Testing CAPM in Real Markets: Implications from Experiments

Tests of the CAPM, the prototype model of equilibrium in financial markets, are usually based on returns computed from end-of-month closing prices. It is reasonable to doubt that these prices always reflect markets that are at equilibrium, thus raising the question whether and how inference is biased. Rather than exploring this issue using one of the many theoretical (but empirically unverified) equilibration models that have been suggested in the literature, this paper starts from an empirical analysis of equilibration in experimental competitive markets. Price and allocation dynamics reveal systematic patterns which can be explained in terms of an equilibration model that builds on a simple and plausible behavioral premise, namely, local optimization. To get a comprehensive understanding of the dynamics of prices and allocations in disequilibrium, an extreme version of the equilibration model is studied, whereby prices move much faster than trades, so that all transactions occur at local equilibrium. A portfolio is identified that remains on the mean-variance efficient frontier throughout the equilibration process, namely, the risk-aversion-weighted endowment (RAWE) portfolio. The RAWE and the market portfolio are closely related (the two eventually converge), so that the latter may stay close to the frontier as well. The experimental data confirm this as well as another result: the off-equilibrium equity premium may vastly over-estimate investors’ risk aversion. Violations of portfolio separation (observed both in experiments and field data) are a natural consequence when equilibration dynamics stop short of (global) equilibrium because further gains from trade are insufficient. Explicit formulation of off-equilibrium allocation dynamics reveals in what direction the RAWE portfolio differs from the market portfolio: if payoff covariances are non-negative and everyone starts out with the market portfolio, then RAWE overweighs securities with low payoff variance. This implies, among other things, that the market portfolio will be out-performed when it is combined with a portfolio that is long in low-volatility and short in high-volatility securities.