Analýza pravděpodobnosti výher týmů v NBA

This diploma thesis focuses on predicting the outcomes of NBA games in real time using a model based on Markov chains, which employs a transition matrix of probabilities between different game states, with the key variable being the current score. The model was applied to a new dataset comprising 516 games from the second half of the 2023/2024 NBA regular season and compared against the odds provided by the Pinnacle sportsbook. The results revealed an expected ROI of 4.37% and a realized ROI of 4.56% across a total sample of 2,734 situations. Based on my empirical observation that Pinnacle might systematically undervalue underdogs in live betting, a subset focusing on teams with a closing odds greater than 5 was analyzed. This subset achieved an expected ROI of 14.03% and a realized ROI of 51.1%, suggesting that Pinnacle could indeed undervalue underdogs in live scenarios. However, this part of the analysis involved only 248 bets, requiring further validation. The thesis also highlights potential issues with the dataset, such as stale or "frozen" Pinnacle odds, which can distort the analysis. The findings of this study demonstrate the potential of Markov chains for real-time prediction, while emphasizing the need for further research and expanded datasets to confirm the robustness of the results.