Experimental economics

• 1. Introduction

  ›

  Experimental economics traces its formal origins to a single credited act. Edward Chamberlin conducted what the field remembers as the first market experiment and the first economic experiment of any kind. That moment planted a question at the center of a new discipline. Do markets, when populated by real people with real money at stake, actually behave the way theory predicts?

  The method built to answer this question is direct in design. Participants receive real cash. The situation they face is structured carefully by the researcher. What follows is recorded and analyzed. Data from those sessions are used to estimate effect sizes, test the validity of economic theories, and illuminate how exchange systems function.

  Vernon Smith's laboratory raised questions that multiplied as the field grew. How much do participants learn from experience, and can that learning be modeled? How much does fairness matter when self-interest is at stake? And what happens when rational analysis alone cannot identify a unique solution?

  In 2002, the Bank of Sweden Prize in Economic Sciences gave the field its clearest institutional recognition. Smith received the prize alongside Daniel Kahneman. He was honored for establishing laboratory experiments as a tool in empirical economic analysis, and for his work on alternative market mechanisms in particular.

• 2. Markets in the Laboratory

  ›

  Vernon Smith drew on Chamberlin's work but modified it in important ways. He wanted to test whether prices and quantities would converge on their theoretical competitive equilibrium values.

  Buyers and sellers were the central figures in Smith's laboratory markets. Each participant was told how much they valued a fictitious commodity. They then competitively bid or asked on that commodity following the rules of real-world market institutions: the Double auction, the English auction, and the Dutch auction.

  What Smith found was that in some forms of centralized trading, prices and quantities did converge on their theoretical values. This held even though the conditions of his markets fell well short of what perfect competition requires. Large numbers of participants and perfect information, two classic assumptions of the theory, were both absent.

  In the 1970s, Charles Plott of the California Institute of Technology joined Smith in developing the experimental approach. Plott extended the work into political science and used laboratory findings to inform economic design and policy.

  Experimental finance built on this foundation. Researchers have designed experiments that manipulate information asymmetry about the holding value of a bond or a share, specifically to study the conditions under which stock market bubbles form. Today, researchers in this area use simulation software to examine trading flows, information diffusion, and price-setting processes across different market environments.

• 3. How Subjects Learn

  ›

  Until the 1990s, simple adaptive models were the standard tools for studying behavioral adjustment in experimental subjects. Cournot competition and fictitious play were the two most common frameworks. Both assumed a kind of mechanical rule-following rather than a richer account of how participants actually adapt.

  Alvin E. Roth and Ido Erev changed that picture in the mid-1990s. They demonstrated that reinforcement learning can generate useful predictions in experimental games. Reinforcement learning describes the tendency of subjects to repeat decisions that earned high payoffs in the past.

  In 1999, Colin Camerer and Teck-Hua Ho introduced a richer framework called Experience Weighted Attraction, or EWA. The model incorporated reinforcement learning alongside a second process called belief learning. In belief learning, subjects form ongoing estimates of what other players are doing and update those estimates as the game progresses. EWA also established that fictitious play is mathematically equivalent to generalized reinforcement, provided appropriate weights are placed on past history.

  Overfitting was the first concern critics raised about EWA. The model uses enough parameters that it risks fitting data too closely to be genuinely predictive. A second concern was that EWA might lack generality across different types of games. A third was that its parameters are difficult to interpret. Researchers addressed overfitting by testing EWA's predictions on experimental data the model had not been trained on. They addressed generality by replacing fixed parameters with self-tuning functions that shift over the course of a game and differ across game types.

  Roberto Weber has investigated learning in settings where participants receive no feedback. David Cooper and John Kagel have examined how subjects learn over similar strategies. Ido Erev and Greg Barron have studied learning in cognitive strategies. Dale Stahl has characterized how participants learn over decision-making rules. Charles A. Holt has examined logit learning across different kinds of games, including games with multiple equilibria. Wilfred Amaldoss has applied EWA to questions in marketing. Amnon Rapoport, Jim Parco, and Ryan Murphy applied reinforcement-based adaptive learning models to the centipede game, one of game theory's most celebrated paradoxes.

• 4. Fairness and Self-Interest

  ›

  The ultimatum game, the dictator game, and the trust game form the core toolkit for studying social preferences. The gift-exchange game and the public goods game round out the set.

  Altruism, spitefulness, tastes for equality, and tastes for reciprocity are the four dimensions that social preferences encompass. The term itself refers to the degree to which people care, or do not care, about each other's well-being.

  Ultimatum game experiments have produced one of the field's most consistent results. When participants are offered low monetary allocations, many choose to reject the offer entirely. They forgo any payment rather than accept a split they consider unfair. This behavior conflicts directly with simple models of self-interest, which predict that any positive offer should be accepted.

  Cultures differ in how much monetary sacrifice their members will make in the name of fairness. Economic experiments have measured this variation across societies. Researchers have continued testing modifications of these canonical game settings to isolate specific aspects of social preference in finer detail.

• 5. The Coordination Problem

  ›

  Coordination games are games with multiple pure strategy Nash equilibria. When no single equilibrium is obviously superior, rational analysis alone cannot guide players to a specific outcome.

  Selecting among Nash equilibria is the core challenge in coordination experiments. The first question asks whether laboratory subjects can coordinate at all on one of the available equilibria. It also asks whether general principles can predict which equilibrium they will reach. The second question asks whether subjects can reach the Pareto best equilibrium, the outcome best for all participants, and what conditions or mechanisms might help them do so.

  Deductive selection principles derive predictions from the properties of the game itself. Inductive selection principles rely on characterizations of how behavior evolves dynamically. Together they represent the two main frameworks researchers use to predict coordination outcomes.

  Pareto-best equilibria, even of the complex, non-obvious, and asymmetric variety, have been reached by groups of experimental subjects. This has happened even when all subjects decided simultaneously and independently, without any communication. How groups reliably converge on these outcomes remains an open question.

• 6. Contracts and the Field

  ›

  Contract theory is concerned with how incentives function when some relevant variables cannot be observed by all parties to an agreement. Testing such theory in the field presents a specific difficulty.

  Laboratory settings solve a problem that field research cannot easily handle. If a researcher can observe and verify the variables that matter for a contract, those variables cease to be truly unobservable. That shift eliminates the very contractual problem that theory aims to explain. The laboratory bypasses this by giving the experimenter direct control over who sees what.

  Moral hazard theory is among the contract-theoretic models researchers have tested this way. Adverse selection theory, exclusive contracting, deferred compensation, the hold-up problem, and both flexible and rigid contract designs have all been examined. Researchers have also tested models with endogenous information structures.

  John A. List pioneered field experiments in economics in the early 1990s. He carried the controlled logic of the laboratory into real-world settings where behavior occurs naturally. The same experimental approach has since been extended to institutions and to legal questions, forming a branch known as experimental law and economics.

• 7. Computational Modeling

  ›

  Agent-based computational modeling is a relatively recent addition to the experimental economics toolkit. Its central participants are not human subjects. They are computational objects designed to interact according to specified rules, and they can represent social or physical entities alike.

  ACE models, as they are known, treat economic processes as dynamic systems of interacting agents. The approach draws on the complex adaptive systems paradigm. Starting from initial conditions the modeler specifies, an ACE model develops forward through time driven entirely by the interactions among agents.

  Empirical validation presents one of the specific challenges that agent-based models face. Some difficulties mirror concerns familiar from experimental economics generally. Others belong to agent-based modeling alone. These include developing a shared framework for empirical validation and resolving the open questions the approach has raised within its own literature.

• 8. Software and Method

  ›

  Urs Fischbacher began developing z-Tree in 1998. Its full name is the Zurich Toolbox for Readymade Economic Experiments. It has become the most widely used software platform for conducting experimental economics research in laboratory settings.

  By February 2020, z-Tree had accumulated roughly 9,460 citation results on Google Scholar. In December 2016, Fischbacher received the Joachim Herz Research prize for best research work, an award recognizing z-Tree's contribution to experimental science.

  The platform runs on a network of computers in a research laboratory. One machine is used by the experimenter; the others are used by subjects. The experimenter designs each session using z-Tree's own programming language. This imperative language allows for a wide variety of experiment types and supplementary surveys.

  A growing list of alternatives has appeared alongside z-Tree. FactorWiz appeared in 2000, Wextor in 2002, EconPort in 2005, oTree and ConG in 2014, and nodeGame in 2016.

  Real monetary payoffs are the first requirement in the field's methodological guidelines. Full experimental instructions must also be published. Deception is prohibited. Experimenters are advised to avoid introducing specific, concrete context into their designs. These guidelines emerged largely in response to two persistent critiques. The first charges experiments with lacking internal validity. The second argues they lack external validity, meaning findings may not generalize to many types of real-world economic behavior. Defenders of the method note that neither critique is unique to experimental economics: both apply equally to theoretical approaches and to purely empirical work. Twelve named platforms now appear in the field's documentation of experimental software, from z-Tree in 1998 to nodeGame in 2016.