Position sizing — when to increase size

This lesson explains when journal metrics justify increasing position size and when they do not. It is designed for a trader who has already run a sample of trades on a small size and wants to know whether the data support a step up, rather than relying on a good streak.

Thesis of this lesson: size grows with evidence, not emotion

After a good series, it is natural to think I will double because I feel in form. That is the shortest path to drawdown. You increase size when journal metrics show a repeatable process: positive expectancy, setup concentration, plan compliance, and acceptable drawdown. Without these four conditions, increasing size is gambling.

Four scale-up conditions

| Condition | How to measure | | --- | --- | --- | | Positive expectancy | average in R per trade, from a sample of dozens of trades | | Setup concentration | per cent of trades complying with one defined setup | | Plan compliance | per cent of trades executed according to rules, not just result | | Drawdown | maximum drawdown in R and per cent — stable and acceptable |

If any of these metrics are not measured, you do not scale up — you return to the journal and the setup definition lesson.

Calculate expectancy before you increase anything

E (in R) = (win_rate x average_win) - (1 - win_rate) x average_loss

Illustrative example: win rate 45%, average win 2R, average loss 1R.

E = 0.45 x 2 - 0.55 x 1 = 0.90 - 0.55 = 0.35R per trade

That sounds good — but the number alone tells only half the truth. The other half is the question of whether this sample even allows you to draw any conclusion.

Why 50 trades is often not enough

Every measurement has uncertainty. For a sample average, it is calculated as standard error:

SE = standard_deviation_of_result_in_R / square_root(number_of_trades)

Illustrative example (standard deviation 1.5R):

This is arithmetic, not opinion, and it explains the most common misunderstanding in scaling: 50 good trades is not proof. Your I feel the system works after two months may be entirely noise. Increasing size on that basis is not optimisation, it is raising the stake on an unknown distribution.

Practical consequence: build a sample before you think about a size jump, and determine sample size from the table above for your standard deviation. The more choppy the system (higher deviation), the larger the sample it needs.

Two numbers worth distinguishing, because they are easy to confuse. Statistical minimum is the smallest sample at which edge stops being indistinguishable from zero — for typical assumptions (standard deviation 1.2R, edge 0.3R), it comes out to about 62 trades; derivation is in the expectancy and drawdown lesson. Observation window is the number of trades after which you allow a size jump — and that is intentionally longer, because you add margin for costs, market regime change, and your own execution errors. That is why stage 0 in the ladder below requires 100 trades, not 62.

Scale-up ladder

The numbers are illustrative — the point is the structure: 25–50% step, not a jump, and a defined return condition. Without a return condition, the ladder is a declaration, not a procedure.

Scaling journal

Decision reason is recorded as numbers, not mood. If the reason column contains words like certainty or good streak, the procedure was broken — and that is the only entry that should really stop you.

Scale-up procedure

1. 25–50% step — for example, from 1% risk to 1.25–1.5% risk (never a jump up).
2. Observation window — next 10–20 trades at new size; metrics must hold.
3. Return on deterioration — drop in compliance, rising drawdown, or emotional deviations means returning to smaller size.
4. Decision recording — in the journal, you record the scale-up date and reason (metrics, not certainty).

Rule: scale-up is a reversible process. There is no shame in reducing size when metrics worsen — that is part of the procedure.

Why a good month is not enough

A month is usually too small a sample, and a good month can be the result of:

• one large win (an anomaly that strongly inflates the average),
• rule-breaking that paid off this time,
• favourable market conditions that do not last.

What counts is process repeatability, not the monthly amount. See expectancy and drawdown — only a series and drawdown tell whether the process holds.

When scaling says do not increase

1. Confidence interval around E includes zero. You have no proof of edge — you have a series.
2. Plan compliance is below the threshold. If you break rules and the result is positive, you are measuring luck, not system.
3. Persistent drawdown. Increasing size during drawdown is doubling the stake at an unknown point in the distribution — the classic path to deeper loss.
4. You changed setup during sample collection. A new setup starts a new sample from zero, and old data no longer describe what you are doing.

Common mistakes

• Scaling after a series of wins. That is emotion, not metric.
• Size jump because I am certain. Certainty is a state of mind, not data.
• Doubling size after losses (chasing). See daily limit.
• Scale-up without an observation window. New size must pass a test before it stays.
• Calculating expectancy without uncertainty. A single average without standard error is a number without a range, that is, a number that cannot be assessed.

Measurable exercise

Assess your own state against the four scale-up conditions. For each, write a concrete number or no data. Then calculate two things: expectancy from the formula and standard error of your sample (standard deviation of result in R divided by the square root of number of trades). Record the interval around E and check whether it includes zero. Only then — if it does not — plan a 25% step with an observation window and a return condition. This is a plan on paper, not a promise to be careful.

Common questions

How large a scale-up step is safe?

Usually a 25–50 percent increase in risk, with an observation window of 10–20 trades. A sudden doubling is a change of game, not a step.

Can I reduce size after losses?

Yes — and that is part of the procedure. Reducing size when metrics worsen is discipline, not failure.

How many trades are enough to consider the system working?

That depends on the variability of results, not on their count alone. Calculate standard error: standard deviation of result in R divided by the square root of the number of trades. As long as the interval around expectancy includes zero, you cannot distinguish edge from noise — regardless of how good the equity curve looks.

What to do when there is not enough data for scale-up?

You do not scale up. First build a sample in the journal; without it, every position size is guessing — and a larger size is faster guessing.