Marginals Versus Copulas: Which Account For More Model Risk In Multivariate Risk Forecasting?, with Simon Fritzsch and Maike Timphus. Journal of Banking & Finance, forthcoming.
Copulas. We study the model risk of multivariate risk models using a comprehensive empirical study on Copula-GARCH models used for forecasting Value-at-Risk and Expected Shortfall. To determine whether model risk inherent in the forecasting of portfolio risk is caused by the candidate marginal or copula models, we analyze different groups of models in which we fix either the marginals, the copula, or neither. Model risk is economically significant, is especially high during periods of crisis, and is almost completely due to the choice of the copula. We then propose the use of the model confidence set procedure to narrow down the set of available models and reduce model risk for Copula-GARCH risk models. Our proposed approach leads to a significant improvement in the mean absolute deviation of one day ahead forecasts by our various candidate risk models.
with Simon Fritzsch and Philipp Scharner, Journal of Risk and Insurance, forthcoming.
We analyze the relation between digitalization and the market value of US insurance companies. To create a text-based measure that captures the extent to which insurers digitalize, we apply an unsupervised machine learning algorithm - Latent Dirichlet Allocation - to their annual reports. We show that an increase in digitalization is associated with an increase in market valuations in the insurance sector. In detail, capital market participants seem to reward digitalization efforts of an insurer in the form of higher absolute market capitalizations and market-to-book ratios. Additionally, we provide evidence that the positive relation between digitalization and market valuations is robust to sentiment in the annual reports and the choice of the reference document on digitalization, both being issues of particular importance in text-based analyses.
with Simon Fritzsch and Felix Irresberger
Portfolio sorts and cross-sectional regressions are standard tools to test the pricing of asset characteristics. We propose the alternative use of non-parametric machine learning methods to estimate quantile curves of the characteristic of interest conditional on a set of controls. Building portfolios based on conditional quantile curves yields characteristic portfolios that should only reflect the priced risk associated with the characteristic. We apply our procedure to the pricing of volatility risk in the cross-section of option returns. The Sharpe ratio of the resultant characteristic portfolios are up to 30% higher than those of comparable strategies.