آیا دلال ها بهتر از شرط بندها نتایج را پیش بینی می کنند؟

In this paper we test the hypothesis that bookmakers display superior skills to bettors in predicting the outcome of sporting events by using matched data from traditional bookmaking and person-to-person exchanges. Employing a conditional logistic regression model on horse racing data from the UK we find that, in high liquidity betting markets, betting exchange odds have more predictive value than the corresponding bookmaker odds. To control for potential spillovers between the two markets, we repeat the analysis for cases where prices diverge significantly. Once again, exchange odds yield more valuable information concerning race outcomes than the bookmaker equivalents.

There have been many studies of the efficiency of horse race betting markets, based either on bettor determined prices (as in pari-mutuel markets) or bookmaker determined prices, the latter mainly based on UK data (see, for example, Smith et al., 2006). Most studies of bookmaker markets indirectly infer the superiority of bookmaker skills over bettor skills from the existence of persistent negative returns to bettors in aggregate. However, if bettors receive consumption utility from placing wagers in addition to utility from monetary returns, bettor superiority may be consistent with aggregate negative returns. Furthermore these studies tell us nothing about the abilities of bettors who choose to refrain from entering the market when they judge that bookmaker prices overstate the true chances of race entrants. In this paper we use matched data from traditional bookmaking and person-to-person exchanges to test the hypothesis that bookmakers display skills superior to bettors in predicting the outcome of sporting events. One might expect on the basis of the already extensive literature in the economics of auctions (e.g. Klemperer, 1999 and Klemperer, 2004) that the decentralised nature of the decision-making processes characteristic of betting exchanges would accomplish the aggregation of dispersed information in a very efficient manner, whereas the bookmaker (however well informed) may fail to match as efficiently the information revealed through such decentralised bidding. The point here is that the decentralised market aggregates information in a way that no-one is able to do individually. Indeed, there is a growing body of literature which shows that decentralised exchange markets are very efficient in providing forecasts of the probability, the mean and median outcomes, and the correlations among a range of future events. Such markets have been used very successfully to predict uncertain outcomes ranging from the box office prospects of Hollywood movies, through vote shares in elections, to the sales of Hewlett-Packard printers (e.g. Wolfers and Zitzewitz, 2004, Snowberg et al., 2005 and Gruca and Berg, 2007). The structure of the paper is as follows. Section 2 gives relevant background information relating to the betting markets analysed in our study. Section 3 describes the data drawn from bookmaker and betting exchange markets. Section 4 outlines the methodology employed. Our results are presented in Section 5, with discussion. Section 6 concludes.

The visual evidence in Fig. 3 supports the conclusion, drawn from our χ2 significance tests, that the exchange odds tend to have greater predictive value than the bookmaker equivalents in our sample. Our expectation that nominal betting exchange odds have more predictive value than bookmaker odds due to the lower degree of favourite-longshot bias in the former is confirmed. After adjusting for bias the exchange odds continue to hold more information concerning race outcomes than bookmaker odds, and this result is also broadly confirmed in those instances where price divergence is higher than the norm; although the advantage at pd ≥ 0.02 disappears and is only marginal at pd ≥ 0.03, the number of cases in these categories is relatively few. In the main, therefore, the exchange odds prove to be superior predictors of the results of the sampled races. Our principal finding contrasts with that of Levitt, who found that bookmakers exhibited superior skills in evaluating objective outcomes in the handicapping contest that was the medium for his study. The observation was made in the introduction that Levitt's methodology made it likely that the preferences of the most skilled or informed bettors might not be revealed if they decided that the terms of the wagers set by bookmakers were unfavourable and, in consequence, chose not to trade. The same would be true of the bookmaker markets studied here. In contrast the betting exchanges offer opportunities for these bettors to trade which are not available in bookmaker markets. For example, skilled traders, insiders, and bettors seeking hedging opportunities are all able to lay odds on the exchanges which may as a result more accurately reflect the chances of the horses concerned than those offered by bookmakers. In these circumstances we might expect the proportion of turnover attributable to casual bettors to be lower in the exchanges than in bookmaker markets, with a consequent tendency for odds to reflect more closely objective probabilities. This account of the differences between the two markets is consistent with recent transaction cost explanations of the structure of betting markets, outlined in Section 2. It is also consistent with the empirical evidence suggesting that decentralised markets are efficient predictors. Differences in the nature of traders and trading activities may therefore explain the greater relative efficiency of the exchanges in reflecting objective outcome probabilities observed in the current study. Similarly, the results presented here may not so much contradict Levitt's findings as reflect a different composition of traders engaged in the respective betting media studied. As the betting exchanges continue to expand in size and liquidity, it will be interesting to monitor how well they continue to predict the outcomes of events for which they offer markets.