Value-at-Risk is a single question — how much could I lose, with what confidence, over what horizon — answered three quite different ways. For liquid books the choice is often academic. For private credit, where return series are short, smoothed and infrequently marked, the method you pick materially changes the number you report.
Parametric VaR: fast, transparent, assumption-heavy
The parametric approach assumes a distribution — usually normal — and scales a volatility estimate by a confidence z-score and the square root of the horizon. It is fast and explainable, which matters when a Chief Risk Officer has to defend a figure to a board. Its weakness is the assumption: credit returns are skewed and fat-tailed, and a normal model will understate the left tail precisely where it matters.
Historical VaR: lets the data speak — if you have enough
Historical simulation makes no distributional assumption; it re-prices the book across actual past moves. The catch for private credit is data. Quarterly marks and short histories produce too few observations for a stable 99% estimate, and appraisal smoothing suppresses measured volatility, flattering the result.
In illiquid books the honest answer is rarely one method — it is one method cross-checked against another.
Monte Carlo: flexible, and only as good as its model
Monte Carlo simulation generates thousands of scenarios from a model you specify — letting you build in fat tails, default intensities and correlation structures a normal model cannot capture. It is the most flexible route and the most demanding: every assumption is a modelling choice you must own and document.
Look past VaR to Expected Shortfall
Whichever method you choose, the 99% VaR tells you a threshold, not the severity beyond it. Expected Shortfall — the average loss in the tail — is the more honest companion for credit, and increasingly the supervisory preference. Report both.
A practical default
For most private-credit mandates we start parametric for daily monitoring, run a Monte Carlo overlay for the tail, and reconcile against whatever historical window the data can support. The discipline is not in choosing one road; it is in knowing why the three disagree.