Remove the best three trades from a 60 trade month: what is left, and what it says

[Fairman]

Remove the best three trades from a 60 trade month: what is left, and what it says

A journal for one month, 60 trades, total +$1,260. That is $21 a trade, which looks good.

The three best trades were +$320, +$290 and +$240, together $850. Without them the month is $410 over 57 trades, or $7.19 a trade.

The three worst were -$150, -$140 and -$120, together -$410. Without them the month is +$1,670.

Without both groups: 1,260 - 850 + 410 = $820 over 54 trades, $15.19 a trade.

So 67% of the month's profit came from three trades, and the same three trades missing would have left a result about a third of the size. That alone says the average does not describe a typical trade.

In a spreadsheet: =SUM(LARGE(range,{1,2,3})) for the best three and =SUM(SMALL(range,{1,2,3})) for the worst three. Then subtract. It takes a minute.

The numbers are made up for the example, but the exercise works on any journal. It is quick enough to do at the end of every month, and the month where the top three are nearly all of the profit is the one to look at hardest.

[FXS HongKong Scalper]

Very instructive. When most of the profit comes from a few trades, the strategy depends on catching them. Missing a few, due to a day off or a platform issue, can change the month dramatically. Checking this every month shows how robust the results are.

[Fairman]

How to read the result depends on why the best trades were big. If all three came from the same setup, tagged in the journal, and that setup occurs about once a week, a month with three of them is normal and not luck. If they came from news spikes that no rule predicted, the month depended on events. Look at the tags and the context of each, and mark whether the entry was by plan. A big win from a trade that broke the plan counts as a plan failure that happened to pay, and it tends to teach the wrong thing if it is simply averaged in with the rest.

[Fairman]

Try the same exercise on a longer window. For 6 months and 360 trades, remove the best 1%, which is 3 or 4 trades, and see what is left. For a trend following style with a payoff of 3 and a 35% win rate, the top handful of trades may well be most of the profit, and that is a feature of the style: the plan depends on rare large moves, and the losses in between are the price. For a mean reversion style with a payoff of 0.8, if the top trades matter that much, something is off. The same figure means different things for different styles, so compare with the style's own expectation.

[Shadow Trader]

Picture a pension fund with a pile of US equities and a pile of European bonds. At the end of the month, the equities have rallied hard while the bonds haven't. The fund's model says the dollar exposure is now too big, so it has to sell dollars, and it has to do it before the benchmark closes.

That kind of rebalancing isn't a view on anything. It's a rule. And because many funds use similar rules and similar benchmarks, their selling lands in the same few hours on the same few days.

What this does to a scalping chart is subtle. Price leans in one direction late in the month, often during the London afternoon, with pullbacks that fail more often than usual. Breakout signals in the opposite direction look fine on paper and then stall.

The flows are large enough that retail-size trades barely matter, so there's no use fighting them. Trades with the flow tend to work a bit better, and trades against it tend to hit stops a bit more often.

It isn't predictable down to the pip. The direction depends on how stocks and bonds performed during the month, and that can be estimated roughly by checking the month's equity move versus bond or index returns.

Over a handful of month-ends, the lean usually becomes visible in the logs.

Has anyone tracked the last three trading days of each month against their usual results?

[Fairman]

A bootstrap gives a sense of how fragile a month is. Take the 60 results, draw 60 at random with replacement, add them up, and repeat a thousand times. If a good share of the totals fall below zero, the month's profit does not rest on much. In Python it is three lines: import random, then totals = [sum(random.choices(results, k=60)) for _ in range(1000)], then print(sum(t < 0 for t in totals) / 1000). With results like +$320 at the top and -$150 at the bottom the share below zero will be modest, but compare it with a month where the profit was spread evenly and see how much larger it is.

[Fairman]

Costs matter in the same way for big and small trades. A $7 commission is 2% of a $320 winner and 6% of a $120 one, but it is the same $7 on a $15 scalp, which is 47%. If the month's profit comes from three big trades and the other 57 are close to zero before costs, the average trade is paying for itself only through the three. Splitting the journal by size band, under $20, $20 to $100, over $100, shows where the costs fall and which band covers them. In this example the bottom band probably carries most of the commission without contributing much profit.