Loading market quotes…

Subscribe free
Market Fieldnotes11 minFreeGBC Research

Portfolio diversification: correlation and hidden exposure

In short

Portfolio diversification depends on shared exposures and correlations as well as the number of stocks. In a simplified model with equal weights and equal volatilities, N stocks with average correlation r have the same variance as N ÷ (1 + (N − 1) × r) uncorrelated stocks of the same volatility: about 1.2 for ten stocks at 0.8 correlation, or 3.6 at 0.2. This is a risk equivalent, not a count of independent bets. To find hidden exposure you need a map of customers, rate sensitivity and end markets, the correlation of returns, and a stress test with written assumptions.

At the end of 2022 Silicon Valley Bank served thousands of technology and life sciences companies: software, biotech, hardware. Different industries, different products, different cities. The review the Federal Reserve published in April 2023 spells out what they had in common: venture capital-backed companies accounted for more than half of the bank’s deposits, and about 94% of SVBFG’s total deposit amount was uninsured.

When venture funding slowed, those clients received less new capital and started drawing down cash to pay salaries and suppliers, all in the same stretch. On March 9, 2023, the bank lost over $40 billion in deposits, and management expected to lose over $100 billion more the next day. On March 10 the bank failed.

Thousands of clients had behaved like a single client, because they relied on the same source of money. A stock portfolio can do the same thing on a smaller scale: five stocks with different sector labels can share exposure to the same spending budget.

The calculation below compares ten equally weighted stocks at an average correlation of 0.8 with uncorrelated stocks of the same volatility. The result measures equivalent portfolio variance, not the number of independent bets.

What is portfolio diversification?

Investor.gov, the SEC’s site for individual investors, defines diversification as the practice of spreading money among different investments to reduce risk, and adds that within each asset category you should look for investments that may perform differently under different market conditions. A shared factor can still dominate the risk across ten highly correlated positions. Counting tickers alone does not measure how much diversification you have gained.

Count the tickers only after you understand what each one depends on. Sector labels help little, because they classify a company by what it makes, while its risk also depends on who pays it.

How do you tell whether several stocks are the same bet?

The simplest tool is an exposure map. For each position, write down the main customers or end markets, how much the company depends on credit and interest rates, which countries its revenue comes from, and which event last moved the price. Where you don’t know, write “unknown”: filling the gap with whatever the company name suggests is the fastest way to get the map wrong.

Many answers sit in the filings companies make with the SEC. US public companies must disclose any single external customer worth 10% or more of revenue, with the amount, although they don’t have to name it. In our Equity X-Ray reports we go through one company at a time, from the catalyst to the structure of risk. For a swing trading portfolio, one line per stock is enough.

Take a hypothetical $10,000 portfolio split across ten positions. Five of them look unrelated: a chip designer at $1,200, a maker of cooling equipment at $1,000, a power utility at $900, a data center REIT at $800 and a fiber network operator at $600. Five different sectors. Yet a large share of demand for all five depends on the big cloud groups’ data center spending. Together they come to $4,500, 45% of the account.

Five sectors, shared exposureFive sectors, shared exposure. Hypothetical $10,000 portfolio in ten positions: five depend on the same spending budgetFIG. 1 · THE EXPOSURE MAPGB.CAPITALFive sectors, shared exposureHypothetical $10,000 portfolio in ten positions: five depend on the same spendingbudgetStock A · Semiconductors$1,200 · 12%Stock B · Cooling equipment$1,000 · 10%Stock C · Power utility$900 · 9%Stock D · Data center REIT$800 · 8%Stock E · Fiber network$600 · 6%Shared driver: big cloudgroups’ data center budget$4,50045% of a $10,000 accountOther 5 positions · $5,500 · not linkedHypothetical companies, no real stocks. Different sectors share a spending driver.Stock-specific risks remain different.
Figure 1 Five different sectors share a spending driver, creating a combined exposure equal to 45% of the portfolio.

That doesn’t mean all five stocks would fall the same way after a spending cut: contracts, margins, debt and valuations differ. The map tells you where to ask questions, and the answers are in each company’s numbers.

What is stock correlation and how do you calculate it?

Correlation is a number between −1 and +1 that measures how closely two series of returns moved together over a given period. Near +1 they move in the same direction, near 0 they have no linear link, near −1 they move in opposite directions. Here is the calculation on six months of hypothetical returns for two stocks, A and B, both averaging 1% a month.

Correlation between monthly returns of two hypothetical stocks (mean of A and of B = 1%)
MonthABA − meanB − meanProduct
1+3%+4%236
2−2%−1%−3−26
3+5%+6%4520
4−4%−5%−5−630
5+2%+1%100
6+2%+1%100
Sumssquares: 56squares: 7462

Correlation is the sum of the products divided by the square root of the product of the sums of squares: 62 ÷ √(56 × 74) = 0.96. In Excel or Google Sheets, =CORREL(B2:B7,C2:C7) does it all. B swings more than A: correlation measures whether two stocks move together, and the size of the moves is a separate measure, volatility.

The number always needs its period and frequency next to it. Daily and monthly returns can give different values, and the same pair can change a lot from one year to the next. A high correlation doesn’t prove a common cause, and a low one doesn’t rule out a shared exposure: the five stocks on the map may have moved independently for years and still share the same customer. Correlations also tend to rise exactly when you need them low. François Longin and Bruno Solnik, in the Journal of Finance in 2001, found that correlation between international stock markets increases in bear markets and not in bull markets.

How many stocks do you need to diversify a portfolio?

In 1987 Meir Statman calculated that a portfolio of randomly chosen stocks becomes well diversified with at least 30 stocks for an investor who borrows and 40 for one who doesn’t, against the then common view that about ten were enough. Statman was talking about randomly chosen stocks, and a trader’s portfolio is anything but random: it tends to fill up with the strongest stocks of the moment, which often belong to the same theme, as relative strength makes plain.

A simple model shows how much correlation weighs. If N equal-weight stocks each swing 30% a year with an average correlation r, portfolio volatility is 30% × √(1/N + (1 − 1/N) × r), and its variance matches that of N ÷ (1 + (N − 1) × r) uncorrelated stocks of the same volatility. Ten stocks at a 0.8 correlation have a risk equivalent of about 1.2 uncorrelated stocks, with 27.2% volatility against 30% for a single stock. At a 0.2 correlation the same ten stocks have a risk equivalent of about 3.6 uncorrelated stocks and volatility drops to 15.9%.

With high correlation, ten stocks are almost as risky as oneWith high correlation, ten stocks are almost as risky as one. Annual volatility of an equal-weight portfolio of stocks that each swing 30% a yearFIG. 2 · HOW MANY STOCKS YOU NEEDGB.CAPITALWith high correlation, ten stocks are almost as riskyas oneAnnual volatility of an equal-weight portfolio of stocks that each swing 30% a yearcorrelation 0.8correlation 0.5correlation 0.2PORTFOLIO VOLATILITY0%10%20%30%40%151015202530NUMBER OF STOCKS10 stocks, correlation 0.8: 27.2%,1.2 uncorrelated stocks10 stocks, correlation 0.2: 15.9%,3.6 uncorrelated stocksEqual weights and volatilities. r = mean correlation. σ = 30% × √(1/N + (1 − 1/N) × r).Uncorrelated-stock equivalent = N/(1+(N−1)r).
Figure 2 In this equal-weight, equal-volatility model, ten stocks at correlation 0.8 have the same variance as about 1.2 uncorrelated stocks.

In this model, with nonnegative r, portfolio volatility approaches this limit as the number of stocks grows: 30% × √r, which is 26.8% at a 0.8 correlation and 13.4% at 0.2. Going from 10 to 30 stocks at a 0.8 correlation takes volatility from 27.2% to 26.9%.

How do you stress test a portfolio?

A stress test is a calculation with stated assumptions. Apply a hypothetical 20% drop to the five linked positions and hold the rest flat: the loss is $900, 9% of the account, which falls to $9,100.

If the shared customer cuts spending: −$900, 9% of the accountIf the shared customer cuts spending: −$900, 9% of the account. Hypothetical 20% drop in the five linked positions, everything else flatFIG. 3 · A SCENARIO, NOT A FORECASTGB.CAPITALIf the shared customer cuts spending: −$900, 9% of theaccountHypothetical 20% drop in the five linked positions, everything else flatStarting account$10,000Stock A −20%−$240Stock B −20%−$200Stock C −20%−$180Stock D −20%−$160Stock E −20%−$120Other 5 positions, flat byassumption$0Account after the scenario$9,100Hypothetical example for illustration. Holding the other positions flat is a restrictive assumption: in amarket decline they usually fall too.
Figure 3 An arithmetic scenario with stated assumptions, to rerun with different numbers: it is not a forecast.

Holding the other five positions flat is a restrictive assumption, and it should be said out loud. If the spending cut arrives alongside a falling market and the other $5,500 drops 10%, the loss grows to $1,450, or 14.5%, and breaking even takes a 17% gain, as our guide to drawdown math explains. Run the scenario again grouping stocks by interest rate sensitivity instead of by customer: different groups often show up.

If 45% of the account answers to a single budget, ask yourself whether you would open one position of that size today. If the answer is no, the fix goes through the size of the individual positions. Check liquidity too: stocks with different business drivers can become hard to sell at the same moment, when volume dries up and spreads widen.

The formula’s bet count is a risk equivalent in a model with equal weights, equal volatilities and a specified average correlation; it does not count genuinely independent events. A correlation of 0.8 indicates a strong positive linear relationship in the sample, not the percentage of days the stocks move in the same direction. With unequal weights and volatilities, calculate variance from the full covariance matrix.

Frequently asked questions

How many stocks do you need for a diversified portfolio?

It depends on how correlated they are. For randomly chosen stocks, Meir Statman’s 1987 study pointed to at least 30 to 40. In the simplified model used here, ten equally weighted stocks with equal volatilities and an average correlation of 0.8 have the same portfolio variance as about 1.2 uncorrelated stocks of that volatility. This measures a risk equivalent under stated assumptions, not genuine independence.

What does a 0.8 correlation between two stocks mean?

That over the period measured their returns had a strong positive linear relationship in that sample, without specifying how often their daily directions matched. It doesn’t say they moved by the same percentage or that they will keep doing so, and it should always come with the period and frequency of the data used.

How do you calculate the correlation between two stocks in Excel?

Calculate each stock’s returns over the same period, weekly for example, with today’s price ÷ previous price − 1. Then use CORREL on the two columns of returns, never on the prices.

Is an ETF already diversified?

A total stock market fund holds thousands of companies, and Investor.gov cites it as an example of a lot of diversification in one investment. A sector or thematic ETF concentrates exposure instead, and holding it alongside its constituent stocks increases exposure to those holdings. Check their combined portfolio weights.

Does diversification eliminate risk?

No. It can limit losses and reduce the swings in a portfolio, but it doesn’t protect against a fall in the whole market, and correlations tend to rise in declines. That is why you also need to decide how much capital to put at risk, on top of how to spread it.

Sources and further reading

Related research
Interest rates and stocks: the mechanisms and 2022Market Fieldnotes · 05.10.2026Bid-ask spread, volume and liquidity: how to read the costMarket Fieldnotes · 05.10.2026Economic moat: how to spot a competitive advantageMarket Fieldnotes · 05.10.2026TC2000 scan finds no stocks or too many? Check the filtersMarket Fieldnotes · 28.10.2026How to compare ETFs: holdings, overlap, costs and the KIDMarket Fieldnotes · 05.10.2026 Full archive →
FREE · EVERY SUNDAY

Get the next Radar
before Monday’s open.

The first name, the reasoning, and our public research.

FREE · EVERY SUNDAYGet the watchlist