Forecasting Odds Movements in Horse Racing
Term
4. term
Education
Publication year
2020
Submitted on
2020-06-02
Pages
63
Abstract
In this thesis we consider the odds movements in horse racing on a betting exchange. Due to the odds structure on a betting exchange we choose to model the odds movements as the tick increments rather than the change in odds. This leads us to construct the odds trajectory model. Since the parameters in the odds trajectory model are not timevarying we explore an alternative state-space model, namely the dynamic Skellam model. The dynamic Skellam model is a nonlinear non-Gaussian model and this presents some challenges in filtering and smoothing the data. As a smoothing method we maximize the likelihood function where the evaluation of the likelihood function is done by using importance sampling method. The filtering of the data also utilizes importance sampling more specifically we use a particle filter, namely the bootstrap filter. Lastly, we forecast the odds movements by combining the estimated filtering values with the odds trajectory model
Keywords
Documents