More about experiments

For an opinion piece on the use of experiments to study asset pricing was published in Foundations and Trends in Finance; see: BOSSAERTS, Peter. The experimental study of asset pricing theory.Foundations and Trends in Finance, 2009, 3.4: 289-361. The article is based on a series of lectures Peter Bossaerts gave at the Austrian Central Bank in June of 2009, which themselves were distilled from lecture notes the author had been using for Ph.D. classes in asset pricing at the California Institute of Technology (Caltech) and the Ecole Polytechnique Fédérale Lausanne (EPFL).

Here are the components of the article.

Experimentation in asset pricing is both rare and novel, so the article starts with justification. The goal of experimentation is twofold. First, experimentation is meant to evaluate the science behind asset pricing theory. Indeed, complete scientific validation at some point does require experimentation, otherwise, to quote the late Hannes Alfven (Nobel prize physics), we are "likely to go completely astray into imaginary conjecture.” The article gives a pointed example of this in Section 2, not in physics, but in the domain of asset pricing. It is interesting to note that Alfven was an astrophysicist, and astrophysicists, like finance scholars, get most of their data from the field. But unlike the latter, astrophysicists do insist on verifying the principles behind their attempts to interpret field data using laboratory experiments, even if the latter are only miniscule compared to real-world (or should one say real-universe?) phenomena. Like astrophysicists, finance scholars interpret field data through the lens of particular theory (e.g., competitive equilibrium), without being able to verify independently that this lens is appropriate. Only laboratory experiments can give us the confidence that our inference is correct. The second goal of experimentation is to come to a deeper understanding of asset pricing theory. Our own experience with experiments confirms that one cannot appreciate all the ramifications (and the beauty) of asset pricing theory without thinking through how one could generate it in the laboratory. To paraphrase the late Richard Feynman (another Nobel prize in physics): "One cannot understand theory if one cannot create it.” Indeed, our lab’s attempts to “create” the Lucas equilibrium (Lucas, 1978) in the laboratory led us to realize that some claims in the literature about it (Judd et al., 2003) were misguided (see a comment in Finance Research Letters by Bossaerts and Zame, 2006). While others (Berrada et al., 2007) evidently did not need experiments to realize the same, in our experience, few students of asset pricing theory appreciate the true nature of our asset pricing theory. For instance, equivalence results like the claim that static, complete-markets allocations can be implemented in incomplete markets by allowing re-trading (Duffie and Huang, 1985), are really not “equivalent,” because vastly different notions of competitive equilibrium are being compared. (This very example is discussed in more detail in the article.)

To set the stage, the article introduces a sample data set that looks very much like the typical data set finance scholars gather from the field; only, it was actually generated in the laboratory. These data are first interpreted as one would do with field data, only to realize that our interpretation is completely false. We know this because we know how the data were generated in the lab…

Subsequently, the article discusses risk aversion. Asset pricing theory builds on risk aversion. Without risk aversion, models like the CAPM or Lucas’ dynamic asset pricing model are vacuous. The issue is: will there be enough risk aversion in the laboratory given typical levels of compensation? And if so, how do we interpret this in view of Rabin’s theoretical claim (Rabin, 2000) that we should not see risk aversion, or, to put it differently, that it cannot be risk aversion?

The next section of the article is equally foundational. Asset pricing theory builds on competitive equilibrium. Competitive equilibrium is an abstract notion; it does not tell us how we can get there. How, then, will we get to it in the laboratory? Here, we will build on the path-breaking work of Vernon Smith and Charlie Plott who demonstrated that certain trading institutions indeed allow us to generate competitive equilibrium [while prior attempts, with different institutions, had failed (Chamberlin, 1948)]. As it turns out, the successful trading institutions are complex, and require powerful software that only recently has become available. (It will be interesting to note that many real-world financial markets have been adopting the very trading mechanisms that we have been using in the laboratory.)

After that, the article discusses the main findings (as of 2009!), first concerning simple static asset pricing models, moving on to dynamic pricing theory, and the implications of ambiguity aversion. Asset pricing theory rarely discusses how markets reach equilibrium (although market microstructure purports to shed light on it), but, as we explain, experiments shed new light on price behavior during equilibration, as well as on off-equilibrium allocation dynamics. Information aggregation is also dealt with (although we prefer to call it information “amplification” for a reason that will be clear when reading the article), and, lastly, markets for loan and insurance contracts, where adverse selection may preclude equilibration, and even when not, the resulting allocations may be Pareto sub-optimal.