Add research methodology documentation for EA-032

Document the research methodology for EA-032 Linear Regression Slope, outlining the purpose, structure, baseline testing, evaluation metrics, and research workflow.
2026-08-30 14:56:15 +07:00
parent 207a440bdb
commit e87e8242d0
1 changed files with 739 additions and 0 deletions
@@ -0,0 +1,739 @@
+# Research Methodology
+
+## Purpose
+
+This repository is designed to research, test, compare and document MetaTrader 5 Expert Advisors for XAUUSD.
+
+The methodology is intentionally simple:
+
+```text
+Strategy idea
+    ↓
+Implement EA
+    ↓
+Run reproducible baseline backtest
+    ↓
+Review results
+    ↓
+Identify one specific weakness
+    ↓
+Run controlled research experiment
+    ↓
+Compare against baseline
+    ↓
+PASS / FAIL
+```
+
+The objective is not to optimize every EA immediately.
+
+The objective is to determine whether the underlying trading logic has sufficient merit to justify further development.
+
+---
+
+# Repository Research Structure
+
+Each EA is separated into three main layers:
+
+```text
+EAs/
+Backtest/
+Research/
+```
+
+These layers serve different purposes.
+
+---
+
+## 1. EAs
+
+Location:
+
+```text
+EAs/
+└── EA-XXX_Strategy_Name/
+```
+
+Each EA folder should contain:
+
+```text
+EA-XXX_Strategy_Name.mq5
+README.md
+```
+
+### Purpose
+
+The EA folder contains the actual trading implementation.
+
+The `.mq5` source file is the authoritative reference for:
+
+* Entry logic
+* Exit logic
+* Position management
+* Risk parameters
+* Filters
+* Trading conditions
+
+The README explains what the current code actually does.
+
+It should not claim profitability unless supported by backtest evidence.
+
+---
+
+# 2. Backtest
+
+Location:
+
+```text
+Backtest/
+└── EA-XXX_Strategy_Name/
+```
+
+This folder contains the empirical Strategy Tester evidence for the EA.
+
+Recommended contents:
+
+```text
+README.md
+Strategy Tester HTML report
+Balance graph
+Trade statistics graphs
+MFE / MAE graph
+Holding-time graph
+```
+
+The exact graph files may vary depending on the MetaTrader report.
+
+---
+
+## Baseline Backtest
+
+Every EA should first receive a baseline test before any optimization.
+
+The baseline test exists to answer:
+
+```text
+Does the current implementation work under a clearly defined test configuration?
+```
+
+A baseline should use one fixed parameter set.
+
+Do not optimize multiple variables before the baseline is recorded.
+
+---
+
+# Backtest Configuration
+
+Every backtest README should record at minimum:
+
+| Category        | Required Information            |
+| --------------- | ------------------------------- |
+| EA              | EA name/version                 |
+| Symbol          | Trading instrument              |
+| Timeframe       | Test timeframe                  |
+| Start Date      | Beginning of test               |
+| End Date        | End of test                     |
+| Data Quality    | Strategy Tester history quality |
+| Initial Deposit | Starting account balance        |
+| Currency        | Account currency                |
+| Leverage        | Test leverage                   |
+| Lot Size        | Position size                   |
+| Stop Loss       | SL configuration                |
+| Take Profit     | TP configuration                |
+| Spread Filter   | If applicable                   |
+| Strategy Inputs | Relevant EA parameters          |
+
+This is required so another researcher can reproduce the test.
+
+---
+
+# Data Quality
+
+Whenever possible, the primary baseline should use:
+
+```text
+100% real ticks
+```
+
+when supported by MetaTrader 5 and available broker history.
+
+The reported history quality must always be documented.
+
+Do not describe a result as comparable if two tests use materially different data quality without explicitly noting that difference.
+
+---
+
+# Core Evaluation Metrics
+
+The following metrics should be recorded for every baseline test.
+
+## Total Net Profit
+
+```text
+Total Net Profit
+```
+
+Shows the net result after all winning and losing trades in the Strategy Tester simulation.
+
+Net Profit alone is not sufficient to determine strategy quality.
+
+---
+
+## Profit Factor
+
+```text
+Profit Factor = Gross Profit / |Gross Loss|
+```
+
+General interpretation:
+
+```text
+PF < 1.00
+```
+
+indicates that gross losses exceeded gross profits during the test.
+
+A Profit Factor above `1.00` is necessary for a profitable historical test, but is not sufficient by itself to establish robustness.
+
+---
+
+## Expected Payoff
+
+Expected Payoff represents the average historical result per trade in the Strategy Tester report.
+
+A negative Expected Payoff is a direct warning that the tested configuration had negative expectancy over the sample.
+
+---
+
+## Maximum Drawdown
+
+Both balance and equity drawdown should be recorded where available.
+
+Drawdown is one of the primary risk metrics.
+
+A strategy with positive net profit but extreme drawdown should not automatically be considered successful.
+
+---
+
+## Win Rate
+
+Record:
+
+```text
+Winning Trades
+Losing Trades
+Win Rate
+Loss Rate
+```
+
+Where useful, also separate:
+
+```text
+Long win rate
+Short win rate
+```
+
+This can help identify asymmetric behavior between BUY and SELL signals.
+
+---
+
+## Average Winner / Average Loser
+
+Record:
+
+```text
+Average Profit Trade
+Average Loss Trade
+```
+
+These values help evaluate the actual payoff structure of the strategy.
+
+---
+
+## Consecutive Losses
+
+Record:
+
+```text
+Maximum consecutive losses
+Average consecutive losses
+```
+
+This helps evaluate whether the trading system produces risk clusters that may be difficult to tolerate operationally.
+
+---
+
+## Holding Time
+
+Where available, record:
+
+```text
+Minimum holding time
+Maximum holding time
+Average holding time
+```
+
+Holding duration helps identify whether the EA behaves primarily as:
+
+```text
+scalping
+short-term
+intraday
+swing
+```
+
+The classification should only be used descriptively and should not replace the actual reported holding-time data.
+
+---
+
+# MFE and MAE
+
+MetaTrader may report:
+
+```text
+MFE = Maximum Favorable Excursion
+MAE = Maximum Adverse Excursion
+```
+
+These metrics can be useful for understanding how trades behave before they close.
+
+They may help investigate:
+
+* Stop Loss placement
+* Take Profit placement
+* Premature exits
+* Large adverse excursions
+* Unrealized profit before reversal
+
+MFE and MAE should be treated as diagnostic tools rather than standalone proof of strategy quality.
+
+---
+
+# PASS / FAIL Framework
+
+Every baseline receives a research status.
+
+The status is not the same as approval for live trading.
+
+---
+
+## Baseline FAIL
+
+A baseline should normally be classified as FAIL when one or more major problems are present, such as:
+
+```text
+Negative Total Net Profit
+Profit Factor < 1.00
+Negative Expected Payoff
+Extremely high drawdown
+Severe balance deterioration
+```
+
+The exact reason should be documented.
+
+Do not modify the result to make the EA appear successful.
+
+---
+
+## Baseline PASS
+
+A baseline may progress to further validation when it demonstrates:
+
+```text
+Positive historical expectancy
+Acceptable drawdown
+Sufficient trade sample
+No obvious structural failure
+```
+
+A PASS means:
+
+```text
+worthy of additional research
+```
+
+It does NOT mean:
+
+```text
+ready for live trading
+```
+
+---
+
+# Research Process After Baseline
+
+After the baseline, research should focus on one hypothesis at a time.
+
+Example:
+
+```text
+Baseline problem:
+Too many weak signals
+
+Hypothesis:
+A minimum slope threshold may remove weak entries
+
+Experiment:
+Change only the slope filter
+
+Compare:
+Baseline vs Experiment
+```
+
+This makes it possible to understand what caused the result to change.
+
+---
+
+# One Variable at a Time
+
+Whenever practical, early research experiments should change only one primary variable.
+
+Example:
+
+```text
+Baseline:
+RegPeriod = 20
+No slope threshold
+
+Experiment A:
+RegPeriod = 20
+Slope threshold added
+```
+
+Avoid simultaneously changing:
+
+```text
+Regression period
+Stop Loss
+Take Profit
+Trailing Stop
+Trading session
+Slope threshold
+```
+
+because it becomes difficult to determine which change caused the improvement or deterioration.
+
+---
+
+# Research Experiment Naming
+
+Recommended experiment naming:
+
+```text
+EXP-001
+EXP-002
+EXP-003
+```
+
+Example:
+
+```text
+EA-032
+├── BASELINE
+├── EXP-001_Slope_Threshold
+├── EXP-002_RegPeriod
+└── EXP-003_Time_Filter
+```
+
+Each experiment should have:
+
+```text
+Objective
+Hypothesis
+Changed variable
+Fixed variables
+Backtest configuration
+Result
+Comparison with baseline
+PASS / FAIL
+Conclusion
+```
+
+---
+
+# Optimization Policy
+
+Optimization should not be the first step.
+
+Use this order:
+
+```text
+1. Baseline
+2. Identify structural weakness
+3. Controlled experiment
+4. Validate whether signal improves
+5. Only then perform parameter optimization
+```
+
+Large parameter searches can create misleading results if the underlying trading logic is weak.
+
+Optimization should therefore come after the strategy demonstrates basic merit.
+
+---
+
+# Avoiding Overfitting
+
+A profitable backtest is not sufficient evidence that an EA is robust.
+
+Research should watch for:
+
+* Excessive parameter tuning
+* Very narrow profitable parameter ranges
+* Large performance changes from small parameter changes
+* Excellent performance only in one short historical period
+* Very small trade samples
+* Selection of only favorable test periods
+
+A robust strategy should ideally remain reasonably stable when assumptions change moderately.
+
+---
+
+# In-Sample and Out-of-Sample Testing
+
+After a strategy produces a promising result, historical data should eventually be separated into:
+
+```text
+In-Sample
+Out-of-Sample
+```
+
+### In-Sample
+
+Used for:
+
+```text
+research
+parameter development
+optimization
+```
+
+### Out-of-Sample
+
+Used for:
+
+```text
+independent validation
+```
+
+The Out-of-Sample period should not be repeatedly optimized against.
+
+---
+
+# Forward Testing
+
+A strategy that survives historical validation should eventually progress to forward testing.
+
+Recommended progression:
+
+```text
+Backtest
+    ↓
+Out-of-Sample
+    ↓
+Demo / Forward Test
+    ↓
+Small controlled live test
+```
+
+Moving to the next stage requires explicit review.
+
+A historical PASS does not automatically authorize live deployment.
+
+---
+
+# Source Code Review
+
+Before interpreting a backtest, verify that the EA implementation matches the intended strategy.
+
+Review at minimum:
+
+```text
+Entry conditions
+Exit conditions
+SL / TP calculation
+Position counting
+Magic Number filtering
+Spread filter
+New-bar / tick execution
+Break Even logic
+Trailing Stop logic
+Unused inputs
+```
+
+If the source code contains an implementation issue, document it before attributing the backtest result entirely to the trading concept.
+
+---
+
+# Reproducibility
+
+Every important research result should be reproducible.
+
+Keep:
+
+```text
+EA source
+Exact EA parameters
+Test symbol
+Timeframe
+Date range
+Broker / server where relevant
+Strategy Tester report
+Charts
+Research conclusion
+```
+
+If the result cannot be reproduced, it should not be treated as strong research evidence.
+
+---
+
+# Documentation Rule
+
+Repository documentation should separate three things clearly:
+
+### Code facts
+
+What the `.mq5` source actually implements.
+
+### Backtest facts
+
+What MetaTrader Strategy Tester actually reported.
+
+### Research interpretation
+
+What is inferred from those results.
+
+Interpretation must never be presented as if it were raw Strategy Tester data.
+
+---
+
+# Research Workflow
+
+Standard workflow:
+
+```text
+STEP 1
+Create / receive EA source
+        ↓
+
+STEP 2
+Review implementation
+        ↓
+
+STEP 3
+Document EA
+        ↓
+
+STEP 4
+Run baseline backtest
+        ↓
+
+STEP 5
+Save Strategy Tester evidence
+        ↓
+
+STEP 6
+Document baseline
+        ↓
+
+STEP 7
+Research failure / strength
+        ↓
+
+STEP 8
+Define one experiment
+        ↓
+
+STEP 9
+Modify EA
+        ↓
+
+STEP 10
+Run the same test conditions
+        ↓
+
+STEP 11
+Compare against baseline
+        ↓
+
+STEP 12
+PASS / FAIL experiment
+```
+
+---
+
+# Minimum Evidence Requirement
+
+A research result should not be considered complete without evidence.
+
+Minimum evidence:
+
+```text
+EA source
+
+Strategy Tester report
+
+documented test parameters
+
+key performance metrics
+
+research conclusion
+```
+
+Screenshots alone should not replace the original Strategy Tester report when the report is available.
+
+---
+
+# Current Example — EA-032
+
+The first documented case in this repository is:
+
+```text
+EA-032_Linear_Regression_Slope
+```
+
+Its baseline was tested using:
+
+```text
+XAUUSD.PRO
+M1
+2026.01.02 - 2026.04.01
+100% real ticks
+```
+
+The baseline produced:
+
+```text
+Net Profit       = -$993.19
+Profit Factor    = 0.91
+Win Rate         = 31.40%
+Maximum Drawdown = 99.36%
+Total Trades     = 5,039
+```
+
+Therefore:
+
+```text
+EA-032 BASELINE = FAIL
+```
+
+This result serves as the reference point for subsequent EA-032 research experiments.
+
+---
+
+# Repository Principle
+
+The repository follows one primary principle:
+
+```text
+Build
+→ Test
+→ Measure
+→ Understand
+→ Change one thing
+→ Test again
+```
+
+Do not optimize a strategy before understanding its baseline behavior.
+
+Do not claim a strategy works without reproducible evidence.
+
+Do not treat a successful backtest as proof of future profitability.