I analyse UK horse racing markets time series data from Betfair - an online sports betting exchange, and derive a range of stylized facts. The dataset used is a historic data package available for purchase on the Betfair website and thus cannot be shared in this repository.
I identify the unconditional distribution of log returns as a generalised Gaussian. Unlike the distribution of financial asset returns which is known to exhibit heavy tails, the distribution of returns from a betting exchange shows signs of light tails with a tail index closer to that of a Gaussian than a power law. Statistical testing and analysis show a lack of gain-loss asymmetry in this distribution. The absolute returns are concluded to be difference stationary while log returns stationary. Linear autocorrelations appear to be negligible with exception to for the first few lags. Nonlinear autocorrelations show faster decay than that typical of financial time series, indicating a lesser degree of volatility clustering. Lastly, the estimation of the Hurst exponent affirms the absence of long-range memory in betting exchange time series, which points to evidence of mean reversion.
Overall, the results suggest higher informational efficiency than that typically exhibited by financial markets, according to the stylized facts about them.