5

I'm in the position to calculate a non-parametric volatility estimator for 15 and 30 minutes intervals of the SPY. I got data sampled on second resolution. However, I checked plenty of papers but, as far as I understood them, all of the proposed models are solely applied to measure daily volatility. All of the proposed kernels or subsampling methodologies to deal with microstructure noise and/or jumps are estimated to get daily volatility. How do I estimate a robust realized volatility measure for the stated intraday frequencies?

SRKX
  • 11,126
  • 4
  • 42
  • 83
nan
  • 61
  • 2
  • 1
    Which papers are you referring to? When they describe the methods they use, they presumably talk about time scales in order to come up with the methods - can you alter these to infer kernels / susampling methodologies to different time periods? – will Jan 04 '17 at 19:58
  • All approaches presented in(page 65ff): http://edoc.hu-berlin.de/dissertationen/gross-klussmann-axel-2012-05-14/PDF/gross-klussmann.pdf – nan Jan 04 '17 at 20:23
  • Have you searched previous question on this site. This looks relevant. http://quant.stackexchange.com/questions/2589/how-to-calculate-historical-intraday-volatility – Will Gu Jan 05 '17 at 05:20
  • 1
    Yes, I checked it but the answers are not really specific. I just realized that probably the mentioned methods for daily volatility could be also applied for intraday volatility. – nan Jan 05 '17 at 08:18
  • How come this - http://cims.nyu.edu/~almgren/timeseries/notes7.pdf - is not specific? – LazyCat Jan 05 '17 at 14:54
  • I checked the document and it solely shows how to compute daily volatility out of intraday data. There is nothing about computing intraday volatility (e.g. 30min) from smaller sampled intraday prices(e.g. 1min). In addition, the presented methods are few, e.g. nothing about realized kernels. – nan Jan 05 '17 at 15:49
  • Don't you think, it's exactly the same: pick larger horizon, e.g. 30min, split it into smaller buckets, e.g. 1min, and use the formulas from the link. – LazyCat Jan 10 '17 at 19:42
  • The paper does say nothing how to estimate the parameters. E.g. if you determine the optimal sampling frequency you need to estimate the noise variance and a somehow noise free estimator of the RV(usually a sparsely sampled RV). The noise free estimator is usually based on 20 min data, see Barndorff-Nielsen et al. "Realised kernels in practice: Trades and quotes", however how can I estimate the optimal parameters on 20min data if I need 15min vol.... – nan Jan 11 '17 at 20:47
  • Sorry, I don't get notices on comments, hence sporadic responses. I still don't understand your problem. Of course, you don't estimate 15min volatility with the data sampled every 20min. Pick your interval, say 15min and use 1sec resolution to get the estimate. The links in the note have the formula for the variance of your estimate. Yes, you can ask if using 1min sampling is better than using 1sec sampling, but unless you go to a really fine scale, the finer the scale, the better is the estimate. To quantify it, you need the formula for the variance of your estimate. – LazyCat Feb 09 '17 at 20:53

1 Answers1

-1

Have a look at the following R package:

https://www.rdocumentation.org/packages/highfrequency/versions/0.4?

I believe it has all the tools you need and good references to papers and methodology.

cJc
  • 207
  • 2
  • 3
  • Thanks! I already know the package, the problem is that they just use spot volatility measures for intraday data. There is no realised approach for intraday data. The difference is that the spot volatility measures, even the non parametric approach proposed by Kristensen, use the information outside the interval to get an estimate of the volatility. I would like to leverage the information contained in the interval. – nan Jan 11 '17 at 20:14
  • Thanks for clarifying. Did you read the following post: http://quant.stackexchange.com/questions/20661/how-to-calculate-volatility-on-intraday-data – cJc Jan 12 '17 at 08:05
  • All of the provided answers are not microstructure noise robust. – nan Jan 12 '17 at 08:22