Choosing VaR Horizons, Confidence Levels, and Historical Windows
Summary
The document discusses common choices for Value-at-Risk calculations used in market risk management and regulation. One response describes a one-day horizon at a 99% confidence level, estimated from two years of historical data, while noting that shorter histories may be accepted in some settings. It also describes the traditional practice of scaling one-day VaR by the square root of ten to estimate a ten-day measure.
The answers caution that horizon and data-window choices should reflect the intended use and the portfolio's holding period. Regulatory capital or asset-liability management may use longer horizons and different confidence levels, while short-turnover strategies may require a different historical window. The discussion reports broad conventions, not a universal standard: scaling assumptions need justification, and the document does not compare VaR methods or provide backtesting evidence.
Key ideas
- A one-day horizon and 99% confidence level are described as common for market risk VaR.
- The discussion cites two years of historical data as a typical estimation window.
- Ten-day VaR has traditionally been estimated by multiplying one-day VaR by the square root of ten.
- The square-root scaling assumption may need justification, with direct horizon calculations as an alternative.
- The horizon and history length should reflect the use case and holding period.
Tags
Full text
# Which lags or percentiles should be run in a batch when calculating Value-at-Risk? # Which lags or percentiles should be run in a batch when calculating Value-at-Risk? Are there any "standard" VaR calculations run in a batch? For example, testing a VaR calculation with a lag of 1,2, 5 or 10 days over 2 years? Same question for the percentile, 1%, 2.5%, 5% etc. ## Answer by SpeedBoots (score 5, accepted) https://quant.stackexchange.com/a/1980 Standard (read: regulators will accept it) could be a one day, 99% VaR calculated with two years of historical data. A minimum of one year of history is needed although this is not the norm. Typically the one-day VaR is transformed into a 10-day VaR by scaling the calculation by sqrt(10). However, the new market risk rule governs that one justify their use of the square-root factor leaving the alternative of an actual 10-day VaR calculation (a lag of 10 days as you suggest). ## Answer by TheBridge (score 4) https://quant.stackexchange.com/a/1967 Usually when it is for (market) risk management purposes it is quite standard to have 1 day horizon with (allegedly ;-) ) 99% confidence level. As far as I know when it is for regulatory or economic capital requirement and/or Asset Liability Management then horizons might be much longer up to one year and confidence levels are usually 99% and 95%. Regards ## Answer by Michael WS (score 3) https://quant.stackexchange.com/a/2016 I think time length should very dependent on the holding period you are looking at.(This is at least how we handle) For example, if you turn your book every ten minutes, a 6 month time frame could be sufficient. If your holding periods are on a monthly basis, you will need much longer holding periods
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.