Research Notes on Deep Learning Fundamentals and Consumer Ratings

Summary

This research digest summarizes two studies on equity signals. The first trains a deep neural network on five years of fundamental data to forecast future fundamentals. The summary reports that, relative to a simple forecast assuming current data persist, the network improved mean squared error, annualized return, and Sharpe ratio in backtests. It argues that relationships among fundamental variables may provide useful predictive information, though the digest does not give sample details, validation design, or precise performance figures.

The second study examines consumer product ratings as an investment signal. It reports that a monthly portfolio buying firms with unusually high consumer ratings and shorting those with unusually low ratings earned a stated excess return range, with no evidence of reversal over the following year. Predictive ability reportedly persisted after controlling for several business and trading measures, and ratings also predicted revenue and earnings anomalies. These are summaries rather than full study methods; data construction, costs, robustness, and generalizability are not detailed.

Key ideas

• A deep neural network trained on five years of fundamentals was used to forecast future fundamental data.
• The digest reports improved forecast error and portfolio metrics versus a persistence-style baseline.
• A long-short strategy based on unusually high and low consumer ratings reportedly generated monthly excess returns.
• The consumer-rating signal reportedly remained predictive after several controls and showed no subsequent-year reversal.
• The document is a research summary and omits important details about data, implementation, costs, and robustness.

Tags

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