Hi Vincenzo,

Thanks for the detailed explanation. That gives much better context to your process.

The SQN > 2.5 with 100+ trades as a pre-filter makes sense as an initial statistical layer, and I agree that this already removes a lot of weak candidates before they reach incubation.

At the same time, I think it also raises an interesting question about the role of data itself in the process.

If data is the foundation of everything we do as algo builders, then the depth and quality of that data become critical. Without sufficient historical data, there is no real basis for analysis, no meaningful statistics, and ultimately no reliable strategies.

From that perspective, I tend to look for broader datasets and higher trade counts, because they expose the strategy to more market conditions and provide a more complete picture of its behavior. In my experience, the more data a system has to live through, the more clearly its strengths and weaknesses become visible.

That is also why I personally prefer to work with stronger SQN levels on extended datasets, combined with deeper validation, rather than relying only on shorter samples that may still be influenced by limited conditions.

At the same time, since the incubator has only been running for about two weeks, I think the current results should still be seen as early-stage observation rather than confirmation. The idea itself is strong, and I believe it will become much more meaningful once more forward data is available.

That is also why I was wondering about the comparison side. Since the goal is a 1:1 DOE between original and grid versions, it would be very interesting, especially in a future update, to see both sides presented together for the same strategies over the same period. Showing the original non-grid version and the grid-overlay version under the same conditions would make it much clearer what the grid layer is actually changing in terms of both performance and risk profile.

Regarding SQN, I also use it as a filter during the building phase, but more as a starting point than as a final validation layer. In my own workflow, I usually pay attention to SQN 3.0, and when I want to be stricter I build with an SQN filter around 3.5. What I have noticed is that stronger strategies tend to maintain or even improve their SQN when additional out-of-sample data is introduced beyond the initial build window.

For example, I build on a long historical period and then extend the data horizon forward to introduce several additional years of unseen data. This creates another validation layer where weaker strategies usually degrade, while stronger ones remain stable or improve.

Combined with that, I rely heavily on detailed Monte Carlo analysis. Not just pass or fail, but the actual structure of the simulations. In my post below, I showed both 50-run and 100-run Monte Carlo tests and explained exactly what I look at in the simulation spread, degradation structure, and confidence levels, using screenshots from a proven system that has already been running for over a year:

https://forexsb.com/forum/topic/10059/monte-carlo-in-ea-studio-how-i-use-it-in-real-workflow/

To give some additional context to what I mean in practice, here are a few examples from my own strategies that have been running forward over time.

These are not just backtest results. They are strategies built on large datasets, with high trade counts and strong SQN values, then observed through forward, demo, and live conditions. Most of the systems shown below have been running for more than a year, which makes the behavior much more meaningful than a short early-stage sample.

So the screenshots are meant to illustrate that the workflow is not based only on backtest filtering, but on a combination of deep historical validation, structural Monte Carlo review, and long forward observation. In other words, not just selecting strategies that look good, but selecting strategies that continue to behave well after extended validation and real execution.

One note on EA 284. This one is a variation of an already proven system, EA 628. The internal logic was not changed. The original version, EA 628, opens only buy trades and occasionally reverses, without a take profit. EA 284 is based on the same internal foundation, but I removed the long-only restriction and adjusted the SL and TP layer so it can trade both directions. So although it is a newer variation, it still comes from a validated base logic rather than being a completely different system.

I also explained that broader portfolio and strategy-variation logic in more detail in this post:

https://forexsb.com/forum/topic/10058/why-multiple-profitable-eas-can-still-hurt-your-account/

For me, this is where the combination of stronger initial filtering, deeper structural validation, and extended forward observation really comes together.

Hi Vincenzo,

Thanks for the detailed explanation. I understand your point and the broader structure behind your process.

At the same time, I think there is an important distinction between describing a workflow and actually demonstrating its long-term output.

From my perspective, everything ultimately comes back to data, not just in backtesting, but especially in forward and live conditions over extended periods.

You can also look at it from a different perspective.

Imagine you are choosing a trader to manage your capital. Would you choose someone who has been trading for 2–3 years, or someone who has been consistently profitable for 15–20 years?

Most people would naturally lean towards the trader with more experience, because they have already gone through multiple market conditions and proven themselves over time.

It is a similar idea when building strategies.

A strategy built on a broader dataset has effectively been exposed to more market conditions before it is even tested further. That additional exposure already acts as a first layer of robustness.

Of course, just like with traders, past performance is never a guarantee. Time will always be the final test. But more exposure during the build phase generally provides a stronger foundation.

For that reason, I place a strong emphasis on both data depth and out-of-sample validation.

In my own workflow, I do not rely on a single validation layer. I build on long historical datasets and then extend the data horizon forward to introduce additional years of unseen data. This extended out-of-sample phase is where weaker strategies usually break down, while stronger ones tend to remain stable or even improve.

After that, I apply Monte Carlo analysis to evaluate structural robustness.

I then validate the strategies again in MT4 using extended tick data over a much longer period, typically from 2003 up to the present.

For me, this is a critical step.

Because if I build strategies from around 2010 onward, then testing them on data going back to 2003 effectively creates an additional out-of-sample layer of several years. It allows me to see how the strategy behaves in conditions it has never “seen” during the build process.

In particular, I want to know if a strategy can remain stable through extreme market environments, including periods like the 2008 financial crisis. That kind of stress testing gives a completely different level of confidence compared to shorter or more recent samples.

The idea is simple: the more market history a strategy has to survive, the more meaningful the validation becomes.

That is also why I see data as the true foundation of everything in this process.

Without sufficient historical depth, there is no real basis for analysis, no meaningful validation, and no reliable way to assess robustness. Indicators, rules, or parameter optimization are secondary, because without enough data they can easily lead to overfitting.

And in my view, overfitting is one of the biggest risks in algorithmic trading.

The only way to reduce that risk is by exposing the strategy to more data, more out-of-sample conditions, and more independent validation layers.

That is why I put strong emphasis not only on initial out-of-sample settings during the build process, but especially on extended out-of-sample testing afterwards, combined with additional validation in MT4.

That combination is what gives me much more confidence that a strategy is not just fitting a limited sample, but is structurally stronger across different market regimes.

Because in the end, what matters is not whether a strategy looks good early, but whether it can remain stable and profitable over long periods of time.

That is also why I tend to look less at early-stage experiments and more at what strategies actually do over 6 to 12 months or longer in real conditions. That is where the real validation happens.

To give a bit more context to what I mean in practice, I apply the same principles not only in EA Studio, but also in StrategyQuant.

The screenshots below are not short-term backtests, but examples of strategies that have been through broader data exposure, structural validation, and forward observation.

They are simply meant to illustrate that the workflow is not only theoretical, but also reflected in actual behavior over time.

So while I understand the idea behind highlighting early behavior, I think the strongest conclusions usually come later, when there is enough data to truly confirm stability and robustness over time.

Regarding the call, I appreciate the offer, but I prefer to keep the discussion here on the forum so everything remains transparent and useful for others as well.

I think that also aligns better with the purpose of these discussions.

Looking forward to seeing how your results develop over a longer period. That will make the comparison much more meaningful.

https://www.myfxbook.com/members/decste … n/11491807

***

Hello, good afternoon everyone. For some time now, I've been following the discussions here. Many of you really have great points, but I would like to highlight some areas where I have profitable experience executing this type of strategy, and I can prove it. I have audited accounts, I provide signals, and I would really like to share my method with you here.

First point: the correct way to generate the strategy is, initially, without money management. No strategy should be created with the grid active during the backtesting period. The grid needs to be static and non-optimizable, and it should be adapted to the strategy later. I will give you an example of the technique I currently use.

First, I take the last 5 years of data. I train it using EA Studio from our dear Popov so it can find all strategies with a **300-point take profit** and a **5000-point stop loss**.

I will select only the strategies that had absolutely no losses during this period—meaning they didn't draw down more than 25% of the account and didn't hit any stop loss. Because, at the end of the day, if the strategy draws down 5000 points, it means it's not worthy of a cost-averaging (grid) strategy, as it will blow my account. I do all the normal steps everyone here is used to doing, using active robustness tools.

After that, once the strategy is ready, I add the grid option to it. I do this directly in the code, sometimes even using a good coding LLM.

It works like this: first of all, so you can continue optimizing the bot using **"Open Prices Only" (Bar Open)**—just like we already do in EA Studio—you need to set the grid system to work solely on the bar open as well. Forget **"Every tick" (OnTick)** once and for all. It's simply not functional; we would waste massive amounts of computing power just to optimize this bot in MT5 on every tick when there's absolutely no need for it. Just give up on OnTick and force the reprocessing of all orders, including the initial grid entries, exclusively on the bar open—exactly how the bot comes out of the box when we export it from EA Studio.

After that, it's very simple. With the cost-averaging already implemented the way I explained, and having created the strategy under the condition of a 300-point take profit and a 5000-point stop loss, all you need to do is add **16 averaging levels spaced 300 points apart**. Then, add an average price reset (take profit target) 100 points above the standard average price. And there you go, your consistent grid EA is ready.

When you are going to run an optimization, turn off the cost-averaging. If you want to optimize using the MT5 backtester, it is crucial that the EA is NOT optimized with the averaging option turned on. The averaging should be added later, and *only* later.

I don't know if everyone was able to grasp the context here, but basically, this is the method I use, and this is what has been working for me. I have 11 champion bots that I use on my own accounts, and I also sell this exact software, which was largely made using EA Studio. I did something even more unbelievable: I used a portfolio strategy. I created 1,000 strategies that didn't draw down 5000 points in the last 5 years, and I added a grid to all of them. This is how I've been doing it. This is how I've been making money. I feel this might help some of you who are still a bit lost on this subject.

I have been working in the financial market for about 10 years, and I've been making a living off it for 7. So, I believe that if you listen to me and apply it exactly the way I just explained, you will achieve, with 100% certainty, profitability with grid bots. I'll tell you upfront that the returns aren't astronomical, but it is truly a game-changer for this technique.

I agree with you, and you raised a great point regarding that. The thing is, as I always like to tell my friends, you have two choices to make. Either you make money with a system and a well-defined strategy that only uses the grid as risk management—and not as the strategy itself—or you stay in a back-and-forth cycle of win-lose, win-lose with strategies that have absolutely no risk management.

There are other types of management out there. For example, I have a system that applies a Martingale to the next trade. It was also created using EA Studio. When I use this trading system, as soon as it loses a trade, it tries to win on the next one. If it loses the next one, it doubles the lot size and tries to win the next. It just keeps applying the Martingale to the subsequent trade. So, with a 1:1 risk-to-reward ratio—for example, a 200-point Take Profit and a 200-point Stop Loss—it works incredibly well too.

So the question I leave you with is this: Isn't our ultimate goal profit? Isn't it to put money in our pockets? That is exactly what I have been doing. That's why I wanted to share this method with you, because we have to make a serious choice: we either use a method that, over the long term, proves to be unfunctional, or we use a method where a crash might eventually happen, but by the time that loss comes, we have already made back our initial risk. Make sense?

Today, I see the financial market as a place where dreams can be realized, you know? I'm not saying this as some fake "trading guru" talk or anything like that. I'm saying it because it's real.

So, what am I trying to say? I don't know if I managed to convey this properly through my previous texts, but the point is that I managed to find my place in the sun using the grid system—not as the core strategy, but as risk management. As I mentioned above, it is crucial that the strategy is trained without the grid, so it can be added later. If the strategy is profitable without the grid and passes all robustness tests without the grid, then we apply the grid to it.

For what purpose? To recover drawdowns and to reduce the strategy's stagnation periods precisely when it stops working well in certain market conditions. Therefore, it functions as a tool, not as the strategy itself. You see countless people blowing their accounts all the time with grids because they use the grid as the actual strategy. They don't focus on training their strategy and putting it through robustness tests without the grid first, to then use the grid merely as a tool if the base strategy proves profitable. Do you get what I mean?

That is exactly my point. The grid needs to be static. The strategy training must happen without the grid activated, and it should only be applied afterward. The best way I found to do this is what I explained above.

I don't know if I was clear enough in my answers, but well, that's what I wanted to say. I am completely open to discussions, including ways to improve my grid system if you guys have any ideas. I update it every single month. Every month, I retrain the strategies, cycle them out of the portfolio, put new ones in, and then add the grid to them. So, while this account has been running solid for 3 years, the reality is that it has been updated every single month since its creation.

Yes, I understand, and I am not going to drag out this discussion. That is your perspective, and I respect it. You manage to make money and have found a method to do so without using risk management for potential recoveries. Unfortunately, I haven't been able to do that.

But the point is, I believe both you and I are good traders. I believe we are good strategy creators, and we've both made money from the market. At the end of the day, that is what matters. So yes, I respect your point. It was a healthy discussion, and thank you for your feedback.