Position Sizing Beats Prediction Most Weeks

Investors love prediction. Directional calls are vivid, shareable, and easy to score in hindsight. Position sizing is quieter. It rarely trends on social media. Yet most weeks, sizing decides outcomes more than the cleverness of the call. A mediocre idea held at a survivable size beats a brilliant idea held at a ruinous size.

What sizing is actually saying

A position size is a compressed statement about conviction, liquidity, correlation with the rest of the book, and how wrong you can be before the mandate breaks. When people skip that compression and jump from narrative to notional, they are not being bold. They are outsourcing risk management to hope.

I ask four questions before size becomes final. What is the maximum loss I accept if the thesis fails quickly? How long might I need to exit without forcing the market? How does this risk stack with existing factor exposures? What happens to the book if three similar ideas fail together? If those answers are vague, the size should be small until they are not.

Volatility scaling is useful and incomplete

Scaling positions by recent volatility is a useful default. It keeps day-to-day noise from dominating decision quality. It is incomplete because calm regimes invite oversized notionals that become violent when noise normalises. I prefer sizing as if volatility can jump toward a longer-history percentile I actually respect. If that size feels “too small to matter,” the strategy may not be robust enough to deserve capital.

Leverage as compressed sizing

Leverage is position sizing with borrowed confidence. Traditional credit desks learn this early; digital-asset venues teach it faster. The question is not whether leverage can amplify returns. It can. The question is whether your process names the funding, liquidation, and gap risks before the amplification looks free.

Prediction theatre versus process

Prediction theatre rewards being loudly right once. Process rewards being quietly not ruined often. In research meetings I would rather hear a clear invalidation level and a size rationale than a poetic macro story with no exit math. Stories matter. Stories without sizing are entertainment.

Systematic tools help here. Rules that map signal strength to capped risk, hard loss limits, and gross exposure ceilings remove some of the ego from the moment of entry. They do not remove judgment. Someone still designs the envelopes. Automation without envelopes is just faster regret.

A bias I keep returning to

Prefer fewer positions you can explain under stress over many positions you can only explain in a bull tape. Prefer sizes that keep you solvent and solvent-minded. Prefer written sizing rules that survive both boredom and euphoria.

Conviction is not a position size

Teams often translate rhetorical conviction into notional. The louder the room, the larger the ticket. That mapping is backwards. Conviction should affect how much research time you spend and how clearly you write invalidation—not how casually you ignore liquidity and correlation. A high-conviction idea can still deserve a modest size if the left tail is fat or the exit is crowded.

I like sizing grids that map signal quality and uncertainty into capped risk units. The grid is imperfect. It stops the worst improvisation. When an idea sits outside the grid, that is a feature: it forces a conscious exception with an expiry, not a silent enlargement.

Path dependence and the second loss

After a winning streak, people size up. After a losing streak, people either freeze or double to “get it back.” Both behaviours are path dependent and usually hostile to edge. A sizing process that resets to rules rather than to recent P&L is a form of emotional hygiene. Journals help here: they show whether size followed the rule or followed the last trade’s mood.

In multi-strategy books, sizing also means budget across strategies so that one “small” idea does not consume the firm’s entire risk appetite through correlation. Prediction of each line can be fine while the portfolio prediction—the joint left tail—is ignored. Most weeks, that joint view matters more than any single clever call.

Teach juniors sizing before teaching storytelling. Stories are abundant. Survivable sizing is scarce.

You will still be wrong. Markets guarantee that. The goal of sizing is not omniscience. It is to make wrongness ordinary rather than existential.

Sizing across correlated ideas

A book full of individually modest positions can still be enormous if the ideas share one factor. Correlation-aware sizing is less glamorous than naming winners, yet it prevents the “death by a thousand small tickets” problem. I prefer a gross factor budget that caps how much total risk can sit in one story, regardless of how many tickers express it.

When uncertain, shrink. Shrinking is reversible. Ruin is not. Prediction can wait for another meeting; solvency cannot.

If your edge is real, sizing lets it compound. If your edge is imaginary, sizing limits the tuition bill. Either way, sizing is the grown-up part of the job.