I already posted a related question a while ago but was unsure if I should post within the same question.
I want to estimate mulitperiod asset return correlations and test if there are significant deviations from the multiperiod correlatio matrices to their one-period counterpart. I use overlapping data to increase the number of observations. However I'm aware that this can induce problems.
It is known that overlapping data is problematic for OLS since it induces autocorrelation in the error terms. OLS assumes homoscedasticity and absence of autocorrelation. In this case, for example Newey West (1987) provide a correction technique.
However, I dont want to perform regression analysis but estimate the pairwise correlation coefficients. To what extend would this approach be affected by the use of overlapping returns. Do I need a correction method like Newey West? Correlation after all is a scale free measure.
Secondly,how would I test the significance of the estimated correlation coefficients? Jennrich (1970) proposes a method based on a chi square distribution to test wether to correlation matrices are identical, but this requires independent observations.
Hope you can help me with this since it has been bothering me for a while now.
Thanks in advance!