Skip to content
Menu
Finabase Learn

Trading · Risk management

Scaling in and out of trades: partial exits, pyramiding and the math

Compare six exit plans on one entry and one stop: how partial profit-taking and pyramiding change win rate, expectancy and variability with and without an edge.

On this page
  1. One trade, measured in R
  2. Six plans with no edge
  3. The same plans with an edge
  4. Scaling in
  5. Practical details
  6. How Finabase does it
  7. Frequently asked questions
  8. Summary
  9. Check your understanding

Scaling out means closing a position in parts: sell half at a first target, hold the rest for a second. Scaling in means building it in parts: buy some now and add more later. Both are popular because they feel better than all-or-nothing decisions. Taking part of a profit relieves the fear of giving it all back, and adding to a winner feels like pressing an advantage.

Feeling better and performing better are different things. This lesson puts six common plans on the same footing, one entry and one stop, and works out exactly what each does to the win rate, the expectancy and the variability of results. The main finding is simple and often resisted: how you divide an exit changes the shape of your results, and it cannot create an edge that the entry did not have.

One trade, measured in R

Every plan below starts the same way. A trade is entered with a stop that defines the planned loss, 1R (Risk-reward ratio and R multiples: the math of a trade). Price then wanders until it hits either the stop or a target. The only thing that differs between plans is what is done at each level.

To compare them we need the probability that price reaches a target before the stop. For a price with no tendency in either direction, a random walk, the classical result is:

Multiply each probability by its payoff and the expectancy is always zero: 50% × 1R − 50% × 1R, or 33.3% × 2R − 66.7% × 1R. Without an edge a farther target is hit less often in exact proportion to its size. That is the baseline against which exit plans should be judged.

Six plans with no edge

Plan Outcomes (probability → result) Trades that end in profit Expectancy Spread of results
All out at +1R 50% → +1R; 50% → −1R 50% 0R 1.00R
All out at +2R 33% → +2R; 67% → −1R 33% 0R 1.41R
Half at +1R, half at +2R, stop unchanged 33% → +1.5R; 17% → 0R; 50% → −1R 33% 0R 1.12R
Half at +1R, stop moved to entry, half at +2R 25% → +1.5R; 25% → +0.5R; 50% → −1R 50% 0R 1.06R
All out at +3R 25% → +3R; 75% → −1R 25% 0R 1.73R
Add equal size at +1R, stop to entry, out at +3R 17% → +5R; 83% → −1R 17% 0R 2.24R

"Spread of results" is the standard deviation of the outcome of one trade.

Share of trades that end with a profit, by exit plan, when there is no edge All out at +1R 50% of trades, expectancy 0R All out at +2R 33% of trades, expectancy 0R Half at +1R, half at +2R, stop stays 33% of trades, expectancy 0R Half at +1R, stop to entry, half at +2R 50% of trades, expectancy 0R All out at +3R 25% of trades, expectancy 0R Add equal size at +1R, stop to entry, out at +3R 17% of trades, expectancy 0R A price with no drift: every plan has the same zero expectancy. Only the share of winning trades changes.
Share of trades that end with a profit, by exit plan, for a price with no drift. Every plan has an expectancy of zero. Exact calculation by Finabase.

The table explains the appeal of scaling out. Compared with holding everything for +2R, the "half off, stop to entry" plan turns a method that wins a third of the time into one that wins half the time, and it narrows the swings from 1.41R to 1.06R. A trader's records look better and the losing streaks are shorter (Losing streaks and drawdown recovery: what a trader should expect). The average is exactly what it was.

That is a real benefit for a person who finds a 33% win rate hard to follow. It should be bought knowingly, because once there is an edge it has a price.

The same plans with an edge

Now give the price a tendency. In the first case price drifts in the trade's direction after entry, as a trend-following entry hopes: the drift is set so that +2R is reached before −1R 40% of the time instead of 33%. In the second case the drift runs against the trade, set so the figure is 28%.

Six exit plans on the same entry and the same 1R stop: expectancy per trade Exit plan Ends above zero No drift Favourable drift Adverse drift All out at +1R 50% 0.00R +0.10R −0.08R All out at +2R 33% 0.00R +0.20R −0.16R Half at +1R, half at +2R, stop stays 33% 0.00R +0.15R −0.12R Half at +1R, stop to entry, half at +2R 50% 0.00R +0.12R −0.10R All out at +3R 25% 0.00R +0.31R −0.23R Add equal size at +1R, stop to entry, out at +3R 17% 0.00R +0.32R −0.23R
Expectancy per trade for six exit plans on one entry and a 1R stop, for a price with no drift, a favourable drift and an adverse drift. "Ends above zero" is for the no-drift case. Exact calculation by Finabase.
Plan No drift Favourable drift Adverse drift
All out at +1R 0R +0.10R −0.08R
All out at +2R 0R +0.20R −0.16R
Half at +1R, half at +2R, stop unchanged 0R +0.15R −0.12R
Half at +1R, stop moved to entry, half at +2R 0R +0.12R −0.10R
All out at +3R 0R +0.31R −0.23R
Add equal size at +1R, stop to entry, out at +3R 0R +0.32R −0.23R

The pattern is consistent.

When the move tends to continue, scaling out costs money. Holding everything to +2R earned +0.20R a trade. Selling half at +1R cut that to +0.15R, and also moving the stop to the entry cut it to +0.12R. The early exit sold part of the position just when the odds had improved.

When the move tends to fail, scaling out saves money. Under the adverse drift the partial plans lost less than holding for the full target.

A partial exit is therefore a smaller bet on continuation. It sits between "all out at +1R" and "all out at +2R" in every column. There is nothing wrong with a smaller bet. The point is that it is a choice about size, with the same trade-off as any other sizing choice, and not a free improvement.

The model is deliberately simple: a steady drift, one stop distance, no costs. A real edge may live in a particular part of the move. Mean-reversion methods often capture most of their profit quickly, and for them early exits can be right (Mean reversion vs trend following: two rules on one market). The honest way to choose an exit plan is to test it on the method's own trades. Kaminski and Lo reach the parallel conclusion for stop rules: under a random walk a stop rule cannot raise expected return, and it adds value only when returns show momentum.

Moving the stop to the entry

The "break-even stop" deserves its own note because it is so widely recommended. Once a trade is 1R ahead, raising the stop to the entry price removes the chance of a loss on the remaining shares. In the no-drift table it also removed half of the +1.5R outcomes and replaced them with +0.5R: the stop is now only 1R below the price and ordinary movement reaches it half the time. It is a tighter stop by another name, with the cost every tight stop has (Stop-loss placement: volatility, structure and time stops).

Scaling in

There are two very different ways to build a position in parts.

Adding to a winner (pyramiding)

The trader starts with a normal position and adds when the trade has moved in its favor, raising the stop on everything so that total risk does not grow. The Turtle rules, for example, added a unit at each further half-ATR move in the trade's direction, up to a maximum.

The last row of both tables is this plan. It has the lowest win rate of all, 17% with no drift, because a trade must first reach +1R and then travel a further 2R without falling back 1R. It has the largest payoff and the widest spread of results. Under favourable drift it earned the most (+0.32R), and under adverse drift it lost about the most. Pyramiding concentrates a trader's results into a few large winners. It suits methods that depend on persistent trends and people who can sit through long strings of small losses.

Adding to a loser (averaging down)

Buying more as the price falls lowers the average cost, and it raises the amount at risk on a trade that is, so far, wrong. If the original stop is kept, the added shares increase the loss when it is hit. If the stop is removed to make room, the trade no longer has a defined risk. Common trading mistakes and the math behind them works through the arithmetic of an averaged-down position.

A long-term investor who buys an index fund at regular intervals whatever the price is doing something else: there is no stop and no thesis to be proven wrong, only a schedule (Lump sum or instalments: every start month since 1993, tested). The distinction is whether the additions were planned in advance for a reason, or made to avoid taking a loss. Shefrin and Statman's classic study of the disposition effect describes the pull to hold and add to losers while selling winners early, the opposite of what a trend requires.

Practical details

  • Costs. Each partial exit is another order. With zero commissions the cost is the spread and any slippage, paid once more per piece (Liquidity and slippage: volume, depth, spread and market impact). On small positions that can outweigh the benefit.
  • Share counts. Half of 15 shares is not a whole number. Small positions cannot be split cleanly, which is one reason scaling suits larger accounts.
  • Orders. A two-target exit is usually set up as two bracket orders, each with its own target and a shared stop level (Bracket, OCO and OTO orders: linking entries, targets and stops). After the first target fills, the remaining stop must cover only the remaining shares.
  • Records. Count a scaled trade as one trade with one result in R, as in the tables above. Counting each piece separately inflates the win rate.
  • Taxes. Each partial sale is a taxable event in a taxable account, and selling at a loss and re-buying within 30 days can trigger the Wash sale rule: the 61-day window, basis adjustment and traps.

How Finabase does it

  • Paper trading. The paper account accepts market and limit orders (Practise with Finabase paper trading on the web), so a position can be bought or sold in parts and each fill appears in the transaction history. It has no stop or bracket orders: the stop side of a plan must be watched or set as a price alert.
  • Backtest. The Backtest tool enters and exits a fixed $10,000 on each signal in one piece. It cannot test partial exits or adds, so it shows the entry's win rate and gain under an all-in, all-out plan only.

The six plans on this page were computed exactly from barrier probabilities by the script named in the sources, not simulated.

Frequently asked questions

Is it better to take partial profits or hold for the full target?

It depends on how the method makes its money. Partial profits raise the win rate and smooth results, and they lower expectancy when price moves tend to continue. If moves tend to reverse quickly, taking part early helps. Only the method's own trade history can say which applies.

Does moving a stop to break-even improve a trade?

It removes the possibility of a loss on the remaining shares and makes it more likely that ordinary movement closes the trade before the target. With no edge the two effects cancel exactly. With a trending edge the break-even stop lowered expectancy in the example here.

What is pyramiding in trading?

Pyramiding is adding to a position that is already profitable, usually while raising the stop so total risk stays about the same. It lowers the win rate, increases the size of the winning trades and makes results depend on a small number of large moves.

Is scaling in the same as averaging down?

No. Scaling in can mean adding to a winning trade with the stop raised, which keeps risk contained, or adding to a losing trade, which increases risk as the trade goes wrong. The second is averaging down.

Summary

Scaling out and scaling in rearrange a trade's possible outcomes. With no edge, every exit plan has zero expectancy: selling half at +1R and moving the stop to the entry lifts the win rate from 33% to 50% and narrows the swings without changing the average. With an edge the choice matters. When moves tend to continue, partial exits and break-even stops give up expectancy, and pyramiding adds to it at the cost of a low win rate and wide swings. When moves tend to reverse, early exits lose less. A partial exit is best understood as a smaller bet on continuation, chosen for the smoother results it gives, and tested on the method's own trades before it is adopted.

Check your understanding

4 questions on this lesson. Each answer comes with the reason, and nothing is scored.

  1. A price has no tendency in either direction. With a stop at −1R, how often does a trade reach a +3R target first, and what is the expectancy?

    1. 33% of the time; expectancy +0.33R
    2. 25% of the time; expectancy 0R
    3. 50% of the time; expectancy +1R
    Show the answer

    B. 25% of the time; expectancy 0R P = stop distance ÷ (target + stop) = 1 ÷ 4 = 25%. Expectancy = 0.25 × 3R − 0.75 × 1R = 0. Without an edge, a farther target is hit less often in exact proportion.

  2. Compared with holding the whole position for +2R, selling half at +1R and moving the stop to the entry does what when there is no edge?

    1. Raises both the win rate and the expectancy
    2. Raises the win rate from 33% to 50% and narrows the swings, with the same zero expectancy
    3. Lowers the win rate and raises the expectancy
    Show the answer

    B. Raises the win rate from 33% to 50% and narrows the swings, with the same zero expectancy The plan changes the shape of the results, not their average. That can make a method easier to follow, but it is not an edge.

  3. A method’s winning trades tend to keep going after they reach +1R. What does scaling out half at +1R do to its expectancy?

    1. Lowers it, because part of the position is sold just when the odds have improved
    2. Raises it, because profit is locked in
    3. Nothing
    Show the answer

    A. Lowers it, because part of the position is sold just when the odds have improved In the favourable-drift case, holding to +2R earned +0.20R per trade and the partial plans +0.15R and +0.12R. A partial exit is a smaller bet on continuation.

  4. You buy 100 shares at $40 with a stop at $38. At $42 you buy 100 more and raise the stop on all 200 shares to $40. If the stop is hit, what is the result in R?

    1. 0R: the stop is at the entry price
    2. −1R: the first lot breaks even and the second loses $200
    3. −2R: both lots lose
    Show the answer

    B. −1R: the first lot breaks even and the second loses $200 1R is 100 × $2 = $200. Stopped at $40, the first lot is flat and the second, bought at $42, loses $200. Raising the stop kept total risk at the original 1R, but the average cost is now $41.

Sources and further reading

  1. William Feller, An Introduction to Probability Theory and Its Applications, vol. 1, 3rd ed. (Wiley, 1968), ch. XIV — probability of reaching one barrier before another
  2. Kathryn M. Kaminski and Andrew W. Lo, “When Do Stop-Loss Rules Stop Losses?,” Journal of Financial Markets 18, 2014
  3. Curtis Faith, The Original Turtle Trading Rules (2003) — adding units to winning positions
  4. Hersh Shefrin and Meir Statman, “The Disposition to Sell Winners Too Early and Ride Losers Too Long,” Journal of Finance 40(3), 1985
  5. Finabase calculation — exact barrier probabilities for a random walk with and without drift (scripts/learn/figures/trading-risk-depth.ts)