Tail-Risk + Capital Modeling
Test your VaR model against extreme value theory, built from your own data, and hold the capital your risk actually requires.
The problem with a model built for calm markets
Many funds and trading desks still rely on a variance-covariance or lognormal VaR model, often built years ago and never tested against a fatter-tailed alternative. Models based on the normal distribution make extreme losses look nearly impossible. When markets stop behaving normally, the model is wrong in exactly the place where being wrong is most expensive.
The result runs in one of two directions. The firm holds more capital than its risk requires, which drags on returns. Or it carries tail exposure large enough to cause a loss the model said was close to impossible.
What we do differently
We benchmark your existing model against an extreme value theory (EVT) alternative, using your own loss and return data.
- Review your current model. Assumptions, data window, confidence levels, and how results are used in limits and capital decisions.
- Examine the tail. Threshold selection, diagnostics, and a Generalized Pareto fit to losses beyond the threshold. Where returns show volatility clustering, we use a GARCH-EVT approach.
- Compare. VaR and Expected Shortfall under your model and the EVT alternative, at several confidence levels.
- Backtest. Exceedance counts and standard coverage tests show which model has actually performed, rather than which one looks better on paper.
- Stress and scenario. Loss estimates under varied market conditions, including those outside your historical window.
- Quantify the capital difference. What your current model understates or overstates, in dollars.
What you receive
- VaR and Expected Shortfall estimates under multiple models and confidence levels
- Backtesting results and diagnostics, with plain-language interpretation
- Stress and scenario loss estimates
- A capital comparison between your current model and the EVT-informed view
- Model documentation written to withstand auditor, investor, and regulator review
- Working code, so your team can reproduce and extend the analysis
Who this serves
Hedge funds, proprietary trading firms, asset managers, and other trading institutions that lack in-house quant capacity to validate or rebuild tail-risk models, or that want an independent view alongside their internal one.
Typical engagements
| Engagement | Typical duration | What it covers |
|---|---|---|
| Tail risk validation | 3 to 4 weeks | Benchmark your existing VaR against an EVT alternative on your own data |
| Capital model build | 6 to 10 weeks | Documented EVT-based tail-risk and capital model, ready for review |
| Ongoing monitoring | Monthly or quarterly | Re-estimation, backtesting, and model governance support |
Questions we hear
Will you replace our risk system? No. This is analysis and model validation. Your system of record stays in place, and we provide the benchmark, the documentation, and the code.
What data do you need? Return or P&L history for the book or desk, and a description of your current methodology. More history improves tail estimates, and we are explicit about how little data the extreme tail really contains.
Is EVT always better? No. It is a tool for the tail specifically. We show where it differs from your current model and where the two agree, and we are open about the uncertainty in any tail estimate.
The outcome
You hold the capital your actual risk requires, not the capital a model built for a calmer market told you to hold.