What a decade of in-house research reveals about the craft of picking stocks
The Shop in Yugawara
Anyone who has ever asked me for a dinner recommendation learns quickly that few things in this world make me tick like a good bowl of ramen. So some months ago, longing to relive my trip to Japan last April, I sat down to watch a documentary about a ramen shop.1 The shop sits in Yugawara, a quiet seaside town about two hours outside Tokyo, and by more than a few accounts it serves what some would call the best bowl in the country. Which means that every morning, its proprietor, a man named Iida Shota, and his apprentices wake before sunrise to face one humble, almost Sisyphean task: to go on making the best bowl of ramen. The day begins at 5:30, and the film follows their ritual: scallions cut, chashu roasted, noodles pressed fresh. Every component is made from scratch. And yet, after thousands upon thousands of hours over two decades, Iida insists he has only just begun. What stayed with me, beyond the mouth-watering footage, was a single remark. Iida says he does not much care about being called the best in Japan. He thinks of himself instead as an artisan whose obligation is less to the customer than to the craft itself. Mastery, in this telling, is not a destination you arrive at and defend, but a direction you keep walking in.
While we might reasonably disagree with Iida on the importance of our duty to clients, I have been turning that idea over in my mind ever since, trying to work out why it struck me so deeply, and how it applies to what we do at Qube. As a wealth manager, I feel as though there is a constant temptation to present the firm as a finished thing: polished, authoritative, and arrived. But the truth about our research is that it resembles that kitchen in Yugawara far more than it resembles a glass tower in downtown Toronto. Most of our work is preparation: those same disciplined motions repeated and refined and improved, far from anyone’s view. In that sense, this commentary is a reflection on what happens when you finally write all those motions down.
Why We Wrote It All Down
For most of our history, Qube’s research has lived the way most good kitchens have lived: in the hands and the heads of the people doing the work. An analyst would study a company from our universe, build a model, and form a view. That view, the distillation of the narrative and the numbers, went to our steering committee, who decided how to act on it.
It is, I am convinced, a sound process, and a track record stretching back to inception feels like fair proof of concept. The bar, though, is higher than it looks. Research from the Institute of Business & Finance finds that nearly nine in ten active managers underperformed their benchmark over fifteen years, net of fees.2 That means the industry’s central promise, that a professional can beat a simple index after costs, is one that most professionals fail to keep. Clearing that bar puts Qube in an ever-narrowing group of managers who have actually delivered sustained alpha, after costs. But I don’t mention this to boast. I say it because the group is small, and we would very much intend to stay in it.
Pull quote: “We are humbled to have won many awards, but I feel uncomfortable when people say we are the best… because I don’t think about our accolades much. I would very much like to improve.” — Iida Shota
And yet the finer details of that record lived in too many places: scattered across reports, spreadsheets, re-ratings, and the memory of people who, in the nature of our analyst program, eventually move on. We could always tell you what we owned and why, but it was far harder to reconstruct, from first principles, everything we had ever considered and everything we declined to own, and the reasoning behind each call.
Figure to rebuild TABLE: Kaleo Full vs. its traditional benchmark since inception: annual total return (%), yearly alpha, and rolling annualized alpha. Original PDF, page 17.
To be clear, the old way has paid off. Since inception, our flagship Kaleo Full model has compounded at roughly 13.7% a year against a traditional benchmark nearer 11.5%. That 2.2% gap may feel modest, but it is the difference between turning a dollar into close to eight over the past fourteen and a half years and turning it into just over six. We have beaten the benchmark in ten of the last fifteen years.
A decade and a half of alpha, however gratifying, is still a record of results, and not an understanding of why. And it is the years that disappointed us, not the ones that flattered us, that have the most to teach, provided we can actually study them. That last part is the crux. It is very hard to learn from your own history when that history is scattered across hundreds of files and the fading recollections of people who have since moved on.
So over the past year we set out to consolidate everything. The result is what we call the Master Research Database: a single record capturing a decade of in-house equity research, from the first cursory screen of a company to the deep, conviction-level review that precedes a purchase or a sale. To borrow the metaphor one last time, it is the recipe book, prep list, and discard pile, finally gathered in one place. Our research program, like a kitchen, cannot improve what it does not measure. That’s why we built the database — so that we could begin asking harder questions of ourselves and answer them with evidence rather than recollection.
Pull quote: “Using the discarded parts is important… As a chef, it is a privilege to cook with these ingredients. Instead of throwing out the leftover vegetable seeds and skins, we dry them and add them to the stock.” — Iida Shota
A Decade, in Four Numbers
Figure to rebuild INFOGRAPHIC: A Decade, in Four Numbers: 1,196 companies studied; 2,222 research reviews; ~15,160 analyst hours; 10 yrs, 2016 to 2026. Source: Qube Master Research Database. Hours estimated at 4 per Phase 1 review, 8 per Phase 2, and 40 per Phase 3. Original PDF, page 18.
Ten thousand hours of deliberate practice is what is said to separate an apprentice from a master. Whether those hours are spread across eleven hundred companies or poured into a single bowl, the work that goes into a craft is mostly invisible in the final product. A client never sees the screens that went nowhere, the models built and set aside, or the long evenings spent talking ourselves out of a too-good-to-be-true decision. They see only the bowl: the twenty-eight to forty positions that made it out of the kitchen. And while these four numbers are a glimpse through the pass, I will attempt to walk you through the kitchen behind it, and through what a decade of working in it has begun to teach us about our own craft.
The Sieve
One of the most revealing things in the record is not how many companies we studied, but how few survived the studying. Our process moves a name through three phases of escalating scrutiny. Phase 1 is the first screen — a fast, structured look to decide whether a company is worth more of our time. Phase 2 is genuine modelling, where a business is taken apart and rebuilt. Phase 3 is the conviction review, representing the deepest level of work that precedes a decision to commit our clients’ capital. At each phase, many candidates fall away, and what remains is a steeply narrowing funnel.
Figure to rebuild CHART: Figure 1 — Of 1,814 Phase 1 reviews, fewer than one in twelve reached the deepest level of analysis. Source: Qube Master Research Database, 2016 - 2026 (1,196 unique companies; 1,814 Phase 1 reviews). Counts reflect reviews instances at each phase. Original PDF, page 19.
Of the 1,814 companies that entered Phase 1 (some for the first time, others revisited as our universe was reset), only 263, about 14.5%, earned the time of a full model. Of those, 145 went on to a conviction review. Put plainly, barely one company in twelve that we look at makes it to the stage where we are seriously prepared to own it, while the rest are set aside. But that’s the entire point; a sieve that lets everything through is not a sieve. In that context, the value of our research program lies precisely in what it declines to recommend — in the eleven bowls poured out so that the twelfth is worth serving.
Conviction Earned, Not Assumed
If depth of work is doing its job, a second pattern should emerge: the deeper we look, the more decisive we become. As expected, the data bears this out almost perfectly. At the first screen we are skeptics, and only about a quarter of rated names are worth further scrutiny; the largest share of prospects are passed over as fairly valued or wildly overpriced. By the conviction stage, three in four of our rated verdicts are buys, and outright sells all but disappear.
Figure to rebuild CHART: Figure 2 — The share of “buy” verdicts climbs from 23% at first screen to 75% at the deepest review. Source: Qube Master Research Database. Among rated verdicts, BUY rises from 23% at first screen to 75% at deepest review. Original PDF, page 20.
It would be easy to read that rising line as growing enthusiasm, but it’s the opposite. Conviction climbs with depth, not because we talk ourselves into companies, but because depth is how we talk ourselves out of the weak ones. While a first screen can be fooled by a good story, forty to sixty hours of modelling cannot. So each phase strips away the names that only looked good from a distance, and what survives to Phase 3 is a small set of companies that has withstood every reasonable attempt to break it. The high buy-rate at the deep end, then, is a measure of how much doubt the survivors had to outlast.
Why does this matter for Kaleo? A portfolio built from names that earned conviction through scrutiny should behave differently from one built on first impressions. Its mistakes are fewer and more explicable, while its winners are better understood, and its drawdowns easier to hold through because we know exactly why we are there. In other words, our early skepticism is what buys us durability later. The sieve, therefore, is not just a filter on quality, but also a source of the portfolio’s resilience.
Beyond Attribution
There is a harder question the database was really built to answer, and one every active manager should eventually face: does our research actually work? The conventional tool for answering it is performance attribution:
the practice of splitting a portfolio’s return relative to its benchmark into the part explained by where you were invested (allocation) and the part explained by which securities you chose (selection), with a small interaction term where the two combine.
We know the tool firsthand, not just from our time spent in the CFA Program, but because we have applied it before. Over the years, we have run formal Brinson-style attribution analysis on Kaleo more than once. While the studies were useful in context, they were ultimately self-defeating. As one of our own internal reviews put it, the whole method assumes the allocation decision (which sector to invest in) comes first. If you take a genuine bottom-up, stock-picker’s view of the world, then there is simply nothing to analyze on the allocation side, because there was never a sector decision to begin with. This industry’s favourite scorecard, in short, is built for a kind of investor we are not.
At first glance, a reader might gloss over a subtle distinction. Qube does not build portfolios by deciding to overweight technology and underweight materials. We build them one company at a time, from the bottom up, and the sector weights that result are a consequence of that stock-by-stock work, and not a strategy in their own right. For a shop like ours, the allocation effect of an attribution analysis measures a decision we never deliberately made, and it can swamp the one thing we actually care about: whether we picked good companies.
Our own year-end numbers show this pitfall in miniature, since the headline return does us a disservice. We picked good stocks. Measured over the three years ended December 31, 2025, the part of our result we genuinely control, stock selection, added +11.65% of cumulative excess return, led by our communication-services, health-care, and information-technology names. What dragged the headline down was allocation, a -6.10% cumulative cost, driven chiefly by an overweight to a soft consumer-staples sector and an underweight to a roaring materials sector. This unfolded during a historic, tech- and AI-fuelled run-up that rewarded precisely the corners of the market a disciplined value process is slowest to crowd into. The net of the two was +5.54%. In other words, our stock-picking worked. We simply ended up light in the period’s hottest sectors as a by-product of letting conviction, not trends, decide where we invested.
Figure to rebuild CHART: Figure 3 — Selection (our craft) added +11.65% over the three years to December 2025; allocation (a by-product) cost -6.10%. The +5.54% net hides the skill. Source: Kaleo Full performance attribution, 3 years to Dec 31 2025 (GICS sector). Selection added +11.65% cumulatively; allocation drag -6.10%; net +5.54%. Original PDF, page 21.
However, I want to be clear that none of this means attribution lies. It means attribution answers the wrong question for a stock-picker. It can tell us whether this quarter’s portfolio beat the benchmark, which is a fact already visible in the statement. It cannot tell us whether our research is getting better. Those are different questions, and only the second one is about the craft.
Pull quote: “The moment I feel like I understand it, I suddenly have no idea. That’s why I think I love ramen so deeply, and want to pursue it for life.” — Iida Shota
Where the Verdicts Go
If attribution cannot show us the craft, then our ledger is one of the few tools that can show us something attribution never could: the full shape of our judgement, sector by sector, as it narrows from a first impression into a committed position. Every name that enters our universe eventually carries a rating, where a 1 or 2 is a buy, a 3 a hold, and a 4 or 5 a sell. Typically, only the buys advance. The diagram below traces that flow across all ten sectors and all three phases.
Figure to rebuild CHART: Figure 4 — Across ten sectors, only Phase 1 buys (295) advance; holds and sells are set aside. Of the buys, 98 survive Phase 2 and 90 reach Phase 3. Source: Qube Master Research Database (rated verdicts). Only Phase 1 buys (ratings 1-2) advance to Phase 2; holds (3) and sells (4-5) are set aside. The pattern repeats into Phase 3. Original PDF, page 22.
What the figure above makes vivid is how unsentimental our process is. Arguably, you might also see that a hot sector does not buy a company a pass, and a cold one does not condemn it. Instead, the verdict travels with the business and its fundamentals, rather than with its peer group.
The Questions Made Possible
Ten years of consolidated history changes what is knowable. With the full record in hand, we can finally ask questions that no single attribution analysis could reach, and, more importantly, begin to answer them:
— Does conviction predict performance? Every recommendation we have ever made carries a rating. We can now test, directly, whether our highest-conviction calls have actually outperformed our lukewarm ones. If they have, we can build portfolios that lean deliberately into conviction rather than treating every buy as equal.
— Has a decade of sharpening our craft paid off? Our valuation work has only grown more sophisticated than it was in 2016. The record lets us ask whether that added rigour has produced more reliable alpha, and whether our edge has genuinely widened as we refined the model, the methodology, and the craft, or whether it has merely produced more elaborate models. This is the difference between a sharper knife and a heavier one.
— If we have an edge, what should we build with it? The early evidence points to real, repeatable skill in certain corners and more mixed results in others. Confirming where that edge lives is the first step toward something larger: new strategies, designed around our demonstrated strengths, so that we have more and better ways to serve what each client actually needs.
None of these is a question about a single bowl, but instead the kitchen, and about whether the craft itself is improving. With this project, we can shift from grading individual results toward studying, and steadily bettering, the process that produces them.
The Shop in Edmonton
For ten years this work existed only as it happened. It was lived, acted on, and then, inevitably, set aside. The remarkable thing is not any single number or diagram meant to showcase what is possible, but that a decade of effort now exists in one place, where it can finally be questioned.
In the coming quarters we will back-test the link between our conviction and our returns and use what we learn to build our existing portfolios with even more intention. Where the evidence is strong enough, the same work can seed entirely new strategies, grounded not in a hunch about what we are good at, but in a decade of recorded proof, and aimed squarely at serving our clients better. We are, in the most literal sense, just getting started. As the ledger continues to grow, one company at a time, each addition makes the next question easier to ask.
Inspiration can come from anywhere, and I never anticipated that a film about a small ramen shop would stay with me the way it has. Perhaps it was the passion of an artisan with no interest in being called the best, only in being a little better tomorrow than he was today. There is a kitchen like Iida’s in Yugawara, and there is also one a great deal closer to home, in Edmonton. One where the same kind of unglamorous, repeated and careful work goes on. As we pass the halfway mark toward our next decade, his posture strikes me as well worth borrowing. Our database, set alongside our returns, is not a trophy to be set on a shelf and admired. It is a way to see whether our knife-work is getting sharper and our palate wider, all in the pursuit of a mastery we will never finally reach. Everything in these pages is simply the kitchen behind us: the early mornings, the discarded parts folded back into the stock, and the thousands of hours no client ever sees. Taken together, they culminate into one more decade-long conviction, and that the work, thank goodness, is never finished.
Important: Figures reflect internal research records and historical Kaleo performance. Past performance is not indicative of future results. Conviction-to-return back-testing referenced above is forthcoming; no claim of realized stock-selection alpha attributable to the database itself is made or implied. This commentary is for informational purposes and is not investment advice.
Notes
1 The Most Legendary Ramen Shop in Japan by Alvin Zhou Films.
2 SPIVA Scorecard: How Active Funds Performed vs. Their Benchmark by the Institute of Business & Finance.