stats_pca | Principal component analysis of several series (e.g. yields across maturities, asset returns): loadings of each column on each component, component scores per row as a series with a chart, eigenvalues with the share and cumulative share of variance explained (scree chart), and the Bai-Ng information criteria IC_p1-IC_p3 for the number of components (the minimum suggests how many to keep). | |
stats_factor_analysis | Exploratory factor analysis of the correlation matrix of several series (e.g. asset returns, macro indicators): rotated and unrotated loadings of each variable on n_factor latent factors, communalities (variance shared with the factors), uniquenesses (idiosyncratic variance), eigenvalues (principal axis) or loading standard errors (maximum likelihood), and the fitted correlation matrix. | |
stats_manova | Multivariate analysis of variance (multivariate linear model): whether each term of the formula moves several responses jointly, e.g. whether the returns of several assets differ across regimes or sectors. | |
stats_canonical_correlation | Canonical correlation analysis between two sets of variables (e.g. a basket of asset returns and a set of macro surprises): the canonical correlations, a Wilks’ lambda F test that each one and all smaller ones are zero, the overall multivariate tests (Wilks, Pillai, Hotelling-Lawley, Roy), and the canonical coefficients of each set. | |
stats_mv_mean_test | Hotelling’s T-squared test of a mean vector: one sample against hypothesised means (e.g. are the average returns of several assets jointly zero), or two independent groups against each other. | |
stats_mv_cov_test | Likelihood-ratio tests of the structure of a covariance matrix estimated from the rows (e.g. asset returns): equal to a hypothesised matrix, spherical, diagonal (no correlation), block diagonal (groups of assets uncorrelated with each other), or equal across groups (Box’s M, e.g. calm vs stressed regimes). | |
