Testing for Dependency, Autocorrelation and Weak Information Efficiency in Horse Race Rankings Time Series

Time series of horse trotting race outcomes (rankings) are analysed and tested with time series methods. Horse race rankings are low valued integers that are typical for Poisson processes. Testing for temporal independency in rankings series is conducted with different tests. We focus on 456 series - both on observed and on expected rankings (odds) - and show that close to 40% of analysed observed series and close to 20% of expected series have significant, mainly non-negative, autocorrelations. These are not sensitive to the average ranking levels that the horses have obtained across the races they have finished. In average the observed rankings series show larger persistence than expected rankings. The close relationship between observed and expected rankings and non-independency results imply that a typical weak-form of information efficiency or forecast unbiasedness testing framework can be applied to horse level rankings series. Interestingly some information efficiency results are found mainly under maintained hypothesis, the difference between observed and expected rankings. It is argued that OLS methods are not necessary suitable in this context as the rankings series are integer valued and expected rankings are sensitive to bettors’ risk aversion.