Forecasting Implied Volatility Returns for At-The-Money Currency Options Using Machine Learning

Abstract

This paper explores the use of machine learning models to predict what characteristics affect illiquidity in stocks using historical data. The paper uses thirteen different regressions, exploring the effects of 43 characteristics. The regressions are run with and without the variable bid-ask spread. The in-sample findings suggest that the oracle, group lasso and enet regressions are outperforming the OLS regression both with and without bid-ask spread. Bid-ask spread is seen to be the variable with the highest correlation in the out of sample analysis. The regressions without bid-ask spread show more variance in the results also showing the variables BM and VolMkt to be most correlated. Concluding that the bid-ask spread is the most correlated characteristic.