How to spot fake EAs from the market place?

[Camilo Mora]

Recently, I watched a video of a person coding an EA that took historical data, offset it by 1 candle and loaded it into an EA that tailor positions based on whether the prior candle went up or down. The result was an EA with amazing performance and very low DD. The point it was trying to make was how easy is to bias the results of the backtesting. Can anyone think of a method by which this type of situation can be identified?

The only I can think of is the use of the live signal, which has its own biases as in most cases it is very recent, and in some other cases authors load several and remove the loser ones over time. The facility by which current EA performance indicators can be manipulated, leaves people vulnerable to scams while rendering useless the work by honorable developers.

While one can buy based on trust on the developer, this will make it superhard for new developers to enter the market.

[Camilo Mora]

While researching this issue of spoting EAs that preload data, I found a similar post talking about EAs that modelled the data, and by loading the model, the EA can also provide good results in the backtesting but very poor on real. I guess it could be similar to overfitting as well, if EA parameters are kept internal.

[Oleksandr Medviediev]

Look for EAs with live forward tests to ensure reliability. Skip the rest.

[Yauheni Korsun]

Камило Мора:

Недавно я посмотрел видео, на котором человек кодировал советник, который брал исторические данные, смещал их на 1 свечу и загружал в советник, который подстраивал позиции на основе того, выросла или упала предыдущая свеча. Результатом стал советник с потрясающей производительностью и очень низким DD. Суть его в том, как легко исказить результаты бэктестинга . Может ли кто-нибудь придумать метод, с помощью которого можно определить такой тип ситуации?

Единственное, что приходит мне на ум, это использование живого сигнала, который имеет свои собственные предубеждения, поскольку в большинстве случаев он очень недавний, а в некоторых других случаях авторы загружают несколько и удаляют неудачные с течением времени. Возможность, с помощью которой можно манипулировать текущими показателями производительности EA, делает людей уязвимыми для мошенничества, одновременно делая бесполезной работу честных разработчиков.

Хотя можно совершать покупки, доверяя разработчику, это крайне усложнит выход на рынок новых разработчиков.

I don't believe that an EA can work for a long time

[Fabio Cavalloni]

The usual rule of the world apply also this.

If a thing look too beautiful to be real, very probably it's not.

Don't forget to check for real account tracking, comparing backtests with pubic signals and so on.

[byantotrade]

I do a lot of backtest of many EAs. Not few of them store historical data/pattern into the EA. So, my methologies are :

1. Look on the EA file size. If the EA file size is more than 500K is a suspect of suspicious EA.

2. Do a back test with a period after latest version release date. Usually, fake EA got bad result after the release date.

3. I created custom symbol and then I downloaded the tick data of target symbol. After that, using my code to add value to the data and I shift the date 2 weeks after. Like as example gold price is 2664 on November 8th, I will change it gold price is 2764 on  November 22. I import the data to the custom symbol and use the custom symbol for back test. All data still keep the pattern in the different time. So, I found fake EA result will not the same with the custom data. I already found some of them.

[Thomas Bradley Butler]

Camilo Mora:

Recently, I watched a video of a person coding an EA that took historical data, offset it by 1 candle and loaded it into an EA that tailor positions based on whether the prior candle went up or down. The result was an EA with amazing performance and very low DD. The point it was trying to make was how easy is to bias the results of the backtesting. Can anyone think of a method by which this type of situation can be identified?

The only I can think of is the use of the live signal, which has its own biases as in most cases it is very recent, and in some other cases authors load several and remove the loser ones over time. The facility by which current EA performance indicators can be manipulated, leaves people vulnerable to scams while rendering useless the work by honorable developers.

While one can buy based on trust on the developer, this will make it superhard for new developers to enter the market.

ones they say $100 to $1 billion, that is how you know

[Camilo Mora]

The suggestion of using simulated data is a good one, and was proposed in a prior post. The problem is that simulated data sort of ruins the legit patterns that the data may have and the EA tries to predict. In statistics the is an old saying that goes like ..."garbage in garbage out". so this solution could be unfair to eas with good predictability potential.

[Camilo Mora]

When one buys a car, one can take it for a ride; when you buy a product on any large store you can return it if it is broken or even if you do no like, and some go as far as to offer money back guarantied for their quality.

Thinking about what good companies offer to their customers, I think Metatrader should allow/force sellers of EAs in their platform to offer free trials for their EAs (may be for a week to a month). The more one read about this, the more it becomes clear that metatrader is leaving customers too vulnerable to scammers in the EA market place.

If you think about this idea of a week or month free trial, it could allow people on the fence to go for it and buy EAs that meet basic quality standards. It will right away  filter all unscrupulousness sellers from the Market place, as one can truly validate the quality of EAs. It will also force developers to sell products of quality, which i think over time will turn into more sales for them and metatrader. It sounds like good business practice, while protecting customers. What you think?

[Sergey Golubev]

Fabio Cavalloni #:

If a thing look too beautiful to be real, very probably it's not.

It was one old thread on the forum (I can not find it now sorry) where OP asked to fix his EA.
Because his EA is too good in backtesting (very profitable) so he asked to fix it.
Some users understood that he was joking ... but he uploaded the source code ... and EA was fixed by them (by the forum users).

[Dz Mak]

Камило Мора:

Недавно я посмотрел видео, на котором человек кодировал советник, который брал исторические данные, смещал их на 1 свечу и загружал в советник, который подстраивал позиции на основе того, выросла или упала предыдущая свеча. Результатом стал советник с потрясающей производительностью и очень низким DD. Суть его в том, как легко исказить результаты бэктестинга . Может ли кто-нибудь придумать метод, с помощью которого можно определить такой тип ситуации?

Единственное, что приходит мне на ум, это использование живого сигнала, который имеет свои собственные предубеждения, поскольку в большинстве случаев он очень недавний, а в некоторых других случаях авторы загружают несколько и удаляют неудачные с течением времени. Возможность, с помощью которой можно манипулировать текущими показателями производительности EA, делает людей уязвимыми для мошенничества, одновременно делая бесполезной работу честных разработчиков.

Хотя можно совершать покупки, доверяя разработчику, это крайне усложнит выход на рынок новых разработчиков.

They are very easy to check