QuarcHome
Model catalogueBack to the model index

EWMA (RiskMetrics)

Forecasts volatility and covariance with exponentially decaying weights on past observations.

How it works

EWMA estimates conditional variance or covariance as a weighted average with geometrically declining weights. A decay factor λ sets the contribution of the old estimate and 1 − λ weights the latest squared shock or cross-product. Recent observations matter most and the recursion is simple to update daily.

Worked example

With λ = 0.94, previous daily variance 0.0001 and a 3% residual, the next estimate is 0.94 × 0.0001 + 0.06 × 0.0009 = 0.000148. Volatility rises from 1% to about 1.22%. The same decay rule applied to each covariance entry requires aligned returns.

Validation and limitations

Test decay and initial conditions on held-out periods and handle missing observations consistently so the covariance remains coherent. The standard recursion lacks explicit mean reversion to a long-run variance. Compare GARCH and examine portfolio error on calm and volatile dates.

Research sources