For example, in a year oil price increase with dollar price during few days, few of days they don't correlate yet all and few days they negative correlation. That means, my two time series have different behavior in time.If i use simple correlation coefficient then i will end up with over all relation between them. But i want to know which days it is correlated and which days it is not correlated.
You need structural breaks and cubic spline to augment your observations to be of a certain n>30 definitely. Then go ahead and use your datasets. It sounds like a lot of work if you plan to do this for a great number of years/days though.
Hey Kishore,
you can compute the correlation coefficient over a moving window. E.g.:
c(i) = corr(s1(i:i+n),s2(i:i+n)); % For time point i
c(i+o) = corr(s1(i+o:i+o+n),s2(i+o:i+o+n)); % For time point i+o
However, this is rather slow, as the corr. coeff needs a lot of multiplications.
Greetings, David
Dear Kishore Kumar Grandhi,
You can use the correlator to solve your problem. In this case the part of the time series (for example, dollar price) to be regarded as a reference signal.
@David Jule Mack,
What about window length?
The window length also influence the results. How to overcome this?
Choose it at the time scale you expect to be correlated (several days). Since you are using the same window size, the results are comparable.
- Badges
- Science topic
- Similar topics
- Statistics
- Time Series