Detecting anchoring effects in a dynamic decision-making environment. An empirical study of the Hong Kong horserace betting market
Anchoring effects refer to a phenomenon whereby people make absolute estimates by starting from an initial point (anchor value) and adjusting upwards or downwards to reach their final answers (target values). A number of studies have demonstrated that these adjustments are often insufficient (e.g., Tversky and Kahneman, 1974) and are subject to many factors, such as the source of the anchor (self-generated or externally given), the relationships between anchors and targets (relevant or irrelevant), the knowledge and experience of decision makers (novices or experts), and features of the decision tasks (conjunctive or disjunctive tasks). Much of the literature has been based on laboratory experiments, but relatively little investigation of this phenomena has occurred in naturalistic environments. Consequently, the aim of this paper is to explore anchoring effects in a dynamic real world decision making environment associated with risk and uncertainty – the horserace betting market to examine whether similar effects are observed in real world as in laboratory settings.
Data from the Hong Kong horserace betting market is used to explore potential anchoring effects by analysing the relationship between bettors’ subjective probability judgments of a horse winning a race and the information that they use to help make decisions. This is achieved by developing ordinal regression and conditional logit models. The factors explored in this study include past performances of horses, jockeys and trainers, and the barrier conditions and official ratings of horses. Tests are firstly undertaken for data from Happy Valley and Sha Tin racetracks combined and then are conducted at the two racetracks separately. Previous research suggests that experienced bettors bet late in the market. Consequently, the odds, which reflect bettors’ subjective probability assessments of a horse winning, are selected at three different time points during the betting period so as to identify the level of anchoring effects associated with experienced as well as less experienced bettors.
Results suggest that the HK betting public as a whole are very well calibrated in terms of their assessments of the winning probability of horses. It appears that they use information efficiently in their decision making models. It is found that bettors do not anchor their subjective judgments on horses’ previous performances, do not anchor their subjective probability of a favourite winning a race based upon pervious performances of favourites, and do not anchor their probability of a jockey and/or a trainer winning the next race based upon their performances in previous races. However, results relating to barrier conditions and official ratings of horses indicate that the HK betting public do anchor their judgments on these factors although their anchoring behaviours at Happy Valley and Sha Tin racetracks are at different levels. Reasons fro these differences are explored in the paper.
In summary, the results presented here challenge the consensus to emerge from laboratory based studies and suggest that anchoring effects may not be as widespread in real world decision making environments as previously thought, suggesting that further investigations in these markets are required.