Okay, Here's where we are:

• We considered a bunch of stocks, namely the DOW 30.
• We constructed a portfolio, devoting fractions x₁, x₂, ... x₃₀ to each stock in our "initial" portfolio.
  (In what follows, we'll devote $x₁, $x₂, ... $x₃₀ so our total portolio is $1,000.)
• We follow this portfolio (without rebalancing).
• We calculate the Variance of this portfolio.
• We modify the allocations in order to maximize the variance of this portfolio.
• This involves determining the principal eigenvector of the covariance matrix.
• We then stare intently at the principal eigenvector - which puts the 30 stocks in order (from top rank to bottom rank).
• We consider just the top 4 in this Principal Component ranking and ask:
    "How well does that top 4 perform, compared to the DOW 30?"

>Pretty darn good, I'd say! And you're convinced by just one or two examples? C'mon! Let's pretend it's January, 2000. We look at the PCA top 4 as determined by the monthly data for the previous eight years. We get Figure 1: the growth of a $1K portfolio made up of the PCA top 4 and the DOW30. The top 4 beat the DOW by a mile, eh? Aah, but what about the past eight years, from Jan,1998 to Jan, 2006? >But that includes some of the 1992-1999 data! So? I'd like to see the transition from the pre-2000 performance to the post-2000 performance so I overlap and ... >Can you just go on! Okay ... we get Figure 2. It shows the same allocation as we used previously, but for eight years extending to year Jan, 2006. >It still beats the DOW by a mile!

Figure 1 Figure 2

>And were they such good performers up to Jan, 2006?
Here they are:

>Mamma mia! Lots of lousy stocks there! So, how about the bottom 4?
We get this, for the Bottom 4 based upon the 1992-1999 data:

PCA Bottom 4 (1998-2005)

... continued to 1998-2005

The Bottom 4 are still pretty good over the years 1992-1999, where PCA determined the ranking ... but not so good in the later years.

>But it still beat the DOW! Okay, what were the top and bottom 4, for ...?
For the eight years ending in Jan, 2006? They were like so:

PCA Top 4 (1998-2005)

PCA Bottom 4 (1998-2005)

>Have you noticed that AXP and C and UTX were also in the top 4 for 1992-1999?
Well, there's some overlap in years ... but it is a wee bit surprising, eh?
>Have you noticed that bottom 4 for 1992-1999 didn't hang around for 1998-2005?
Interesting, eh?

Here's another one where, again, we consider some 8 year period, determine the PCA top 4 allocation, then just buy-and-hold to see what happens with the same dollars devoted to that top 4 ... but over the next 8 years.

PCA Top 4: based upon Jan, 1990 - Jan, 1998 data

Same 4, but continued to Jan, 2006

Although these Top 4 (based upon 1990 -1997 inclusive) looked good, even over the next eight years, the first few years would have been pretty scary for an investor

The first few years of the 1990-1997 "Top 4"

Of course, in looking at the current list of DOW stocks, some weren't around ten or twenty years ago ... so the list of stocks in the spreadsheet PCA-DOW2.xls should be modified to reflect the actual DOW list for the years being investigated.
>Or you could pick any 30 stocks, right?
Yeah, I guess you could ...
>But how would I go about plotting those Top 4 stocks, assuming I wanted ...
There's a spreadsheet which is similar to the earlier one, except it has a button to see the Top 4, like this:

You can download it by clicking on the picture.
It gives charts like we got above in, say, Figure 1.