plotirf() now plots up to three sets of impulse responses in the same panels. Each set is supplied as an array or as an estimator result (a list with irf and optionally se and boot) and is drawn without bands, with normal bands from standard errors, or with percentile bands computed from bootstrap draws and recentred on the point estimate (boot_type = "recentred"; "percentile" gives raw percentile bands). The default is a single 90% band (ci = 0.90); up to two confidence levels can be supplied. New arguments names, colors, boot_type and legend. plot2irf() is now a thin wrapper around plotirf() kept for backward compatibility. Fixed: two-dimensional IRFest with IRFse no longer errors.
svariv() returns the bootstrap draws (boot) also when details = FALSE.
svariv() now computes the reduced-form residual covariance on the full VAR sample, as in Mertens and Ravn (2013); only the 2SLS step uses the observations with non-missing proxies. With details = TRUE it reports, for each policy residual, the homoskedastic first-stage F statistic (FStat) and the robust effective F statistic of Montiel Olea and Pflueger
(FStatRobust, new argument cov_type). Identification failures now raise an informative error.
hetiv() and proxyiv() with details = TRUE now report weak-instrument diagnostics on the sample of every local projection: for E = 1, a data frame FirstStage with the homoskedastic first-stage F statistic (squared t statistic of the instrument) and the robust effective F statistic of Montiel Olea and Pflueger (2013) by horizon; for E > 1, a data frame WeakTest with the generalized Lewis and Mertens (2025) test from gweakivtest() by shock dimension (target), horizon and outcome variable. Both diagnostics always use the HAC (Newey-West) covariance estimator, irrespective of cov_type. WeakTest reports, for each shock dimension e, the test for the system with the first e policy variables and instruments, targeting the coefficient on policy variable e, with the relative bias criterion. hetiv() now also returns the OLS residual models in OLSRes, and proxyiv() computes Tiv on impact as documented.
Removed the Hstep argument from hetiv(), proxyiv() and recols(). All horizons 0, ..., H - 1 are always estimated. Previously, Hstep > 1 combined with hpredict > 1 selected the wrong horizon for the shock prediction.
New argument pooled in hetiv(), proxyiv() and recols(): with pooled = TRUE the shock prediction stacks the forecast errors of all horizons 0, ..., hpredict and uses them jointly (pooled prediction), instead of the single horizon hpredict. kfpredict() accepts Psi and et as arrays with one slice per horizon and stacks them, documents the pooled prediction, and warns when the residual covariance is singular. Control days now enter the horizon-h residual covariance only if no event falls in their window t + 1, ..., t + h. The columns of et and the elements of OLSRes are now named by variable and horizon (y1.h0, OLS.n1.h0, …).
Breaking: hpredict in hetiv(), proxyiv() and recols() is now a horizon on the scale of the row labels of irf (0, ..., H - 1, default 0 = impact) instead of a 1-based row index. Callers who passed a non-default value must subtract one. Psi equals irf[hpredict + 1, , ].
hetiv 0.1.2
New svariv(): proxy SVAR (SVAR-IV) following Mertens and Ravn (2013). Estimates a VAR(P) by OLS, identifies the first E shocks with external instruments, computes IRFs by recursion, and optionally returns wild-bootstrap percentile bands and standard deviations that work with plotirf() and plot2irf().
New recols(): recursive (Cholesky) local projections estimated by OLS, with the same interface and output as hetiv(), as a benchmark identification.
New arrangeirf(): arranges IRF panels from plotirf()/plot2irf() into one figure per shock and optionally saves them as PDF. Adds gridExtra to Imports.
Various smaller improvements and additional options in hetiv() and proxyiv()
hetiv 0.1.1
Changes to default HC settings for local projections
hetiv 0.1.0
Initial CRAN-ready release.
Provides heteroskedasticity-based and proxy-based instrumental variable local projection estimators for event-study settings.
Includes impulse response plotting helpers, weak instrument testing, Kalman-filter shock extraction, and simulation utilities.
Adds an introductory vignette with a worked end-to-end example.
