Bug fix: the bound parameters were computed with the sample mean in place of the initial observation X_0. Cavaliere and Xu (2014, eq. 4.10 and Remark 4.1) use X_0 and show that the mean makes the estimators inconsistent.
Bug fix: the ADF statistics were taken from a regression with a constant on the raw series, and ADF-alpha omitted the division by alpha(1). They now come from the regression of the de-meaned series without deterministic terms, with ADF-alpha = T * pi / alpha(1), as in equation (3.7).
Bug fix: MZ-alpha and MZ-t omitted the -X_0^2 / T term of the numerator, so they did not share the limiting distribution of ADF-alpha and ADF-t.
Bug fix: the MSB p-value was taken in the wrong tail; MSB rejects for small values.
Bug fix: the Monte Carlo null distribution now follows Algorithm 1 (a random walk regulated at the estimated bounds, de-meaned before the functionals are formed); the previous version used mirror reflection and did not de-mean.
Bug fix: the MAIC lag selection omitted the tau_T(k) term of Ng and Perron (2001) and used varying samples; it now uses the full MAIC on a common sample.
All five statistics and both bound parameters agree to four decimals with the Stata command boundedur (SSC) on the same data.
boundedur 1.0.2
Corrected the DOI of Cavaliere and Xu (2014) to 10.1016/j.jeconom.2013.08.026 in DESCRIPTION, README and all R and Rd files.
boundedur 1.0.0
Initial CRAN release.
Features
boundedur(): Main function for bounded unit root tests
ADF-alpha and ADF-t tests
M-type tests (MZ-alpha, MZ-t, MSB)
Monte Carlo p-value computation
Support for one-sided and two-sided bounds
select_lag_maic(): MAIC lag selection (Ng & Perron, 2001)