Value at Risk (VaR) answers a specific question: what is the maximum loss you can expect over a given time horizon at a given confidence level? A one-day 95% VaR of $100,000 means that on 95 out of 100 days, your losses should not exceed $100,000. It says nothing about what happens on the other 5 days.
That silence about tail events is VaR's most important limitation. A portfolio can have a low VaR and still be exposed to catastrophic losses. The 5% of days that exceed the VaR threshold can include losses that are 2x, 5x, or 10x the VaR amount. Conditional VaR (also called Expected Shortfall) addresses this by averaging the losses that exceed the VaR threshold, giving you information about the severity of tail events rather than just their probability.
There are three main approaches to calculating VaR. Parametric VaR assumes returns follow a normal distribution and uses mean and standard deviation to calculate the threshold. Historical simulation uses actual past returns and identifies the relevant percentile. Monte Carlo simulation generates thousands of random scenarios based on assumed distributions and correlations. Each method has trade-offs between computational simplicity and accuracy.
The normality assumption in parametric VaR is particularly problematic for crypto. Crypto returns exhibit significant skewness and kurtosis (fat tails and asymmetry), meaning extreme events occur far more frequently than a normal distribution would predict. A 99% VaR calculated assuming normality can dramatically underestimate the actual frequency and magnitude of large losses in crypto portfolios.
Historical simulation seems more intuitive because it uses actual past data, but it has its own problems. It assumes the future will look like the past, which is a particularly poor assumption during regime changes. If your historical window does not include a crisis, your VaR will not reflect crisis-level losses. If your window is dominated by a crisis, your VaR will be overly conservative during calmer periods.
VaR aggregation across a portfolio depends on the correlation structure between positions. When correlations are stable, portfolio VaR is lower than the sum of individual position VaRs (reflecting diversification benefit). When correlations spike toward 1 during crises, the diversification benefit evaporates, and portfolio VaR can increase dramatically even without any change in individual positions.
Despite its limitations, VaR remains useful as a communication tool and a basis for risk limits. Setting a maximum VaR limit for a portfolio creates a constraint that forces position sizing discipline. When VaR approaches the limit, positions must be reduced, which imposes systematic risk management that prevents large concentrated bets.
The practical recommendation is to use VaR as one input among several rather than as a standalone risk measure. Complement it with maximum drawdown analysis, stress testing against specific scenarios, and Expected Shortfall calculations. No single risk metric captures all dimensions of portfolio risk, and over-relying on any one number creates blind spots that tend to matter most during exactly the conditions you are trying to protect against.