forex-scalping.com

Use a MAE Table to Choose a Stop

A trader records the maximum adverse excursion, in pips, of ten winning trades: 4, 6, 5, 9, 7, 5, 12, 6, 8, 5. The current stop is 20 pips. Is it the right distance?

Sort the values: 4, 5, 5, 5, 6, 6, 7, 8, 9, 12. The largest dip against any winner was 12 pips. The median is 6. Nine of ten dipped 9 pips or less.

What does that say about the distance between 12 and 20 pips? No winner in the sample used it. Anything in that band was occupied only by trades that went on to lose, or that would have lost. A stop at 14 pips, with a small margin over the worst winner, would have kept all ten winners and cut the average loss from 20 pips to 14, a 30 percent reduction.

The payoff: with the same dollar risk, a 14-pip stop allows a position about 43 percent larger, since 20 divided by 14 is 1.43. Or it allows the same size with 30 percent less risk.

Now the warnings. Ten trades is tiny. The next winner might dip 16 pips. The data also covers only winners, and losers reveal whether a tighter stop would have changed them, which it would only by exiting sooner.

Moreover, a stop set exactly at the edge of the sample invites overfitting. A cleaner approach is to choose a level tied to structure and volatility, then check that it lies near what the data suggests.

Exercise: collect MAE for thirty winners before adjusting a stop.

Nice and simple. One addition: run the same table on the losers and look at their MFE. If many losers were in profit by 5 or 6 pips before reversing, a tighter stop alone will not fix them, but a time-based exit or a partial might.

Also, ten winners is a small sample. With 12 as the worst case, the next 50 winners may contain a few at 14 or 15. We usually take the 90th or 95th percentile of winners' MAE plus a small buffer, and recheck every 50 trades.

The analysis is a good illustration of the MAE method. All ten winners dipped 12 pips or less against the entry, so a stop at 14 pips would have kept every one of them while cutting the loss on losing trades by 30 percent. With the same dollar risk, the position could be 43 percent larger, which would scale up the winners too. That's a substantial improvement from a simple change. But the warnings that follow are equally important, since ten trades is a small sample and the next winner might well need more room than any of these did.

I built a MAE table from my own journal about a year ago, and the first surprise was how different winners and losers looked. For my breakout trades, the median adverse move on winners was only 3 pips, but the stop I had been using was 15 for no reason beyond habit. Nobody had ever told me that the upper part of that range was empty space.

Following the suggestions above, I took the ninetieth percentile of the winners and added two pips, which gave me 9. I also looked at the MFE of the losers, as mentioned, and found that about a third had been four or more pips in profit before turning. That led me to add a break even move after 4 pips, which fixed more than tightening the stop would have.

I agree the sample matters. I redo the table every 40 trades and only change anything if the number moves by more than a couple of pips.

The point about losers is subtle. A tighter stop doesn't change whether a losing trade loses; it just makes it lose less. But it might also convert some winners into losers, if a future winner dips more than 14 pips before recovering. The MAE table of winners shows how much room winners needed in the past. To estimate how often a tighter stop would cost a winner, you need enough winners to see the tail of the MAE distribution. With thirty or more, as the exercise suggests, the 90th or 95th percentile of MAE becomes a more reliable guide than the single largest value.

Combining MAE data with structure, as the post recommends, guards against overfitting. If the 14 pip stop lands in the middle of nowhere on the chart, it might be hit by a normal swing that the sample didn't happen to include. If a 15 or 16 pip stop sits just beyond a recent swing high or low, it has a structural reason as well as a statistical one. Choosing a level that satisfies both is more robust than relying on either alone. Volatility matters too, since the MAE of winners in a calm market might be smaller than in a volatile one.

Collecting MAE for thirty winners before adjusting a stop is good discipline. Many platforms don't record MAE directly, but it can be calculated from the lowest price during a long trade or the highest during a short, relative to the entry. Some journal tools calculate it automatically from trade data. Has anyone here used MAE data to tighten or widen a stop? I'd be interested in how much the stop changed and whether the results improved over the following months, since live results are the real test of any adjustment.