How Many Stocks Should I Own? What the Research Says

The honest short answer: the classic studies put the number between 10 and 40. Evans and Archer in 1968 built random portfolios and watched the standard deviation flatten within the first handful of stocks, which became the "10 stocks is enough" rule. Statman in 1987 weighed the benefit against the cost and found at least 30 for a borrowing investor and 40 for a lending one. Campbell, Lettau, Malkiel and Xu in 2001 found the number needed had risen over 1962 to 1997 because individual stocks had become less correlated with each other. Bessembinder in 2018 reframed the question entirely: most stocks lose to Treasury bills over their lifetimes and the market's gain comes from a small minority, so the risk of a small portfolio is missing the winners, not only volatility. And the number changes with what the stocks are, how they are weighted, what the funds next to them hold, and whether the reader's paycheck already depends on one of them.

This post reports what each study measured and found, and where they disagree. It does not tell anyone what to hold.

What the studies actually measured

"How many stocks" is really four different questions, and each study answered a different one.

Evans and Archer, 1968: the origin of "10 stocks"

Evans and Archer drew random portfolios from S&P 500 constituents, sized from one stock up to 40, repeated the draw 60 times per size, and plotted the average standard deviation of each set. The curve fell steeply at first and then flattened. Their conclusion, that the economic justification for going much beyond about 10 securities was doubtful, is where the 10-stock rule of thumb comes from. The measure was standard deviation of equal-weight portfolios drawn at random, so the result describes blind picking, not a portfolio anyone actually holds.

Statman, 1987: 30 to 40 once cost enters

Statman's paper is titled with the exact question and opens by contradicting the 10-stock view. His method was to ask when the marginal reduction in risk from the next stock stops paying for the cost of holding it, using an index fund as the comparison. The abstract states the finding: "a well-diversified portfolio of randomly chosen stocks must include at least 30 stocks for a borrowing investor and 40 stocks for a lending investor." The difference between 10 and 30 is not a disagreement about the shape of the curve. Evans and Archer looked at where the curve flattens visually; Statman priced the remaining gap and found it was still worth closing at 10, 20 and beyond.

Campbell, Lettau, Malkiel and Xu, 2001: the number is not fixed

Campbell, Lettau, Malkiel and Xu split stock volatility into market, industry and firm-level pieces from 1962 to 1997 and found "a noticeable increase in firm-level volatility relative to market volatility." Because the firm-specific piece is what diversification removes, its growth means more holdings are needed to reach the same level of risk. The abstract puts it directly: "correlations among individual stocks and the explanatory power of the market model for a typical stock have declined, while the number of stocks needed to achieve a given level of diversification has increased." A count that fit 1968 data did not fit 1997 data; the number moves with the correlation structure of the market.

Bessembinder, 2018: the count is the wrong question

Bessembinder measured the lifetime buy-and-hold return of every US common stock in the CRSP database from 1926 to 2016 against one-month Treasury bills. Most stocks lost that comparison. The ASU summary of his work states that the best-performing 4 percent of listed companies account for the entire net gain of the US stock market since 1926, while "the remaining 96 percent of stocks collectively matched one-month T-bills." The reason the market still beats T-bills is positive skewness: a few enormous winners offset thousands of losers.

This changes what a small portfolio risks. The earlier studies treated risk as variance, and variance falls fast with the first few stocks. Bessembinder's data says the more expensive risk is missing the handful of companies that produced the return, and a 10-stock portfolio drawn from thousands has a real chance of holding none of them. The ASU page draws the same conclusion: "active strategies, which tend to be poorly diversified, most often underperform."

Why there is no single number

The measure of risk

Standard deviation, the measure in the classic papers, converges quickly. Drawdown and the chance of missing the winners do not. A portfolio can have a standard deviation close to the market's and still lag it by a wide margin over a decade because it never held the few names that mattered. Which of those two failures the reader cares about decides which literature applies.

Correlation between the stocks chosen

Here is the arithmetic the studies share, with a worked illustration. For N equal-weight positions, portfolio variance = average variance / N + (1 - 1/N) x average covariance. The first term is the piece that shrinks as stocks are added. The second is the floor.

Illustration, not a measurement of any real market: set every stock's variance to 1 and every pairwise correlation to 0.3, so average covariance is 0.3.

• 1 stock: variance 1.00, standard deviation 100 percent of a single stock
• 5 stocks: 0.2 + 0.8 x 0.3 = 0.44, standard deviation about 66 percent
• 10 stocks: 0.1 + 0.9 x 0.3 = 0.37, standard deviation about 61 percent
• 20 stocks: 0.05 + 0.95 x 0.3 = 0.335, standard deviation about 58 percent
• 40 stocks: 0.025 + 0.975 x 0.3 = 0.318, standard deviation about 56 percent
• Floor (infinite N): 0.30, standard deviation about 55 percent

At 10 stocks the gap to the floor is about 6 points; at 40 it is about 1.5 points. That is the Evans and Archer curve, and it is why 10 looks like enough on a chart.

Now drop the correlation to 0.15, which is the direction Campbell and co-authors found the market moving:

• 10 stocks: 0.1 + 0.9 x 0.15 = 0.235, standard deviation about 48 percent
• 40 stocks: 0.025 + 0.975 x 0.15 = 0.171, standard deviation about 41 percent
• Floor: 0.15, standard deviation about 39 percent

The gap at 10 stocks is now about 10 points instead of 6. Lower correlation makes diversification more valuable per stock and pushes the count up at the same time. The reverse also holds: ten stocks in one sector, with a pairwise correlation nearer 0.7, sit almost on the floor at 10 because the floor itself is high. The count looks diversified and the portfolio is not. The diversification guide covers correlation across sectors and asset classes in more depth.

Position weights

Every study above used equal weights. Real portfolios are not equal-weighted, and the effect is larger than the count. With unequal weights, the shrinking term becomes the sum of squared weights instead of 1/N, so the effective number of positions is 1 divided by that sum.

Illustration: 13 equal positions have a sum of squared weights of 1/13, an effective count of 13. One position at 40 percent plus twelve at 5 percent has a sum of 0.16 + 12 x 0.0025 = 0.19, an effective count of about 5.3. The statement lists 13 lines; the portfolio behaves like a 5-stock book, and its drawdown is dominated by one name. The concentration problem post works through the same weight math using the Herfindahl index, which is this sum of squared weights under another name.

What the funds next to the stocks hold

A portfolio of 8 stocks plus an S&P 500 fund is not 8 positions. The fund holds roughly 500 companies at market-cap weights, so the look-through count is in the hundreds, and any of the 8 direct holdings that is also a large index constituent is owned twice, with the two weights adding. The same look-through cuts the other way with several funds: an S&P 500 fund, a total-market fund and a Nasdaq-100 fund share their largest names, so three fund lines can amount to one bet on the same ten companies. The ETF overlap post covers how that duplication is measured, and it is why the stock count alone says little about a portfolio that also holds funds.

Employer stock and RSUs

The last variable is outside the brokerage account. An investor whose salary comes from the same company as a large vested RSU position holds that company twice: once as equity and once as income. A layoff and a stock decline at that company tend to arrive together, a correlation no count of other stocks addresses. For many people in tech the employer is the single largest position, often well above the 40 percent in the illustration, and it raises the sum of squared weights more than anything else in the book. The RSU tax post covers the tax side of that position.

The one place the number is written down

Regulated investment companies, the legal form of most mutual funds and ETFs, do have a minimum. Section 851(b)(3) of the Internal Revenue Code requires that at the close of each quarter "at least 50 percent of the value of its total assets is represented by" cash, government securities, other RIC shares and other securities "limited, in respect of any one issuer, to an amount not greater in value than 5 percent of the value of the total assets" and "not more than 10 percent of the outstanding voting securities of such issuer." A second clause caps any single issuer at 25 percent of total assets. Arithmetically, the 5 percent limit means the tested half of the book cannot hold fewer than 10 issuers, and in practice funds hold far more.

This is a tax-status test for funds, not a research finding about individuals. What is notable is that when a number had to be written down, it was a weight cap, not a count: no one issuer above 5 percent on half the assets.

Where that leaves the question

The studies agree on the shape and disagree on the number because they measured different things. Variance flattens within about 10 to 20 equal-weight stocks and keeps improving slowly to 40 and beyond, faster when correlations are low and barely at all when the stocks share a sector. Priced against trading costs, the improvement was still worth paying for at 30 to 40 in 1987. The count needed rose through the 1990s as firm-specific volatility grew. And the return distribution is skewed enough that the expensive risk of a small portfolio is not its volatility but the winners it never held.

Every one of those findings assumes directly held, equal-weight, randomly drawn stocks and nothing else. Real portfolios have a 40 percent position, three overlapping index funds and an employer's stock. For those, the ticker count is the least informative number on the statement; the look-through weight per company is the one that describes the risk.

Seeing the count with the funds included

I build Helm Terminal. It reads the positions in connected brokerage accounts, read-only, or positions typed in by hand, and reports concentration by position, by sector and by the companies shared across funds, so the count includes what the funds hold rather than the number of lines on the statement. For the fund look-through question on its own, the ETF overlap tool is free and needs no account. The free tier includes the portfolio dashboard, AI stock analysis, one monitored thesis and the general morning brief. Pro is $20 a month or $149 a year. None of this is financial advice, and Helm does not tell anyone how many stocks to own; it shows the weights so the reader can see what the number actually is.

Related reading

• The Concentration Problem: Why Your Portfolio Probably Has a Single Point of Failure
• Portfolio Diversification Is Broken (Here's What Actually Works in 2026)
• ETF Overlap: Why Two Funds Own the Same Companies
• Your Portfolio Is More Concentrated Than You Think
• RSU Tax Strategies: A Complete Guide for Tech Employees