Options Backtesting Architecture

Six-stage options backtesting pipeline mechanics: surface IV calibration, natural execution modeling, phantom alpha from mid-price fills, physical vs. cash settlement assignment risk, and CPCV out-of-sample validation.

Overview

Options backtesting is categorically more complex than equity simulation. Where equity engines evaluate a single scalar price trajectory, options engines must navigate a high-dimensional surface parameterized across strikes (K), expiries (τ), and option rights (ω). Every tick evaluation requires implied volatility calibration, Greek computation, margin allocation, and settlement tracking. The consequence of shortcuts — especially midpoint execution fills — is phantom alpha that can artificially inflate annualized returns by 16% or more on a standard $50,000 account.

Key Concepts

Phantom Alpha from Midpoint Execution

The single most common source of backtest distortion in options research. When a backtest fills orders at the midpoint of the bid-ask spread rather than at the natural bid (for sales) or ask (for purchases), it captures half the spread width as unearned income on every leg. A simple iron condor with 200 annual trades across 4 legs and a $0.10 mid-spread width generates$ 8,000 in fictitious profit — a 16% annualized boost on a $50,000 account. Accurate backtesting requires natural execution: selling at the bid, buying at the ask.

Surface Calibration

Raw option chain data often omits implied volatility and Greeks. Before any analysis can proceed, the engine must invert the Black-Scholes-Merton (for European contracts) or Bjerksund-Stensland (for American contracts) pricing model to extract implied volatility (σ) from observed market prices. This is a numerical root-finding problem solved using Brent's method. Crossed markets — where P_bid > P_ask — must be purged before calibration.

Natural Execution and the Six-Stage Pipeline

A complete options backtesting engine operates as a six-stage state machine:

Chain Ingestion & Normalization — Load raw ticks, purge crossed markets, calibrate IV surface
Leg Discovery & Delta Targeting — Find strikes matching target delta using root-finding (Δ ≈ ∂V/∂S)
Execution Modeling & Margin Allocation — Apply natural fills, deduct Reg-T or Portfolio Margin
Mark-to-Market & Position Lifecycle — Track Greeks, NLV, DTE exits, and stop-loss triggers
Assignment, Exercise & Settlement — Process OCC auto-exercise (≥ $0.01), ex-dividend assignment checks, cash vs. physical settlement
Risk Diagnostics & Out-of-Sample Tests — Compute Sortino, CVaR, Tail Ratio, Max Drawdown; run CPCV validation

Assignment and Settlement Risk

Options on cash-settled indices (SPX, NDX) expire to a single cash transfer with no overnight equity exposure. Options on physically-settled equities and ETFs (SPY, AAPL) convert to 100 shares per contract upon expiration. This distinction is critical:

Early exercise risk: Only relevant for American-style physically-settled options. Counterparties exercise calls early before ex-dividend dates when D > C_extrinsic, and exercise puts early when r·K·τ > P_extrinsic.
Pin risk: When an underlying closes at-the-money at 4:00 PM EST, assignment uncertainty persists until 5:30 PM EST when OCC contrary exercise notices clear.

CPCV (Combinatorial Purged Cross-Validation)

Standard cross-validation breaks down for options strategies because overlapping contract lifecycles create data leakage between training and test sets. CPCV solves this by purging training samples that overlap with test contract windows and embargoing subsequent observations affected by volatility shocks from the test event.

Formulas

Surface Calibration Root-Find:

C_market - C_model(S_t, K, τ, r, q, σ) = 0

Solved numerically via Brent's method to extract implied volatility σ.

Natural Execution with Spread Penalty:

P_fill,buy = P_mid + α × ((P_ask - P_bid) / 2)

Where α ∈ [0, 1] is execution friction (α = 1.0 = full natural fill at the ask).

Ex-Dividend Call Assignment Trigger:

D > C_extrinsic = C(S, K, τ) - (S - K)

When the declared dividend D exceeds the call's remaining extrinsic value, rational holders exercise early.

Cost-of-Carry Put Exercise Trigger:

r · K · τ > P_extrinsic

When rising interest rates make carrying cash proceeds more valuable than holding the put's remaining extrinsic value.

Key Takeaways

Mid-price fills are fictitious: Even a $0.05 spread error per leg across 200 trades compounds into thousands of dollars of phantom P&L.
Calibrate before you compute: Any engine omitting implied volatility inversion will compute incorrect Greeks and delta targets.
Cash-settled vs. physically-settled is not cosmetic: Physical settlement introduces overnight assignment risk, margin shock, and ex-dividend hazards that cash-settled indices eliminate entirely.
Data scale is the engineering bottleneck: US listed options generate ~400 million 1-minute NBBO bars per day. Efficient engines require Parquet partitioning, sparse surface grids, and incremental gap-fill caching.
Validate on broad parameter surfaces: If strategy performance peaks sharply at 18-delta and collapses at 17 or 19, the result is statistical noise, not a robust edge.