Skip to content
Menu
Finabase Learn

Trading · Risk management

Correlated positions: when five trades are really one bet

How correlation between open trades concentrates risk, how to count the independent bets in a book, and what eight stocks did on the worst days of 2024 to 2026.

On this page
  1. What correlation measures
  2. From correlation to the number of bets
  3. The worst days in real data
  4. What to do about it
  5. Limits of the measure
  6. Correlation and related ideas
  7. How Finabase does it
  8. Frequently asked questions
  9. Summary
  10. Check your understanding

A trader risks 1% on each of five trades and feels diversified. The five are a chip maker, another chip maker, a cloud software company, a social media company and a phone maker. On the day the market drops 4%, all five hit their stops. The account loses 5% in a session, on trades that were each sized to lose 1%.

Nothing went wrong with the sizing of any single position. What failed was the assumption that five trades are five separate risks. Stocks that move together behave like one large position. This lesson shows how to measure that with correlation, how to convert a list of positions into a number of independent bets, and what a real set of stocks did on the worst days of the last two years.

What correlation measures

The correlation coefficient compares two series of returns. A value of +1 means they always move in the same direction by proportional amounts. Zero means the moves are unrelated. A value of −1 means they move in opposite directions. Correlation coefficient (CORREL): how two series move together gives the formula and Correlation in a portfolio: what a correlation matrix tells you applies it to asset classes in a long-term portfolio.

For a trader the measure answers a narrower question: if this position has a bad day, how likely is it that my other positions have one too?

The figure shows the answer for eight widely held US stocks over the two years to November 2026, using daily percentage changes.

Correlation of daily returns, eight US stocks, 2024-11 to 2026-11 NVDA NVDA AMD AMD MSFT MSFT META META AAPL AAPL JPM JPM XOM XOM KO KO 1 0.53 0.40 0.42 0.29 0.33 0.02 −0.26 0.53 1 0.29 0.34 0.26 0.29 0.02 −0.22 0.40 0.29 1 0.37 0.31 0.29 −0.03 −0.11 0.42 0.34 0.37 1 0.33 0.36 −0.04 −0.12 0.29 0.26 0.31 0.33 1 0.40 0.14 0.13 0.33 0.29 0.29 0.36 0.40 1 0.13 0.03 0.02 0.02 −0.03 −0.04 0.14 0.13 1 0.13 −0.26 −0.22 −0.11 −0.12 0.13 0.03 0.13 1 Shaded darker: 0.50 or more. Lightly shaded: 0.25 to 0.49. Correlation of daily percentage changes.
Correlation of daily returns, eight US stocks, November 2024 to November 2026 (500 trading days). Source: Finabase chart data; correlations computed by Finabase.
Pair or group Correlation of daily returns
NVDA and AMD (two chip makers) 0.53
NVDA and META 0.42
NVDA and MSFT 0.40
AAPL and JPM 0.40
Average of the five technology stocks 0.35
NVDA and XOM 0.02
NVDA and KO −0.26
Average of all 28 pairs 0.18

Two points are worth noticing. The strongest link is inside one industry: the two chip makers. And each of the five technology stocks and the bank had a correlation above 0.5 with the S&P 500 fund SPY (from 0.54 to 0.69). Each of them is partly a bet on the market as a whole, which is the reason they are correlated with each other.

From correlation to the number of bets

The risk of several positions held together depends on their correlations. For N positions of equal risk, each with the same standard deviation of result and the same correlation ρ with every other, the variance of the total follows from Markowitz's formula for a sum, and it gives a compact measure.

How many independent bets a set of equal positions is worth 0 2 4 6 8 10 12 1 2 3 4 5 6 7 8 9 10 11 12 average correlation 0.0 average correlation 0.2 average correlation 0.4 average correlation 0.6 average correlation 0.8 Number of open positions, each with the same risk Effective number of independent bets = N ÷ (1 + (N − 1) × correlation)
Effective number of independent bets = N ÷ (1 + (N − 1) × correlation), for equal-risk positions. Calculation by Finabase.

The lines flatten, and that is the second lesson of the formula. As N grows, the effective number of bets approaches 1 ÷ ρ and stops. At an average correlation of 0.4, ten positions are worth 2.2 bets and a hundred would be worth 2.5. Diversifying within a correlated group runs out quickly. Only positions with low correlation to the rest add much.

What it does to the size of a bad day

The same arithmetic gives the spread of the combined result. If each of six positions has a result with a standard deviation of 1R, the total has a standard deviation of:

Average correlation Standard deviation of six positions together Compared with independent
0 2.45R 1.0×
0.2 3.46R 1.4×
0.4 4.24R 1.7×
0.6 4.90R 2.0×

The planned total loss is the same in every row: 6R if every stop is hit (Portfolio heat: adding up the risk of every open trade). What changes is how often the outcome is near that extreme. With unrelated positions, losing all six together is a rare coincidence. At a correlation of 0.6 it is an ordinary bad day.

The worst days in real data

Averages hide the days that matter. The table lists the ten largest one-day falls of SPY in the two-year sample and what the eight stocks did on each.

Date SPY Stocks that fell (of 8) Average of the eight Average of the five technology stocks
Apr 4, 2025 −5.9% 8 −6.4% −6.4%
Apr 3, 2025 −4.9% 7 −5.9% −7.5%
Nov 5, 2026 −4.4% 8 −5.3% −7.1%
Apr 10, 2025 −4.4% 7 −4.4% −5.5%
Dec 18, 2024 −3.0% 8 −2.4% −2.7%
Oct 10, 2025 −2.7% 7 −3.1% −4.4%
Mar 10, 2025 −2.7% 6 −2.9% −4.3%
Jun 5, 2026 −2.6% 6 −3.0% −5.3%
Apr 21, 2025 −2.4% 8 −2.2% −2.9%
Apr 16, 2025 −2.2% 7 −3.3% −5.1%

On every one of the ten days at least six of the eight fell, and on four days all eight did. That includes Exxon and Coca-Cola, whose correlation with the technology stocks over the whole period was close to zero or negative. A low average correlation did not prevent them from falling on the days the whole market was sold.

The five technology stocks fell more than the market on nine of the ten days, by an average of 5.1% against 3.5% for SPY. Five long positions in that group would have taken their largest losses on the same handful of days.

A large body of research describes the same thing: correlations between stocks and between stock markets tend to be higher in sharp declines than in calm periods. Longin and Solnik documented it for international markets. The practical reading is that the diversification you measure in ordinary times is larger than the diversification you get when you need it.

What to do about it

None of this says how many positions to hold. It gives a way to count them honestly.

  1. Group positions that share a driver. Same industry, same theme, same sensitivity to interest rates or to one commodity. A long position in a stock and in the sector fund that holds it is one group. So are five stocks bought on the same breakout signal on the same day.
  2. Give each group one risk allowance. If the per-trade risk is 1% and a group is treated as one bet, three positions in it share 1%, or a somewhat higher group cap, not 3%. The Turtle rules did exactly this: 4 units for one market, 6 for closely correlated markets together, 10 for loosely correlated ones.
  3. Count direction. A book that is entirely long is one bet on the market on its worst days, as the table shows. Long and short positions in related stocks offset part of that common risk, though a short carries risks of its own (Short selling explained: borrowing, fees, rules and unlimited risk).
  4. Check with returns, not labels. Sector names mislead. Apple's correlation with the bank JPMorgan in this sample (0.40) was higher than with Microsoft (0.31). Comparing the price paths of two holdings over a few months is a quick test.
  5. Re-measure. Correlations drift. The values on this page describe one two-year window.

Limits of the measure

  • Correlation is an average over the period chosen. A different window gives different values, and a calm window understates what happens in a sell-off.
  • It measures direction and consistency, not size. Two stocks can be highly correlated while one moves three times as far as the other. Sizing by volatility (Position sizing: how many shares to buy for a set risk) deals with that part.
  • The effective-bets formula assumes equal risk and equal correlations. Real books have neither. It is a rule of thumb for orders of magnitude, not a precise count.
  • Daily data misses overnight links. Two stocks that report earnings the same evening, or react to the same news before the open, can gap together in a way their ordinary daily correlation does not predict.
  • Low correlation is not protection on market-wide days. The table above is the evidence.
Idea What it tells a trader
Correlation between positions How often they win and lose together
Beta How far a stock tends to move for a 1% market move
Portfolio heat The total planned loss if every stop is hit
Diversification How combining unrelated holdings lowers overall swings

Heat gives the size of the worst planned outcome. Correlation gives its likelihood. A book needs both numbers.

How Finabase does it

Finabase does not print a correlation matrix for a watchlist or a portfolio. Several tools show the same information in other forms.

  • Comparison. The comparison tool draws several symbols on one chart as percentage change from a common start. Lines that rise and fall together over a few months are correlated; this is the quickest visual check before adding a position to a group.
  • Heatmap and sectors. The heatmap and the sector page show which groups moved together on a given day. On a broad sell-off the whole map is one color, which is the table on this page in picture form.
  • Portfolio. A portfolio lists holdings and their values, so the share of the account in one industry can be added up.

The correlations and the worst-day table here were computed from frozen daily bars by the figure script named in the sources.

Frequently asked questions

How many correlated positions is too many?

There is no fixed number. The formula on this page shows that beyond three or four positions in a group with an average correlation of 0.4 or more, each additional position adds risk without adding much diversification. Many traders therefore cap the combined risk of a group instead of the count.

Are stocks in different sectors uncorrelated?

Usually less correlated, not uncorrelated. In this sample the average correlation across all eight stocks was 0.18, and all of them fell together on four of the ten worst market days. Most stocks share exposure to the market itself.

Does hedging with a short position remove correlation risk?

It reduces the shared market component. A long position in one chip maker and a short in another leaves the difference between the two companies as the main risk. That difference can still be large, and the short side adds borrowing costs and the risk of a squeeze.

What correlation is considered high for trading purposes?

A common working scale treats daily-return correlations above about 0.5 as high for individual stocks, 0.25 to 0.5 as moderate and below 0.25 as low. The cut-offs are conventions. What matters is that the higher the figure, the more the positions should be sized as one.

Summary

Positions that move together concentrate risk even when each is sized carefully. For N equal positions with average correlation ρ, the book is worth N ÷ (1 + (N − 1) × ρ) independent bets, which stops growing at 1 ÷ ρ however many positions are added. Over the two years to November 2026 large technology stocks had daily correlations of roughly 0.3 to 0.5 with each other, and on the market's ten worst days at least six of eight stocks from four sectors fell together. Treating related positions as one bet with one risk allowance, checking direction as well as sector, and remembering that correlations are highest in declines keeps a diversified list of trades from behaving like a single oversized one.

Check your understanding

4 questions on this lesson. Each answer comes with the reason, and nothing is scored.

  1. A trader holds 5 positions of equal risk whose average correlation is 0.5. About how many independent bets is that?

    1. 5
    2. 2.5
    3. 1.7
    Show the answer

    C. 1.7 Effective bets = N ÷ (1 + (N − 1) × ρ) = 5 ÷ (1 + 4 × 0.5) = 5 ÷ 3 ≈ 1.7. Five related trades concentrate risk almost as much as two unrelated ones.

  2. At an average correlation of 0.4, what is the most independent bets any number of equal positions can be worth?

    1. About 2.5
    2. About 10
    3. There is no limit
    Show the answer

    A. About 2.5 As N grows the formula approaches 1 ÷ ρ = 1 ÷ 0.4 = 2.5. Adding more positions from a correlated group soon stops adding diversification.

  3. Over two years a consumer staples stock had a correlation near zero with a group of technology stocks. On the market’s worst days it fell with them. What does this show?

    1. The correlation was calculated wrongly
    2. An average over all days understates how much stocks move together in sharp market declines
    3. Staples stocks are riskier than technology stocks
    Show the answer

    B. An average over all days understates how much stocks move together in sharp market declines Correlation is an average over the whole period. On market-wide sell-offs most stocks fall together, so the diversification measured in ordinary times is larger than what is delivered on the worst days.

  4. Which is the more reliable way to decide whether two holdings belong to one risk group?

    1. Check whether they have the same sector label
    2. Compare how their prices have actually moved against each other over recent months
    3. Check whether they have similar share prices
    Show the answer

    B. Compare how their prices have actually moved against each other over recent months Labels mislead. In the lesson’s sample Apple’s correlation with JPMorgan (0.40) was higher than with Microsoft (0.31). Returns, not sector names, show what moves together.

Sources and further reading

  1. Harry Markowitz, “Portfolio Selection,” Journal of Finance 7(1), 1952 — the variance of a sum of correlated holdings
  2. Attilio Meucci, “Managing Diversification,” Risk 22(5), 2009 — the effective number of bets
  3. François Longin and Bruno Solnik, “Extreme Correlation of International Equity Markets,” Journal of Finance 56(2), 2001
  4. Curtis Faith, The Original Turtle Trading Rules (2003) — limits for closely and loosely correlated markets
  5. Finabase chart data — daily bars of NVDA, AMD, MSFT, META, AAPL, JPM, XOM, KO and SPY, November 18, 2024 to November 17, 2026; correlations computed by Finabase