I want to find relation between 2 time series: a and b. I followed two ways:
1) Differentiated both 'a' and 'b' at lag 1 to remove trend (giving differenced time series 'ad' and 'bd') and then used xcorr in 'normalized' mode in MATLAB. (refer to figures)
2) I used corrplot (which in turn uses corrcoef): corrplot([a,b]) in matlab
Apparently corrcoef uses mean subtraction by default for detrending, while in xcorr I had to detrend the series manually by differentiating. (See: https://groups.google.com/forum/#!topic/comp.soft-sys.matlab/hFyEasUnKl8 )
Both the methods are conveying correlation cofficients at the end, but what can I understand from these 2 different outcomes?
The purpose of my analysis is two-fold:
1) To see if the time series have a relation or not
2) To use the correlation coefficients for feature selection in a multi-variate problem (several time series).
I would say that exhibiting a relation between your time series is much more complex than that. Using a first difference is good but you have clearly a seasonal pattern. It is quite normal that the two methods you describe give different results. Please contact me directly.
Guy
اخي العزيز
اذا كانت لديك سلسلتين فمن الواجب أولا التحقق من استقرارية السلسلتين عن طريق ما يعرف بـ معاملات الخلط لتتاكد من stationary for each first and second series .
after this work, you see differences .
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series