Does implied volatility systematically overstate realised moves at earnings? This study analyses 37,508 earnings events across S&P 500 constituents from 2010 to 2025, measuring the IV run-up, the IV-RV spread at announcement, and the profitability of selling ATM straddles at earnings, segmented by sector, market cap, and earnings surprise magnitude.
The central question is whether the implied volatility priced into options before earnings announcements exceeds the subsequent realised move on a systematic, measurable basis. If it does, selling ATM straddles at earnings represents a structural rather than speculative edge. If it does not (or if the edge is consumed by transaction costs), the strategy has no rational basis.
The universe is constructed from CRSP's S&P 500 constituent list (crsp.msp500list),
using point-in-time membership windows to avoid survivorship bias. Every permno that held membership between
January 2010 and August 2025 is included for the quarters it was in the index. Earnings announcement dates
come from IBES (ibes.actu_epsus, quarterly, pdicity='QTR'), and implied volatility
from the OptionMetrics standardised surface (optionm_all.vsurfd, 30-day maturity, 50-delta call).
For each event, we define an event window from T-20 to T+5 trading days relative to the announcement date T. The IV baseline is the mean of T-20 to T-15 (six pre-announcement days well before options begin to price the specific event). All IV values are normalised to this baseline (baseline = 100), making the run-up comparable across stocks with different volatility levels. The straddle P&L uses a BSM ATM approximation: theoretical straddle value equals the 30-day IV times the square root of two trading days over one trading year times the square root of two over pi, applied to a $10,000 notional position.
The chart below shows the average normalised IV from T-20 to T+5, where T is the earnings announcement date. The ribbon represents the 25th to 75th percentile across all events. The baseline of 100 represents the mean IV from T-20 to T-15, a period far enough before the announcement that options have not yet begun pricing the specific event risk.
The run-up is consistent across the full sample. IV typically bottoms at the baseline level around T-15 to T-12, then rises steadily, peaking at T-1 or T=0 (the announcement date itself). The crush is abrupt: IV collapses by the close of T+1 in the majority of events, typically reverting to below-baseline levels within two to three trading days.
The sector breakdown reveals a clear hierarchy. Consumer Discretionary leads at a T-1 peak of 110 normalised IV, followed closely by Technology and Consumer Staples at 109 (the latter is counterintuitive given that staples earnings are typically uneventful, but the IV build reflects systematic option positioning ahead of any scheduled announcement rather than genuine binary uncertainty). Healthcare and Industrials sit in a mid-tier around 107. Financials (105) and Energy (102) carry the lowest pre-earnings build. Energy is the sharpest outlier: oil-linked names carry macro uncertainty that does not resolve at the earnings date, so IV barely compresses at T+1, unlike the 21-22 point collapses in Consumer Discretionary, Technology, and Consumer Staples. For straddle sellers, the highest-crush sectors offer the most reliable post-event IV decay, but their elevated run-ups also reflect genuinely binary outcomes on guidance and product cycles, consistent with the larger left tails in those sectors.
The straddle P&L is modelled using a BSM vega-gamma decomposition. Selling a 30-day ATM straddle one day before earnings and closing at T+1 generates two offsetting effects: a vega gain from the IV crush post-announcement, and a gamma loss proportional to the square of the actual 2-day move. The structural question is not whether you win more often than not, but whether IV systematically overprices the expected move. Across this sample, IV exceeded realised vol by -16.5pp on average, a persistent premium that translates into a positive expected value for sellers in 69% of individual events.
The premium is not uniform across sectors or years. A large IV-RV spread is a necessary but not sufficient condition for profitability: if the actual move is fat-tailed and unpredictable, high IV overshooting can coexist with low average P&L (as in Energy and small-cap names). The heatmap below shows average P&L per trade by sector and year, a more complete picture of where the edge is concentrated and where it breaks down.
| Sector | 2010 | 2011 | 2012 | 2013 | 2014 | 2015 | 2016 | 2017 | 2018 | 2019 | 2020 | 2021 | 2022 | 2023 | 2024 | 2025 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Technology | -7 | +17 | +61 | +106 | +84 | -10 | +45 | +113 | +69 | +75 | +54 | +43 | +57 | +32 | +83 | +118 |
| Consumer Discretionary | +8 | +30 | +55 | +90 | +121 | +38 | +107 | +70 | +44 | +133 | +88 | +79 | +115 | +33 | +92 | +99 |
| Communication Services | +4 | -25 | +41 | +28 | +44 | +77 | +66 | -12 | +41 | +102 | +4 | +58 | -59 | +7 | -32 | +12 |
| Consumer Staples | +11 | +8 | +39 | +43 | +48 | +20 | +40 | +28 | -22 | +95 | +88 | +103 | +50 | +60 | +61 | +75 |
| Healthcare | +15 | -34 | +42 | +23 | +39 | +30 | +2 | -8 | +59 | +32 | +79 | +75 | -49 | +44 | +62 | +18 |
| Industrials | +4 | -14 | +8 | +51 | +15 | +18 | +6 | +33 | +24 | +60 | +44 | +49 | +30 | +5 | +13 | +14 |
| Materials | -1 | -21 | -25 | +59 | +18 | -36 | -12 | -27 | +26 | +24 | -9 | +12 | -18 | -20 | -18 | +1 |
| Financials | +8 | +30 | +21 | +36 | -20 | +13 | -12 | +39 | +26 | +33 | +26 | +43 | -8 | -16 | +49 | +88 |
| Energy | -11 | -63 | -26 | +3 | +14 | -11 | -22 | -26 | +8 | -35 | -60 | -122 | -2 | +28 | -68 | +35 |
| Utilities | -71 | -74 | +20 | +25 | +29 | -10 | +6 | +47 | +48 | +18 | +13 | -1 | -8 | +22 | +14 | +6 |
| Real Estate | +25 | +38 | +44 | -1 | +42 | +35 | +2 | +26 | +24 | +38 | -23 | +19 | -9 | +53 | +46 | +22 |
T-1 entry, T+1 exit straddle, $10K notional per trade. Avg P&L per $10K notional. Color centred at $0: green = profitable, red = loss-making. Cells with fewer than 5 events shown as –.
The two charts show the mean IV-RV spread by GICS sector (sorted by spread magnitude) and by market cap quintile. A more negative spread means IV overpriced the expected move by a larger margin: a larger structural premium for the seller. Quintiles are assigned at each event date using market capitalisation at T-1.
Technology, Consumer Discretionary, and Communication Services show the largest spreads (around -25pp), meaning options priced in roughly 25 percentage points more vol than subsequently realised. Real Estate and Utilities sit near zero: options there price the expected move with far greater accuracy, leaving little systematic premium. The market cap picture is analytically the most interesting: the smallest quintile (Q1) carries the largest IV-RV spread (-22pp), yet delivers the lowest average P&L (+$7) and win rate (64%). High IV overshooting does not guarantee profitability when the move distribution is fat-tailed: the premium is consumed by occasional large losses that thinner option markets price imprecisely.
EPS surprise quantifies how far reported earnings deviate from the IBES consensus at announcement. For a straddle seller, what matters is not the direction of the surprise but whether the resulting stock move exceeds the IV-implied breakeven. Events are split into four absolute categories: large miss (below -10%), slight miss (-10% to 0%), slight beat (0% to +10%), and large beat (above +10%). The chart below shows average P&L per trade and the IV-RV spread by category: two complementary lenses on the same edge.
Large misses (consensus miss of more than 10%) are the single damaging category: despite carrying the widest IV-RV spread of any group (-22pp), average P&L collapses to -$41. The IV overprice is real in aggregate but stock moves on severe misses frequently clear the straddle breakeven by multiples of the implied range. Slight misses (0 to -10%) barely break even (+$4), with the stock move typically landing just inside the breakeven zone. Slight beats are the most profitable (+$48, 72% win rate), showing the narrowest IV-RV spread (-13pp): option markets price these low-uncertainty confirmations with the least excess IV, yet the move is almost always contained. Large beats also perform well (+$38), with a wider spread (-18pp) that reflects greater uncertainty priced into pre-announcement IV for big upside events. The pattern is asymmetric: on the downside, only large misses are damaging; on the upside, both slight and large beats are consistently profitable.
The backtest sells a theoretical ATM straddle for every constituent with sufficient IV data on the last trading day before each earnings announcement (T-1 close), then closes the position at T+1 close. The P&L is aggregated by calendar quarter to control for cross-sectional correlation within earnings seasons.
The chart below shows total P&L per calendar quarter (the sum of all individual trade P&Ls within that quarter) alongside the cumulative P&L line. Each bar represents one earnings season for the full S&P 500 universe. The quarterly view is more informative than the per-trade view because earnings are correlated within quarters: a macro shock in one quarter tends to move all names simultaneously.
The table below shows how performance changes if the straddle is entered earlier (T-5, T-3, T-2) rather than at T-1. For a short straddle, time decay works in your favour regardless of entry. The risk of entering earlier is vega exposure: IV continues running up between entry and the event, and as a short seller, rising IV works against you. Earlier entries also carry more pre-announcement delta risk if the stock drifts before the number drops.
| Entry | N Events | Win Rate | Avg P&L / $10K | Sharpe |
|---|---|---|---|---|
| T-5 | 37,505 | 64.5% | $+15 | 0.55 |
| T-3 | 37,506 | 65.6% | $+16 | 0.74 |
| T-2 | 37,508 | 66.6% | $+20 | 1.14 |
| T-1 best | 37,508 | 69.0% | $+31 | 2.05 |
The positive aggregate P&L and mean IV-RV spread of -16.5pp confirm a genuine structural premium at earnings. However, the quarterly Sharpe of 2.05 is modest, and it is computed before any transaction costs. ATM straddles on short-dated earnings contracts typically carry bid-ask spreads of 5 to 15 percent of the premium on liquid names, and materially more on smaller names. A realistic net Sharpe after costs, for names outside the top 100 by liquidity, would likely be close to zero or negative.
The strategy has a second structural problem: capacity. Executing a straddle sell across hundreds of names every earnings season requires significant option market access, margin, and operational infrastructure. For a systematic fund, this is feasible. For an individual investor, the effective universe shrinks dramatically to names with liquid near-term option markets, which concentrates the exposure in a handful of mega-caps.
| Dimension | Detail |
|---|---|
| Universe | S&P 500 point-in-time constituents from CRSP msp500list; survivorship-bias-free membership windows used throughout |
| Period | 2010-01-01 to 2025-08-29 (15+ years); IV data pulled from 2009 to provide T-20 buffer for January 2010 events |
| Earnings dates | IBES actu_epsus, pdicity='QTR'; announcement date = anndats; deduplicated on permno + anndats |
| Implied volatility | OptionMetrics vsurfd, 30-day maturity, delta=50, cp_flag='C'; secid matched to CRSP permno via 8-char CUSIP through optionm_all.secnmd |
| IV baseline | Mean of T-20 to T-15 (six trading days); events with fewer than 3 valid baseline days excluded |
| Normalised IV | IV at each T-day offset divided by baseline mean, multiplied by 100 |
| Actual move | Absolute 2-day compound return from T-1 close to T+1 close: |((1+ret_T) * (1+ret_T+1)) - 1|; CRSP dsf |
| P&L model | BSM vega-gamma decomposition for a short 30-day ATM straddle held for 2 days: vega_pnl = $10K × 2 × φ(0) × sqrt(30/252) × (IV_T-1 − IV_T+1); gamma_loss = $10K × φ(0) × actual_move² / (IV_T-1 × sqrt(30/252)); net P&L = vega_pnl − gamma_loss; where φ(0) = 1/sqrt(2π) ≈ 0.3989 |
| P&L basis | Gross, approximated from OptionMetrics standardised IV; excludes bid-ask spreads, commissions, and margin costs |
| Sharpe ratio | Computed on quarterly aggregate P&L (sum of all trades in each calendar quarter); annualised as mean / std × sqrt(4); disclosed limitation: within-quarter cross-sectional correlation inflates this statistic |
| EPS surprise | (actual_eps − consensus_mean) / |consensus_mean|; IBES statsum_epsus fpi='6', most recent estimate before anndats |
| Sectors | GICS codes from Compustat company table via CCM link; current (most recent) classification used for all historical events |
| Market cap | abs(prc) × shrout at T-1; quintiles assigned cross-sectionally at each event date |
| Exclusions | Events with missing IV baseline, missing T-1 or T+1 price/return, or no OM secid match are excluded from all statistics |
This study uses 30-day standardised IV as a proxy for the near-term earnings option. The actual option market prices near-term earnings contracts at a significant premium to the 30-day surface, meaning the true premium available to straddle sellers is likely higher than reported here, but the actual option data is noisier and harder to standardise across 500 names. The choice of 30-day IV understates the premium slightly while improving data reliability and comparability.
The 2-day return window (T-1 to T+1) may miss the full earnings move for pre-market announcements, where the stock gaps on T rather than T+1. Using T to T+1 as the event window would slightly increase measured actual moves for this subset. The aggregate effect is small but directionally reduces reported avg P&L and win rate marginally.
GICS sector classifications are as of the most recent Compustat update rather than point-in-time. Several large-cap reclassifications occurred during 2018 (Communications Services sector creation) and will cause some Technology and Consumer Discretionary names to appear under their current sector rather than their historical classification. This affects interpretability of the sector-level results for the 2010-2018 sub-period.