Predicting Skew Changes Ahead of Earnings Events

The volatility skew—the pattern of implied volatility across strikes—undergoes predictable changes around earnings announcements. Before earnings, volatility rises and often becomes more skewed (out-of-the-money puts become expensive relative to calls, reflecting jump-down risk). Machine learning can predict these skew changes by analyzing pre-earnings market signals and historical patterns.

The Earnings Volatility Effect

Earnings introduce uncertainty. A company might beat estimates (stock rallies) or miss (stock declines). This binary-event nature creates jump risk: large discrete moves possible, not just continuous diffusion.

Options markets price this jump risk. Out-of-the-money puts become expensive (they protect against the jump-down scenario). This creates skew: volatility is higher at lower strikes than at higher strikes.

The magnitude of skew increase depends on expected earnings surprise size (more uncertain earnings → more skew) and investor risk aversion (higher risk aversion → more put demand → more skew).

Pre-Earnings Signals

Machine learning can predict post-earnings skew by analyzing pre-earnings signals:

• Recent volatility history: stocks with high recent volatility often have larger earnings-driven moves
• Implied volatility term structure: how much does IV increase closer to earnings?
• Put-call ratios: elevated put buying indicates expected downside
• Order-book positioning: unusual activity in OTM puts signals hedging demand
• Sentiment data: analyst revisions, options flow, social media signals
• Historical patterns: some stocks show more skew increases at earnings

A predictive model combines these signals to forecast how skew will change.

Feature Engineering for Earnings Prediction

Effective features include:

• IV rank: where is current volatility relative to historical range?
• Implied move: derived from ATM straddle price, indicates market's expectation of move size
• Realized volatility on prior earnings: how much did stock move last earnings?
• Analyst consensus and dispersion: more dispersion → larger expected move
• Put-call option flow skew: are traders buying more puts than calls?
• VIX term structure: overall market risk aversion level

Modeling Skew Dynamics

The skew can be characterized by a few parameters: ATM IV, skew angle (how steep the IV curve is), skew convexity (curvature). Predicting skew change means predicting how these parameters evolve.

Time-series models (ARIMA, GARCH) applied to skew parameters can forecast changes. Neural network models can learn nonlinear relationships between pre-earnings signals and skew parameters.

An alternative: directly forecast IV at multiple strikes simultaneously, using a multivariate model that ensures smooth surface output.

Recurrent Models for Temporal Dynamics

Skew evolution before earnings often shows patterns: skew gradually increases as earnings approach, with acceleration in final days. RNN models capture these temporal dynamics.

By observing skew evolution over several days pre-earnings, the model can predict the likely skew path forward.

Practical Skew-Trading Strategies

Predictions enable several strategies:

• Skew flattening trades: if predicting skew to flatten post-earnings, sell OTM puts and buy ATM puts
• Straddle strategies: long straddles if predicting large moves; short straddles if predicting small moves
• Calendar spreads: profit from changing term structure as earnings approach then pass
• Volatility selling: sell options if prediction indicates IV will decline

Cross-Asset Relationships

Skew predictions improve by considering related assets. If competitors are reporting earnings, their skew changes often correlate with the stock in question. Index volatility skew also provides information.

Multi-asset models that jointly forecast skew for multiple related stocks perform better.

Post-Earnings Analysis

Comparing predicted skew changes to realized changes enables model improvement. Some earnings involve surprises (earnings beat/miss that weren't priced in); the model should learn which pre-earnings signals are most predictive in these cases.

Conclusion

Volatility skew follows predictable patterns around earnings events. Machine learning models that analyze pre-earnings signals can forecast skew changes, enabling profitable options strategies. The intersection of options analytics and earnings-driven event prediction is a natural domain for machine learning application.