Financial Market Applications of LLMs

The Gradient · 2 years ago

Researchers and trading firms are exploring whether Large Language Models can predict financial market movements by modeling sequences of prices and trades, similar to how they predict word sequences. Hudson River Trading estimated that stock market data provides approximately 177 billion tokens annually, comparable to the 500 billion tokens used to train GPT-3, but financial markets are fundamentally harder to predict than language because prices are noisier and subject to competitive trading that eliminates predictable patterns. While pure price prediction with LLMs faces challenges, the approach may succeed in areas like synthetic data generation, multimodal analysis combining price data with alternative sources, and supporting fundamental analysis rather than high-frequency trading.