The econometrics module assembles the analyses that usually mean Stata or a shelf of R packages: interrupted ARIMA, structural breaks, panel models with Hausman tests, difference-in-differences, Granger causality and cointegration, limited dependent variables and GARCH volatility. Each is a tab, and each writes its tables and figures into the report.
The module covers the analyses an applied paper in economics or policy usually needs: whether an intervention changed a series (ARIMA with an exogenous intervention term, estimated per group), whether a series breaks and where (Bai-Perron, Chow, supF, CUSUM), how to model panel data (pooled, fixed or random effects, with the Hausman test), difference-in-differences, Granger causality and cointegration, probit/Tobit/quantile regression, and GARCH volatility.
The estimators follow the definitions of the canonical R packages rather than inventing conventions — the panel estimators match plm, the breaks match strucchange, cointegration follows urca, instrumental variables follow AER, and GARCH follows rugarch. Where a technique has a standard test attached (Hausman, Breusch-Pagan LM, Sargan, Wu-Hausman), the test is reported with the estimate.
It does not do everything Stata does. If your identification strategy depends on the full Stata ecosystem — margins, the many ivreg2 options, or a specific estimator you rely on — stay in Stata. Magic Stat's point is doing these standard analyses, with journal-ready figures and a report, from a spreadsheet and without writing xtreg or arima syntax.
Method availability: the optional packages arch, linearmodels and ruptures unlock advanced paths (GJR/EGARCH, clustered/robust panel SEs, PELT/Binseg break search). Without them, the module uses its own implementation, which is the default.
Panel: fixed or random effects with the Hausman decision, and standard errors clustered by unit. Breaks: minimum segment length and, optionally, a set of regressors.
No. It covers the common applied analyses — intervention ARIMA, breaks, panel, DiD, Granger, cointegration, probit/Tobit/quantile, 2SLS and GARCH — but not the whole Stata surface. If your identification relies on a specific Stata procedure, stay in Stata.
The implementations follow the same definitions as plm, strucchange, urca, AER and rugarch, and were built to reproduce their numbers rather than to introduce new conventions. Standard errors can still differ from another software's convention (for example, ARIMA with MA terms) — that is a convention difference, not a different model.
No. The module runs on the app's existing scientific stack. Installing the optional packages enables GJR/EGARCH with Student-t, clustered/robust panel standard errors and the PELT/Binseg break search; without them, the built-in implementations are used.
Yes — the panel tab runs DiD with a group×post interaction, reports the pre/post means by group and draws the parallel-trends figure.
No. The analysis runs locally. The only signal the app sends automatically is an anonymous installation counter — no data, no names, no e-mail.
Time series analysis → Linear regression → Robust regression →