SINGLE GAME PLAYER PREDICTION AND AN OPTIMIZATION FOR DAILY FANTASY BASEBALL

Abstract

This thesis is concerned with developing a model to accurately predict a baseball player’s performance in his next game within the framework of the DraftKings Major League Baseball scoring system. The objective was to see if this model could earn a profit playing DraftKings 50-­‐50 leagues. DraftKings assigns a cost to each player in action on a given night. Batters and pitchers earn points for accumulating certain positive statistics, and pitchers can lose points for accumulating certain negative statistics. The statistics are classified into two groups depending on their complexity. The “true” rates of the simple statistics are estimated by using a combination of each player’s preseason projections and his performance in the current season. These true rates are then reconciled with true rates of the batters or pitchers faced. The simple statistic rates are then converted into totals and used to predict the more complex statistics. When all the necessary statistics have been predicted for each player, the DraftKings scoring system is applied to obtain a single value for each player. A linear programming algorithm is then used to determine the value-­‐maximizing roster of players whose total cost is below the budget constraint. The results of 88 DraftKings leagues were tracked over a two-­‐month period and the model produced a lineup for “virtual” entry into each one. The model results were computed and its placement was determined in each league. The resulting prize money totaled $1,719.00, a 4.38% return on the $39,261.00 total investment in entry fees. However, since this $39,261.00 would have been invested over two months, it was determined that just $5,561.00 was initially necessary, boosting the return on investment to an impressive 30.91%.