Part 3: Size, signals and outcome · Chapter 8 of 14 · 8 min
Size after streaks
Changing the size according to how the last trade went is the most expensive deviation. On the plan (with an edge in the data) doubling after a loss takes capital from 176% to 336% but the drawdown from 5.9% to 17.8% (65.5% with a 5% base risk). On a rule with no edge the same doubling gives a median capital of 60% against 86%, and in 42% of cases the account ends below 50%. Halving the size after every loss reduces the drawdown (3.5%) and the growth (150%).
Three ways of changing size
After a loss, or after a win, the size of the next trade seems something to decide on the spot. Three common behaviours:
- Revenge: after a loss you double, to recover (martingale: book 11, chapter 5).
- Confidence: after two wins you increase, because "I am in form".
- Prudence: after a loss you halve, so as not to get worse.
As with all deviations, the cost is measured on the same trades. The plan, base risk 1% on current capital, in the order in which the trades happened (79):
| Size rule | Final capital | Drawdown |
|---|---|---|
| Fixed size (plan) | 176% | 5.9% |
| Doubling after a loss (maximum 4 times) | 336% | 17.8% |
| Doubling after 2 wins in a row (then back to 1) | 188% | 6.9% |
| Half size after a loss | 150% | 3.5% |
| 1.5 after a win, 0.5 after a loss | 166% | 4.6% |
Read the chart in words
Daily gold chart with, below, three capital curves: the one with doubling much higher and more irregular, the half-size one lower and smoother.
Doubling after a loss gave almost double the capital (336% against 176%), with a drawdown three times larger (17.8% against 5.9%). It seems a good idea. It is not, and the reason is in the next two sets of numbers.
When doubling works
The plan has an edge in the data (+0.74 R, not proven) and the longest losing streak is 6 (with doubling limited to 4 times, the maximum risk is 4%). In this series losses were followed by a win fairly soon, and doubling amplified the gain. Shuffling the order of the 79 trades 5,000 times:
| Size rule | Median capital (5th percentile) | Median drawdown (95th percentile) |
|---|---|---|
| Fixed size | 176% (133%) | 6.0% (10.8%) |
| Doubling after a loss | 320% (177%) | 17.8% (33.1%) |
| Doubling after 2 wins | 193% (134%) | 6.9% (11.6%) |
| Half size after a loss | 151% (121%) | 4.0% (6.9%) |
| 1.5 after a win, 0.5 after a loss | 169% (124%) | 5.5% (9.2%) |
With an edge and a losing streak that does not get long, doubling returns more and risks more: the 95th percentile of the drawdown is 33.1% against 10.8%. With a 5% base risk doubling gives 8,471% and a drawdown of 65.5% (fixed size: 1,237% and 26.8%): the same pattern, with enormous numbers and a step from ruin.
When it does not work
If there is no edge, doubling does not create one. Take the no-edge rule of book 12 (20-candle breakout, 1 ATR stop, 1 R target: 307 trades, −0.02 ± 0.11 R), with a base risk of 2%:
| Size rule | Final capital (real order) | Drawdown |
|---|---|---|
| Fixed size | 86% | 24.6% |
| Doubling after a loss | 53% | 57.1% |
| Doubling after 2 wins | 68% | 44.6% |
| Half size after a loss | 95% | 18.2% |
| 1.5 after a win, 0.5 after a loss | 96% | 25.3% |
And shuffling the order of the 307 trades 4,000 times:
| Size rule | Median capital (5th percentile) | Median drawdown (95th percentile) | Final under 50% |
|---|---|---|---|
| Fixed size | 86% (48%) | 36.5% (58.5%) | 6% |
| Doubling after a loss | 60% (13%) | 68.9% (91.5%) | 42% |
On a rule with no edge doubling takes the median capital from 86% to 60% and makes the account end below 50% in 42% of cases (against 6%). The median drawdown is almost 69%.
The difference between the two cases is what the trader does not know: whether their plan has an edge (and how much), and how long the next streak of losses will be (book 11, chapter 3: from 4 to 8 in a long run). Doubling earns if you are right about the first and loses everything if you are wrong about the second.
Prudence and confidence
Half size after a loss reduces the drawdown (3.5% instead of 5.9%) and reduces growth (150% instead of 176%): it is like having a smaller risk, with its advantages and its costs. On the no-edge rule it loses less (95% against 86%). It is not irrational if it is a rule of the plan; it is if it is decided after the loss, with fear, differently each time.
Confidence after two wins gave results close to the plan (188% against 176%, drawdown 6.9% against 5.9%) on this series, and on the one with no edge it lost (68% against 86%, drawdown 44.6%): it rewards the plan when wins really come in streaks, and punishes it when there is no memory (chapter 4: after a win the winning share is 44%, not more).
What to do with it
- Fix the size in the plan (book 11, chapter 5: fixed fraction, 0.5-2%) and do not change it according to the last result.
- If you want to reduce size after losses, use a written drawdown rule (book 11, chapter 13: half risk at −10%).
- Never double after a loss: it requires an edge and a short streak, and you know neither.
- Record the actual size of every trade and compare it with the plan's: the deviation is the datum.
- Change the size only at the review, in writing.
Procedure in four steps
- Write in the plan the size rule (fixed, or with drawdown steps) and the base risk.
- Record the actual size of every trade.
- Mark as a deviation every size different from the plan's and its result.
- Every month compare the real capital with what you would have had at fixed size.
Exercises
Chapter exercises
1. On the plan, doubling after a loss gives 336% capital against 176%, but with a drawdown of 17.8% against 5.9%. How is it read?
2. On a rule with no edge, doubling after a loss (2% risk) makes the account end below 50% in what share of cases?
3. Half size after a loss takes capital from 176% to 150% and the drawdown from 5.9% to 3.5%. What is it, in substance?
4. After two wins in a row the winning share of the next one is 44%. What does it imply for the rule "doubling after two wins"?
Have a question or want to share your exercise?
Post in the community, or join the free signals room on Telegram.
Educational content, not financial advice. Trading involves risk.