A random entry benchmark for testing your exits

[Fairman]

A random entry benchmark for testing your exits

Here's a test that's humbling and useful. Take your strategy's exit rules, stop, target, trailing, time exit, and pair them with random entries instead of your real entry signals. Run it many times with different random seeds. Then compare the results with your actual strategy.

If your real strategy does much better than the random entries with the same exits, your entries are adding value. If the results are similar, your entry signal might not be doing much, and most of the result comes from how you manage trades, or from the market's drift over the test period.

It also tells you something about the exits. If random entries with your exits are roughly break even before costs, the exits are neutral. If they lose heavily, the exits might be poorly suited to the market, and even good entries would struggle.

The random entries should match your real strategy's frequency and direction mix roughly. If your strategy takes 200 trades over the test, generate 200 random entries over the same period, with a similar long and short balance.

It's not a perfect test, since randomness has its own noise, but running it a few hundred times gives a distribution to compare against.

It's one of the cheapest ways to find out whether your entries actually help.

[Fairman]

A variation is to keep your entries and randomise the exits, for example by using random holding times. That tells you whether your exits add value on top of your entries. Doing both tests separately helps you understand which part of the strategy carries the edge. Sometimes people find that their careful entry rules add little and the exits do all the work, or the other way around. That's useful for deciding where to spend time improving the strategy. If neither part beats random, that's a strong hint the strategy's results came from the test period rather than any real edge, and it may need rethinking.

[Fairman]

In Python this is quick with numpy. Pick random bar indices within the trading hours your strategy uses, assign random directions with the right ratio, apply the exit logic, and record the results. Repeat with different seeds and collect the distribution of total R. Then see where your actual result falls. If it's in the top few percent of the random distribution, that's a decent sign the entries matter. If it's near the middle, the entries aren't doing much on their own. A few hundred runs takes seconds in numpy, so there's no real reason to skip it. The histogram of results is also a good picture to keep with the strategy notes.

[FXS Prop Desk]

The random entry test is humbling for exactly the right reason. If random entries with your exits produce results close to your real strategy, most of the edge is in the exits, or there is no edge at all.

Running a few thousand seeds and plotting the distribution of results also gives a natural benchmark. If the real strategy sits in the top 5 percent of that distribution, the entries are probably adding something. If it sits in the middle, they are not.

[Fairman]

The random benchmark can also show how much the overall market direction affected results. If the test period had a strong trend, random entries in the trend direction might do well just because of the trend. Running separate random tests for long only and short only, and comparing with your strategy's long and short results, separates the effect of the trend from the effect of your entries. If your longs only beat random longs by a small margin in a strong uptrend, the entry signal isn't adding much. It's one of the clearest ways to see whether the strategy has an edge beyond the drift.

[Shadow Trader]

A trade goes against the position by two pips. The price is now below the entry. The mind fixes on the entry level and wants to get back to it. A bounce of one pip feels like relief. Closing at breakeven feels like victory.

Anchoring ties decisions to the entry price, which the market doesn't care about. The relevant question is where price is likely to go next, not where the position started.

On short timeframes, anchoring shows up in two forms. Holding a loser until it returns to breakeven, turning a small loss into a larger one. And exiting a winner just above entry out of fear of losing the gain.

A useful mental exercise is to imagine having no position. At this moment, with the same chart, would a new trade be taken in this direction? If not, the position shouldn't be held.

It's a cheap way to strip the anchor away, and it works faster than any rule.

When was the entry price the reason a loser got held far too long?

[Fairman]

How many random runs are enough? A few hundred usually gives a stable distribution, and more than a thousand rarely changes the picture. The key number is where your real result falls in the distribution. If it's better than ninety five percent of random runs, the entries probably add value. If it's in the middle, they probably don't. I'd also look at the shape of the distribution. A very wide spread means the exits are doing very different things depending on luck, which is a sign of a high variance strategy regardless of entries.