LLM Trading Strategies Fall Short Against Buy-and-Hold Benchmark

A comprehensive academic study has found that large-language-model (LLM) trading strategies fail to beat the oldest trick in investing - buying and holding.

The research, titled 'Can LLM-based Financial Investing Strategies Outperform the Market in Long Run?', tested AI-driven trading approaches across more than 20 years of market data and over 100 stock symbols. The conclusion was blunt: LLM strategies consistently underperformed a simple buy-and-hold benchmark.

During bullish market phases, the LLM strategies were excessively conservative, failing to capture the full upside of rising markets. When markets turned bearish, the models flipped to being overly aggressive, absorbing significant losses precisely when capital preservation mattered most.

The researchers built a backtesting framework called FINSABER specifically designed to address biases that plagued earlier studies. However, the study focused entirely on traditional equity markets and did not test AI trading in crypto assets, leaving open the question of whether these strategies work in digital currencies.