optimizer
Entropic Value-at-Risk (EVaR)
A coherent tail risk measure built from exponential loss moments. At the same confidence level it is never smaller than CVaR or VaR.
Entropic Value-at-Risk (EVaR) is a tail risk measure built from exponential moments of the loss distribution. At the same confidence level it is never smaller than VaR or CVaR, so it gives the most cautious reading of the three. Amir Ahmadi-Javid introduced it in 2012.
EVaR builds on the moment generating function, which summarizes a whole return distribution, and on the Chernoff bound from probability theory. For a chosen confidence level, it is the tightest upper bound on VaR and CVaR that the Chernoff inequality can deliver, found by searching over a scaling parameter that controls how heavily large losses are weighted. It is finite only when the loss has finite exponential moments, which some very heavy-tailed distributions lack. Under a normal distribution, 95% EVaR equals the mean loss plus about 2.45 standard deviations, compared with 2.06 for CVaR and 1.645 for VaR. Like CVaR it is coherent and convex, so an optimizer can minimize it efficiently.
Lower is safer. VaR is never above CVaR, and CVaR is never above EVaR, so EVaR is the most cautious of the three. The gap between CVaR and EVaR reflects how much weight the far tail carries, though its size depends on the whole distribution and does not grow in the same way for every fat-tailed case. Because EVaR draws on the whole distribution, a few extreme days in a history can dominate an estimate, which makes it more sensitive to the sample than CVaR. It is also the newest of the three, with a shorter record in practice.
Example with illustrative numbers: a portfolio's daily returns are normally distributed with a 1% standard deviation and an average near zero. At 95% confidence its one-day VaR is about 1.64%, its CVaR about 2.06% and its EVaR about 2.45%. With fat-tailed returns the three numbers shift by different amounts, so compare them on your own data.
On GNG, the Portfolio Optimizer offers an Entropic VaR engine in its tail-risk group, next to Conditional VaR and Conditional Drawdown. It minimizes EVaR at the 95% level across the historical daily returns in your lookback window, with your caps, sector limits, yield floor and turnover limit built into the problem; if the solver fails, GNG falls back to inverse-volatility weights and flags the run. For universes of 1,500 names or more, it first narrows the list to a normal budget of 500 names: your current holdings and any names covered by your holding rules stay in, the remaining slots go to the highest expected returns, and the results say so. GNG describes it as stricter than CVaR and best for fat-tailed return distributions.
Conditional Value-at-Risk (CVaR) is the less strict neighbor, and Conditional Drawdown-at-Risk (CDaR) targets drawdowns instead of single-day losses.
More optimizer terms
Get full access to GNG Research
Create a free account to access portfolio tracking, advanced tools, and more.