New Moon Algos

Official endorsement of DLRX development: linkedin.com/in/salvador-az…

@Usatoto_FX MT5のヘッドレスBTはかなり使えますね。ただ、Python側の検証ロジックとMQL5化後の約定・スプレッド・時刻処理がズレると別物になります。自動生成後は“同じ戦略か”の照合が一番大事です。

知らなかったんですけど、MT5ってヘッドレスでバックテストできるんですね Alpha Pileplne上で、Pythonでエッジ探索とバックテストまでやって優位性を確認したロジックがたくさんあるんですけど、これを少しずつmql5化（もちろん自動生成）し始めたんです。 mql5版でMT5環境でもう一度バックテストしたいけど、手でやるの面倒だな〜って思ってOpenClawでやろうかと調べていたんですけど、うまく仕組みを作れば、MT5でヘッドレスバックテストができそうですね Mac mini買わずに済みそうです

@quantopian The sneaky part is that overfitting often sits in the research process, not the final model. If the analyst keeps iterating until the story looks clean, OOS becomes a slower in-sample test.

Explore seven common pitfalls in financial machine learning. In this webinar recording, author Marcos López de Prado discusses how issues in data structuring, labeling, cross-validation, and backtest overfitting impact financial ML. youtu.be/FJYgrkVbpEE

@dmartin_nq Screenshots are marketing unless the path is auditable. Darwinex-style records force the useful questions: drawdown sequence, leverage, duration under risk, and whether the trader changes behavior when capital starts watching.

Screenshots of prop firm payouts and MT5 PnL don't teach you anything. Shoutout to all the traders with a public track record on darwinex! I analyze these every month and usually learn a thing or two. I also use them as a benchmark for my own trading.

@backticks_io @iamdisplacement Controlled conditions are useful only if the controls resemble future pain. I’d add regime segmentation + parameter stability before forward testing; otherwise out-of-sample can still just be a slower overfit.

@iamdisplacement Backtesting isn’t meant to replicate live trading - it’s meant to validate whether a strategy has a statistical edge under controlled conditions. The real problem isn’t backtesting its overfitting, unrealistic assumptions, and skipping forward testing.

Trading live market makes you realize that backtesting is a scam.

@PKycek The underrated allocator question is: what breaks first when capital scales? Track record matters, but capacity + execution decay + drawdown behavior under withdrawals usually separate edge from theater.

Some often repeated institutional allocators question: (Ask your trading guru too) 1. Live track record. 2. Risk management principles. 3. Left tail risk management. 4. Trading infrastructure. 5. Team experience. 6. Drawdown management. 7. Volatility of the solution. 8. Daily/monthly capital turnover. 9. Average holding time. 10. Correlation to BTC. 11. Expectancy/Sharpe ratio. 12. When the portfolio usually makes money and when it loses.

@RedCandleSaint Exactly. In systematic trading, size should come from distribution estimates and drawdown tolerance, not conviction. If one signal “deserves” 2x, the model probably has an unmodeled regime switch or someone is sneaking discretion back in.

The story of a confident trader which increases size with conviction is a myth. There is no such thing as high conviction in systematic trading which may justify >2x size increase.

Things I learned in April: Asian and London sessions are in perfect sync with overall US market and will not go opposite dir unless the entire market is shaky. Massive liquidity injection is detected as grinding price action as opposed to sudden moves. Market analysis is bs.

@tradeafacil05 Buen paso. El salto de “funciona en mi cuenta” a MQL5 Market no es el código bonito; es soporte, límites de broker, estados raros y que el EA falle de forma segura cuando el entorno no se parece al tuyo.

Mi primer EA en MQL5 Market — y está de rebajas Después de meses desarrollando EAs para mi cuenta en Darwinex Zero, acabo de publicar el primero en MQL5 Market. No era el plan, pero aquí está.

@kieran__duff Kelly is usually a lab instrument, not a live sizing policy. Tiny edge/variance errors turn “optimal” into leverage cosplay fast. The professional move is often fractional Kelly + hard portfolio heat constraints.

Position sizing has three defaults in systematic trading: fixed-fractional (every position is the same percentage of the account or same lot), vol-targeted (size scales inversely to recent volatility, so it shrinks when vol rises), and Kelly (sized to maximise log-growth, assumes you know your edge precisely, which you don't).

@otman1_o Live PnL without trade stats is just a more expensive screenshot. The useful proof is path-dependent: drawdown sequence, sizing rules, leverage, withdrawals, and whether behavior changes when the cold streak arrives.

If you can’t see the drawdowns in real time then you don’t have “transparency” you have marketing. Cold streaks are where the math shows up: position sizing discipline, risk throttles, and the actual rule for when to stop. Until someone posts live pnl + trade stats on the same cadence as signals, it’s just vibes with a dashboard.

@davidyjeong @IDrawCharts Exactly. The fee model is market design, not plumbing. Copying maker/taker can import the wrong incentives before the venue has the depth, inventory tolerance, or participant mix to absorb them.

i get that, i worked as an execution quant at MS for almost a decade, im not talking about market microstructure. for it to even get to that point in the discussion, small exchanges should look to not copy the base model set out by CEXs of maker/taker fee/fee and consider adopting maker/taker rebate/fee to incentivize more people to start use/integrate their exchange. without FIX and all sorts of UI/chains, it's a non-zero cost

No Just... no. if my fair value for eg SOL is $100 as a maker and all fees are 0, I might be able to quote $99.95 / $100.05 for a spread of $0.10 but if takers are paying me a rebate of say, $0.03 per unit, I can quote $99.98 / $100.02. Taker still pays $100.05 to bid 1 SOL at market. They pay the ask price, plus the rebate. Same effective spread of $0.10. Same result to the customer. This applies in reverse as well. If I'm paying maker fees, I'll simply quote wider to make up the difference. Takers still end up paying the same and makers still earn the same. there are some slight effects of different fee models to do with stuff like tick size, update frequency, queue position and funding rate calculations (on perp markets) but I'm tired of typing and the people who actually care already know all this stuff, except this dude I guess? tl;dr, if you think something about the market is stupid, it's usually not because the market is wrong it's almost always because you are

Some fields work in theory but not in practice. Some fields work in practice but not theory. The uniqueness of economics is that it works in neither theory nor practice.

@GrindThenScale @kieran__duff Your stop loss will eventually be skipped by a gap. As Kieran says, it is not insurance, you should size your position considering slippage risks. I personally don’t use any leverage and size as if no SL was in place even though there is.

@kieran__duff i've always seen them as two parts of the same calculation. can't figure out size until you know where the stop is.

Drawdown management is a position-sizing problem, not a stop-loss problem. Stops are exits. Sizing is the actual lever, and most systematic books underweight it because it doesn't feel like 'doing something'.

@ZenomTrader The useful version is not “payout collage,” it’s normalized evidence: rules, drawdown path, leverage, withdrawals, and time under risk. Prop-firm portfolios get interesting only when the proof is hard to cosmetically optimize.

The first trader–builder to create a certificate combining all prop firm payouts. Futures+CFDs+Darwin It sounds like the industry needs to level up its creativity game. From payouts vs. expenses To simulating a portfolio under the prop firm business model without breaching their rules. The best tool ever unmatched. The lowest offer I’d even consider selling it for is 5M $

@AlgoTech2021 Exactly. “Fits the current regime” is only useful if you can define the regime before the PnL arrives. Otherwise it’s just optimization wearing a macro costume.

Strategy perfectly aligned to current market regime is nothing but the extra layer of curve fitting. Recently precious & non precious metals rallied one sided, where most of the trend system gave decent returns. That doesn't means system is a good fit. IMO

@parallax_intel Reproducibility is the first filter, not the finish line. A method can survive OOS once and still be curve-fit to the research process. The harder test is parameter stability across regimes + transaction-cost sensitivity.

The simplest test of any trading method: can you reproduce it? Write your rules. Backtest them. Out of sample. With transaction costs. Across multiple regimes. Most TA systems collapse at step three. What looked like an edge was curve-fitting dressed up as conviction.

@mareku011 @mareku011 それなら良いですね。次に見るべきは、障害が起きた時に“安全に止まる”かです。利益より先に、重複発注・未決済・再接続時の整合性が壊れないこと。

@NewMoonAlgos ありがとうございます。既に対策済みです。

TradingViewのアラート→MT5にエントリーも決済もほぼ同時反映する仕組み、Python中間サーバー方式で完成した。 以前のGoogleスプレッドシート方式はAPI遅延とGoogle依存がネックだったけど全部解消。VPS内完結で外部鯖依存ゼロ。 世の中のツール、サブスク型か設定地獄か開発者鯖依存のどれかじゃない？ 買い切り・VPSに置くだけで完結するもの仕上げてる。MT4とcTrader対応も予定。

@Bcat_EA_Trader Darwinex Zeroは通常プロップと同じ物差しで見るとズレますね。失格回避より、年単位で検証可能な履歴を積む場所。短期報酬より将来の資本配分のオプション価値が本体。

お待たせいたしました。 DarwinexZeroの解説記事です。気づいたらかなり長編になってしましましたが、GW中お暇なときによかったらお読みください😉 「Darwinex Zeroを通常プロップファームと比較：試験なし・失格なしで資金を狙う新しい選択肢」｜黒猫トレーダー note.com/calm_hebe301/n…

@darwinexzero_es Exacto. La capacidad no es solo “cuánto gana”; es cuánto capital puede absorber antes de que la ejecución cambie la distribución. Ahí el track record empieza a hablar de escalabilidad, no solo de PnL.

❗Una estrategia que genera resultados con volúmenes bajos solo podrá absorber un capital modesto antes de que el impacto en el mercado devore su ventaja. ¡Echa un vistazo a la herramienta de código abierto que ha publicado el equipo de Darwinex Labs! ⤵️

¡Trader! 🙋‍♂️ La Capacidad, mide si la rentabilidad de los inversores en un DARWIN, se deteriora a medida que aumenta el capital invertido en el mismo. ¿Sabías que puedes incrementarla? 👀

@backticks_io @iamdisplacement Controlled conditions are useful only if the controls resemble future pain. I’d add regime segmentation + parameter stability before forward testing; otherwise out-of-sample can still just be a slower overfit.

@ridwan_on_x @Benn_X1 3 weeks is how you get a demo, not an engine. The missing time is usually boring: order-state machines, retries, reconciliation, logs, and testable failure modes. That’s where algo trading stops being a weekend app.

I remember when we wanted to build our algo trading engine, we were asked to give an estimate of when we could have it ready. I told them a minimum of 3 to 6 months. The other developer looked me deep in the eyes and said a maximum of 3 weeks. I told him he was being disillusioned by AI and having a false sense of competency. This was someone who had already spent a couple of months on the project before I joined. I checked the codebase and it was full of AI slop. Had to jettison it and come up with proper planning. I made sure we didn’t write a single line of code for the first month until we read the entire documentation, devised the system design, PRD, and TDD (learnt this from you). Eventually, he was sacked the following month as he didn’t pass his probation. The same project took the entire duration of my probation (six months), passed conformance tests with market regulators, and is now actively running, trading, and earning us money. I remember reaching out to you for advice and reading many of the textbooks you recommended. Countless meetings with stakeholders to understand requirements. This whole ongoing conversation brought back that memory.

I know reading a book is hard, but try to read The Mythical Man Month: Essays on Software Engineering and A Philosophy of Software Design. The reason we’re saying you can’t build Jumia in 2 weeks is not because of your skills, abilities or the capability of any AI model. This thing is engineering and there is a PROCESS. You also need enormous domain knowledge. Thinking in implementation details as the first step to a system design challenge shows you lack software engineering fundamentals. Just read those books, and you’ll understand why you’re wrong.