VaR, Expected Shortfall, and Risk Gates
Read daily tail-risk estimates, distinguish return thresholds from loss magnitudes, and understand the limits of position sizing.
Category: Risk
Content updated: . Return and loss signs, the hypothetical example, and the scope of historical controls were revised. This is not independent model validation.
What question does VaR answer? Value at Risk describes a tail threshold for a stated horizon and confidence level. It is not the largest possible loss. Historical estimates describe the supplied sample; they do not guarantee future loss frequencies.
Read the sign convention first
Let r be a daily return. The return threshold is q = quantile(r, 0.05). Return-based ES95 = mean(r | r <= q). These values are usually negative. A loss convention reverses the signs: VaR_loss95 = -q and ES_loss95 = -ES_return95. The same tail can therefore appear as a negative return or a positive loss magnitude. Check the panel’s units and convention before comparing numbers.
Worked example: a hypothetical daily sample
Suppose the 5th-percentile return is -3%, and the average return at or below that threshold is -4.5%. The return-based display is VaR95 -3% and ES95 -4.5%. The corresponding loss magnitudes are 3% and 4.5%. These invented numbers illustrate the sign convention; they are not a current company estimate.
Can I add each position’s VaR? No. A weighted standalone tail estimate is not automatically a portfolio contribution. Portfolio risk depends on joint returns, correlations, weights, and the chosen model. Simple sums can misstate diversification and shared downside. Read the portfolio method separately from a company’s historical statistic.
Does a risk label mean an automated veto is active? No. This article explains risk measurements and historical control designs. It does not establish that an automated agent approves, rejects, or sizes a current position. Use each tool’s displayed availability, model, and source dates. Historical agent records are not current trading decisions.
What can the estimate miss? A calm sample can omit a future shock. Short samples, missing returns, changing correlations, and data errors can distort the tail. Compare the horizon, confidence level, observation window, and source date. Expected shortfall describes tail severity but does not remove these limits.