Skip to content
Canli Capital

Research

Options dispersion: literature and implementation boundary

By , Canli Capital

Reviewed: 2026-08-16
Family: options_dispersion
Claim state: mechanism supported; investable return untested
Return identities spent: zero

What the evidence supports

The defensible mechanism is a priced difference between index correlation/variance insurance and the corresponding single-stock option basket. It is not a generic claim that any difference between index and constituent implied volatilities is arbitrage.

Driessen, Maenhout, and Vilkov document a materially higher average option-implied correlation than realized correlation in S&P 500 and DJ30 data and interpret the gap as a negative correlation risk premium. Their implementation combines index and constituent option-implied variances with point-in-time index weights. The study uses OptionMetrics options data, CRSP returns, historical index composition, and 30-, 60-, and 91-day standardized maturities. Missing constituent variance estimates are handled by rescaling available weights, a choice ALPHAC will report explicitly rather than inherit silently.

Carr and Wu establish the model-free option-strip construction used to synthesize variance-swap rates. This supports the measurement layer, not a net trading return: discrete strikes, stale or zero bids, dividends, rates, jumps, and execution all remain implementation problems.

Buss and Vilkov show that option-implied correlations can improve forward-looking beta estimates. That is useful evidence that option cross-sections carry information, but it is adjacent evidence; it does not prove a dispersion portfolio earns an implementable premium.

Marshall directly studies dispersion trading and finds that candidate opportunities decline as transaction costs rise. The sample is only 2005–2007, so it is a warning about friction sensitivity, not adequate evidence for a modern strategy.

The exchange benchmark is a measurement reference, not a backtest

The Cboe S&P 500 Dispersion Index (DSPX) measures 30-day expected dispersion as the square root of the nonnegative difference between a float-market-cap-weighted basket of constituent expected variances and VIX squared. Its universe requires S&P 500 index options and eligible constituent options; valid constituent variances require near- and next-term strips and at least three valid out-of-the-money calls and puts in each strip. The methodology filters invalid and zero-bid quotes.

This is an excellent independent accounting reference, but three boundaries matter:

  1. DSPX is an indicator, not a self-financing, executable options portfolio.
  2. Its published back-tested history starts in June 2014, while the live index launched in September 2023.
  3. Live levels may pull a constituent variance forward from a recent valid observation, whereas the back-tested end-of-day series does not. ALPHAC may never treat a pulled-forward value as a contemporaneously tradable quote.

The methodology also changes option classes for splits, special dividends, mergers, spin-offs, and delistings. Those are mandatory lineage events in our implementation.

Data conclusion

Alpaca is suitable for prospective OPRA capture and paper execution, but its official historical options coverage begins only in February 2024. That is too short to estimate a correlation-risk premium across calm, crisis, rate, volatility, and liquidity regimes.

OptionMetrics IvyDB US is the preferred research spine because it provides end-of-day bid/ask, volume, open interest, option sensitivities, rates, dividends, corporate actions, permanent IDs, and volatility surfaces from January 1996. Cboe DataShop quote intervals are the preferred execution-calibration source because they include historical NBBO and sizes from January 2012; the full one-minute market is very large and must be scoped to the frozen basket.

ALPHAC interpretation

The conventional position—long constituent variance and short index variance—is short implied correlation and can lose sharply when correlations and index downside insurance reprice together. It is therefore not presumed near-uncorrelated to AlphaMax, equity beta, or existing volatility exposure. Admission requires crisis conditional correlations and expected shortfall, not merely a low full-sample correlation.

The first eligible return identity, if the no-return data audit passes, is one 30-calendar-day, monthly-rebalanced, point-in-time S&P 500 top-50 basket. Legs are ex-ante vega balanced. Every short option exposure must have frozen disaster wings; daily delta hedging, early assignment, dividends, gaps, margin, partial fills, quote size, and corporate actions are charged explicitly. Seven-day, 60-day, sector-basket, reverse-direction, and threshold variants remain in the same family trial account and cannot be presented as independent discoveries.

Contrarian conclusion

This family is economically credible but operationally hostile. Its many legs create a large gap between a clean implied-correlation chart and an executable portfolio. If institutional history, point-in-time membership, quote-size calibration, or defined-loss replication cannot be sealed, the correct result is DATA_GATED or KILL_FEASIBILITY, not a midpoint backtest.

Rerun inputs

  • workflow requested: firecrawl-research-papers
  • collection result: Firecrawl unavailable because FIRECRAWL_API_KEY was not configured
  • fallback: original papers, publisher records, official Cboe methodology, and official vendor docs
  • return data opened: no
  • return hypotheses spent: zero