Ask most people how investors make money and they'll say: pick the right stocks. Ask a professional and they'll tell you the pick is only half the job — often the smaller half. The other half is a quieter question: how much? This is position sizing, and it is where two investors holding the exact same list of stocks can end up with completely different results. Below, we make that concrete with the actual math — the same winning strategy, run at four different sizes, ends anywhere from tripling your money to losing nearly all of it.
Why the same idea can be a great bet or a terrible one. Imagine two people who both think a company will do well. One puts 2% of their money in it; the other puts 40%. They own the identical stock on the identical day for the identical reason — but they are not making the same decision. If the idea works, the second person wins big. If it fails, the second person may not survive to make the next bet. The pick was the same. The sizing made them different investors.
A formula from a telephone lab. In 1956, a Bell Labs scientist named John Kelly Jr. — a colleague of information-theory pioneer Claude Shannon — published a formula meant to describe noise on long-distance phone lines. It turned out to answer a gambling question just as well: given an edge, what fraction of your money should you risk to grow fastest over the long run without going broke? The Kelly criterion says bet more when your edge is larger and your odds are better, and less when they're thin. For the cleanest possible case — a coin that lands your way 55% of the time, at even money — Kelly's answer is exactly 10% of your bankroll per bet.
Too big is worse than too small — and that's arithmetic, not opinion
Here's what makes sizing unforgiving: the penalty for oversizing is not symmetrical. Take that genuine 55% edge and place 250 bets, starting from $10,000, at six different sizes. No simulation is needed for this chart — the median outcome follows directly from the growth formula:
Read the right side of that chart carefully. At 2× Kelly (20%), the median bettor wins 55% of 250 bets — a genuinely winning record — and ends with roughly the $10,000 they started with. All of the risk, none of the reward: the growth math cancels to zero. At 3× Kelly (30%), the same winning edge ends with about $170 — a 98% loss, from a strategy that was right more often than wrong. Losses compound against you faster than wins compound for you: −20% then +20% is not break-even, it's −4%. Undersizing costs you some growth. Oversizing costs you the game.
The same coin, three bankrolls
Medians hide the ride. So here is a virtual simulation — not real prices — of three bankrolls betting on the identical sequence of 250 coin flips (55% winners, deterministic seed), differing only in how much they stake per flip:
| Bet size | vs Kelly | Median after 250 bets | Growth / bet | Max drawdown* | The trade-off |
|---|---|---|---|---|---|
| 2.5% | ¼ Kelly | $17.3k | +0.22% | −17% | Safe, slow — leaves compounding on the table |
| 5% | ½ Kelly | $25.6k | +0.38% | −31% | Roughly ¾ of Kelly's growth for a fraction of the pain — the practitioner favorite |
| 10% | Kelly optimum | $35.0k | +0.50% | −57% | Fastest possible compounding — if your edge estimate is exactly right |
| 15% | 1.5× Kelly | $25.4k | +0.37% | −82% | Same median as ½ Kelly, with far deeper drawdowns |
| 20% | 2× Kelly | $9.7k | −0.01% | −95% | All of the risk, none of the reward — growth math cancels to zero |
| 30% | 3× Kelly | $174 | −1.61% | −100% | A genuine winning edge, bet into near-certain ruin |
Notice what the drawdown column does: half Kelly keeps roughly three-quarters of full Kelly's growth rate while cutting the swings dramatically, and 1.5× Kelly earns the same median as half Kelly while living through far deeper crashes. That's why serious practitioners — from card-counting teams to quant funds — habitually bet a fraction of Kelly: the formula assumes you know your edge exactly, and being wrong about your own edge is the normal condition of investing. If your true win rate is 52% while you believed 55%, full Kelly is already overbetting. Small samples make edges look bigger than they are — which is precisely when oversizing does its damage.
How our engine turns this into actual weights
This isn't theory to us — sizing is most of what our engine does each morning, and the recipe is public. Each strategy's confidence in a stock is first calibrated downward (a temperature adjustment that tempers overconfidence — a lesson straight from the paragraph above). Calibrated confidence then sets a half-Kelly-style cap: a position's weight may not exceed the calibrated edge over a coin flip. Each candidate is scored as confidence divided by downside volatility, so a jumpy stock needs more conviction to earn the same weight. Finally, everything is squeezed under each strategy's hard single-position cap — 15% of the book for Conservative, 20% for Sector, 25% for Core, 35% for Aggressive, fixed by strategy character rather than tuned to a backtest — and whatever doesn't fit stays in cash.
One honest note: this conservatism has a price, and we've measured it. In a July 2026 review our engine's caution was leaking upside — in strong rallies it deployed less than intended and lagged on the way up. The first fix we designed failed its pre-registered backtest gate, so we kept the cautious sizing. That is the Kelly trade-off, live: we would rather leave some growth on the table than discover, one bad month, that we had been overbetting an edge we misjudged.
Why it matters for reading any track record. When you look at a set of investment decisions, don't stop at the tickers. Look at the weights. A big position tells you where the real conviction — and the real risk — lives. A small one is a maybe. The list of names tells you what someone is thinking about; the sizes tell you what they actually did.
1. Sizing beats picking more often than intuition suggests: the same 55% edge triples money at Kelly (10%), gains nothing at 20%, and loses ~98% at 30%. Oversizing destroys winning strategies.
2. Betting a fraction of Kelly (half is common) keeps most of the growth with far shallower drawdowns — and protects you from the near-certainty that you've overestimated your own edge.
3. Weights are the honest sentence in any portfolio: they say what an investor actually did. These are mathematical trade-offs, not predictions or a betting recommendation.
Informational and entertainment content only. Not investment advice. The coin-flip figures are pure arithmetic and a labeled virtual simulation — not market data, backtests of real assets, or return forecasts. Sizing frameworks like Kelly describe trade-offs; they are not a guarantee of returns, and every position carries the risk of loss.