Hi!
I am trying to use a stationary non.financial corporate debt to GDP ratio on some level basis (differencing would not be possible) in R. How to make the ratio stationary? I tried STL and DECOMPOSE packages, but they de-trend the variable and it then loses all the significance in the models. I also tried subtracting various long-term moving averages (from 4Y to 10Y), but that does not help either. I need to use the level information of the ratio and specifically, I need to use period from 1967 - 1995 (selected as in sample) and 1995 - now (out of sample).
Thanks!
Michael
1. Why to make a debt ratio stationary? Any such transformation reduces the information content of a series.
2. It seems that detrending makes the variable stationary, but then there is no significant effect in your estimated model. This may be a hint that the model is wrong (in the sense that the debt ratio has no effect on the endogenous variables.
3. Obviously, you put too much weight on the method than on theory.
Hi Anton, Thanks for answering:
1. For a long time period, level stationarity would be necessarity since there is a clear exponential trend in debt to GDP. This would totally ruin the model. The coefficient of debt to GDP would not be correct.
2. STL and Decompose do not both make sense as they scrap the useful information, but long-term moving average subtraction does make sense as it is statistically significant in the model. According to graph of detrended debt to GDP with long term moving average, it seems only logical to retain this series. But for some reason, it does not pass the ADF, PP and KPSS tests for stationarity.
3. Yes, I need to use Linear OLS model and Vector Autoregression models where stationarity is a requirement. For the others (probit), it`s not such a big deal as far as I understand.
I think it's best is to find a variable that cointegrates with the debt to gdp ratio and then you can include them both in the model.
Have you tried unit root tests that take into account possible structural breaks in the timeseries?
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R