Evaluating Foundation Time-Series Models in Systematic Options Compounding

A prevailing belief in quantitative finance is that deploying high-capacity foundation models inevitably generates superior predictive edge (Gu, Kelly, & Xiu, 2020). This paper presents an empirical evaluation testing whether zero-shot foundation forecasting (Google TimesFM 1.0-200M; Das, Kong, Sen, & Zhou, 2024) and kinematic derivatives physics enhance strike selection and compounding in systematic options selling across Amazon (AMZN), Alphabet (GOOG), and NVIDIA (NVDA) over a continuous 10-year cycle (October 2016 to October 2026 | 2,512 trading days), strictly partitioned into a 5-year In-Sample calibration epoch and a 5-year Out-of-Sample holdout period. We simulate mechanical options compounding under an idealized Zero-VRP invariant (pricing all contracts at zero implied-volatility premium via analytical Black-Scholes-Merton) to isolate pure directional and timing alpha from structural volatility premium capture. On this invariant benchmark, baseline rule-based mechanics achieve terminal equity of $2,349,600 on AMZN (23.50x), $821,300 on GOOG (8.21x), and $27,998,700 on NVDA (280.0x). Integrating TimesFM 10-day ahead forecasts to write defensive call overlays produces zero statistically detectable forecast edge: directional accuracy is indistinguishable from coin flips (50.8% AMZN, 51.2% GOOG, 49.8% NVDA; Diebold-Mariano p-values 0.27, 0.11, 0.42), while call truncation inflicts catastrophic opportunity cost on explosive trend legs (NVDA terminal wealth collapses from 280.0x to 10.9x). Furthermore, kinematic dip-buying rules (velocity-acceleration phase triggers) display severe parameter decay out of sample: on AMZN, out-of-sample terminal wealth sits at the 72.4th percentile of an uninformed random-put placebo (p = 0.28). Deflated Sharpe Ratios (DSR; Bailey & Lopez de Prado, 2014) across all evaluated machine learning enhancements fall to 0.00, demonstrating that the observed historical outperformance was a product of selection bias across non-stationary market regimes.