There is no such thing as a perfect estimator and Value at Risk methodologies are not without limitations. And different VaR approaches each have their own advantages and drawbacks. Generally, Value at Risk calculations are based on estimations that rely on fitting the distribution of returns to a statistical distribution. These are referred to as “parametric approaches.” The problem with parametric approaches is that real life returns do not always match a theoretical distribution very well, so estimations inherently have a form of model risk due to the difference between the theoretical and actual distribution sets. The most common distribution type, the normal distribution, has a mean of 0 and a standard deviation of 1 and is used often in financial theory. But returns in real life rarely fit the normal distribution, and have been shown to demonstrate ‘fat tails’ (returns far away from the mean value). The variation of real world returns requires risk practitioners to match the real world distribution of returns to a theoretical distribution type in order to make statistical references.
CryptoDataDownload avoids the drawbacks of parametric approaches to VaR in its Historical VaR and Expected Shortfall report. Historical VaR, a non-parametric approach, estimates the Value at Risk by ranking the known distribution of returns for the data set, thereby eliminating the need to make assumptions about the distribution of returns. And although this removes the need to make assumptions, it has its own unique drawbacks. For instance, VaR is limited to the worst known returns that have occurred. This means that it would be impossible to calculate a Value at Risk metric that is worse than the worst return in the data population. If the return distribution is not very large, the Historical VaR calculation will likely severely underestimate the true Value at Risk. This drawback is especially an issue if the data set is relatively small. A newly created cryptocurrency, for instance, would not have any previous history as it did not exist. This hinders statistical estimations and investors should always be wary of a full loss of investment capital.
It is important to remember that regardless of the VaR estimation technique used ... VaR is still only an estimation about the potential loss.
There are improvements that can be made on the Value at Risk metric accuracy by using simulations and filtering techniques; but it is important to remember that regardless of the VaR estimation technique used, parametric or non-parametric, VaR is still only an estimation about the potential loss.