Conviction-Weighted Position Sizing: A Better Portfolio Construction Method

Most portfolio analyzers and portfolio construction software default to two sizing methods: equal weight or volatility-adjusted (risk parity). Both are easy to compute, both are defensible — and both ignore the single most important input a discretionary investor actually has: how confident they are in each idea.

Conviction-weighted position sizing flips that. Instead of asking "how do I divide the pie evenly?", it asks "how much of the pie does this idea actually deserve?" This guide walks through the math, the behavioural advantages, and how to implement it without blowing up your risk budget.

The three common sizing methods

• Equal weight. Every position gets 1/N of the book. Simple, but a 10% position in your highest-conviction idea contributes the same as a 10% position in a name you barely tracked.
• Volatility-weighted (risk parity). Position size is inversely proportional to volatility, so each name contributes equal risk. Mathematically elegant, but it routinely shrinks the names you understand best (often higher-vol growth or single-stock ideas) and grows boring names you have no edge in.
• Conviction-weighted. Position size scales with your stated confidence in the thesis, capped by hard risk limits. The name you'd happily buy twice gets sized like it; the speculative starter gets sized like a starter.

The math

Give every idea a conviction score c_i on a fixed scale (we use 1–5). The target weight for position i is:

w_i = (c_i ^ k) / Σ (c_j ^ k)   subject to   w_i ≤ w_max

The exponent k controls how aggressively conviction translates into size. At k = 0 you get equal weight. At k = 1 sizing is linear in conviction. At k = 2 a 5-conviction idea gets 25x the weight of a 1-conviction idea — usually too aggressive, but useful for showing the shape.

The w_max cap (e.g. 8–12% per single name) is what keeps the method honest: high conviction expresses itself by being at the cap, not by becoming an uncapped concentration bet.

Why it beats equal-weighting and vol-weighting in practice

For a discretionary investor with any real edge, the distribution of outcomes is not uniform across ideas. A handful of theses do most of the work. Equal weighting taxes those winners — you literally own less of your best ideas than you should. Vol-weighting often makes it worse by underweighting exactly the high-vol asymmetric bets that justified the research in the first place.

Conviction-weighting is a way of letting your research show up on the P&L line. If you spent 80% of your work on three names, those three names should dominate sizing — not because the math says so, but because the work says so.

The behavioural advantage

Writing down conviction before you size the trade does two things:

• It forces you to separate the idea (a 5) from the execution moment(do I really want this here, today). Most sizing mistakes are conviction-execution mismatches.
• It gives you an audit trail. Six months later you can ask "did my 5s actually outperform my 2s?" If they didn't, your conviction scoring is calibrated wrong — and that is the most valuable feedback loop in discretionary investing.

How ConvictionOS implements it

ConvictionOS treats conviction as a first-class field on every position, alongside thesis, catalysts, and exit criteria. Target weights are computed from conviction and a per-basket cap, then compared against your actual weights to surface drift in the Rebalancing Center. The Thesis Tracker reminds you to revisit conviction on a cadence — because the worst position-sizing mistake isn't the initial size, it's the size you forgot to update after the thesis changed.

Where to start

1. Pick a conviction scale (1–5 is enough) and write down what each level means.
2. Set a single per-name cap (e.g. 10%) and a per-basket cap (e.g. 40% in any single theme).
3. Start with k = 1 (linear). Only crank k higher once you have evidence your highest convictions actually outperform.
4. Re-score conviction on a fixed cadence (monthly or after every earnings print), and let target weights drift accordingly.