Skip to content

The cross-section of individual equity option returns

Bibliographic record. Follow the original-source link for the publication.

Field Value
Primary domain Option Returns
Other domains Volatility, Hedging Exposure Risk, Microstructure
Methods Research Methods
Facets Instrument Single Stock Options
Authors Mobina Shafaati, Don M. Chance, Robert Brooks
Published 2026-06-25
Source Journal of Empirical Finance
Identifiers doi:10.1016/j.jempfin.2026.101748
URL Open original source

Editorial synthesis

Why it matters

The study offers an interpretable sparse-model baseline for individual-equity option returns, examining both forecast performance and whether selected characteristics survive resampling and reselection. It is useful for feature evaluation in delta-hedged option-return research, not as a model of two-leg terminal payoffs, max-loss-normalized returns, or an executable trading strategy. (full_text:S27, full_text:S136, full_text:S138, full_text:S179, full_text:S385)

Main author claims

  • The authors start with 100 stock/firm and 30 option characteristics, remove 22 highly collinear variables, and apply LASSO to monthly cross-sectional ranks. They repeat selection and post-selection Fama–MacBeth estimation across 1,000 resamples of whole months. Historical-minus-implied volatility (hv_iv) and idiosyncratic volatility are selected in every resample, and turnover in roughly 98%–100%. Conditional on the other selected predictors, hv_iv and turnover have positive associations with subsequent delta-hedged returns, while idiosyncratic volatility has a negative association. (full_text:S27, full_text:S108, full_text:S137, full_text:S138, full_text:S230, full_text:S232, full_text:S256, full_text:S271, full_text:S273, full_text:S282)
  • The authors report: In Table 5, the authors reselect and re-estimate models in rolling 120-month training windows and report 159 monthly out-of-sample forecasts. Adaptive LASSO with BIC gives realized-on-predicted slopes of 0.979 for calls and 1.045 for puts, with cross-sectional regression R-squared near 2.1% for both. Predicted-return decile high-minus-low portfolios have annualized Sharpe ratios of 3.351 and 4.228. The higher reported maxima of 3.929 and 4.723 use other penalty settings; they should not be conflated with BIC performance or interpreted as evidence of returns net of full execution costs. (full_text:S398, full_text:S399, full_text:S402, full_text:S408, full_text:S409, full_text:S410, full_text:S416, full_text:S177)
  • The authors' double sorts associate larger hv_iv return spreads with information frictions, demand pressure, and illiquidity, while idiosyncratic-volatility spreads strengthen under several arbitrage and hedging-friction proxies. Turnover behaves differently, with stronger patterns in volatile and smaller-stock segments rather than a consistent arbitrage-cost pattern across liquidity proxies. These are conditional associations compatible with proposed mechanisms, not causal identification of those mechanisms. (full_text:S460, full_text:S462, full_text:S480, full_text:S486, full_text:S528, full_text:S529, full_text:S530)

Data, method, or discussion scope

The main paper uses US individual-equity options from 1996–2019, combining OptionMetrics, CRSP, Compustat, and I/B/E/S, with approximately 200,000 call and 160,000 put contract-month observations. It retains near-ATM calls and puts per stock where available. Actual initial maturities are 37–52 days; positions are held for approximately one month with daily delta rebalancing, not to expiry. Returns divide hedged gains by the absolute initial net cash position |O₀−Δ₀S₀|, not the premium, margin, or maximum loss. Annual and quarterly accounting inputs are lagged six and four months. First online publication was 2026-06-25, not the end of the empirical sample. (full_text:S21, full_text:S141, full_text:S144, full_text:S145, full_text:S159, full_text:S170, full_text:S174, full_text:S175, full_text:S177, full_text:S179, full_text:S205, full_text:S218)

Main limitations

Table 6 fixes characteristics using the full-sample bootstrap and only rolls coefficient estimation, giving feature choice a hindsight advantage; it is not the rolling-reselection design of Table 5. The main text does not establish whether Table 5 also repeats VIF pruning within each training window or how CV folds respect time. Reported R-squared comes from cross-sectional regressions with an intercept, not forecast-error OOS R-squared against zero returns; a slope near one alone does not establish unbiasedness. Returns use bid/ask midpoints and deduct risk-free financing, but the main paper does not report portfolios net of spreads, daily hedge execution, stock borrowing, margin constraints, and impact. Resampling whole months preserves within-month dependence, not serial dependence across months, and does not automatically establish coverage for nonregular post-selection intervals. Double sorts do not identify causality, and the historical ATM monthly sample does not establish validity for today's full option surface, 0DTE, or two-leg combinations. All 25 supplied pages were read; the separate Online Appendix was unavailable, and the authors' data and code were not independently replicated. (full_text:S52, full_text:S53, full_text:S137, full_text:S140, full_text:S145, full_text:S175, full_text:S177, full_text:S179, full_text:S205, full_text:S230, full_text:S231, full_text:S239, full_text:S386, full_text:S404, full_text:S405, full_text:S411, full_text:S441, full_text:S460, full_text:S553)

Relationships

  • None recorded.