Convolutional neural network improves monthly skewness estimation from high-frequency data
Estimating "Realized" Skewness using Convolutional Neural Network
Machine Learning
Summary
Estimating how financial returns are skewed over a month is tricky when only looking at low-frequency data. The authors propose using a special kind of AI called a convolutional neural network that learns from detailed 5-minute data to better estimate monthly skewness. Their tests show it reduces errors compared to traditional methods and stays reliable even if the model is not perfect. They also confirm it works well on real financial portfolios sorted by skewness.
What this means in practice
- •For quantitative finance teams: Use the neural network estimator to get more accurate monthly skewness measures from intraday market data for portfolio construction.
- •For financial risk managers: Improve risk assessments by integrating a more stable and less biased skewness estimator based on high-frequency data into risk models.
Authors
Haozhe Jiang, Ostap Okhrin, Michael Rockinger
Abstract
We propose a new estimator of low-frequency skewness (monthly) that exploits high-frequency data (5 min) through a direct functional mapping consisting of layers of Convolutional Neural Networks followed by Multi Layer Perceptrons. Monte Carlo experiments involving stochastic volatility models, with and without jumps, show that the estimator reduces finite-sample bias relative to existing realized-skewness estimators and remains accurate under model misspecification. Empirically, our estimator exhibits temporal stability and delivers cross-sectional performance in actual skewness-sorted portfolios. We characterize the finite-sample bias of conventional plug-in skewness estimators and establish risk consistency properties of the proposed NN estimator under standard regularity conditions.