Bayesian Quantile Forecasting Using Realised Volatility

This was my first project when I started my PhD in March 2013. I took me close to a year to complete and put it into a paper.

I did have to learn Bayesian statistics, MATLAB and Markov Chain Monte-Carlo methods, hence the the huge time span! The learning curve was certainly steep, but it has given me a solid foundation going forward, both skill wise and mindset wise. To be frank, it was very daunting at the beginning; the papers I was given by my supervisor were borderline illegible to me. It took a lot of reading, highlighting and explanation by my supervisor and colleagues to manage to get into the academic paper mindset.

This paper builds on the theoretical work by Hansel et al. (2011) and the quantile forecasting of Watanabe (2012), and adds a Bayesian flavour to it, as well as implementing different distributions. It takes 10 data sets, all major international stock indexes (Data provided by Oxford-Man Institute), both returns and various realised volatility measures from 1 January 2000 up to now, updated daily. Technical details are provided in the paper below, but the jist of it is that by using realised volatility, you can improve quite drammatically on the forecast performance for Value-at-Risk and Conditional Value-at-Risk.

This paper was presented at the Australian Statistical Conference (ASC) in July 2014 and is currently under review in the ‘Journal of Forecasting’ (JoF).

If you have questions about any aspect of this project (code,theory etc), please shoot me an email and I will do my best to get back to you as soon as possible.

Download this Paper

Leave a comment