Optimalizace denní expozice indexu S&P 500 s využitím strojového učení

This bachelor's thesis deals with the design and implementation of a machine learning model for optimizing dynamic daily exposure to the S&P 500 equity index within the range of 0 to 2. The objective is to evaluate the validity of the semi-strong form of the efficient market hypothesis through the algorithmic analysis of publicly available market indicators. The data foundation consists of a proprietary time series dataset from Hull Tactical. The proposed prediction system combines XGBoost and LightGBM regression models, with a subsequent transformation of excess return estimates into the final allocation via a logistic function. The testing results on out-of-sample data covering 504 trading days show that the tactical capital allocation, with an annualized Sharpe ratio of 1.0961, did not outperform the passive buy-and-hold strategy, which achieved a value of 1.1310. The obtained results are consistent with the semi-strong form of the efficient market hypothesis for highly liquid assets and, in the context of the adaptive markets hypothesis, indicate a high degree of informational extraction from the used indicators by the market ecosystem.