Position Sizing Explained: Kelly Criterion, Volatility-Adjusted Sizing, and Portfolio Risk Management

Most investors spend the majority of their research time on stock selection - figuring out which companies to own. They spend far less time on position sizing - figuring out how much of each company to own. This is a significant error. The evidence from both quantitative finance and the track records of professional investors suggests that sizing decisions drive a substantial fraction of long-run portfolio performance.

Getting an investment right (the company outperforms) but sizing it too small produces a fraction of the potential return. Getting an investment wrong but sizing it too large produces catastrophic loss. The mathematical compounding consequences of sizing errors are asymmetric: a single oversized position that loses 80% can destroy years of returns from correctly sized winners.

This guide covers the major position sizing frameworks available to a retail investor, from simple equal-weighting to Kelly-based conviction sizing to volatility-adjusted approaches, and shows how to integrate them into a coherent portfolio construction process.

Why Position Sizing Matters More Than Most Investors Think

Consider two investors with identical security selection skill. Both pick stocks that outperform the market by 5 percentage points annually over 10 years. Investor A uses equal weighting and never deviates from it. Investor B sizes positions based on conviction, with top ideas representing 15 to 20% of the portfolio and less-certain ideas at 2 to 3%.

If Investor B's sizing correctly reflects the actual distribution of their edge - if the high-conviction positions truly outperform by more than the lower-conviction ones - Investor B will compound substantially faster than Investor A over time. The compounding of good sizing decisions is powerful and cumulative.

The reverse is equally true. If Investor B's large positions systematically perform worse than their small ones (a warning sign of overconfidence), their returns will fall far below Investor A. Sizing that does not reflect genuine edge amplifies mistakes rather than advantages.

Equal Weighting: The Baseline

How It Works

Equal weighting is the simplest sizing framework: every position receives the same allocation. If you hold 20 stocks, each gets 5% of the portfolio. New positions are sized at the same fraction as all other positions.

The Case For It

Equal weighting has more to recommend it than it appears. Research on equal-weighted stock portfolios has shown they often outperform cap-weighted indexes over long periods. This is partly because equal weighting provides automatic rebalancing: as winners appreciate, you trim them, and as losers decline, you add to them. This produces a systematic sell-high/buy-low discipline.

Equal weighting also removes the need to make judgments about relative conviction - judgments that research suggests investors make poorly, for reasons connected to overconfidence and anchoring. If you do not trust your ability to assess your own edge accurately, equal weighting avoids the compounding of sizing errors.

The Case Against It

Equal weighting treats a position where you have exceptional conviction and deep research identically to a position where you have moderate confidence and surface-level analysis. This is only rational if you believe all your investment ideas have the same expected value and risk profile - an assumption that most serious investors would reject.

Conviction Weighting: Sizing by Expected Edge

The Core Concept

Conviction weighting allocates more capital to positions where the investor's analysis suggests greater expected return per unit of risk. A position where the model estimate shows 40% upside to fair value, with high confidence in that estimate and a durable business, might receive 3 to 5 times the allocation of a position where the differential is 15% and the business is more uncertain.

How to Define Conviction Operationally

The challenge with conviction weighting is that "conviction" is a subjective feeling that is easily contaminated by recent price action, narrative appeal, and overconfidence. The most disciplined form of conviction weighting anchors sizing to specific, measurable factors:

Valuation differential: How large is the estimated gap between current price and estimated intrinsic value? A larger differential implies a larger margin for error and a higher expected return even under pessimistic scenarios.

Estimate confidence: How narrow is the range of reasonable fair value estimates? A business with highly predictable cash flows (recurring revenue, long-term contracts, dominant market position) supports a tighter valuation range than a cyclical commodity producer. Wider uncertainty should mean smaller size.

Thesis maturity: Is this a fresh idea where your analysis is still developing, or a position you have held and monitored for years with deep institutional knowledge of the business? Newer positions deserve smaller initial sizing.

Variant perception: Do you hold a view that is materially different from the market consensus? If so, where does your analysis diverge, and what is your evidence for the divergence? Without a clear answer, the "conviction" may be noise.

The Kelly Criterion: Mathematical Sizing

The Formula Revisited

The Kelly Criterion, developed by John Kelly in 1956, calculates the fraction of capital that maximizes the long-run growth rate of wealth. For investments with continuous returns rather than simple win/lose outcomes, the formula is often expressed as:

f = (expected return) / (variance of return)

Or in edge-odds form: f = (bp - q) / b, where b is the net profit per unit risked, p is the estimated probability of success, and q is the probability of failure.

For a worked example: an investor analyzes a software company and estimates:

• 65% probability the stock gains 60% over three years (thesis plays out)
• 25% probability the stock is flat (thesis stalls but no permanent loss)
• 10% probability the stock loses 50% (thesis is wrong, business deteriorates)

Expected return = (0.65 x 0.60) + (0.25 x 0.00) + (0.10 x -0.50) = 0.39 + 0 - 0.05 = 0.34 or 34% cumulative

Variance calculation requires more steps, but the key intuition is: the Kelly fraction scales up with expected return and scales down with variance. High expected return and low variance justify a large position. Lower expected return or high variance justify a smaller one.

Half-Kelly in Practice

Full Kelly sizing is theoretically optimal but practically extreme. A full-Kelly portfolio can require position sizes of 30 to 50% of capital in high-conviction situations, and it will produce drawdowns that most investors find psychologically unbearable even if mathematically justified.

Professional investors who use Kelly-based approaches - Ed Thorp, Bill Gross, and others in the quantitative tradition - standardly use half-Kelly or less. Half-Kelly sacrifices roughly 25% of long-run growth rate relative to full Kelly but cuts volatility approximately in half.

A practical implementation for a retail investor:

1. Estimate the edge for each position using the probability-weighted return calculation above.
2. Calculate the full Kelly fraction.
3. Multiply by 0.5 (half-Kelly) or 0.33 (third-Kelly) depending on how confident you are in your probability estimates.
4. Cap any single position at a maximum threshold (typically 15 to 20% for most investors) regardless of what Kelly suggests.

The cap is important. Kelly assumes you know your edge accurately. In reality, your probability estimates will often be overconfident. The cap prevents a single overconfident position from being catastrophic.

Volatility-Adjusted Position Sizing

The Risk Parity Concept

Equal weighting by dollar is not the same as equal weighting by risk. A position in a volatile small-cap biotech company contributes far more volatility to a portfolio than the same dollar amount in a stable consumer staples company. Volatility-adjusted sizing attempts to equalize the risk contribution of each position rather than the dollar contribution.

The simplest form: calculate the annualized standard deviation (historical volatility) of each position. Divide a target risk budget by that volatility to get the position size.

If your target volatility contribution per position is 1% of portfolio volatility, and a stock has annualized volatility of 40%, the position size is 1% / 40% = 2.5% of the portfolio. A stock with 20% volatility would receive 5% of the portfolio to achieve the same risk contribution.

Volatility-Adjusted Sizing Table

This approach naturally limits exposure to high-volatility names without requiring explicit conviction judgments about each one.

Blending Conviction and Volatility

The most sophisticated retail investors blend conviction sizing with volatility adjustment. The process:

1. Start with a conviction-based sizing estimate (from Kelly or qualitative assessment).
2. Adjust for volatility: if the conviction sizing suggests 10% but the stock has 60% annualized volatility, scale back toward 5 to 6%.
3. Check correlation: if the position is highly correlated with existing large positions, scale back further to avoid duplicating risk exposure.
4. Apply a hard cap on any single name.

Maximum Drawdown and Position Sizing

Why Drawdown Matters Separately from Volatility

Volatility measures how much a portfolio moves around. Maximum drawdown measures the peak-to-trough decline in portfolio value over a given period. These are related but distinct concepts.

A portfolio can have moderate annualized volatility but a catastrophic drawdown if one large position declines 80% or more. This is relevant to position sizing because the mathematics of loss recovery are non-linear: a 50% loss requires a 100% gain to recover. A 75% loss requires a 300% gain to recover.

A single position that represents 25% of your portfolio, if it goes to zero, creates a 25% drawdown. Recovering from that requires a 33% gain from the remaining portfolio before you are back to even. If the loss takes three years to develop (as many investment theses do), you may have lost three years of compounding time as well.

This math argues for hard caps on single positions. Most practitioners suggest 15 to 20% as the maximum for any individual name in a concentrated portfolio, and lower for early-stage or speculative positions where zero is a realistic outcome.

Stop-Loss Rules and Their Tradeoffs

Some investors use stop-loss rules as a sizing control: if a position falls more than 15 or 20%, it is exited regardless of the thesis. This limits maximum drawdown mechanically.

The problem with stop-losses in a thesis-driven investment approach is that short-term price declines are often disconnected from fundamental value. A 20% decline might represent a better entry opportunity if the business is intact. Exiting on price alone forces you to conflate "the stock fell" with "my thesis is wrong" - which is often not the case.

A more nuanced approach: distinguish between thesis-driven exits (the fundamental case has deteriorated) and loss-limit exits (the position has reached a maximum acceptable loss, regardless of the thesis, because capital preservation matters). Set explicit thesis-driven review triggers based on fundamental metrics, not price alone.

Correlation Between Positions and Its Impact on Portfolio Risk

Why Correlation Changes the Math

When building a portfolio, individual position sizes do not simply add up to a risk total. The correlation between positions determines how much of each position's individual risk offsets versus compounds.

Two positions with 30% individual volatility and correlation of +0.9 (highly correlated) together contribute nearly 60% volatility to the pair. The same two positions with correlation of -0.3 contribute about 30% to the pair. The diversification benefit of holding two positions is entirely dependent on their correlation.

For a concentrated portfolio, this means simply counting positions understates risk if those positions are highly correlated. A 10-stock portfolio where all 10 stocks are technology companies with similar macro exposures may behave like a 3 to 4 stock portfolio from a diversification standpoint.

Building a Sizing Framework with Correlation in Mind

A practical approach to correlation-aware sizing:

Identify factor exposures: List the major sensitivities of each position (market beta, sector, interest rate sensitivity, dollar sensitivity, economic cycle sensitivity). Group positions by their dominant exposure.

Limit cluster concentration: Within any single factor cluster (e.g., high-growth technology, rate-sensitive financials, commodity producers), limit the total allocation to that cluster. Even if each individual position passes a sizing test, the cluster as a whole should not represent more than 30 to 40% of the portfolio.

Seek genuine diversifiers: Positions that behave differently from each other in different environments provide real diversification benefit. They do not need to be in different industries - a technology company selling to government clients may behave very differently from one selling to consumer discretionary sectors.

Putting It Together: A Practical Sizing Framework

The following framework is designed for a self-directed investor managing a concentrated stock portfolio of 10 to 25 positions.

Step 1 - Categorize the position by conviction tier. Assign each position to a tier based on depth of research, estimated valuation differential, and confidence in the thesis.

Step 2 - Apply Kelly-based sizing. Estimate expected return under base, bull, and bear scenarios with probabilities. Calculate the full Kelly fraction. Apply half-Kelly. Cap at the tier maximum.

Step 3 - Adjust for volatility. If the volatility-adjusted size is materially smaller than the Kelly size, bring the position down. Higher-volatility names require smaller sizes to contribute comparable risk to the portfolio.

Step 4 - Check correlation. Review how the position correlates with existing large positions. If it adds concentrated exposure to a factor cluster that is already large, reduce the size or replace a correlated position with this one.

Step 5 - Apply hard caps. No single position above 15 to 20%. No single factor cluster above 30 to 40%. Speculative positions (zero is a realistic outcome) capped at 2 to 3%.

Key Takeaways

• Position sizing is as important as security selection for long-run portfolio returns. Correct sizing amplifies edge; incorrect sizing amplifies mistakes.
• Equal weighting removes the need to assess relative edge and protects against overconfidence, but it treats high-conviction and low-conviction ideas identically, which is rarely optimal.
• Conviction weighting allocates more to positions with larger estimated value differentials, higher confidence in the estimate, and more mature thesis development. It requires honest self-assessment.
• The Kelly Criterion maximizes long-run wealth growth mathematically but requires accurate edge estimation. Half-Kelly or third-Kelly is standard practice because edge estimates are imprecise.
• Volatility-adjusted sizing equalizes risk contribution rather than dollar contribution. Higher-volatility positions receive smaller allocations to avoid disproportionate contribution to portfolio risk.
• Drawdown mathematics are non-linear: a 50% loss requires a 100% gain to recover. Hard position caps prevent single-position failures from requiring extreme recovery gains.
• Correlation between positions determines how much of each position's individual risk compounds versus diversifies. Build a concentrated portfolio with attention to which positions actually behave differently from each other.
• A practical sizing framework integrates conviction tier, Kelly-based sizing, volatility adjustment, correlation review, and hard caps into a systematic process applied consistently before and during each position.