I've got a datafile which has bond returns (30, 20, 10, 5, 1 year) and time. The bond variables are returns on the different time bonds. I need to perform PCA and EFA on the data. I tried using time as one of the factors but it returns Heywood case as the uniqueness is 0. Should I disregard time? What i don't understand is if I disregard it then my analysis will have no practical interpretation. Already I've two principal components using PCA and I'm unable to interpret them. Please help me in understanding the practical implications of the data. Thanks in advance for your time and kind consideration... P.S. I'm working in Stata 12.1. Solutions in SPSS will be helpful too but I won't be able to perform them as my SPSS 20 does not accept datafile from this version of stata.
Exclude time and re-run it. For each eigenvector that you're using to build the factors, plot it to get a better sense of what they mean. There is a significant literature on a PCA of bond yields (rather than bond returns since bond returns are driven by yield curve dynamics), which normally describes the first component as corresponding to the interest rate level, the second corresponding to a shift in the slope of the yield curve, and the third corresponding to a twist in the yield curve. My guess is that the factors constructed from the returns correspond to the impact on bond returns from those changes.
I agree with you regarding the dependence of bond returns on interest rates and yield curve but I think these components are not enough to answer the question as the PCA and EFA tend to give different analyses so to base both the answers on the same parameters will be a bit faulty. Other factors that might influence it are inflation, credit rating, portfolio diversification methodology etc. Courtesy- Copeland, Weston and Shastri's "Financial theory and Corporate policy"