Optimizing Short and Long Term Investment Returns Using Multi-Armed Slot Machine Algorithms

Abstract

Listen

Translate article

This study focuses on comparing the effectiveness of UCB, Thompson sampling and Epsilon-Greedy algorithms in multi-armed slot machine algorithms for short-term and long-term investment return optimization in financial markets. This analysis examines stock performance data from Tesla, General Motors, and Ford over specific periods: five years of weekly data (2019-2024) and six months of daily data (February to August 2024). From the results, it is shown that for long-term investments, Over the five-year period, Thompson Sampling outperformed UCB and Epsilon-Greedy algorithms in terms of stability and overall return, demonstrating a consistent upward trend with fewer fluctuations. For short-term investments, although the UCB algorithm performs similarly to the Thompson sampling algorithm, the returns in the UCB algorithm grow progressively slower as the timeline is lengthened. In the six-month daily data analysis, the Epsilon-Greedy algorithm excelled in capturing short-term gains due to its aggressive exploration mechanism, though it also showed higher volatility in response to market fluctuations, and the algorithm should be chosen to be able to keep up with the fluctuations in the stock market in real time, and the investor should take into account the cost of time as well as changes in the market in order to formulate an optimal investment strategy.The findings of this study will provide a practical type of approach to investors when developing investment strategies. Future research will explore other classes of new algorithms and consider more market factors so that they can be better adapted to the complex investment environment in the future.