Forecasting Realized Volatility with Time Series Foundation Models: A Comparison with Econometric Benchmarks
Updated
Updated · arxiv.org · Jul 6
Forecasting Realized Volatility with Time Series Foundation Models: A Comparison with Econometric Benchmarks
1 articles · Updated · arxiv.org · Jul 6
Summary
A new study compared nine time series foundation models (TSFMs) to eight econometric benchmarks for forecasting realized volatility across 50 financial assets.
Results show that only the Tiny Time Mixers (TTM) foundation model consistently outperformed the established Log-HAR benchmark, and only by a narrow margin.
The findings highlight that model architecture choice matters more than simply using foundation models, and combining TTM with Log-HAR yields robust results across assets.