1
$\begingroup$

I have a daily time series data spanning over 22 years. I need to compute some meaningful yearly standard deviation statistics / generate probability distribution and estimate tail risk. 22 years obviously is not enough, so was wondering about perhaps generating overlapping 1-year change time series and analyzing the daily change in it - that would give me around 5200 observations. Struggling a bit on how to interpret the distribution that I generated - it appears to be a daily change of the yearly change, so was wondering if it could be used at all or if there is a better way to use overlapping data?

Thanks a lot for the inputs,

D>

|improve this question
$\endgroup$
  • $\begingroup$ Hi: The crucial question is what distribution you are interested in ? The daily distribution, the weekly, the yearly ? Then, that determines how to calculate the returns etc. Also, don't think of it as a daily return of a yearly return. Just think of it as a daily return and then sum of those daily's result in whatever other return ( weekly, yearly etc ). $\endgroup$ – mark leeds Sep 17 '18 at 17:25
  • $\begingroup$ Hi - thanks for the response. I want yearly distribution, but believe limitation is that I have only 22 end of year data points - not sure how to overcome that if I want to estimate tail risk? $\endgroup$ – 291890964 Sep 17 '18 at 17:27
  • $\begingroup$ Note, to be more precise, it's not really the sum. For small returns, the sum is okay. Generally speaking, you take the smaller timescale return and calculate the cumulative product of that return ( over whatever period you're interested in ) and then subtract 1.0. $\endgroup$ – mark leeds Sep 17 '18 at 17:28
  • $\begingroup$ Hi - the data is built from delta in 2 yield curves - based on the 1 year data point in each curve for each day since 1996 $\endgroup$ – 291890964 Sep 17 '18 at 17:29
  • $\begingroup$ yearly, is a problem. maybe it's best to take some smaller time scale estimate ( say daily ) and then assume independence so that you can use the $n \times \sigma^2$ estimate to get the time scale comprising $n = 252$ periods of the daily. Others may have better ideas. Obviously, you'd prefer more years. $\endgroup$ – mark leeds Sep 17 '18 at 17:30

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.