Division of the Humanities and Social Sciences California Institute of Technology Pasadena. California 91125 Price-conveyed Information vs. Observed Insider Behavior a Note on Rational Expectations Convergence

The recent experimental results of Plott and Sunder (1982) and Friedman, Harrison and Salmon (1983) on the ability of single commodity markets to "reveal" the underlying state to initially uninformed traders were potentially influenced by a design in which the set of informed traders was held constant throughout the life of the market. Hence the performance of uninformed traders in the market might have been predicated on their knowledge of, and the observed behavior of, the informed traders. The experiment discussed below is a replication of one market in Plott and Sunder (1982), with the added feature that the traders who were to be informed of the state differed fr. om period to period. The results are equivalent to those of Plott and Sunder (1982) in the price dynamics, while less conclusive regarding the acquisition and use of the state-price correspondence by uninformed traders. f'fi'i\iffMCf'JT OF LCOf\JOi\11 Ui\Jlvrn::-;!Tj' OF IVill\11\JESOTN PRICE-CONVEYED INFORMATION VS. OBSERVED INSIDER BEHAVIOR: A NOTE ON RATIONAL EXPECTATIONS CONVERGENCE * Jeffrey S. Banks California Institute of Technology 1 . INTRODUCTION In a recent paper on behavior in experimental securities markets, Plott and Sunder (1982) concluded that the rational expectations (RE) model was superior to the traditional prior information (PI) model in predicting equilibrium prices and holdings. In particular, given a market with one commodity, three possible states of the world, and three groups of trader "types" each with differing valuations on the commodity per state, initially uninformed traders were able to infer the underlying state from the current market price and act accordingly. In a related paper, Friedman, Harrison and Salmon (1983) observed that, given the existence of a futures market, the RE model outperformed the PI model in multiperiod single commodity markets as well. One source of potential misinterpretation, however, comes from the fact that, although one half of the traders of each type were informed of the true state at the beginning of each period in Plott and Sunder (1982) and one third of the traders of each type in Friedman, Harrison and Salmon (1983), the same traders were informed in every period. In a more recent paper, Plott and Sunder (1983) constructed markets where all traders received partial information (i.e. given possible states X, Y, and Z, a 2 trader's private information would be either "not X" or "not Y" if the state were Z), yet the combinations of traders receiving a certain message in any period was determined randomly. Hence no trader could be identified as having some constant bit of information. In the single commodity markets, Plott and Sunder (1983) concluded that the RE model did not perform as well as the PI model in their respective predictions. Thus, given the fixed information structure of Plott and Sunder (1982) and Friedman, Harrison and Salmon (1983), one cannot ignore the possibility that, in these experiments, initially uninformed traders guessed the identities of the informed traders and then used their observed behavior to gain knowledge of the state, as opposed to gaining knowledge from the market bids, offers, and prices. In the experiment discussed below, an attempt was made to overcome this problem. The experiment consisted of replicating market 5 of Plott and Sunder (1982), altering only the information structure so that each trader was informed of the state in at least two periods (out of a possible ten) and then randomizing over which periods a trader was informed. This structure was designed to limit the ability of uninformed traders to infer the identities of the informed traders, since in each period different combinations of traders were informed. Note that a single replication, with equivalent empirical results, / will be sufficient to conclude that holding the informed traders constant did not generate the results reported in Plott and Sunder (1982) and Friedman, Harrison and Salmon (1983).