Statistics Glossary → ARCH-LM test (conditional heteroscedasticity)
ARCH-LM test (conditional heteroscedasticity)concept
A Lagrange multiplier test of whether, in the residuals of a time series model, today's volatility depends on the volatility of past periods (an ARCH effect).
In series such as financial returns, calm periods tend to follow calm periods and turbulent ones follow turbulent ones; this 'volatility clustering' shows that the variance changes over time in a series whose mean looks constant. Engle's ARCH-LM test regresses the squared residuals on q of their own lags. If past squared residuals explain the current one, the variance is changing conditionally. The test statistic is n·R², which under the null hypothesis of no ARCH effect approximately follows a chi-square distribution with q degrees of freedom.
A significant result shows that the standard errors and forecast intervals of models assuming constant variance are unreliable and that conditional variance models such as GARCH are needed. After fitting a GARCH model, re-applying the test to the standardized residuals checks whether the model has captured the volatility structure adequately. The test assumes the mean equation is correctly specified; omitted structure in the mean or isolated large outliers can also produce a significant result.
An everyday example
Looking at a stock's daily returns, days with big falls are followed by days of large swings, and calm days by calm days. Applying the ARCH-LM test with 5 lags gives p < 0.001. This shows that the risk is not constant and that daily volatility can be forecast with a GARCH model.
In MerQur
In GARCH, the 'Ljung-Box lag and ARCH-LM lag' option runs Engle's ARCH-LM (n R²) test on the standardized residuals; in VAR, 'Diagnostic lags' also include the lag for the multivariate ARCH-LM test.
Analyses where this concept appears
VAR (Vector Autoregression)GARCH (Volatility Model)
Related concepts
Conditional variance (volatility) and GARCH ordersLjung-Box (portmanteau) test