A Practical Framework for Position Sizing

Position sizing is the part of investing that gets the least attention relative to how much it actually matters.

Most of the discussion in investing focuses on selection, finding the right business, the right market, the right entry price. Sizing is treated as an afterthought, a percentage figure attached at the end of the analysis almost as a formality. This is backwards. Two investors can be equally right about the same business and produce very different outcomes purely because of how much capital each of them committed to the idea.

There are two common failure modes in how people approach sizing, and they sit at opposite extremes.

The first is the purely formulaic approach, often derived from something like the Kelly criterion or a volatility target, where a mathematical model outputs a specific percentage based on expected return, variance, and correlation assumptions. The problem with this approach in practice is that the inputs are estimates dressed up as precision. Expected return is a guess. Historical volatility may not reflect forward looking risk. Correlation assumptions break down exactly when they matter most, during periods of genuine market stress. A model that outputs 7.3 percent as the optimal position size is providing false confidence, not real precision.

The second failure mode is the opposite, sizing purely by feel, where conviction alone determines how large a position becomes. This tends to produce portfolios where the largest positions are simply the ones the investor talked themselves into most enthusiastically, without any structural check on whether that enthusiasm is proportionate to the actual risk being taken.

A practical framework sits between these two extremes, using structure to inform judgment rather than replace it.

The first input is genuine conviction, honestly assessed. This means asking a specific question: if this position went to zero, would that be consistent with something you failed to anticipate, or something you knew was a real possibility going in. A position sized as if failure were nearly impossible should be reserved for theses with genuinely low probability of complete loss. A position where you can articulate a plausible path to zero should never be sized as if that path does not exist.

The second input is correlation with the rest of the portfolio, assessed at the level of underlying exposure rather than asset class labels. Two positions that are both, in substance, bets on a single country's currency stability are correlated in the way that matters, even if one is a bank stock and the other is a government bond. Sizing decisions need to account for what happens to the whole portfolio if the underlying driver common to multiple positions moves against you, not just what happens to each position in isolation.

The third input is liquidity, specifically the ability to exit the position at a reasonable price under both normal and stressed conditions. A position that represents a meaningful percentage of daily trading volume in its market cannot be sized the same way as an equivalent conviction position in a highly liquid, deeply traded name. The exit is part of the position, not a separate consideration to be dealt with later.

Putting these three inputs together produces a workable process rather than a single formula.

Start with a maximum size ceiling based on liquidity, the largest position you could realistically exit within an acceptable timeframe and price impact under stressed conditions. This is your hard constraint, not your target.

Within that ceiling, size according to conviction, using a small number of tiers rather than a continuous scale. A highest conviction tier for theses you have researched most thoroughly and would be genuinely surprised to see fail. A middle tier for good ideas with real uncertainty attached. A smaller, exploratory tier for positions you are still building conviction on. Three or four tiers is usually enough. More granularity than that tends to imply a precision the underlying judgment does not actually have.

Then adjust down, never up, for correlation. If a position shares its primary risk driver with other significant positions already in the portfolio, reduce its size below what conviction alone would suggest, because the portfolio's actual exposure to that risk driver is the sum of all correlated positions, not any single one.

This framework will not produce the same answer every time for the same investor, and that is appropriate. Position sizing is not meant to be a formula that removes judgment. It is meant to be a structure that makes judgment more consistent, more honest about what is actually known versus assumed, and more resistant to the kind of enthusiasm that leads good ideas to become oversized positions.

The investors who compound most effectively over long periods are rarely those with the highest hit rate on individual ideas. They are those whose sizing discipline means no single wrong idea can meaningfully damage the portfolio, while their best ideas are sized large enough to actually matter when they work.