Research
Job Market Paper (Working Paper)
Abstract
Multi-period volatility forecasting is crucial for financial decision-making. We consider a scenario where the decision-maker specifies an ex-ante loss function, such as the QLIKE, to assess the accuracy of multi-period volatility forecasts from a candidate volatility model. To reduce the impact of model misspecification on forecast accuracy, we introduce an estimator that is `matched' to the specification of the forecast evaluation loss function. We examine the estimator's performance under a bias-variance trade-off, highlighting conditions where it is likely to offer improvements over standard estimation methods. We also propose a model misspecification test based on the Hausman principle, which exploits the fact that our estimator and the standard estimator are consistent for the true parameter under the null of correct specification but converge to different pseudo-true values under the alternative. In a Monte Carlo study, we examine the misspecification with respect to long-memory dynamics. Our results show that the misspecification test is reasonably sized and has power that increases with the degree of long-memory misspecification. Additionally, we recover multi-period volatility forecasts and find that under correct specification, both estimators perform equivalently; however, under misspecification, our estimator provides superior forecast accuracy. Finally, an out-of-sample analysis across ten return and realised measure series from 2001 to 2010 suggests three key findings: first, it is optimal for our estimator to match the estimation loss function to a shorter horizon than the forecasting horizon; second, our estimator provides greater accuracy gains for GARCH-type volatility models applied to realised measures of volatility compared to those applied to returns; and third, our estimator leads to greater accuracy gains for underparameterised models (which are more likely to be misspecified), highlighting the bias-variance trade-off.Other Working Papers
Long Memory Realised GAS Model (2022, draft available soon)
Summary
We introduce a univariate score-driven model that explicitly incorporates long-memory dynamics in the conditional variance of daily returns. We model the conditional variance both as a fractionally integrated process and as a heterogeneous autoregressive model. The new model accommodates heavy-tailed densities for both daily returns and realized measures. This choice of observational densities ensures automatic correction for influential observations through the score function. Our out-of-sample analysis identifies that accounting for long memory is particularly useful for volatility level evaluation and return risk assessment during non-crisis periods.Work in Progress
Simulation-Based Method for Quantiles of Cumulative Variables (2024)