Assessing bettors' ability to process dynamic information: policy implications.

• Author: Johnson, Johnnie E.V.
• Position: Symposium

1. Introduction

   Success in many areas of human endeavor stems from the ability to convert rapidly changing information into accurate probability judgments. Dynamic information environments are often subject to sporadic adjustments resulting from structural instabilities (e.g., in the business world, from announcements concerning impending acquisitions or innovations). There is a wealth of laboratory-based evidence that individuals base their forecasts on human judgment rather than statistical methods when faced by such dynamic information (e.g., Dalrymple 1987; Taranto 1989). These forecasts are often based on heuristics, which result in systematically biased judgments (Kahneman et al. 1982; Timmermans 1993; Baranski and Petrusic 1995). These problems are exacerbated if the information remains implicit (e.g., rumor). Consequently, in an effort to counteract the adverse impact these effects might have on market efficiency, regulators attempt to ensure that market participants receive information in a timely, explicit, and consistent fashion. While the need for information to promote market efficiency is not under dispute, there is growing evidence that individuals may not require explicit information. For example, Remus et al. (1996, p. 23) indicate that "humans have the ability to detect and react to structural instability that characterizes many business forecasting tasks" and there is evidence that judgmental forecasters can develop well-calibrated subjective probability judgments under appropriate conditions (e.g., Johnson and Bruce 2001).

   In summary, there are conflicting views concerning decision makers' ability to handle structurally unstable, implicit information. However, much of the evidence that questions the reliability of individuals' probability judgments has been derived from the laboratory. Consequently, this article examines to what extent and in what manner individuals' probability judgments in a real-world betting market account for dynamic, often implicit, information. The aim is to provide betting market regulators with evidence to help them decide which information is necessary to promote market efficiency and how it needs to be disseminated. This may help avoid the pitfalls of an over-regulated, stifled market whilst allowing for an adequate flow of appropriate information.

   The remainder of the article is organized as follows. Section 2 reviews the literature addressing probability judgments in dynamic information environments and outlines the article's research questions. Section 3 describes the data, explains the methodological advantages of the chosen setting, and describes the procedures used to explore the research questions. Section 4 presents the results, which are discussed in section 5. Some concluding remarks follow in section 6.

2. Probability Judgments in Dynamic Environments

   Existing Literature

   Complexity exacerbates the difficult task of processing data and increases when the information required for judgments remains uncertain and changes through time. Under such conditions, individuals often rely on their own judgments rather than on statistical forecasts (e.g., Kleinmutz 1990; Sanders and Manrodt 1994). These judgments are often less accurate than forecasts based on simple statistical models (e.g., Mocan and Azad 1995; Remus et al. 1995) because the latter forecasts act as a form of task information feedback, which can improve judgments even more than outcome feedback (Balzer et al. 1994; Remus et al. 1996; Sanders, 1997).

   Individuals' assessment of dynamic information can be hindered by their limited cognitive capacity (Hogarth 1987). This capacity is challenged by increased complexity, arising, for example, from implicit information. Under such conditions individuals increasingly rely on heuristics (Bolger and Harvey 1993), which can result in systematic biases (Kahneman et al. 1982; Fildes 1991; Cohen 1993; Harvey et al. 1994) and a reduction in decision quality (Malhotra 1982; Ford et al. 1989; Timmermans 1993) and probability judgment accuracy (Baranski and Petrusic 1994, 1995; Suantek et al. 1996). Furthermore, poor calibration is exacerbated if the information is uncertain or changing (Griffin and Tversky 1992; Chinander and Schweitzer 2003). The assessment of dynamic information is also hindered by the tendency to desire consistent information (Soll 1999). This leads to dissonant information being discounted (Harries et al. 2004), which reduces the ability to react to structural shifts.

   From the previous discussion it is clear that there are a number of factors that can reduce an individual's ability to effectively handle dynamic information. However, research suggests that these findings may arise from the artificial nature of some experimental calibration studies (Gigerenzer et al. 1991; Ayton and Wright 1994). The excellent calibration observed in some naturalistic settings appears to support this view (Murphy and Brown 1985; Keren 1991; Johnson and Bruce 2001). There is also evidence that certain factors can support probability judgments under dynamic, naturalistic information conditions. For example, individuals alter behavior based on outcome information (Kopelman 1986; Jones et al. 1997) and they tend to make judgments based on the most recent evidence in a sequence of contradictory evidence (Ashton and Kennedy 2002). "Recency" may foster appropriate reaction to structural shifts in an evolving information set. In addition, it has been proposed that evolution has equipped individuals to process probabilistic information from frequencies observed in a natural environment (Gigerenzer 2000; Hoffrage et al. 2002). In summary, there is mixed evidence concerning individuals' ability to handle dynamic information.

   Can individuals make full use of dynamic information even if this is not made explicit? The answer to this question may affect how policy makers frame their regulations concerning the provision of information to market participants.

   Regulation Policy and Research Questions

   The British Horseracing Authority (BHA), the official regulatory authority governing U.K. horse racing, administers the rules of racing. For example, they ensure that racetracks adhere to a set of common standards, including the minimum levels of information they must provide to the betting public (e.g., explanations required of trainers if their horses perform unexpectedly badly, etc.). The BHA is mindful of the need to develop regulations that ensure the sport is run effectively and efficiently and in the best interests of a range of stakeholders (e.g., racehorse and racetrack owners, and the betting public). This is achieved by maintaining a balance between sufficient and overly restrictive regulation (which may stifle or reduce the attraction of the sport and the betting market on which it depends).

   Racetrack operators can cause structural shifts in the information provided to bettors (which they require to make accurate probability judgments). For example, racetrack operators can alter the ground conditions between races (e.g., through harrowing or rolling tracks with artificial surfaces). This could, for example, change the advantage of a particular post position (hereafter PP: the barrier position from which a horse starts a race). Previous laboratory-based research, discussed above, suggests that bettors are unlikely to make good probability judgments in the face of such structural shifts, particularly if the information remains implicit or is only allowed to leak out in the form of rumor. If regulators believe that bettors will not effectively process implicit, dynamic information created by such management changes, they might require racetrack operators to minimize the degree to which they engage in these practices and/or to announce their actions in advance. Consequently, to assist the regulators in making such judgments, two key research questions are addressed: To what extent bettors, when faced by a dynamic and implicit information set, (i) account for the full information content of evolving data in their probability judgments, and (ii) form good probability judgments based on heuristics that employ a linear model of current information or a historical model that accounts for the full complexity of information from previous trading periods.

3. Methodology

   This section explains why the horserace betting market is an ideal setting for studying the manner in which dynamic information is used in forming probability judgments and introduces the dataset and procedures that will be used to address the research questions.

   Characteristics of the U.K. Horserace Betting Markets

   In discussing the characteristics of the U.K. horserace bookmaker betting market, the following topics will be addressed: (i) the manner in which subjective probability judgments are revealed in betting markets, (ii) the similarities between decisions made in betting markets and those made in other environments, (iii) how the availability of an unequivocal outcome to a horserace facilitates the analysis of the manner in which information is used by bettors, and (iv) the nature of the particular dynamic information set that is used in this study.

   Revelation of Probability Judgments in Betting Markets

   Bettors in U.K. bookmaker horserace betting markets purchase assets (place bets), the returns to which depend upon the result of the horserace to which the particular market relates.

   "In its simplest formulation, the market for bets in an n-horse race corresponds to a market for contingent claims with n states in which the ith state corresponds to the outcome in which the ith horse wins the race" (Shin 1993, p. 1142). The odds of horse i in race j ([O.sub.ij]) are determined initially by bookmakers' perceptions of the probability of each horse winning. They then adjust the odds as new information becomes available, much of which arrives in the form of bets placed on each horse. Bettors have an...