Co-Authors: Dr. Lendie Follett, Dr. Steven Kou, Dr. Cindy Yu
Abstract: Although the leverage effect, i.e., a negative correlation between the return and volatility, and the inverse leverage effect have been suggested for equities and commodities, respectively, the existing studies suffer from an identification problem because they only model one asset. Using a comprehensive multivariate model with jumps and heavy tail distribution for both a market index and the asset, we find the inverse leverage effect for cryptocurrencies and meme stocks. The same effect claimed previously in commodities markets based on a univariate model is insignificant when using the bivariate model. While the existing literature finds that assets exhibiting a leverage effect benefit from being re-weighted proportionally to the inverse of the variance in a managed portfolio, we show that an asset with an inverse leverage effect should be re-weighted proportionally to the variance. To handle over 18,000 latent variables, a particle Gibbs with an ancestor sampling algorithm is extended to estimate parameters efficiently.
Ongoing research with data science graduate student, Carol Jiang.