Combining Setups With a Filter: Turning a Coin-Flip Setup Into an Edge

Part of the Setups That Work series.

Most setups aren’t the problem. The lack of a filter is.

A setup with a 50% win rate and even risk-to-reward isn’t a losing setup, it’s an unfiltered one. Somewhere inside that data is a subset of conditions where it wins a lot more than half the time, sitting right next to a subset where it barely wins at all, and averaging them together hides both. A filter is how you separate the two.

This post closes out the first stretch of the Setups That Work library by showing how to actually build and test one.

What a filter is (and isn’t)

A filter is a single added condition that has to be true before you’re allowed to take a setup, a piece of context that tells you whether today’s version of the pattern is more likely to belong to the winning subset or the losing one.

A filter is not a second setup layered on top of the first, and it’s not a vague feeling like “the chart looks strong.” It’s a specific, checkable condition: a trend direction on a higher timeframe, a volatility threshold, a time-of-day window, a news blackout. If you can’t check it in a few seconds before you enter, it’s not a usable filter, it’s just a new source of hesitation.

Where filters actually come from

The best filters come from your own data, not from a hunch. Pull your last 30 to 50 occurrences of a setup, traded or just logged in your journal, and look for what was different between the winners and the losers. Was the winning group mostly trending days? Mostly the first hour? Mostly low-volatility mornings versus high-volatility afternoons?

That comparison usually surfaces one or two candidate filters worth testing formally, instead of guessing at filters that sound reasonable but have no connection to how your specific setup actually behaves.

Testing a filter properly

Once you have a candidate filter, split your existing sample into two groups: trades where the filter condition was true, and trades where it wasn’t. Compare the win rate and average result of each group directly.

A real filter shows a clear, meaningful gap between the two groups, not a marginal difference you could explain away as noise. If the filtered group wins 68% of the time and the unfiltered group wins 45%, that’s worth adopting. If it’s 53% versus 49%, you’ve probably found statistical noise, not an edge, and adding it as a hard rule will just mean missing trades for no real benefit.

This is also where sample size matters. A filter that looks great across eight trades is not evidence. Give it enough occurrences, ideally 30 or more per group, before trusting the comparison.

Stacking filters without breaking the setup

It’s tempting, once one filter helps, to keep adding more: a trend filter, plus a volatility filter, plus a session filter, plus a news filter. Each one sounds reasonable on its own. Stacked together, they often shrink your sample of qualifying trades down to almost nothing, which means you can no longer tell whether the setup has an edge or you’ve simply carved out a handful of lucky historical trades and called it a strategy.

One well-tested filter that clearly separates winners from losers is worth more than four filters that each sound smart but were never tested together. If you want to add a second filter, retest the combined rule set from scratch rather than assuming the individual results just stack.

The seasoning analogy

Think of a base setup like a plain, well-cooked piece of chicken, solid, functional, not exciting. One good filter is like adding the right seasoning: it doesn’t change what the dish fundamentally is, it just brings out what was already there and makes the result noticeably better. But dumping in five different seasonings at once, each chosen because it sounded good on its own, usually doesn’t compound into something five times better. It usually just ruins the dish. Filters work the same way: a couple of well-chosen ones enhance the setup you already have. A pile of untested ones bury it.

Putting it together

The process, every time: identify a candidate filter from your own trade data rather than a guess, split your historical sample by whether the filter condition was true, compare win rate and average result across a large enough sample to trust, and only adopt the filter if the gap is real and meaningful. Add filters one at a time, and retest whenever you combine them, instead of assuming each one’s benefit simply adds up.

This is also the natural point where this series connects back to Strategy Blueprint. A setup gives you the trade idea, but the discipline to backtest and filter it properly is what turns “trading setups that work” into a repeatable process, rather than another rotation of ideas that stop working the moment the market shifts.

Takeaway: A mediocre win rate is usually a mixed sample, not a bad setup. Find the one filter your own data actually supports, test it against a large enough sample to trust, and resist the urge to stack filters you haven’t proven work together.

This closes out the first stretch of the Setups That Work library. Go back to the hub to pick your next setup to test.