Hi there,I have seen it suggested is that one cure for high multicollinearity is to take first differences of the data.Now usually when you have integrated processes, there is no relationship between the regression coefficients of the first differenced process and the coefficients of the original process (indeed OLS regression of the original process doesn't make sense in this situation).However in this case where I am first differencing to reduce multicollinearity, can I take the results of the first difference regression and use them to find the coefficients for the regression of the original variable? If so how would I do this?ThanksBaz

This sounds a bit strange to me... Generally people take first differences of the data not to remove multicollinearity but to remove stochastic trend from time series (and hopefully end up with stationary data so that you can apply your preferred statistical technique).The best way to remove collinearity is to isolate the time series that is redundant and remove it from your analysis. You can look at the test statistics on the individual parameters of the regression for example to do so.Alternatively, Carol Alexander describes using PCA to transform your data into an orthogonal data set.

Hi there,Thanks for that, but the problem with PCA is that you usually have to take first differences there too, so I am still stuck for a way to get from those first difference dynamics to the dynamics of the actual curve?