Add research methodology for XAUUSD MT5 EA

This document outlines the research methodology for evaluating MetaTrader 5 Expert Advisors for XAUUSD, detailing the structured process from strategy implementation to final assessment.
2026-08-27 12:07:23 +07:00
parent 474e05cbd4
commit b98ffd9706
1 changed files with 892 additions and 0 deletions
@@ -0,0 +1,892 @@
+# XAUUSD MT5 EA Research Methodology
+
+## 1. Purpose
+
+This document defines the research methodology used in the `xauusd-mt5-ea-research` repository.
+
+The objective is to evaluate MetaTrader 5 Expert Advisors for XAUUSD through a structured and reproducible process.
+
+The methodology separates:
+
+```text
+Strategy Implementation
+        ↓
+Baseline Backtest
+        ↓
+Research
+        ↓
+Controlled Experiments
+        ↓
+Validation
+        ↓
+Final Assessment
+```
+
+A strategy is not considered successful simply because one backtest is profitable.
+
+Likewise, a failed baseline does not automatically mean the research idea should be discarded.
+
+Every conclusion should be supported by reproducible test evidence.
+
+---
+
+# 2. Repository Structure
+
+Each part of the repository has a separate responsibility.
+
+```text
+xauusd-mt5-ea-research/
+│
+├── EAs/
+│   └── EA-<>/
+│       ├── EA-<>.mq5
+│       └── README.md
+│
+├── Backtest/
+│   └── EA-<>/
+│
+├── Research/
+│   └── README.md
+│
+├── docs/
+│   └── methodology.md
+│
+└── GitHub_Profile/
+    └── README.md
+```
+
+### `EAs/`
+
+Contains the actual MQL5 implementation and documentation describing what the EA does.
+
+The EA README should document:
+
+* Strategy concept
+* Entry logic
+* Exit logic
+* Risk management
+* Inputs
+* Position management
+* Known implementation characteristics
+
+It should describe the source code as implemented rather than how the strategy was theoretically intended to work.
+
+---
+
+### `Backtest/`
+
+Contains reproducible Strategy Tester evidence for each EA.
+
+Examples:
+
+```text
+Backtest/
+└── EA-028_Ichimoku_Cloud/
+    ├── README.md
+    ├── ReportTester-XXXXXX.html
+    ├── ReportTester-XXXXXX.png
+    ├── ReportTester-XXXXXX-hst.png
+    ├── ReportTester-XXXXXX-mfemae.png
+    └── ReportTester-XXXXXX-holding.png
+```
+
+Original test evidence should be retained.
+
+Later tests must not silently replace earlier results.
+
+---
+
+### `Research/`
+
+Contains:
+
+* Observations
+* Research questions
+* Hypotheses
+* Experiment plans
+* Experiment conclusions
+
+Research hypotheses must be clearly separated from findings supported by evidence.
+
+---
+
+### `docs/`
+
+Contains repository-wide methodology and standards.
+
+This file belongs here because the methodology applies to all EAs rather than one specific strategy.
+
+---
+
+# 3. Research Workflow
+
+Every EA should follow the same basic workflow.
+
+## Stage 1 — Strategy Definition
+
+Define exactly what the strategy is intended to test.
+
+Record:
+
+* Indicator or strategy concept
+* Entry conditions
+* Exit conditions
+* Position sizing
+* Stop Loss
+* Take Profit
+* Trade management
+* Filters
+* Relevant parameters
+
+At this stage, no profitability claim should be made.
+
+---
+
+# 4. Source Code Review
+
+Before interpreting a backtest, inspect the actual `.mq5` implementation.
+
+The objective is to answer:
+
+> What does the EA actually execute?
+
+This distinction is important because the implemented strategy may differ from the conceptual strategy.
+
+For example:
+
+```text
+Tenkan > Kijun
+```
+
+is not equivalent to:
+
+```text
+Tenkan crosses above Kijun
+```
+
+The first describes a persistent state.
+
+The second describes an event.
+
+Research documentation should reflect the implemented behavior.
+
+---
+
+# 5. Baseline Backtest
+
+Every EA begins with a baseline test.
+
+The baseline establishes the reference result against which future experiments are compared.
+
+The baseline should use a documented configuration and should not be optimized after seeing the result.
+
+At minimum record:
+
+| Category   | Required Information   |
+| ---------- | ---------------------- |
+| EA         | EA name/version        |
+| Symbol     | Tested symbol          |
+| Timeframe  | Tested timeframe       |
+| Period     | Start/end dates        |
+| Data       | History/model quality  |
+| Deposit    | Initial capital        |
+| Leverage   | Account leverage       |
+| Inputs     | Complete EA parameters |
+| Trades     | Number of trades       |
+| Profit     | Net Profit             |
+| PF         | Profit Factor          |
+| Expectancy | Expected Payoff        |
+| DD         | Maximum Drawdown       |
+| Evidence   | Original tester report |
+
+---
+
+# 6. Baseline Evidence
+
+The original MetaTrader 5 Strategy Tester report should be preserved.
+
+Where available, retain:
+
+```text
+HTML report
+Balance chart
+Trade distribution charts
+MFE / MAE chart
+Holding-time chart
+```
+
+The purpose is reproducibility.
+
+A README summary is not a replacement for the original Strategy Tester evidence.
+
+---
+
+# 7. Baseline Assessment
+
+A baseline should be classified according to the evidence.
+
+Possible states:
+
+```text
+PASS
+FAIL
+INCONCLUSIVE
+```
+
+### PASS
+
+The baseline satisfies the predefined acceptance criteria for the research stage.
+
+PASS does **not** mean the EA is ready for live trading.
+
+---
+
+### FAIL
+
+The baseline fails one or more critical acceptance criteria.
+
+A failed baseline may still continue into research if there is a clear hypothesis worth testing.
+
+---
+
+### INCONCLUSIVE
+
+The available evidence is insufficient to make a reliable assessment.
+
+Examples include:
+
+* Insufficient trade sample
+* Incomplete historical data
+* Invalid test configuration
+* Missing evidence
+* Execution errors
+
+---
+
+# 8. Research After Baseline
+
+After the baseline, identify the main research problem.
+
+Example:
+
+```text
+Observation:
+Profit Factor = 0.92
+
+Research question:
+Why does the strategy have negative expectancy?
+```
+
+Do not convert an observation directly into an unsupported explanation.
+
+Incorrect:
+
+```text
+The EA loses because M1 is noisy.
+```
+
+unless an experiment has demonstrated this.
+
+Correct:
+
+```text
+Hypothesis:
+M1 may contain too much noise for this strategy.
+
+Required test:
+Compare the same strategy across multiple timeframes.
+```
+
+---
+
+# 9. Hypothesis-Driven Testing
+
+Every significant strategy modification should correspond to a research hypothesis.
+
+Recommended format:
+
+```text
+Hypothesis
+    ↓
+Experiment
+    ↓
+Evidence
+    ↓
+Result
+    ↓
+Conclusion
+```
+
+Example:
+
+```text
+H1:
+The strategy performs poorly because M1 produces excessive low-quality signals.
+
+Experiment:
+Run the unchanged strategy on M1, M5, M15 and H1.
+
+Evidence:
+MT5 Strategy Tester reports.
+
+Conclusion:
+Accept or reject H1 based on the results.
+```
+
+---
+
+# 10. One Variable at a Time
+
+During diagnostic research, change as few variables as possible.
+
+Avoid:
+
+```text
+Change timeframe
+ change indicator settings
+ change SL/TP
+ add ATR
+ add session filter
+ change trailing stop
+```
+
+in a single experiment.
+
+If performance changes, the cause would be unclear.
+
+Prefer:
+
+```text
+Baseline
+   ↓
+Change one component
+   ↓
+Compare
+   ↓
+Record result
+```
+
+This rule can be relaxed later during explicit optimization stages, but not during initial diagnosis.
+
+---
+
+# 11. Experiment Identification
+
+Experiments should have unique IDs.
+
+Recommended format:
+
+```text
+EXP-<EA ID>-<Experiment Number>_<Description>
+```
+
+Example:
+
+```text
+EXP-028-001_Timeframe
+EXP-028-002_Crossover
+EXP-028-003_Regime
+EXP-028-004_ExitManagement
+EXP-028-005_Session
+EXP-028-006_VolatilityExit
+```
+
+This makes experiments traceable to their evidence.
+
+---
+
+# 12. Experiment Record
+
+Every experiment should record at least:
+
+```text
+Experiment ID
+EA Version
+Hypothesis
+Symbol
+Timeframe
+Test Period
+History Quality
+Parameters
+Net Profit
+Profit Factor
+Expected Payoff
+Maximum Drawdown
+Total Trades
+Evidence
+Result
+Conclusion
+```
+
+A result without its configuration is not considered reproducible research evidence.
+
+---
+
+# 13. Backtest Comparison
+
+Experiments should primarily be compared against their designated baseline or control.
+
+Important metrics include:
+
+### Profitability
+
+```text
+Net Profit
+Gross Profit
+Gross Loss
+Profit Factor
+Expected Payoff
+```
+
+### Risk
+
+```text
+Maximum Balance Drawdown
+Maximum Equity Drawdown
+Recovery Factor
+```
+
+### Trade Behavior
+
+```text
+Total Trades
+Win Rate
+Average Winner
+Average Loser
+Consecutive Wins
+Consecutive Losses
+Holding Time
+```
+
+### Stability
+
+Where appropriate:
+
+```text
+Sharpe Ratio
+Balance curve behavior
+Performance by period
+Performance by direction
+Performance by session
+```
+
+No single metric should be used as the sole evidence of strategy quality.
+
+---
+
+# 14. Profit Factor
+
+Profit Factor is calculated conceptually as:
+
+```text
+Gross Profit
+────────────
+Gross Loss
+```
+
+A Profit Factor below `1.0` means gross losses exceed gross profits.
+
+A value above `1.0` means gross profits exceed gross losses.
+
+However:
+
+> Profit Factor > 1 does not by itself prove that a strategy is robust.
+
+It must be evaluated together with drawdown, trade count, test period and other performance characteristics.
+
+---
+
+# 15. Drawdown
+
+Drawdown is treated as a critical risk metric.
+
+A profitable strategy with unacceptable drawdown should not automatically PASS.
+
+Research should consider:
+
+```text
+Return
+AND
+Risk
+```
+
+rather than maximizing profit alone.
+
+Very high drawdown may indicate that the strategy is not practically survivable even when final net profit is positive.
+
+---
+
+# 16. Trade Sample Size
+
+A very small number of trades can produce misleading performance statistics.
+
+Trade count should therefore always be reported.
+
+There is no universal minimum number of trades that proves a strategy is valid.
+
+Instead, sample adequacy should be considered together with:
+
+* Strategy frequency
+* Test duration
+* Market coverage
+* Distribution of trades
+* Stability of results
+
+---
+
+# 17. Parameter Optimization
+
+Parameter optimization should not be the first response to a failed baseline.
+
+The preferred sequence is:
+
+```text
+Understand failure
+        ↓
+Form hypothesis
+        ↓
+Test hypothesis
+        ↓
+Identify useful behavior
+        ↓
+Then optimize
+```
+
+Optimization performed too early increases the risk of fitting parameters to historical noise.
+
+---
+
+# 18. Overfitting Control
+
+A strategy should not be accepted merely because an optimized parameter set performs well on the same historical data used to discover it.
+
+Research should distinguish between:
+
+```text
+Development Data
+```
+
+and:
+
+```text
+Validation Data
+```
+
+Parameters developed using one historical sample should later be evaluated on data that was not used to select those parameters.
+
+---
+
+# 19. In-Sample and Out-of-Sample Testing
+
+Once a strategy shows enough promise to justify validation, historical data should be separated conceptually into:
+
+### In-Sample
+
+Used for:
+
+* Strategy development
+* Hypothesis testing
+* Parameter selection
+
+### Out-of-Sample
+
+Used to evaluate the resulting strategy without further adjustment.
+
+If the strategy performs well in-sample but fails substantially out-of-sample, the research result should not be considered robust.
+
+---
+
+# 20. Robustness Testing
+
+After a candidate strategy survives initial research, additional validation may include:
+
+```text
+Different historical periods
+Different market regimes
+Parameter sensitivity
+Spread sensitivity
+Execution assumptions
+Out-of-sample testing
+Forward testing
+```
+
+These tests belong to a later validation stage.
+
+They should not be added before the basic strategy has demonstrated enough evidence to justify them.
+
+---
+
+# 21. Parameter Sensitivity
+
+A potentially robust strategy should not depend entirely on one exact parameter combination.
+
+For example, if:
+
+```text
+Parameter = 26
+```
+
+is profitable while:
+
+```text
+25
+and
+27
+```
+
+fail dramatically, the result deserves additional investigation.
+
+Broad regions of acceptable performance are generally more informative than one isolated historical optimum.
+
+---
+
+# 22. Transaction Costs
+
+XAUUSD strategies can be sensitive to execution conditions.
+
+Where relevant, research should document:
+
+```text
+Spread
+Commission
+Slippage
+Execution assumptions
+Broker symbol specifications
+```
+
+This is particularly important for strategies with:
+
+```text
+High trade frequency
+
+Short holding time
+```
+
+because transaction costs can materially affect expectancy.
+
+---
+
+# 23. Failed Experiments
+
+Failed experiments must be retained.
+
+A failed experiment provides information about what was tested and prevents the same unsupported idea from being repeatedly rediscovered.
+
+Therefore:
+
+```text
+FAIL ≠ Delete
+```
+
+Instead:
+
+```text
+FAIL
+ ↓
+Record evidence
+ ↓
+Record conclusion
+ ↓
+Decide whether another hypothesis is justified
+```
+
+---
+
+# 24. Evidence Rule
+
+A research claim should be traceable to evidence.
+
+Preferred relationship:
+
+```text
+Claim
+ ↓
+Experiment ID
+ ↓
+Test configuration
+ ↓
+Original MT5 report
+```
+
+If evidence is missing, the claim should be marked as:
+
+```text
+UNVERIFIED
+```
+
+rather than treated as established fact.
+
+---
+
+# 25. Version Control
+
+Changes to trading logic should result in an identifiable EA version or experiment reference.
+
+Avoid changing source code and then overwriting the evidence from the previous version.
+
+The intended relationship is:
+
+```text
+EA Version
+    ↕
+Experiment
+    ↕
+Backtest Evidence
+```
+
+This allows a result to be reproduced later.
+
+---
+
+# 26. Research Decision Levels
+
+The following levels are used to avoid confusing a successful experiment with a production-ready system.
+
+```text
+LEVEL 0 — IMPLEMENTED
+EA compiles and executes.
+
+LEVEL 1 — BASELINE TESTED
+A reproducible baseline exists.
+
+LEVEL 2 — RESEARCH CANDIDATE
+There is enough evidence to justify further investigation.
+
+LEVEL 3 — CANDIDATE STRATEGY
+Controlled research shows potentially useful behavior.
+
+LEVEL 4 — VALIDATED CANDIDATE
+The strategy survives additional robustness and out-of-sample testing.
+
+LEVEL 5 — FORWARD TEST CANDIDATE
+The strategy is approved for controlled forward/demo testing.
+
+LEVEL 6 — PRODUCTION CANDIDATE
+Evidence is sufficient for a separate decision about live deployment.
+```
+
+Progression between levels requires evidence.
+
+A profitable backtest alone does not automatically advance an EA to production.
+
+---
+
+# 27. Current Example — EA-028
+
+EA-028 provides the first documented example of this methodology.
+
+Baseline:
+
+```text
+EA:            EA-028_Ichimoku_Cloud
+Symbol:        XAUUSD.PRO
+Timeframe:     M1
+Period:        2026.01.02 – 2026.03.01
+Real Ticks:    100%
+Trades:        6,457
+
+Net Profit:    -$953.45
+Profit Factor: 0.92
+Expected:      -$0.15
+Equity DD:     95.63%
+```
+
+The correct conclusion is:
+
+```text
+BASELINE = FAIL
+```
+
+The incorrect conclusion would be:
+
+```text
+Ichimoku does not work on XAUUSD.
+```
+
+The baseline only establishes that:
+
+> This implementation, using this configuration, on this symbol/timeframe and historical period, failed the baseline test.
+
+Further claims require further evidence.
+
+---
+
+# 28. EA-028 Current Research Path
+
+The current planned sequence is:
+
+```text
+EA-028 Baseline
+      ↓
+EXP-028-001
+Timeframe Comparison
+      ↓
+EXP-028-002
+Entry / Crossover Test
+      ↓
+EXP-028-003
+Market Regime Analysis
+      ↓
+EXP-028-004
+Exit Management
+      ↓
+EXP-028-005
+Session Analysis
+      ↓
+EXP-028-006
+Volatility Exit
+      ↓
+Candidate?
+   ↙       ↘
+ NO        YES
+Stop /     Validation
+Archive
+```
+
+Each stage should only proceed when the preceding evidence justifies the next experiment.
+
+---
+
+# 29. Core Research Principles
+
+The repository follows five core rules:
+
+```text
+1. Preserve original evidence.
+
+2. Separate observation from hypothesis.
+
+3. Change one major variable at a time during diagnosis.
+
+4. Do not optimize before understanding the baseline.
+
+5. Do not claim robustness without independent validation.
+```
+
+---
+
+# 30. Final Standard
+
+The purpose of this repository is not to produce the highest possible historical backtest result.
+
+The purpose is to answer:
+
+> **Does the trading idea demonstrate a repeatable and sufficiently robust edge to justify further testing?**
+
+Every EA should therefore progress through:
+
+```text
+IMPLEMENT
+   ↓
+TEST
+   ↓
+UNDERSTAND
+   ↓
+EXPERIMENT
+   ↓
+VALIDATE
+   ↓
+DECIDE
+```
+
+with the original evidence preserved at every stage.