Backtesting Cooldown Length Before You Commit to a Number

The cooldown is the setting people are least able to defend. Ask someone why theirs is sixty minutes and the answer is almost always that it was sixty minutes when they found it. Which would be fine if the parameter did nothing, but it decides whether your profile is allowed to re-enter an instrument it just exited, and re-entries are the part of most momentum strategies where the outcomes are widest in both directions.

It is also one of the few parameters where the tradeoff is genuinely legible. A short cooldown lets the profile chop itself to pieces in a range. A long one saves you from that and eventually starts deleting the second entry that made the whole run work. Somewhere in between is a range rather than a point, and you can find the range in an afternoon.

What the cooldown is actually preventing

Separate it from the other limits before you test it, because it gets confused with them constantly.

A concurrency or exposure limit controls how much you have on at once, across instruments. A cooldown controls how soon you may go back to the same instrument after an exit, and it does nothing about breadth. If your profile takes nine positions in nine different names on one morning, a cooldown of any length will not stop it, and if you thought it would you have been running with less protection than you believed.

What it stops is the loop. A signal fires, price ticks against you, the stop takes you out, price ticks back, the signal fires again, and you pay a full round trip in fees and spread for the privilege of arriving back where you started. In a range that can happen four or five times in a session on the same instrument, and it is the single most reliable way for a technically sound profile to lose money in a market that went nowhere.

So the question the sweep answers is narrow and worth stating precisely. At what interval does blocking re-entry stop being protection and start being an opinion about the market that you did not intend to express.

Running the sweep by hand, because the panel does not do sweeps

The Backtest a Profile panel on Autopilot takes three inputs and returns one result. A profile, a number of days, and a start equity. It replays historical signals against that profile with no real orders placed, and it honors all filters, sizing, leverage caps, and risk guards, which is what makes changing a parameter and re-running a valid comparison.

There is no parameter sweep on the screen. Nothing offers a range, and nothing holds two runs side by side. So a sweep here means six separate runs with a value edited between each, and a spreadsheet you maintain. I also cannot confirm from the panel that a cooldown exists as a named setting on your profile at all, so open the profile editor and check before planning the work. If there is no such control, the decision still belongs to you, and the same six run structure applies to whichever setting on your profile governs re-entry.

Three rules keep the sweep honest. Do every run on the same day, so the day count resolves to the same window. Use the same start equity throughout, and make it your real account size rather than a default. And change only the cooldown between runs, which is harder than it sounds, because the temptation once a value looks promising is to nudge something else at the same time.

The shape of the curve, and where in it to read

Run values that span an order of magnitude rather than clustering around your current setting. On a profile working off four hour bars, something like no cooldown, one bar, four hours, twelve hours, twenty four hours and forty eight hours gives you the whole shape.

Record four columns for each run, because total return alone will send you to the wrong answer.

That table is a sketch of a shape, not a result from any particular profile, and your numbers will differ. The shape is the thing to look for: a steep improvement as you move off zero, a broad flat middle where nothing much changes, and a decline at the long end that arrives alongside a collapse in trade count.

Read the middle of the plateau, never the peak. The single best value in a six point sweep is the one that suited this window, and choosing it is how you end up with a parameter that works beautifully on the six months you tested and unremarkably thereafter. If four, twelve and twenty four hours all produce similar results, the honest conclusion is that anything in that band is fine and the parameter is not very sensitive, which is good news. Take the middle and stop tuning.

If instead there is a sharp peak with worse values on either side, treat that as a warning rather than a discovery. Genuine parameter effects are usually broad. Sharp ones are usually the window.

The second entry that carried the run

Before you commit to the long end, go and look at what the long cooldowns removed, because the trade count column will not tell you.

Take the run at your longest value and the run with no cooldown, put both trade lists in a spreadsheet, and pull out the entries present in the short run and missing from the long one. Those are your blocked re-entries. Sum their results and look at their distribution.

Two patterns show up. Either the blocked set is a mass of small round trips with a slightly negative average, which is the loop the cooldown exists to prevent and confirms the setting is earning its place. Or the blocked set is mostly small losses plus two or three unusually large gains, which is the pattern that matters. A trend that pauses, stops you out, and then resumes is one of the most common ways a good trade is structured, and a long cooldown is exactly the rule that guarantees you miss the resumption. If the two or three large gains dominate the sum, your cooldown is not filtering noise, it is filtering the signal.

The fix in that case is not a shorter cooldown across the board, because that hands the loop back. It is usually to make re-entry conditional rather than timed: allow the profile back in when the original condition re-triggers cleanly rather than when a clock expires. Whether your profile can express that is a question for the profile editor, and if it cannot, the timed cooldown at the short end of the plateau, with the loop cost accepted, is the honest compromise.

When the cooldown is fixing a data problem instead

One case is worth ruling out before you spend an afternoon on any of this. If the no cooldown run shows the same instrument entering several times within a few minutes, you may not have a strategy problem. You may have duplicate or repeated signals arriving from the source, and a cooldown set long enough to suppress them is a workaround that hides the real fault and will keep hiding it.

Check the entry timestamps in the shortest run for near identical pairs on the same instrument. If they are there, fix the feed side before tuning anything, because a cooldown chosen to mask duplicates will be wrong for every other purpose.

Then carry the whole thing into live carefully. The replay places no real orders, and the module says plainly on the same page that with paper mode off it places real orders on your connected exchanges, that execution prices may differ from signal prices because of market conditions and latency, and that you are solely responsible for the trades. Its stated route is paper first, verify the behaviour, then live at small size. Watch one thing specifically during the paper period: whether the profile ever re-enters faster than your cooldown should allow. If it does, the setting is not being applied the way you assumed, and every run in your sweep was measuring something else. The pause and emergency kill switch controls at the top of the page are what you use in the meantime, and a profile that is looping on an instrument at two in the afternoon is precisely the situation they are there for.