0

enter image description here

Given that my variables exhibit non-stationary (i.e lrgdp and lop) and I intend on estimating a VAR model, would it make sense to correlate them in their first differences instead (dlrgdp and dlop) and how does one then interpret the results? For instance, 5 lags of dlrgdp would mean 6 lags of lrgdp?

Shan
  • 41
  • 4
  • 1
    In brief, yes. Correlation between non-stationary series can be problematic. See here for more: https://stats.stackexchange.com/questions/133155/how-to-use-pearson-correlation-correctly-with-time-series/133171#comment730305_133171 What is the reason for doing this correlation analysis? How does it relate to your intention to build a VAR model? – ColorStatistics Jan 26 '19 at 17:00
  • If your original variables are measured each year, say, then their first-order differenced versions represent year-to-year changes in the values of the original variables. – Isabella Ghement Jan 26 '19 at 18:22
  • This is because I want to help characterise temporal sequence of influences in VAR and I'm trying to figure this out through cross-correlations and the granger causality analysis. – Shan Jan 26 '19 at 18:48

0 Answers0