Aqueduct Racetrack appears in a study of 1,825 horse races run in 1970 at Aqueduct and Belmont Park, where researchers compared betting-market odds with the horses’ actual results. The findings made the track part of a particularly precise demonstration that market prices can aggregate dispersed judgments into predictions that closely match outcomes.

The 1970 study

Arthur Hoerl and Herbert Fallin examined every race at the two tracks during the year. They asked whether the odds, understood as subjective probabilities, predicted how frequently horses won, and whether the market’s ranking of horses predicted the order in which they finished. The races had fields ranging from five to twelve horses, allowing the researchers to compare forecasts across different field sizes.

Rankings and outcomes

With only a few exceptions, the market’s ranking corresponded to the finishing order: the favorite finished first most often, the second-favored horse finished second most often, and the same pattern continued down the ranking. The average finishing position also declined consistently as a horse’s market rank fell, without a single mistake in that measure. The results thus went beyond the simple observation that favorites tend to win; the ranking captured the broader ordering of performance.

How well the odds were calibrated

The odds also closely matched the actual frequency of victories. In the 312 seven-horse races, the favorite was forecast to win 33 percent of the time and did win 34 percent. The second-ranked horse’s forecast and actual win rates were 22 and 21 percent; for the third-ranked horse, both were 16 percent. The fourth-ranked horse was forecast at 12 percent and won 12 percent of the time; the fifth was forecast at 9 percent and won 8 percent; the sixth was forecast at 6 percent and won 8 percent; and the seventh was forecast at 3 percent and won 2 percent. Results in races with other field sizes were similar, though not quite as close.

Significance as a prediction market

The Aqueduct and Belmont findings are presented as unusually strong evidence about the predictive performance of betting markets. They show both ranking accuracy and probability calibration: the odds indicated which horses were likelier to finish ahead, while the predicted chances of winning closely tracked observed frequencies. The example supports the book’s broader account of collective judgment, while remaining a specific historical test rather than proof that every crowd or market is reliably accurate.