Tutorial: Semi-parametric financial tail risk forecasting incorporating realized measures
Venue: Information Science Building (No. 36) of the NCHU (see campus map).
Room: Room U414 (4F of the Information Science Building).

Prof. Richard Gerlach and Dr. Chao Wang, The University of Sydney Business School, Australia.
Email: Contact


Quantitative financial tail risk measurement and forecasting provide a fundamental toolkit for financial risk management, investment decisions, capital allocation and external regulation. Value-at-Risk (VaR) and Expected Shortfall (ES) are tail risk measures that are employed, as part of this toolkit, to measure and control financial risk. In this tutorial, we begin with a brief introduction to the most common tail risk measures: Value-at-Risk (VaR) & Expected Shortfall (ES) and to the three main types of the financial tail risk forecasting models in the literature: parametric, non-parametric and semi-parametric. We also introduce various realized measures of volatility, commonly used in literature nowadays. Second, we focus on introducing and implementing several semi-parametric tail risk forecasting models, starting with the well-known Conditional Autoregressive Value at Risk by Regression Quantiles (CAViaR) model (Engle and Manganelli, 2004), then the Conditional Autoregressive Expectile (CARE, Taylor, 2008) model which jointly estimates quantiles and expectiles (and implicitly ES too). A brief introduction to estimation, both frequentist via loss functions and Bayesian is given to round out the first 2 hours. In the 2nd two hour slot, we present a semi-parametric Realized-CARE framework of models and its implementation. This framework extends the CARE model by incorporating a measurement question that contemporaneously links the latent conditional expectile with the realized measure. Next, the recent finding of a class of joint VaR and ES loss functions motives the development of semi-parametric tail risk models that jointly estimate and forecast VaR and ES. We start this part with introducing a joint ES and quantile regression framework (based on the ES-CAViaR, Taylor, 2017). Then two innovative frameworks extending that model class, which allow separate dynamics for the ES equation and/or allow a separate measurement equation are presented. The tutorial will be conducted using Matlab, and many illustrations through real forecasting examples of financial return series will be presented and discussed. Though it is not compulsory, participants are encouraged to bring their laptops with Matlab installed as some code will be shared for participants to replicate our work and examples.

Programme - Friday, 28th of June 2019
08:30 - 10:30Session I
10:30 - 11:00Coffee break
11:00 - 13:00Session II