Relationships that can change

Two pair names can look different while hiding the same currency direction. A long EUR/USD position buys euro exposure against dollars. A long GBP/USD position buys pound exposure against dollars. Both are short the dollar side, although the euro and pound can move differently. Start by writing those directions before asking whether a chart statistic makes the positions diversified.

A relationship is an observed connection under stated conditions. It is not an instruction to buy two pairs, a permanent economic law or proof that one movement causes another. This lesson connects the pair mechanics from earlier levels with sample measurements, cross-asset context and changing conditions. You can do every activity with paper, invented prices or historical observations; no live order is needed.

Keep three questions separate: what currency directions do I hold, what did these series do in a particular sample, and what joint adverse cash scenario could affect my account? Exposure answers the first, correlation helps describe the second, and a consistent scenario worksheet addresses the third. None alone settles all three.

A correlation calculation needs paired observations taken on a consistent basis. Choose the instrument, provider, price side, time zone, return interval and start/end dates. Pair the return for one interval with the other series over that same interval. A London close compared with a different session boundary may include different news. Mixing a daily movement with a monthly movement asks an unclear question.

Comparing raw price levels can be misleading when both series drift over time. Returns are usually a clearer starting point for studying co-movement, but their choice still matters. Daily, weekly and monthly returns can produce different coefficients. Overlapping multi-day returns reuse observations and introduce dependence; do not call each overlapping row an independent new experiment.

Do not fill missing days with a made-up zero return just to obtain a number. State a consistent matching policy, retain exclusions and explain how they change the sample. If the observations cannot be aligned reliably, an honest “not measured” is more useful than a confident-looking coefficient.

The method has roots in nineteenth-century statistical work rather than retail trading. Karl Pearson’s 1896 Royal Society paper discussed regression and correlation in a biological research setting. Today the coefficient is used across many fields. Its long history does not turn a sample relationship into a market law; the definition and data still need checking.

Pearson correlation measures linear association between two varying series in a chosen sample. Its coefficient runs from −1 to +1 when defined. Positive values mean above-average movements in one tend to accompany above-average movements in the other; negative values describe the opposite tendency. A value near zero means little measured linear association under that setup, not that the two series have no possible relationship.

The calculation centres each series by subtracting its own sample mean. It multiplies the paired deviations and adds them, then divides by the square root of the product of the two sums of squared deviations. This normalisation gives a dimensionless coefficient. If either series has no variation in the sample, the denominator is zero and this correlation is undefined; it is not automatically zero.

“Perfect” here concerns the recorded sample, not the next price change. An unusual observation can affect the coefficient substantially, especially in a small sample. Inspect the actual paired rows or a scatter plot as well as the summary number. There is no universal coefficient threshold that turns two leveraged positions into a safe portfolio.

In window one, B is exactly twice A, so the correlation is +1 for these four nonconstant paired rows. In window two, B is exactly minus twice A, so it is −1. These are deliberately simple teaching sequences, not reported EUR/USD or GBP/USD returns. They show how a relationship can depend on the chosen window without pretending four rows provide reliable forecasting evidence.

For window one, A has mean 0.5 and B has mean 1. Their centred cross-products total 10; their squared-deviation totals are 5 and 20. Thus r = 10/√(5 × 20) = 1. Reversing B changes the numerator sign and gives −1. You do not need to calculate many coefficients by hand, but you should be able to explain why the paired rows, means and denominator matter.

When comparing actual nonoverlapping windows, keep the frequency, data source and matching rules the same. Record both results, not only the stronger one. Changes may reflect different economic conditions, a few large observations, sampling variation or data problems. The number alone cannot tell you which explanation is correct.

This is the course’s deliberately invented teaching matrix, not current market data and not calculated from the four-row exercise above. Each diagonal is +1 because a varying series matches itself. The matrix is symmetric: EUR/USD versus GBP/USD has the same coefficient as GBP/USD versus EUR/USD. It uses raw quoted-pair monthly returns under one hypothetical common sample.

Do not multiply one trade’s planned loss by 0.65 to estimate the other’s loss, or call +0.65 a 65% chance of simultaneous gains. Cash results depend on direction, size, entry/exit, account conversion and costs. A matrix assembled from mismatched date windows may also be internally inconsistent; a valid collection of pairwise-looking numbers is not automatically one valid joint portfolio model.

Read negative raw-pair coefficients alongside the actual position directions. If one position is short rather than long, its directional exposure changes. A negative coefficient between EUR/USD and USD/JPY does not mean every combination of those trades offsets risk. Long EUR/USD and short USD/JPY both sell USD exposure.

EUR/USD describes dollars per euro; its inverse describes euros per dollar. A rise in one corresponds to a fall in the other. But the exact arithmetic depends on the return definition. For a simple return r expressed as a decimal, the inverse-rate simple return is −r/(1+r), not exactly −r. At small changes the difference may be modest; at larger changes it matters.

Logarithmic returns, defined as the logarithm of ending price divided by starting price, change sign exactly when the rate is inverted. Therefore consistently inverting one log-return series flips its correlation sign, when the coefficient is defined. With simple returns, inversion is nonlinear, so do not promise an exact sign-reversed coefficient for every finite-change sample.

Separately, taking a short exposure rather than a long changes the sign of a fixed-size directional price result under consistent conversion assumptions. It does not reverse the quote itself. Financing, differing execution prices and changing conversion rates may keep the actual net account-result series from being a simple negative copy. Label what is being compared before interpreting a coefficient.

Cross-asset analysis compares currencies with such things as bond yields, commodity prices or equity prices. Possible links include expected interest-rate paths, import/export income, financing conditions and broad changes in risk-taking. A plausible explanation is a question to examine, not proof that an asset always controls a currency.

A bond yield is not the same variable as a bond price, and a yield change in percentage points is not the same unit as a currency percentage return. Define the variables and timing clearly. Comparing a US yield with USD/JPY without considering Japanese yields, expectations and other events gives only part of the context.

Do not turn a familiar label such as “risk-on currency” or “safe haven” into a rule that must work during every event. Such descriptions depend on period and context. Record when the idea failed as well as when it appeared to fit; a post-event story can sound convincing without being a rule available beforehand.

A yield or interest-rate difference can influence the cost or income associated with holding currency exposure. It does not settle the total home-currency result. The exchange-rate movement, funding expense, conversion, product terms and other charges all matter. A central-bank policy rate is also not automatically the interest rate a customer receives or the financing rate a retail leveraged product applies.

If the position is financed with borrowing, the amount still owed and its interest must be considered separately. A falling foreign asset can compound the currency loss. If a hedge exists, its size, term, execution and cost change the calculation; merely saying “hedged” does not establish a risk-free result.

Retail overnight credits or debits depend on the exact contract, provider schedule, direction and dates. They may reflect markups and settlement conventions, and can change. Do not enter two central-bank rates into a worksheet and call their difference the guaranteed daily income on a leveraged account. The next lesson examines carry trades more directly.

A market regime is a description of conditions over a period, such as relatively quiet movement, a sustained directional trend, a range or unusually large price changes. Different characteristics can overlap: a trend can also be volatile. Labels are useful for organising evidence only when the measurement, observation window and classification rule are clear.

For an observation exercise, you might define a completed-bar range measure over an explicit lookback, compare it with a previously available reference and state what value counts as elevated under that project. Choose the threshold before examining evaluation outcomes. This is a research definition, not a recommended universal setting. ATR, reviewed in Level 4, measures movement magnitude under specified inputs; it does not predict direction or guarantee a fill.

A label based on the next month’s prices was not available at today’s decision. Distinguish a retrospective description from a rule that could classify conditions in real time. Conditions near a boundary may be uncertain, and a delayed measure can change only after the market already moved.

A rule that appeared successful in one setting can fail in another. That observation may justify a new hypothesis, but adding regime filters after seeing the losses is development work. Preserve the old result and test the new version on genuinely later or unviewed observations as Level 9 requires.

A shared shock can affect several currencies, asset classes and funding markets at once. Participants reducing exposure may sell positions that previously seemed unrelated. A historical coefficient from calm conditions may then give a poor picture of joint risk. Conversely, local events can break a previously strong connection. Neither outcome is inevitable.

Changing volatility, spread and execution conditions matter alongside changing co-movement. A plan that assumes independent stop fills at exact prices may underestimate losses when several positions face gaps or poor execution together. A scenario is a stated “what if,” not a predicted probability or guaranteed worst case.

Do not use a chosen stress label as a reason to select only impressive history. Keep all predeclared windows, including weak or contradictory evidence. A sample can suggest a concentration worth investigating without providing enough evidence to estimate its future frequency.

List each proposed or open position, direction, size, account currency and exit/fill assumptions. Calculate each cash result under the same scenario, then add amounts expressed in the same account currency. Separate fees and financing consistently. This is more informative for a beginner than treating a correlation coefficient as a fixed loss multiplier.

Opposite-looking positions also need arithmetic. Long EUR/USD and short GBP/USD may partly offset a broad dollar movement at particular sizes, but they retain euro-versus-pound exposure and product/cost risk. Calling them a hedge does not prove exact offset under every exchange movement. A historical coefficient cannot replace the contract and scenario details.

If the combined scenario conflicts with your written practice policy, revise the hypothetical plan or record no action. Do not invent a lower correlation to make the worksheet pass. Keep essential household money outside trading experiments; a well-labelled worksheet still cannot cap a live leveraged loss.