Capital scaling – readiness, controls and reversing a size change
A larger account and a larger position are different decisions. Additional capital can provide a greater safety buffer, while increasing the risk percentage can deepen drawdowns without improving the strategy. This article compares three ways to change scale using common assumptions and examines their effect on cash risk, volume and execution costs. We explain how to prepare a decision card, specify rollback conditions and control aggregate exposure. We also discuss small accounts, alternative funding arrangements and a development plan based on skills. These examples are neither profit forecasts nor a prescription to increase positions. They help you assess whether a new scale fits your limits and whether you can evaluate it using reliable records.
Name the change before calculating the new size
Adding money, changing a risk percentage and increasing volume are different operations. In this hypothetical worksheet, account value is USD 18,000 and the fixed planned-risk fraction is 0.4%, giving a USD 72 budget. A later USD 6,000 deposit would raise the stated account value to USD 24,000. The fraction stays unchanged, giving a USD 96 budget.
The deposit is a cash flow, not a trading profit. It does not establish readiness for a larger position. We use these assumed values only to examine calculation consistency; they are not recommended account sizes or risk fractions.
The instrument specification in this example permits a0.05-lot volume step. Planned stop distance is 15 pips and assumed pip value is USD 10 per lot per pip. The separately added all-in round-trip charge is initially USD 8 per lot. These parameters must come from the actual contract and cost model in a real calculation.
Round the volume while respecting the stated budget
Estimated stop-distance loss is 15 ×10 = USD 150 per lot. Adding the USD 8 charge gives USD 158 per lot. At the original budget, raw volume is 72/158 ≈0.4557 lot. The allowed 0.45 lot gives estimated risk 0.45 ×158 = USD 71.10. Rounding up to 0.50 lot would give USD 79, above the USD 72 budget.
| Account value | Budget | Allowed candidate | Estimated risk |
|---|---|---|---|
| USD 18,000 | USD 72 | 0.45 lot | USD 71.10 |
| USD 24,000 | USD 96 | 0.60 lot | USD 94.80 |
After the deposit,0.60 lot fits the new budget under the same assumption because 0.60 ×158 = USD 94.80. The example rounds down on the stated volume grid; rounding to the nearest arbitrary decimal may create an invalid quantity or exceed the budget. Margin and other limits still need their own checks.
Recalculate when the new-size cost assumption changes
Suppose a separate hypothetical execution-cost review changes the charge estimate at the larger size from USD 8 to USD 20 per lot. The stop-distance component stays USD 150 per lot. Estimated all-in risk is now USD 170 per lot, so 0.60 lot implies USD 102. That exceeds the unchanged USD 96 budget by USD 6.
On the same 0.05-lot grid,0.55 lot implies 0.55 ×170 = USD 93.50. It fits this particular calculation. This does not prove the estimate will cover an actual gap, fill or changing pip value. The calculation is a conditional planning model, not a guaranteed maximum loss.
The important change is not the arithmetic alone: the old quantity was approved under a different assumption. Keep the cost-source version and time beside the candidate quantity. A revised input requires a fresh check before treating the earlier result as usable.
Define what is reversed if the trial is stopped
A rollback rule should identify the operation being reversed. Reducing a candidate volume is different from withdrawing the deposit, lowering the risk fraction or disabling automation. State which action is authorised and what happens to existing positions, pending requests and protective exits. No order is placed in this worksheet.
Retain comparable observations before and after a real size trial, including failed or rejected requests. A comparison of unusually quiet small-size trades with stressed large-size trades cannot isolate size as the cause. If the necessary evidence is unavailable, record the limitation rather than promoting a successful short run into a capacity estimate.
The final decision can remain “do not increase yet.” A correct calculation only addresses the assumptions supplied. It does not establish a durable edge, enough liquidity or financial ability to bear the loss. Continue below for the original readiness evidence, aggregate exposure checks and reversal procedures.
Three exercises with worked answers
The exercise data are hypothetical. Choose an answer, then check the calculation.
From a conditional quantity to a controlled size trial
The detailed lesson retains the original diagrams, comparative scaling cases, readiness card and rollback procedures. Use them to separate cash flows, portfolio exposure, execution quality and evidence about the strategy itself.
1. Four different decisions: deposits, reinvestment, volume and risk percentage
The most common question on trading forums is: “I have USD 5,000 and a good quarter behind me. Can I increase my risk from 1% to 2%?” That question mixes four separate decisions. Adding capital changes the denominator, not the rule. Reinvesting profits changes the denominator without a new deposit or a decision date. Increasing volume within the same risk budget does not change the denominator – it changes execution requirements. There is one qualification: if the increase comes from a tighter SL, it also changes the exit rule and therefore the strategy itself; you are changing two things at once. A higher risk percentage changes a system parameter and the entire distribution of drawdowns – position size is a parameter of the system, not a decoration[1]. Each decision needs different evidence before implementation and a different way of reversing it.
| Decision | What changes | What stays unchanged | Evidence required before the change | How to reverse it |
|---|---|---|---|---|
| Adding capital | Account size and the cash amount of 1R | Risk percentage, playbook, exit rules | A complete journal sample reconciled with the account statement; an interval for expectancy whose lower bound is above the threshold recorded on the card – merely having the threshold outside the interval is insufficient, because an interval entirely below the threshold also excludes it; no open drawdown | Withdraw the excess to return to the previous balance; restore the previous volume in the same session |
| Reinvesting profits | The cash amount of 1R rises continuously, without a decision or a date | Risk percentage, playbook, source of capital | A defined trigger for recalculating position size, such as every 5% change in balance, recorded in the plan before the period, not after a good month | Freeze the sizing denominator at the level recorded at the last review |
| Higher volume | Order size within the same risk budget. If it comes from a tighter SL, the exit rule also changes: that is already two changes | Capital, risk percentage, cash loss limit | An execution audit at the new size: slippage, SL/TP asymmetry and the execution provider's volume step | Return to the previous size; no change in capital is required |
| Higher risk percentage | A system parameter: the entire distribution of drawdowns and risk of ruin | Capital, playbook, exit rules | A separate experiment with its own sample, interval and simulated paths (15.4) – not a good quarter | Return to the previous percentage immediately when a stopping criterion is breached |
Change one thing at a time. Two simultaneous changes make it impossible to separate their effects: if performance deteriorates, you do not know which change failed. In practice: you add capital → the old risk percentage and playbook remain in place throughout an observation period planned in advance. You want to change the risk percentage → keep the old capital level and a journal documenting the change. Without isolating variables, the equity curve is a collection of coincidences rather than the result of an experiment.
The probability of a losing streak – putting a number on “statistically inevitable”
A streak of n consecutive losses with win rate WR has probability (1 − WR)n in one specific window. The probability that such a streak occurs at least once in N trades is calculated recursively over the length of the current streak, using a Markov chain. It is not given by the closed-form expression 1 − (1 − (1 − WR)n)N−n+1: that expression treats overlapping windows as independent and overstates the result. The figures below were calculated recursively:
| System WR | 4 consecutive losses in 100 trades | 5 consecutive losses in 100 | 6 consecutive losses in 200 | 8 consecutive losses in 200 |
|---|---|---|---|---|
| 50% (e.g. mean reversion with RR 1:1) | 97.3% | 81.0% | 80.1% | 32.0% |
| 55% (example: RR 1:1.5) | 91.5% | 64.7% | 60.2% | 16.5% |
| 40% (trend following with RR 1:3) | 99.9% | 97.6% | 98.5% | 75.0% |
In other words, at a 55% WR over 200 trades, representing a year of active intraday trading in this example, the probability of six consecutive losses is approximately 60%. This is not a system failure: it is a boundary condition to account for when choosing your size. With 1% risk, that streak costs 5.85% of capital, calculated with compounding; with 2% risk, it costs 11.42%. The system, market and trader are the same; position sizing alone makes the difference.
The table assumes independent trades (i.i.d.) and is exact under that assumption, not an approximation. The closed-form expression gives substantially higher values for the same parameters, such as 80.3% instead of approximately 60% for a 55% WR and six losses in 200 trades, and should not be used. In a real journal, outcomes may be autocorrelated, with losses clustering in unfavourable regimes, which can increase these probabilities. The appropriate tool is then a block bootstrap of your own R-multiple vector, described in 15.4, rather than an analytical formula.
The chart is not a forecast and does not show that risking 2% is worthwhile. It shows one thing: with the same sequence of outcomes, doubling the risk percentage almost doubles drawdown depth – 5.85% versus 11.42%, or 1.95×, less than twice as much because losses compound. It says nothing about whether the sequence will recur or whether the trader will maintain execution at the trough; that is a question about process, not arithmetic. The second panel shows the same arithmetic from the other side: reversing each outcome's sign is enough for a higher risk percentage to deepen the loss in the same proportion. The multiplier works both ways and does not create an edge by itself.
2. The scaling card – nine fields and a rollback rule
Changing size without a card leaves the decision to discretion. The card requires something that a drawdown makes much harder: recording before the change how you will recognise that the experiment has failed and exactly how you will return to the previous size. Nine fields, one page, a version and a date.
| Field | What it must contain |
|---|---|
| 1. Current position | Balance, risk percentage, cash value of 1R, playbook version, number of trades since the last change, data source (statement + journal) |
| 2. Reason for the change | What is missing at the current size and what should improve. Not “I want to earn more”, but a statement that can be checked through observation |
| 3. Parameter being changed | Exactly one of the four in section 1, with before and after values. Entering two invalidates the card |
| 4. Budget and limits | Both a percentage and a cash limit per trade; daily and weekly limits from the plan (15.2); a limit on total exposure to one risk factor |
| 5. Evidence required before the change | A dated list of the records in section 4: statement, trade export, journal with R values, playbook version, and the result of reconciling the journal with the statement |
| 6. Metrics being monitored | What you measure during the trial and how often. At minimum: expectancy with an interval, drawdown on the account-value curve, entry and exit slippage, and the percentage of trades with rule breaches |
| 7. Stopping criteria | Conditions that end the experiment regardless of profit or loss: exceeding the drawdown limit, rule breaches above the threshold in the plan, or deterioration in execution beyond the threshold in section 5 |
| 8. Rollback method | The size you return to, the deadline, the action taken (withdraw excess capital / restore the previous risk percentage / restore the previous order size), and the number of trades you remain there before the topic can even be reconsidered |
| 9. Review date or condition | A date or a trade count – with the latter matched to signal frequency (section 6), not to the calendar |
Illustrative example using synthetic data. A completed card using synthetic data for the decision to add capital:
| Field | Entry |
|---|---|
| Current position | USD 10,000, risk 1% (1R = USD 100), playbook v3.2 frozen for 4 months, 180 trades since the freeze, journal reconciled with the statement (difference: 0 positions) |
| Reason for the change | At 1R = USD 100, calculated volume is 0.3333 lots, and the 0.01 step is 3.0% of that volume. At 1R = USD 200, calculated volume is 0.6667 lots, and the same step is 1.5%. The maximum possible unused risk budget from rounding down – a full volume step – therefore falls from 3.0% to 1.5% of the budget. The actual unused amount depends on the remainder when volume is divided by the step; at this SL it is approximately 1% in both cases (USD 1.00 out of 100 and USD 2.00 out of 200). The bound on the deviation improves, not that particular observation. Testable by comparing planned and actual R0 across 30 trades with different SL distances |
| Parameter being changed | Capital USD 10,000 → USD 20,000. Risk percentage unchanged (1%), playbook unchanged (v3.2), volume calculated from the sizing formula |
| Budget and limits | 1% and no more than USD 200 per trade; daily limit of 2 full losses; total exposure to one factor, such as USD direction, limited to 2 positions |
| Evidence before the change | 12 months of statements, CSV export, journal with net_R/MAE_R/MFE_R/slippage, playbook v3.2 with change log, reconciliation record dated 2 days earlier |
| Metrics being monitored | Expectancy with an interval after 30 and 60 trades; drawdown on the account-value curve; median and 90th-percentile slippage at SL and TP; percentage of trades with rule breaches |
| Stopping criteria | Drawdown deeper than 8% from the peak — the trial threshold is tighter than the reduction trigger in section 7 (10%) because it should end the experiment before the emergency rule activates — or rule breaches in more than 8% of trades or median SL slippage more than 0.5 pips above the previous-size baseline. |
| Rollback method | A withdrawal bringing the balance back to USD 10,000 on the first business day after a stopping criterion is met. If the trial stops at an 8% drawdown, the balance is USD 18,400, so withdraw 8,400, not 10,000. If the balance is below 10,000, withdraw nothing and recalculate 1R from current capital: 1% of 8,400 is USD 84, not 100. Restore the previous volume in the same session; complete 60 trades at the old size without rule breaches before reconsidering the change on the card |
| Review condition | 60 closed trades at the new size (at 3–4 opportunities per week, approximately 4 months); no earlier, regardless of results. |
“Do not increase yet” is a procedural outcome, not a failure
The card has three possible outcomes, not two. Increase: every field is completed, evidence is complete, and stopping criteria and rollback methods are recorded. Do not increase yet: at least one field lacks support in the data. The journal has not been reconciled with the statement, the expectancy interval includes the decision threshold, the playbook changed during the trial, or you cannot describe how to reverse the change. The decision is then “maintain the current size”, to be reconsidered after a recorded number of trades. Reduce: follow the separate path in section 7.
“Do not increase yet” is operationally less costly than either alternative because it changes nothing that would subsequently have to be reversed. For it to be chosen, it must be recorded as a fully valid outcome on the card, alongside the next review date.
3. Three scaling variants: A, B and C – the cash-risk calculation
Three versions of the same ambition, “I want to trade a larger amount”, produce three different drawdown distributions. Below is an explicit calculation with common execution assumptions, without forecasting a profit: EUR/USD, planned SL 30 pips, pip value USD 10 per standard lot, volume step 0.01 lots, and an exchange rate of PLN 4.10 per USD for the Polish currency examples in this article. This is a single scenario assumption, not a current exchange-rate quote. 1R is the planned risk at entry; an outcome worse than −1R because of a gap or slippage does not change that denominator.
| Variant | Capital | Risk | 1R in USD | Volume with a 30-pip SL (0.01 step, rounded down) | Six consecutive losses | Effect on execution |
|---|---|---|---|---|---|---|
| A. More capital, same percentage | USD 20,000 | 1% | USD 200 | 0.66 lots (0.6667 rounded down) | −5.85% = −USD 1,170 | Order size doubles: check slippage and the volume step at the new size, not the old one |
| B. Same capital, higher percentage | USD 10,000 | 2% | USD 200 | 0.66 lots (0.6667 rounded down) | −11.42% = −USD 1,142 | The same execution size as A, but drawdown is approximately twice as deep as a percentage of capital |
| C. More capital, cash risk limit | USD 20,000 | 1%, capped at USD 100: effectively 0.5% | USD 100 | 0.33 lots (0.3333 rounded down) | 0.5% fractional comparison: −2.96% ≈ −USD 593. Active USD 100 cap: −3.00% = −USD 600. Rounded to 0.33 lots: −2.97% = −USD 594. | Order size stays at the previous baseline, so execution requirements do not change |
Reading the volume column. Divide 1R by the product of SL distance and pip value: 200 / (30 × 10) = 0.6667 and 100 / (30 × 10) = 0.3333 lots. Round down to the 0.01 step: 0.67 lots would mean USD 201 of risk, exceeding your own USD 200 limit. The fractional comparison uses planned 1R of 1%, 2% and 0.5% of current capital. A cash cap is a different rule: at USD 20,000, min(1% of current equity, USD 100) remains USD 100 for all six losses, so the loss is USD 600 (3.00%), not the USD 592.55 (2.9627%) produced by continuously risking 0.5%. At 0.33 lots with the stated SL, each loss is USD 99 and six cost USD 594 (2.97%). Rounding volume down makes realised model drawdowns less than 0.15 percentage points shallower than their respective unrounded calculations. Keep the cash-cap and fractional models separate rather than treating them as interchangeable.
The calculation reveals three things that “a larger account means a larger profit” does not. First, A and B have the same cash value of 1R (USD 200) and the same volume; they differ only in the fraction of capital represented by that amount. The same losing streak costs almost twice as much in B as a percentage of capital: 11.42% versus 5.85%, or 1.95×, not 2×, because losses compound. It costs less in dollars, 1,142 versus 1,170, because the same initial cash risk is applied to a smaller account. Second, C is the only variant that leaves execution unchanged from the previous baseline: the order amount remains at the size tested earlier and only the buffer grows. Third, none of these variants tells you anything about profit. Changing volume multiplies outcomes in both directions; it does not create an edge.
Use both percentage and cash limits. In practice, C combines two rules: “1% of capital, but no more than X USD per trade”. The cash ceiling protects against a percentage producing an order amount beyond the size for which execution has been tested – precisely the risk introduced by scaling. Record both figures in field 4 of the card and check both for every order; the smaller amount applies.
Shared portfolio exposure – the sum of stop-loss budgets does not shrink
A larger account makes simultaneous positions more tempting. Four positions, each risking 1%, in EUR/USD, GBP/USD, AUD/USD and NZD/USD with the same USD direction represent 4% risk if all stops are hit. The sum of those budgets does not shrink at any correlation. Correlation does something else: a high positive correlation, around +0.7 or above (an illustrative procedure threshold, not a universal market standard), increases the probability that the stops are hit together in one move. Operationally, treat that basket as one position risking 4% on one factor, USD direction, and record a total exposure limit per factor on the card, not only a per-trade limit. A correlation matrix for simultaneously open pairs is a separate review record, independent of the assessment of an individual setup.
4. Evidence required before changing size
A scaling decision rests on a sample intended to describe what actually happened. The two most common reasons it fails to do so are an incomplete journal and an interval too wide relative to the decision threshold.
Auditing sample completeness – survivorship bias in your own journal
A journal is vulnerable to two kinds of missing records that inspecting the journal alone will not reveal. You need a comparison with the execution provider's statement:
- Random omissions: a forgotten entry, a bad day, insufficient time. They reduce the number of observations without systematically shifting the result
- Selective omissions: missing losing trades that “should never have happened”, positions taken during tilt, or “test” entries. The sample then systematically overstates performance and the audit describes a fantasy
The procedure takes 15–30 minutes and determines whether further calculations are worthwhile: download the official statement for the period, compare the number of closed positions with the number of entries, reconcile the lists and check where the omissions lie. If the missing trades are mostly losses, journal expectancy is an artefact of record-keeping rather than a measure of the process. The full procedure for reconciling the journal with the execution provider's report, including deduplication, corrections and cost differences, belongs in 15.9. Here, you need its result as evidence on the card.
- The execution provider's statement for the trial period: an official file, not a platform screenshot
- A trade export with date, instrument, direction, volume, entry, SL, TP, exit, commission and swap
- A journal with net_R, MAE_R, MFE_R and slippage for each trade, not merely the result in account currency
- The current playbook version with its freeze date and changes from the previous version
- A reconciliation record for journal and statement: position counts on both sides, a list of differences and a date
- Path-simulation results for the proposed size. The procedure, code and seed are in 15.4; put the result and parameters on the card without repeating the method
Process stability – freeze the playbook before the trial
A sample describes one configuration only if that configuration remained unchanged. If you altered SL distance, added a filter or changed an entry condition during the trial, you do not have one sample of n observations. You have several smaller samples from different versions, each with a correspondingly wider interval. That is why field 1 contains the playbook version and freeze date, and field 5 contains the change log from the previous version. The card itself also needs an identifier and version: otherwise you cannot later say which scaling decision you are reversing.
The other essential figure is the percentage of trades with rule breaches. Results collected with frequent breaches do not describe the system alone; they describe the system plus improvisation, and it is improvisation you would be scaling. The checks before each scaling decision always follow the same order: first “are the data and execution reliable?”, then “is this the same regime?”, and only then “what does the result say?”. The detailed sequence and the criteria for maintaining, monitoring, reducing or disabling a setup belong in 15.10.
How many observations? The question is incomplete
No trade count decides anything on its own. It matters only through the width of the interval relative to the threshold recorded on the card. Calculate the expectancy interval as E ± 1.96 × std(R) / √n. This is an approximation: 1.96 comes from the normal distribution and assumes a known std(R) and independent observations. With n below roughly fifty and standard deviation estimated from the same sample, a t-distribution multiplier is appropriate; with clustered outcomes, the interval is wider than this formula suggests. The following values use std(R) = 1.4R, a scenario assumption; your value will differ:
| n trades | Half-width of the 95% expectancy interval | Implication for the card |
|---|---|---|
| 30 | ±0.50R | A direction, nothing more. The interval is wider than most decision thresholds |
| 50 | ±0.39R | Still wider than a typical threshold; the decision is usually deferred |
| 100 | ±0.27R | The interval starts to establish the sign, rarely the magnitude |
| 200 | ±0.19R | At expectancy +0.3R, the lower bound is approximately +0.11R; zero is excluded |
| 500 | ±0.12R | A narrower interval, but still one sample from particular regimes |
The wording on the card should look like this: “After n = 200 trades, the 95% expectancy interval is +0.11R to +0.49R. Method: E ± 1.96 × std(R)/√n. Assumption: independent observations; streaks and shared events make the interval wider. The lower bound is above zero, so a necessary condition is met – this is not proof of an edge.” If the interval is not entirely above the decision threshold, defer the decision and keep the current size until the next review. Repeatedly checking the interval during the trial, for example after 30 and 60 trades, increases the chance that it will eventually fall on the favourable side of the threshold. That is why the number of reviews and the decision point are recorded in advance rather than selected afterwards. Stopping trading because of operational risk, such as a loss limit or execution error, is always permitted earlier, regardless of the interval.
The metrics used alongside expectancy – profit factor, SQN, downside deviation, Calmar, drawdown duration and execution drift – each have one definition in this section of the course, published in 15.9. That article also provides the SQN interpretation scale and its observation-count cap. Enter values calculated using those definitions, together with n and the interval. A Sharpe ratio without its frequency, period and reference rate is not a threshold and does not belong in scaling criteria.
Risk of ruin at the new size – one formula and one warning
The analytical formula ((1 − A)/(1 + A))U[2] applies to a wager with fixed win and loss amounts, advantage A = 2p − 1, and U units to ruin. Substituting A = expectancy / |avgLoss| introduces a different assumption that must be stated. For A = 0.3 and U = 25, where each unit is a fixed cash amount equal to 2% of initial capital and ruin is a 50% loss, the formula gives 0.00002%. With 2% of current balance instead, 35 losses are needed to reach or exceed a 50% loss because losses compound (after 34, the loss is still below 50%). The analytical figure is so small that its useful message is the formula's inadequacy for the situation: with fat tails and autocorrelated outcomes, simulated paths can produce results orders of magnitude higher. A bootstrap of your own R vector comes closer to answering the question about risk of ruin at the new size. The procedure, code, seed and horizon are in 15.4; 200 versus 1,000 trades changes the thresholds.
5. Execution at the new size: costs, slippage, liquidity and capacity
A change in size is primarily a change in execution conditions. The entry rule stays the same, but an order for a different amount reaches a different part of the order book, rounds differently to the volume step and looks different to the execution provider. That is why every new size needs its own execution audit, starting with the first 30 trades, rather than inheriting a pass from the previous size.
Three figures to compare before and after the change
- Median slippage, measured separately at entry, SL and TP, with its sign. Warning signal: SL versus TP asymmetry greater than 0.3 pips
- 90th-percentile slippage: the distribution's tail, showing what happens around releases and in low liquidity. Warning signal: the gap between the 90th percentile and the median grows relative to the previous size. Report the gap in pips, not as a multiple of the median: signed median slippage can be zero or negative, making a “threefold” comparison undefined or misleading
- Execution drift in R: entry and exit slippage, commission and swap, divided by 1R. The definition and formula are in 15.9; here, you need the same figure calculated twice, before and after the change
Compare samples under identical conditions: the same setup, the same hours and a similar mix of context, including quiet sessions and release windows. Thirty morning trades at one size and thirty evening trades at another compare sessions, not size.
Count costs in R once, using the correct units. Spread is already included in fill prices: a long enters at ask and exits at bid. Do not subtract it again. To convert pips, cost in R = cost in pips / SL distance in pips. A 0.8-pip spread with a 50-pip SL is therefore 0.016R, equivalently 1.6% of the 1R risk unit. If journal expectancy falls from +0.40R in a test to +0.24R in the live account, that 0.16R decline corresponds to 8 pips with a 50-pip SL, ten times what the spread alone explains. Break the remainder into commission, swap and slippage; otherwise you do not know what you are scaling. Tax is not a trade-execution cost and is not included in R.
A passive test of execution asymmetry
A warning about the B-Book model remains theoretical until there is a way to observe execution. Conduct this test passively, without increasing risk; do not open larger “test” positions:
- Log for every trade: time from order submission to confirmation; signed slippage at entry, SL and TP; spread at entry; and context, distinguishing a quiet session from a release window
- Separate the records into quiet conditions and high-volatility windows. Thirty trades do not guarantee 15 in each group. If either group has fewer than roughly a dozen observations, keep collecting rather than comparing a complete group with an incomplete one
- Compare asymmetry: if release-window slippage is several times larger and systematically adverse, with greater slippage at SL than TP, you have a signal of asymmetric execution. Possible causes include the provider's model, last look, liquidity-provider quality or your own infrastructure. The trader can observe the effect, not establish the cause from it alone
- To investigate size effects, compare execution quality at small volumes, such as 0.01 versus 0.1 lots, rather than comparing profit
The test is an observation of journal records, not an invitation to take a larger position “for testing”. Without it, “I have good results at 1 lot” may mean “I have results at 1 lot because my size has not yet required different handling”.
Capacity – when size starts to move the price
On the most liquid pairs during the London session, a retail account's effect on quotes is negligible; daily FX turnover is measured in trillions of dollars[6]. Problems arise in specific places: exotic pairs during the Asian session, less-liquid crosses outside London hours, and instruments displaying a narrow spread. A narrow displayed spread alone does not establish available depth. Diagnose as above by comparing requested and obtained prices across 30 trades in the instrument. If average slippage increases with position size, you have a hypothesis of a capacity limit. Thirty trades do not distinguish that hypothesis from noise: report dispersion within each size group and check whether the difference between medians exceeds that dispersion before calling it capacity. Countermeasures include splitting orders into tranches or dropping the instrument at that size. Record either as a restriction on the card before increasing the amount.
What belongs in other course sections
- Workstation and redundancy – backup power, a spare connection, a server and configuration backups – belong in 15.1. For scaling, the relevant point is that workstation cost is approximately fixed, so its share of capital falls as the account grows. Retaining the Polish worked example: about PLN 1,300–3,300 annually is approximately 1.1–2.7% of a USD 30,000 account and 0.3–0.8% of a USD 100,000 account at the scenario rate of PLN 4.10/USD. Local equipment and service costs will differ
- Execution-provider selection, jurisdiction and compensation schemes belong in section 13. As a specifically Polish example, KDPW operates a scheme covering 100% of the first EUR 3,000 and 90% of the remainder, with a maximum payout of EUR 20,100 on eligible funds up to EUR 22,000. This is not a worldwide limit. Other jurisdictions' limits and insolvency procedures are discussed in 13.3; check the scheme applicable to your broker's legal entity
- Tax reporting belongs in section 14. Article 14.1 discusses after-tax results and the need to apply the rules of the relevant tax jurisdiction. The journal supplies records; it is not itself a tax calculation, and tax is not included in R
- Electing professional-client status under EU MiFID II requires meeting two of three criteria in Annex II[4]: 10 significant-size trades per quarter over four quarters, a portfolio above EUR 500,000, and one year working in the financial sector. All three are not required, but this is not arithmetic alone: the investment firm must assess knowledge and experience and follow a written procedure. The employment must involve a role requiring knowledge of the relevant transactions. The consequence is leaving the scope of retail product-intervention measures and their safeguards. Details are in 13.5; this particular test should not be assumed to apply outside its jurisdiction
6. One parameter at a time, with observation matched to opportunity frequency
A staged change starts by creating a queue. Section 1 explains which changes must not be combined; this is the operational side of that rule. If you want to change more than one thing – add capital, recalculate sizing and raise the risk percentage – put them in a queue and open a card for the first item only. Create the next card only after closing the previous one: after the review in field 9 and one of the three decisions. Order the changes by how cheaply they can be reversed. A deposit can be reversed by a withdrawal in one session, a risk percentage can return immediately after a stopping criterion is breached, while higher volume forced by a tighter SL changes the exit rule and takes longest to reassess because it requires a new sample. Opportunity frequency determines the length of each stage.
The observation period after changing size is neither a fixed number of months nor a universal trade count. It depends on signal frequency: you need enough observations for the interval in section 4 to narrow relative to the card's threshold, and that takes as long as your setup takes to produce the opportunities.
| Signal frequency | 30 observations | 60 observations | Implication for field 9 |
|---|---|---|---|
| 3–4 opportunities per week (intraday) | About 2 months | About 4 months | Schedule the review by trade count; the calendar date follows from it |
| 1 opportunity per week (swing) | About 7 months | More than a year | The same count implies a very different horizon: either accept less precision or accept a longer observation period |
| 2–3 opportunities per month (position trading) | About 10–15 months | About 1.5–2.5 years | At this frequency, a scaling decision based on your own sample is effectively a multi-year decision; complete cards less often and more cautiously |
The uncomfortable consequence deserves a direct statement: the rarer the signal, the longer it takes to gather evidence – not the less evidence you need. Shortening observation because “the setup offers few opportunities” changes the decision threshold, not the trading size.
Returning to a smaller size as an observation tool
A smaller cash amount is a legitimate setting for observations you do not want to make at full risk, and the place to return after an unsuccessful scaling experiment. The condition is that the arithmetic must work; otherwise “small size” is only an impression. Under the cent-account contract specification used in this example, 0.01 cent lots means 10 actual currency units, with an EUR/USD pip value of approximately USD 0.001, or USD 0.10 per cent lot. In a standard account, 0.01 standard lots means 1,000 units and about USD 0.10 per pip. A risk budget of 1% of USD 200–500, or USD 2–5, with a 20–50-pip SL therefore requires 0.4–2.5 cent lots. Check your provider's contract specification and volume step and round down. Rounding up exceeds the budget and, at such small amounts, can exceed it by tens of percent. One risk figure, with one volume calculated from it – not three different numbers in three places in the plan.
Market regime as a journal label
“A sample from more than one regime” remains an abstract requirement until regimes are labelled. Four dimensions can be recorded for each trade in a regime_tag column. The thresholds are strategy parameters, not market standards (illustrative procedure thresholds, not universal market standards):
| Dimension | Regime A | Regime B |
|---|---|---|
| Directionality | Trend: ADX(14) D1 ≥ 25 for more than 50% of the period | Range: ADX D1 < 20 |
| Volatility | High ATR: ATR(14) D1 > 1.3× its 90-day average | Low ATR: ATR D1 < 0.8× its 90-day average |
| Macro context | A period with high-impact releases | A period without high-impact releases |
| Risk appetite | Risk-on proxy: a falling dollar index | Risk-off proxy: a rising dollar index |
A label is not evidence by itself. It lets you calculate expectancy separately for each regime and check whether the difference fits within the interval. If it does not, you have a regime-dependent system and a separate sizing question for each regime. Mechanical-filter thresholds and their calibration on your own data are developed in 15.6.
7. Pausing, reducing risk and reversing a size increase
Reducing size is a downward change and requires the same discipline as increasing it: a trigger, an amount and a return condition recorded before the event. The table below is this course section's single set of figures for that procedure. The “reduce” decision comes from a review (15.10); its implementation belongs here. These values are scenario assumptions (illustrative procedure thresholds, not universal market standards). Yours follow from the limits recorded in your plan (15.2).
| Trigger | Action | Condition for returning to the baseline size |
|---|---|---|
| Drawdown deeper than 10% from peak account value | Halve position sizing, from 1% to 0.5% | 30 trades without rule breaches at the reduced size; not recovery of the loss |
| Three consecutive losses | Pause for 60 minutes and check entry conditions | The pause is over and the next entry satisfies the necessary conditions |
| The plan's weekly loss limit | Pause until the next week begins and halve sizing for that following week | One week without rule breaches (breach rate = 0), regardless of results |
| Deteriorating execution: median SL slippage rises by more than 0.5 pips | Halve sizing and conduct the execution audit in section 5 | Median and 90th-percentile slippage return to pre-change values, confirmed over 30 trades. Changing providers alone is not a return condition; a new environment needs its own audit from the first 30 trades |
| A regime change outside the card's permitted range | Halve sizing until the regime ends, or suspend the setup | The regime indicator returns to the range recorded on the card |
| A life event limiting time and attention | Pause or cut sizing to one quarter | Working conditions stabilise, confirmed by a full week of the complete workflow without shortcuts |
Reversing a size increase is separate from reducing sizing. Return to the previous cash value of 1R by withdrawing excess capital or restoring the previous risk percentage. Stay there for the trade count recorded in field 8. Only after that period can the topic return to the card, with a new rationale rather than the same one. The return condition is never “after recovering the loss”. A pause is not punishment and must not depend on the result, because that would reward risky attempts to win losses back.
Three traps that look like scaling
The underlying psychology – pressure to recover, martingale, tilt and the chain of decisions after a loss – is described in 10.5. The procedure for a limit already breached is in 15.8. Here, one thing matters: record the trigger, reduction amount and return condition before they are needed.
8. A development plan: skills and research quality, not a target balance
A development plan expressed as balances becomes a timetable the market has no obligation to follow. A plan expressed as skills can be carried out in any regime because its goals depend on what you do, not on what price does. Four goals can be assessed without looking at profit:
| Goal | Measure | Check |
|---|---|---|
| Reproducible decisions | The percentage of trades for which the journal explains entry and exit without relying on memory | Randomly select 10 entries from a month ago and reconstruct the decisions from the records alone |
| Complete records | Difference in position counts between statement and journal; percentage of entries with every required field | Reconcile with the statement after each accounting period |
| Execution quality | Median and 90th-percentile slippage, execution drift in R, and rate of rule breaches | Compare the same three measures with the previous period, not someone else's benchmark |
| Research quality | Number of hypotheses tested and rejected; trial register; number of playbook changes during the period, with reasons | Use the trial register from 15.4: were the specification and rejection threshold recorded before the test? |
None of these goals says “how much I will earn”. All ask whether the data behind the next scaling card will be worth anything. Progress depends on available time, not ambition. With three hours per week covering analysis, journal entries, screenshots and a weekend review, a few dozen well-documented trades per year is a realistic order of magnitude (estimate based on stated assumptions: check against your own time records: the source example allows approximately 135 working hours annually, about 45 weeks at three hours; some of that goes to reviews and analysis without trades, while the full cycle of preparing, executing, capturing screenshots, recording and reviewing one trade takes roughly an hour). If your setup needs several hundred observations, the problem is fitting the strategy to your life rather than motivation. Workflow variants for different time budgets are described in 15.5.
A specialist audit of readiness to change size
Illustrative example using synthetic data. Review these six dimensions before completing the card. Entries in columns 4–6 are examples and must be based on your own data:
| Readiness dimension | Evidence | Evidence limitation | Scaling decision | Observation | Rollback condition |
|---|---|---|---|---|---|
| Data quality | Journal-to-statement reconciliation record | Reconciliation cannot reveal errors shared by both sources | Permitted | Reconcile after each period | Position-count difference > 2% |
| Process stability | Rule-breach rate and number of playbook changes during the period | Measures compliance with a rule, not the quality of the rule | Permitted with the playbook frozen | Check breaches weekly | Breaches exceed the plan's threshold |
| Scope of evidence | Expectancy with an interval, observation count and regime labels | One sample from particular regimes; independence is an assumption | Permitted when the interval's lower bound exceeds the card's threshold | Interval after 30 and 60 trades | Lower bound reaches or falls below the threshold |
| Execution | Median and 90th-percentile slippage, execution drift in R | Measured at the previous amount | Only with an audit at the new amount | First 30 trades at the new size | Median SL slippage increases by more than 0.5 pips from baseline |
| Instrument capacity | Relationship between slippage and position size | Visible only after enough trades in one instrument | Permitted except in instruments with limited liquidity | Slippage by size group | Slippage increases linearly with size |
| Capital and goals outside the market | Source of funds, financial buffer and liabilities | Self-reported; cannot be verified from the journal | Only with funds whose loss would not change living circumstances | At each deposit | Deposit funded by debt or money earmarked for another purpose |
Capital you can put at risk – and external funding
There is no amount suitable for everyone, and this article will not supply one. It depends on your execution provider's fixed costs, volume step, your setup's SL distance and how much you can lose without changing your living circumstances. Two boundary conditions can be checked. First, if the minimum volume step exceeds the risk budget at your capital level, you are trading a plan forced by rounding rather than your own plan. Second, use only funds whose loss would not change your living circumstances: not a loan, a credit card or money earmarked for another purpose.
External funding, meaning accounts funded by an outside firm, should be considered as a parallel route, never a shortcut to more capital, and only after reading the actual contract. Check at least the daily and total drawdown limits and their calculation; the profit target and deadline; restrictions on weekend positions, release-window trading and hedging; refundability of the assessment fee; and conditions for profit-share payouts. The account's stated limit is not your own capital available to lose and does not replace your own sample. Some firms' business models depend on assessment fees. Regulatory proceedings against one such firm began in the United States and Canada in 2023; the US regulator's case was dismissed in 2025.
Keep the wider context in view. In research on day traders on the Taiwan stock exchange[7], about 1.6% were profitable after costs in an average year and fewer than 1% were predictably profitable. The paper circulated in several working versions with different dates and measures, so check figures against the version you use. KNF data[5] for Poland's CFD/forex market in 2025 show 72.2% of active clients losing money; that is a Polish finding, not a global estimate. These figures do not tell you that development is pointless. They show that a plan built around a target balance relies on being an exception, whereas a skills-based plan can be carried out whichever side of the outcome distribution you finish on.
FAQ – frequently asked questions
How does adding capital differ from reinvesting profits?
Can I add capital when the account is in drawdown?
How much money do I need to start trading a live account?
Can the same system work at every size?
I make money in a smaller account but start losing after increasing size. Where should diagnosis start?
Does an externally funded account replace my own capital?
How often can I change size?
What if my personal circumstances make the plan impossible to follow?
Sources and bibliography
- Tharp V.K., Trade Your Way to Financial Freedom, 2nd edition, McGraw-Hill, 2006/2007. R-multiple and position size as a system parameter underpin section 1's distinction between changing capital and changing the risk percentage.
- Vince R., The Mathematics of Money Management, Wiley, 1992. Position-sizing formulas and analytical risk of ruin, including their assumptions, underpin section 4's warning that a closed-form expression does not replace path simulations.
- European Securities and Markets Authority (ESMA), Decision (EU) 2018/796 – Product intervention measures on contracts for differences, 22 May 2018, Official Journal L 136, 1 June 2018. The restrictions applied as a temporary measure from 1 August 2018 and were not renewed beyond 31 July 2019. Poland subsequently applied KNF Decision DAS.456.2.2019 from 1 August 2019. The 74–89% range of losing accounts comes from ESMA's 27 March 2018 announcement.
- Directive 2014/65/EU (MiFID II), Annex II – criteria for elective professional clients. Two of three criteria: 10 significant-size trades per quarter over four quarters, a portfolio above EUR 500,000, and one year working in the financial sector.
- Polish Financial Supervision Authority (Komisja Nadzoru Finansowego, KNF), Wyniki klientów na rynku Forex za rok 2025 [Client results in the forex market for 2025], March 2026 annual publication. The finding that 72.2% of active retail clients incurred losses provides the specifically Polish reference point in the opening box and section 8.
- Bank for International Settlements (BIS), Triennial Central Bank Survey: OTC Foreign Exchange Turnover, April 2025 survey, announcement dated 30 September 2025. Global turnover of approximately USD 9.5 trillion per day provides context for the capacity discussion in section 5.
- Barber B.M., Lee Y-T., Liu Y-J., Odean T., “Do Day Traders Rationally Learn About Their Ability?”, working paper, UC Davis / Berkeley, circulated in several versions with differing dates and measures. Taiwan stock-exchange day traders, 1992–2006: approximately 1.6% profitable after costs in an average year and fewer than 1% predictably profitable.
Current data scope: the BIS final December 2025 summary gives USD 9.5 trillion in average daily April 2025 turnover, replacing the preliminary USD 9.6 trillion figure. KNF's 2025 data cover active clients of surveyed domestic firms, including Polish and foreign residents and a broad set of OTC instruments; 72.2% is not a worldwide percentage of all CFD accounts.
