Value Betting Engine: del modelo de probabilidad a la apuesta de valor con el criterio de Kelly

Value betting is a misunderstood concept in sports betting. People often believe it's about predicting the winner, but it's actually finding situations where the market's odds deviate from the true probability. Value betting is mathematically equivalent to identifying undervalued assets in the stock market. The key to success lies in two components: (1) estimating the probability of each outcome accurately and (2) comparing it with the market's odds to detect value and determine the appropriate bet size without risking ruin.

To estimate the probability, I utilize the Sports Performance Engine, a model built on LightGBM and XGBoost using StatsBomb data. This reliable model forms the foundation for the value betting strategy. To identify value, I compare the market's odds against my estimated probability. The simple but decisive signal of value is when the expected value (EV) of the bet is positive: EV = p · (odds - 1) - (1 - p), where p represents my estimated probability. Only bets with positive EV are considered.

The next crucial step is determining the appropriate bet size. Here, I employ the Kelly criterion, which calculates the fraction of capital that maximizes long-term geometric growth. However, to avoid excessive risk, I use a fractional Kelly approach, betting a smaller portion of the recommended amount. This reduces exposure to errors in probability estimation while still capitalizing on profitable opportunities.

To ensure the strategy's robustness, I conducted backtesting from 2022 to 2024. By simulating the bankroll over time, I verified that the strategy can withstand bad runs, even with positive EV. The lessons learned are clear: proper risk management, akin to the Kelly criterion, is as essential as accurately detecting opportunities. A great model with poor sizing will fail, but a modest model with disciplined sizing will endure. This principle applies to any trading environment.