So if one were to look at your data, what would the value of X be?
Perhaps the value that eliminates the bottom 10% by ROE? Do you know
what this value was on average in your test?

Looks like it's generally the negative numbers, which are more common that I suspected.
Fraction of population of 1700 Value Line stocks by ROE since Jan 1997.

10.2% negative
 4.6% zero or unpopulated
 1.1% from  0 to  1
 1.4% from  1 to  2
 1.6% from  2 to  3
 1.9% from  3 to  4
 2.3% from  4 to  5
 2.6% from  5 to  6
 3.3% from  6 to  7
 3.5% from  7 to  8
 4.1% from  8 to  9
 4.6% from  9 to 10
 4.8% from 10 to 11
 4.9% from 11 to 12
 5.1% from 12 to 13
 4.6% from 13 to 14
 4.2% from 14 to 15
 4.0% from 15 to 16
 3.4% from 16 to 17
 3.0% from 17 to 18
 2.9% from 18 to 19
 2.5% from 19 to 20
 2.0% from 20 to 21
 1.8% from 21 to 22
 1.7% from 22 to 23
 1.5% from 23 to 24
 1.2% from 24 to 25
 1.0% from 25 to 26
 0.9% from 26 to 27
 0.8% from 27 to 28
 0.7% from 28 to 29
 0.6% from 29 to 30
 0.6% from 30 to 31
 0.4% from 31 to 32
 0.4% from 32 to 33
 0.4% from 33 to 34
 0.4% from 34 to 35
 0.3% from 35 to 36
 0.3% from 36 to 37
 0.2% from 37 to 38
 0.2% from 38 to 39
 0.2% from 39 to 40
 3.9% over 40

To recap the returns
Negative ROE = 6.11%/year
Zero or unknown ROE = 10.56%/year
Lower half of positive ROE = 11.52%/year
Top half of positive ROE = 12.77%/year

Bad negative numbers aren't worse than small negative numbers.
The lowest returns are simply the subset "<0".
If you wanted the highest return 30 stock portfolio given only the ROE
field you'd choose the 30 stocks with ROE closest to 32%, which gives
15.6%/year before trading costs, perhaps a bit of a lucky outlier.
I guess there are a few absurdly high ROE values that are a bad omen?
A monthly-constituted portfolio of all stocks 35% ROE or better returned 12.6%.

Jim