Semi-nonparametric cointegration testing

This paper considers a semi-nonparametric cointegration test. The test uses the LM-testing principle. The score function needed for the LM-test is estimated from the data using an expansion of the density around a Student t distribution. In this way, we capture both the possible fat-tailedness and the skewness of the innovation process. Model selection criteria are employed to select the appropriate order of the expansion in finite samples. Using a Monte-Carlo experiment, we show that the semi-nonparametric cointegration test has good size and power properties. The test outperforms previous testing procedures in terms of power over a broad class of distributions for the innovation process.