Financial risk models rely on historical data and assumed distributions. When estimating Value at Risk the parameters that govern volatility, correlations, and tail behavior determine the reported capital metrics. Empirical research and regulatory practice show that errors in those parameter estimates do not remain innocuous under stress: they amplify, sometimes materially, the difference between reported VaR and actual losses.
Why estimation errors amplify under stress
Jon Danielsson London School of Economics has documented that VaR is particularly sensitive to assumptions about tails and dependence. Small misestimation of tail thickness or correlation can produce large shifts in extreme quantiles because VaR is a function of the far tail of the loss distribution. Paul Embrechts ETH Zurich has emphasized that tail modeling faces severe data scarcity, so parameter uncertainty is intrinsic and hard to reduce even with sound methodology. The Basel Committee on Banking Supervision introduced stressed Value-at-Risk precisely because ordinary VaR calculations tended to underrepresent risk during periods of market turmoil, when parameters move away from their historical averages.
Evidence and mechanisms
Parameter errors matter for two linked reasons. First, tail parameters have high estimation variance because extreme events are rare, so fitted models may understate tail risk. Second, stress events often change the underlying dynamics: correlations increase and volatility regimes shift, making historical parameter estimates stale. Together these effects create model risk where the model’s point estimate of VaR understates both expected shortfall and the uncertainty around it. Backtesting studies and regulatory reviews repeatedly find that institutions relying solely on point-estimated VaR experienced larger realized losses during crises than predicted.
Practical consequences and contextual nuances
For firms this means potential undercapitalization and unexpected margin calls. For regulators the consequence is systemic: many institutions using similar models and data amplify common underestimation, increasing systemic vulnerability. In emerging markets and territories with thin markets the problem is worse because data limitations raise parameter error. Environmental shocks such as major climate events can invalidate historical parameter stability, and cultural differences in risk governance affect how aggressively institutions account for parameter uncertainty. Mitigations include stress testing, conservative parameter choices, Bayesian or robust estimation that account for parameter uncertainty, and regulatory requirements like stressed VaR to force explicit consideration of model fragility. Even then, residual uncertainty remains and must be managed as a feature of risk governance rather than a numerical nuisance.