NEWS

qardlr 1.1.2

qardl() and qardl_bic_select() stop with an error when y or X contains missing values; the message gives the number and position of the incomplete rows.
summary.qardl() prints, under the gamma(tau) table, that the covariance of gamma(tau) has rank one (Cho, Kim and Shin, 2015, Theorem 1(ii)) and that its t statistics equal the t statistic of rho(tau) in absolute value. The documentation of qardl() states the same.

Removed an empty item from the author list of qardlr-package.Rd, which caused an HTML validation NOTE ("trimming empty
"). No changes to code.

Standard errors and Wald tests now follow Cho, Kim and Shin (2015): the covariance of the long-run parameters is Theorem 2 (with (X'PX)^-1 and the error density estimated by a Gaussian kernel with the Bofinger bandwidth), the covariance of phi and gamma is Theorem 1 with the estimator of equation (13), and the Wald tests across quantiles use the joint covariances of Theorems 3 and 4. The previous version assumed independence across quantiles, computed the long-run standard errors by a delta method without covariance terms, and fell back to a fixed covariance of 0.01 when the quantile regression covariance failed.
Fixed the contemporaneous coefficients with several covariates and q > 1: the coefficient of x1 at lag 1 was reported as the coefficient of x2 at lag 0.
gamma is now the level coefficient gamma(tau) of Cho, Kim and Shin (2015), the sum of the coefficients on x_t, ..., x_{t-q+1}, so that beta = gamma / (1 - sum(phi)) as printed; the coefficients on x_t are in gamma0. Constancy tests for gamma are carried out covariate by covariate because its covariance has rank one at each quantile.
qardl_simulate(): the true long-run parameter is now gamma_true / (1 - sum(phi_true)), the value implied by the simulated process (the default beta_true of 1 was not the parameter of the DGP).
New outputs cov_joint and fhat.