What do you check before buying an EA? These are the 7 things I look at - page 2

[Christian Kotschote]

Tasi Churchill Mofor #:
This is the checklist I wish every buyer used. It's also the standard I try to build to as a developer.

Two additions:

8. Does the logic match the risk? Ask the developer how the EA exits losing trades. If the answer is "it waits," that's your grid/martingale warning from point 2. A hard stop loss on every trade means the equity curve can't hide anything.

9. Is the settings page honest? Check whether the developer lists the recommended broker, spread limit, and risk per trade, and whether results were tested at those exact settings. Backtests on the default setfile with ideal spreads tell you little.

On point 3, I'd also look at the same period on a different broker's data. A real edge survives that. A fitted one often doesn't.

I build XAUUSD EAs, so these are things I check on my own products first: fixed risk per trade, daily loss limit, spread and news filters, and live results on Myfxbook. If I can't show it, I don't claim it.

Which of the 7 do you find sellers fail most often? My guess is #5.

Hi Tasi,

thanks, nice to hear it from the developer side.

Your point 8 is a good question to ask. I'd add one thing from my own experience: a hard stop on every trade helps a lot, but it's not a full guarantee. When the market opens with a big gap, the position gets closed far past the stop.
That's how i lost a prop account, the daily limit was 5 % and the day ended at about 7 %. So I'd also look at how many positions can be open at the same time, not only whether each one has a stop.

The same period on a different broker's data is a good test, I hadn't put that on the list.

To your question: for me it's #3. Most backtests I've seen look a lot better than the live signal of the same EA, especially the drawdown. #5 is common too, but #3 is the one I run into almost every time.

Best Regards,

Chris

[Abdullah Uygar Tuna]

Two items I would add: compare execution, not only results, and test the worst day, not only max drawdown. For execution, run the EA at minimum lot on a demo or small live account and log, per trade, requested price versus fill price, spread at entry, and time from OrderSend to the deal. Look at the distribution of slippage and the 95th percentile, not the average, then re-run the tester with a 50-100 ms execution delay and your broker's commission; if the edge disappears, it will not survive live. For risk, find the largest single-day loss on floating equity (not closed balance) in the tester report and check it against your own daily limit, especially if you trade a prop account where equity is monitored continuously. Finally, require the live trade list to be close to the tester over the same dates, with similar trade count and average trade; a big divergence in either is the warning sign.

[Muhammad Andiya Nafhan]

I’d add reproducibility to the checklist.

A useful backtest should identify the EA version, input settings, broker/server, symbol, date range, modelling mode and trading costs. The buyer should be able to run the Market demo with those settings and investigate any differences.

For an EA with several strategies or optional recovery modes, I’d also want separate reports for each configuration. One attractive combined result does not explain which strategy generated the return or how exposure changed after losses.

Clear documentation makes the developer’s claims easier to evaluate.

[Christian Kotschote]

Muhammad Andiya Nafhan #:

I’d add reproducibility to the checklist.

A useful backtest should identify the EA version, input settings, broker/server, symbol, date range, modelling mode and trading costs. The buyer should be able to run the Market demo with those settings and investigate any differences.

For an EA with several strategies or optional recovery modes, I’d also want separate reports for each configuration. One attractive combined result does not explain which strategy generated the return or how exposure changed after losses.

Clear documentation makes the developer’s claims easier to evaluate.

Thanks Muhammad,

the EA version is the part i would underline. Every update is basically a new optimization, so a backtest from version 2 says little about version 5. I'd check the date of the last update and only count the live results from that day on.

The separate reports make a lot of sense too. If a recovery mode is optional, thats exactly where a grid can hide behind one nice combined curve.

Chris :)

[[Deleted]]

On #14, an EA update isn't necessarily a new optimization. MetaQuotes describes updates for bug fixes and new features too. I'd ask what changed: a logging fix and a change to entries or position sizing have different implications for comparing versions. Earlier results still describe the earlier version; they don't automatically validate the new one.

MetaQuotes: https://www.mql5.com/en/articles/385

[Christian Kotschote]

Den Martin #:

On #14, an EA update isn't necessarily a new optimization. MetaQuotes describes updates for bug fixes and new features too. I'd ask what changed: a logging fix and a change to entries or position sizing have different implications for comparing versions. Earlier results still describe the earlier version; they don't automatically validate the new one.

MetaQuotes: https://www.mql5.com/en/articles/385

Good morning Den,

fair correction, "basically a new optimization" was too strong. A logging fix is not the same as a change to the entries or the lot size.

My problem is that the changelog often just says "bug fixes and improvements". Then i cant tell which of the two it was. So I'd ask the seller directly like you say, and until I know, I'd treat the old live results as the history of a slightly different EA, not as proof for the new one.

Best regards

Chris

[Henry Djuhanda]

Good list, Chris. Point 2 is the one I see ignored most often. A few things I'd add, from the side of building EAs as well as buying them:

8. Size of the worst single trade
Not just max drawdown, but how much one losing trade costs in % of the account. If one trade can take 10–15% of the balance, the EA is one bad day away from a prop firm limit no matter how good the curve looks. I'd also check that every trade has a real SL at the broker, and not a "virtual" or very wide emergency one only.

9. How many positions are open at the same time
Five trades of 0.2 lots opened within the same minute in the same direction are really one 1.0 lot trade. A signal can show "small lots" while the total exposure is huge. I look at the trade history for clusters of entries at the same time and price.

10. Swap and holding time
On gold especially, long positions held for several days can pay a lot of swap. A strategy that looks fine in a backtest with old swap values can lose a big part of its edge at today's rates. Swap is usually shown in the signal's history, so it's easy to check.

11. What the risk setting does to the numbers
Many EAs come with low, medium and high risk presets, and the signal is often run at the highest one. I'd scale the signal's drawdown down to the setting I'd actually use, and see if the return is still worth it. If the seller doesn't explain how the presets differ, I'd ask.

On point 7, I agree with the idea but would add one nuance: for a swing system that takes only a few trades per week, 200 trades can take more than a year. In that case I'd rather check whether the live trades match the backtest trade by trade over the same period. If they do, the long backtest history means more than it would otherwise. If they don't, even 200 trades won't help.

Best regards,
Henry

[Ryan L Johnson]

I would never buy a trading EA from anyone without a full live trading ledger obtained directly from the EA seller's broker-dealer. That would require a signed authorization letter from the seller─for you to forward to the broker-dealer. Then do the legwork of verifying the ledger against the stats posted by the seller. Obviously, AI can speed up that legwork. That would put an abrupt end to the BS train. If the seller ghosts you, you immediately know why. Simple due diligence.

[Christian Kotschote]

Hi @Henry Djuhanda , @Ryan L Johnson ,

thanks Henry, thats a really useful addition. Point 9 is the one I'd underline. Five 0.2 lot trades at the same price are one 1.0 lot trade with five tickets, and the signal still looks like "small lots". Point 11 goes in the same direction. If the signal runs on the high preset and I would use the low one, I have to scale the drawdown and the return down before I compare anything.

On point 7 I agree. For a swing EA with a few trades a week, checking the live trades against the backtest for the same weeks says more than waiting for 200 trades.

Ryan, a signed ledger from the broker would be the cleanest proof, no doubt. I just don't think many sellers will sign something like that for an EA that costs a few hundred dollars. A smaller step I'd ask for is the investor (read-only) password of the live account. Then you log in yourself and compare the history with what the seller posts. If they refuse that too, it tells you about the same as ghosting.

Have a good start into the week

Chris :)

[Ryan L Johnson]

Christian Kotschote #:
Ryan, a signed ledger from the broker would be the cleanest proof, no doubt. I just don't think many sellers will sign something like that for an EA that costs a few hundred dollars. A smaller step I'd ask for is the investor (read-only) password of the live account. Then you log in yourself and compare the history with what the seller posts. If they refuse that too, it tells you about the same as ghosting.

What an excellent idea─assuming that remote read connections don't slow down MT5.

And on second thought, an average broker-dealer likely wouldn't want to be bothered with paperwork nor even digital paperwork.

Happy trading.