Testing Treasury Yield Features for S&P 500 Price Forecasting

Summary

This article tests whether five-year U.S. Treasury yield data can improve forecasts of the S&P 500, motivated by the idea that stocks and government bonds may reflect shifts in risk appetite. It compares models predicting a future S&P 500 close using index OHLC data, Treasury data, and combined inputs. Evaluation uses time-series cross-validation without random shuffling, with a forecast horizon of 20 steps.

Treasury-only inputs performed worse across the tested models, with greater error variation, and feature selection did not retain Treasury variables. The selected stochastic-gradient regressor was then tuned with L-BFGS-B, but the tuning overfit training data and failed to beat default settings. The article concludes that index data appeared more useful in this experiment. Its findings concern the chosen data and modeling setup; the observed relationship is not stable evidence of a general trading edge, and the excerpt does not establish live trading profitability.

Key ideas

• The study compares S&P 500 forecasting with index OHLC data, Treasury yield data, and both sources together.
• It uses chronological cross-validation with a gap matched to the stated forecast horizon.
• Treasury-only inputs had worse reported forecast performance and greater error variation.
• Feature selection did not keep Treasury variables in the chosen model.
• Hyperparameter tuning overfit the training data and did not beat default settings.

Tags

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