A Bayesian Ensemble Framework for Financial Event Detection via Reward-Bayesian Stochastic Search and Combinatorial Thompson Sampling

Abstract

In the financial market, economic players raise funds from various financial markets, such as the money market and capital market, according to their purposes and manage those funds. The direction of raising and operating these funds may change depending on financial market events. Financial market events include a variety of events, including stock market stress events related to stock market crashes, credit risk events related to bonds, inflation events related to commodities, and liquidity-related events. In order to accurately detect financial market events, key variables with high information power must be selected, and the variable selection method must reflect the dynamic characteristics of the time series. This study proposes a financial event detection framework that applies a variable selection process to macroeconomic variables and sector indices. To demonstrate that our methodology works well in various financial markets, we considered stock and bond markets as the test markets. For these test markets, we demonstrate that our framework outperforms other benchmarks across a variety of test periods.