Krest
See it live →
← The overview
The Krest Test · The full teardown
We took it to Krest.

In a bull market, everything wins. So did the Coffee Can actually pick winners?

The overview showed a screen that beat the market handsomely across five bull market years. This is the harder question. When almost every stock is rising, a strategy can look brilliant while its actual stock picking does nothing. So we asked whether the ranking earned the result, or just rode the tide.

Educational6 min readPart two of two

The Coffee Can screen ranks companies by the quality of their business and buys the top of the list. The promise is that the highest ranked names are genuinely better and will earn more. In a rising market that is a hard promise to test, because nearly everything earns more. So we sorted every stock into ten buckets by the screen's own ranking and looked at what each bucket actually returned.

Everything won. The ranking added nothing.

Here is the tell. Every single bucket, from the highest ranked to the lowest, returned somewhere between thirty and forty eight percent a year. The stocks the screen rates worst still made a fortune. And the highest ranked bucket was not the best; it sat mid pack, below buckets it looks down on. When the losers make forty percent, the ranking is not sorting winners from losers. The tide is lifting every boat, and the screen is taking credit for the water.

Return by the screen's own ranking, best bucket to worst
Full run return a year · ten buckets · D10 = highest ranked ↗ See it live on Krest
Highest ranked bucketNifty 500
Every bucket beat the market, so the chart looks like a triumph. But a working ranking would still slope down from left to right. This one has no slope, and the top bucket is mid pack.

The metric it ranks on barely predicts anything

The screen's core sorting tool is return on capital. So we measured, across every stock and every date, how well a company's return on capital actually predicted the returns that followed. A useful signal would score well above zero. It scored below it: higher quality on this measure went with slightly lower subsequent returns. The screen's own rank barely cleared zero.

How well the ranking predicted future returns
Correlation with forward return · 0 = no predictive power
A number near 1 would mean a strong signal. Return on capital, the metric the whole screen ranks on, points the wrong way; the combined rank is a whisker above zero.

Trusting the ranking made it worse

If the ranking picked winners, holding only its very top names would pay most. It paid least. The broad basket and the top thirty both compounded near thirty four percent. Concentrate into the top ten and the return falls to twenty nine; constrain it further and it collapses to eighteen, well under the market. Even in a market this kind, leaning harder on the screen's best ideas cost you money.

Return by how tightly you concentrated
Full run return a year · vs Nifty 500
The screen's most concentrated, highest conviction versions did worst. Whatever earned the return, it was breadth and a rising market, not the sharpness of the pick.

How often did it beat the market?

On paper the base rates look strong, especially over five years. But hold the caveat in your mind as you read them: with only five years of history, there are just a handful of genuine windows here, and every one of them overlaps inside the same bull market. A high hit rate drawn from one good market is not the same as a high hit rate drawn from many.

Even here, one wobble hints at the risk

With only five bull years, a regime split cannot prove much, and every bucket below is drawn from the same short, kind sample. But it offers one quiet hint. Across this history the strategy did well whenever growth was rising, and in the one stretch where growth turned down, its median month went negative. The very companies sold as all weather compounders slipped the moment the weather changed, in the only glimpse we have of it.

Median monthly return by macro regime
Top decile · a single five year bull sample, read with caution
Read this lightly. Five years cannot separate skill from weather, but the lone growth-down stretch is the only time it lost money.

Strip the name off. What is it really?

Ask which real, buyable fund this screen most resembles and the answer is a small cap blend: month to month it tracked a basket of small cap and midcap funds at a correlation of about 0.83. Its apparent edge over an index does exist on paper, but it rests entirely on one broad index across one bull run; measured against most other indices the same "alpha" turns negative. The flattering number is a property of the sample, not the skill.

0.83
correlation to a small cap fund blend · effectively a small cap bet
1
bull cycle behind every headline number · no bear market in the sample

So what actually worked?

Put it together and the spectacular record thins out. The ranking did not sort winners from losers, because in this window there were barely any losers. The metric it ranks on did not predict returns. Concentrating into its best ideas hurt. What remained was a broad basket of decent companies carried upward by the strongest bull market in a generation, and a screen that got to stand in front of the result and take a bow.

That is not proof the Coffee Can is empty. Quality may compound beautifully over a full cycle. It is proof that five winning years cannot tell you, and that the only honest verdict waits for a market that falls. A respected name and a dazzling five year chart told you one story; the data underneath told a quieter one. The way to know the difference, for any idea you believe, is to test it across the bad years too.

The one test it has not had

Every number here is drawn from a single bull market. The honest way to judge the Coffee Can is to run it across the years that hurt, and to try the tweaks its defenders suggest: different quality thresholds, tighter or looser concentration, longer holds. Until it meets a falling market, the record is a promise, not a proof. That test is a few clicks away.

So we took it to Krest, and ran it through the whole test.

KREST TESTED · RUN ON REAL HISTORY ·
Method mark
Krest Tested
We took the Coffee Can ranking apart, bucket by bucket and factor by factor, across its full history. The rigour is ours; the verdict is yours.

Free · no account needed

Test before you trust.

Don't take our word for any of it. Every figure in this teardown came from a few clicks on Krest, and each is a click from the full, live analysis. Reading and exploring is free.

More Krest Research

For education only. Not investment advice or a recommendation to buy, sell, or hold any security, strategy, or product. Past performance does not guarantee future results, and all investing carries risk, including the possible loss of capital. Make your own decisions, and consider consulting a SEBI registered investment adviser.

Best effort analysis. Prepared on a best effort basis from historical data and may contain errors, omissions, or assumptions. Shared for information and discussion only, and should be independently verified before you rely on it. Krest accepts no liability for any decision made or loss incurred based on it.

Figures reflect a quality screen in the Coffee Can spirit (rank by return on capital, quality and growth filters, market cap above ₹100 cr), reconstructed yearly, over its available history since June 2020, measured against the Nifty 500 total return index. This is a five year sample set entirely within a bull market; decile, correlation and base rate figures are computed across all rolling windows within it.

Krest · mykrest.com