On Wed, 23 May 2012, Sven Schreiber wrote:
Am 22.05.2012 23:42, schrieb Allin Cottrell:
> On Tue, 22 May 2012, Claudio Shikida wrote:
>> Here is a strange thing. I took the real GDP per capita (chain) from China
>> and USA (Penn World Tables, Gretl, Database). I used the whole sample
>> (1950-2004, if I am not wrong)
>> So I started a basic work with time series. I applied an ADF test in the
>> China's variable using constant and trend, with general-to-particular
>> approach in lags, starting with 10 lags.
>> Here is the result.
>> Dickey-Fuller test for CHN_RGDPCH
>> sample size 52
>> unit-root null hypothesis: a = 1
>> with constant and trend
>> model: (1-L)y = b0 + b1*t + (a-1)*y(-1) + e
>> 1st-order autocorrelation coeff. for e: 0.092
>> estimated value of (a - 1): 0.0677971
>> test statistic: tau_ct(1) = 7.65922
>> p-value 1
>> Look, why is this p-value equal "one"? Any tips?
> It's equal to (well, approximately equal to) 1 because this sample
> of 52 observations provides no evidence whatever against the null
> hypothesis that Chinese real GDP per capita is a unit-root process
> (allowing for a linear trend). That's not very surprising if you run
> an OLS regression of CHN_RGDPCH on a time trend and look at the
> residuals. They are anything but white noise; it's obvious to the
> eye they are non-stationary.
Follow-up: From the point estimate the process would be explosive (a>1).
That's not really covered by the one-sided unit-root test.
Yes, I should have mentioned that the coefficient is on the wrong
side of zero to count against the null.
But what puzzles me is the large absolute value of the test
statistic -- what is driving this is unclear to me, presuming that
the precision of the estimate is not too high with 52 obs.
Fair enough, but it's just OLS and is easily verified. Chinese
growth over the sample period has been very emphatically faster than
ols d_CHN_RGDPCH 0 CHN_RGDPCH(-1) time