The main purpose of value-at-risk (VaR) measures is to quantify potential losses under "normal" market conditions, where normal is defined by the confidence level, typically 99 percent. In principle, increasing the confidence level could uncover progressively larger but less likely losses. In practice, VaR measures based on recent historical data can fail to identify extreme unusual situations that could cause severe losses. This is why VaR methods should be supplemented by a regular program of stress testing. Stress testing is a non-statistical risk measure because it is not associated with a probability statement like VaR.

One other reason to stress test is that VaR measures typically use recent historical data. Stress testing, in contrast, considers situations that are absent from historical data or not well represented but nonetheless likely. Alternatively, stress tests are useful to identify states of the world where historical relationships break down, either temporarily or permanently.