Correlation can tell you whether two return series moved together in a selected sample. It cannot tell you that one asset caused the other to move, that the relationship will continue, or that a high-correlation pair will mean-revert on command.

For traders, the useful question is not simply “What is the correlation?” It is “What exactly did I measure, over which period, at what frequency, and what decision am I trying to improve?”

This guide owns the general trading use of correlation: measuring two market relationships, comparing rolling relationships, interpreting hedges, and understanding why correlation alone is not enough for pair trading. For a portfolio-level audit that combines a correlation matrix with weights, volatility, covariance, rolling windows, and stress periods, use the portfolio correlation guide.

Key Takeaways

Pearson correlation ranges from -1 to +1 and describes linear co-movement in a defined sample.
For trading analysis, compare aligned return series, not two raw price charts that may both trend over time.
Correlation is sample-dependent: changing the lookback, bar interval, session, currency basis, or market regime can change the result.
A negative correlation does not guarantee a hedge, and a positive correlation does not prove two positions are economically identical.
High correlation alone is not a complete pair-trading signal. Relative-value strategies also need a defined spread, sizing method, stability test, costs, and an exit rule.
Correlation does not prove causation and can miss nonlinear, asymmetric, or tail relationships.

This article is educational. It does not recommend a security, hedge ratio, pair trade, or portfolio allocation.

What Is Correlation in Trading?

Correlation measures the direction and strength of a linear relationship between two variables. In trading, the variables are usually two aligned return series.

The Pearson correlation coefficient is commonly written as:

r = covariance(X, Y) / (standard deviation of X × standard deviation of Y)

Its value lies between -1 and +1.

Correlation	Statistical meaning	Trading interpretation
Near +1	Strong positive linear relationship in the sample	Returns often moved in the same direction and proportionally enough to create high co-movement
Near 0	Weak linear relationship in the sample	There was little linear co-movement; another relationship may still exist
Near -1	Strong negative linear relationship in the sample	Returns often moved in opposite directions

The coefficient is not a slope. A correlation of 0.70 does not mean Asset B rises 0.70% whenever Asset A rises 1%. It means the two standardized return series had a relatively strong positive linear relationship in the selected observations.

NIST's statistical documentation treats Pearson correlation as a measure of linear association. That limitation matters in markets because relationships can be nonlinear, regime-dependent, or dominated by a few extreme observations.

Calculate Correlation From Returns, Not Just Price Levels

Two assets can both trend upward for years and produce an impressive-looking price relationship even if their day-to-day returns are only weakly related.

For most trading correlation work, start with aligned percentage or log returns.

A simple percentage return is:

return_t = (price_t / price_t-1) - 1

Then calculate correlation between the two return columns.

Example

Suppose two assets have these daily returns:

Day	Asset A	Asset B
1	+0.8%	+0.6%
2	-0.5%	-0.4%
3	+1.1%	+0.9%
4	-0.2%	+0.1%
5	+0.4%	+0.3%

A correlation calculation summarizes the linear co-movement of those observations. It does not tell you why the moves occurred or whether the same relationship will hold next month.

Before calculating anything, freeze the data rules:

price field: close, settlement, midpoint, or another source;
return definition: simple or log;
bar interval: 5-minute, hourly, daily, weekly;
session: regular session, extended hours, or full electronic session;
currency basis;
corporate-action adjustment for equities;
timestamp alignment;
lookback window.

If those rules change, you are measuring a different relationship.

Why Alignment Matters

Correlation requires comparable observations.

If one asset closes at 4:00 p.m. New York time and another market's reference price is taken hours later, the same calendar date may represent different information sets. Intraday correlation can be even more sensitive to asynchronous timestamps and thin trading.

Common alignment errors include:

comparing U.S. equity closes with foreign-market prices from a different information window;
mixing adjusted and unadjusted equity data;
filling missing bars without documenting the method;
comparing 24-hour crypto returns with regular-session stock returns;
using different timezone conventions for two intraday series.

A clean correlation calculation begins with a clean common timeline.

Rolling Correlation Is Usually More Useful Than One Lifetime Number

A full-history coefficient compresses years of observations into one number. That can hide changing market regimes.

A rolling correlation recalculates the coefficient repeatedly over a trailing window. For example, a 60-observation rolling correlation uses observations 1–60, then 2–61, then 3–62, and so on.

The purpose is not to find a universally correct window. It is to see whether the measured relationship is stable enough for your use case.

What to compare

Instead of declaring one window “best,” compare at least:

a shorter window that reacts faster;
a medium window that smooths short-lived noise;
a longer reference window;
a stress or event sample when the relationship matters most.

If a relationship looks strong only in one carefully selected window, treat it cautiously.

Correlation Changes With Market Regimes

Market relationships are driven by shared exposures, policy expectations, liquidity, positioning, and risk sentiment. Those drivers change.

Examples of relationships that can vary materially over time include:

equities versus government bonds;
equity sectors versus the broad index;
commodities versus currencies;
Bitcoin versus technology equities;
two currency pairs sharing a common currency leg.

A relationship can move from positive to weak, from weak to negative, or become unstable without either asset being “wrong.”

This is why a hedge or pair trade should not be justified by one historical correlation coefficient.

Correlation Does Not Equal Causation

If two assets have a correlation of 0.80, several explanations are possible:

Asset A influences Asset B;
Asset B influences Asset A;
both respond to a third driver;
both load on the same market factor;
the relationship is temporary;
a small number of extreme observations created much of the measured association.

Correlation alone cannot distinguish these cases.

For example, two semiconductor stocks may move together because they share broad equity beta, sector exposure, interest-rate sensitivity, or a common earnings cycle. The coefficient does not identify which mechanism dominated.

Treat correlation as a descriptive statistic, not a causal model.

Correlation Is Not the Same as Beta

Correlation and beta answer different questions.

Correlation standardizes both series and measures how closely their returns move together linearly.

Beta measures how sensitive one asset's returns have been to movements in a chosen benchmark.

Two assets can have high correlation but different betas because one is much more volatile than the other. That distinction matters when traders try to construct a hedge or relative-value position.

A high correlation does not tell you the correct hedge ratio.

Correlation Is Also Not the Same as Cointegration

This distinction is critical for pair trading.

Two return series can be highly correlated without their price spread having a stable long-run relationship. Conversely, a statistically useful relative-value relationship may require more than a high Pearson coefficient.

Correlation asks whether two series moved together linearly in the selected sample.

Cointegration asks whether a combination of non-stationary price series forms a more stable long-run relationship.

That means:

high correlation ≠ guaranteed mean reversion

If you want to trade a spread, you need a separate spread hypothesis and evidence for that hypothesis.

How Traders Can Use Correlation Without Overreaching

Correlation is most useful when it changes a specific risk or research question.

1. Detect duplicated directional exposure

Suppose you are considering two trades that both depend heavily on the same market factor. A high recent return correlation is one clue that the trades may fail together.

It is not enough by itself to calculate total portfolio risk, because position size, volatility, options convexity, leverage, and event exposure also matter.

For the complete weighted portfolio workflow, use the portfolio correlation and diversification audit.

2. Compare a proposed hedge with the risk being hedged

A hedge should be evaluated against a defined loss source.

Questions to ask:

What exact exposure am I trying to offset?
Was the historical relationship measured on the same horizon as the risk?
Does the hedge have its own basis, liquidity, volatility, financing, or option risk?
How did the relationship behave during the stress conditions that matter?
What happens if the correlation changes sign?

FINRA notes that diversification can help manage concentration risk but does not eliminate market risk. The same principle applies to a hedge: historical negative correlation is evidence, not insurance.

3. Map shared drivers across markets

Correlation can be a starting point for intermarket analysis.

CME's education on intermarket spreads shows how traders may pair related futures contracts to trade relative performance instead of outright direction. The relationship still has to be tied to the economics and contract specifications of the two legs.

A useful intermarket review asks:

why should these markets be related?
what fundamental driver is shared?
what driver is different?
when has the relationship broken historically?
are both legs equally liquid at the intended time?
what are the margin and transaction-cost implications?

4. Screen candidates for deeper pair analysis

Correlation can be used as an initial screen for pairs that often move together.

But it should not be the final entry rule.

A pair-trading research process should separately define:

the economic rationale for the relationship;
the return correlation and its stability;
the spread or ratio definition;
the sizing or hedge-ratio method;
whether mean reversion is actually supported;
the entry and exit rule;
fees, spread, slippage, borrow, financing, and two-leg execution risk;
out-of-sample validation.

CME's intermarket-spread material also warns that a spread can lose money on both legs. Relative-value does not mean risk-free.

A Better Pair-Trading Example

Assume two related futures or stocks have shown positive return correlation over several samples.

Do not jump directly to:

“Asset A is high relative to Asset B, so short A and buy B.”

Instead build a testable hypothesis:

define the spread using a documented hedge ratio;
calculate the spread historically;
test whether the spread behavior is sufficiently stable for your model;
record how often and how far it diverged;
include execution costs for both legs;
test periods when the relationship structurally changed;
define what invalidates the trade.

If the relationship is only “these two charts look similar,” the research is incomplete.

For a concrete FX example, the GBP/USD and AUD/USD correlation guide shows why two USD-quoted pairs can create shared dollar exposure and a synthetic cross view without guaranteeing neutral risk or a fixed correlation.

How to Calculate Pearson Correlation in a Spreadsheet

A simple spreadsheet workflow is enough for basic two-series analysis.

Put timestamps in Column A.
Put Asset A prices in Column B.
Put Asset B prices in Column C.
Calculate Asset A returns in Column D.
Calculate Asset B returns in Column E.
Remove rows where either return is missing.
Use the spreadsheet CORREL function on the two return columns.
Save the lookback, frequency, data source, and observation count with the result.

Example formula:

=CORREL(D2:D101,E2:E101)

The result is meaningful only with its metadata. “Correlation = 0.64” is incomplete. “Daily close-to-close adjusted returns, 100 aligned observations, ending 2026-08-15, correlation = 0.64” is reproducible.

How to Calculate a Correlation Matrix in Python

For several assets, a return matrix can be useful for screening.

import pandas as pd

prices = pd.read_csv("prices.csv", index_col="date", parse_dates=True)
returns = prices.pct_change().dropna(how="any")
correlation_matrix = returns.corr(method="pearson")
print(correlation_matrix)

Before using the output, verify that the columns are aligned, the price series use compatible adjustments, and the data sample matches the decision you are studying.

For a portfolio-level interpretation of a matrix, including weights and covariance, use the portfolio correlation guide.

How to Read a Correlation Matrix

A correlation matrix places each series on both axes.

The diagonal is 1 because every series is perfectly correlated with itself. The off-diagonal cells show pairwise relationships.

Do not reduce the entire matrix to one rule such as:

“anything above 0.70 is the same trade”;
“anything below 0.30 is diversified”;
“negative correlation is a hedge.”

Those thresholds ignore volatility, sizing, factor exposure, nonlinear risk, and sampling error.

Use the matrix to ask better questions, not to automate a universal decision.

Common Correlation Mistakes

Mistake 1: Treating 0.70 as a universal danger threshold

There is no universal coefficient at which two trades suddenly become “one position.” A 0.70 relationship can matter very differently depending on size, volatility, leverage, and the loss scenario.

Mistake 2: Assuming near-zero correlation means independence

Pearson correlation measures linear association. A nonlinear or state-dependent relationship can exist even when Pearson correlation is near zero.

Mistake 3: Using a negative correlation as a guaranteed hedge

A negative historical relationship can weaken, disappear, or reverse. A hedge also has basis risk, sizing risk, execution risk, and sometimes financing or option-premium cost.

Mistake 4: Using correlation to predict direction

If two assets are positively correlated, that does not tell you whether either one will rise tomorrow.

Mistake 5: Using only one lookback window

A result that appears only in one window may be unstable. Compare several predefined samples rather than selecting the window that makes the relationship look strongest.

Mistake 6: Ignoring outliers

A few crisis observations can dominate covariance and correlation. Inspect the underlying scatterplot and the actual return series.

Mistake 7: Confusing correlation with pair-trade mean reversion

A pair can move together for years and still have a spread that trends instead of reverting. Correlation is not a substitute for spread-stability testing.

Mistake 8: Forgetting two-leg execution risk

A theoretical relationship can disappear after bid-ask spread, slippage, borrow, financing, partial fills, latency, or legging risk.

Correlation and Diversification: Keep the Boundary Clear

Diversification is broader than correlation.

FINRA and Investor.gov describe diversification as spreading exposure across investments and asset classes to help manage concentration risk. Correlation can help diagnose whether holdings have historically moved together, but a correlation coefficient does not by itself determine a suitable allocation.

Portfolio risk also depends on:

weights;
volatility;
covariance;
concentration;
factor exposure;
liquidity;
leverage;
currency exposure;
time horizon;
investor or strategy objectives.

That deeper portfolio analysis belongs to the portfolio correlation matrix and diversification guide.

A Reproducible Correlation Workflow for Traders

Use the following sequence whenever a market relationship matters to a decision.

Step 1: State the decision

Examples:

“Do these two planned trades share too much directional exposure?”
“Has this hedge historically offset the loss scenario I care about?”
“Is this pair worth deeper spread research?”

Step 2: Define the data

Write down symbol, venue, data source, adjustment method, currency, session, bar interval, and lookback.

Step 3: Calculate aligned returns

Do not rely on chart appearance.

Step 4: Calculate Pearson correlation

Store the coefficient and observation count.

Step 5: Inspect a scatterplot

Look for outliers, clusters, nonlinear shapes, and obvious regime splits.

Step 6: Compare rolling windows

Ask whether the relationship is stable or changing.

Step 7: Review stress periods

If the relationship is supposed to reduce risk, inspect the periods when risk actually mattered.

Step 8: Add economic context

Document why the two markets might move together and why that relationship could break.

Step 9: Match the statistic to the use case

exposure check → combine correlation with size and volatility;
hedge → test the target loss and hedge ratio;
pair trade → test spread behavior and execution;
diversification → use the dedicated portfolio audit.

Step 10: Recalculate on a predefined schedule

Do not update only after the relationship fails. Use a schedule tied to the strategy's horizon and data frequency.

Can ChartMini Calculate Correlation for You?

No. ChartMini is a historical chart-replay and trading-practice tool, not a portfolio analytics or statistical-arbitrage platform.

ChartMini does not currently:

calculate live Pearson or rolling correlation;
build a correlation matrix across your positions;
optimize hedge ratios;
test cointegration;
generate pair-trading signals;
recommend uncorrelated assets;
monitor a live portfolio for correlation changes.

You can use ChartMini to practice discretionary decisions on historical charts and review whether your market-context assumptions were useful, but correlation calculations should be performed in a spreadsheet, statistical package, or dedicated analytics workflow.

If you are testing a rule-based relationship, document the rule first and validate it using a reproducible backtesting process.

FAQ

What is a good correlation for trading?

There is no universal “good” value. The useful range depends on the purpose. A pair-trading screen, hedge study, and portfolio-concentration review need different evidence. Always store the sample, lookback, frequency, and use case with the coefficient.

Is 0.8 correlation very high?

A Pearson coefficient of 0.8 indicates strong positive linear co-movement in that specific sample. It does not prove the assets are interchangeable, equally volatile, causally linked, or guaranteed to remain highly correlated.

Does negative correlation make an asset a good hedge?

Not automatically. Historical negative correlation is only one input. A hedge should be evaluated for sizing, basis risk, liquidity, costs, stress behavior, and the exact loss it is intended to offset.

Can two assets be highly correlated but bad for pair trading?

Yes. High return correlation does not guarantee that their price spread or ratio mean-reverts. Pair trading needs a separate relative-value model, spread definition, stability evidence, execution assumptions, and exit rule.

Should I use prices or returns for correlation?

For most trading and portfolio applications, use aligned return series. Raw price levels can create misleading relationships because trending assets may rise together over long periods without strong period-to-period co-movement.

What window should I use for rolling correlation?

There is no universally correct window. Choose windows that match the strategy horizon, predefine them, and compare short, medium, long, and stress samples rather than selecting a lookback after seeing the result.

Is correlation useful for forex pairs?

Yes, but shared currency legs create exposure that must be mapped explicitly. The GBP/USD and AUD/USD correlation guide shows why two USD-quoted pairs can have changing correlation and unequal two-leg risk.

Practical Next Step

Pick one relationship that already matters to your trading. Freeze the data rules, calculate aligned returns, record Pearson correlation for more than one predefined window, inspect the scatterplot, and write down what would make the relationship fail.

Do not change a position simply because a coefficient crossed an arbitrary threshold. First identify the decision the statistic is supposed to improve.

Sources and Verification Notes

NIST, Pearson Product-Moment Correlation Coefficient — definition and statistical interpretation of Pearson correlation: https://itl.nist.gov/div898/software/dataplot/refman2/auxillar/pear_dis.htm
CFA Institute, Portfolio Mathematics, 2026 Curriculum — covariance and correlation as portfolio return/risk measures: https://www.cfainstitute.org/insights/professional-learning/refresher-readings/2026/portfolio-mathematics
CME Group, Equity Intermarket Spreads — relative-value trading between related futures contracts and the risks of spread trading: https://www.cmegroup.com/education/courses/introduction-to-equity-index-products/understanding-equity-intermarket-spreads.hideSubnav.educationIframe.html
FINRA, Asset Allocation and Diversification — diversification, uncorrelated exposure, and concentration-risk context: https://www.finra.org/investors/investing/investing-basics/asset-allocation-diversification
Investor.gov, Asset Allocation and Diversification — diversification as a risk-management concept rather than a guarantee against loss: https://www.investor.gov/introduction-investing/getting-started/asset-allocation

Correlation Trading Guide: How to Measure Market Relationships Without Fooling Yourself