Interpreting Zero Value-at-Risk Breaches as Overestimation

Summary

The document explains how to interpret a period with no Value-at-Risk (VaR) violations. Under the unconditional coverage view, if observed returns never reach or fall below the forecast loss threshold, the model has produced fewer breaches than expected, which is evidence that risk may be overestimated. The answer confirms this basic interpretation.

The discussion also cautions that breach frequency is only one part of VaR evaluation. A model should also be checked for independence: violations should not cluster or otherwise show dependence over time. The answer points to Christoffersen’s interval forecast evaluation as relevant context. No data, formal test calculations, or confidence levels are supplied, so zero breaches alone does not establish definitive miscalibration; the conclusion depends on the sample and the VaR confidence level.

Key ideas

• Zero VaR violations can indicate that forecast risk is overstated under an unconditional coverage assessment.
• Breach frequency should be compared with the rate implied by the VaR confidence level.
• VaR evaluation should also assess whether violations are independent over time.
• A finite sample with no breaches is not, by itself, a complete model validation.

Tags

# Overestimating or underestimating risk?


# Overestimating or underestimating risk?












This question might be silly, but I want to be sure of myself.

If one has Value-at-Risk forecasts and there are zero VaR breaches (i.e. no return value is smaller than or equal to the VaR value) then the risk is said to be overestimated right?

So in a time series we would get zero observations for which holds $$r_t\leq -VaR_t.$$

In this particular case, I guess the risk is said to be overestimated. Hope someone can confirm this and if this case is called underestimation I would like to hear the reason for why this case is called underestimation.

Thanks

## Answer by Malick (score 5, accepted)

https://quant.stackexchange.com/a/27877

Yes, it is correct.





Please note that here we only focus on the so called “unconditional coverage hypothesis” (the frequency of VaR violations). You should also check that the VaR violations are independently distributed (the independence hypothesis) .

See Christoffersen, P. (1998). Evaluating interval forecasts. International Economic Review, 39(4), 841–862. Retrieved from http://www.jstor.org/stable/2527341

see also here on ssrn

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.