Some Time Series Questions

[danishussalam]

Dear all,

Would really appreciate if someone can answer:

I am working on a report the aim of which is to investigate the link b/w infrastructure and job creation. My data variables are roads (KM) , electricity generation (GHW) and gas generation. Whereas job creation variables include unemployment rate (%), employment in the transport sector (mill), construction sector (mill) and electricity + generation sector (mill).

1) When I plot roads against a time variable, it shows a clear upward trend. But when I use the dfuller test my computed t-stat is very high. High enough to reject the null hypothesis of stationary. How come it can be stationary when the graph shows a clear upward trend? I even tried taking log of the variable, t-stat becomes even higher.

2) If the variable is stationary at level, is it okay if we still take a first difference of it? (assume first difference is stationary too)

3) Iv'e heard about some detrending filters like hprescott. Is it okay to apply both that i.e first difference and a trend filter to make a variable stationary?

[nishantvats12]

Hi,

1. The trend that you might be seeing might be cause of the scale that you are taking to plot. It is just a guess and it is impossible to say anything for definite about the data without seeing it.

2. No, it is not advisable to take a difference of a stationary variable.