A Polynomial Heston–GARCH Return Model: Quadratic Extensions and Comparative Pricing
Year: 2026
Type: preprint
Abstract: We propose a polynomial Heston–GARCH (HG) joint return specification with quadratic instantaneous variance and separate square-root and proportional leverage channels. The underlying Pearson state diffusion is established; our contribution concerns its return coupling, financial admissibility, weighted moment representation and empirical diagnostics. We prove a sufficient true-martingale condition and derive exact conditional-moment calculations. As an analytical benchmark, a displaced representation of Extended Hull–White volatility links its stationary tail index, short-horizon skewness and moment eigenvalues. Illustrative parameter scenarios show that higher-cumulant growth is sensitive to proportional noise. A reproducible public-index experiment separates factor count, diffusion family and observation information. One-factor HG loses to EHW on reconstruction. An exploratory Heston plus squared-CIR specialization improves on expanded-domain two-factor Heston by approximately 13%, but capping Heston volatility-of-volatility at four preserves the training advantage while reversing the later-period ranking. Withheld-SKEW errors and state-increment diagnostics do not establish predictive or dynamic consistency. A separately sourced 2022 SPXW panel evaluates 6,643 held-out quotes using frozen structural parameters and contemporaneous state fitting. Follow-up tests include the proposed one-factor HG and Heston plus mixed-HG specifications. HG outperforms the saved one-factor Heston fit; the mixed extension improves modestly on squared-CIR, while independent Heston plus GARCH remains the best tested stochastic specification. Nevertheless, local calibration-strike implied-volatility interpolation achieves 99.4% bid–ask coverage against 3.0% for that structural winner. The evidence supports a tractable joint construction and precise conditional comparisons, while identifying parameter-domain sensitivity and poor absolute pricing portability. It does not establish universal model superiority, accurate dynamic filtering or production-ready pricing.
Author: Noguer i Alonso, Miquel
Language: English
Subfield: Finance
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7225870406
