Proposal for Improving MQL5 Market Transparency - A Mandatory Trading Account for Expert Advisor Publishers – - page 2

New comment
Ioannis Christopoulos #:

the MT5 tester.... Α backtest cannot be trusted.

I can hard-code exact prices, timestamps and specific symbols, whether Forex pairs or stocks, and make the EA execute trades at precisely those prices and times. Controlled testing,

This allows to reproduce specific market scenarios and test exactly how an EA behaves under controlled conditions. Overfitting.

Controlled testing lead to overfitting.

This is exactly what I believe is missing from the MQL5 Market: something similar to Myfxbook, but with an important difference.

An independently hosted demo account, running the actual Market version of the EA, with no possibility for the developer to modify the account, interfere with its trading history or reset it after a drawdown.

Even if the user changes the EA's parameters, the demo should continue running against future, previously unknown market prices, maintaining a continuous and independently verifiable record.

In my opinion, this is the only meaningful way to verify whether an EA's claimed behaviour holds up beyond its backtest.

This means you don't yet know how to conduct credible backtests. Furthermore, as I mentioned earlier, the most reasonable approach would be to offer products exclusively via a monthly rental model (similar to how signals work) rather than a one-time, purchase for ethernity. Logically, no system will remain profitable forever. Moreover, clients should be able to switch between different EAs whenever they wish, without incurring major costs or being forced to pay for a "lifetime" license. It should function just like the signal service: you choose what you want for a specific duration.
Ghenadie Tumco #:
This means you don't yet know how to conduct credible backtests

I think we are discussing two different things.

I am not questioning the usefulness of credible backtesting. I am questioning whether a backtest, regardless of how well it is conducted, can independently verify that a commercial EA behaves as claimed.

We may not understand each other
Ioannis Christopoulos #:

I think we are discussing two different things.

I am not questioning the usefulness of credible backtesting. I am questioning whether a backtest, regardless of how well it is conducted, can independently verify that a commercial EA behaves as claimed.

We may not understand each other

Possibly, and we can also drop the discussion about backtests.

The only remaining, is live verification methods, as I’ve mentioned, the best solution is a monthly rental at a reasonable price, this allows you to rent whatever you want each month, rather than being tied to an EA forever.

Ghenadie Tumco #:

Possibly, and we can also drop the discussion about backtests.

The only remaining, is live verification methods, as I’ve mentioned, the best solution is a monthly rental at a reasonable price, this allows you to rent whatever you want each month, rather than being tied to an EA forever.

Or a 1 time 1 month free trial of an EA. Id be happy to provide that

12
New comment