Hi again, I discovered that I had confused the significance level. It should be 5% significance in all cases, both bartlett and boot-strapped p-values (1896-1929 and 1982-2008).

Best,
Lars

Från: Lars Ahnland <lars.ahnland@outlook.com>
Skickat: den 28 oktober 2022 13:10
Till: gretl-users@gretlml.univpm.it <gretl-users@gretlml.univpm.it>
Ämne: [Gretl-users] Re: small Johansen test - minimum sample size
 
Hi Sven,

Thanks for the reply! Here are the excel files (datafile "6 + China 1896-2008"; and results). The datafile is a weigthed average of two variables for a number of countries. It is the sheet 1 I have used with Gretl. As you can see, I have used sub-sets of the time series, referring to specifiuc macroeconomic policy regimes, in 1896-1929, 1946-1973 and 1982-2008.

Best regards,

Lars

Från: Sven Schreiber <sven.schreiber@fu-berlin.de>
Skickat: den 27 oktober 2022 16:02
Till: gretl-users@gretlml.univpm.it <gretl-users@gretlml.univpm.it>
Ämne: [Gretl-users] Re: small Johansen test - minimum sample size
 
Am 27.10.2022 um 12:44 schrieb Lars Ahnland:
Hi,

I am using the package johansensmall, for johansen cointegration test with small sample sizes. I have a dataset with only 29 samples. Now, I get a significant reslut at the 5% level with the Bartlett correction, and at 1% level with the boot-strapped p-values. I report both values. Would you say these results are valid, considering the very small sample size?

Hi Lars,

indeed I'd say it's unusual that the bootstrap version rejects the null at a stricter level than the Bartlett correction. For such a small sample size I would typically expect that the bootstrap test is unable to reject. If you could share this particular dataset and your test specification (deterministics, lags), maybe we can have a better idea of what's going on.


cheers

sven