Handling Stale Risk Data and Holidays in Historical VaR

Summary

The document discusses two practical choices in historical simulation VaR: how to handle repeated zero returns caused by stale risk-factor observations, and whether to include holiday dates with no portfolio price change. It explains that stale observations should be assessed according to the factor’s liquidity and observability, since an apparent gap may reflect either unchanged values or missing market movement.

The response cautions that filling gaps with a smooth path can understate risk if the factor actually jumped, while treating a smoothly moving factor as having a jump can overstate risk. For a one-day VaR intended to represent a trading day, the answer recommends excluding holidays rather than counting a zero-P&L date. The discussion is brief and gives no specific imputation procedure or quantitative comparison, so the appropriate treatment of stale data remains dependent on the factor and the cause of the gap.

Key ideas

• Assess stale risk-factor data in light of the factor’s liquidity and observability.
• A missing observation may conceal either gradual movement or a genuine market gap.
• Smoothing a real jump can understate VaR, while assuming a jump in smoothly moving data can overstate it.
• Exclude holidays when calculating one-day VaR for trading days.

Tags

# Value at Risk Calculation


# Value at Risk Calculation












- In historical simulation VaR, if we have stale data for some of our risk factors, which means daily returns on a particular risk factor is 0 on multiple occasions. Is there any way to solve this other than changing the data source to get more updated data.

- In historical VaR,for time t=0,1,2, if t=1 is a holiday then the pnl of the portfolio at EOD t=1 will be 0 as there was no change between t=0 and t=1. Should we consider this 0 pnl in VaR calculation or directly compute pnl based on t=2 and t=0?

## Answer by Lliane (score 1, accepted)

https://quant.stackexchange.com/a/43142

- It depends what kind of parameter it is. Is it observable and liquid or not ? There could be true gaps in the data. If you have a gap in your data while the parameters moved in a linear manner in real life, you might overestimate your VaR. While conversly if you adjust your data in a linear way while there was an actual gap you might underestimate it.

- By convention you should exclude holidays from your calculation, if you're computing a 1 day var, it's for a trading day.

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.