Does a Time-Series Foundation Model Add Economic Value to Systematic Trading? A Six-Study Falsification Test of TimesFM

Publication date: 2026-09-24

Contributors: Ogdn Ames, Ames Investment Systems

Abstract

Time-series foundation models promise strong zero-shot forecasts, but their value to systematic trading depends on incremental information beyond conventional models and on whether forecast gains survive implementation costs. This paper evaluates Google's TimesFM 2.5 across five distinct designs and one deployment study spanning 2000-2026, using walk-forward estimation, same-slot controls, explicit trading costs, and later-date evaluation windows. TimesFM improves some volatility forecasts but provides no robust evidence of incremental economic value in these designs.

Keywords: TimesFM, time-series foundation models, volatility forecasting, systematic trading, volatility targeting, backtest overfitting, deflated Sharpe ratio

Published by Ames Investment Systems

License: All rights reserved
Permission is required for reuse beyond applicable legal exceptions.