Track record

Did the quality grade separate companies in the past? Measured on 243 US companies from 2011 to 2023, using only the data available on each cut-off date. This page shows the result and, right next to it, what the test does not show.

Past performance is not a reliable indicator of future results.

The result

After 36 months the top quality third had a median return of +44.3 %, the middle third +25.9 %, the bottom third +12.7 %.

Swipe sideways for all columns.

Quality third	Companies	Median 36 months	Versus market	Above market
Top third	69	+44.3 %	+5.0 pp	61 %
Middle third	69	+25.9 %	-14.8 pp	30 %
Bottom third	69	+12.7 %	-28.0 pp	13 %

Median: the middle value when all are sorted by size. Versus market: distance from the average of all stocks on the same cut-off date. Above market: share of companies in the group that beat that average.

How it was computed

On 13 cut-off dates, each 30 April from 2011 to 2023, every company was graded under the rules on this site, using only reports filed with the SEC by then. Later corrections did not flow back.
The share price return over the following 36 months was measured. Together that makes 2,976 observations.
Each company counts once, not once per observation: the median of its quality total across all cut-off dates, provided there are at least 5. That applies to 207 of the 243 companies. Otherwise a single company would count up to thirteen times.
Computed with method version 2.6.0. The change to 2.7.0 only concerns how report dates are estimated, not the grades.

Why two numbers are needed

2011 to 2023 was mostly a rising market: all observations together had a median of +29.7 %. A group with a positive return is no achievement in that setting. That is why the distance to the market stands next to it.

This distance is a strict yardstick, for an arithmetic reason: the group is measured as a median, the market as an average. A few large winners lift the average above the median, in this data +41.0 % versus +29.7 %. That is why even good groups often fall below it. What carries meaning is the order of the groups, not the level of any single number.

What a stricter threshold does

Each row contains all companies from a given quality total upwards. The higher the threshold, the better the result and the smaller the selection. The number of companies is therefore in every row; a threshold that leaves only a few is correspondingly uncertain.

Swipe sideways for all columns.

Quality total	Companies	Median 36 months	Versus market	Above market
from 17 points	18	+53.0 %	+13.9 pp	67 %
from 16 points	28	+46.1 %	+9.9 pp	68 %
from 15 points	45	+43.6 %	+8.1 pp	64 %
from 14 points	80	+43.2 %	+4.0 pp	57 %
from 13 points	111	+38.1 %	-3.1 pp	47 %
from 12 points	142	+35.1 %	-5.1 pp	46 %
from 11 points	165	+31.8 %	-6.6 pp	41 %
from 10 points	179	+30.1 %	-8.8 pp	39 %
from 8 points	190	+29.6 %	-10.7 pp	37 %
from 6 points	201	+29.2 %	-12.1 pp	36 %
alle points	207	+27.8 %	-13.4 pp	35 %

The counter-check: what did not work

The same calculation for the distance between the price and the company’s own valuation range, split into five equal groups. Measured against the market, the cheapest shares did worst. The distance to the range is therefore not a buying order. A company trading far below its own history has usually fallen for a reason the multiple does not know. That is why valuation does not enter the total.

Swipe sideways for all columns.

Valuation fifth	Cases	Median 36 months	Versus market	Above market
1 of 5, most expensive	418	+29.2 %	-9.2 pp	43 %
2 of 5	415	+25.0 %	-10.3 pp	39 %
3 of 5	403	+29.3 %	-12.0 pp	40 %
4 of 5	392	+31.9 %	-14.5 pp	39 %
5 of 5, cheapest	373	+24.1 %	-18.0 pp	36 %

Does it hold without outliers?

Without the 15 best and 15 worst companies and with the thirds rebuilt: top third +4.3 percentage points versus the market, bottom third -27.1 percentage points. The gap remains; a few large winners do not carry it.

What this test does not show

A period, not a law. 2011 to 2023, large US companies, a phase in which quality and growth clearly beat cheaply valued shares.
Selection bias. The test universe consists of companies still listed today. Acquired companies are missing, which tends to lower the results; bankrupt ones are missing too, which tends to raise them. Which effect dominates cannot be told from this data.
No costs, no taxes, no dividends in the return measurement.
No out-of-sample test. The rules were developed on the same data they are measured on here. A test that only runs forward from a fixed date is still pending.
No statement about a single share. Group figures over 36 months say nothing about what an individual company does.