Correlation in Trading: Positive, Negative and Changing Correlations

What is correlation in trading, and how do positive, negative and changing correlations differ?

Short answer: Correlation in trading measures the strength and direction of the relationship between two instruments’ returns over a defined period. Positive correlation means returns tend to vary together; negative correlation means they tend to vary oppositely. Correlations can change as market conditions change, so historical relationships should not be treated as fixed trading signals.

What does correlation measure in trading?

Correlation describes how two variables vary together. In trading, a common measure is Pearson’s correlation coefficient, usually written as r, calculated from two sets of returns over matching intervals.

A return is a price change expressed relative to the starting price. If an instrument rises from 100 to 102, its simple return is 2%. Using returns helps compare instruments with different price levels.

For this purpose, correlating raw price levels can be misleading: two unrelated series may appear closely connected because both trend over time. Correlation of returns is generally a more useful starting point for examining whether their movements are related.

Pearson correlation measures a linear relationship. It does not establish that one market causes another to move, identify which market moves first, or specify how many points one instrument should move when the other changes.

What do positive, negative and zero correlation mean?

Positive correlation means returns tend to vary together across observations. Negative correlation means they tend to vary oppositely. More precisely, the calculation compares each return with its own series’ average; it does not simply count sessions when both prices rise or both fall.

Values closer to either −1 or +1 indicate a stronger linear relationship. Values nearer zero indicate a weaker one. Labels such as “strong” or “moderate” depend on context, sample size and intended use; there is no universal trading threshold.

Changing correlation describes how the estimate evolves over time. It can strengthen, weaken or change sign.

Why do correlations change?

Correlations change when the forces influencing markets change. Shared exposure to a currency, interest rates or investor risk appetite may dominate one period. A company event, supply shock or different central-bank policy may dominate another.

Inflation can also change asset relationships. A December 2023 Bank for International Settlements study documented US equity–government bond return correlation turning positive from mid-2021. It explained how inflation and expected monetary tightening can put pressure on both asset classes. This is a historical example, not a current correlation reading.

Gold and the US dollar do not have a permanently negative relationship either. The World Gold Council documented a period of positive gold–dollar correlation in its May 2022 analysis. Different drivers can affect both markets at once.

An estimate can also change without a lasting economic shift. A short sample or a few unusually large moves may alter the result. A changing coefficient is therefore a reason to examine the data and market context, not proof that a new regime has begun.

How can traders measure and monitor correlation?

A useful correlation reading needs four details: the instruments, return frequency, lookback length and measurement date. “These markets are correlated” is incomplete without them.

1. Choose matching observations. Compare returns over the same intervals, using consistent timestamps. Different market closing times, holidays or missing observations can distort the comparison.
2. Calculate returns consistently. A simple percentage return is (current price ÷ previous price − 1) × 100. Sixty daily returns require 61 consecutive price observations for each instrument.
3. Calculate the coefficient. With 60 aligned returns in each spreadsheet column, =CORREL(B2:B61,C2:C61) calculates Pearson correlation. The input series must vary; a series with zero standard deviation has undefined correlation.
4. Compare and update windows. Recalculate as new observations arrive, and compare a shorter window with a longer one.

Sampling frequency and lookback length are different choices. Hourly versus daily returns changes the movement being measured. Twenty versus 60 daily observations changes the history included.

Rolling correlation uses a moving window: the newest observation enters and the oldest leaves. A short window responds quickly but can fluctuate substantially. A longer window smooths some fluctuations but can conceal a recent change.

For example, a hypothetical pair of instruments might show +0.82 over 60 daily returns and +0.24 over the latest 20. The recent sample shows a weaker linear relationship. Those illustrative numbers do not establish why it weakened or which reading will describe the future.

How do correlated instruments affect open trades?

The correlation between instruments is only part of the exposure calculation. Long and short positions respond differently to the same underlying price movement.

This table assumes straightforward directional positions and a relationship that persists. It does not account for different sizes, volatility, contract specifications or costs. The same correlation can produce very different monetary outcomes in different accounts.

For example, a long EUR/USD position benefits from euro strength against the dollar. A short USD/CHF position benefits from dollar weakness against the franc. Both positions therefore include exposure to a weaker dollar. If the two pairs have negative return correlation over the measured period, taking opposite trade directions can reinforce exposure rather than offset it.

For clients of a multi-asset broker such as NordFX, the practical question is whether several open positions depend on the same market outcome. Counting instruments alone does not show how diversified the account is.

What does correlation risk look like in practice?

Consider a hypothetical account with $10,000 equity. A trader opens long EUR/USD and long GBP/USD positions. Each position is sized so that a fill at its planned stop price would lose $50, before trading costs.

If the dollar strengthens against both currencies, both pairs fall and both positions may lose. The symbols differ, but the trades share dollar exposure.

The $100 total comes from adding the two planned losses. Correlation does not reduce that sum, and a coefficient does not give the probability that both stops will be reached. Actual losses may exceed the planned amounts because of gaps, slippage or costs.

The decision is therefore about combined exposure: whether the position sizes and shared driver fit the account’s risk limits. This example is arithmetic for illustration, not a recommended allocation or a forecast.

What mistakes should traders avoid?

1. Treating correlation as a success rate. A coefficient is not a win probability or the percentage of matching price moves.
2. Confusing correlation with movement size. Highly correlated instruments can have different volatility. Equal lot sizes need not create balanced monetary exposure.
3. Using mismatched or outdated data. Different timestamps, frequencies or stale readings can answer a different question from the one relevant to the trade.
4. Assuming low correlation eliminates risk. A near-zero historical coefficient does not rule out shared losses, nonlinear dependence or a future increase in correlation.
5. Expecting every divergence to close. Historical co-movement alone does not prove a stable price spread or guarantee that two markets will return to a previous relationship.

Frequently asked questions

What does a correlation of 0.7 mean?

It indicates a positive linear relationship in the measured data, closer to +1 than to zero. It does not mean that the instruments move in the same direction 70% of the time. Interpretation also depends on the sample and return frequency.

Can positive correlation become negative?

Yes. Changes in economic conditions or market-specific drivers can reverse a relationship. Sample selection can also change the estimate, so compare periods and examine the underlying data before drawing a conclusion.

Which currency pairs are the most correlated?

There is no permanent ranking. Pairs sharing a currency or economic drivers may be closely related, but the ranking depends on the measurement date, frequency and lookback period. Any “most correlated” list should state those details.

Are gold and the US dollar always negatively correlated?

No. Their relationship varies, and both can rise or fall during the same period. The return series and measurement window matter more than a fixed rule about their relationship.

Does negative correlation guarantee a hedge?

No. The effect depends on trade direction, size, volatility and whether the relationship persists. Buying one negatively correlated instrument and selling the other can reinforce exposure.

What is the best time frame for correlation analysis?

There is no universally best choice. Use a return frequency relevant to the holding period and compare more than one lookback length. A daily correlation estimate may not describe an intraday relationship.

Can correlation predict the next price move?

Not by itself. A correlation coefficient describes a measured relationship; it does not establish causation, a lead–lag relationship or a reliable entry signal. Any predictive use needs separate testing.

What should traders take from correlation analysis?

Correlation is useful when its context is explicit: what was measured, over which intervals and period, and how the result relates to actual positions. Reviewing those details alongside size, volatility and shared market drivers can reveal concentrated exposure. The coefficient remains a historical estimate, and its relevance must be reassessed as conditions change.