ABSTRACT
The earliest experimental studies of multi-period assets are those of Forsythe et al. (1982, 1984) and Friedman et al. (1984). These studies describe the behavior of experimental markets for assets with a life of two and three periods (a period is defined as a unit of time between two dividend payments). Though these horizons are short, traders do face a situation where they have incentives to arbitrage intertemporally and to form expectations about prices in future periods. After a number of replications of two-or three-period asset markets, prices in the last period converge to approximately the rational expectation equilibrium level. However, convergence is slower and less reliable for period prices, the longer the period precedes the final one. Rational expectation equilibrium prices cannot be discovered until the price for the last period stabilizes, and the price discovery process unravels backward. The presence of futures markets aids and accelerates convergence to rational expectations equilibrium (Forsythe et al., 1984; Friedman et al., 1984). However, markets for longer-lived assets have a strong tendency to generate price “bubbles.” This result is originally due to Smith et al. (1988), but it has been widely replicated and shown to be robust to numerous modifications of the experimental design. In the original design of Smith et al., markets are created for assets with a life of a finite number of periods (usually 15 or 30 periods). The asset pays a dividend in each period, which (other than in a few sessions where there is a final fixed terminal value for the asset) is the only source of intrinsic value. The dividend payment is identical for all traders and the distribution of dividends is common knowledge to all traders. The time-series of transaction prices in markets with this structure does not track the fundamental value, but rather is characterized by price bubbles and crashes. A bubble is an extended period of time, during which prices are much higher than fundamental values, while a market crash is a sudden and rapid fall in prices. Some factors are known to mitigate bubbles in experimental markets. Haruvy and Noussair (2006) have shown that allowing short-selling reduces prices, but if the short-sale constraints are too loose, prices are below fundamental values. They also replicate an earlier result by Caginalp and Smith (1998), who show that the more liquidity traders have available to make purchases, given the fundamental value and the total stock of the asset, the higher the price level relative to fundamentals. If short-sale constraints are set at an appropriate level, which may or may not be plausible given the information available to a market designer, the possibility of making short-sales can push prices toward fundamentals (Ackert et al., 2006). Noussair and Tucker (2006) show that the addition of a sufficient number of derivative futures markets has a strong tendency to reduce, and indeed
often completely eliminate, spot market bubbles and crashes. Haruvy et al. (2007) show that when individuals are required to predict (privately) the prices that will occur in the future, individuals do not anticipate bubble and crashes, but rather tend to extrapolate previous trends. Thus, crashes are typically a surprise to market participants. However, all of the studies that examine changes in asset market behavior as traders acquire more experience find that market prices are closer to fundamentals, the more experience traders have in the same environment. In addition to the fact that they typically trade long-lived assets, prediction markets have the feature that some traders have better information about the value of the asset than others. One function of a market, and indeed the primary purpose of a prediction market, is to reveal this information with the market price. Several early experimental studies indicate that markets have a strong tendency to disseminate private information, provided that enough individuals hold the information. For assets that have a life of only one period, and have a common though uncertain value, Plott and Sunder (1982) observe that when insiders who know the true value of the asset are present, prices in continuous double-auctions reveal the insider information. This result shows that there exist conditions where it is possible to use a decentralized market to disseminate privately held information. Later studies illustrate the limits of the ability of markets to reveal information. Plott and Sunder (1988) study the issue of whether markets can aggregate privately held information. They endow insiders with a portion of – but not all of – the information needed to determine the true value of the asset. Only the aggregation of all of the information held by insiders would allow the state of nature, and therefore the fundamental value of the asset, to be deduced with certainty. The results on information aggregation are mixed. In a setting in which there are markets for contingent claims and in which dividends differ between agents, prices tend toward the level corresponding to rational expectations. However, when only one security is exchanged, prices do not correctly reflect the available information. Forsythe and Lundholm (1990) show that, for the same environment, sufficient trader experience, in conjunction with common knowledge of payoffs, enables the market to reliably aggregate and reveal the inside information. Another observation emerging from early experimental research is that markets may price as if they reveal information that is not actually held by any traders. If the presence of insiders is uncertain, market activity can lead to convergence of prices to levels that are consistent with the presence of insiders, even when no insiders actually exist. This occasional failure of markets to reveal the absence of information is termed an “information mirage” (Camerer and Weigelt, 1991). Individuals may trade on the basis of inferences they make from the trades of others, creating price paths that falsely reveal information that traders do not actually have. Camerer and Weigelt usefully distinguish between mirages, which are caused by uncertainty about the information of others, and bubbles, which appear to be caused by uncertainty about the rationality of others. The experimental literature on prediction markets can be classified as belonging to one of two branches. One branch, discussed in Section 3, focuses on the
operation of open contingent-claims markets, in which agents buy and sell securities that payoff in the event that specific future events occur. The prices are taken as measures of the likelihood of events or the expectation of the magnitude of an outcome variable to be realized in the future. The second branch of the literature, which is the topic of Section 4, discusses research on pari-mutuel betting markets. These are markets in which traders can make irrevocable bets on a future event, and the market odds change in response to betting behavior. The odds are interpreted as a measure of the likelihood of future events. Contingent-claims and pari-mutuel markets have a close theoretical link. However, because of differences in framing, in the institutions of price formation typically present, and in the revocability of purchases, they are discussed separately here.
