LIVE · SPY IV SURFACE · MLP

Data real SPY end-of-day option chains, 2010–2023 (Kaggle EOD) · OTM-filtered & liquidity-cleaned → ~5.9M quotes · Training 80,000-quote sample, 20,000 held-out validation (random i.i.d. split) · Inputs → target moneyness × maturity features (+ macro) → implied volatility · 60 epochs

Architecture

Features

Activation

architecture

parameters

train RMSE

val RMSE

Network architectureinputs → IV

Implied volatility surface prediction → real surface (web)

Training dynamics · RMSEtrain vs validation

epoch 1 / 60

@davidarias_cfa · trained on real SPY option chains 2010–2023

Model comparison · all variants click a row to load it · ranked by val RMSE