This study introduces an advanced approach to stock market prediction following earnings reports by leveraging Large Language Models (LLMs) fine-tuned with instruction-based techniques and quantized low-rank adaptation (QLoRA). By integrating both financial base factors (e.g., earnings transcripts, financial growth metrics) and external market factors (e.g., analyst ratings, market indices), the model significantly outperforms traditional methods, including benchmarks like GPT-4. The research highlights the llama-3-8b-Instruct-4bit model as a top performer and explores future enhancements, such as incorporating a ‘Hold’ option and extending prediction horizons to accommodate diverse investment strategies.
Emmanuel Hauptmann is CIO and Head of Systematic Equities at RAM AI. He co-founded the company in 2007 and has led the development of the firm’s systematic investment and AI platform since.
