stats_heteroskedasticity_test | Tests a regression’s residuals for heteroskedasticity (error variance that changes with the regressors or over time): Breusch-Pagan (Koenker), White (squares and cross products), Goldfeld-Quandt (variance in two sub-samples) and Engle’s ARCH LM test (volatility clustering in time series). | |
stats_residual_autocorrelation_test | Tests a regression’s residuals for autocorrelation: Durbin-Watson (first order. | |
stats_vif | Multicollinearity check: the variance inflation factor of every regressor (how much its coefficient variance is inflated by correlation with the other regressors) and its tolerance (1/VIF), with the design’s condition number. | |
stats_influence | Finds observations that drive a regression: leverage (hat values), Cook’s distance, DFFITS and externally studentized residuals for every row, the Bonferroni outlier test, and the top rows by Cook’s distance with flags (Cook’s > 4/n, leverage > 2k/n, |DFFITS| above its threshold, Bonferroni p < alpha). | |
stats_nearest_correlation | Repairs a correlation or covariance matrix that is not positive semi-definite (from pairwise estimates, stressed or hand-edited correlations) so it can be used in risk models, simulation and portfolio optimization: the nearest valid matrix by iterative projection, or eigenvalue clipping. | |
stats_specification_test | Tests whether a linear regression’s functional form is right: Ramsey’s RESET (added powers of the fitted values or regressors), the rainbow test (fit on the central rows versus all), Harvey-Collier (mean of recursive residuals), White’s specification test (heteroskedasticity or misspecification) and the LM test for squared regressors. | |
stats_structural_break_test | Tests a regression’s coefficients for stability over the row order (time): the CUSUM test on OLS residuals (Ploberger-Kramer), Hansen’s joint stability test of coefficients and variance, and the recursive-residual CUSUM path with its Brown-Durbin-Evans bands (series and chart. | |
stats_nonnested_test | Compares two non-nested linear models of the same response (neither is a special case of the other, e.g. two competing factor sets): the Cox test, the Davidson-MacKinnon J test and the encompassing F test, each in both directions. | |
