Clustered trades are one bet, not five

[Fairman]

Clustered trades are one bet, not five

A scalper takes five trades in ten minutes, all long EURUSD, each from a slightly different level. The journal shows five trades. The market sees one bet.

The outcomes are highly correlated. If the direction is wrong, all five lose. If right, all five win. The effective sample size is closer to one than five.

This matters for statistics. A win rate of 60 percent over 100 trades might consist of 20 independent decisions with five trades each. The confidence interval is much wider than it appears.

It matters for risk too. Five clustered trades carry five times the exposure of one, concentrated on a single idea.

Grouping trades by time window or by signal, and counting each group as one observation, gives a more honest picture. The number of independent bets in a month is often small.

For evaluation, using group-level results shows whether the strategy truly has an edge or whether a few lucky clusters drive the results.

Sizing accordingly means treating a cluster as a single position and judging its total risk, not each leg separately.

A practical fix is to define a cooldown: after a trade is opened on a given pair and direction, no new trades in that direction on that pair for a set number of minutes unless the first has closed. That naturally limits clustering and forces each new trade to be a separate decision.

Another is to report all statistics at the cluster level. Win rate per cluster, average profit per cluster, and the standard deviation across clusters give a much more honest basis for judging whether a strategy is working. The numbers are usually less impressive and far more useful.

Group clustered trades together and count again: how many independent decisions are in the last month?

[Fairman]

Taking several trades at once on correlated pairs or in the same direction on one pair can look like diversification, but it is often one bet repeated. If you buy EUR/USD, GBP/USD and AUD/USD at the same moment because the dollar looks weak, all three trades depend on the same outcome. If the dollar strengthens, all three lose together. The total risk is closer to three times a single trade than to three independent trades. Treating the cluster as one position, with a combined risk limit, keeps the account from taking a much bigger hit than the plan intends.

[Amanhasnoname]

I only understood this after I plotted my own trades by entry time and saw tight little bunches of orders on the same side in a few minutes. In one month I had 112 trades, but when I merged anything opened within eight minutes of each other in the same direction on the same pair, there were only 37 groups. Suddenly my "great" win rate looked far less convincing, because several of the wins came from one lucky push.

The most useful change was to measure the group's worst moment, not each leg. I now add up the open risk across the cluster and set a ceiling of 1.2 times my normal single trade risk, so adding a second or third entry means the earlier stops must already be at break even or better. That turned my scale-in habit into something I could defend.

It also changed how I review. I look at the average result per group and how many consecutive groups lose, which is a much quieter picture than the raw trade list.