Principle components construction

[mad_academic]

Hi forum. I’m stuck, and in need of help! (If context helps we can use stocks.wf1 from within eviews...)

Background
stocks.wf1: 5 stocks.
- I can generate a principle components series, e.g. “pc1”.
- I can get the table output showing the loadings (i.e the weight/importance of each series in forming that PC).

Question
Taking these weights and the 5 stocks, I am unable to recreate the eviews generated pc1 manually (my series is different). I guess this is due to some prior transformation on the 5 stocks thus the weights are not weights directly applicable to the original data?

My aim
What I want to do is to take the loadings and be able to create “pc1” manually (e.g. in excel or with eviews programming). Can anyone help?

Many thanks!

[EViews Glenn]

If you are computing the pc from the correlation matrix, the data must be standardized prior to applying the loadings. As discussed in the manual...

Note that when computing scores using Equation (12.33), EViews will transform the to match the data used in the original computation. For example, the data will be scaled for analysis of correlation matrices, and partialing will remove means and any conditioning variables. Similarly, if the preliminary analysis involves Spearman rank-order correlations, the data are transformed to ranks prior to partialing. Scores may not be computed for dispersion matrices estimated using Kendall's tau.

If that's not what you are doing, I'm not certain where the discrepancy lies.

[mad_academic]

HI Glenn,

Thanks for the steer. I've looked at the help file you quoted. I think for my application, it seems I have to go another way.

The actual context I'm using PCs for forecasting. I've done 1 step ahead PC forecasting - straightforward. It is multi-step forecasting that is causing me the problems.

E.g. for each of the i=1-5 series you would have Yi(t)=c+PC1(t-1)+PC2(t-1)+ut

I want to forecast from the end of the in-sample period the next two observations. The first forecast is easy as PC data exists. For the second forecast I need the 5 forecasts from the first forecast in PC form. I was hoping (naively) to be able simply apply the insample PC weights to transform the 5 forecasts for the 1st forecast observation into a PC-observation. On reading the help file this does not seem doable.

It seems the way forward is to include this first forecast with the in-sample data and reestimate the PCs then get the second period forecast as a 1 step ahead forecast.

Many thanks for the reply.