Building Agent-Based Models: A New Lens on Economic Reality

In the complex world of economics, traditional models often fall short of capturing the intricacies of human behavior and market dynamics. This talk introduces the power of Agent-Based Models (ABM) as a revolutionary approach to economic modeling, shifting the focus from oversimplified, equilibrium-driven theories to dynamic, agent-based simulations that mirror real-world complexities. ABM allows us to explore alternative scenarios, understand causal relationships, and generate deeper insights into economic outcomes. By examining housing markets and other real-world examples, this talk will demonstrate how ABM can provide a more realistic understanding of economic phenomena and open new pathways for research and policy-making.

1. Building ABM Models to Understand Economics

2. The central battleground is Knowledge Extreme Misunderstandings exist of the nature of models and modeling. Before learning to build good ABM models, we must go back one step to understand what models are, and how they help us to understand the world.

3. Very Brief Historical Review Centuries of Religious Battles led to rejection of scholastic school of thought, which built social theory on theological foundations Reject the Bible led to the reconsideration of Epistemology: how can we know what is true and what is not? Having been deceived by faith, the Enlightenment philosophers vowed to build knowledge on observations and reason. The rejection of the unseen continues to be a fundamental principle of modern epistemology. This leads to vast misunderstandings.

4. Rejection of Unseen => Positivist Models Reality Nominalist Model Attempt to match observables Realist Model Attempt to match unobservables Match ? Match? Model Model

5. Nominalist Versus Realist Models The complex and unknown reality generates observable data. A model is a GUESS at this hidden reality. The model also generates data. A nominalist model is successful if it generates data similar to the observed actual data. A realist model is successful if the structure of hidden reality matches the guess about the hidden reality represented by the model. This success can almost NEVER be verified directly. Indirect verifications of many different kinds are possible.

6. Illustration: Macroeconomic Models Observables = Data on C, I, Y and other macro data. Unobservables = Causal Relationships among these data. Standard Pre-Keynesian Supply Side Model: K, L => Y = F(K,L) => C= cY , I=(1-c)Y => GI = gI and PI= (1-g)I Government Investment will crowd out Private Investment. Keynesian policy of increasing GI will not increase production. Keynesian Demand Side Model has a different, more complex, set of causal relationships.

7. Keynesian Model Y*=F(K,L*), where L* ≤ L. Unemployment is possible. Y* is the anticipated (expected) demand that producers think will obtain next period. Anticipated Demand = C E + I E = Y* Simplifying Assumption: Consumption demands are predictable. C E = c Y* => c Y*+ I E = Y* => Y* = I E / (1-c) If producers anticipate larger demand for investment, then they will produce more, until they reach maximum possible output at full employment. Lower levels of investment demand will lead to unemployment. Government policy can create full employment

8. What is a Model? A model is a map of the underlying reality which produces the observables (data). A model ALWAYS creates the possibility of exploring alternative universes – by making plausible changes in the underlying unobservables, one gets corresponding changes in the observables. Think about causal structure. For supply side, the only way to change GNP is to change K,L. For demand side, the root causal factor is producer expectations about demand.

9. Demand Side & Supply Side Models differ in terms of causal relationships Causal relationships are NOT directly observable. Nominalist modeling cannot capture these. Therefore, standard modeling procedure in Econometrics allows you to assume any causal relationship you want. The only requirement on the model is fitting the data well. Because NEITHER party (demand siders or supply siders) talk about causal relationships, the entire debate is extremely confused and confusing.

10. Success Criteria for Nominalist Models Searching for Patterns in the Data (Finding the Model). Prediction: How well the model can predict on data sets not used to construct/fit the model. Explanation: Does the model “explain” the data? Goodness of Fit.

11. Success Criteria for Realist Models Search for Causal Relations (Hidden Entities or Powers) Abductive Inference: Best Explanation for observed patterns. Causal Explanation, or Understanding how the deeper reality produced the observed patterns (and what else was possible).

12. ABM Models are Realist Models Conventional Economics uses Nominalist

13. Goal in Building ABM models is to arrive at understanding of reality Conventional economic models lead to seriously distorted understanding of reality, because equilibrium and optimization are not characteristics of the real world Beauty of ABM is that they allow us to build realistic simulations of real world situations. The goal is to arrive at understanding. Therefore, we always start with simplest models with minimal number of factors, and play with the model to understand how it works. After achieving this intuitive understand, we proceed to complexify the model.

14. We will explore the Supply and Demand model of Manikiw in some detail

15. What happens in the Real World? Price Students Owners 1 10 101 9 1 201 8 2 301 7 3 401 6 4 501 5 5 601 4 6 701 3 7 801 2 8 901 1 9 1001 10

16. Theory: Equilibrium Price between (500,600) What happens in the real world which leads to this price? How do Agents behave? What are the rules of the game? (Micro-structure) That is, what information do the agents have? What information do the Owners have? How do they make decisions? How is trading organized? What is the legal framework for enforcement of commitments?

17. Walrasian Auctioneer Story Imagine there is a student housing office. All trades must take place in this SHO. Owners list their housing and rental demand, and Students list their requirements and housing budget. SHO Manager announce a housing price: $350 : Students 1,2,3 cannot afford this price. 7 Students raise their hands. Owners 1,2,3 raise their hands. Mismatch: Demand Exceeds Supply. SHO Manager adjusts price until Demand matches Supply. This is the Walrasian Auction Micro-Structure. But it is highly implausible. Does not correspond to how the real world operates.

18. ABM is about creating plausible models, which picture reality. What is wrong with the Walrasian story? Real World Housing Markets do not display equilibrium price – that is, identical houses may have different prices, for a number of reasons. Walrasian auctioneer asks the question: If there is an equilibrium price – that is, if all houses have the same rental price – then what should it be? This is the WRONG question to ask about the Manikiw rental market. It assumes an ideological stance, assumes it is true, and asks how to FORCE it to happen.

19. Justification for Walrasian Auctioneer Story Price $350 has been announced. 7 Students are scrambling to find houses, while only 3 Owners are willing to offer houses. Competition will take place among students. Seeing that only three houses are available, some of the students will offer a higher price. Suppose one student offers 450. Now one additional Owner will enter market. Critical Implausible Element of Story: Re-contracting. Once one student negotiates with new entrant to rent at 450, ALL must raise prices to 450 – there is only ONE market price. This does not match reality. Auctioneer achieves this artificially, by making it a rule.

20. Finding Better Stories to tell about the world Look at what actually happens, and try to model it. This is impossible via conventional economic models. This is precisely what Agent Based Models are designed to do. Look at ANY economic model, and you can re-do it using ABM, and achieve much deeper insights into reality. The goal of modeling is to UNDERSTAND the real world outcomes, after specifying micro-structure and agent behavior. Both micro-structure and agent behavior come from our knowledge of the real world.

21. What drives the story that equilibrium is 550? The central element is that there can only be one rental price for all of the houses. Once this rule is accepted, then 550 is actually a plausible outcome, which matches the demand and supply of houses. Modeling is ALWAYS about imagining alternative realities: What else might happen (which did not happen) but was a possible outcome according to the rules which reality follows (our model). The S&D has a very limited set of alternate realities – all equilibrium prices from 0 to 1000 are possible alternatives. All are worse, in providing less students with housing.

22. This is an Economist-Oriented Exposition Non-economists would find the assumption of equilibrium price puzzling. It is not what we see in the real-world housing market. QUESTIONS to explore via model: What kind of assumptions about agent-behavior and micro-structure lead to (roughly) equilibrium price outcomes as predicted by Supply and Demand theory? Is this a GOOD outcome? Are there better outcomes? What kind of micro-structures and agent behaviors would lead to better outcomes? What happens in ABM models matched most closely to real world agent behaviors and micro-structures?

23. Walk-Around With Zero Information Model FIRST ROUND: 10 Students Randomly Walk around the neighborhood, and each one arrives at ONE of the 10 houses. If Owner Demand is greater than Student Budget, then No Deal. If Owner Demand is less than or equal to Student Budget then we have two possibilities: Posted Prices: Deal Takes place at Owner’s posted price. Negotiations: Both sides negotiate, arriving at midpoint. Second Round starts with unmatched Owners and Students.

24. I will use worksheet to demonstrate many possibilities We will see how different assumptions about micro structure lead to different outcomes. Three types of demonstrations, listed below, will be shown on worksheets, to illustrate how ABM thinking works. Note that the simplest ABM models are simulations which do not require any programming. This is my recommended way to build an ABM model. Build simple models and simulate by hand to develop understanding. With this understanding you can go on to build more complex models using NetLogo or Python.

25. Demonstration 1 What assumptions are needed to get to equilibrium in one-shot game? Full Information, Zero Transaction Cost, and Re-Contracting What assumptions are needed about behavior? Homo Economics: Cold, Callous, and Calculating: No Honor – will break his word if he can earn a little more by doing so. Without recontracting , no decentralized mechanism can get to equilibrium in one-shot. There is no sense in repeated games either.

26. Demonstration 2: What is a good welfare measure of the outcome? Consumer Surplus? It does not take into account the results of those left out of the market. Considering failure to match as an adverse outcome, the matching equilibrium is far superior to the market outcome. Many kinds of micro-structure can lead to matching outcomes. Many kinds of behavioral assumptions can also lead to matching outcomes. Understanding these micro-structures and behaviors gives us deep insights into the nature of rental markets.

27. Demonstration 3: Effects of Transaction Costs Given time to gather information, examine houses, negotiate, one can fail to find opportunities even if they exist. Famous experiments by David Card show that employment increases when minimum wage is increased This effect can be demonstrated via ABM. Setting a minimum wage creates an information effect where searching for best price is not necessary, because it is known in advance. In these cases, matching finishes quickly, and no one fails to find match due to transaction costs.