Experiments with Financial Markets: Implications for Asset Pricing Theory
Bossaerts, Peter, American Economist
This article surveys financial markets experiments from a particular vantage point, namely, asset pricing theory. The goal is to assess to what extent these experiments have (and could) shed light on the validity of the basic principles of asset pricing theory, namely (i) that markets equilibrate to the point that expected returns are proportional to covariance with aggregate risk, (ii) that markets aggregate dispersed information. There appears to be solid support for (i), yet the evidence regarding (ii) is mixed. The reason for the latter is hard to determine, because of features in the experimental design that are at odds with standard asset pricing theory (e.g., the payoff on a security depends on the identity of the holder). Where we can interpret the results, the article demonstrates that the occasional aggregation failures ("mirages") agree with rational learning. More specifically, their number is consistent with the mistakes one expects a Bayesian to make even when she has full knowledge of the likel ihood function (of any signals conditional on the value of the state variable that she is learning).
This paper surveys experiments of financial markets that were designed with the competitive paradigm in mind. The results will be analyzed from a particular theoretical angle, namely, asset pricing theory. That is, we discuss to what extent a given financial markets experiment can shed light on the validity of asset pricing theory.
Modem asset pricing theory has strong roots in economics and probability theory. Its models are logically compelling, and the derivations elegant. Many models are widely used in industry and government, in applications of capital budgeting, industry rate regulation, performance evaluation, etc. Yet, there is surprising little evidence in support of the theory, and what has come forth is controversial. But tests of asset pricing models have almost exclusively been based on econometric analysis of historical data from naturally occurring markets. That type of empirical analysis is very difficult, because many auxiliary assumptions (homogeneous, correct ex-ante beliefs, stationarity, unbiased samples, etc.) have to be added to the theory for it to become testable.
Experimentation would provide an alternative way to verify the principles of asset pricing theory, because many auxiliary assumptions are under the control of the experimenter. That is, experimentation provides one way to gauge the validity of what would otherwise remain mere elegant mathematics. This article reports on what has been accomplished so far.
Not all experiments on financial markets were designed with the idea that they should verify theoretical principles. Often, the link with the theory is vague. Sometimes, the outcomes of loosely designed experiments were ambiguous, and, because of the absence of a solid theoretical foundation, difficult to interpret. As it turns out, this will include some widely cited experiments, and, consequently, our analysis will be provocative. But that is meant to generate renewed interest in experimentation with financial markets. Indeed, after much activity in the eighties, interest in financial markets experiments disappeared almost entirely. Only recently have experiments re-appeared, and the successes may be attributed to their solid asset pricing theoretic foundation.
Asset pricing theory studies the pricing and allocation of risk in competitive financial markets. (Although it could widen its scope to other mechanisms, the competitive market is studied almost exclusively.) At the outset is the presumption that there are risk averse agents who invest, to smooth consumption over time, and to diversify risk. The latter implies that portfolio analysis (allocation of wealth across several securities) take a core position.
Asset pricing theory also studies the ability of competitive financial markets to aggregate diverse information about uncertain future events. …