|
2 years
experience
|
4
products
|
100
demo versions
|
|
0
jobs
|
1
signals
|
0
subscribers
|
Finance and International Business professional with a specialization in Financial Management. Self-taught MQL5 and Python developer focused on algorithmic trading, multi-asset portfolio construction, and quantitative risk management.
My work centers on designing, optimizing, and validating Expert Advisors that operate as a coordinated portfolio rather than isolated strategies. I apply correlation analysis, temporal coverage mapping, and asset class diversification to build systems that do not rely on a single instrument or a single approach.
I currently manage algorithmic portfolios spanning forex, indices, metals, energy, and US equities, operating across multiple sessions and timeframes simultaneously.
I share my experience through technical articles and open-source tools in this community. I believe the shift from "building individual EAs" to "engineering portfolios" is the transition that separates retail from institutional thinking, and that principle guides everything I publish here.
My work centers on designing, optimizing, and validating Expert Advisors that operate as a coordinated portfolio rather than isolated strategies. I apply correlation analysis, temporal coverage mapping, and asset class diversification to build systems that do not rely on a single instrument or a single approach.
I currently manage algorithmic portfolios spanning forex, indices, metals, energy, and US equities, operating across multiple sessions and timeframes simultaneously.
I share my experience through technical articles and open-source tools in this community. I believe the shift from "building individual EAs" to "engineering portfolios" is the transition that separates retail from institutional thinking, and that principle guides everything I publish here.
Cristian David Castillo Arrieta
Published post A 90% Win Rate can still loss Money:
On many Expert Advisor pages, the first number you see is the win rate: 85%, 90%, 95%. It is the easiest number to sell, because being right most of the time feels like safety. It is also the number that hides the most risk...
38
Cristian David Castillo Arrieta
I researched 39 markets for AbacuQuant Portfolio. I rejected 22 of them.
Crude oil, the S&P 500, the DAX, ten US stocks. Some looked great on paper, then failed the real-tick test or the statistical screening. AUDJPY missed one threshold by 0.004 and was rejected anyway. The rules don't bend to let a market in.
That's the point. Many EAs are sold on the best backtest that could be found. This one is built from what survived a process designed to say no.
What made it in:
• 11 validated markets: EURUSD, USDJPY, AUDUSD, EURJPY, gold and six indices (Dow Jones, Nasdaq 100, Nikkei 225, Australia 200, Euro Stoxx 50, FTSE 100)
• 71 independent configurations running from one H1 chart
• Account-level risk control: global limits on total open risk, margin and equity drawdown across every market
• Stop loss and take profit attached to every order. No martingale, no grid, no averaging, no recovery logic
The numbers (Strategy Tester, real ticks, 0.5% risk per trade, Jan 2020 – Jul 2026):
• 11,333 trades, 80.3% win rate
• Deepest equity decline: 6.4%
• Most recent period (May 2024 – Jul 2026): profit factor 1.38, 24 of 27 months positive, all 11 markets in profit, none above 14% of the total
The trade-off, stated plainly: the average loss is about 2.8x the average win. You will see losing trades. If you want an equity curve with no visible losses, this is not your EA.
Why test it: the demo is free. Run it with "Every tick based on real ticks", on your own broker, with the balance you actually plan to trade. The EA tells you at startup what balance each configuration needs. Look closely at the losing trades.
Why get the full version: it's a research framework, not just a set of presets. Manual mode gives you the same engine, and a custom optimization criterion ranks results by the win rate's margin over its own break-even level instead of net profit. New markets are added only if they pass the same tests, and you get direct support from me in English and Spanish. I have been running this engine on my own real-money account since August 2026.
Free demo, full methodology and pricing (rental from 49 USD/month):
https://www.mql5.com/en/market/product/191885
Backtests describe the past, not the future. Trading involves risk of loss.
Crude oil, the S&P 500, the DAX, ten US stocks. Some looked great on paper, then failed the real-tick test or the statistical screening. AUDJPY missed one threshold by 0.004 and was rejected anyway. The rules don't bend to let a market in.
That's the point. Many EAs are sold on the best backtest that could be found. This one is built from what survived a process designed to say no.
What made it in:
• 11 validated markets: EURUSD, USDJPY, AUDUSD, EURJPY, gold and six indices (Dow Jones, Nasdaq 100, Nikkei 225, Australia 200, Euro Stoxx 50, FTSE 100)
• 71 independent configurations running from one H1 chart
• Account-level risk control: global limits on total open risk, margin and equity drawdown across every market
• Stop loss and take profit attached to every order. No martingale, no grid, no averaging, no recovery logic
The numbers (Strategy Tester, real ticks, 0.5% risk per trade, Jan 2020 – Jul 2026):
• 11,333 trades, 80.3% win rate
• Deepest equity decline: 6.4%
• Most recent period (May 2024 – Jul 2026): profit factor 1.38, 24 of 27 months positive, all 11 markets in profit, none above 14% of the total
The trade-off, stated plainly: the average loss is about 2.8x the average win. You will see losing trades. If you want an equity curve with no visible losses, this is not your EA.
Why test it: the demo is free. Run it with "Every tick based on real ticks", on your own broker, with the balance you actually plan to trade. The EA tells you at startup what balance each configuration needs. Look closely at the losing trades.
Why get the full version: it's a research framework, not just a set of presets. Manual mode gives you the same engine, and a custom optimization criterion ranks results by the win rate's margin over its own break-even level instead of net profit. New markets are added only if they pass the same tests, and you get direct support from me in English and Spanish. I have been running this engine on my own real-money account since August 2026.
Free demo, full methodology and pricing (rental from 49 USD/month):
https://www.mql5.com/en/market/product/191885
Backtests describe the past, not the future. Trading involves risk of loss.
Cristian David Castillo Arrieta
Can you reproduce this equity curve?
I’m sharing the backtest setup I used with AbacuQuant Portfolio so anyone can download the demo and try to reproduce the results independently.
You can test the instruments one by one if you want to analyze each market separately.
However, if you want to reproduce the portfolio configuration shown in the backtest, set the following instruments to TRUE:
* EURUSD
* USDJPY
* AUDUSD
* XAUUSD
* EURJPY
* N225
* WS30
* NDX
* STOXX50
* UK100
All other instrument presets should remain FALSE.
Risk settings used
Risk per trade: 0.5%
General protection stop: 30%
The idea is simple: don’t just look at the equity curve — download the demo and test it yourself.
Use the same instruments and settings shown above and see what results you get on your own MT5 environment.
Different brokers, spreads, commissions and execution conditions can affect the results, so your backtest may not be identical to mine.
I would actually encourage you to challenge the results.
Test the instruments individually. Test the portfolio. Change the parameters. Stress-test it.
The goal of AbacuQuant Portfolio is not to ask you to blindly trust a backtest.
It’s to give you something you can test for yourself.
👉 Download the demo and start your own backtest.
AbacuQuant Portfolio — available on MQL5.
I’m sharing the backtest setup I used with AbacuQuant Portfolio so anyone can download the demo and try to reproduce the results independently.
You can test the instruments one by one if you want to analyze each market separately.
However, if you want to reproduce the portfolio configuration shown in the backtest, set the following instruments to TRUE:
* EURUSD
* USDJPY
* AUDUSD
* XAUUSD
* EURJPY
* N225
* WS30
* NDX
* STOXX50
* UK100
All other instrument presets should remain FALSE.
Risk settings used
Risk per trade: 0.5%
General protection stop: 30%
The idea is simple: don’t just look at the equity curve — download the demo and test it yourself.
Use the same instruments and settings shown above and see what results you get on your own MT5 environment.
Different brokers, spreads, commissions and execution conditions can affect the results, so your backtest may not be identical to mine.
I would actually encourage you to challenge the results.
Test the instruments individually. Test the portfolio. Change the parameters. Stress-test it.
The goal of AbacuQuant Portfolio is not to ask you to blindly trust a backtest.
It’s to give you something you can test for yourself.
👉 Download the demo and start your own backtest.
AbacuQuant Portfolio — available on MQL5.
Cristian David Castillo Arrieta
Anyone who has spent a few months with automated trading has seen the same pattern: a backtest with a smooth, steep equity curve, followed by a live account that behaves nothing like it. In most cases, the cause is one of three things: Loss recovery...
81
Cristian David Castillo Arrieta
Published code Edge Drift Detector
Tests whether live closing deals still match the backtest of an Expert Advisor. A bootstrap-calibrated CUSUM detector raises an alarm and estimates when the drift began, a two-window check separates a short dip from a lasting decline, and an expectancy decomposition shows whether the win rate or the trade sizes changed. The script generates demonstration data on the first run and is written in pure MQL5.
390
47
Cristian David Castillo Arrieta
Published article Building a Neural Loss-Pattern Auditor in MQL5 
Aggregate metrics like win rate or profit factor miss sequence-dependent behavior, such as sizing up right after a loss. This MQL5 script trains a small native neural network on closed-deal history to estimate loss probability from behavioral and market-context features. It reports accuracy uplift over a baseline, probability calibration, and permutation feature importance, then combines them into a configurable A-F grade with concise, plain-language recommendations.
573
Cristian David Castillo Arrieta
Published article Did Your Scale Outs Actually Help? A Scale Out Value Analyzer in MQL5 
The article presents an MQL5 tool that tests whether scaling out improved results rather than only appearing disciplined. It reconstructs positions from closing-deal history and reprices the full volume at the first, last, and best exit rates actually achieved, producing a Value-Add Ratio, a Scale-Out Win Rate, and an Efficiency measure. A single-trade dependence check and a configurable A+ to F grade turn these into clear, decision-ready feedback.
382
Cristian David Castillo Arrieta
These remarkable optimization results were generated using the AbacuQuant Portfolio EA, available here:
What makes this interesting is that this is not a single strategy being optimized over and over. AbacuQuant Portfolio is designed as a portfolio engine, capable of running multiple independent configurations across different instruments and market conditions from a single MT5 chart.
The process starts by testing different combinations of strategies, trading conditions, parameters, stop/target structures and risk settings. Instead of simply looking for the configuration with the highest profit, AbacuQuant uses a custom optimization criterion designed to evaluate whether the historical win rate provides a meaningful margin above the break-even level implied by the stop and target structure.
The resulting configurations are then combined into a portfolio, allowing different strategies and instruments to operate during different market sessions while a shared risk engine controls total exposure at the account level.
The objective is not to find one “perfect” setup. It is to build a collection of independent configurations that complement each other and create a more robust portfolio.
This image shows one of those optimization and portfolio-development results. The robot used to generate it is the AbacuQuant Portfolio EA.
🔗 https://www.mql5.com/en/market/product/191885
What makes this interesting is that this is not a single strategy being optimized over and over. AbacuQuant Portfolio is designed as a portfolio engine, capable of running multiple independent configurations across different instruments and market conditions from a single MT5 chart.
The process starts by testing different combinations of strategies, trading conditions, parameters, stop/target structures and risk settings. Instead of simply looking for the configuration with the highest profit, AbacuQuant uses a custom optimization criterion designed to evaluate whether the historical win rate provides a meaningful margin above the break-even level implied by the stop and target structure.
The resulting configurations are then combined into a portfolio, allowing different strategies and instruments to operate during different market sessions while a shared risk engine controls total exposure at the account level.
The objective is not to find one “perfect” setup. It is to build a collection of independent configurations that complement each other and create a more robust portfolio.
This image shows one of those optimization and portfolio-development results. The robot used to generate it is the AbacuQuant Portfolio EA.
🔗 https://www.mql5.com/en/market/product/191885
Cristian David Castillo Arrieta
Published code Neural Loss-Pattern Auditor
Neural Loss-Pattern Auditor trains a small feed-forward neural network, written from scratch in native MQL5, on closed-deal history to test whether behavioral and market-context features predict which trades are more likely to lose. It reports an accuracy uplift over a naive baseline, a probability-calibration table, a permutation feature-importance ranking, and a configurable A-F composite grade with recommendations. On first run it uses a built-in synthetic demo, so the output is visible immediately with no setup; switch one input to InpUseDemoData=false to analyze real account history instead. Pure MQL5: no external libraries, no Python, and no AI service of any kind.
704
163
Cristian David Castillo Arrieta
Introduction AbacuQuant Portfolio is a multi-configuration Expert Advisor for MetaTrader 5. Instead of running one strategy with one set of parameters on one instrument, it runs up to 48 independent configurations at the same time, across EURUSD, GBPUSD, USDJPY and AUDUSD, from a single H1 chart...
102
Cristian David Castillo Arrieta
After months of research, I've published AbacuQuant Portfolio — a multi-configuration Expert Advisor for MetaTrader 5, and I wanted to share it here first.
Why it's different
Most EAs run one strategy with one set of parameters on one instrument. This one runs up to 48 independent configurations at once, across EURUSD, GBPUSD, USDJPY and AUDUSD, from a single chart. Each configuration owns its own trading hours, its own combination of ten built-in strategies, and its own ATR-based stop and target. A configuration built for the London session simply doesn't trade outside it — and risk is measured across the whole account, not per configuration, so running four instruments together doesn't quietly multiply your exposure.
And it's built the honest way: no martingale, no grid, no averaging into a losing position, no hidden loss-recovery logic. Every stop is attached the moment the trade opens. If you've been burned by an EA with a beautiful equity curve that fell apart the first time the market disagreed with it, this was built specifically against that failure mode.
How it decides a trade
Each active configuration waits for its window, reads only completed H1 bars (never the forming candle), requires several of its enabled strategies to agree, and sizes the position from the stop distance rather than a fixed lot — so a wider stop always means a smaller position, and the money at risk stays constant. Every configuration passed the same acceptance procedure on real-tick data before it shipped: statistical significance over its own break-even line, a minimum number of losing trades to actually measure it, and stops that were genuinely reached rather than just theorized.
Try it before you take my word for it
The demo runs in full inside the Strategy Tester — same presets, same logic, no time limit. Turn on the validated instruments, select "Every tick based on real ticks," and look at the trade list yourself: the stops, the position sizing, the risk ceilings holding across instruments. I'd rather you find out in the tester, for free, than after committing real money.
Link to the product page: https://www.mql5.com/en/market/product/191885
Why it's different
Most EAs run one strategy with one set of parameters on one instrument. This one runs up to 48 independent configurations at once, across EURUSD, GBPUSD, USDJPY and AUDUSD, from a single chart. Each configuration owns its own trading hours, its own combination of ten built-in strategies, and its own ATR-based stop and target. A configuration built for the London session simply doesn't trade outside it — and risk is measured across the whole account, not per configuration, so running four instruments together doesn't quietly multiply your exposure.
And it's built the honest way: no martingale, no grid, no averaging into a losing position, no hidden loss-recovery logic. Every stop is attached the moment the trade opens. If you've been burned by an EA with a beautiful equity curve that fell apart the first time the market disagreed with it, this was built specifically against that failure mode.
How it decides a trade
Each active configuration waits for its window, reads only completed H1 bars (never the forming candle), requires several of its enabled strategies to agree, and sizes the position from the stop distance rather than a fixed lot — so a wider stop always means a smaller position, and the money at risk stays constant. Every configuration passed the same acceptance procedure on real-tick data before it shipped: statistical significance over its own break-even line, a minimum number of losing trades to actually measure it, and stops that were genuinely reached rather than just theorized.
Try it before you take my word for it
The demo runs in full inside the Strategy Tester — same presets, same logic, no time limit. Turn on the validated instruments, select "Every tick based on real ticks," and look at the trade list yourself: the stops, the position sizing, the risk ceilings holding across instruments. I'd rather you find out in the tester, for free, than after committing real money.
Link to the product page: https://www.mql5.com/en/market/product/191885
Cristian David Castillo Arrieta
Published product
Thirty-nine markets were researched for this Expert Advisor. Twenty-two were rejected. Eleven made it in. AbacuQuant Portfolio trades the eleven that passed: three currency majors, the euro-yen cross, gold, and six stock indices that follow the trading day from Tokyo and Sydney to London, Frankfurt and New York. One H1 chart. Seventy-one independent configurations. One risk engine that measures the whole account, not one trade at a time. Every market was put through the same research process
Cristian David Castillo Arrieta
Published article Building a Hidden Risk of Ruin Auditor in MQL5 
Aggregate metrics alone do not reveal how a trade sequence manages risk. This MQL5 tool analyzes closed positions to flag four structural patterns: post-loss volume escalation, overlapping same-direction entries, asymmetric payoffs, and a classical risk-of-ruin figure. The results are merged into a configurable A-F grade with concise recommendations to guide further review.
1356
Cristian David Castillo Arrieta
Published code Scale Out Value Analyzer
A native MQL5 tool that reconstructs closed positions from deal-level history, flags the ones closed through more than one exit, and reprices each one at its own first, last, and best exit rates to measure whether scaling out actually added value. Reports a Value-Add Ratio, a Scale Out Win Rate, an Efficiency figure, and a single-trade dependence check, combined into an A+ to F score with recommendations. Runs out of the box against a built-in demonstration data set; a companion script exports the real input file from your own account history. Pure MQL5, no external libraries.
896
131
Cristian David Castillo Arrieta
Published code Hidden Risk of Ruin Auditor
Reads a closed-position trade history (a CSV file, or one generated automatically from the current account's deal history by the companion RuinExport.mq5 script) and reports four independent risk fingerprints: volume escalation after a loss, overlapping same-direction exposure that averages into a worse price, payoff asymmetry between wins and losses, and a classical risk-of-ruin estimate at a stated risk per trade. The four scores combine into a single A-to-F grade with plain-language recommendations. If no CSV is found, the script generates a reproducible demonstration book automatically, so the report is visible on the first run.
1373
253
Cristian David Castillo Arrieta
Published article Execution Cost and Slippage Sensitivity Analyzer 
Backtests often understate spread, commission, and slippage. This MQL5 analyzer loads closing deals and simulates rising execution costs to measure robustness. It computes the breakeven cost per deal, the cushion over an assumed cost, the net profit and profit factor at that cost, and how many winners turn into losers, then summarizes the result with an A+ to F grade and targeted guidance.
1845
Cristian David Castillo Arrieta
Published article Creating a Profit Concentration Analyzer in MQL5 
Net profit and win rate tell you how much a strategy made, not how the result is distributed. This article builds a native MQL5 script that reads your closed trades and measures profit concentration: the top-N trade share, the Gini coefficient of the winners, an outlier-dependence stress test that removes the best few winners, and the largest day against a prop-firm consistency limit. It combines these into one A+ to F score with recommendations, running inside MetaTrader 5.
1245
Cristian David Castillo Arrieta
Published code Portfolio Correlation and Margin Risk Calculator
Computes the historical Pearson correlation between any set of instruments and the combined margin your account would need to hold all of them at once, as a percentage of your equity. Runs natively in MetaTrader 5 with no external libraries, no Python, and no AI — set your symbol list and lot sizes as inputs and it reports the full matrix in the Experts tab and on the chart, refreshing on a timer.
1269
264
Cristian David Castillo Arrieta
Why the same trailing stop breaks the moment Gold changes character
I was in a forum thread today about trailing stops on XAUUSD, and it made me put into words something I've been building my whole approach around for a while: almost every trailing method traders compare — EMA cross, Chandelier, ATR multiples, swing-structure trails — gets judged on a single backtest run over one continuous chunk of history. The "best" multiplier or ladder step that wins that test isn't actually the best method. It's the method that happened to fit whatever mix of trend and chop was sitting in that sample.
The fix I use is simple to describe and annoying to implement properly: split the history into volatility regimes first (I use ATR percentile over a rolling window, expansion vs. compression), then optimize and validate each piece of logic separately per regime instead of once over the whole dataset. A structural trail wins clearly in expansion. In compression it just gets chopped up by noise, and something tighter does better there. Neither method is "the winner" — the regime decides which one applies.
That's the same principle I ended up building AbacuQuant around, just scaled up from one exit rule to an entire portfolio. Instead of one strategy tuned to look good on one backtest, the logic behind each strategy is walk-forward tested and optimized (genetic optimization, not a single curve-fit) across different market regimes and asset classes, forex, metals, indices, energy, ETFs, individual stocks — and then combined into a portfolio specifically to keep cross-asset correlation low (the current version sits under 0.4 correlation across most pairs in the book). The idea isn't "find the one strategy that beats the market." It's "find enough structurally different, regime-validated pieces that the portfolio doesn't fall apart when one regime ends," which is exactly the failure mode people are describing in that XAUUSD thread, just at the position level instead of the portfolio level.
It also runs entirely inside your own MetaTrader account nothing custodial, your funds never move to a third party and the newer version adds the drawdown/consistency rules prop firms check for, since that's become how a lot of people are actually trading it live.
If any of this is useful for how you're thinking about your own trailing logic or portfolio construction, happy to go deeper in the comments. And if you want to see what the regime-validated approach looks like applied across a full portfolio rather than one exit rule, it's at abacuquant.com.
Cristian David Castillo Arrieta
I was in a forum thread today about trailing stops on XAUUSD, and it made me put into words something I've been building my whole approach around for a while: almost every trailing method traders compare — EMA cross, Chandelier, ATR multiples, swing-structure trails — gets judged on a single backtest run over one continuous chunk of history. The "best" multiplier or ladder step that wins that test isn't actually the best method. It's the method that happened to fit whatever mix of trend and chop was sitting in that sample.
The fix I use is simple to describe and annoying to implement properly: split the history into volatility regimes first (I use ATR percentile over a rolling window, expansion vs. compression), then optimize and validate each piece of logic separately per regime instead of once over the whole dataset. A structural trail wins clearly in expansion. In compression it just gets chopped up by noise, and something tighter does better there. Neither method is "the winner" — the regime decides which one applies.
That's the same principle I ended up building AbacuQuant around, just scaled up from one exit rule to an entire portfolio. Instead of one strategy tuned to look good on one backtest, the logic behind each strategy is walk-forward tested and optimized (genetic optimization, not a single curve-fit) across different market regimes and asset classes, forex, metals, indices, energy, ETFs, individual stocks — and then combined into a portfolio specifically to keep cross-asset correlation low (the current version sits under 0.4 correlation across most pairs in the book). The idea isn't "find the one strategy that beats the market." It's "find enough structurally different, regime-validated pieces that the portfolio doesn't fall apart when one regime ends," which is exactly the failure mode people are describing in that XAUUSD thread, just at the position level instead of the portfolio level.
It also runs entirely inside your own MetaTrader account nothing custodial, your funds never move to a third party and the newer version adds the drawdown/consistency rules prop firms check for, since that's become how a lot of people are actually trading it live.
If any of this is useful for how you're thinking about your own trailing logic or portfolio construction, happy to go deeper in the comments. And if you want to see what the regime-validated approach looks like applied across a full portfolio rather than one exit rule, it's at abacuquant.com.
Cristian David Castillo Arrieta