Comparing Time Series using Semblance Analysis

This program shows the correlation between two datasets as a function of both time and wavelength.

Only in very simple situations can we describe the correlation between two time series as a single number. Often it varies both as a function of time, and of wavelength. Datasets may be uncorrelated on short timescales due to noise, but strongly correlated on larger wavelengths (eg annual).

This program will compute (using wavelets) the correlation as a function of both time and wavelength. Anticorrelation (s=-1) is displayed in blue, zero correlation (s=0) in green, and positive correlation (s=+1) in red.

The image shows the correlation between the oil price and the gold price, which is mostly red (implying strong positive correlation) except for a period between 1985 and 1995.

Note that it is valid to compare datasets that have been measured in different units using this method.