I'm a masters student looking for a thesis topic. Please note, I'm studying Finance and whilst the econometrics component of my course is rather lacking when compared to a straight Econometrics degree, thanks to the internet there is a vast wealth of resources out there for me to try and grasp it on my own - with some friendly help along the way of course :)

I'd like to use the thesis as an opportunity to learn a new regression model. Specifically GARCH models used to forecast volatility.

As I understand it, the GARCH-MIDAS model (as described by Engle et. al 2013) can and has been used with daily stock return data and macroeconomic data (typically quarterly or monthly) to produce volatility forecasts that contain both long and short-run components. The GARCH model encompasses the mean-reverting short-run fluctuations in volatility, whilst the MIDAS component captures the long-run effects.

This is what I gather from briefly looking over the paper late last night. Any corrections are welcome!

In terms of time-series experience, I can implement AR, ARDL, VAR and VECM models. I have never used ARCH, GARCH, or any of their variations; though as I said I would really like to learn.

I guess all I am asking is whether this is feasible option for me as a student with no formal teaching, but a desire to learn by himself. If so, can anyone point me to some books with practical expositions of the GARCH and MIDAS regression methods? I presume an understanding in each method separately would greatly aid combining the two.

Your thoughts and opinions would be greatly appreciated!


Sean Becketti has a very readable book called Introduction to Time Series Using Stata that has a chapter on (G)ARCH models. Unfortunately, it does not discuss mixed-data sampling (MIDAS), as that is a more advanced topic. I am not aware of any introductory treatment of that material.


Forecasting with Mixed Frequencies by Michelle T. Armesto, Kristie M. Engemann, and Michael T. Owyang has a nice overview of several mixed frequency prediction techniques (including MIDAS) and evaluates them on several macroeconomic series:

For our forecasting experiments, the data we use are log growth rates of the seasonally adjusted annual rate of nominal GDP from the Bureau of Economic Analysis, seasonally adjusted nonfarm payroll employment and seasonally adjusted CPI from the Bureau of Labor Statistics, and seasonally adjusted IP from the Board of Governors of the Federal Reserve System.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.