Hi All,I would like your opinion on the following.We can create an Index following portfolio with co-integration. It is well documented in public domain. Or we can use orthogonal regression to create a index tracking portfolio. Which one you feel would provide reliable results? RegardsVK

Why not do both?Typically I have found orthogonal regression to be unstable at times so I stay clear of it. Also I don't think orthogonal performs that much better than OLS considering the extra complexity of it.

Can you provide a couple of links/cites to good papers or books on this topic that you mention are public?Thanks!

Hi,In fact i have both these methods implemented. I am in favor of cointegrated ECM model.ECM model using johansen method to find number of co-integrating time series is extention of Engle-granger procedure to multivariate world. However I read in few papers that Engle-granger for two time series is preferred over johansen procedure? I would like to know if Johansen procedure is less accurate compared to Engle granger? RegardsVK

Have you looked at the KPSS test?

Hi marine,Yes I did look at KPSS test. However the results are not conclusive with respect to the time series being trend stationary or difference stationary.It would be an better idea to indentify cointegrating time series first and perform OLS regression than performing ortho PCA on all the time series.regardsVK

Hi,One more question on Index Tracking...I was reading carol alexander - ISMA paper on "Cointegration and Asset allocation". It mentions that Optimum replicating portfolio is the one that give maximum stationarity in the residuals for a combination of best assets. It also mentions that in classical regression dependent and independent variables are assumed stationary and thus the residuals are stationary. Now i form a portfolio from assets which are individually cointegrated with the index. Then I OLS regress the portfolio against the index to find the coefficients. it this a better method for selecting weights? or i need to try Cointegrating regression to find optimum coefficients that give stationary residuals.regardsVK

to add to the previous post. In the OLS regress, the individual asset is tested for stationarity first and then tested for cointegration with the index.regardsVK

Hi,I tested stationarity of residuals of index against tracking portfolio with ADF and PHILLIPS Perron tests, which both reject unit root null and accept stationarity. But KPSS test rejects trend stationarity and accepts unit null. I am bit confused on the test results. RegardsVK