Predicting Stock Price Moves from Earnings Call Transcripts

Summary

The document summarizes research on using corporate earnings-call transcripts to predict subsequent stock-price movements. Earnings calls contain management discussion of financial performance and information that investors and analysts may use when making decisions. The described method applies a deep-learning framework to the transcript language: an attention mechanism encodes text into vector representations, which a discriminative network classifier uses to predict price direction.

The summary says the proposed model outperformed traditional machine-learning models in the authors’ empirical work and that earnings-call information improved stock-price prediction. It does not provide details about the data set, prediction horizon, validation design, benchmark models, or magnitude of improvement. The result should therefore be read as a brief account of a research finding rather than a complete evaluation of a deployable trading strategy. The document describes a prediction method, but does not specify how signals would be converted into trades or whether any returns account for execution costs.

Key ideas

• Earnings-call transcripts are treated as textual inputs that may contain information relevant to future stock-price movement.
• An attention mechanism encodes transcript language into vectors for a classifier.
• The classifier predicts the direction of subsequent stock-price movement.
• The summary reports better empirical performance than traditional machine-learning models, without giving evaluation details.
• The document does not describe a complete trading or execution framework.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.