Combined exposure and losing streaks

Three small positions can create one large shared risk. Your per-idea budget is only the starting point: add the risks of ideas open together, pending orders that could activate together and costs that may accumulate across the account. Think in scenarios. Which event could hurt several positions at once, and what would happen if their intended exits filled worse than planned?

This lesson connects currency exposure, correlation, daily and weekly rules, losing streaks and the possibility of becoming unable to continue. It finishes with a usable demo policy. Its numbers are teaching assumptions, not suggested live limits. A policy is useful even when its conclusion is to pause learning trades or never use real money.

A conventional long EUR/USD position has exposure to euro strength against the dollar: describe it as long EUR and short USD. A short EUR/USD reverses those directions. Use this directional inventory before relying on a correlation table. It helps reveal overlap without pretending the two legs are separate cash accounts or that every derivative involves a physical exchange.

A long EUR/USD and a short USD/JPY both have short-USD directional exposure, even though “long” and “short” appear on different tickets. A long EUR/USD and a short EUR/USD may offset some price exposure under particular sizes and account rules, but spread, financing, provider margin rules and closing order remain. A netting account may combine them instead of retaining two positions.

Put risks into one account currency before summing. If two ideas each have US$10 planned stop losses, their combined planned loss is US$20 if both fail under those assumptions. The sum is not a worst possible account loss. It omits any charges you have not included, changes in conversion and exits that miss the intended price. Even apparently unrelated pairs can lose together.

Also identify concentration in a single dealer, platform, connection, holding window or news event. A correlation matrix covers price relationships in its sample; it does not measure every common failure. A total cap, a shared-currency cap and a rule for pending orders address different parts of the problem. Record gross and net exposure so offsetting tickets do not hide large commitments.

Pearson correlation summarises the direction and strength of a linear relationship between two data series over a stated sample. It lies from −1 to +1 when both series have nonzero variation. Positive means above-average observations in one tend to align with above-average observations in the other; negative means they tend to align with below-average observations. Perfect ±1 describes an exact straight-line relationship in the sample, not equal cash gains or a permanent law.

For market comparisons, use aligned returns or price changes with a clearly defined timeframe and window. Correlation between raw trending price levels can be misleading. Check timestamps, missing observations, holiday effects, data source and whether both series represent the same time intervals. A 20-observation daily calculation and a 200-observation hourly calculation answer different sample questions.

A matrix shows many pairwise coefficients at once. Read the row for one pair and the column for another. The diagonal is +1 because each varying series is compared with itself; the table is symmetric. These are mathematical properties, not evidence of a profitable rule. The native table below uses seven synthetic log-change series created from eight invented observations only to practise reading; it contains no live or historical market measurements.

For reproducibility, these are the input currency-versus-USD log changes in arbitrary scaled teaching units, not current percentages or market records. USD against itself is zero. Derive USD/CAD as minus CAD/USD, USD/JPY as minus JPY/USD, EUR/GBP as EUR/USD minus GBP/USD, and EUR/JPY as EUR/USD minus JPY/USD. Then use the coefficient formula above; round only the displayed result.

For this practice matrix, a number closer to +1 indicates a stronger positive linear relationship in the invented sample, and a number closer to −1 indicates a stronger negative one. There is no universal coefficient cutoff that guarantees useful diversification. A number near zero means little measured linear relationship in that sample; it does not prove independence, no relationship of another kind or safety during a common shock.

Position direction changes the interpretation. A pair-return coefficient describes the two quoted return series, not the cash P&L of any possible combination of long/short positions. If one return series is multiplied by −1 to model opposite direction under a simple constant-exposure assumption, the sign of its correlation changes. Different sizes, account conversions, nonlinear products and costs need further analysis.

A correlation estimate belongs to its sample window. Central-bank news, changing interest-rate expectations, risk sentiment and currency-specific events can alter relationships. Moving the window changes which observations are included. Stress periods may cause losses to cluster even when an earlier average relationship looked weak. Do not copy a coefficient once and keep it as a permanent fact.

A small sample can create a striking coefficient by chance. Looking through many windows until one supports the desired position introduces selection bias. Record the window before testing the claim, compare another time period and state how uncertain the estimate is. Correlation does not prove which series causes the other, does not create an entry signal and does not establish the chance of both stops being hit.

A loss-limit policy controls your own actions. It can require no new orders after a defined loss amount, an agreed number of errors, a weekly threshold or a process failure. It cannot instruct the market to stop losing at that amount. Define the day’s timezone, when the week starts, whether the measure is gross realised losses or net P&L, how open exposure is counted and what happens to positions and pending orders at the trigger.

Define whether profitable trades restore room. A policy can choose not to let profits increase the day’s allowed risk. Do not redefine the rule during an emotional session or move the boundary to a new timezone to make a loss disappear. Resetting a personal counter does not repair account drawdown.

A fixed-cash method uses the same planned cash loss each time; as equity falls, that cash amount becomes a larger percentage of what remains. A fixed-fraction method recalculates a chosen fraction of current equity for each idea. Under ideal exact fills, its cash size shrinks after losses. Minimum volume, steps, charges and conversion can prevent the neat percentage from being achievable.

The table holds the loss sequence fixed and varies the sizing rule. It does not estimate how often six losses occur or prove that the smallest fraction is suitable. Fixed-fraction ideal mathematics may never hit zero in a finite number of losses, yet a real account can become unable to place its minimum order or meet its margin requirement much earlier.

Several losses can occur even under an assumed model with a favourable average outcome. The next trade does not owe repayment. Believing a win becomes inevitable because losses already happened is the gambler’s fallacy. In an independent model with constant probabilities, the chance of the next outcome does not change because of the previous sequence. Real trades may also be dependent, so the simple model needs evidence rather than faith.

Journal the sequence, costs, rule adherence and market conditions. A streak may reflect ordinary variation, a changed environment, a flawed rule or implementation errors; the sequence alone cannot choose among them. A pause lets you investigate. It is more useful than taking an extra position just to force the day’s outcome back to zero.

A doubling plan, often called a martingale-style plan, increases the next cash risk after a loss to try to recover earlier losses with one later win. That argument assumes unlimited funds, allowed size, sufficient margin, predictable payouts and no damaging costs or gaps. Real accounts have limits. The next required stake can become unaffordable long before a hoped-for win.

Adding to a losing position, moving its stop or opening another highly overlapping ticket can create a similar escalation even without exact doubling. Recalculate the whole idea and all open exposure. A risk policy should specify what changes are permitted and what requires stopping; an emotional recovery aim is not a replacement for that rule.

Ruin means reaching a condition where the activity cannot continue under the stated constraints. It might mean zero or negative equity, inability to meet minimum size or margin, a personal hard-stop drawdown or losing money that should never have been committed. State the threshold. Different definitions can produce different probabilities, so a percentage labelled “risk of ruin” is incomplete without its model.

A meaningful model needs initial equity; size rules; a ruin threshold; outcome probabilities and sizes; costs; a time horizon; dependencies between outcomes; and treatment of gaps, conversion and funding. A win rate alone cannot provide all that. The same 45% win rate can describe tiny wins and large losses or large wins and small losses, with very different implications.

This foundations lesson deliberately does not present an invented ruin-probability table from a win rate alone. The consecutive-loss table above is transparent arithmetic with stated assumptions. Lower size can reduce the damage of a specified loss sequence, but it does not guarantee survival, remove a negative average outcome or protect against every gap. A useful beginner decision is to remain in demo whenever the model, product or personal budget is unclear.

Turn the ideas into a short document you can read before practice. Protect essential money; name the account currency and reference; list allowed demo instruments and contract assumptions; define per-idea risk, aggregate risk, shared-currency gates, daily/weekly rules, event/overnight handling and a process for stops and pending orders. Describe what you will record when the planned and actual results differ.

Use a change log. A policy revision should follow review of the records, not occur while trying to keep a losing position open. A missed stop, misunderstood unit, unexplained margin figure, unreliable connection or urge to recover urgently can be a process trigger to pause. Account for open exposure and cancel unwanted pending orders using your predefined procedure; simply closing the notebook does not close positions.

For review, distinguish decisions you can control—whether to enter, intended size, rules and record keeping—from outcomes you cannot guarantee—price movement, fill quality or future relationships. Level 6 will build on this foundation by describing price-action rules; those rules still need the risk boundaries established here.