Please rotate your device to landscape mode to view the charts.

Option Return Predictability with Machine Learning and Big Data

Journal: Review of Financial Studies

Date: 20230901

Author: Bali, Turan G.; Beckmeyer, Heiner; Morke, Mathis; Weigert, Florian

Abstract:
Drawing upon more than 12 million observations over the period from 1996 to 2020, we find that allowing for nonlinearities significantly increases the out-of-sample performance of option and stock characteristics in predicting future option returns. The nonlinear machine learning models generate statistically and economically sizable profits in the long-short portfolios of equity options even after accounting for transaction costs. Although option-based characteristics are the most important standalone predictors, stock-based measures offer substantial incremental predictive power when considered alongside option-based characteristics. Finally, we provide compelling evidence that option return predictability is driven by informational frictions and option mispricing. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

Link: Google Scholar

Background and Context

The Research Question

This study investigates whether machine learning techniques can predict individual stock option returns and identify which characteristics drive these predictions.

Data Coverage

The analysis examines over 12 million observations of U.S. equity options from 1996 to 2020, using 273 different option and stock characteristics as predictors.

Methodology

The researchers compare linear and nonlinear machine learning models to predict delta-hedged option returns, with particular focus on allowing for complex interactions between different predictive factors.

Superior Performance of Nonlinear vs Linear Models

  • Nonlinear models achieve more than twice the predictive power of linear models
  • The nonlinear ensemble achieves 2.04% out-of-sample R² compared to 0.9% for linear models
  • This demonstrates the importance of capturing complex relationships between predictive factors

High-Minus-Low Portfolio Returns Across Model Types

  • Trading strategy based on nonlinear model predictions generates significantly higher returns
  • Nonlinear ensemble achieves 2.04% monthly returns compared to 1.30% for linear ensemble
  • Returns remain significant even after accounting for transaction costs

Predictability Across Information Friction Quintiles

  • Predictability increases monotonically with information frictions
  • Highest information friction quintile shows 5.32% R² compared to near-zero for lowest quintile
  • Suggests predictability is driven by informational inefficiencies

Option Mispricing and Return Predictability

  • Higher levels of mispricing lead to greater predictability and returns
  • High mispricing portfolio generates 2.74% monthly returns vs 0.76% for low mispricing
  • Confirms that option return predictability is linked to market inefficiencies

Contribution and Implications

  • First comprehensive study applying machine learning to predict individual option returns using both option and stock characteristics
  • Demonstrates that nonlinear interactions between characteristics are crucial for predicting option returns
  • Shows that predictability is economically significant and remains robust after accounting for transaction costs
  • Provides evidence that option return predictability is driven by informational frictions and market inefficiencies

Data Sources

  • Model comparison chart based on Figure 1 and Table 2
  • Portfolio returns visualization based on Table 3
  • Feature importance chart based on Figure 7
  • Information friction analysis based on Figure 12
  • Mispricing effect visualization based on Table 9