Home/Risk management

Risk and capital management – eleven articles on how much you are allowed to lose and how to calculate it

After a 10% drawdown, returning to the previous peak requires an 11.11% return; after 50%, it requires 100%. The calculation's base changes, and so does the scale of recovery. Risk management extends further: choosing the loss amount, accounting for execution costs and minimum volume, and checking the shared exposure of several positions. Eleven lessons take you from these calculations to a plan with your own limits, halt conditions and rules for resuming trading. You will also distinguish arithmetic identities from results that depend on the chosen model and data quality. The examples show the consequences of explicit assumptions; they establish neither a strategy edge nor a guarantee of capital protection.

Hypothetical exit-scenario loss calculation, rather than portfolio standard deviation. All amounts refer to the same moment, without counting loss already included in equity again: 250−180=PLN 70. Meeting this limit does not replace checks of other constraints or guarantee an exit price.
What will you take away from this section?
  • A number instead of a rule copied off the internet. A risk percentage derived from the drawdown you are willing to accept and from the length of the losing streak the account is meant to survive, plus a volume computed all the way to the end – with the currency conversion and with rounding down to the step available on your account. Every number states its base: current equity or settled balance.
  • Several portfolio measures, each answering its own question. The sum of full exit losses describes an execution scenario, σ describes the dispersion of returns, and notional currency legs show shared exposure. The number of positions alone cannot replace any of these measures.
  • Decisions written down before you need them. Daily, weekly and monthly limits calculated from the same type of base, but each measured at a different time—with explicit precedence and the conflicting amounts that follow from that difference; a trading halt threshold with a condition for resuming; and a plan in which every field identifies the article that supplies its number. A written rule does not enforce execution—it removes the need to decide these matters at the moment of a loss.

1. Why the section starts with arithmetic and not with an entry rule

The return needed to regain the peak is DD/(1−DD). DD is the loss relative to that peak; the return uses the smaller base after the loss, in the same units of capital and without cash flows. For DD of 10%, 20%, 33%, 50%, 75% and 90%, the required returns are 11.1%, 25%, 49.3%, 100%, 300% and 900%, respectively. If DD is calculated net of costs, the required return is also net; the formula itself is neither a “lower estimate” nor a promise of recovery time. The derivation and full table are in 9.2 [2].

The second number that shapes this section concerns streaks. At a hit rate of 55%, assuming a constant hit rate and independence of successive outcomes – and that is a simplification of the model, not a property of the market – the probability that three trades named in advance all end in a loss is (1 − 0.55)³ = 0.45³ = 0.091125, that is 9.11%. That is not the answer to the question of whether such a streak will happen to you: over several hundred trades, blocks of three outcomes overlap, and the probability of meeting such a streak anywhere in that sequence approaches certainty. Confusing those two questions is the most common error in reasoning about losing streaks. The figures for both readings – for a block named in advance and for a streak met anywhere in the sequence – are given in Article 9.1; the length of the streak the account is meant to survive then enters as an assumption in the derivation of the risk percentage in Article 9.3.

On 27 March 2018, ESMA reported a historical range of 74–89% of retail CFD accounts losing money in analyses by national supervisors [5]. The studies covered different periods and jurisdictions. This is not a single current Forex sample, an individual's probability of loss, or evidence of the effect of a chosen risk percentage. These data support treating the product cautiously; they do not determine a budget.

This section covers calculations and procedures. It does not teach you how to make money or supply ready-made limits. It calculates the consequences of the parameters you provide—capital and its base, account currency and conversion rate, protective exit distance, and the win rate and reward-to-risk ratio measured on your closed trades—and shows where each parameter ceases to be reliable.

Loss-recovery table: a 10% decline requires 11.1%, a 50% decline requires 100%, and a 70% decline requires 233.3%.
Return to the same peak, without deposits or withdrawals. Table values are rounded to one decimal place. Enter the decline as a fraction in the formula, for example 0.50 for 50%; the required return does not determine recovery time.
15 January 2015: the boundary of planned execution. The SNB discontinued the minimum exchange rate of CHF 1.20 per euro [8]. This illustrates why an order level and an execution price need to be distinguished. Direct sources for market conditions and prices, and the loss calculation including costs, are in 9.10 [9]. The central bank's announcement alone does not document the day's low or closing price.

First establish which number you need

Four questions, four kinds of result
QuestionResultLesson
How much will I lose in the assumed exit scenario?Full loss in PLN; no price guarantee9.9
How does the portfolio return vary?σ in a specified currency and horizon9.9
How much can I allocate to a new order?Remaining shared budget after reservations9.3
Does the instrument's minimum fit the limit?Cost-adjusted volume rounded to its step, then checked backwards9.5

Numbers can look similar while answering different questions. Without definitions, it is easy to compare a loss scenario with a standard deviation and obtain a result with no practical meaning. Before calculating, specify the unit, base, horizon and execution model.

2. What a risk rule changes in an account, and what it does not

A risk rule defines the budget and the action to take when it is breached. It does not guarantee execution prices or turn a negative mean per unit into a positive one in a fixed model of net outcomes with proportional costs. Changing position size can, however, change costs in R units if the service charges a fixed or minimum fee. Therefore, check the budget alongside execution assumptions rather than assuming every cost scales with one percentage.

The difference shows on a single account. The data are illustrative and deliberately dimensionless – capital normalised to one, ten losses in a row, risk computed from current equity updated before every trade; the base matters here, because with a fixed risk amount the result would be different. After n losses (1 − r)n of the capital remains, the shortfall is DD = 1 − (1 − r)n, and the return needed to get back to the peak is Z = DD / (1 − DD).

  • r = 1%: 0.9910 = 0.904382 → DD = 9.5618%, that is 9.56%; Z = 0.095618 / 0.904382 = 0.105727 → 10.57%.
  • r = 5%: 0.9510 = 0.598737 → DD = 40.1263%, that is 40.13%; Z = 0.401263 / 0.598737 = 0.670183 → 67.02%.

Five times as much risk per trade deepens the loss by 40.1263 / 9.5618 = 4.20 times, but increases the return needed to recover it by 67.0183 / 10.5727 = 6.34 times. These two ratios differ because Z = DD / (1 − DD) grows faster than drawdown itself—the derivation and table are in Article 9.2, which does not calculate this particular ratio. Here, r is the full net loss in the scenario of the next close, and converting DD into the required return is a capital identity. If you start with a notional loss from the price move, account for costs first; if r already includes them, do not add them again. Neither this scenario nor a historical observation is a lower bound on future drawdown. Deposits, withdrawals or a change in the percentage base during the streak invalidate the calculation.

Between the number in the spreadsheet and the amount that actually leaves the account stand the spread on both sides, the commission, swap points, slippage and the price gap. Planned risk and realised risk are two different quantities, and the difference between them is the substance of the calculation, not a caveat attached to it – it is computed by Article 9.3, and the case in which price jumps over the exit level by Article 9.10.

With several positions, the sum of planned full losses is a useful scenario, but not a measure of portfolio volatility. Long EUR/USD and GBP/USD positions share a short USD leg; reversing direction changes its sign. Exposure size requires notionals and currency conversion, while dispersion requires the dependence between outcomes. 9.9 separates these calculations so that a sum of percentages does not stand in for correlation or σ.

Chart of ten losses: at a 1% fraction, 90.44% of starting capital remains; at 5%, 59.87% remains.
A specified sequence of ten full net losses from current capital. At a 1% fraction, 90.44% of starting capital remains; at 5%, 59.87%. Shared linear scale; an arithmetic model rather than market quotes or a strategy result.
Position size does not remove the source of a loss. With linearly scaled outcomes and costs, changing a positive position size does not change the sign of the mean per trade. A fixed fee, minimum commission or different execution can break that assumption, so recalculate the full net mean for the chosen V. In a model of independent outcomes with a fixed distribution, a positive capital multiplier and a finite negative mean log return, capital decreases asymptotically; any positive threshold below initial capital is reached with probability 1 over an unlimited horizon. This is a result of that model, not a guarantee about a real account's path or a measure of the time needed. Nor must a smaller r always extend every defined hitting time. The sequence therefore runs in one direction: first check whether your closed trades have positive expected value after costs—the calculation is in Article 9.6—and only then choose your position size. Reversing the order gives you a precisely calculated rate of loss.
A written rule is not an executed rule. This section addresses the first: how to derive a threshold, which base it uses, what happens after a breach and under what condition trading resumes. It does not assume that writing a rule enforces it or explain why rules are broken—that belongs to the psychology section. The practical consequence for rule design is nevertheless computational: a threshold requiring interpretation when triggered is not unambiguous, and an ambiguous rule can neither be executed consistently nor checked later against the records.

Two definitions of “1%”, two different orders

For capital of PLN 10,000, 1% is PLN 100. If it means the notional price move R₀, with an assumed cost of 0.13R₀, the full scenario loss is PLN 113. If PLN 100 is the total budget, the exact upper limit of the price-move unit in this model is 100/1.13 ≈ PLN 88.495575. Rounded down to the nearest grosz, it would be PLN 88.49; 88.50 × 1.13 = PLN 100.005, so that amount already exceeds the PLN 100 limit. Round volume down to the instrument's step and check the full scenario loss again. One definition includes the cost; the other adds it afterwards.

These are not two adjustments to the same order: the result depends on the percentage's definition. Calculating the full volume requires the actual cost components from 9.5. The first step is therefore to define R before comparing any rates.

3. Retail account parameters in Poland

The following parameters apply to retail CFDs covered by Polish KNF Decision DAS.456.2.2019, effective from 2 August 2019. Its operative provisions and KNF information were checked on 5 October 2026 [6][7]. The intermediary's home country alone does not determine the scope: the intervention also covers activities carried out in Poland from another country. Client status, instrument and the terms of the specific service require separate checks. This section uses these values in risk calculations; the full regulatory explanation belongs to the relevant section.

The table below shows the quantity most often confused with risk. Margin follows from the notional value of the position and the leverage limit; the risk amount follows from the distance of the protective exit and the volume. The exit distance does not enter the margin calculation at all, and margin does not enter the risk calculation – they are answers to two different questions, and that is why the last column is identical in both rows.

Required margin for 1 lot of EUR/USD (EUR 100,000 notional), compared with the change in position value over a 50-pip move. The first row assumes an illustrative specification of exactly 30:1, distinguished below from the legal minimum margin of 3.33%. The second assumes 100:1 where the instrument conditions are met and the broker grants experienced retail client status under Poland's KNF decision. That status is restricted to retail clients whose registered office or habitual residence is in Poland when the contract is concluded. The conversion rates EUR/PLN 4.28 and USD/PLN 3.95 are assumed scenario constants, not market quotes—they have no date, price side or provider and serve only to show the order of magnitude. Regulatory parameters: KNF Decision DAS.456.2.2019 of 1 August 2019, effective from 2 August 2019; operative provisions and KNF information checked on 5 October 2026. The intermediary calculates margin using its own conversion rate and account policy.
Regulatory parameterMargin in EURMargin in PLNChange in gross value on a 50-pip move
Illustrative specification of exactly 30:1; legal minimum for the specified pairs: 3.33%100 000 / 30 = 3 333.33 EUR3,333.33×4.28 = PLN 14,266.6750 pips × USD 10 = USD 500; ×3.95 = PLN 1,975
Illustrative specification of 100:1; minimum 1% once status and instrument conditions are met100 000 / 100 = 1 000 EUR1 000 × 4.28 = 4 280 PLNSame notional and move: PLN 1,975 gross

The table assumes exactly 30:1: 100,000/30 = EUR 3,333.33. The literal legal minimum of 3.33% gives EUR 3,330 [6]; these parameters must not be treated as identical. The last column gives the size of the price change before costs, rather than the settlement of a particular position direction. The full loss with the account conversion rate and costs is calculated in 9.5 [3], and the permitted budget and its remaining amount in 9.3 [1].

The KNF decision's mandatory close-out condition applies when account equity falls below half the total required initial margin for open CFDs [6]. It does not mean a 50% loss from the peak or half the margin for one position. Nor does it specify a “weakest first” sequence. A particular service may impose additional restrictions, so check your own threshold and the margin condition separately.

Negative balance protection limits liability for all CFDs in the covered account to the funds in that account [6]. It does not guarantee a stop-loss price, prevent a temporarily negative reading or preserve deposited capital. It does not cover every instrument or client status. A loss budget remains necessary regardless of this rule.

Account currency can change the cost of a loss. An illustrative USD 500 loss gives PLN 1,975 at a conversion factor of 3.95 and PLN 2,075 at 4.15—a PLN 100 difference. These are two exchange-rate scenarios, not two current quotes or an automatic sum of differences over a year of trades. Your own conversion rate has a price side, observation time and application to the loss; its change is included in the calculation in 9.5.

Time of day is a separate matter. Execution conditions change throughout the day and vary between intermediaries. On less liquid pairs—including those involving the Polish zloty—windows with a thin order book occur at different times from those on major pairs. For risk calculations, this means only that a cost estimated for one window may not describe another; the direction and size of the difference require data or an explicit scenario. The sessions themselves are described in the trading hours and liquidity section.

A locked wallet represents margin; a long position with PLN 30,000 notional stands in the centre, and scissors marked R represent a separate exit rule. Leverage of 10:1 requires PLN 3,000 margin; 30:1 requires PLN 1,000. A −1% price move means −PLN 300 at the same notional.
Margin and loss budget answer different questions. In this illustrative model, the same long PLN 30,000 notional position requires PLN 3,000 margin at 10:1 leverage or PLN 1,000 at 30:1. A −1% price move gives −PLN 300 in either variant. The model omits costs and currency conversions; it describes neither regulations nor a particular provider’s specification. The margin threshold does not define planned R, and an ordinary stop guarantees no fill price.

The percentage of losing accounts: check the denominator

A provider's warning is not a forecast of your loss. The KNF decision specifies the calculation for retail CFD accounts: the result also includes open positions and costs; cash flows are not profit, and inactive accounts are excluded from the denominator [6]. One person may hold several accounts. Before comparing two percentages, check the period, product and counting method.

ESMA's historical range from 2018 covers a different population from a particular provider's current warning. Neither number says how changing your budget by half a percentage point will affect your results.

4. Section map – eleven articles, one order

The section starts with the meaning of risk and the arithmetic of loss (9.1–9.2). It then separates budget, invalidation condition and cost-adjusted volume (9.3–9.5). Next come net outcomes and changes to open positions (9.6–9.7). The last lessons cover drawdown, the portfolio, events and the plan (9.8–9.11). You can read in sequence or select an article by the quantity you need; the map below identifies which lesson owns each calculation.

Two dependencies are forced by the arithmetic, not by the order of exposition: the volume in 9.5 is a function of the distance of the protective exit (9.4) and of the base of the risk percentage (9.3), so both of those decisions must come earlier. Beyond that, every article can be read on its own, provided you take the input parameters from the article that settles them – which is why every number is given with its owner.

Map of eleven lessons in five groups: foundations and the mathematics of loss (9.1–9.2), parameters before an order (9.3–9.5), outcomes and open positions (9.6–9.7), the whole account (9.8–9.9), and events and the plan (9.10–9.11). Calculating volume in 9.5 requires the budget from 9.3 and the exit distance from 9.4; the grouping itself does not impose a single reading order.
Map of eleven lessonsFive topic groups. Only two arrows represent inputs: the budget from Lesson 9.3 and the exit distance from 9.4 feed into volume in 9.5. Choose the other groups according to your question. Where are the inputs for your decision? Foundations 9.1 Risk versus strategy 9.2 Mathematics of loss Before an order R d V 9.3 Budget 9.4 Exit 9.5 Volume Outcome and position 9.6 Expected value 9.7 Open position Whole account 9.8 Drawdown 9.9 Portfolio risk Events and plan 9.10 Events and VaR 9.11 Risk plan

9.1. Capital protection: risk versus strategy

9.1 separates trade selection from the budget that determines the scale of its effect on capital. It also shows why a favourable mean per unit is insufficient when the stake is excessive. These are different questions, even though some of the data used to answer them may overlap. The article does not evaluate signals; it uses an explicit model to examine the effects of different rates and the freshness of the information used to size a position.

You will find here the separation of two meanings of the word risk – variability of the result and permanent loss of capital – and of a third use which is neither: the risk amount R, that is, the number entered into the volume calculation. The article also indicates which of those meanings the section deals with. Then a hierarchy of decisions set in one direction – how much you may lose, where the invalidation point lies, what volume follows from that, and only at the end the entry signal – and a map of the whole section tying every decision to the article that settles it and to the parameter that comes out of it. The article also says outright that positive expected value does not rule out the loss of capital, but it does not carry out that calculation: recovery asymmetry and risk of ruin belong to 9.2.

9.2. The maths of losses: recovery asymmetry and risk of ruin

The computational core of the section and the only place in which the formula for the return needed to recover a drawdown is derived. The article begins with the derivation of Z = DD/(1 − DD) and a table from −5% to −90%, denser in the shallow range than in the deep one, and then separates what is an arithmetic identity from what is the output of a model – because the two groups of statements have a different status in this section and must not be mixed.

Risk of ruin is given with the variant of the model, the formula and the list of assumptions, and every cell of the table can be reproduced from that formula; separately stands the calculation for two levels of risk per trade with the remaining parameters unchanged, and a variant for a reward-to-risk ratio other than one to one.

There is also a settlement the section previously lacked: the same set of outcomes arranged in a different order gives the same end point, but a different maximum drawdown along the way and a different moment at which the limit is breached. The simulation of the distribution of outcomes is described from the procedural side – what is drawn, from which distribution, how many repetitions – together with a passage saying what its result does not cover. The article closes with a self-check task and its solution.

9.3. Risk per trade, aggregate risk and loss limits

The article answers the question that the phrase "I risk 1%" does not yet contain: one percent of what. The base of the calculation is settled once for the whole section – with a table showing what the same rule amounts to in money on the settled balance, on current equity and on the peak of the equity curve, and in which direction each of those bases shifts the position size. The column on volume gives the direction only – the conversion into lots is handed over to 9.5 and deliberately not repeated.

9.3 derives the percentage from an acceptable drawdown and a streak scenario, and specifies the bases for the limits. It separates the sum of planned losses from the mean outcome and its dispersion. Each measure can support a different decision; the sum is a scenario budget, not σ. Daily, weekly and monthly limits each use a base observed at their own time, while a drawdown threshold uses the peak base. Examples of conflicting limits show how to calculate the remaining shared budget before opening another position. The exercise requires a clearly specified action after a threshold is breached.

9.4. The protective exit and the invalidation point

The placement of the protective exit follows from the question "at what price does the reason for entering stop holding", not from the amount it is convenient to lose. The article sets side by side both orders of determination – from invalidation through distance to volume, and from amount through volume to distance – and shows the numerical result of each, because that is the only way to settle the dispute about order rather than merely declare it.

The variants of placement are described as a designer's choice, with the cost expressed in distance and in the volume they permit at constant risk – not as a ranking of methods. The exit distance is an input here for the volume calculation, which belongs to 9.5; the way extremes, zones and volatility measures are determined belongs to the section on technical analysis.

A time exit is a separate operational condition. By itself it does not give a known price loss in advance, so the plan must also specify the budget and execution scenario. 9.4 distinguishes the hypothesis condition, activation and execution price.

What a protective exit provides, and what it does not. It makes the closing decision before entry and defines the intended loss used to calculate volume. It does not ensure an execution price: after activation, the order is executed at the first available price, and if a gap separates the order level from the next price, the loss is calculated from the actual price. With constant pip value, the actual distance equals the planned distance plus the execution deviation; a nominal level guarantees neither that deviation nor a loss limit. The converse is also false: having no protective exit does not mean an unlimited loss, but the absence of the exit scenario defined by that stop. Possible loss depends on the instrument and service terms. Negative balance protection in a covered retail CFD account limits total liability to that account's funds; it neither protects the deposit nor guarantees the price. Mandatory close-out is a separate mechanism dependent on equity and margin. The loss calculation for a gap greater than the exit distance is in Article 9.10.
A schematic price path separates hypothesis invalidation, stop activation after a gap and actual execution.
Schematic long-position scenario. The hypothesis is invalidated at Bid 100.00. The assumed service activates the stop when Bid reaches 99.90 or lower. After the last Bid of 99.95, a gap leads to the first available Bid of 99.60, triggering the order; execution occurs at 99.55. The dashed line does not represent available prices within the gap. Check the activation condition in the agreement; an ordinary stop guarantees no price.

9.5. Position size: models for choosing volume

The article runs the conversion chain to the end, with no jumps: capital and its base, the risk percentage, the exit distance in pips, the pip value, the move into the account currency, the raw volume, the volume after rounding down to the step available on the account, and the real risk recomputed after that rounding. A separate example shows a calculation in which the output of the formula falls below the minimum volume – and the honest conclusion is to skip the trade, not to raise the risk.

9.5 compares position-sizing models as answers to different questions, then works through costs, minimum size, volume step and margin [3]. The Kelly criterion requires a distribution model, rather than just two averages from any sample. In the two-outcome version, its dependence on p at a fixed RR is linear; joint errors in both estimates can substantially change the result. The article separates the theoretical result, a fraction of that result, and your own loss limits. Examples with different holding horizons show why a shorter holding period alone does not imply a smaller budget.

9.6. Reward-to-risk, hit rate and expected value

This is the only place in the section where these three concepts are defined; the remaining articles take them from here as input parameters. The definitions are given so that it is clear which number is which: the reward-to-risk ratio is the quotient of the average win and the average loss measured in the same unit on closed trades, and not an announcement from a trade plan. The article derives expected value and the break-even hit rate, shows the substitution with units, and states by how many percentage points the cost of execution shifts that threshold. Two sets of parameters with the same expected value and a different path show why expected value alone is not enough to assess an account, and an extended section on "what expected value does not tell you" lists one by one the questions that number does not answer – from the single trade, through the stability of parameters over time, to the share of a handful of trades in the whole result. The article asserts nothing about an edge: it computes the consequences of parameters that come from your own records and carry uncertainty. It is closed by a task with a model solution.

9.7. Managing an open position: break-even, trailing exit and scaling

Every intervention in an open position changes the distribution of its outcomes, but the directions of that change are not a property of the operation itself: in the article's flagship calculation the hit rate computed after costs falls while the reward-to-risk ratio rises – the opposite of what intuition suggests. The question is therefore not whether intervention helps, but what exactly it does to three parameters at once and in which direction expected value moves once the numbers are substituted.

Changes to an open position must be compared on the same paths. 9.7 settles a full exit, moving the protective level and a partial close relative to initial R₀, including the cost of every operation. It does not assume that a partial close always raises the win rate or reduces dispersion. The trailing exit's parameter and update time are part of the rule. Moving the level further away or enlarging the position requires another check of the full loss and shared limits; the original entry approval does not justify it by itself.

9.8. Drawdown of capital: two definitions and a halt procedure

Drawdown has no single definition, and the choice of definition changes the number describing one and the same path. The article measures it twice on a single account path. From a peak of 51 200 PLN and two different bottoms – 46 800 PLN on the settled balance and 45 200 PLN on current equity – come (51 200 − 46 800) / 51 200 = 8.59% and (51 200 − 45 200) / 51 200 = 11.72%. With the halt threshold set in that example at −10%, the first definition does not trigger it and the second does; at a threshold of −15% neither does. Those are results from 9.8, carried over here together with the data. Only after that settlement does it make sense to compare anybody's results or to set any threshold at all.

A budget threshold is not evidence that a rule has failed, and a historical maximum is not a guaranteed lower bound on future drawdown. 9.8 compares history with an explicit simulation model and horizon. Quantiles follow from that model's assumptions; future results can exceed them or remain below them. The procedure has four fields: condition, scope, duration, and resumption including exposure. Recovering part of the capital does not replace a parameter review. The exercise requires both drawdown definitions on one path and your own action conditions.

9.9. Pair correlation, hidden exposure and portfolio risk

9.9 calculates three separate portfolio quantities: notional currency legs, volatility with covariances, and the sum of losses in an exit scenario. Three positions can share exposure, but their directions and notionals must be known. Stop-loss amounts do not stand in for standard deviations; a purported correlation of one does not justify adding those amounts. Matrix checks cover symmetry and the validity of the whole set of correlations, as well as the range of each cell.

Correlation is treated here as a measurement, not as a constant: every value given has a period, an observation frequency, a transformation of the series and a number of observations, and the article shows how changing each of those four parameters changes the result for the same pair. The sign is settled separately: the correlation of price series is not the same thing as the correlation of the results of two positions, because the directions of those positions also decide the sign. A set of several rules run at once is treated like a portfolio – with a shared limit, a division of the risk budget, and behaviour when two rules signal at the same time.

Positive EUR and GBP nodes connect to one negative USD node, showing the notional legs of two long EUR/USD and GBP/USD positions.
Buying EUR/USD and GBP/USD creates positive EUR and GBP legs and two negative USD legs. The shared node shows notional-exposure concentration. This illustration is not a correlation measurement; different-currency notionals require conversion before comparison.
A liquidity episode requires its own scenario. The Reserve Bank of Australia describes the event on 3 January 2019: at about 9:36 AEDT, the yen appreciated against USD by around 3% in thirty seconds; during the episode, AUD fell by around 7% against JPY. The report discusses low liquidity and the possible contribution of automated orders and strategies amplifying the move [11]. This is not a measurement of changing correlation: several simultaneous moves are insufficient to estimate it or disprove a constant parameter. The practical implication for the plan is that a dispersion matrix and a liquidity-loss scenario describe different things. Prices and the event calculation are in 9.10, and the definition of the matrix and signed exposures in 9.9.

9.10. Event risk, price gaps and the limits of value at risk

The article deals with the case in which the market passes through the protective exit level without an intermediate price. The mechanism is described from the execution side: triggering when the level is breached, execution at the first available price, no quotes in between. The loss is then computed separately, as volume × pip value × the actual distance of the fill price, and compared with the planned distance and execution deviation – shown on two jump scenarios, with a comparison against the planned risk and against capital.

The second part is the value-at-risk card: the definition, the confidence level, the horizon, the estimation method, the window and the number of observations, the full list of assumptions, and a separate list of the things the measure does not tell you. The two fields describing the measurement are required on the same footing as the rest – as long as they remain empty, the row with the result is a calculation on an assumed parameter, not a reading from any series.

VaR describes a loss quantile for a specified method, confidence level and horizon [4]. It does not describe the size of every exceedance. The error of the normal-distribution version depends on the distribution and quantile level; its direction cannot be assumed from the phrase “fat tails” alone. 9.10 also compares expected tail loss and stress, which answer different questions. Historical examples of disrupted liquidity have direct sources and a qualification about the type of price.

9.11. The risk management plan

9.11 transfers the methods' results into eleven plan fields, grouped into seven operational parts [10]. It adds a check that they are jointly feasible: a full budget definition, remaining limits, reservations for open and pending positions, and update times. Completing one field is insufficient if several fields use the same free budget without reserving it. The example shows both an incorrect notional volume and the correct cost-adjusted result.

A separate clash section takes pairs of parameters that can be written down independently but not executed that way: how many positions fit within the daily limit at the assumed execution overhead, what happens to the monthly limit after entering the month in the red, and how much closing half of an open position really costs. Alongside, six consistency tests the plan must pass before it comes into force.

Beyond the template itself stand a pre-entry control card, restricted solely to risk parameters, and a version card: one row per parameter change, with the day it takes effect and a list of the numbers recomputed along with it. The article contains no entry rules and no trade journal – a risk management plan is a part of a trading plan, not the whole of it, and the format of the records belongs to the workshop section; the recording standard here covers only the versioning of the plan itself. The task at the end consists of filling the plan in with your own numbers and pointing to the parameter that a losing streak does not change automatically – because the percentage stays the same and the amounts recompute themselves from the base.

A valid position can be too large for the rest of the day

An explicit budget example, without notionals or correlations: after an earlier loss, PLN 250 of the day's limit remains. Two positions require a further total of PLN 180 from current equity in the scenario. PLN 70 remains for a new loss. An order with its own PLN 100 limit may satisfy the percentage per trade while exceeding the shared budget by PLN 30.

Neither the strategy's win rate nor the number of symbols settles this decision. 9.3 determines the remaining limits, 9.7 the loss on the existing position, and 9.11 combines the readings in the plan. Calculate every amount from the same moment, without adding a loss already included in equity again.

5. How to read this section and what to have at hand

If you are starting from zero, read in order: 9.1 and 9.2 require no experience at all and set up every number that appears later. If you are looking for a specific settlement, go straight to it – with the caveat that you will only perform the volume calculation from 9.5 correctly if you have first settled the base of the percentage (9.3) and the placement of the invalidation point (9.4). A calculator that computes a volume without those two settlements returns a number that cannot be interpreted.

  • You do not know how much to risk per position or what to compute that percentage from → 9.3, then 9.5.
  • You do not know where to put the protective exit and you set it from the amount you are willing to lose → 9.4.
  • A losing streak is under way and you do not know whether it falls within the range your parameters explain → 9.2 and 9.8.
  • You have several open positions and do not know what the aggregate risk is → 9.9, the halt threshold → 9.8.
  • You do not know whether your closed trades after costs give positive expected value → 9.6.
  • You want to write the decisions down straight away → 9.11; you will fill the fields with numbers from the articles the plan refers to at each of them.
Three routes through eleven lessons: starting from the basics, from a specific question or from your own plan.
A map of eleven lessons with their current numbers. Arrows help select a route to the needed calculation; they replace no definitions and prescribe no compulsory reading order.

What to have at hand before you start calculating

The calculations in this section only become feasible once you have the full set of inputs. Without them every result will look correct and will mean nothing.

  • Capital and its base – a specific amount in the account currency together with the settlement of whether you compute from the settled balance or from current equity, and the moment at which you measure it.
  • The account currency and the conversion rate – a rate with a date and an indication of whether it is the bid, the ask or the mid; if the quote currency is the account currency, you write a conversion factor equal to one rather than omitting it.
  • The instrument specification – contract size, pip value, minimum volume and volume step. The last two decide whether a percentage rule is feasible on your account at all.
  • The cost components separately – spread, commission on opening and closing, swap points, assumed slippage. Not one aggregate amount, because each of those components enters the calculation at a different place.
  • Parameters from your own records – hit rate, average win and average loss computed on closed trades after costs, together with the number of trades on which you estimated them. The number of trades matters here as much as the values themselves: without it there is no telling whether the measured hit rate is distinguishable from the break-even threshold or falls within the error of your own measurement – and that is settled by calculation, not by judgement. The calculation is run by Article 9.1, and the comparison with the threshold is carried out by Part C of the task.

What this section does not contain

This section takes quantities from the neighbouring sections as inputs and does not set them out a second time. The table below says where to look for each of them and what exactly the risk section takes from it.

The boundaries of the section – matters settled elsewhere and the way they enter the risk calculation
MatterWhere it is settledWhat the risk section takes from it
Pip value, contract size, volume step, required margin, current equityForex basicsInput quantities of the calculation, recalled in a single sentence and never derived afresh
Price structure, local extremes, the average true range, the choice of intervalTechnical analysisThe distance of the protective exit as an input number for the volume formula
The calendar of macroeconomic releases and the character of the readingsFundamental analysisOnly the information that in a release window execution conditions depart from the typical
Positioning reports and the background of participant positioningPositioning reportsNothing – they enter no risk calculation in this section
Entry rules, trading styles, testing methodology and sample splittingTrading strategiesHit rate, reward-to-risk ratio and expected value as input parameters
Emotions, the pressure after a losing streak, returning to trading after a breakTrader psychologyNothing – the risk section writes down a rule, a threshold and a condition, it does not describe the experience
Order execution models, choosing and assessing a broker, execution infrastructureChoosing a broker, platforms, regulationSpread, slippage and the gap as parameters that worsen the result of the calculation
The full trading plan, the trade journal and the format of the recordsThe trader's workshopNothing – the section records only its own risk parameters and their successive versions

What you will be able to do after the whole section

The list is checkable: you carry out every item on your own numbers, not on an example from the text.

  • Name the base of every risk figure and show how far the results diverge under each of the possible bases.
  • Compute a position size from capital, the risk percentage and the distance of the protective exit, with the currency conversion and with rounding to the volume step.
  • Tell planned risk from realised risk and compute the difference arising from spread, commission, slippage and the gap.
  • Establish the invalidation point before the volume, not after it.
  • Compute how much has to be recovered after a drawdown of a given size, and explain why the relation is non-linear.
  • Estimate risk of ruin in a specified model with a barrier and horizon, and identify what its assumptions omit; the direction of error cannot always be determined without data.
  • Compute expected value and the break-even hit rate after costs, and state by how much cost moved that threshold.
  • Write down the limits – per trade, daily and weekly – together with the halt condition and the condition for resuming trading.
  • Assess the aggregate exposure of several positions and identify hidden exposure to a shared currency.
  • List what value at risk does not tell you, and state which measure answers the question about the tail of the distribution.
  • Recognise when a result comes from an arithmetic identity and when from a model resting on an assumed distribution and estimated parameters.
  • Assemble every decision into one written plan in which each parameter has its own number and its own condition for change.

If any of those items sounds like a foreign language, start with the section on the basics – not because of the level of difficulty, but because the formulas in articles 9.2–9.5 operate on quantities defined precisely there, and without them you will substitute numbers that do not mean what you think they mean. The inputs to the calculations – hit rate, reward-to-risk ratio, expected value – come from your own records; the way they are measured is described in Article 8.2.

What updates after a loss, and what waits for a review

Under a fixed-fraction policy of 1%, PLN 20,000 of capital gives PLN 200. After a full net loss of PLN 200, the new PLN 19,800 base gives PLN 198. Neither the percentage nor the rule changed—the data changed. The daily limit may still refer to its own opening base, and the drawdown threshold to the previous peak. Do not recalculate all three from current equity simply because it is the one visible number.

A percentage change following a review creates a separate policy version. Its date and scope also matter for positions already open. Amounts and versions are organised in 9.11 [10].

6. Task: assemble the data and name the owner of the question

This task does not compute a position size or a halt threshold – those calculations have their owners in the later numbers. It checks two things that are the competence of this page: whether you have the full set of inputs together with their status, and whether you can name the quantity that assigns a question to the right article – including when two numbers sound equally sensible. The third part is a single comparison: whether your own hit rate is measured accurately enough to enter the later formulas at all – and which article runs the calculation that shows it.

Part A. The data card

Fill in eight items for your account. Next to each one add a status: measurement, account parameter, regulatory parameter or assumed value. An item without a status counts as missing.

  1. Capital in the account currency, its base – settled balance or current equity – and the moment of measurement.
  2. The account currency.
  3. The conversion rate into the account currency: value, pair, side of the rate (bid, ask, mid), date and time. If the quote currency is the account currency, enter 1.0000 rather than leaving the field empty.
  4. Contract size and pip value for one lot on your instrument.
  5. Minimum volume and volume step from your account's specification table.
  6. The cost components separately, with units: spread, commission on opening and on closing, swap points, assumed slippage.
  7. Hit rate p, average win, average loss and the number of closed trades N on which you computed them.
  8. The distance of the protective exit you intend to use, and the way it was determined.

Part B. Who settles it – and how you can tell

Each of the six cases gives two defensible article numbers. Choose the one that owns the question and add one sentence: which quantity decides the issue. A number alone does not count as an answer—the ability to name the criterion is precisely what is being checked.

  1. I closed half the position after a move in my favour and I do not know whether I improved the arithmetic or merely smoothed the path. Candidates: 9.6 or 9.7.
  2. The percentage rule gives a position smaller than the minimum on my account and I want to know from what level of capital it is feasible at all on this instrument. Candidates: 9.3 or 9.5.
  3. The daily limit is used up, the weekly one is not – and I have an open position and a signal for another. Candidates: 9.3 or 9.8.
  4. Three weeks ago I changed my risk percentage and I do not remember from when it applies or which numbers I recomputed along with it. Candidates: 9.3 or 9.11.
  5. The value at risk computed on my portfolio came out smaller than the loss from a single day last month. Candidates: 9.9 or 9.10.
  6. Two of my rules signalled on the same day, on pairs sharing one currency. Candidates: 9.3 or 9.9.

Part C. How data uncertainty changes the decision

The win rate enters expected-value calculations and position-sizing models as a number, but it is a sample estimate with its own error. Uncertainty in that estimate—the standard error SE = √(p(1 − p) / N), an interval around the measured win rate and the assumptions under which it holds—is covered by Article 9.1; the break-even win rate p* = 1 / (1 + RR) and its version after execution costs are defined in Article 9.6. Here, both quantities are inputs, leaving one decision: take your win rate p and number of closed trades N from Part A and check where the threshold lies relative to the interval boundaries. If it lies inside the interval, this test under its assumptions does not rule out p on either side of break-even. That does not prove that no further conclusion can be reached with another valid method. RR must have one consistent definition in this comparison: with mean wins and losses measured net, do not add the same costs again. For notional gross endpoints and a positive cost, derive a separate threshold after costs. Uncertainty in the means and changing conditions require additional assessment.

Solution

Part A. There is no solution and there cannot be one – the numbers are yours. What is checkable is the completeness of the card, and four omissions come back most often: a rate without a side and without a date; a risk percentage without an indication of the base and the moment of measurement; cost entered as one aggregate amount instead of four components that enter the calculation in different places; and a hit rate given without N. Each of those omissions immobilises the later calculations regardless of how carefully they are performed.

Part B. There is one criterion, and it, not the number, is the substance of the answer: what settles the matter is the quantity that has to be computed, not the topic the question is about. The second number in each pair usually supplies an input – which is why it sounds credible.

  1. 9.7. What settles it is the change in the distribution of outcomes caused by intervening in an open position – the same sequence has to be settled with the intervention and without it. 9.6 only defines the three quantities by which the result is measured.
  2. 9.5. What settles it is the minimum volume and the volume step from the specification table: the question concerns the capital threshold below which the smallest available position breaks the limit at the very moment of opening. 9.3 establishes how much you may risk, but it does not check whether that rule is feasible on the instrument.
  3. 9.3. What settles it is precedence in the cascade of periodic limits and which threshold binds when two say different things. 9.8 handles a threshold computed from the drawdown relative to the peak – a different counter, a different base, a different moment.
  4. 9.11. What settles it is the record: the day the new value takes effect and the list of parameters recomputed along with it. The value itself is derived by 9.3, but a derivation does not say from when it applies.
  5. 9.10. What settles it is that the question concerns the behaviour of the measure beyond the assumed confidence level – and a quantile is by definition silent about that. 9.9 will compute the aggregate exposure, but it will not answer why the number came out smaller than the observed loss.
  6. 9.9. What settles it is the shared factor: two signals on pairs sharing a currency may be a single directional bet. 9.3 will give a shared limit in money, but it will not say whether that is one position or two.

If at any point you chose the second number and can name the quantity behind it, read both articles – the pairs are chosen so that the boundary between them is thin and runs exactly where one calculation ends and the other begins.

Part C—an explicit educational sample: 36 wins in 80 closed trades, giving p̂ = 0.45, with an assumed RR = 1.5. This is not the result of a real trading account. The parameters are dimensionless. Independent outcomes and a constant win rate are assumed only to illustrate uncertainty in a model input.

The approximate 95% Wald interval under those assumptions is p̂ ±1.96√[p̂(1−p̂)/N] = 34.10–55.90%. The notional break-even win rate without costs for RR = 1.5 is 40%; its definition and net version are in 9.6. These numbers show a range of possible inputs. They are not a standalone test of strategy quality.

The 40% threshold lies within this approximate interval, so this procedure does not rule out parameters on either side of notional break-even. It does not give the exact probability that the rule loses money. Uncertainty is particularly relevant to a fraction-sizing model based on p and RR, such as Kelly; it does not prevent setting a conservative monetary budget. Position sizing remains the responsibility of 9.5.

With dependent outcomes, uncertainty can increase or decrease depending on the covariance structure. The model excludes uncertainty in RR, costs and changes in the distribution over time. A rule change requires identifying a separate period, rather than automatically deleting earlier records. Testing methodology belongs to Section 8; here we identify only the boundary of the data used for a risk decision.

A number in a table is not always a measurement

1−0.99¹⁰ = 9.5618% is the result of an explicit rule of net losses of 1% of the current base. Streak probability also requires a model of win rate, dependence and horizon. The percentage of losing accounts requires a specific study and population. These three kinds of number have different foundations and should not all receive the same “market data” label.

The lessons distinguish conditional arithmetic, model illustrations and observations from a source. A simulation has code, a distribution and a number of repetitions. History has a source identifying the price type. Your own plan parameter needs justification, rather than the appearance of a statistic.

7. What comes next

After going through the whole section you should have a written document containing specific numbers: the risk base, the percentage per trade with its derivation, the way the invalidation point is determined, the cascade of periodic limits, the halt threshold and the resumption condition, and the limit on correlated exposure. Each of those items has to follow from a calculation you can repeat on changed data – because parameters change, and a plan you cannot recompute ages along with them.

Two closing caveats. First: risk management does not turn a loss-making set of rules into a profitable one and does not eliminate the possibility of a permanent loss of capital – it changes the distribution of outcomes and lengthens the sample on which anything can be assessed. Second: none of the numbers in this section is a recommendation. The risk percentage, the halt threshold and the exposure limit are the results of a calculation carried out on your assumptions, and when you change an assumption, the result changes.

Eleven plan fields from lesson 9.11: risk meaning and survival, limits, invalidation, volume, outcome model and open position, then drawdown, portfolio, events and updates; alongside the check 250−180=PLN 70.
The eleven names correspond to lesson 9.11’s plan fields. Enter your own parameters, bases, reading times and update conditions. A halt threshold does not imply automatic resumption. PLN 250, 180 and 70 form a separate hypothetical example, rather than recommended limits; eight daily checks verify plan execution, rather than strategy profitability.

The next section – trader psychology – deals with what this section by design does not describe: the gap between a rule written down and a rule executed. Here only the first side of that equation is settled.

We start with the demarcation – with the question of what exactly a risk rule does, given that it does not improve the quality of an entry rule.

Risk management tools: Calculate exact lot sizes in the Position Size Calculator, measure recovery math in the Drawdown Calculator, check overlapping currency legs in Portfolio Exposure, and stress-test sequences in the Monte Carlo Simulator.

FAQ

How do I choose a risk percentage?
Record the full budget definition, base and measurement time. Derive the fraction from the acceptable loss and streak scenario, then check costs and minimum size. The illustrative rates are not a standard; start with 9.3.
Does a stop-loss guarantee the loss limit?
An ordinary order does not guarantee an execution price. Separate the hypothesis condition, activation and closing price. An incorrect quote side or a gap requires a full calculation—9.4 and 9.10.
Does Kelly choose a safe stake?
The model maximises a specified growth objective under distribution assumptions. It is not a universal loss limit. Estimates of p and RR can both be wrong, and any sample with the same averages is not automatically a two-outcome model. The assumptions and sensitivity are shown in 9.5.
Is historical drawdown a boundary for the future?
No. The value depends on the path, measurement and sample length; a future period may have a smaller or larger drawdown. A budget threshold and a statistical signal of model change are separate decisions—9.8.
Does trading several currency pairs spread risk?
You need the directions and notionals of the currency legs and the dependence between outcomes. The sum of full losses is a scenario, rather than sigma or notional exposure. Three symbols alone do not establish diversification—9.9.
Does a risk plan replace a profitable strategy?
No. It sets exposure size and actions after a loss, but guarantees neither profitability nor survival. Entry and exit rules are assessed in Section 8. Data on p, RR and the net mean are evaluated in 9.6, and the combined document in 9.11. Its calculations also require the concepts from Section 2.

Sources

  1. Method and educational assumptions: 9.3—deriving the percentage. External sources and their scope of use are listed in the relevant article.
  2. Method and educational assumptions: 9.2—explicit ruin models. External sources and their scope of use are listed in the relevant article.
  3. Method and educational assumptions: 9.5—volume and costs. External sources and their scope of use are listed in the relevant article.
  4. Method and educational assumptions: 9.10—VaR and stress. External sources and their scope of use are listed in the relevant article.
  5. ESMA, announcement of 27 March 2018, ESMA71-98-128. Historical analyses of retail CFD accounts in different jurisdictions. This is not the “Final Report” numbered ESMA35-43-1135.
  6. KNF, Decision DAS.456.2.2019, operative provisions, points 1–15. PDF pages 1–6 read on 5 October 2026. The expired temporary ESMA decision is not cited as current law for these parameters.
  7. KNF, product interventions. Territorial scope and the nature of national decisions; page updated on 25 June 2026 and read on 5 October 2026.
  8. SNB, announcement of 15 January 2015. The full document was read: it confirms the date and decision, but not the time or price of a retail-market low.
  9. Historical execution description and direct price sources: 9.10—event risk. The hub does not provide unverified EUR/CHF closing prices.
  10. Recording results, checking the shared budget and versions: 9.11—risk management plan. Examples of an assumed scenario remain illustrative.
  11. Reserve Bank of Australia, Box B: The Recent Japanese Yen Flash Event, Statement on Monetary Policy, February 2019. The complete HTML article was read on 5 October 2026; it describes the event and liquidity without estimating correlation or providing execution data for a specific account.

Related glossary terms

Use these definitions while working through this section.

Jarosław Wasiński

Editor-in-chief, ForexMechanics.com • Macro & FX analyst

Independent analyst and practitioner with over 20 years of experience in financial markets. Actively involved in Forex since 2007, with a focus on fundamental analysis, OTC market structure, and disciplined capital risk management.

  • Creator and editor-in-chief of MyBank.pl, providing reliable financial education since 2004.
  • Author of hundreds of in-depth market commentaries, structural analyses, and educational materials for traders and investors.
  • Advocate for market transparency, promoting education grounded in hard data and institutional reports (BIS, ESMA, CFTC).

All content is educational and informational only – it does not constitute investment advice or a recommendation. Remember: leveraged trading (Forex/CFD) carries a high risk of rapid capital loss.