Identifying and modelling online betting patterns

Abstract

Online betting can be viewed as a stochastic process through a gambler's perspective. A sequence of games are played in succession till the gambler logs off only to start another session later. The gambler might decide to stop playing. In this dissertation several statistical and stochastic constructs are proposed, studied and used to model the betting patterns of an individual. A database containing anonymous records of 967 customers was used as testing ground. Distributional fits for a number of variables concerning the gaming history of each gambler yielded various parameter estimates. These estimates were then subjected to clustering techniques which gave us interesting agglomerates. In particular mixture distributions were considered at length. Betting events vary over time - the amount of time each game lasts, the amount of money staked and other variables suggest a stochastic setting. A continuous-time Markov Chain setting was created and fitted to the data. A suitably fitted model leads naturally to phase-type distributions which describe the random time taken for a Markov chain to reach its absorbing state - in our case for the gambler to stop playing indefinitely.