainty by weighing evidence and prior beliefs to update their understanding of the world. Our cognitive process approximate bayesian reasoning. Other researchers in neuroscience like Karl Friston take this approach also. He also takes the resourc-rational approach, meaning that human thought and behavior should be understood as the best possible use of limited computational resources. So instead of optimizing, like in economics, people are bound by computational resources such that hueristics and approximations take place of a potentially resource intensive optimization. Think of it this way, economists assume that preference are complete. This means that agents are capable of enumerating all possible choices and rank ordering them according to which choices maximize utility. They do this intertemporally, meaning they are aware of how the choice space will look N years in the future. They also are aware of the secondary effects of choosing action A over action B, meaning they have counterfactual knowledge of the decision space. For any given decision, they can compute the optimal solution. Kind of hilarious when you pose it this way. But a more realistic approach, is to model people as cognitively resource constrained, not just physical resource constrained. Griffiths emphasizes computational-level analysis, following David Marr’s idea that we should ask: What is the goal of the computation? What is the optimal solution, and how close are humans to achieving it under real-world limits? His work connects to the idea that human cognition is adapted to the structure of the environment; we make the best decisions possible based on the patterns we’ve learned from experience (like machine learning models trained on data). This is known as ecological rationality

Obviously, I'm not here to reconcile the differences between Griffiths and Sperber/Wilsons theories. Strictly speaking there isn't anything to reconcile, the latter aren't advancing a model of rationality. But I think their communicative model implies many elements in Griffiths approach to rationality. For sperber/wilson, agents seek to maximize relevance and minimize cognitive load during communication. From a rational agent perspective, agents must engage with other agents to acquire information relevant to their decision problem. They obviously do not exhaustively inquire with other agents over the space of possible information sources and level of depth when engaging with the information. They are bound by cognitive resources and update their posterior beliefs accordingly. The mind is a relevance engine that is evolved to make inferential leaps based on minimal effort; I think this is literally what a heuristic is. I suppose the main difference between the two is that, since communication is inherently contextual and social, interpretation is bound by expectations of relevance, which are socially generated. In other words, rationality itself is something bound to the social. Griffiths approach is still individual based. 

Below are some more non-classical approaches to modeling rationality that have inspired this section:
  1. Herbert Simon – Bounded Rationality: Humans are not fully rational due to cognitive and informational limits. Instead, they are boundedly rational; they make satisficing (satisfy + suffice) decisions rather than optimizing. People choose the first acceptable solution, not necessarily the best one. Memory, attention, and time restrict decision-making. Behavior is shaped by the structure of the environment; people adapt, rather than optimize.
  2. Gigerenzer – Ecological & Heuristic Rationality: In many environments, simple heuristics can be more effective than complex reasoning. Rationality is adaptive, not absolute. Fast and frugal heuristics are quick, efficient mental shortcuts that exploit environmental structure. Ecological rationality means What’s rational depends on the match between the mind’s heuristics and the structure of the environment.
  3. Amos Tversky & Daniel Kahneman – Heuristics and Biases: Humans rely on heuristics, which often lead to systematic biases, deviations from ideal rationality. There are many catalogued biases and heuristics including the Availability heuristic (Judging likelihood by ease of recall) and Representativeness heuristic (Judging similarity over base rates). Prospect theory is also central. This states that people evaluate gains/losses relative to a reference point and are loss-averse. This is where the idea "predictably irrational" comes from. 
Implicit within all of this is the concept of self interest. I didn't designate a section specifically for this concept because its normally covered by many definitions of rationality; although sometimes glossed over. Self interest does not mean selfish. Let's just establish that. It primarily refers to a specific vantage point of an individual within a broader economic system. This really goes back to Adam Smith. In The Theory of Moral Sentiments, Smith argues that our moral sentiments (feelings of empathy, sympathy, approval etc.) are stronger toward people who are closer to us, whether emotionally, socially, or physically. Smith talks about a hierarchy of concern, such that the literal proximity determines how much we care; the closest is ourselves, then family/friends/community, and then countrymen/strangers, finally humanity at large. Smith is providing an explanatory account of emotional distance, not evaluative. He is saying that it takes work to extend our sympathies beyond our immediate circles. Self interest refers to the fact that individuals attend to things that are more immediately salient, more directly important to them, or within a more proximal frame of reference. Smith never referred to self-interest as a dog-eat-dog kind of behavior; on the contrary, he repudiated that.

Self interest and altruism tend to operate simultaneously. Likewise, competition and cooperation tend to operate in conjunction. Many people confuse these concepts, juxtaposing them against one another, thinking they are mutually exclusive. However, economists understand these concepts like the following: for a rational agent pursuing self preservation, very often its necessary to engage in altruistic and cooperative behavior. We engage in this behavior not in begrudgingly, but because its part of a broader self preservation goal that encompasses a wide range of ethical behaviors. But can we be perfectly altruistic, by showing a high level of empathy for those outside your immediate proximity? The answer is probably no. We are oriented toward our own immediacy, and sometimes this comes at the expense of showing empathy towards the out group. Since we cant empathize with all possible vantage points (we cant put ourselves in everyones shoes), we tend to place our cognitive effort towards our immediate groups. This is in essence how economists think of self interest. I should qualify that last statement actually. Many economists who have actually read Adam Smith will think this. Other economists reduce self-interest to mere utility maximization. Smith’s insight that our sympathy weakens with distance shows that he saw self-interest not as a cold, calculating force, but as entangled with the limits of our emotional imagination. The economist’s self-interest is abstract and constant; Smith’s self-interest is human, fallible, and bound up with emotion and moral perception.

Opportunity Cost

This is a fundamental concept used by economists to understand the allocation of some scarce resource. The resource can be time, energy, money, something physical like land, etc. Opportunity cost refers to a decision about the scarce resource and the implicit tradeoff you are making relative to alternative uses. For example, suppose you have some resource called X, that can be used for purposes, ranked in descending order, (a,b,c). By selecting option (a), your opportunity cost is (b); it represents what you have to "give up" in order to devote those resources to option (a). The opportunity cost is the actual value (usually defined in terms of utility, something we will define later) someone must give up. 

Let's concretize this with an example. Suppose you have a portfolio of financial assets, with a mixture of fixed income and equities. You can imagine a scenario where you have 1000$ to spend on different mixtures, or combinations of these assets. Call these mixtures M1, M2 and M3. Lets say M1 gives expected profits of 100$, M2 90$ and M3 85$. Your opportunity cost will be the difference between M1 and M2. 

This seems like a rather useless concept at first, but I cannot stress how essential it is within the economists toolkit. It is fundamentally about how agents identify, rank, and select among alternative uses of some resource; which is at the core of economic behavior. Opportunity cost focuses on the single most valuable thing you had to give up, when making your choice. Here is an example you might see in a textbook. Suppose a company owns a building. It can choose to rent it out for 50k a year or use it for its own operations. If the company decides to use it, it is forgoing 50k dollars of rent they could have earned. So they will not use the building unless they expect they can make more than 50k on the alternative use. 

Consider another example. Imagine a group of citizens under an oppressive regime. They’re frustrated, hungry, overtaxed, but they haven’t revolted yet. Why might this be? Revolting involves opportunity costs. To participate in a revolution, people must give up their current income (even if its low), relative safety, and the time/effort that could go into something else like fleeing. So the opportunity cost of revolting is the stability and limited benefits of not revolting. Now suppose those current benefits diminish, the opportunity cost of not revolting will increase. So from an economists perspective, when the opportunity cost of action becomes lower than the opportunity cost of inaction, a revolution might spark. 

I'm not here to argue about whether revolutions are made based on an economic calculus. All I am attempting to do is show how the concept is applied. Obviously, ideological factors strongly influence the nature of revolutions. Also, even within a business context, someone might not pick the better alternative (from a financial perspective) for reasons such as sentimentality, or emotional connection to one alternative. Opportunity costs are intimately connected with incentives, which we'll learn about next.

Incentives

These are broadly thought to be factors that influence the choices people make by altering the perceived costs and benefits of different actions. Economists use an extremely broad definition; incentives are anything that motivate or influence human behavior by means of altering the cost-benefit structure. When an incentive lowers the opportunity cost of an action, that action becomes more attractive. If it raises the opportunity cost, it's less likely to be chosen. Opportunity cost is what’s given up when making a choice. Incentives are what shift the relative attractiveness of those choices by influencing opportunity costs.

From a systemic perspective, we normally refer to "incentive structures". This refers to the framework of rewards and penalties built into a system that shapes how people behave. It is the underlying setup that determines which actions are encouraged and discouraged, and what outcomes are rewarded or punished. The system itself can be formal, in the form of laws, contracts and policies. It can also be informal, in the form of cultural norms, social pressure, and status. 

For example, imagine a corporate bonus system. Managers might get bonuses for short term profits. This might result in cost cutting behavior like laying off workers or slashing R&D. So the stock prices of that corporation rise in the short term, at the expense of long term innovation and employee maturation. This would be characterized as an incentive structure flaw. Or consider an environmental policy that subsidizes gasoline to the poor. The well intentioned incentive structure might have the unintended consequences of increasing pollution; since people are incentivized to drive more. Also, if there is no long term plan to phase out the program, driving can become so entrenched within the economic system, that transitioning to renewals becomes near impossible in the future because there are no incentives to adjust. 

This might be one of the most fundamental concepts economists use to characterize human behavior. We are constantly asking whether a certain policy distorts economic incentives. There is even a subset of economics called Mechanism Design, which studies how to construct rules or institutions, to reverse engineer incentives to get the outcomes you want. It studies how to create systems or rules (mechanisms) so that individuals, acting in their own self-interest, will still produce a desirable overall outcome. A traditional "economics as an observational science" approach is interested in studying how people behave within a given system, while a "economics as a field of engineering" uses mechanism design to design the system itself. Given the outcome we want, what rules should we write? 

Directly related to mechanism design, and more broadly, incentive structures, is the field of Law & Economics. The field is fundamentally about designing legal rules and institutions that align individual incentives with socially desirable outcomes. For example, in tort law (accidents, negligence, and liability), our goal might be to reduce harm while not stifling productive activity. A mechanism to achieve this, might be to select liability rules that create optimal care incentives for injurers and victims. This would be highly relevant in a situation where employers might be accountable for employee injuries. The main idea is that we want to write rules such that people can act in their own self interest but this does not result in a tragedy of the commons (more on that later). Property rights are also a fundamental concept in economics. I'll write more about that later, but the rules of ownership will dramatically determine the resulting allocation and distribution of resources. 

I hope this begins to show how these basic concepts have fundamentally shaped our modern institutions. Interestingly, I became aware of Law & Economics through its critics. I frequently read a legal theorist named Richard Wright, who specializes in tort law, and is highly critical of what he sees as the degradation of legal theory due to the introduction of economic concepts into the discipline. In particular, he is critical of Richard A. Posner, who might be the most influential figure in Law & Economics. Essentially, Wright and others argue that Posner's view reduces law to cost-benefit analysis, ignoring justice, rights, and duties. Law is not just a tool for efficient outcomes—it’s a normative system rooted in justice, fairness, and moral responsibility.  I am not going to elaborate any further on this, but if interested, I definitely recommend reading about these two people. You'll begin to see how prolific economic thinking has penetrated the most fundamental institutions governing our lives. The nature of the legal system directly impacts incentive structures and opportunity costs, and hence how the economic metabolism of society functions.

Utility

This might be one of the most underappreciated concepts in economics, among non-practitioners. Utility is a core concept in decision theory, economics, and operations research, representing a way to model and quantify preferences, satisfaction, or value that an agent (individual or organization) assigns to outcomes. It's used to guide rational choices under conditions of uncertainty, scarcity, or competing alternatives. Utility is a numerical representation of preferences. Higher utility values represent more preferred outcomes. Its used to model rational choices under uncertainty (e.g., expected utility theory) and to describe consumer behavior, market demand, and welfare economics. A major principle in economics is that people try to maximize utility; agents choose options that yield the highest point on the utility curve. Another important concept is marginal utility; the additional utility derived from consuming one or more unit of a good or service. This is typically understood by analyzing the derivative of the utility function; the function mapping input combinations (typically combinations of goods and services) to utility space. A utility function is typically written as:
u:XRu: X \to \mathbb{R}

where:

  • XRnX \subseteq \mathbb{R}^n is the consumption set or set of possible bundles (e.g., combinations of goods).

  • u(x)Ru(x) \in \mathbb{R} is the real-valued utility assigned to a bundle x=(x1,x2,,xn)Xx = (x_1, x_2, \dots, x_n) \in X.

This function represents the preferences of a consumer over bundles of goods. Here are a few examples of utility functions used in economics:

  1. Cobb-Douglas Utility:

u(x1,x2)=x1αx2β,where α,β>0u(x_1, x_2) = x_1^\alpha x_2^\beta, \quad \text{where } \alpha, \beta > 0
  1. Perfect Substitutes:

u(x1,x2)=ax1+bx2,where a,b>0u(x_1, x_2) = a x_1 + b x_2, \quad \text{where } a, b > 0
  1. Perfect Complements:

u(x1,x2)=min{ax1,bx2},where a,b>0u(x_1, x_2) = \min\{a x_1, b x_2\}, \quad \text{where } a, b > 0
  1. Quasilinear Utility:

u(x1,x2)=ln(x1)+x2u(x_1, x_2) = \ln(x_1) + x_2
  1. CES (Constant Elasticity of Substitution) Utility:

u(x1,x2)=(ax1ρ+bx2ρ)1/ρ,where ρ0u(x_1, x_2) = \left( a x_1^\rho + b x_2^\rho \right)^{1/\rho}, \quad \text{where } \rho \neq 0

The famous "demand curve" in economics is indirectly derived from the utility function. Utility functions represent a consumers preferences over a bundle of goods. Since economics is about scarcity, this utility is constrained by the household budget (we do not have infinite resources to satisfy every desire). The global maximum U(.) might be outside the feasible set. So the consumers goal is to maximize utility subject to the budget constraint (which is a constrained optimization problem). Solving these equations, gives the optimal quantities as functions of price and income, which correspond to the demand curve. Economists typically assume diminishing marginal utility, meaning the first derivative of the utility function is a decreasing function. The utility function increases at a decreasing rate, while the marginal utility function (first derivative) asymptotically approaches some constant. The demand curve falls out of the budget constraint problem (p*x < I). In plain English, the demand curve (partly) is downward sloping because of diminishing marginal utility. 

This might sound extremely abstract and unremoved from practice but its at the very core of how economists model the economy. Consider a real example. The Federal Reserve has a core set of monetary policy tools it can use to influence interest rates, inflation, employment, and economic stability:

1. The Federal Funds Rate (Main Tool): The interest rate banks charge each other for overnight loans of reserves. The Fed doesn't directly set this rate but targets it by adjusting the supply of reserves in the banking system via open market operations (OMO) or interest on reserves. This rate affects all other short-term interest rates — from savings accounts to business loans to mortgage rates. Changing it influences consumption, investment, and inflation.

2. Open Market Operations (OMO): Buying or selling government securities (like Treasury bills) in the open market.

- Buy securities → inject money → lower interest rates (stimulate economy)

- Sell securities → pull money out → raise interest rates (cool economy)

3. Interest on Reserve Balances (IORB): The interest rate the Fed pays banks on their reserves held at the Fed. Banks won’t lend at rates lower than what they can earn risk-free from the Fed. So this effectively sets a floor for the Fed Funds Rate.

4. Discount Rate (Lender of Last Resort): The interest rate the Fed charges commercial banks for short-term loans from the Fed's discount window. Mostly used in emergencies when banks can’t get funding elsewhere. It’s a backup liquidity tool, not a regular lever for setting monetary conditions.

5. Forward Guidance (Communication Tool): Public statements about future policy intentions, e.g. "Rates will remain low for an extended period. Expectations drive behavior. Clear guidance can influence long-term interest rates and financial market conditions.

Based on a model of consumer utility, the fed adjusts these policy tools, influencing consumption choices, by changing the incentives embedded within the utility function. In monetary theory, utility functions shape how people make choices about consumption today vs. in the future. The Euler equation formalizes that tradeoff. 

The Core Euler Equation (in real terms):

u(ct)=βEt[u(ct+1)(1+rt)]u'(c_t) = \beta \, \mathbb{E}_t \left[ u'(c_{t+1}) \cdot (1 + r_t) \right]

Where:

  • u(ct)u'(c_t): marginal utility of consumption today

  • β\beta: discount factor

  • rtr_t: real interest rate (nominal rate minus expected inflation)

It says households will only give up consumption today if they’re compensated by enough expected utility tomorrow , which depends on the real return (how much they can get by saving). The curvature of the utility function directly controls how much consumption will shift if rates change. Let’s map the tools to the equation:

1. Federal Funds Rate (nominal rate iti_t)

Appears in the Euler equation indirectly, via the real interest rate:

rtitEt[πt+1]r_t \approx i_t - \mathbb{E}_t[\pi_{t+1}]

So:

  • If the Fed raises iti_t → higher rtr_t → future consumption is more attractive → households reduce current consumption.

  • That’s how rate hikes cool demand.

2. Open Market Operations (OMO)

OMO are used to hit the Fed’s interest rate target. So they influence iti_t, and thus rtr_t, the same way. They’re the operational tool to implement changes that show up in the Euler equation.

3. Interest on Reserve Balances (IORB)

IORB affects bank incentives to lend, which influence the actual market interest rates (including iti_t). If banks can earn more by parking money at the Fed, they tighten lending → market interest rates go up → higher rtr_t in the Euler equation.

4. Forward Guidance

Even if the Fed doesn’t change iti_t today, if it signals future rate hikes, that changes:

Et[(1+rt+1)] affects today’s consumption ct\mathbb{E}_t[(1 + r_{t+1})] \Rightarrow \text{ affects today's consumption } c_t

So expectations of higher rates reduce current consumption — directly through the expectations operator in the Euler equation.

5. Quantitative Easing (QE)

QE works by reducing long-term interest rates, including rtr_t, especially when short-term rates are at zero.

Lower rtr_t → higher current consumption ctc_t → economic stimulus.

The form of u(c)u(c) determines how strongly households respond to changes in rtr_t. For example:

  • If u(c)=ln(c)u(c) = \ln(c), then:

    u(c)=1cPeople are fairly responsive to changes in interest rates.u'(c) = \frac{1}{c} \Rightarrow \text{People are fairly responsive to changes in interest rates.}
  • If utility is more curved (higher σ\sigma in CRRA), they are less responsive.

So Fed economists calibrate utility functions in DSGE models to match observed behavior, how much households change consumption in response to interest rate changes. Households maximize utility subject to constraints (budget, cash-in-advance, etc.). Solving this yields Euler equations and money demand functions, which help predict how interest rates affect savings/consumption, how inflation expectations influence money holdings (which affects output), and how policy changes influence economic behavior. 

So as you can see, utility builds off of the more basic notions of incentives and opportunity costs. 

Supply/Demand Functions and Ceteris Paribus

I think we have all the basics necessary to derive the supply and demand model. This is cornerstone in economics. Before diving into any derivations, I'd first like to explain two concepts that are critical for understanding how economists analyze the resulting model. Economists want to answer questions such as:
  1. How does a change in one variable affect others?
    • What happens to employment if the minimum wage increases?
  2. What are the effects of policy interventions or external shocks?
    • How does a tax affect consumer prices and producer revenue?
  3. What are the conditions for equilibrium?
    • At what price do supply and demand balance?
  4. How do markets and agents react to changes in incentives or constraints?
    • How do consumers respond to a change in interest rates?

In economic analysis, comparative statics and ceteris paribus are fundamental analytical and conceptual tools used to understand how changes in one variable affect others within an economic model. Comparative statics is the method economists use to compare two different equilibrium states resulting from a change in an exogenous variable (something determined outside the model, like government policy or consumer preferences). It compares before-and-after outcomes and does not focus on the process of adjustment or dynamics over time. For example, if the price of gasoline increases, comparative statics would look at how this affects the quantity of gasoline demanded, holding everything else constant. It compares the original equilibrium (before the price change) with the new equilibrium (after the price change). Ceteris paribus is a Latin phrase meaning “all other things being equal” or “holding everything else constant.” It is a simplifying assumption used to isolate the effect of a single variable. For example, when analyzing how the demand for apples changes with price, economists assume ceteris paribus; that income, prices of other goods, and consumer preferences remain unchanged. Very often, comparative statics often asks counterfactual questions that explore outcomes in alternative states of the world. Comparative statics uses counterfactual reasoning to compare equilibria before and after a change in an exogenous factor, holding everything else constant (ceteris paribus). This is implemented by analyzing the partial derivatives of the resulting model, which I'll show shortly. Ceteris paribus is powerful: it lets you isolate these effects clearly, and see whether your model produces intuitive or surprising behavior. Partial derivatives are the mathematical formalization of a comparative static analysis.

So now that we understand what we are trying to do, lets derive a supply curve, demand curve, and analyze equilibrium by doing comparative statics. Remember, we are assuming economic agents are maximizers. Individual consumers will be seeking to maximize utility and firms will be seeking to maximize profit (minimize cost). Choice variables are at the heart of any optimization problem. They represent the things the decision maker can control or choose to adjust in order to acheive these goals. For a consumer, the choice variables are quantities of a good to buy. For a firm, these are the factors of production (labor or capital) or the output level. For a policy maker, it might be tax rates or subsidy levels. These are "variables" because the decision maker is free to choose the values, under some set of constraints, to optimize the objective.

Remember, an optimization problem has three main components:

  1. Objective function: what you're trying to maximize or minimize (e.g., profit, utility)
  2. Choice variables: the variables you control (e.g., \(x_1, x_2, \dots, x_n\))
  3. Constraints: limits or requirements (e.g., budget, production technology)

Example 1: Consumer Choice

\[ \max_{x_1, x_2} \ u(x_1, x_2) \quad \text{subject to } p_1 x_1 + p_2 x_2 \leq I \]

  • Choice variables: \(x_1\), \(x_2\) — quantities of goods
  • Objective: maximize utility
  • Constraint: budget

Example 2: Firm Profit Maximization

\[ \max_{L, K} \ \pi = p f(L, K) - wL - rK \]

  • Choice variables: \(L\) (labor), \(K\) (capital)
  • Objective: maximize profit
  • Constraints: often none in short-run, or include capacity or cost restrictions

Choice variables enter the objective function, so changing them changes the value the decision-maker cares about. They are adjusted subject to constraints; you can’t just pick anything, but you choose what’s best from what’s allowed. In mathematical optimization, they are what you're solving for. Without choice variables, there's no decision to make. The solution to an optimization problem is always a value (or set of values) for the choice variables that maximizes or minimizes the objective.

Let’s derive the market supply curve from the firm’s profit maximization problem using the Lagrangian method:

  1. Start with a single firm facing a cost constraint (cost minimization problem as dual)
  2. Derive the firm's supply function
  3. Aggregate to get the market supply curve

1. Setup: Firm’s Profit Maximization Problem

Let’s assume the firm has:

  • A production function: \(f(x)\), where \(x\) is the input
  • Input price: \(w\)
  • Output price: \(p\)

The firm chooses input \(x\) to maximize profit:

\[ \max_{x} \ \pi(x) = p \cdot f(x) - w \cdot x \]

2. Cost Minimization Problem via Lagrangian

We now minimize cost subject to producing a fixed amount \(q\):

\[ \min_{x} \ w \cdot x \quad \text{subject to} \quad f(x) \geq q \]

This is suitable for Lagrangian optimization. Define the Lagrangian:

\[ \mathcal{L}(x, \lambda) = w x + \lambda (q - f(x)) \]

First-Order Conditions (FOCs):

  • \(\frac{\partial \mathcal{L}}{\partial x} = w - \lambda f'(x) = 0 \Rightarrow \lambda = \frac{w}{f'(x)}\)
  • \(\frac{\partial \mathcal{L}}{\partial \lambda} = q - f(x) = 0 \Rightarrow f(x) = q\)

So, the cost-minimizing input \(x^*(q)\) solves:

\[ f(x) = q \quad \Rightarrow \quad x = f^{-1}(q) \]

Then cost function:

\[ C(q) = w \cdot f^{-1}(q) \]

3. Derive Supply Function

Now switch back to the profit maximization side.

Profit:

\[ \pi(q) = p q - C(q) = p q - w \cdot f^{-1}(q) \]

Maximize with respect to \(q\):

\[ \frac{d\pi}{dq} = p - w \cdot \frac{d}{dq} f^{-1}(q) = 0 \]

But from inverse function rule:

\[ \frac{d}{dq} f^{-1}(q) = \frac{1}{f'(x)} \quad \text{where } x = f^{-1}(q) \]

So:

\[ p = \frac{w}{f'(x)} \Rightarrow p \cdot f'(x) = w \]

This matches the first-order condition from profit maximization: value of marginal product = input price.

Solve this equation for \(q\), via \(x\), to get the firm's supply function:

\[ q = f(x^*(p)) \quad \text{where } f'(x^*) = \frac{w}{p} \]

4. Example: Cobb-Douglas Production

Say: \(f(x) = x^\alpha\), \(0 < \alpha < 1\)

  • \(f'(x) = \alpha x^{\alpha - 1}\)
  • Set: \(p \cdot f'(x) = w \Rightarrow p \cdot \alpha x^{\alpha - 1} = w\)
  • Solve for \(x\):

\[ x = \left( \frac{p \alpha}{w} \right)^{\frac{1}{1 - \alpha}} \]

So output: \(q = f(x) = \left( \frac{p \alpha}{w} \right)^{\frac{\alpha}{1 - \alpha}}\)

This is the individual firm’s supply function:

\[ q(p) = A p^{\frac{\alpha}{1 - \alpha}}, \quad A = \left( \frac{\alpha}{w} \right)^{\frac{\alpha}{1 - \alpha}} \]

5. Market Supply Curve

If all firms are identical (simplifying assumption by economists) and there are \(N\) of them:

\[ Q(p) = N \cdot q(p) = N A p^{\frac{\alpha}{1 - \alpha}} \]

This is the market supply curve. The inverse supply function expresses price as a function of quantity supplied rather than the other way around. This is useful because it tells us what price is required to induce a given level of output. It’s particularly helpful for market analysis, equilibrium modeling, and comparative statics.

Recall from earlier:

\[ q(p) = A \cdot p^{\frac{\alpha}{1 - \alpha}}, \quad \text{where } A = \left( \frac{\alpha}{w} \right)^{\frac{\alpha}{1 - \alpha}} \]

Let’s solve for \(p\) in terms of \(q\): Start with:

\[ q = A \cdot p^{\frac{\alpha}{1 - \alpha}} \]

Solve for \(p\):

  1. Divide both sides by \(A\):

\[ \frac{q}{A} = p^{\frac{\alpha}{1 - \alpha}} \]

  1. Take both sides to the power \(\frac{1 - \alpha}{\alpha}\):

\[ p = \left( \frac{q}{A} \right)^{\frac{1 - \alpha}{\alpha}} \]

Substitute back in the expression for \(A\):

\[ A = \left( \frac{\alpha}{w} \right)^{\frac{\alpha}{1 - \alpha}} \quad \Rightarrow \quad \frac{1}{A} = \left( \frac{w}{\alpha} \right)^{\frac{\alpha}{1 - \alpha}} \]

So:

\[ p(q) = \left( q \cdot \left( \frac{w}{\alpha} \right)^{\frac{\alpha}{1 - \alpha}} \right)^{\frac{1 - \alpha}{\alpha}} \]

Or more cleanly:

\[ p(q) = B \cdot q^{\frac{1 - \alpha}{\alpha}}, \quad \text{where } B = \left( \frac{w}{\alpha} \right) \]

(Note: the exponent is positive because \(0 < \alpha < 1\).)

How do we interpret this? As \(q\) increases, \(p\) must increase to induce more output. This reflects increasing marginal cost; typical of most realistic production settings. The elasticity of supply depends directly on the exponent \(\varepsilon\), which comes from technology (i.e., how easily output increases with input).

\[ p(q) = B \cdot q^{\varepsilon}, \quad \varepsilon = \frac{1 - \alpha}{\alpha} > 0 \]

If market supply is \(Q(p) = N \cdot q(p)\), then:

\[ Q(p) = N A p^{\frac{\alpha}{1 - \alpha}} \Rightarrow p(Q) = \left( \frac{Q}{N A} \right)^{\frac{1 - \alpha}{\alpha}} \]

Now economists can make statements about what a market of firms will do under various conditions such as an unforseen policy shock or perhaps a natural disaster.

Now let’s now derive the market demand function from the consumer’s utility maximization problem.

1. Conceptual Setup: Consumer Choice

A consumer wants to choose a bundle of goods that gives the highest utility subject to a budget constraint.

Let:

  • \(x_1, x_2\): quantities of two goods
  • \(p_1, p_2\): prices of goods 1 and 2
  • \(I\): income
  • \(u(x_1, x_2)\): utility function

2. Consumer’s Problem (Utility Maximization)

\[ \max_{x_1, x_2} \ u(x_1, x_2) \quad \text{subject to } p_1 x_1 + p_2 x_2 \leq I \]

We’ll assume the constraint binds (no free money), so:

\[ p_1 x_1 + p_2 x_2 = I \]

We’ll use the Lagrangian:

\[ \mathcal{L}(x_1, x_2, \lambda) = u(x_1, x_2) + \lambda (I - p_1 x_1 - p_2 x_2) \]

3. Solve Using Cobb-Douglas Utility (Example)

Let’s use a classic Cobb-Douglas utility function:

\[ u(x_1, x_2) = x_1^{\alpha} x_2^{1 - \alpha}, \quad \text{where } 0 < \alpha < 1 \]

First-Order Conditions (FOCs)

  1. \(\frac{\partial \mathcal{L}}{\partial x_1} = \alpha x_1^{\alpha - 1} x_2^{1 - \alpha} - \lambda p_1 = 0\)
  2. \(\frac{\partial \mathcal{L}}{\partial x_2} = (1 - \alpha) x_1^{\alpha} x_2^{-\alpha} - \lambda p_2 = 0\)
  3. \(\frac{\partial \mathcal{L}}{\partial \lambda} = I - p_1 x_1 - p_2 x_2 = 0\)

4. Solve the FOCs

Take the ratio of (1) and (2):

\[ \frac{\alpha x_1^{\alpha - 1} x_2^{1 - \alpha}}{(1 - \alpha) x_1^{\alpha} x_2^{-\alpha}} = \frac{\lambda p_1}{\lambda p_2} \Rightarrow \frac{\alpha}{1 - \alpha} \cdot \frac{x_2}{x_1} = \frac{p_1}{p_2} \]

Solve for \(x_2\) in terms of \(x_1\):

\[ x_2 = \frac{1 - \alpha}{\alpha} \cdot \frac{p_1}{p_2} \cdot x_1 \]

Now plug into the budget constraint:

\[ p_1 x_1 + p_2 x_2 = I \Rightarrow p_1 x_1 + p_2 \left( \frac{1 - \alpha}{\alpha} \cdot \frac{p_1}{p_2} \cdot x_1 \right) = I \]

Simplify:

\[ p_1 x_1 \left(1 + \frac{1 - \alpha}{\alpha} \right) = I \Rightarrow p_1 x_1 \cdot \frac{1}{\alpha} = I \Rightarrow x_1 = \frac{\alpha I}{p_1} \]

Plug back into \(x_2\):

\[ x_2 = \frac{(1 - \alpha) I}{p_2} \]

5. Marshallian (Ordinary) Demand Functions

\[ x_1(p_1, p_2, I) = \frac{\alpha I}{p_1}, \quad x_2(p_1, p_2, I) = \frac{(1 - \alpha) I}{p_2} \]

These are the individual demand functions: they show how much of each good the consumer will demand, as a function of prices and income. This function is Homogeneous of degree 0 in \((p_1, p_2, I)\): If all prices and income double, demand stays the same. It is Downward sloping: As \(p_1\) increases, \(x_1\) decreases. Demand for each good is proportional to the budget share: \(\alpha\) and \(1 - \alpha\).

Now lets derive the Inverse demand function, expressing price as a function of quantity demanded, holding income and other prices constant and Market demand, aggregating individual demand functions across consumers. From earlier, we have the Marshallian demand, these express the maximum price the consumer is willing to pay for a given quantity. Demand is downward-sloping: as quantity increases, the price the consumer is willing to pay falls. These functions reflect the marginal utility per dollar, the more you have of a good, the less you're willing to pay for more.

Suppose there are \(N\) identical consumers with the same utility function and income \(I\). Then the market demand is just the sum of individual demands. These are aggregate demand curves for goods 1 and 2 in the market.

\[ X_1(p_1) = N \cdot x_1 = N \cdot \frac{\alpha I}{p_1} = \frac{N \alpha I}{p_1} \]

\[ X_2(p_2) = N \cdot x_2 = N \cdot \frac{(1 - \alpha) I}{p_2} = \frac{N (1 - \alpha) I}{p_2} \]

Rewriting in inverse form:

\[ p_1(X_1) = \frac{N \alpha I}{X_1}, \quad p_2(X_2) = \frac{N (1 - \alpha) I}{X_2} \]

We are almost there. We have the two major ingredients for establishing an equilibrium needed to do comparative statics! In a competitive market, equilibrium occurs at the price \(p^*\) and quantity \(q^*\) where:

\[ \text{Quantity demanded} = \text{Quantity supplied} \quad \Rightarrow \quad Q_d(p^*) = Q_s(p^*) \]

It’s the point where buyers and sellers agree: no excess demand, no excess supply. Assume a single good, \(q\), and a Cobb-Douglas utility and production structure. Recall the market demand and market supply functions from earlier:

\[ Q_d(p) = \frac{N \alpha I}{p} \]

\[ Q_s(p) = N A p^{\frac{\alpha}{1 - \alpha}} \]

This shows upward-sloping supply and downward-sloping demand. Set quantity demanded equal to quantity supplied:

\[ Q_d(p) = Q_s(p) \quad \Rightarrow \quad \frac{N \alpha I}{p} = N A p^{\frac{\alpha}{1 - \alpha}} \]

Cancel \(N\) on both sides:

\[ \frac{\alpha I}{p} = A p^{\frac{\alpha}{1 - \alpha}} \]

Multiply both sides by \(p\):

\[ \alpha I = A p^{\frac{\alpha}{1 - \alpha} + 1} \]

To solve for equilibrium price \(p^*\), let’s simplify the exponent:

\[ \frac{\alpha}{1 - \alpha} + 1 = \frac{\alpha + (1 - \alpha)}{1 - \alpha} = \frac{1}{1 - \alpha} \]

So:

\[ \alpha I = A p^{\frac{1}{1 - \alpha}} \Rightarrow p^{\frac{1}{1 - \alpha}} = \frac{\alpha I}{A} \Rightarrow p^* = \left( \frac{\alpha I}{A} \right)^{1 - \alpha} \]

Now we want to find the Equilibrium Quantity \(q^*\). Plug \(p^*\) back into either supply or demand. Using demand:

\[ q^* = Q_d(p^*) = \frac{N \alpha I}{p^*} = N \alpha I \cdot \left( \frac{A}{\alpha I} \right)^{1 - \alpha} = N \cdot A^{1 - \alpha} \cdot (\alpha I)^\alpha \]

What determines the equilibrium?

  • On the demand side:
    • \(N\): number of consumers
    • \(I\): income
    • \(\alpha\): preference weight on good 1
  • On the supply side:
    • \(A\): a function of technology (\(\alpha\)) and input cost \(w\)
    • \(N\): number of firms

Comparative Statics:

  • Increase in income \(I\):
    • Demand shifts right
    • \(p^*\) increases
    • \(q^*\) increases
  • Decrease in input cost \(w\):
    • Supply shifts right (A increases)
    • \(p^*\) falls
    • \(q^*\) rises
  • More consumers or firms:
    • Increasing \(N\) on the demand side raises \(q^*\) and \(p^*\)
    • Increasing \(N\) on the supply side increases \(q^*\) but reduces \(p^*\)

This is a theoretical model that makes predictions about what will happen empirically. Typically what an economist would do next is collect data and fit an econometric model to test whether the theoretical model matches reality. This is also known as the "Validation" stage of applied mathematical modeling. If the empirical estimates align with the predictions made by the theoretical model, this would act as confirmation of the model; it describes economic reality to a sufficient degree. I'll say more about empirics in later posts. For now, just remember that many economists consider an empirical model to be successful if the signs of the coefficients align with the theoretical model. For example, if the empirical model suggests that an increase in income decreases demand for a product, that would indicate something is wrong with the model. If the model fits poorly to the data (based on econometric criteria of model fit), economists might revise or relax their assumptions, consider alternative functional forms, or move to more advanced methods like non-parametric estimation.

It should also be clear that policy implications flow naturally from economic models. Once a model is formalized and possibly validated empirically, it becomes a tool for counterfactual reasoning; "What would happen if something changes"? The "something" is typically a policy. If something shifts, it gives us a predictive framework for understanding how the system will respond. The model identifies key variables that influence market outcomes so economists can use this to figure out the direction and magnitude of effects of some policy based on the structure of the model. This is where the common textbook examples of policy derive: tax policy, price controls, subsidies etc. I've not yet talked about efficiency, but this is a common theme in economics. Efficiency is a deviation from the optimal market solution. But efficiency is conceptualized very specifically in economics and derives from the assumptions governing these standard models (welfare economics). More on that later. These models also help economists run cost-benefit analyses, compare trade-offs between the short-run and long-run, and allow them to draw conclusions about who will be effected by something.

The model provides clarity about mechanisms and outcomes, it makes our assumptions explicit, and allows simulation of alternatives without trial-and-error in the real world. Policy implications are not just guesses; they are disciplined logical consequences of the structure of the model. The more grounded the model is in both theory and data, the more reliable its guidance becomes for shaping public policy. Often in economics, much dispute arises at the level of assumptions. Differing assumptions lead to differing policy implications, which have a real effect on people. These assumptions ought to be interrogated. Unfortunately, in the public sphere, economic discourse is polluted by ideology. Models are just maps of reality; they are necessarily wrong but can be useful. They are useful for gaining analytical clarity and are fundamental tools of inquiry, not conversation stoppers. Often times, these models become justifications for dogmatic economic ideology. This was the major motivation driving me to write about this because nothing pisses me off more than ideologues at think tanks who pollute the broader public dialogue.

Assumptions

We have made reference to the notion of "assumption" many times throughout this post. There are many assumptions fundamental to what might be referred to as "orthodox economics". Often, these assumptions are what divide economists into varying heterodox "schools of economics". They are essentially different starting points of analysis which lead to different sets of methods and many times, rapidly divergent descriptive and prescriptive conclusions. I first want to discuss more generally the role of assumptions in everyday life. Then we will dive into the most core assumptions fundamental to much of economics. 

An assumption is a proposition that is accepted without proof (at least temporarily), often as a starting point for further reasoning, action, or communication. In modeling, assumptions are simplifications or idealizations that are used to abstract away complexity, make problems mathematically tractable, and define boundary conditions of the model. They are often not intended to be true in the real world, but are meant to produce useful insights under controlled or ideal conditions. Many times, they are required because we have insufficient information or data about the system under consideration. There are different types of assumptions that arise in modeling contexts. Structural assumptions define the form of the model; like linear or nonlinearity, more broadly functional form. There are simplifying assumptions that are used to eliminate complications, such as ignoring friction in simple physical models. There are boundary and initial condition assumptions that define the system limits; such as defining edge cases. There are assumptions about homogeneity or independence; such as iid assumption in statistical models to ignore network effects. In everyday situations, we often have culturual and social assumptions such as "smiling signalling friendliness"; necessary for navigating social life. We have ontological assumptions such as objects persisting through time. We have foundational epistemic assumptions such as the existence of other minds. We assume object permanence, causal relations, and spatial continuity; these are "hardwired"assumptions. 

Assumptions are a necessary part of all human reasoning. We cannot scrutinize every proposition at every moment. If we do not assume something, reasoning cannot begin; we simply couldn't infer anything because all reasoning, argument, and proof requires a starting point. Assumptions are crucial for communication because conversations require shared background context, definitions, and social norms. In modeling, you cannot model everything; and thats the very point of a model anyway, we want to simplify and abstract irrelevant details. Modeling allows us to make our assumptions explicit; often times disagreement arises because of unknown divergent implicit assumptions. Assumptions are heuristics, mental shortcuts that allow rapid decision making in uncertain environments. But unexamined assumptions can lead to bias, error, or dogma. To assume is to bet on something being provisionally true so that we can function, think, or communicate. But all assumptions are contextual, and being aware of them is the key to clarity and critical thinking.

Economics is assumption-heavy due to its abstract nature and the complexity of what we are modeling (human behavior). We often make behavioral assumptions about agents, such as what information they have access to, their goal structure, and whether they're inclined to cooperate or compete. We make assumptions about the environments agents operate within; such as market assumptions about transaction costs, adjustments, and structure. We often make many technical assumptions about continuity, convexity, differentiability, and equilibrium. We also make many assumptions about the dynamics of economic systems. All theories rest on assumptions (uniformity of nature, observer independence etc.); there is simply no way around it. The necessity of assumptions comes with many benefits. Models become tractible; solving equations and running simulations becomes feasible. Assumptions provide clarity; we can focus our attention on core mechanisms and not irrelevant details. Probably one of the most important benefits is comparaibility; shared assumptions make our models comparable. They also enable counterfactual analysis. But there are known risks as well. We could be oversimplifying reality, missing essential real world dynamics. Assumptions may give us a sense of false precision. In economics, there can often be normative smuggling; assumptions about rationality, efficiency, or fairness may be ideologically laden. Incorrect assumptions can also lead to disasterous policy misguidance. 

Given the known risks of assuming something incorrect, the obvious next question becomes: When should we revise our assumptions? Assumption revision is the process of modifying, rejecting, or reaffirming the background beliefs or premises on which a system of thought or action is based. It arises under various scenarios. For example, when new evidence contradicts existing conclusions. This is an application of Poppers falsifiability criterion. If a model fails, we revise the assumption. But this obviously begs the questions; what does it mean for a model to fail? This is not always clear cut. Another example is when internal inconsistencies emerge. If two or more assumptions yield incompatible outcomes, we must revise at least one in order to maintain consistency. More generally, its rational to revise an assumption when New data falsifies the predictions derived from the assumption, The assumption leads to incoherence or contradiction within the model, The assumption creates systematic prediction errors across contexts, Better alternative assumptions produce more robust or simpler explanations, The assumption is based on outdated concepts or technologies (e.g., assuming infinite computation power), or when Stakeholders or society reject the normative basis of the assumption (e.g., utility maximization at all costs). It is irratinoal to stick to an assumption when you ignore contrary evidence due to ideological commitment/convenience/tradition, the assumption is unfalsifiable or immune to revision (dogma), the models predictive failures are blamed on everything else except the assumptions, or when the assumption serves to protect power structures and not explain reality. This is often driven by motivated reasoning, status quo bias, or confirmation bias; all forms of epistemic irrationality. Being rational means being attuned to evidence, willing to adapt, and humble about certainty. Assumption revision is not a weakness; it's a sign of intellectual maturity and rigor. There are many different theories of belief revision; since it is outside the scope of the purpose of my objective here, we can move on to the core assumptions in economic theory. 

Microeconomic theory is based on a set of fundamental assumptions about consumer and firm behavior. These assumptions provide the foundation for economic models and help in deriving demand and supply functions, equilibrium conditions, and welfare implications. For now we will just look at microeconomic fundamentals, but the assumptions underlying modern macroeconomic models (DSGE models and Rational Expectations Theory) are very similar. 
  1. Rationality: Economic agents (consumers and firms) act rationally, meaning they seek to maximize their utility or profit given constraints. For example, a consumer chooses the combination of goods that provides the highest satisfaction within their budget. For a consumer maximizing utility:

    \[ \max_{x} U(x) \]

    subject to:

    \[ p \cdot x \leq w \]

    where:
    \(U(x)\) is the utility function,
    \(x\) is the consumption bundle,
    \(p\) is the price vector,
    \(w\) is the consumer’s wealth.

    For a firm maximizing profit:

    \[ \max_{q} \pi(q) = R(q) - C(q) \]

    where:
    \(q\) is output,
    \(R(q)\) is revenue,
    \(C(q)\) is cost.

  2. Complete Preferences: Consumers can rank all possible consumption bundles. For example, a consumer can compare apples and oranges and decide whether they prefer one to the other or are indifferent. A preference relation \(\succsim\) satisfies completeness if:

    \[ \forall x, y \in X, \quad x \succsim y \text{ or } y \succsim x. \]

    where \(X\) is the set of all consumption bundles.

  3. Transitivity: If a consumer prefers bundle \(A\) to \(B\) and \(B\) to \(C\), then they must prefer \(A\) to \(C\). If a consumer prefers coffee to tea and tea to soda, then they should prefer coffee to soda.

    \[ \forall x, y, z \in X, \quad x \succsim y \text{ and } y \succsim z \Rightarrow x \succsim z. \]

  4. Non-Satiation (Monotonicity): More of a good is always preferred to less, assuming no negative effects. A consumer prefers 5 chocolates over 4 chocolates. If \(x'\) has at least as much of each good as \(x\) and strictly more of at least one good, then:

    \[ x' \succ x. \]

  5. Convex Preferences: Consumers prefer a balanced mix of goods rather than consuming only one type. A consumer prefers a mix of apples and bananas over consuming only apples or only bananas. If \(x \sim y\), then for \(0 \leq \lambda \leq 1\),

    \[ \lambda x + (1-\lambda)y \succsim x, y. \]

  6. Diminishing Marginal Utility: As consumption of a good increases, the additional utility gained from consuming one more unit decreases. The first slice of pizza is highly satisfying, but the tenth slice provides much less additional satisfaction. Mathematically:

    \[ \frac{\partial^2 U}{\partial x^2} < 0. \]

  7. Perfect Information: All economic agents have full knowledge of prices, product quality, and available alternatives. For example, a consumer knows the price of apples at every store and always buys from the cheapest source.
  8. Perfect Competition: Markets have many buyers and sellers, no single agent has market power, and goods are homogeneous. For example, the wheat market has many sellers, and no single farmer can influence the price. Mathematically, each firm is a price taker:

    \[ P = MC(q) \]

  9. No Externalities: All costs and benefits of a transaction are borne by the buyer and seller, with no third-party effects. For example, a factory polluting a river affects nearby residents, violating this assumption. Mathematically, market efficiency is:

    \[ MC_{\text{private}} = MC_{\text{social}}. \]

  10. No Barriers to Entry or Exit: Firms can freely enter or exit the market based on profitability. For example, if the coffee shop industry becomes highly profitable, new competitors can enter the market. Mathematically this means long-run equilibrium requires zero economic profits:

    \[ P = AC. \]

  11. Time-Invariant Preferences: Consumer preferences do not change unpredictably over time. If a person prefers Coke to Pepsi today, they will likely prefer it tomorrow. Mathematically:

    \[ U(x,t) = U(x) \text{ for all } t. \]

  12. Well-Defined Property Rights: Resources have clear ownership, allowing markets to function efficiently. A farmer owns land and can decide how to use or sell it. If \(x\) is owned by agent \(i\), then:

    \[ x \in X_i. \]

  13. Continuity of Preferences: A small change in consumption does not cause abrupt changes in preferences. For example, if a consumer slightly increases the quantity of an orange, their utility does not change dramatically. Mathematically, this means:

    \[ \lim_{x \to x'} U(x) = U(x'). \]

  14. Production Function Assumptions: Firms use inputs efficiently and experience diminishing returns to inputs. Doubling workers in a factory may not double output due to inefficiencies. For a production function \(f(L, K)\), where \(L\) is labor and \(K\) is capital:

    \[ \frac{\partial^2 f}{\partial L^2} < 0, \quad \frac{\partial^2 f}{\partial K^2} < 0. \]

Many of the assumptions in standard microeconomic theory do not hold in real-world settings. Violating these assumptions has significant implications for economic models, often necessitating alternative approaches. Many standard microeconomic assumptions fail in practice, leading to market failures, inefficiencies, and suboptimal decision-making. Alternative models—behavioral economics, game theory, contract theory, and institutional economics—provide more realistic approaches to understanding real-world behavior. Relaxing these assumptions makes economic models more complex but also more applicable to real-world problems. Below are key microeconomic assumptions that are frequently violated and the implications of their violations:

  1. Rationality (Utility and Profit Maximization): Behavioral economics shows that individuals frequently exhibit bounded rationality (Simon, 1955). People rely on heuristics and biases (Kahneman & Tversky, 1979) rather than maximizing expected utility. Firms may satisfice instead of maximizing profit, meaning they aim for a satisfactory rather than the optimal outcome. Standard demand and supply models may fail when consumers make suboptimal choices.
    • Standard models may overpredict rational behavior in decision-making
    • Markets may not clear efficiently due to systematic errors in decision-making
    • Firms can exploit consumer biases for profit
    • Bubbles and irrational behaviors (such as excessive risk-taking) may arise
  2. Perfect Information: Consumers and firms often have incomplete or asymmetric information. A few examples are the lemons problem (Akerlof, 1970), moral hazard, and adverse selection in insurance markets. Search costs and information-processing limitations can also affect choices.
    • Markets may fail due to information asymmetries
    • Firms may exploit consumers using advertising, leading to suboptimal consumption
    • Government intervention may be required (e.g., regulations on truth-in-advertising or financial disclosures)
  3. Transitivity of Preferences: People’s preferences are inconsistent over time (Tversky & Kahneman, 1981). Often they have context-dependent preferences (e.g., framing effects): choices depend on how alternatives are presented. They frequently have cyclical preferences: People may prefer A over B, B over C, but C over A.
    • Standard utility maximization fails because preference orderings are not well-defined
    • Choices may be intransitive, leading to unstable market equilibria
    • The Revealed Preference Theory fails when choices change with context
  4. Perfect Competition (Price Taking Behavior): Many markets have monopolies, oligopolies, and monopolistic competition. Firms engage in strategic pricing, branding, and product differentiation. Market power allows firms to set prices above marginal cost.
    • The first welfare theorem (which states that competitive markets lead to efficiency) does not hold
    • Price distortions lead to deadweight loss
    • Firms engage in rent-seeking behavior (more on this later)
  5. No Externalities: Real-world markets generate negative externalities (pollution) and positive externalities (innovation spillovers). Firms and consumers do not internalize the full social cost/benefit.
    • Market outcomes are not Pareto efficient
    • Public goods (e.g., clean air) are underprovided
    • Tragedy of the commons occurs (Hardin, 1968)

Obviously, there are economists that relax these assumptions if the situation calls for it. You can think of these as default assumptions, or a base case. Economists will make adjustments if this methodological individualism fails to describe the system. For systemic risk, macro crises, and institutional evolution, individual-level analysis is insufficient. Newer models increasingly blend individual and collective dynamics, incorporating social networks, heuristics, and institutional effects. For example, complexity economics rejects many of these assumptions. Here is a table for comparison:

Assumption Standard Micro & Macro Theory Complexity Economics
Rationality Agents are fully rational, optimizing utility or profit Agents have bounded rationality and use heuristics (Adaptive Expectations)
Homogenous Representative Agent A single agent represents an entire sector or economy Heterogeneous agents interact and adapt (Agent Based Models)
Equilibrium Systems converge to a stable equilibrium Systems are out-of-equilibrium, evolving over time (Positive Feedback Dynamics)
Perfect Information Agents have full or at least rational expectations Information is localized and incomplete
Linear Dynamics Small shocks lead to small effects (predictable responses) Systems exhibit nonlinear dynamics and tipping points
Exogenous Shocks Crises are caused by external factors (e.g., policy mistakes) Crises emerge endogenously from network effects (Financial Contagion, Technology Diffusion)
Aggregate Behavior Macro outcomes result from simple aggregation of individual behavior Emergence: Macro outcomes arise from micro interactions (Non-Linear and Evolutionary)

When I began heavily questioning the assumptions of economic models in graduate school, I came across a well-known result in computer science and computational complexity theory that demonstrates the inherent computational difficulty of finding equilibrium in economic models. The result shows that many equilibrium problems in economics, particularly those based on fixed-point theorems, are NP-hard or even in the complexity class PPAD-complete (Polynomial Parity Argument on Directed graphs). It took a while for me to wrap my head around this because Economists typically do not study computational complexity, and hence would never imagine questioning the equilibrium assumption in economics. Theoretical economic models assume that equilibrium exists (e.g., Walrasian general equilibrium, Nash equilibrium). However, from a computational complexity perspective, actually computing these equilibria is often infeasible in practice. Several studies have shown that finding general equilibrium prices or Nash equilibria is at least NP-hard. This means It may take exponentially long computation time to find an equilibrium, making it impractical for real-world markets. This implies that markets may not reach equilibrium in reasonable time scales, meaning market equilibrium assumptions might not hold in practice. I think Kenneth Arrow proved that under certain assumptions, economies described by the competitive model will have a unique equilibrium. But then (Daskalakis, Goldberg & Papadimitriou, 2006) showed that, even if an equilibrium exists (by Nash’s theorem), finding it is as hard as any problem in the PPAD class. Nash's theorem relies on Brouwer's fixed-point theorem but Daskalakis et al. showed that computing such a fixed point is PPAD-hard, meaning there is no polynomial-time algorithm unless PPAD problems are easy. In large markets or games, even if an equilibrium theoretically exists, it may be computationally impossible to find. This questions the practical relevance of equilibrium-based economic models. (https://people.cs.pitt.edu/~kirk/CS1699Fall2014/lect4.pdf)

I also began asking about what happens if we have heterogenous utility functions across the collection of consumers and whether utility functions are strictly independent from one another. How can we do any aggregation? Turns out, aggregation under heterogenous preferences is a known issue. We cannot aggregate individual behavior cleanly unless very specific conditions hold. If each consumer \( i \) has their own utility function \( U_i(x_i, y_i) \), we can’t, in general, assume that aggregate demand behaves like a “representative” consumer’s demand. Why?

  • Income Effects Differ Across Consumers: Suppose one consumer has a strong income effect, another has a weak one. As total income or prices change, aggregate demand won’t behave like any single demand function — the composition of demand changes.
  • Preferences Might Not Be Homothetic: If utilities are non-homothetic (e.g., demand depends on income in nonlinear ways), the shape of aggregate demand depends on the income distribution — not just total income. That makes aggregate demand non-representable by a single utility function in general.
  • Non-Separability & Interdependence: If consumers' utilities interact (e.g., network effects, social preferences), you can’t even write their problem as separate maximization problems. For example: \( U_i(x_i, y_i; x_j) \) — consumer \( i \)'s utility depends on what consumer \( j \) does. Aggregation fails hard in this case; the economy is strategic, not just additive.

In comes The Sonnenschein-Mantel-Debreu (SMD) Theorem: Almost any shape of market demand can be generated by aggregating rational individual demands, even if each consumer behaves "nicely" (i.e., maximizes utility, has well-behaved preferences). Aggregate demand functions need not satisfy The law of demand, Uniqueness, Smoothness, or Downward-sloping structure. Even if all individual preferences are convex, continuous, monotonic, etc., aggregate demand can be wild. There are special cases where you can cleanly aggregate.

Case 1: Identical, Homothetic Preferences: Everyone has the same utility function, and it’s homothetic (e.g., Cobb-Douglas, CES). Then aggregate demand depends only on total income, not its distribution. A representative consumer exists.

Case 2: Gorman Polar Form: The Gorman form shows when individual demands can be aggregated: If each consumer’s indirect utility function is quasi-linear in income, i.e.,

\[ v_i(p, m_i) = a_i(p) + b(p) \cdot m_i \]

and all consumers share the same marginal utility of income function \( b(p) \), then aggregate demand can be represented as if from one representative consumer. This condition is very restrictive. If utility functions differ, especially non-homothetically, and income effects vary, then aggregation fails. You cannot represent aggregate demand with a single utility function. The distribution of income and preferences matters deeply. And if preferences are not independent (e.g., they depend on others), then you are no longer even in a “representative agent” world, you’re in game-theoretic territory. This is fascinating; Arrow and Debreu have shown that in most cases (because of aggregation problems), there will probably not be a unique stable equilibrium. In the ideal case, where there is a stable equilibrium, Papadimitriou essentially shows that it will take an economy exponentially long to find it. So what are economists doing with their time? Who knows. Earlier I mentioned the rationality of assumption revision, and the general drawbacks of assuming something false. I think the implications are clear. I should specify that these results are applicable in macroeconomics, but I don't see how they wouldn't also be applicable to single markets which are just localized instantiations of the general problem.

"Almost a century and a half after Léon Walras founded general equilibrium theory, economists still have not been able to show that markets lead economies to equilibria. We do know that — under very restrictive assumptions — equilibria do exist, are unique and are Pareto-efficient. But — what good does that do? As long as we cannot show that there are convincing reasons to suppose there are forces which lead economies to equilibria — the value of general equilibrium theory is nil. As long as we cannot really demonstrate that there are forces operating — under reasonable, relevant and at least mildly realistic conditions — at moving markets to equilibria, there cannot really be any sustainable reason for anyone to pay any interest or attention to this theory."

In order to have a coherent market demand function, you typically need these assumptions:

Assumption Why It's Needed What Happens if It Fails
1. Homothetic preferences So Engel curves (income → quantity) are straight lines through the origin — demand only depends on relative prices and total income. Demand becomes sensitive to income distribution, not just total income — aggregation fails.
2. Identical preferences Ensures that income effects and substitution effects are similar across individuals. Different preferences create non-canceling income effects, making demand shapes unpredictable.
3. No wealth effects (or quasi-linear utilities) Gorman polar form requires linear Engel curves with the same slope across consumers. Consumers react differently to income changes → aggregation fails.
4. Independent preferences Each consumer’s utility is independent of others’. Interdependent preferences (e.g., externalities, network effects) destroy separability. No meaningful aggregation.
5. Complete markets & no rationing Ensures each agent optimizes fully. If constraints exist (e.g., liquidity, quantity rationing), individual demands don’t reflect preferences alone.
6. Convex preferences Makes individual demands well-behaved (single-valued, continuous, responsive). Non-convexity introduces multiple optima, discontinuities, or non-monotonic demand — invalidates aggregation.
7. Perfect information and price-taking behavior Ensures individual demands are responsive only to prices and income. If strategic behavior or uncertainty exists, individual choices reflect beliefs, not pure preferences.

Generally these are the issues:

  • Different marginal propensities to consume across individuals → income redistribution changes aggregate demand.
  • Presence of luxuries and necessities → demand depends on income distribution, not just totals.
  • Social preferences, peer effects, positional goods → utility is not separable.
  • Price-dependent wealth effects → demand depends on who holds wealth, not just how much exists.
  • Differential exposure to prices (e.g., subsidies, taxes, discrimination) → different consumers face different effective prices.
  • Behavioral heterogeneity (bounded rationality, reference dependence, etc.) → violates utility maximization assumptions.
  • Multiple equilibria or discontinuities → market demand can't be summarized by a stable function.

Efficiency

This concept is actually what prompted me to write this post. The word is so misunderstood and so misused that I figured it best to not address it in isolation, but in conjunction with related economic terminology to provide a richer context. It's bad enough listening to non-economists talk about policy or economic systems, but when DOGE was propped up during this new presidential term, I simply lost my mind. I am not trying to gatekeep, but words have meaning, and when "inefficient" is conflated with "thing i don't like", we are beyond the realm of reasonable discussion and have traversed into "you simply need an education." Combine this with the fact that my literal occupation revolves around long term strategic planning in a massive enterprise, where at the core we are fundamentally concerned with optimality and efficiency, hearing the general public be fooled by such rhetoric made me realize most of the population operates with folk economic ideologies. There you go, this is why I decided to write a post no one will read; perhaps it was to convince myself that I am not the crazy one. Anyways, in this section, we will define efficiency in a very broad sense, then see how economists understand the term in relation to economic methodology. 

Efficiency refers to the ability to accomplish a task or achieve a goal with minimal waste of time, effort, or resources. It often implies maximizing output while minimizing input, whether in terms of energy, cost, or labor. Broadly speaking, it is "doing more with less" or "getting the most out of inputs". In economics, efficiency refers to the optimal use of scarce resources to maximize output, wealth, or welfare. It generally falls into three main types: 
  • Allocative Efficiency – Resources are distributed in a way that maximizes societal welfare, meaning goods and services are produced according to consumer preferences. It occurs when the price of a good equals its marginal cost (P = MC).
  • Productive Efficiency – Goods and services are produced at the lowest possible cost, meaning no additional output can be achieved without increasing input costs. This happens when firms operate at the lowest point of their average cost curve.
  • Pareto Efficiency (Pareto Optimality) – A state where no one can be made better off without making someone else worse off. It represents an ideal allocation of resources where improvements in one area would require trade-offs in another.
Economists often analyze efficiency to determine whether markets are functioning optimally or if government intervention is needed to correct inefficiencies (e.g., externalities, monopolies, or public goods). The economic concept of efficiency, particularly allocative and productive efficiency, is deeply rooted in the assumptions of the competitive equilibrium model. The first fundamental theorem of welfare economics states that, under the assumptions of perfect competition (no externalities, complete markets, perfect information, and price-taking behavior), a competitive market equilibrium leads to a Pareto-efficient allocation of resources. This means that no one can be made better off without making someone else worse off. The second fundamental theorem states that, any Pareto-efficient allocation can be achieved through market mechanisms, given appropriate redistribution of initial endowments (e.g., through lump-sum transfers or taxation). These efficiency outcomes rely on strong assumptions that do not hold in real-world economies; market failures such as externalities, information asymmetries that disrupt price signals, transaction costs and barriers to entry, and behavioral biases of the economic agents. Think of the welfare theorems as benchmarks by which to compare actual economic outcomes against. I think I made my bias clear earlier, these assumptions are fictitious and impossible to instantiate, therefore I find it highly suspect to conclude that actual markets are operating in an economically efficient sense. 

In the real world, many business that are of sufficient scale tend to hire specialists that have an expertise in a discipline that can assist in improving business processes and decision making. This can include disciplines like operations research, data science, statistics, industrial engineering, or something related, where mathematical modeling of real world systems takes precedence. Operations research (OR) is a toolkit for making better decisions with limited resources. It focuses on modeling systems and using mathematics and computation to find efficient and often optimal decisions. Below is an overview of key OR methods and how they help with efficiency, optimization, and decision support:
  • Linear Programming (LP): Formulates decision problems as a linear objective (such as maximizing profit or minimizing cost) subject to linear constraints that capture limited resources, capacities, and business rules. LP solvers search the feasible region and identify the best combination of decision variables (e.g., production quantities, workforce levels, shipping amounts) that satisfies all constraints. By providing optimal allocations and sensitivity information (like shadow prices and reduced costs), LP helps managers understand trade-offs, identify bottlenecks, and make efficient, data-driven decisions.
  • Integer & Mixed-Integer Programming (IP/MIP): Extends linear programming by requiring some or all decision variables to be integers (e.g., 0–1 yes/no choices, counts of trucks, machines, or facilities), making it ideal for planning and design problems where fractional decisions are meaningless. IP/MIP models capture logical relationships (open/close, assign/not assign, either-or) and complex constraints that reflect real-world operational rules. By searching over discrete combinations of decisions, these models uncover high-quality, implementable plans for scheduling, routing, location, and capacity expansion, directly supporting strategic and operational decision-making.
  • Nonlinear Programming (NLP): Handles optimization problems where the objective or constraints are curved (nonlinear), such as diminishing returns, nonlinear costs, physical laws, or risk measures. NLP can capture realistic relationships like nonlinear fuel consumption, power flows, or portfolio risk-return profiles that linear models cannot accurately represent. By optimizing over these nonlinear relationships, NLP provides decision makers with feasible, often more realistic “best possible” solutions that balance performance, cost, and risk in engineering, energy, and financial applications.
  • Multi-Objective Optimization: Simultaneously considers several objectives (such as cost, service level, environmental impact, and risk) instead of collapsing everything into a single metric. These models generate a set of Pareto-efficient solutions that represent different trade-offs among objectives, rather than one “one-size-fits-all” answer. Decision makers can then explore this efficient frontier, compare scenarios, and select the alternative that best reflects their strategic priorities and preferences, enabling transparent and structured decision support.
  • Network Optimization: Represents systems as nodes (locations, facilities, servers) and arcs (roads, pipelines, communication links) with capacities and costs, then finds the most efficient routes, flows, or configurations. Classic problems include shortest paths, minimum-cost flows, and network design for logistics, transportation, and telecommunication networks. By optimizing how goods, information, or resources move through the network, these models reduce transportation and operating costs, improve reliability, and support strategic decisions about where to invest in new capacity.
  • Stochastic Programming: Explicitly incorporates uncertainty in parameters like demand, prices, processing times, or failures by modeling multiple scenarios with associated probabilities. Decisions are typically split into “here-and-now” choices (made before uncertainty is realized) and “recourse” actions (adjustments made after outcomes are known). This framework produces solutions that hedge against risk, perform well on average across scenarios, and help decision makers understand the value of flexibility and contingency plans under uncertainty.
  • Robust Optimization: Focuses on solutions that remain feasible and effective across a range of uncertain parameter values, without requiring detailed probability distributions. It defines uncertainty sets (e.g., demand within a range, costs varying within a band) and finds decisions that perform well under worst-case or adverse conditions. This approach is especially valuable when data is scarce or noisy, delivering plans that are more resilient to surprises and reducing the likelihood of costly constraint violations or service failures.
  • Queueing Theory: Models systems where customers, jobs, or tasks arrive, wait, and receive service (e.g., call centers, hospital ERs, repair shops, checkouts). Using arrival and service rate assumptions, it predicts key performance measures such as average waiting time, queue length, utilization, and probability of delay. These insights help managers choose staffing levels, number of servers, and priority rules that balance service quality with cost, enabling efficient capacity planning and service-level management.
  • Discrete-Event Simulation: Builds a time-based, virtual model of processes where events (arrivals, service completions, breakdowns, shifts) occur at discrete points in time. By simulating the system many times under different configurations, managers can test “what-if” scenarios—such as adding staff, changing layout, or modifying rules—without disrupting real operations. Simulation reveals bottlenecks, variability effects, and system-wide interactions, providing rich decision support for improving throughput, reducing congestion, and designing more efficient workflows.
  • Monte Carlo Simulation: Uses random sampling from probability distributions to propagate uncertainty through models and generate distributions of outcomes (e.g., costs, profits, completion times). This technique helps quantify risk by estimating metrics like the probability of meeting a target, worst-case losses, or expected overruns. Decision makers can then compare alternatives not just on single-point estimates but on their full risk profiles, leading to more informed choices about buffers, contingency plans, and risk-reward trade-offs.
  • Heuristics & Metaheuristics: Provide general-purpose search strategies (such as genetic algorithms, simulated annealing, tabu search, or GRASP) designed to quickly find good, near-optimal solutions to very large or complex problems where exact optimization may be too slow or impossible. These methods explore the solution space intelligently, using rules of thumb and guided randomness to escape local optima and improve solutions over time. They are widely used for routing, scheduling, design, and planning problems, where they deliver high-quality, implementable decisions within practical time limits, enhancing operational efficiency.
  • Decision Trees: Represent sequential decisions, chance events, and outcomes in a branching tree structure, with probabilities and payoffs attached to each branch. By “rolling back” the tree using expected values, managers can systematically compare strategies, account for uncertainty, and determine the optimal decision at each stage. Decision trees also support the evaluation of information-gathering actions (such as tests or market research), helping quantify the value of information and clarify complex, multi-step decision problems.
  • Multi-Criteria Decision-Making (e.g., AHP, TOPSIS): Helps evaluate and rank alternatives when multiple, often conflicting criteria matter, such as cost, quality, risk, sustainability, and strategic fit. Methods like AHP structure the problem into a hierarchy and elicit relative importance weights from decision makers, while techniques like TOPSIS rank alternatives based on their closeness to an ideal solution. These approaches transform subjective preferences into transparent, quantitative scores and rankings, enabling consistent, defendable decisions in complex evaluations like vendor selection, project prioritization, or technology choice.
  • Inventory Models (EOQ, reorder points, (s,S) policies): Determine how much to order and when to reorder to balance ordering costs, holding costs, and the risk of stockouts. Models like EOQ give optimal lot sizes under stable demand, while reorder point and (s,S) policies incorporate variability and service-level targets. By linking inventory decisions to demand patterns, lead times, and cost parameters, these models support efficient inventory control, reduce excess stock, and improve product availability and customer service.
  • Data Envelopment Analysis (DEA): Evaluates the relative efficiency of comparable decision-making units (DMUs)—such as plants, branches, hospitals, or service centers—by comparing multiple inputs (e.g., labor, capital, materials) and outputs (e.g., units produced, patients treated, services delivered). DEA constructs an empirical “efficient frontier” and measures how far each unit is from this frontier, identifying efficient peers and quantifying potential input savings or output increases. This information supports performance benchmarking, target setting, and resource reallocation, guiding managers toward more efficient operations and best practices across the organization.
Private enterprises employ these methods specifically with the goal of ultimately maximizing profits. In the private sector, efficiency is relatively straightforward because it is tied to profit maximization—a company is considered efficient if it can produce more output with fewer inputs while maintaining or increasing profitability. The metrics are clear: cost reduction, revenue growth, return on investment (ROI), and shareholder value. In contrast, government efficiency is much harder to define and measure, primarily because:
  • Multiple, often conflicting objectives – Governments aim to maximize social welfare, which includes economic growth, healthcare, education, infrastructure, and public safety. These goals don’t always align or have a clear financial return.
  • Lack of a single performance metric – Unlike profit, government performance is measured through qualitative and quantitative indicators like literacy rates, life expectancy, poverty reduction, and public satisfaction. These are difficult to compare directly.
  • Political constraints – Decision-making in government is influenced by political pressures, elections, interest groups, and bureaucracy, which can lead to inefficient allocation of resources.
  • Equity vs. Efficiency Trade-offs – Governments often prioritize fairness and accessibility over pure efficiency. For example, providing healthcare to all citizens may be less cost-efficient than a private insurance model but is justified on social grounds.
It's not to say that measuring the efficiency of a governing body is impossible, its just that government decisions are of a different category with different considerations other than profit maximization. They still seek to minimize costs but these policy decisions are often more complex with objectives that are much more vague. Some of which include impact assessment, program evaluation, and various methods of policy analysis that often include econometric approaches. Some argue that governments should adopt private-sector efficiency principles (e.g., performance-based funding, lean management). Others caution that governments have unique responsibilities, such as protecting vulnerable populations, which means efficiency cannot be the sole priority. I obviously sit in the latter camp. Many mixed economies do some combination of the two philosophies. 

Allocative efficiency in a government setting is inherently tied to values and political choices, unlike productive efficiency, which can be optimized through technical methods. In a private market, allocative efficiency is relatively clear: resources should go to their highest-value use as determined by consumer demand and willingness to pay. But in a government setting, it is far more complex because there is no price mechanism, social benefits are subjective, ethical considerations are involved, and equity concerns may override efficiency concerns. This means that, there may be instances where governments might prioritize a program with little to no direct or immediate economic benefits. For example, investment in infrastructure might have higher return than education, but the government might prioritize that for long-term social benefits. Some programs such as national defense are inherently difficult to measure. Ultimately, allocative efficiency in government is not purely a technical question but a normative one—it requires an argument from values, not just numbers. What should the government prioritize? Economic growth? Social welfare? National security? What trade-offs are acceptable? Should a program be cut if it helps fewer people but in a deeply meaningful way? Who decides? Democratically elected officials, expert committees, or public opinion? 

Evaluating government programs involves assessing their design, implementation, and impact to determine effectiveness and efficiency. Different evaluation methods help policymakers decide whether to continue, modify, or discontinue programs. Below are the main types of program evaluation methods, categorized based on the evaluation focus.

1. Program Design Evaluations

Program design evaluations focus on whether a program’s design is logical, feasible, and likely to achieve its objectives before it is fully implemented.

  • A. Needs Assessment. A needs assessment determines whether a problem exists and if a government program is truly necessary, often using surveys, focus groups, and statistical analysis of demographic or economic data to clarify the nature and extent of the issue. For example, before launching a job-training program, a government might conduct a needs assessment to determine whether unemployment stems from a lack of skills or from other factors such as limited job availability or structural barriers in the labor market.
  • B. Theory of Change / Logic Model Analysis. Theory of change or logic model analysis evaluates whether the program’s underlying logic—how inputs and activities are expected to produce outputs and ultimately outcomes—is coherent and realistic. This type of evaluation often relies on stakeholder interviews, expert reviews, and literature reviews to assess whether the proposed causal pathways are supported by evidence. For instance, a homelessness prevention program should clearly explain how providing housing subsidies and related services will lead to reduced homelessness rates, and a logic model analysis tests whether that pathway is plausible.
  • C. Feasibility and Pilot Studies. Feasibility and pilot studies test a program’s viability on a small scale before broader implementation, using small-scale trials, randomized controlled trials (RCTs), or process simulations to identify potential problems and refine the design. For example, a new voting system might first be piloted in select districts so that issues with technology, logistics, or voter behavior can be addressed before the system is scaled up to a national level.

2. Implementation Evaluations

Implementation evaluations examine how well a program is being delivered in practice and whether it follows its intended structure and procedures.

  • D. Process Evaluation. Process evaluation investigates whether program activities are being carried out as planned by looking closely at program operations through site visits, administrative data analysis, and interviews with program staff. For example, in a government food assistance program, a process evaluation would examine whether benefits are reaching the intended recipients on time and in the correct amounts, and whether administrative procedures are functioning effectively.
  • E. Fidelity Assessment. Fidelity assessment checks whether a program is being implemented according to its original design and standards, relying on methods such as direct observation, program audits, and comparisons of implementation data with program manuals or guidelines. For instance, if a tutoring program for struggling students is designed to provide 10 hours of weekly instruction but records show that students receive only 5 hours, a fidelity assessment would identify this gap as a deviation from the intended model.
  • F. Capacity and Resource Evaluation. Capacity and resource evaluation determines whether a program has sufficient funding, personnel, and technology to operate effectively, typically using budget analysis, workforce capability studies, and infrastructure assessments. For example, a government healthcare program might have a strong policy design but still fail to achieve its goals because there are not enough doctors, nurses, or clinics in rural areas to deliver services to the target population.

3. Outcome and Impact Evaluations

Outcome and impact evaluations assess whether a program achieves its intended goals and whether observed changes can be attributed to the program itself rather than to other factors.

  • G. Performance Monitoring (Key Performance Indicators – KPIs). Performance monitoring tracks ongoing program success using predefined key performance indicators (KPIs), often presented through data dashboards, regular reports, and trend analyses. For example, a job placement program may monitor the percentage of participants who find employment within six months of completing training, using that metric to gauge whether the program is performing as expected over time.
  • H. Cost-Benefit Analysis (CBA). Cost-benefit analysis compares the total social benefits of a program to its total costs in monetary terms, using economic modeling and comparisons with historical or alternative data to estimate net gains or losses. For example, a public preschool program might cost $10 million to operate but generate an estimated $50 million in future economic benefits through higher lifetime earnings, reduced crime, and lower remedial education costs, leading analysts to judge the program as highly cost-effective.
  • I. Cost-Effectiveness Analysis (CEA). Cost-effectiveness analysis compares the costs of a program to its non-monetary benefits—such as lives saved, illnesses prevented, or students educated—through statistical analysis and, often, longitudinal studies. For instance, two public health campaigns might be compared based on their cost per life saved or cost per case of disease prevented, allowing policymakers to choose the option that achieves similar or better outcomes at a lower cost.
  • J. Randomized Controlled Trials (RCTs). Randomized controlled trials test cause-and-effect relationships by randomly assigning participants to a group that receives the intervention and a control group that does not, and then tracking outcomes for both groups over time. Using rigorous experimental design and outcome measurement, an RCT can determine whether a program truly causes the observed effects. For example, to test a job training program, participants are randomly assigned either to receive training or not, and later comparisons of employment rates reveal whether the program significantly improves job prospects.
  • K. Quasi-Experimental Designs. Quasi-experimental designs estimate program impact when random assignment is not feasible, using methods such as difference-in-differences, propensity score matching, or regression discontinuity. These techniques compare outcomes across groups or time periods that differ in exposure to the program but are otherwise similar, to approximate causal effects. For instance, analysts might compare crime rates before and after a new policing strategy is implemented in one city, using another city without the strategy as a comparison group.

4. Policy and System-Level Evaluations

Policy and system-level evaluations look beyond individual programs to assess broader, long-term policy effectiveness and system-wide impacts.

  • L. Equity and Distributional Analysis. Equity and distributional analysis examines whether program benefits are fairly distributed across different income groups, racial or ethnic groups, and geographic regions by analyzing disaggregated data and engaging stakeholders from affected communities. For example, an equity analysis of a government housing program might reveal that most benefits flow to urban areas while rural communities receive relatively little support, prompting reforms to address these disparities.
  • M. Sustainability Assessment. Sustainability assessment evaluates whether a program can continue to deliver benefits over time without relying on unsustainable levels of funding or external support, often using long-term financial modeling and impact forecasting. For instance, a government solar panel subsidy program might be assessed to determine whether households will continue adopting solar technology once subsidies are reduced or removed, and whether the program’s financial structure can be maintained over the long run.
  • N. Comparative Policy Analysis. Comparative policy analysis compares a program’s design and performance with similar policies or programs in other regions or countries, using benchmarking and policy literature reviews to identify best practices and areas for improvement. For example, analysts might compare U.S. healthcare policy outcomes with those of universal healthcare systems in Europe to understand differences in cost, access, and health outcomes, and to inform potential reforms.

Selecting the Right Evaluation Method

The appropriate evaluation method depends on the stage and purpose of the assessment. Before launching a program, tools like needs assessments, theory of change analyses, and feasibility or pilot studies help determine whether a program is necessary and likely to work. During program operation, process evaluations, fidelity assessments, and capacity and resource evaluations reveal whether the program is being implemented as intended and whether it has the means to function effectively. To assess impact, approaches such as randomized controlled trials, quasi-experimental designs, cost-benefit analysis, cost-effectiveness analysis, and performance monitoring with KPIs help determine whether the program is achieving its goals and doing so efficiently. For long-term policy effectiveness and system-wide understanding, equity and distributional analysis, sustainability assessment, and comparative policy analysis provide insights into fairness, durability, and how a policy compares to alternatives in other jurisdictions.

I'll elaborate on these empirical approaches in another post. We will discuss econometric methods of policy evaluation that utilize the counterfactual framework. Some methods include: Difference-in-Differences, Regression Discontinuity Design, Instrumental Variables, Propensity Score Matching, and Synthetic Control Method. This is standard methodology for any applied econometrician or microeconomist with an empirical emphasis. 

At the beginning of this section, I mentioned various causes of market inefficiency. Governments can potentially correct these market failures through mechanism design or well crafted policy. There are market fundamentalists however, that hold a dogmatic view that markets always allocate resources most efficiently. This belief is rooted in the assumption that "free markets" are the only mechanism that can achieve optimal resource allocation through voluntary exchanges, price signals, and competition. Market fundamentalists define allocative efficiency strictly in terms of free-market distribution. Mainstream and heterodox economics recognizes that government can correct market failures, leading to a more efficient overall allocation. The real debate is not whether government intervention is inefficient, but where and how it should intervene. This is why this topic infuriates me. When DOGE strips funding to the Consumer Financial Protection Bureau or USAID, this has nothing to do with efficiency in any of the senses I've listed. They are not conducting cost analyses or implementing operations research methods to determine the sources of inefficiency. These are value-laden decisions, masked in official sounding terminology like "inefficiency", driven by market fundamentalist ideology, which ultimately supports and reinforces our crony capitalism

Risk and Uncertainty

I'll start by saying that this section is not about risk analysis, risk assessment, or risk management per se; although there will be significant conceptual overlap. A detailed investigation of those subjects would fall under actuarial sciences, but could be found in sub-domains of economic departments that have strong mathematical emphases. Instead, this section will focus on how economists incorporate these concepts into their models of the economy. While risk analysis, assessment, and management might occur within the context of an enterprise trying to understanding its exposures, economists directly incorporate risk concepts into their models of the economy; the "enterprise" economists are modeling is itself "the economy". So for example, an economist might model an economy with agents having certain risk profiles, and economic processes that are subject to varying degrees of uncertainty due to exogenous shocks. Economists might then use Monte Carlo Methods to characterize the uncertainty of their model outputs under varying parametric assumptions. So in this section, we will first conceptualize risk and uncertainty, then see how economists incorporate these concepts into their models, and then finally understand how they inform conclusions. 

Risk is often described very simply as consequence multiplied by probability. If an event has no probability of occurrence, then there is no risk. Likewise, if there is no consequence or undesirable outcome, there is no risk. A hazard is a thing that causes the potential for some adverse consequence. Opportunities are causes that have the potential for a positive consequence. If there is no hazard or opportunity, there will be no consequence and therefore no risk. The range of possible consequences can range in frequency, magnitude, severity, and duration. Notice that a key component of risk is probability. Uncertainty is a distinct but related concept, described by a situation where we cannot assign a probability to an outcome, and is ultimately related to limitations in our knowledge. Risk often refers to situations where there is natural variability which can be described by a probability distribution, and where the hazards are fully enumerated (we have a relatively complete understanding of the data generating mechanism). Uncertainty involves situations where we cannot even assign a probability distribution. You might have vague, conflicting, or incomplete info, or think the environment could change in ways you can’t summarize with one probability measure. For example, new technology or political regime change where you don’t even know the relevant scenarios, let alone their probabilities. This is referred to as Knightian Uncertainty within economics, named after economist Frank Knight. 

Let's now see how economists treat risk in standard economic models.

a. Expected utility over lotteries

The workhorse representation:

  • There’s a set of states of the world (\(S = \{s_1, \dots, s_n\}\)).
  • A lottery (or risky prospect) is a vector of outcomes with probabilities, e.g.:
    \[ L = (x_1, p_1; x_2, p_2; \dots; x_n, p_n) \]
  • Under von Neumann–Morgenstern expected utility:
    \[ U(L) = \sum_{i} p_i \, u(x_i) \]
  • Risk attitude is captured by the curvature of \(u(\cdot)\):
    • Concave → risk-averse
    • Linear → risk-neutral
    • Convex → risk-loving

So risk here = lotteries with known \(p_i\).

b. Subjective expected utility

Savage’s subjective expected utility theory allows:

  • Probabilities to be subjective beliefs rather than “objective frequencies.”
  • But still a single probability measure \(P\) over states.
  • Preferences over acts (functions from states to outcomes) can be represented as:
    \[ V(f) = \int u(f(s)) \, dP(s). \]

Here, risk is still “well-defined”: the agent behaves as if they have a single coherent probability distribution.

c. General equilibrium and Arrow–Debreu

In Arrow–Debreu general equilibrium:

  • There are state-contingent commodities and state-contingent securities.
  • Uncertainty enters as multiple states \(s\), and a security might pay 1 unit in state \(s\) and 0 in others.
  • If markets are complete, agents can fully insure against risk by trading these securities.

Risk is a structure on the state space plus a probability measure over it; prices encode how the market aggregates these risks.

d. Finance: mean–variance, CAPM, etc.

In finance, risk often appears as:

  • Random returns \(R\) with known distribution.
  • Variance or standard deviation as a summary of risk.
  • E.g. Markowitz mean–variance: investors choose portfolios to trade off expected return vs variance.
  • CAPM, APT, etc., treat shocks as random variables with known distributions; risk that cannot be diversified away is priced.

e. Macroeconomics and dynamic models

In macro (RBC/DSGE models):

  • There are shocks to productivity, preferences, etc., modeled as stochastic processes (e.g. \(AR(1)\) with normal innovations).
  • Agents know the stochastic law of motion and form rational expectations (their subjective distribution = true distribution).

Again: risk = stochastic shocks with known distributions, embedded in dynamic optimization.

f. Game theory

In games:

  • Nature’s moves (e.g. a type or state) are modeled with a known probability distribution.
  • Players have beliefs about others’ types and actions, often represented as a probability measure.

This is risk about unknown but probabilistically well-characterized events.

I actually can't describe in great detail how economists go about modeling Knightian Uncertainty; that's typically not standard curriculum for masters students. From what I can gather, it often involves relaxing assumptions, incorporating model uncertainty, and allowing for multiple priors. Different groups of economists treat risk and uncertainty somewhat differently, depending on their methodological commitments, empirical focus, and philosophical views about probability and knowledge. The same words—“risk,” “uncertainty,” “ambiguity”—can therefore mean slightly different things in different subfields.  

Mainstream micro / finance

In mainstream microeconomics and modern finance theory, the default approach is to model virtually all randomness as risk in the sense of well-defined probabilities. Preferences are typically represented using von Neumann–Morgenstern expected utility or close variants (such as subjective expected utility or models with time or state separable utility). If the true data-generating process is not objectively known, it is standard to assume that agents hold subjective probability distributions over states of the world and then maximize expected utility with respect to those beliefs. In this framework, saying that “you don’t know the true probabilities” is resolved by the idea that “your beliefs are your probabilities”: whatever uncertainty you face is captured by a single coherent subjective prior.

Ambiguity and Knightian uncertainty are acknowledged, but they are usually treated as more specialized topics rather than as the baseline. They tend to appear in decision theory, certain parts of asset pricing (e.g., models with ambiguity-averse investors), or specialized macro and finance papers. The core textbooks and canonical models, however, remain firmly within the expected-utility, single-prior framework, where risk is always representable by a probability measure (objective or subjective) over a fixed state space.

Behavioral and experimental economics

Behavioral and experimental economics place strong emphasis on documented deviations from standard expected-utility behavior under risk. A central example is prospect theory and its later refinements, which incorporate features such as reference dependence, loss aversion, and probability weighting to better match observed choices. Other models, like rank-dependent utility and cumulative prospect theory, similarly modify how probabilities enter the utility calculation, capturing the empirical finding that people tend to overweight small probabilities and underweight moderate to large ones.

This literature also pays particular attention to ambiguity aversion: robust evidence from experiments (such as Ellsberg-type choices) shows that people often treat unknown odds differently from known odds, even when expected payoffs are matched. To capture this, behavioral and decision-theoretic models frequently use non-additive probabilities (capacities), probability weighting functions, or multiple-priors (max–min or smooth ambiguity) representations. These approaches allow the formal separation of “risk” (known probabilities) from “ambiguity” or “uncertainty” (imprecise or non-unique probabilities). In practice, behavioral economists are often the ones pushing hardest on the claim that, at the level of actual human behavior, “risk ≠ uncertainty”: people react systematically differently when probabilities are themselves obscure or ill-defined, not just risky.

Macroeconomics

In traditional real business cycle (RBC) and New Keynesian DSGE (dynamic stochastic general equilibrium) models, all randomness is typically treated as risk with known probability distributions. Shocks to productivity, preferences, policy, or financial frictions are modeled as stochastic processes (e.g., AR(1) processes with normally distributed innovations), and agents form rational expectations: their subjective beliefs coincide with the true model probabilities. Uncertainty is thus “fully probabilistic” and embedded in the stochastic structure of the model; agents know the distribution of shocks, even if actual realizations are unknown ex ante.

More recent strands of macroeconomics, however, explicitly introduce richer notions of uncertainty. “Uncertainty shocks” refer to periods in which the dispersion, volatility, or perceived ambiguity about future outcomes rises, often modeled via time-varying volatility, stochastic volatility, or changes in the cross-sectional dispersion of shocks. Another strand uses robust control and model uncertainty: policymakers or firms are modeled as fearing that their baseline model may be misspecified, so they behave cautiously or choose policies that are robust to worst-case scenarios. In this context, “uncertainty” can mean either a higher variance of shocks (which is still risk in the classic sense) or deeper doubts about the correctness of the model itself (which is closer to true Knightian uncertainty). Macroeconomists thus use the word “uncertainty” in both a narrow, variance-based sense and in a broader, model-uncertainty sense, depending on the specific framework.

Keynesian / Post-Keynesian / radical uncertainty views

Starting with John Maynard Keynes and later developed by some Post-Keynesian and “radical uncertainty” authors, a different perspective argues that much of the economic world is characterized by fundamental or radical uncertainty. Here, the future is viewed as genuinely non-ergodic: it does not behave like a stable, repeatable statistical process that can be inferred reliably from past data. Many crucial events—wars, financial crises, institutional shifts, technological breakthroughs, or changes in social norms—are seen as unique, path-dependent, and not well-described by known or even well-approximated probability distributions.

In this view, the expected-utility apparatus with well-defined probabilities is not just an approximation but can be actively misleading for analyzing major economic decisions, especially in areas like investment, innovation, and high-level financial or policy choices. Instead of rational optimization under a single known prior, people are thought to rely heavily on conventions, social norms, narratives, “animal spirits,” and simple rules of thumb to navigate an inherently unknowable future. Confidence, sentiment, and storytelling play a central role in driving investment and spending, and instability can emerge endogenously as these conventions and narratives shift. This contrasts sharply with mainstream models that treat all randomness as risk, suggesting that for many important questions, uncertainty is qualitatively different from—and more profound than—probabilistic risk.

Austrian, evolutionary, complexity perspectives

Austrian, evolutionary, and complexity-oriented economists emphasize Knightian uncertainty as central to entrepreneurship, innovation, and economic change. They view the economy as a complex, adaptive, and evolving system in which new technologies, products, institutions, and even entirely new states of the world emerge over time. Because these future states cannot be fully anticipated or listed in advance, it is impossible to assign complete probability distributions over a fixed state space. The standard “state space + probability measure” framework is therefore seen as too static and limited for understanding genuine novelty and creative destruction.

In these perspectives, entrepreneurs are rewarded precisely for bearing non-insurable, fundamentally uncertain outcomes and for discovering opportunities that others have not foreseen. Formal models with multiple priors or non-additive probabilities are less central here; instead, the focus is often on qualitative arguments, historical examples, and computational or agent-based models that highlight open-ended evolution and feedback effects. Conceptually, however, these schools are firmly in the camp that insists “uncertainty is not just risk”: economic dynamics are shaped by unknown unknowns, structural change, and the ongoing creation of new possibilities that cannot be captured by a single, stable probability distribution over a fixed set of states.

Now that we have somewhat surveyed how economists think about risk and uncertainty, let's unpack the process by which economists model decision problems. It more or less follows this general template:

Decision Problems Under Risk in Econonomics

A (single-agent) decision problem under risk is typically described within a formal framework that makes all sources of randomness probabilistic and well specified. In this setting, the economist defines a set of possible choices and a set of possible states of the world, together with a probability distribution over those states and a way to map choices and states into outcomes that the decision maker cares about. Preferences over these risky outcomes are then represented by some utility functional, most commonly expected utility.

  • A set of actions \(A\) (also called policies or choices), which captures what the decision maker can choose. These actions may be discrete (e.g., buy vs. not buy insurance) or continuous (e.g., portfolio shares, effort levels), and they can be thought of as the feasible decisions given technology and institutional rules.
  • A set of states of the world \(S\), which represents all the relevant ways the world might turn out once uncertainty is resolved (e.g., high return vs. low return, good weather vs. bad weather). States are assumed to be mutually exclusive and collectively exhaustive, so that exactly one state occurs.
  • A probability measure \(P\) over \(S\), which is what makes the problem one of risk rather than pure uncertainty. This measure may be interpreted as an objective distribution (e.g., long-run frequencies) or as a subjective probability distribution representing the agent’s beliefs. In either case, \(P\) is assumed to be a single, coherent probability measure defined on all relevant events.
  • A set of consequences/outcomes \(X\), such as consumption bundles, wealth levels, or payoff vectors, that describe what the agent ultimately cares about in each state. Outcomes can be multidimensional (e.g., consumption today and tomorrow, different goods, leisure and labor) and may incorporate both material and non-material aspects of wellbeing.
  • A consequence function \(f : A \times S \to X\), which specifies what outcome you get if you choose action \(a \in A\) and state \(s \in S\) occurs. Formally, for each pair \((a,s)\), the consequence is \(x = f(a,s)\). This function captures how technology, market structure, and constraints translate decisions and states into realized outcomes.
  • A preference relation over random outcomes, usually represented by a utility function and an aggregation rule, most commonly expected utility. In the canonical case, the agent has a von Neumann–Morgenstern utility function \(u : X \to \mathbb{R}\), and evaluates risky prospects (lotteries over \(X\)) by their expected utility with respect to \(P\).

In risk models, the key assumption is the existence of a single, well-defined probability distribution \(P\) over states (objective or subjective). Everything else—optimal choices, comparative statics, welfare analysis, and equilibrium concepts in multi-agent settings—is built on top of that probabilistic structure. The distinction between risk and deeper forms of uncertainty (where probabilities may be ill-defined or non-unique) is deliberately set aside in this framework.

The “standard” modeling pipeline under pure risk

You can think of the standard approach to modeling decisions under risk as a structured checklist that an economist implicitly runs through when translating a verbal description of a problem into a formal model. The pipeline below is written for a single-agent problem, but the same logic extends to multi-agent environments and general equilibrium models.

Step 0 — Verbal problem statement

Example: “A household chooses how to allocate its wealth between a risky asset and a risk-free asset to maximize expected lifetime welfare.”

At this step, the economist formulates the problem in plain language. This involves:

  • Identifying the agents (e.g., household, firm, bank, government).
  • Clarifying the key decisions they make (e.g., how much to save, what portfolio to hold, what effort level to exert).
  • Describing the main constraints they face (budget, technology, institutional rules).
  • Highlighting what is random (prices, returns, shocks) and what is taken as given.
  • Specifying the time scale (one-period vs. multi-period, short run vs. long run) and the general context in which the decision is made.

The goal of Step 0 is to distill the economic intuition and narrative into a form that can then be mapped to a formal model in subsequent steps.

Step 1 — Choose the economic environment

The next step is to choose the basic structure of the economic environment in which the decision is embedded. This requires answering several high-level questions:

  • Unit of analysis? Is the relevant decision maker an individual consumer, a representative household, a firm, a financial intermediary, or a government? This choice affects how we interpret utility, constraints, and objectives.
  • Time structure? Is the problem a one-shot (static) choice, a finite-horizon problem, or an infinite-horizon (dynamic) problem? In dynamic settings, we must also decide on the length of periods and whether decisions are made in discrete or continuous time.
  • Interaction? Is this a single-agent problem taking prices and other variables as given, or a strategic interaction (game) where other agents’ behavior matters and must be modeled explicitly? In multi-agent settings, equilibrium concepts (e.g., Nash or competitive equilibrium) come into play.

For exposition, it is common to start with the simplest case: a single agent with a static decision, taking prices and probabilities as given, and then later extend the framework to dynamic or multi-agent environments.

Step 2 — Specify states, probabilities, and information

Once the environment is chosen, the next step is to formalize uncertainty via states and probabilities, and to clarify what the agent knows at the time of decision.

  • Define a finite or continuous state space \(S\), such as \(\{s_1, s_2, \dots, s_n\}\) for discrete states, or a subset of \(\mathbb{R}^k\) for continuous uncertainty. Examples might include “high return vs. low return,” “good weather vs. bad,” or a continuum of possible productivity shocks.
  • Assign probabilities \(P(s)\) for each state in the discrete case, or a density \(p(s)\) with respect to some reference measure in the continuous case. These probabilities satisfy the usual axioms (non-negativity and summing/integrating to one).
  • Specify what the agent knows at the time of decision. Under risk, the agent is assumed to know (or behave as if they know) the probability measure \(P\). Information can be summarized by an information set or sigma-algebra, but in simple models we often just say “the agent knows the distribution of shocks.”

If probabilities are subjective (beliefs rather than physical frequencies), the standard risk framework still imposes a single coherent subjective probability measure \(P\). The agent may update these beliefs over time via Bayes’ rule as new data arrive, but at each point in time, her uncertainty is summarized by a single prior (or posterior) distribution over states.

Step 3 — Specify actions and constraints

Next, we define what the agent can do and what limits those choices.

  • The action set \(A\) describes the feasible decisions (e.g., portfolio weights in different assets, effort levels, consumption–savings choices, production plans). The action set may be continuous, discrete, or a mixture, and is often assumed to be closed and bounded for technical reasons (to guarantee the existence of optimal choices).
  • Constraints link actions and states to feasible outcomes:
    • Budget constraints (e.g., wealth cannot be negative, spending cannot exceed income).
    • Technological constraints (e.g., production functions that limit output given inputs).
    • Resource or institutional constraints (e.g., borrowing limits, regulatory constraints, participation constraints).

Mathematically, for each action \(a \in A\) and state \(s \in S\), the outcome is \(x = f(a,s)\) and must lie in some feasible set \(X(a,s)\). The function \(f\) together with the constraints embodies the technological and institutional structure of the problem, and determines which lotteries over outcomes are attainable.

Step 4 — Specify preferences under risk (expected utility)

We then specify how the agent evaluates risky prospects. The standard assumption is expected utility under risk.

  • A utility function \(u : X \to \mathbb{R}\), usually:
    • Increasing in “good” things (e.g., more consumption, higher wealth, better health yields higher utility).
    • Concave in core arguments (e.g., consumption), which reflects risk aversion: the agent prefers the expected outcome of a lottery to the lottery itself.
  • A preference functional over risky prospects (lotteries) induced by expected utility. If the agent chooses action \(a\), the random outcome is \(f(a,S)\), and her expected utility is \[ U(a) = \mathbb{E}_P\big[ u(f(a,S)) \big] = \sum_{s \in S} P(s)\, u(f(a,s)) \] in the discrete case, or the corresponding integral in the continuous case: \[ U(a) = \int u(f(a,s)) \, dP(s). \]

Here, the risk attitude of the agent is encoded in the curvature of \(u\); probabilities themselves are typically taken as given and enter linearly inside the expectation. Alternative but related specifications (such as constant relative risk aversion or Epstein–Zin preferences in dynamic settings) still treat risk via a single probability measure but may separate attitudes toward risk, intertemporal substitution, and other dimensions of choice.

Step 5 — Define the optimization problem

Given the primitives above, the agent’s problem is to choose an action that maximizes expected utility subject to constraints:

\[ \max_{a \in A} \; \mathbb{E}_P\big[ u(f(a,S)) \big] \]

subject to the technological, budgetary, and institutional constraints described in Step 3.

For dynamic problems, this optimization is embedded in a dynamic programming framework. A typical setup involves:

  • State variables \(z_t\) (e.g., wealth, capital, productivity, information).
  • Control variables (actions) \(a_t\) chosen at each date.
  • A transition equation \[ z_{t+1} = g(z_t, a_t, \varepsilon_{t+1}), \] where \(\varepsilon_{t+1}\) is a random shock with a known distribution under \(P\).
  • A Bellman equation of the form \[ V(z) = \max_{a \in A(z)} \Big\{ u(x(z,a)) + \beta \, \mathbb{E}_P\big[ V(z') \mid z,a \big] \Big\}, \] where \(0 < \beta < 1\) is the discount factor and \(z' = g(z,a,\varepsilon')\) denotes next period’s state.

Expectations are always taken with respect to a known probability distribution, either objective or subjective, and the agent is assumed to correctly anticipate how the state evolves given her choices and the stochastic environment.

Step 6 — Solve the model: derive optimal policy and comparative statics

The next step is to actually solve the optimization problem, either analytically (when possible) or numerically (which is common in more complex or dynamic models).

  • Find the optimal action \(a^*\) in the static case or the policy function \(a^*(z)\) in dynamic settings, which describes the optimal choice as a function of the state.
  • Derive comparative statics, asking how optimal choices change when exogenous parameters change:
    • How does \(a^*\) change when wealth increases?
    • How do optimal portfolio shares change when risk aversion rises?
    • How do choices respond to changes in probabilities, interest rates, or the distribution of shocks?

These comparative statics provide the main qualitative predictions of the model. For example: “If risk aversion increases, the optimal share of wealth in the risky asset falls,” or “If the probability of a bad state increases, precautionary saving rises.” In dynamic models, we may also analyze stability, convergence to a steady state, or the behavior of the system in response to different shock processes.

Step 7 — Map the model to observables

To use the model empirically or for policy analysis, we must connect its abstract objects to real-world data.

  • Identify what is observable:
    • Choices: portfolio shares, insurance purchases, consumption–savings decisions, labor supply, etc.
    • Outcomes: consumption levels, income, asset returns, default events, realized shocks when they can be measured.
    • Some state variables: wealth, employment status, prices and interest rates, sometimes beliefs (via surveys).
  • Identify what is not observable and must be inferred:
    • Utility parameters: risk aversion coefficients, discount factors, habit parameters.
    • Beliefs: subjective probabilities over states or over future variables.
    • Latent state variables or shocks that are not directly measured but influence behavior.

This step typically involves specifying an observation equation: a statistical relationship that maps from model variables (true actions, states, and shocks) to actual data (which may be measured with noise or only partially observed). This mapping is crucial for estimation, identification, and for comparing the model’s predictions with empirical patterns.

Step 8 — Estimation and empirical verification

Finally, we ask: does the model describe actual behavior under risk? This involves taking the model to data and assessing how well it fits, explains, or predicts observed outcomes.

Several strategies include:

  • Structural estimation: Specify the full model, including functional forms and distributions, and estimate its parameters (e.g., risk aversion, discount factors) by fitting the model to observed choices and outcomes. Methods include maximum likelihood, generalized method of moments (GMM), and the method of simulated moments.
  • Calibration: Choose parameter values so that the model matches key empirical moments (e.g., average consumption growth, volatility of returns, portfolio shares). Calibration is common in macro and finance when full structural estimation is difficult.
  • Experimental and field evidence: Use laboratory experiments, surveys, or field experiments to estimate risk preferences and beliefs directly, and then compare these with the implications of the model. This can reveal systematic deviations from expected utility or from the assumed probability structure.
  • Reduced-form tests and model comparison: Derive testable implications (e.g., Euler equations for consumption, portfolio choice conditions) and check whether they hold in the data. Compare competing models (e.g., different utility specifications or different belief assumptions) by their empirical performance.
  • Out-of-sample prediction and policy evaluation: Assess whether the model can predict behavior in new environments (e.g., after a policy change or under different risk scenarios) and use it to evaluate counterfactual policies.

Through these empirical exercises, economists assess the adequacy of the pure risk framework, identify where it works well, and pinpoint situations where more complex notions of uncertainty, behavioral deviations, or richer institutional details may be needed.

Finally, incorporating risk and uncertainty into models informs the conclusions that can be drawn. Conclusions are often qualified and understood differently in light of uncertainty; it constraints the decision making space of the modeler. 

1. Deterministic vs. Risky/Uncertain Models: What Actually Changes?

In a purely deterministic economic model, everything is known with certainty. Once you fix the parameters and initial conditions, the model delivers a single, precise path for outcomes like consumption, investment, GDP, or employment. Conclusions in that world naturally take the form of sharp statements such as, “If we raise taxes by X, consumption falls by Y,” or “The optimal savings rate is s*.” There is no distinction between what happens on average versus what happens in good or bad states, no notion of volatility or tail events, and no need to think about how spread out the possible futures might be. Optimization in such a setting is simply about choosing the best point on a known path.

Once you incorporate risk and uncertainty, the entire nature of the conclusions changes, because the model now describes distributions of outcomes rather than single trajectories. Instead of saying “consumption will be 100,” the model might say “consumption is 100 on average, but it could be 70 or 140 with certain probabilities.” Economic decisions become trade-offs between expected outcomes and the riskiness of those outcomes: more return vs. more downside risk, higher average growth vs. more volatility, or higher expected welfare vs. worse outcomes in some states or for some groups. Agents now maximize expected utility (or some other risk-sensitive criterion) rather than a deterministic utility function, and policy evaluations hinge on how an intervention shifts not just the mean, but the entire distribution of possible futures. In short, with risk and uncertainty, the key object of interest is no longer a point prediction but the whole probability distribution around it.

2. How Conclusions Look Different Once Risk and Uncertainty Are in the Model

The move from a risk-free to a risky or uncertain model doesn’t just add noise; it fundamentally alters what “optimal” behavior and “good” policy look like. One clear example is precautionary behavior. In a deterministic life-cycle model, a household chooses savings based on time preference and the interest rate alone, leading to a simple plan: “save this much to smooth consumption over time.” When future income and health are risky, however, households face bad potential states and respond by saving more than in the deterministic benchmark to self-insure against shocks. This precautionary saving can be reduced by social insurance (like unemployment benefits or health insurance), which changes the conclusion from “insurance is a distortion” to “insurance can raise welfare by reducing downside risk.” Without risk, the entire notion of saving as self-insurance and the interaction with social policy would be invisible.

Another major shift occurs in asset pricing and risk premia. In a deterministic world, two assets with the same expected return are equally attractive once you account for time, because there is no risk to compensate. With risk and risk aversion, investors demand a risk premium to hold assets whose payoffs are uncertain, especially those that do poorly in bad aggregate states. This leads to central conclusions in finance: assets with higher covariance with bad times must yield higher average returns, and policies or institutions that change the risk environment (such as financial regulation) can alter risk premia, not just levels of returns. Similarly, the presence of risk explains why insurance, diversification, and other risk-sharing institutions exist at all. In a deterministic setting, insurance markets, diversified portfolios, and complex state-contingent contracts would have no role; once risk is present, complete markets allow Pareto-efficient risk-sharing, and incomplete markets or borrowing constraints lead some agents to bear too much risk, affecting their consumption, investment, and labor supply. This underpins conclusions such as “social insurance can raise welfare despite distortions” by providing risk-sharing that private markets fail to deliver.

Risk and uncertainty also transform how economists understand investment and the timing of decisions. In a simple deterministic net-present-value (NPV) framework, any project with positive NPV should be undertaken immediately, and waiting only reduces value. When future demand, costs, or regulation are uncertain and investment is irreversible, there is an option value of waiting to gather more information. Firms optimally delay investment when uncertainty is high, even if expected NPV is positive, and policy conclusions change accordingly: reducing regulatory or macroeconomic uncertainty can stimulate investment simply by lowering the option value of waiting, even if expected profitability is unchanged. Lastly, in macroeconomic policy, stochastic shocks make stabilization policy meaningful. A deterministic model either has no fluctuations or only predictable cycles, making policy about levels and steady states. Introducing shocks and risk aversion shows that reducing volatility and the probability of severe recessions can raise expected welfare. Policies that slightly lower average output but significantly reduce crisis risk may be desirable, and the trade-off becomes one between mean outcomes and volatility or tail risk. None of these conclusions emerge in a stripped-down deterministic world where uncertainty is absent by construction.

3. How Conclusions Are Qualified and Interpreted in Light of Risk and Uncertainty

Once risk and uncertainty are explicitly modeled, economists rarely present conclusions as unconditional and exact. Instead of saying, “Policy X increases employment by 1%,” they say things like, “Given the assumed distribution of shocks, parameters, and model structure, policy X raises the expected level of employment by about 1%, with a confidence interval around that estimate.” This kind of qualification emphasizes that results depend on how shocks are modeled, how risk aversion is specified, and what data the model was calibrated or estimated on. If the stochastic process for shocks is misspecified or if the degree of risk aversion is very different from what the model assumes, then optimal choices and recommended policies may change substantially. In other words, conclusions are conditional on a particular characterization of risk and uncertainty embedded in the model.

Economists also use intervals and robustness checks to express how fragile or stable their conclusions are under uncertainty. Forecasts and policy effects are often presented with confidence or credible intervals, predictive ranges, or fan charts that show not just a central estimate but the range of plausible outcomes. These intervals help assess whether an effect is statistically meaningful and how much uncertainty surrounds it. Beyond that, sensitivity analysis is used to test how conclusions respond to alternative parameter values (such as different degrees of risk aversion or shock variances), different modeling assumptions (for example, alternative frictions or shock processes), or even different decision criteria (like expected utility versus robust control or max–min preferences). If a conclusion only holds under very narrow assumptions about risk and uncertainty, it is treated more cautiously than a result that survives a wide range of plausible scenarios. Risk-aware analysis pushes economists to express results as “under these assumptions and within these ranges, we find X,” rather than as universal, context-free claims.

4. How Risk and Uncertainty Constrain and Inform Actual Decisions

For individuals and firms, explicitly recognizing risk and uncertainty reshapes how they manage their finances and real decisions. Households in uncertain environments choose portfolios that reflect their tolerance for risk instead of simply chasing the highest expected return, and they willingly pay for insurance (such as health, disability, or unemployment coverage) even when expected payouts are lower than premiums, because they value protection in bad states. They also hold precautionary savings and avoid excessively leveraged or fragile positions that could fail under adverse shocks. Firms behave similarly: they hedge with derivatives, diversify across products or markets, maintain cash buffers or credit lines, and often stage or delay large investments when uncertainty about future demand, costs, or regulation is high. None of this behavior makes sense in a fully deterministic model, but it follows naturally once risk and uncertainty are part of the modeling framework and the decision criterion.

For policy-makers and central banks, formal models of risk and uncertainty lead to prudential, robust, and intertemporal trade-offs. Regulatory authorities design capital buffers, liquidity requirements, and stress tests for financial institutions precisely because they want the system to withstand rare but severe adverse states. Central banks consider not just the most likely inflation and output outcomes under different policy rules, but also the distribution of possible paths and the probability of hitting constraints like the zero lower bound. Governments evaluate social insurance, disaster relief, and climate policy by trading off current costs against the reduction in the probability or severity of future crises and catastrophic outcomes. This often motivates robust policies that perform reasonably well across many possible models and parameter values, rather than policies that are optimal under one precise, but possibly misspecified, model. Once uncertainty is explicit, the value of information and flexibility also becomes central: it can be optimal to invest in data, research, or experimentation to reduce uncertainty, and to design decisions that preserve options (for example, reversible or phased policies) rather than locking into a single course of action. In this way, risk and uncertainty directly constrain what is considered acceptable policy and guide both private and public decision-makers toward strategies that balance expected benefits with resilience to adverse scenarios.

5. Summary: How Risk and Uncertainty Reframe Economic Conclusions

Incorporating risk and uncertainty into economic models fundamentally changes both the content of conclusions and the way conclusions are expressed. Instead of focusing only on point predictions or steady states, economists emphasize distributions of outcomes, risk premia, precautionary motives, option values, and the role of volatility and tail events. Concepts like precautionary saving, demand for insurance, asset risk premia, delayed investment due to uncertainty, and welfare-improving stabilization policy all arise directly from the explicit treatment of risk. Without these ingredients, many of the most important real-world behaviors and policy issues simply cannot be captured: why people insure, why firms hedge and hold cash, why financial regulation exists, why macro volatility matters, and why reducing the risk of crises can be more valuable than marginally raising average growth.

At the same time, risk and uncertainty push economists to qualify their claims and to speak in terms of conditional, probabilistic statements. Model-based conclusions become “under these assumptions, here is how policy X shifts the distribution of outcomes,” accompanied by confidence intervals, sensitivity checks, and robustness analyses across alternative parameter values and model specifications. For both private decision-makers and policy-makers, decisions are framed as balancing expected gains against risk exposure, downside protection, and robustness to model error. Thus, the explicit incorporation of risk and uncertainty does not just add technical complexity; it reshapes what “good decisions” and “good policies” mean, making resilience, variance, and tail risks central considerations alongside expected outcomes.



Nash Equilibrium


Elasticity


Time Preference and Discounting (Present Value, Discount Rate)

I made reference to this concept earlier without explicitly describe what it is. The discount rate in economics is a conceptual tool used to determine the present value of future cash flows. It reflects the idea that a dollar today is worth more than a dollar tomorrow due to time, uncertainty, and opportunity cost. At its core, the discount rate is the rate at which future values are adjusted to reflect their value in todays terms. This is usually described in financial contexts as cashflows, but it can be applied to any benefit or cost, including utility. 

The discount rate is important for understanding the time value of money. This refers to the fact that there is normally greater benefit to recieving money now versus an identical sum of money later. The discount rate allows you to quantify the future value of money in present value terms. The time value of money is essentially one factor considered when weighing the opportunity cost of spending rather than saving or investing. This is not to be conflated with interest rates; you can think of it as the justification for the existence of interest rates. Interest rates are implied by the underlying discount rate, which is itself informed by, and a reflection of, peoples intertemporal decision making preferences. The discount rate shows up everywhere. In capital budgeting or cost-benefit analysis, the discount rate is used to compare the present value of expected returns with the initial investment (via Net Present Value, or NPV). A higher discount rate makes future earnings look less valuable. Governments use a discount rate to evaluate infrastructure, environmental policies, or social programs. It helps compare immediate costs with long-term benefits or consequences (e.g., in climate change modeling). Higher discount rates are often used to reflect higher risk. The more uncertain the future cash flow, the more it is discounted.

The discount rate plays a central role in many areas of economic theory, especially in models that involve intertemporal choices — decisions that involve trade-offs across time. Here's how it shows up conceptually and formally across key areas:
  1. Intertemporal Choice Theory (Microeconomics)

    In consumer theory, individuals make choices between consumption today and consumption in the future. The discount rate reflects how much a person values present consumption over future consumption.

    • Utility Function Example: In a two-period model, the utility function might be:
      \[ U = u(C_0) + \frac{1}{1 + \rho}\, u(C_1) \]
    • \( C_0 \): consumption today
    • \( C_1 \): consumption in the future
    • \( \rho \): subjective discount rate (how impatient the person is)

    The higher \( \rho \), the less value is placed on future utility — i.e., more impatience.

  2. Ramsey Growth Model (Macroeconomics)

    This model studies optimal savings and consumption over time in an economy. It includes a social discount rate, reflecting how a planner values future utility compared to present utility.

    \[ U = \int_0^\infty e^{-\rho t} u(c(t)) \, dt \]

    • \( \rho \): pure rate of time preference (discount rate)
    • \( u(c(t)) \): utility from consumption over time

    This determines optimal paths of capital accumulation, saving, and consumption. A higher \( \rho \) leads to more present consumption and less saving.

  3. Cost-Benefit Analysis & Public Economics

    Governments use a social discount rate to evaluate long-term projects (infrastructure, environmental protection, etc.). A high discount rate may make long-term benefits look trivial, which can discourage investments in sustainability or climate action.

    Debate:

    • High discount rate → undervalues future generations
    • Low discount rate → emphasizes intergenerational equity
  4. Environmental and Climate Economics

    The discount rate is crucial in climate models, such as the Stern Review or DICE model.

    • Stern Review used a very low discount rate (around \( 1.4\% \)), emphasizing long-term climate costs.
    • Critics (like Nordhaus) use higher rates (around \( 3\%-5\% \)), leading to more moderate action now.

    Small differences in the discount rate can drastically change climate policy recommendations.

  5. Finance Theory

    In asset pricing and discounted cash flow (DCF) models, the discount rate is used to evaluate the present value of uncertain future returns.

    \[ PV = \sum \frac{E(R_t)}{(1 + r)^t} \]

    • \( r \): discount rate = risk-free rate + risk premium
    • The discount rate affects firm valuation, stock prices, and investment decisions.
The discount rate shapes how we trade off present and future outcomes — whether as individuals, investors, or societies. Governing institutions like the Fed use discount rates to inform their thinking on monetary policy. The Fed must constantly balance stimulating the economy today (e.g. by cutting interest rates) and avoiding long-run inflation, financial instability, or overheating. The Fed "discounts" the future less when it fears long term damage, when it wants to support investment in long-term productivity, and when it values future inflation stability very highly. In the macro models used by central banks (e.g. DSGE models), agents — households, firms, governments — make decisions over time. The models include discount factors to capture time preferences. The Fed uses these models to simulate how changes in the policy rate affect future inflation, output, employment, etc. The representative agent’s discount rate affects how sensitive consumption and investment are to interest rate changes. Fed policy aims to influence real interest rates, which directly affect behavior in these models — and those effects are discounted over time to compute the welfare effects of policy paths. Even though the operational "discount rate" is a tool, the theoretical concept of discounting is built into the models, forecasts, welfare functions, and trade-offs that guide monetary policy decisions at the Fed.

As alluded to, the discount rate is implied by market rates. In simple consumption-savings models, the Euler equation describes the intertemporal condition for optimal consumption: 

\[ 1 = \beta (1 + r) \quad \Rightarrow \quad r = \frac{1}{\beta} - 1 \]

The lower \( \beta \) is (i.e. the more impatient the agent), the higher the interest rate must be to compensate for waiting. People prefer consumption now, to shift that consumption into the future, they must be rewarded. That reward is the interest rate on capital. So, the discount factor determines the "price of waiting", and the interest rate is that price (determined by markets) in real terms. 

In intertemporal choice theory, individuals make decisions involving consumption (or utility) at different points in time. The discount factor, denoted \( \beta \in (0, 1] \), captures how much less they value future utility relative to present utility.

\[U = \sum_{t=0}^{\infty} \beta^t u(c_t)\]

Here:

* \( u(c_t) \): utility from consumption at time ( t )
* \( \beta \): discount factor
* \( 1 - \beta \): degree of impatience

A lower \( \beta \) means the agent values the future less, they’re more impatient. It is a common assumption that people naturally prefer gratification sooner rather than later; this is held axiomatically within economics. If preferences are complete, transitive, stationary, and satisfy continuity and separability, then utility must take the discounted utility form. Koopmans showed this implication follows from standard assumptions in consumer choice. The idea is not that people are irrational, but that waiting has a cost (due to uncertainty, mortality, or biological drive). Below are some common variations you'll see depending on the nature of the cashflows:
Concept Formula Role of Discount Factor ( \( \beta \) )
Basic PV \( PV = \frac{FV}{(1 + r)^t} \) \( \beta = \frac{1}{1 + r} \), so \( PV = FV \cdot \beta^t \)
Basic FV \( FV = PV \cdot (1 + r)^t \) \( FV = \frac{PV}{\beta^t} \)
Multiple Payments \( PV = \sum \beta^t C_t \) Time-weighted sum of cash flows
Perpetuity \( PV = \frac{C}{r} = \frac{C}{1 - \beta} \) Implies \( \beta \to 1 \) as \( r \to 0 \)
Continuous Discounting \( PV = FV \cdot e^{-rt} \) Discount factor = \( e^{-rt} \)



More Reading (Highly Recommended):









Comments

Popular posts from this blog

The Nature of Agnosticism Part 5

MAGA Psychology and an Example of Brain Rot