Position Sizing by Conviction: A Quality-Weighted Approach

Assume two investors own identical holdings — the same five companies, purchased at the same prices on the same day. Investor A distributes capital evenly at 20% each. Investor B weights by conviction: 35% in her highest-confidence name, then 25%, 20%, 12%, and 8% in descending order of certainty. Five years later, the highest-conviction pick has returned 23% annually; the other four average 11%. Equal weighting produces a blended return of roughly 14.2%. Conviction weighting produces 17.5%. Three percentage points of annual difference — from no research advantage whatsoever, no better stock selection, no different information.

Position sizing is where equity research converts to actual results. It's also where most investors are least disciplined. Stock selection — which companies to own — attracts nearly all the intellectual energy in portfolio management. How much to own each company attracts almost none. The result is positions sized by feel: a little more in the names you were excited about when you bought them, a little less in the ones where uncertainty lingered, no clear framework behind the differences, and no systematic process for updating the weights as conviction changes.

There is a theoretically optimal answer to the sizing question — the Kelly criterion — and there is the practical framework most investors should actually use instead. The gap between them is more instructive than either one alone.

What the Kelly Criterion Gets Right — and Where It Breaks Down

John Kelly Jr., a researcher at Bell Labs, published his optimal-bet formula in 1956 in the Bell System Technical Journal. The formula: f* = (p(b+1) − 1) ÷ b, where p is your estimated probability of a favorable outcome and b is the net gain per unit risked on a win. The insight it encodes is elegant — position size should scale precisely with edge. If the probabilities favor you more strongly, allocate more. If the edge is thin, allocate less. Hold nothing where expected value is zero or negative.

f* = (p × (b + 1) − 1) ÷ b p = estimated probability of a favorable outcome b = net gain per unit risked on a winning bet

What Kelly gets right: conviction and allocation should move together. This sounds obvious. The gap between knowing it and implementing it systematically is wider than most investors admit — which is why most portfolios are effectively equal-weighted even when the underlying conviction is not.

Where Kelly breaks down: it requires probability estimates that simply do not exist in equity investing. What is p for the hypothesis that Visa's global payments network maintains structural dominance over the next decade? Is it 0.74? 0.88? The honest answer is somewhere in 'meaningfully above 50%, but I cannot assign it a precise number.' Run the formula with an overestimated p and it recommends dangerously concentrated positions — Kelly-consistent sizing can suggest 25% or more of capital in a single name when you're highly confident. No portfolio-level consequence of that confidence being wrong.

Bill Ackman's Pershing Square Capital Management has historically run 8 to 12 positions, with the largest often representing 15–25% of the portfolio at cost — sizing that is broadly consistent with high-Kelly allocations for an investor with Ackman's depth of research. It also produced a loss of roughly 40% in calendar year 2015, when the Valeant Pharmaceuticals thesis — which Pershing Square's Q3 2015 investor letter still defended at length — collapsed alongside the business. Kelly's math was not wrong. The probability estimate was. And there was no diversification buffer to soften the landing.

A Quality-Weighted Framework for Position Sizing

The practical alternative replaces Kelly's precise probability input with a qualitative conviction tier. Instead of estimating that your probability of success is 0.78 — precision you don't have — you assess whether your conviction is high, moderate, or exploratory, and translate that tier into a position size range. The underlying analytical work is the same. You're just not pretending to quantitative precision that the input data cannot support.

Three tiers work for most individual portfolios. High-conviction positions — wide-moat businesses with well-understood competitive dynamics, quality scores that have held across multiple economic cycles, and valuations offering a genuine margin of safety — warrant allocations in the 7–10% range. These are the holdings where your research is deepest, your understanding of the downside scenarios is most explicit, and your basis for the thesis has been stress-tested across at least several years of operating history.

Moderate-conviction positions sit in the 3–6% range. These might be narrower-moat businesses where the competitive position is real but more contested, or wide-moat companies where valuation provides less cushion than you'd like, or genuinely high-quality businesses in industries you understand less thoroughly than your core holdings. Not disqualified. Not fully researched to high conviction. Sized accordingly.

Exploratory positions at 1–2% serve a different function entirely. Owning a small stake in a business you're actively researching changes your attention in useful ways — you read the filings differently when capital is deployed. At 1–2%, a complete loss is noise. You're not making a portfolio decision at that size. You're making a research decision.

The tier assignments matter less than the discipline of making them explicitly — and updating them when conviction changes. A position you bought at 7% might now represent 5% of portfolio purely because the stock underperformed. Or it might warrant deliberate trimming because something in the most recent 10-K reduced your confidence in the competitive position. Those are completely different situations with different implications. Only an explicit framework lets you tell them apart in real time.

A Framework for Deciding Your Conviction Tier

The questions that move a holding into the high-conviction tier are specific. Can you articulate the competitive moat in two sentences without using the word 'quality'? Have you read at least three annual filings — including one from a difficult operating year — and emerged with your thesis intact? Does the quality score hold across margin cycles, not just favorable ones? And do you have a clear picture of what would have to be true for the thesis to be durably wrong?

That last question is the one most investors skip. High conviction does not mean 'I think this goes up.' It means: I have thought carefully about the paths to permanent loss of capital, assessed their likelihood, and sized the position as though those paths are genuinely unlikely — not merely uncomfortable to contemplate. A wide-moat business can fall 30% in a broad market selloff with no change to the underlying competitive position. That is noise — uncomfortable noise, but noise. A business quietly losing pricing power to a better-capitalized competitor is a different situation, and treating it as a 'temporary drawdown' is the mistake that turns a bad position into a portfolio-defining loss.

For risk caps: even the highest-conviction position should have an explicit ceiling. At 10%, a complete loss on a single holding reduces your portfolio by 10 percentage points — painful but recoverable. At 25%, the same scenario becomes a defining event for your long-term returns, not a setback. Most individual investors should cap any single position at 10–12% regardless of conviction. The cap is not a concession to uncertainty. It is a recognition that even correct analysis can collide with outcomes no analysis fully prices.

Portfolio-level context shapes what the tiers look like in practice. A 12-stock concentrated portfolio should carry smaller individual position ceilings than a 20-stock diversified one, because the same loss is proportionally larger at the portfolio level. The rebalancing rules question — how to maintain target weights over time — interacts directly with the conviction tier framework. The two should be designed together, not in isolation.

Conviction Isn't Certainty

The caveat deserves its own section. I'm genuinely less confident about the right ceiling for high-conviction positions than I'd like to be. The academic and practitioner literature on concentrated portfolios is fragmented — some studies document persistent outperformance from 10–15 position funds; others find no consistent advantage after controlling for survivorship bias and style tilts. The methodology differences matter enough that a clean, universal conclusion is probably unavailable. The right answer depends on your research depth, your temperament under drawdown stress, and your actual — not hypothetical — ability to distinguish thesis changes from market noise.

What the evidence is clearer about: investors who've run concentrated portfolios successfully over decades share one discipline that has nothing to do with confidence. They are as rigorous about defining what would change their view as they are about building the positive thesis. Conviction without an explicit falsification condition isn't conviction. It's attachment. And attached investors hold deteriorating positions too long, add to weakening theses because the price is lower, and confuse averaging down with adding to conviction. Those are not the same thing.

The quality tier framework helps here — but only if you apply it honestly. A position that was high-conviction three years ago but where subsequent filings have weakened the basis for the thesis should not remain at 8% because you are reluctant to recognize a change in view. If the current evidence would place it in the moderate tier, size it at 3–6%. The position sizing decision should reflect your current assessment of the business, not the confidence you had when you first bought. Updating the tier is not admitting a mistake. It's decision discipline.

And quality scores are an input, not the final answer. They surface the businesses worth deeper conviction analysis — filtering out obviously lower-quality names and directing attention toward the ones with durable economics. But conviction requires engaging with the actual filings: the competitive landscape section of the 10-K, the capital allocation commentary in the MD&A, the pattern of guidance versus actual results across four or five years. The score tells you where to look harder. It does not substitute for looking. When a company scores well on all the quantitative dimensions but you cannot yet articulate the competitive moat simply, that is an exploratory position — not a high-conviction one. Size it like one.

Key Takeaways

• Position sizing is where research converts to results. Two investors with identical stock selections but different conviction weighting can produce meaningfully different returns over a decade. The math of concentration compounds — it works in your favor when conviction is correct, and against you when it isn't.
• Kelly-criterion sizing captures the right intuition — size proportional to edge — but requires probability estimates equity investors don't possess. In practice, overconfident inputs recommend dangerously concentrated positions. A conviction-tier framework (high: 7–10%, moderate: 3–6%, exploratory: 1–2%) is more honest about the limits of what you actually know.
• The conviction tier is a current assessment, not a historical one. A position weakened by subsequent information needs to be reassessed at the tier level, not held at the original weight because you're reluctant to acknowledge the change. Decision discipline means updating the size to reflect what you currently believe.
• Even the highest-conviction position deserves an explicit size cap — most individual investors should limit any single holding to 10–12%. And the test that separates a genuinely high-conviction position from one that merely feels comfortable is whether you've explicitly identified the paths to permanent impairment and found them genuinely unlikely — not just optimistically dismissed them.