Portfolio Concentration Explained: Diversification Math, Kelly Criterion, and Active Share

There is a persistent tension at the center of portfolio construction theory. On one side: diversify broadly to eliminate idiosyncratic risk. On the other: concentrate in your best ideas to maximize the compounding of superior analysis. Both camps have serious intellectual foundations, and understanding the mathematics behind each is essential for building a coherent portfolio strategy.

This article walks through the quantitative case for diversification, the equally rigorous case for concentration, how to measure concentration with tools like active share and tracking error, and how to think about the tradeoffs between the two approaches for a self-directed retail investor.

The Mathematics of Diversification

How Many Stocks Eliminate Idiosyncratic Risk?

Modern Portfolio Theory, formalized by Harry Markowitz in 1952, established that portfolio risk has two components: systematic risk (market risk that cannot be diversified away) and idiosyncratic risk (company-specific risk that can be reduced by holding multiple uncorrelated assets).

The key insight is that idiosyncratic risk declines rapidly with the first few positions added, then the marginal benefit of each additional position shrinks. The math works as follows.

If each stock has a standard deviation of roughly 35% annually (reasonable for a mid-cap individual stock) and the pairwise correlation between stocks averages around 0.30 (a realistic assumption for a diversified sector mix), the portfolio standard deviation as a function of number of positions looks like this:

The numbers show why the standard textbook advice is "20 to 30 stocks." Moving from 1 to 20 positions reduces volatility by more than half. Moving from 20 to 100 positions reduces it by only another 1.5 percentage points. The bulk of the diversification benefit is captured in the first 20 to 30 names.

What remains after 20 to 30 stocks is almost entirely systematic (market) risk. No amount of diversification removes it. You can reduce it through asset allocation - holding bonds, real estate, cash - but within equities alone, 20 to 30 positions captures most of what diversification has to offer.

The Caveat: Correlation Is Not Static

The academic model assumes average pairwise correlations. In practice, correlations rise sharply during market stress. In the 2008 financial crisis, assets that appeared uncorrelated during normal markets moved in the same direction simultaneously. This is called correlation breakdown, and it means that diversification protects you less exactly when you want it most.

This does not invalidate diversification as a strategy, but it does mean the protection from holding 30 names is easy to overestimate if those 30 names all share the same macro sensitivities (all technology, all rate-sensitive, all China-exposed).

The Case for Concentration

Buffett and Munger on Diversification

Warren Buffett and Charlie Munger have been among the most vocal critics of excessive diversification. Buffett has said that broad diversification is "required when investors do not understand what they are doing." His portfolio has frequently been concentrated in fewer than 15 to 20 names, with a large fraction of the value in just a handful of positions.

The logic behind this view is straightforward. If an investor has genuine analytical skill - the ability to identify businesses trading below intrinsic value with reasonable reliability - then spreading capital across 80 or 100 names reduces the weight in their best ideas and increases the weight in their less-certain ones. Diversification provides protection against ignorance. For an investor who has done rigorous work, it dilutes the advantage of that work.

This argument is not just opinion. Several studies of institutional fund performance have found that the highest-conviction positions in active managers' portfolios (typically their largest 5 to 10 holdings) outperform the market by a wider margin than their full portfolios. The additional positions, added for diversification or career-risk reasons, drag on performance. The implication is that if managers simply held their best ideas, they would do better.

The Kelly Criterion: The Mathematics of Optimal Bet Sizing

John Larry Kelly Jr. published his formula in 1956 while working at Bell Labs, developing it as a solution to information theory problems. Its application to gambling and investing was recognized quickly. The Kelly Criterion tells you the optimal fraction of your capital to allocate to a bet with a known edge.

The formula is:

f = (bp - q) / b

Where:

• f is the fraction of capital to allocate
• b is the net odds received (profit per unit staked)
• p is the probability of winning
• q is the probability of losing (q = 1 - p)

For a simple example: you have an investment where you estimate a 60% probability of a 50% gain and a 40% probability of a 20% loss.

f = (0.5 x 0.6 - 0.4) / 0.5 = (0.30 - 0.40) / 0.5 = -0.20 / 0.5 = -0.20

In this case Kelly actually says avoid the investment. The edge is negative.

Now change the parameters: 60% probability of a 100% gain, 40% probability of a 40% loss.

f = (1.0 x 0.6 - 0.4) / 1.0 = (0.60 - 0.40) / 1.0 = 0.20

Kelly says allocate 20% of capital to this investment.

Why Half-Kelly Is Standard Practice

Full Kelly maximizes the long-run growth rate of capital mathematically. But it comes with brutal volatility. In the example above, a 20% Kelly bet will generate periods where the portfolio falls 40% or more. Most investors, including professionals, cannot psychologically tolerate full Kelly drawdowns.

The standard practical recommendation is half-Kelly: allocate half what the formula suggests. This gives up some long-run growth rate in exchange for substantially lower volatility and drawdown severity. Ed Thorp, the mathematician who pioneered Kelly-based betting and investing strategies, used half-Kelly as his standard approach.

An important constraint: Kelly sizing assumes you know your edge precisely. In reality, investment edge estimates are uncertain. If you are overconfident in your probability estimate, Kelly will tell you to oversize, and the result is ruin. The correct response to uncertainty in your edge estimate is to use a fraction of Kelly, not full Kelly.

Active Share: Measuring Portfolio Concentration Against the Benchmark

What Active Share Is

Active share, developed by Martijn Cremers and Antti Petajisto in a widely cited 2009 paper, measures how different a portfolio is from its benchmark index. It is calculated as half the sum of the absolute value of the differences between the portfolio weight and the benchmark weight for every security.

If a portfolio held exactly the S&P 500 in index proportions, its active share would be 0. If a portfolio had zero overlap with the S&P 500 - holding nothing that is in the index - its active share would be 100.

Most "active" mutual funds have active share between 60 and 90. True concentrated active managers often have active share above 90. Closet indexers - funds that charge active fees but hold a portfolio close to the benchmark - typically have active share below 60.

What the Research Found

Cremers and Petajisto found that funds with active share above 80 to 90 outperformed their benchmarks by a statistically significant margin before fees. Funds with low active share were indistinguishable from the index, net of fees - which means they delivered benchmark returns minus the cost of active management.

This finding has a clear implication: if you are going to pay for active management (including your own time as a self-directed investor), you need to hold a portfolio that is meaningfully different from the index. An almost-index portfolio with extra costs is the worst outcome.

Tracking Error: The Risk Side of Active Share

Tracking error measures how much a portfolio's returns deviate from its benchmark on a period-by-period basis. It is the standard deviation of the difference between portfolio returns and benchmark returns.

A portfolio with 2% tracking error will, in most years, return within roughly 2% of the benchmark in either direction. A portfolio with 10% tracking error could easily be 10% above or below the benchmark in a given year.

High active share and high tracking error go together, but they measure different things. Active share measures how different the holdings are. Tracking error measures how different the returns are over time.

For a self-directed investor, tracking error matters psychologically as much as mathematically. A concentrated portfolio can underperform its benchmark for 2 to 3 years even if its long-run edge is real. Investors who cannot tolerate multi-year periods of underperformance often abandon concentrated strategies at exactly the wrong moment, realizing the worst of both worlds: concentrated risk during the underperformance, then a switch to diversification before the outperformance arrives.

Concentration vs. Diversification: The Decision Framework

The Core Question: What Is the Source of Your Edge?

The right level of concentration depends on the honest answer to this question. If you have a genuine, articulable analytical edge - an ability to identify undervalued businesses that the market is mispricing - then concentration in your highest-conviction ideas is rational. The more positions you add beyond your best ideas, the more you dilute the advantage of that edge.

If you do not have a reliable analytical edge, or if you are investing in areas where picking individual stocks is genuinely difficult (highly efficient markets, complex financials, early-stage companies with uncertain futures), then broader diversification is rational.

Most retail investors fall somewhere in the middle. They may have genuine insight into 5 to 10 companies they know deeply but limited edge in the remaining 40 or 50 positions that make up a "fully diversified" portfolio.

The Practical Middle Ground

A practical framework for a self-directed retail investor who believes they have analytical skill in a limited domain:

Tier 1 - Highest conviction: 5 to 10 positions, each representing 5 to 15% of the portfolio. These are the companies you have analyzed most deeply, where your edge is clearest, and where the valuation differential is largest. This is where Kelly-based sizing is most applicable.

Tier 2 - Research ideas with less certainty: 10 to 20 additional positions, each at 2 to 5% of the portfolio. These are companies with interesting characteristics but where your analytical conviction is lower or the thesis is earlier stage.

Tier 3 - Diversification layer: If desired, broad index exposure for the remaining allocation. This provides market participation in areas outside your analytical domain without diluting your best ideas.

This kind of structure produces active share of 70 to 90 on the stock-picking portion, captures most of the diversification benefit for idiosyncratic risk, and still allows meaningful concentration in the ideas where the investor's edge is strongest.

Correlation and Its Effect on Concentration Risk

One consideration that neither active share nor Kelly directly captures is correlation between positions. A concentrated portfolio of 10 stocks sounds manageable until all 10 are technology companies that move together. In effect, the diversification benefit is much less than the raw position count implies.

When building a concentrated portfolio, think about correlation in at least two dimensions:

Sector and industry correlation: Two companies in the same industry, even if individually researched, will respond similarly to industry-level news. Their correlation will be high.

Factor correlation: Two companies may be in different sectors but share the same sensitivity to interest rates, the dollar, or economic growth. This creates correlation through a common factor rather than a common industry.

A concentrated portfolio with genuine diversification benefit holds companies that respond differently to different economic environments. Some positions that do well in inflationary environments, others in deflationary ones. Some sensitive to economic growth, others defensive. This does not require holding 50 stocks - it requires holding 15 to 20 stocks that genuinely behave differently from one another.

Comparing Approaches

Key Takeaways

• The diversification benefit of adding positions is front-loaded. Moving from 1 to 20 positions eliminates most idiosyncratic risk. Moving from 20 to 100 adds minimal further reduction.
• Concentration makes mathematical sense when you have a genuine, articulable analytical edge. Diversification without edge just means paying the cost of analysis to hold something close to an index.
• The Kelly Criterion provides a mathematical basis for position sizing based on estimated edge and odds. Full Kelly is too volatile for most investors; half-Kelly is a common practical standard.
• Accurate edge estimation is the critical input to Kelly. Overestimating your probability of being right leads to dangerous oversizing. When uncertain, use a fraction of Kelly.
• Active share measures how different your portfolio is from the benchmark. Research suggests high-active-share portfolios have outperformed before fees, while low-active-share portfolios underperform after fees.
• Tracking error is the risk price of active share. Concentrated portfolios can underperform for multi-year periods. If you cannot tolerate this psychologically, moderate your concentration level.
• Correlation between positions reduces the practical diversification benefit of a given position count. Build concentration with attention to which positions actually respond differently to the same macro environment.