Imagine you are seeking to develop a model that can be used to forecast the volatility of the S&P500
index returns. Assume you are at December 31, 2013.
You are provided with two data sets. The spreadsheet “S&P500.xlsx” contains daily index values
from January 2, 1996 to April 9, 2019 (source Yahoo finance). The second spreadsheet ”Realised
vol.xlsx” contains daily realised volatilities from January 3, 2000 to April 9, 2019 (source
https://realized.oxford-man.ox.ac.uk/).
The realised volatility may be used for model building, but is also to be used as the actual volatility
for forecast evaluation. Note that the realised volatility estimates are calculated using continuously
compounded intraday returns – they have not been scaled by 100.
Using the models and techniques discussed in class,
i) select the most appropriate model(s) for your data in sample and report the estimates
and suitable diagnostic test results. You should only select 3 to 4 models (at most), not
anymore.
ii) evaluate the out of sample forecasting performance of your selected model(s) relative to
alternative models and benchmarks. You are only required to consider the MSE and
QLIKE loss functions.
Write a brief three to four page document (double spaced, including Tables) outlining the models
estimated and your justifications for your selected model(s). I will stop reading after 4 pages.
Note: you do not have to compare the out of sample forecasting performance of all the models
examined in part i). Use your judgement to arrive at a subset of the models to be evaluated in part
ii). For example, you may consider the GARCH(1,1), GARCH(2,1), and GARCH(2,2) processes in part i).
Your out of sample evaluation in part ii) should only probably consider one of these models (the best
one as evaluated in sample).
Do not just provide me with screen dumps from the Eviews output. I want you to think carefully
about how to best present your results. It is also not enough for you to just state that you have
estimated the models. Tables summarising the appropriate statistics for each alternative model
need to be incorporated into the discussion.
There is no right or wrong answer here. I am more interested in seeing your discussion of the results,
methodology, model limitations, table presentation etc…