Betting markets, whether markets for bets on horse races, sports, or lotteries, offer an ideal naturalistic environment in which to explore biased decision-making. A key pragmatic advantage is the availability of extensive, rich, and detailed quantitative data relating to bettors’ decisions. Since markets are finite in nature, there is a continually expanding set of “completed” markets, that is, a time period during which betting continues up to a defined endpoint, at which time all bets are settled in an unambiguous manner.2 Furthermore, there is potential for comparative analysis across different types of event or bet, according to such recognized criteria as the quality (e.g., Smith et al. 2006), time of day (e.g.,
McGlothlin 1956), or complexity (e.g., Johnson and Bruce 1998) of the event. Thus it is possible to control for some aspects evidence of biased decision-making in betting markets 491 of the decision setting. Most importantly, betting markets include many of the factors regarded as distinctive to naturalistic decision-making (Orasanu and Connolly 1993): uncertain dynamic environments, poorly-structured problems, high stakes, time stress, action/feedback loops, and multiple players.
Each element of the decision-making event (i.e., the bet) is unique: no two horse races or football matches are the same. Thus the outcome is uncertain, and the information relating to that outcome is often (as it is in many real-world decision environments) ambiguous, vague, or redundant. For example, it is not obvious how to combine the various factors that might enable one to predict participants’ performance.
The dynamic nature of betting markets is evidenced by the constantly updating prices as bettors with diverging opinions participate in the market. Bettors, like many decision-makers in real-world environments, often risk meaningful amounts of money while under stress from time pressures (the window of opportunity in a betting market may last only minutes, or even seconds).
A further important feature of these markets is the repetitive nature of betting. Since events take place regularly and often, there is potential for gaining familiarity and expertise with the task. Betting markets involve action-feedback loops; once bets have been placed and a market is closed and decided, bettors receive relatively unambiguous feedback on the success of their decisions, and this can be incorporated into future decisions (Goodman 1998).
Also, betting markets involve multiple players, and it has been shown that the interaction between individuals in markets can significantly reduce errors (Wallsten et al. 1997). This results from a variety of causes, not least the fact that different individuals use different decision-making procedures and have diverse information gathering skills.
As a result, their reaction to the same information may vary. Consequently, the final prices that emerge in these markets take into account a wide range of information and the forecasts of many individuals, and studies show that combining diverse forecasts generally leads to significantly more accurate predictions (e.g., Grant and Johnstone 2010; Vlastakis, Dotsis, and Markellos 2009).
In addition, betting markets are not subject to several of the limitations of laboratory investigations listed above. For example, bettors are unaware that their decisions may be scrutinized, as they are not directly volunteering to take part in an experiment; instead, betting patterns are analyzed in such a way as to observe their decisions unobtrusively.