TSMT Change Log

The following is a list of changes from the previous version of GAUSS.

4.0.0

  1. New function: svarFit() estimates structural vector autoregressive models. Supports zero short-run restrictions, zero long-run restrictions, and sign restrictions for structural identification.
  2. New function: plotIrf() plots impulse response functions after estimating SVAR models using svarFit().
  3. New function: getFEVD() computes forecast error variance decomposition using IRFs after svarFit().
  4. New function: plotFEVD() generates area plot of forecast error variance decompositions after svarFit().
  5. New function: plotHD() generates stacked bar plot of historical decompositions after svarFit().
  6. New function: sbreakFit() estimates m-break structural break models. Includes improvements over sbreak(), such as formula string inputs, user-configurable settings, dataframe metadata support, and structured output formatting.
  7. New function: getP0() computes prior covariance initialization for use with kalmanFilter(). Supports both diffuse and stationary initialization methods, with automatic root modulus inspection to determine the appropriate approach.
  8. New standardized output printing across all models, including a header with model details and summary evaluation statistics.
  9. Enhancement: Added tsmtModelDesc() structure to store model metadata such as dependent variable name, time span, number of observations, and degrees of freedom.
  10. Enhancement: Added tsmtSummaryStats() structure to hold model diagnostics including SSE, MSE, RMSE, SEE, R-squared, Adjusted R-squared, SSY, and the Durbin-Watson statistic.
  11. Enhancement: All modeling functions now compute and return summary statistics (SSE, MSE, RMSE, SEE, R-squared, Adjusted R-squared, SSY, Durbin-Watson).
  12. Expanded functionality: Models now accept time series date variables for dependent and independent variables. Date ranges are automatically detected and reported.
  13. Expanded functionality: Standard errors are now reported for the constant term in the arimaFit() model.
  14. Enhancement: autoregmt() now accepts a scalar input to specify the same lag length for all independent variables.
  15. Enhancement: Unit root test outputs have been updated to include null hypotheses in the header and test conclusions in the footer (where applicable).
  16. Enhancement: autoregmt() now checks for redundant constants or non-varying variables in the independent variable matrix.
  17. Enhancement: selectLags() now accepts optional arguments p_max, method, and printout, all with sensible internal defaults.
  18. Enhancement: arimaSS() now accepts optional arguments for p, d, q, constant, and trend, with documented default values.
  19. Enhancement: Improved stationarity and invertibility enforcement in arimaSS() using a tanh transformation approach. This method improves numerical stability, supports higher-order AR and MA terms, and enhances convergence behavior.
  20. Enhancement: The time trend component for arimaSS() is now centered and scaled for improved numerical conditioning.
  21. Enhancement: Special handling added to arimaSS() for the case with no ARMA terms. In these models, MLE is skipped and closed-form OLS estimates with valid standard errors are returned.
  22. Enhancement: New improved starting values implemented for arimaSS() using a naive regression-based approach.
  23. Improved covariance estimation in arimaSS(): Implemented the delta method (a Jacobian-adjusted sandwich estimator) to compute standard errors that properly account for parameter transformations used to enforce stationarity and invertibility.
  24. Enhancement: arimaSS() covariance computation now falls back to a pseudo-inverse when the Hessian is singular or near-singular.
  25. Enhancement: arimaPredict() now checks for metadata and prints dates and variable names.
  26. Enhancement: arimaPredict() now supports optional graph generation.
  27. Bug fix: kalmanFilter() previously mishandled trend components. This has been corrected.
  28. Bug fix: arimaSS() now properly supports models with no AR or MA terms.