Market Efficiency in Racetrack Betting: Further Evidence and a Correction

We recently reported (Asch, Malkiel, and Quandt 1984) the results of simulated betting strategies based on logit analysis of racetrack betting data. Morning line odds, final odds, and marginal odds (defined as the odds implied by the moneys wagered during the last few minutes prior to each race) were used to predict the winning probability of each horse in a race; and a variety of strategies involving bets on the highestprobability horses was attempted. Our main conclusions were that we were unable to devise profitable strategies for win betting but that such strategies could be employed in place and show betting. We have now run the same simulations on a new data set comprising 706 races for the summer 1984 harness meet at the Meadowlands Racetrack. The logit estimations, in which winning probabilities are a function of combinations of morning line odds, final odds, and marginal odds, show coefficients remarkably similar to those of our earlier study, which was based on 1978 data for the Atlantic City Race Course. The rates of return to the place and show betting strategies, however, are substantially lower than those estimated with the 1978 data. Rerunning the simulations on the earlier data disclosed a computer programming error that had inflated the returns to place and show bets.' For each data set the sample of races was divided in half. Coefficients from each half sample were used to estimate probabilities of winning. These probabilities were used in simulated betting strate-