Sometimes I have mentioned stock screens on the Berkshire board, and somebody asked me about how they have done in the many years since.
Here is a slight edit of my reply, which some might find interesting.

The most interesting failure is a screen called YLDEARNYEAR, invented in 2003. (not by me)
It outperformed the market through thick and thin for a dozen years after it was published, the sort of thing that gives a lot of confidence that there is something "real" going on.
Then it promptly turned around and started underperforming the market for the next 5-6 years, by a lot.
So, what happened?

The YLDEARNYEAR screen seeks firms with both high earnings yields and high dividend yields.
That population of stocks itself seems to go in and out of fashion, not necessarily on a cycle matching bull and bear markets.
To summarize my current view:
With hindsight, it appears that the screen magnifies (very successfully) the degree to which the dividend population is in vogue.

Some numbers:
To get considered by the screen, a stock has to be profitable and have a dividend, so we can look at that universe as a baseline.
In the first 11 years after the screen was published, the set of ALL moderately profitable firms with dividends outperformed the broad market by about 4.4%/year.
(for that I looked at all stocks in the Value Line 1700 database with P/E < 33 and any dividend yield, equally weighted)
Consequently, during that period, it was a very fertile hunting ground for any yield screen: there was a tail wind.

The YLDEARNYEAR screen outperformed the market by about 14.1%/year in this stretch. (5 fresh stocks bought each month, each held two months, for a 10 stock "dozens" portfolio).
But in the next 6.1 years, the set of all profitable dividend earners underperformed the S&P by -5.6%/year, and the screen underperformed by -17.1%/year.
To overgeneralize, at some point we all noticed this bad streak and gave up on the screen as a fabulously overtuned beast that happened to have a pretty long lucky streak after publication.

But dividend payers have come back into fashion again, outperforming the S&P by +5.4%/year in the last 3.25 years.
And, as you might now expect, the screen has come back into its own: it has outperformed the S&P by 17.7%/year in that same period.
Revenge of the coupons, indeed.

During the stretch that dividend payers were out of fashion, almost all humans using this screen would have thrown in the towel.
Yet, had one stuck with it, it has still outperformed overall since its invention 20 years ago.
The S&P returned 10.1%/year, the set of passably profitable dividend payers returned 11.5%/year, and the screen returned 14.3%/year.

Maybe it shouldn't be thrown out entirely.
One could potentially check to see if dividend earners are in or out of fashion lately, and use the screen only when the omens are good--tie situation doesn't change often. But it's hard to know in advance how that attempt would do long term.
e.g, check to see if no-div stocks have outperformed or underperformed hi-div stocks in the last 2 months. If so, don't use YEY, use something that picks no-div stocks, e.g. a growth screen.
This is just a very crude approximation of Zee's old aggressive/conservative signal.

Jim