Skip to content
All library documents

PRIIPs Category 2 Stress Volatility and Historical Data Windows

Article Quant Q&A · Author: Matthew
Read the original

Summary

The document discusses calculating stressed volatility for PRIIPs Category 2 products. The question describes forming subinterval returns relative to an average, squaring deviations, aggregating them over a window, taking a square root, and using a percentile in the stress calculation. The author is unsure whether to use the full historical dataset or only the latest year and cannot reproduce the regulator's published results exactly.

The response recommends the full dataset for stressed volatility and says the one-year horizon is treated differently from longer horizons. It also argues that the cited flow diagram contains errors and that the three- and five-year stressed volatilities should match under the referenced annex methodology. This is a forum answer pointing to regulatory material, not a complete worked calculation; the implementation details and claims should be checked against the applicable PRIIPs rules and source documents.

Key ideas

  • The question describes estimating stressed volatility from rolling subinterval return deviations.
  • The response recommends using the full available dataset to calculate stressed volatility.
  • It interprets the methodology as distinguishing one-year horizons from longer horizons.
  • Under that interpretation, three- and five-year stressed volatilities should be identical.
  • The response disputes parts of a flow diagram, so readers should verify the regulatory source.

Tags

VolatilityRisk managementStatistics
Full text
# PRIIPS Category 2 Stress Scenario Calc


# PRIIPS Category 2 Stress Scenario Calc












Good evening

This is where I am currently at with regards to calculating the sub interval details.

I have split the calculation into 2 steps, firstly I am taking the return for date x minus the average of date x to date y (for the RHP 1 year using 21 days) so for example, if my first return date was 20/09/2016, I would average the returns for 20/09/16-18/10/16 and power the result to 2 (ie (rti-M1)^2).

The second aspect is the wtiσs, which I have calculated as the Sum of the above results for 20/09/16-18/10/16 divided by the sub interval of 21 then square rooting that result.

Finally, I take the 99th percentile of the results.

However, using the ESMA guidelines and recently released flow diagram, I am unable to come up with the same results.

There appears to have been some discussion on whether or not you should use the latest years worth of data for the 1 year or if you should be using the full data set. Personally I believe it should be the full data set otherwise if the latest years data had lower volatility you wouldn't see the full effect of the stress over a bad years data set.

However, if this was the case, then why is the 3 and 5 year RHP stressed volatility different when they both use the same sub-interval and in theory the same full data-set.

Can anyone help me understand where I am going wrong and/or point me in the right direction please as I have managed to work everything else out and whilst I can get to within 4-5 decimal places of the ESMA results, the next step pushes the final figure for the stress scenario even further out (it works if I manually type in the figures for Wσs√N from ESMA) so I know that the last bit is not incorrect.

Apologies for the long question.

## Answer by Tim Wilding (score 1)

https://quant.stackexchange.com/a/37693

I think this question is dealt with partially here PRIIPs Stress Scenario for Category 2. I don't think you are going wrong and I believe you should use the full dataset to calculate the stressed volatilities.

The ESMA Flow Diagram ( https://esas-joint-committee.europa.eu/Publications/Technical%20Standards/JC%202017%2049%20(PRIIPs_flow_diagram_risk_reward).pdf) you refer to appears to have a number of errors in it. The stress scenarios for the 3 & 5 year period on page 24 appear to be incorrect because they should be identical. Annex 4 pt. 10 (https://ec.europa.eu/transparency/regdoc/rep/3/2017/EN/C-2017-1473-F1-EN-ANNEX-1-PART-1.PDF) is very clear that there is only a distinction between the stressed volatility for a 1 year horizon and other horizons. Hence, the 3 & 5 year stressed volatilities should be identical.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.