Evgeniy Scherbina / 个人资料

Look at the difference: boring standard bars with fixed intervals vs. dynamic Volume Bars. 📊 While traditional candles close simply because "time is up," volume bars only form when actual trading volume hits a specific target. ⚡️ This completely eliminates market noise and highlights the true activity of institutional players. 💰 This is the cutting-edge approach championed by Marcos López de Prado—the most modern innovation in market analysis. 😎

Meet Volume Reaper. 🦾 It’s a game-changer that replaces artificial timeframes with a purely volume-based structure. 🎯 By removing time distortions, the neural network sees the market's real structure, delivering pinpoint accuracy on trend reversals. 🚀 Our AUC metrics prove that this logic far outperforms any standard timeframe. 📈 To top it off, we're integrating an RL model for intelligent risk and capital management. 🧠 Stop trading time—start trading real volume! 🔥

已发布文章Why the Live Candle EA Is Different from Everything I Have Built So Far

👋 Hello everyone!

The conclusion is becoming clearer: only ensembles seem to work properly in Forex trading. There are two possible ways to build them.

🧩 The first option is a universal ensemble — one powerful system that handles all currency pairs together. The second option is an individual ensemble created separately for each symbol. 🎯

Right now, I am leaning more and more toward the second approach. After all, USDCHF does not behave the same way as GBPUSD. 💱 They both use the US dollar as fuel, but their personalities, movements, and market reactions are completely different. So why should they share exactly the same ensemble? 🤔

At this moment, I have completed 3 tests, which means 18 validations for the universal ensemble (6 major pairs × 3 tests). With the individual ensemble approach, it is only 3 validations.

That is why I decided to add 2 more tests. ⏳ It will take exactly 4 days. Then, on July 13 — or July 14 at the latest — I should be ready to launch the "Live Candle" update with a real signal. 🚀

And this one is going to be absolutely insane! Stay tuned… the next chapter is coming! 🔥

已发布文章Forex Lessons: When the First Month Lies — and the Fifth Month Tells the Truth

👋 Hello! I’ve discovered the natural next evolution of ensemble trading systems — the “Live Candle” Expert Advisor 🚀. This newest-generation EA takes everything powerful about ensemble intelligence and pushes it to the next level. And yes — Live Candle ensembles are simply next-class awesome ⚡. They combine some of the most advanced time-series forecasting architectures used today: N-HiTS (Neural Hierarchical Interpolation for Time Series) 📈, TFT (Temporal Fusion Transformer) 🧠, and of course the reliable classic LSTM (Long Short-Term Memory) 🔥.

Each model brings its own edge to the ensemble. LSTM delivers robust sequential memory and captures long-term market dependencies with proven stability 🏛️. N-HiTS specializes in sharp multi-horizon forecasting, making it exceptionally strong at detecting complex trend structures and market turning points 🎯. TFT adds adaptive attention mechanisms, allowing the ensemble to dynamically focus on the most relevant temporal patterns and hidden interactions across market states 🔍. Together, they create a highly diversified forecasting engine where each model strengthens the others 🤝.

But the real breakthrough — exactly as the name suggests — is the live candle itself 🕯️. Instead of waiting for a candle to close, Live Candle treats the current forming bar as part of the active prediction chain. It can work with any timeframe, from M1 to D1, effectively estimating how a higher timeframe candle is likely to close before it actually does ⏳. Since markets move incredibly fast 🌍⚡, this creates a genuine decision-making advantage in many real trading conditions. Early testing shows that Live Candle can be in a league of its own 🏆.

✨ Coming soon — and built to redefine ensemble trading. 🚀📊

Here’s the new standard: train models across different timeframes and architectures on separate historical periods. Then select only those models—or ensemble meta-parameters—that can survive multiple tests spaced out in time. When you train a single model, no matter how many runs you do, one setup might pass one test but fail the next. If you run six major symbols, typically four will fail the second test, and out of those, two will fail both. The final result after two tests is negative. The new approach is not about overfitting to a single backtest. It’s about tuning ensemble meta-parameters so the system remains stable across different time segments. “Gold Chaser” EA update has just been released. Next up: “Edge Ensemble.”

已发布文章Mastering the Search: Unlocking 40,000+ Ensemble Possibilities 🚀

Hey folks,

Below is the plan for updates of my EAs:

LSTM Ensemble – done!
Gold Chaser – April
Edge Ensemble – April
Latte – May
Your False Hope – May

Stay tuned for the upcoming releases.

And one more exciting news — I'm also working on a brand new EA called "Deep Aurum", which will bring a fresh approach to trading Gold and Bitcoin. More details coming soon! 🚀

Hey folks! 🎉 I'm excited to share the latest update before a massive system overhaul drops!

The last thing I'm testing right now is the application of a metamodel to the ensemble — and the results are looking really promising. 🚀

So here's the idea: the system runs an ensemble of base models, each trained on different timeframes and architectures (LSTM, TCN, Transformer). Once the ensemble is assembled, a metamodel is trained on top of it — its job is to learn the optimal way to combine the predictions of all base models into one final signal. Think of it as a smart coordinator that figures out which models to trust more and when. 🧠

What's really cool is the test results. ⬇️ I deliberately picked a tough 5-month test period where a single model simply struggles to perform. The ensemble alone already manages to grind out a profit where a solo model fails. But the ensemble with the metamodel? It takes things to another level — not only is the total profit significantly higher, but the equity curve is noticeably smoother 📈 — which matters just as much as raw profit.

The two charts below show this clearly — you can see the difference with your own eyes. 👀

So... is this the last piece of the puzzle that could still be improved? 🤔 It's starting to look that way! Stay tuned — big updates are coming! 💥

Uff! Two-headed training — one head for forecasting, one for forecast quality. Uncertainty weighting as a loss function, Z-scores as new features, and a revamped multi-timeframe + multi-model ensemble. I'll be updating almost everything over the coming days. The best part? This new approach works not only on major symbols but on Gold too — it's finally becoming truly universal. If anyone's going to crack ensembles on forex, it's going to be me! AI has been a huge help in exploring all these opportunities, and I've been running tests day and night... 🚀

What's up folks! My "Gold Chaser" EA delivered an impressive +40% this month. The main driver of success was oil, which surged sharply, and the EA correctly caught that trend. Gold short positions also contributed nicely to the overall result. Yes, "Gold Chaser" can simultaneously trade oil, gold, silver, and Bitcoin — all in one system. Bitcoin, however, hasn't been performing as expected just yet. This means the model needs some tuning specifically for Bitcoin's unique behavior. And this is exactly where my newest approach could come in handy — model ensembling. An ensemble for Gold Chaser is currently in development and should be arriving in the near future. Stay tuned!

What's up, folks! Another month is wrapping up, and "LSTM Ensemble" kept delivering — up +10% this month. Right now the EA is making a big bet on JPY strengthening, so let's see how March plays out!

In the coming days I'll be dropping updates for both "LSTM Ensemble" and "Edge Ensemble" — and there's a lot to cover. The headline news is the switch to ONNX-based computations. This makes neural network inference significantly faster and cleaner, and most importantly, adding Transformer models to the ensemble is no longer a headache.

Speaking of which — Transformer models are now joining the party! More diverse architectures mean a more robust ensemble, and that's always a win.

And here's the one you've been waiting for: next month "Edge Ensemble" is finally getting a live real-money signal. I've been building toward this for a while, and it's almost here. Stay tuned — March is going to be a big one!

Hi everyone! In the tests I’m still running, I’m coming to a simple conclusion: the "LSTM Ensemble" outperforms the "Edge Ensemble". In theory, a more complex ensemble architecture should deliver better results — but in practice, that doesn’t seem to be the case. Sometimes complexity adds noise instead of value. Here’s the rule I’m following: if a simple model performs as well as or better than a complex one, choose the simple model. Clarity beats complication. Even more exciting, in live mode the "LSTM Ensemble" looks very convincing. In less than two months, it has delivered nearly 30% growth. That’s the kind of performance that speaks for itself.