Statistics Glossary → Jarque-Bera test
Jarque-Bera testconcept
A normality test that jointly measures how far the skewness and kurtosis of the data depart from the normal values (0 and 3).
The test statistic is JB = n/6 · [S² + (K - 3)²/4], where S is the sample skewness and K the sample kurtosis. Since S = 0 and K = 3 for a normal distribution, JB is close to zero. The null hypothesis is 'the data are normally distributed'; in large samples JB follows a chi-square distribution with 2 degrees of freedom, and normality is rejected if p < α.
The chi-square approximation is poor in small samples, where the test can have low power and an actual Type I error rate that differs from α. For small n, Shapiro-Wilk is usually a better choice. Because Jarque-Bera looks only at the third and fourth moments, it can miss distributions whose skewness and kurtosis resemble the normal but which differ in other ways. It is widely used in econometrics and time series work, especially to check the normality of model residuals.
An everyday example
An investor wants to know whether 500 days of stock returns are normally distributed. The returns have skewness -0.4 and kurtosis 6.2, meaning sharp down and up days are far more frequent than a normal distribution predicts. JB comes out at about 227 and normality is rejected with p < 0.001; a normal assumption in risk calculations would understate extreme losses.
In MerQur
One of the tests in the Normality Tests analysis. It is also reported in the residual diagnostics of STL seasonal decomposition and ARIMA forecasting, and among the OLS assumption tests of the SAR model.
Analyses where this concept appears
Normality TestsSAR (Spatial Autoregressive Lag)STL Seasonal DecompositionARIMA Forecast