my question relates to posts found at the following links: Dynamic conditional correlation in Stata and Estimating correlation with DCC GARCH which provide great detail on calculating and graphing DCC's in Stata.

I'm wondering why the initial correlation values when using this method are so low? If we consider the example provided, performing the analysis on a sub-set of this data beyond a specific date and graphing the corresponding DCC's yields the same general shape with initial values very low followed by a spike.

Essentially, I'm looking for an explanation of why the analysis yields two different results for correlation at the same specific time, when the starting point for the analysis is a little later.

I'm sorry if I'm not explaining my question very well, I'm sure there's a straightforward explanation that I'm just failing to grasp! Many thanks.