Ronen Feldman

ProntoNLP's AlphaLLM Aspect-Themes with Consensus signal introduces a groundbreaking Cartesian approach to analyzing earnings calls. This innovation combines horizontal Aspects (Forecast, StrategicPosition, CurrentState, Surprise) with vertical Themes (FinancialPerformance, OperationalPerformance, StrategicInitiatives, etc.) to create a rich multidimensional feature space. 

By representing corporate events as coordinates in this Aspect x Theme x Importance x Polarity grid, the model captures nuanced relationships between sentiment, thematic emphasis, and market impact that traditional linear NLP approaches miss. 

• This novel framework, built on the Llama 3 architecture and fine-tuned on 10,000 expertly tagged paragraphs, further enhances its predictive power by incorporating analyst consensus data from S&P. The resulting 124-feature model delivers impressive backtested results across the S&P 1500 universe since 2010, with the long/short strategy achieving a 216.92% return and a Sharpe ratio of 1.52. The model's ability to process earnings calls on a paragraph-by-paragraph basis, weighting different document sections (prepared remarks vs. Q&A) differently, provides a depth of analysis previously unattainable with conventional NLP techniques.