Position Sizing Methods: Equal Weight, Conviction Weight and Risk Parity

Position sizing is the decision about how much capital each holding receives. It matters because a portfolio’s outcome is the weighted sum of its positions, so the weight applied to an idea determines how much of that idea’s result actually shows up. Getting the names right and the sizes wrong produces a portfolio that does not reflect its own research.

There are four sizing families in common use. None of them is correct in the abstract. Each encodes a specific belief about what the investor does and does not know, and each pays for that belief somewhere.

Equal weight: assuming you cannot rank beyond inclusion

Equal weight gives every position the same share of capital. Twenty holdings means five percent each, reset back to five percent on some schedule.

The belief underneath it is modest and often defensible: the selection process is good enough to say a name belongs in the portfolio, but not precise enough to say this name deserves three times the capital of that one. If the ranking inside the selected set is mostly noise, equal weight is the honest expression of that.

What it buys is robustness and auditability. There are no estimated inputs, nothing to calibrate, and no way for a single position to quietly grow into a dominant exposure between rebalances without it being obvious.

What it costs is threefold. Equal weight requires ongoing rebalancing, since prices immediately push the weights apart, and rebalancing costs money. It tilts the portfolio toward smaller and less liquid names compared with a capitalisation-based scheme, because a small company gets the same rupee allocation as a large one, which is a real exposure decision rather than a neutral one. And it deliberately gives up any information the ranking did contain. The full comparison against the capitalisation alternative is in equal weight versus market-cap weight.

Conviction weight: sizing on the strength of the view

Conviction weighting scales position size with how strong the case for a holding is. In practice this is usually implemented as tiers, such as a core tier at a higher weight and a satellite tier at a lower one, rather than as a continuous function of a score.

The belief underneath it is that the research process produces a genuine ordering, not just a pass or fail. If that ordering is informative, conviction weighting concentrates capital where the edge is and improves the portfolio’s return per unit of risk taken.

The cost is symmetrical and often underappreciated. Conviction weighting amplifies the ranking in both directions. If the ordering is partly noise, the method systematically places the largest bets where the noise happens to be most flattering. It also introduces a subjective input that is hard to review after the fact, because “high conviction” is not measurable in the way volatility is, and it has a documented tendency to be highest exactly when a position has already performed well.

Two disciplines usually accompany it. The first is a hard cap on the largest position regardless of conviction, which bounds the damage from a single wrong call. The second is writing down the conviction rationale at the time the size was set, so that later review can distinguish a considered bet from a drifted one.

Risk-based weight: equalising contribution rather than capital

Risk-based sizing sets weights so that positions contribute comparable amounts of risk rather than comparable amounts of money. In its simplest form, a position’s weight is set inversely to its volatility, so a name that moves twice as much gets roughly half the weight. Risk parity extends this by accounting for correlations, so that positions which move together are not double counted.

The belief underneath it is that rupees are the wrong unit. A ten percent allocation to a stable, low-volatility business and a ten percent allocation to a highly volatile one are not equivalent exposures, even though they look identical on a holdings sheet.

What it buys is a portfolio whose realised volatility is less dominated by its most erratic holdings, and a more even distribution of where the portfolio’s movement actually comes from. Understanding this properly requires knowing what standard deviation of returns does and does not capture.

What it costs is dependence on estimates. Volatility is measured over some past window, and the choice of window changes the answer. Correlations are less stable still, and they have a well-documented habit of rising together in market stress, which is precisely when the diversification the model assumed was there stops being there. A risk parity portfolio calibrated in a calm period can carry more concentrated risk in a violent one than its own model reports.

There is a second, subtler cost: low measured volatility is not the same as low risk. A stock can be quiet for a long time and then gap. Sizing on realised volatility can systematically overweight positions whose risk has not yet shown up in the price series.

Benchmark-relative weight: sizing as a deviation

Benchmark-relative sizing starts from index weights and expresses every decision as an overweight or underweight. A holding at eight percent against a five percent index weight is a three percent active position, and the portfolio is the sum of those active positions.

The belief underneath it is that the portfolio will be judged against a benchmark, so the meaningful decisions are the differences from it, not the absolute weights. Under this framing the natural risk measure is tracking error, and the natural performance measure is active return.

What it buys is coherence between how the portfolio is built and how it is evaluated. It also makes an important category of risk visible: a large index constituent that a portfolio does not hold at all is an active underweight, and it can hurt as much as a bad holding.

What it costs is a structural anchor to the index. The scheme discourages positions far from benchmark weights even when the research supports them, and it imports whatever concentration the benchmark itself carries. That matters in indices where a handful of names dominate, a point covered in how Indian indices are constructed.

The trade-offs side by side

Method	Inputs needed	Main strength	Main vulnerability
Equal weight	None beyond the holdings list	Robust, transparent, no estimation error	Ignores real differences in risk and liquidity
Conviction weight	A trusted ranking	Concentrates capital where research is strongest	Amplifies ranking error; subjective and hard to audit
Risk-based weight	Volatility and correlation estimates	Evens out contribution to portfolio movement	Estimates are unstable and fail together in stress
Benchmark-relative	Index weights	Aligns construction with how performance is judged	Inherits the benchmark’s own concentration

The table is a summary of properties, not a ranking. The right question is which failure mode an investor is best placed to live with, given the reliability of their own inputs.

What position sizing does not tell you

Sizing is a distribution rule, not a source of edge. It is worth being explicit about the boundaries.

It contains no forecast. No sizing method knows which position will work. Every method distributes capital across an unknown future, and the arithmetic that follows is entirely dependent on the quality of the selection that preceded it.

It does not control drawdown. Weights are set at a point in time, and correlated declines move everything at once. A carefully sized portfolio still experiences the market’s maximum drawdown in a broad fall, roughly in proportion to its market exposure.

Measured risk is backward-looking. Every risk-based method reads history and assumes some persistence. When the regime changes, the sizing was calibrated on a world that no longer exists.

It ignores implementation unless forced to. Sizing schemes produce target weights, and reaching those targets involves trading. Small weights in illiquid names can be impossible to fill at the modelled price, and frequent resets accumulate transaction costs and taxes. A sizing policy that does not account for the cost of maintaining itself will look better on paper than in an account.

It cannot be evaluated over short periods. Whether a sizing method helped is a question about many outcomes, not one. Over a single year, the difference between schemes is dominated by which specific holdings happened to work.

Reading a sizing policy honestly

The useful test of any sizing method is not how good its historical numbers look but whether its assumptions are ones the investor actually holds. If the ranking is trusted, conviction weighting is consistent with that trust. If it is not, equal weight is consistent with that scepticism. If risk contribution is the thing being managed, a risk-based scheme follows, provided its estimation fragility is accepted rather than ignored. And if the portfolio is measured against an index, benchmark-relative sizing at least makes the scoreboard and the construction speak the same language.

What breaks portfolios is not choosing the wrong method. It is choosing one method and then quietly operating another, which is what happens when target weights are never compared against actual ones.

Related reading

• Portfolio metrics explained: the hub for the measures used to evaluate a sized portfolio.
• Portfolio construction basics: where sizing sits in the wider construction process.
• Equal weight vs market-cap weight: the two default schemes and how differently they behave.
• Measuring portfolio drift: how sized weights move away from their targets over time.
• Concentration risk in portfolios: measuring concentration by position, sector and factor.

This article is educational. Altys Labs is not a registered research analyst or investment adviser, and nothing here is investment advice or a recommendation to buy, sell, or hold any security.