Articles on data analysis and statistics in MQL5

An MQL5 script reconstructs closed trades from deal history using a two-pass SL/TP lookup and exports them to an Excel-compatible XLSX file without third-party libraries. Four cooperating classes handle trade data, history reconstruction, SpreadsheetML XML generation, and ZIP assembly via .NET's ZipFile class through a direct ShellExecuteW call with marker-file polling. The output opens in Excel and Google Sheets with correct numeric types, formatted date columns, and a bold header row.

An LLM agent with SQL memory in MetaTrader 5 — an architecture that eliminates the main limitation of traditional LLM Expert Advisors (EAs): the loss of context between queries. Each agent decision is recorded in SQLite along with the market conditions at the time of the signal — RSI, MA alignment, volatility — and after the position is closed, the outcome is written to the database. When the next request comes in, the agent reviews its own history — similar past RSI conditions and the last three agent decisions — and only then analyzes the current market. The result is a system that remembers its errors between restarts and adapts its behavior based on accumulated statistics, rather than starting from scratch with each new bar.

We decode ONNX files in pure MQL5 by implementing a Protocol Buffers reader from scratch. We generate a sample network in Python, verify it in Netron, and then parse the same binary to recover graph nodes, connections, weight tensors, and input/output shapes. The result is a MetaTrader 5 program that inspects a trained model's structure before inference, without any external libraries.

This part focuses on practical data analysis in MQL5 with dataanalysis.mqh. We prepare a labeled dataset from bars, apply normalization, explore redundancy with PCA, and train a decision forest to classify future bar regimes. The article shows how to obtain out-of-bag estimates and permutation importance, helping you validate the model and understand which inputs matter most.

This article explains Monte Carlo simulation and analysis for trading and guides you through a Python tool that ingests MetaTrader 5 HTML reports. It generates many randomized equity paths, then summarizes them with max drawdown, bust/profit rates, and percentile envelopes around the mean curve. The workflow helps you assess uncertainty, separate normal behavior from outliers, and size positions accordingly.

The framework evaluates boundary sensitivity by rebuilding shifted target-timeframe bars from completed M1 sources and running one Strategy Tester pass for each boundary offset. It includes a ShiftedBarBuilder, a demonstration EA, and a deterministic self-test. An XAUUSD M15 case study reports net profit, Profit Factor, maximum relative equity drawdown, and total trades across all 15 offsets to quantify dependence on the bar grid.

Direct calls to the MQL5 History API inside analytics components create hidden terminal dependencies that make isolated testing structurally impossible. This article constructs an ITradeRepository abstraction layer with CLiveTradeRepository and CMockTradeRepository implementations, enabling the same analytics engine and equity curve panel to operate identically against live account data or a deterministic in-memory dataset. Repository injection eliminates direct API coupling, supports offline validation, and confines data source changes to a single implementation class.

A Sharpe ratio read off the best of many optimization runs is not the number it looks like. This article ships a reusable native CDeflatedSharpe class that turns a raw Sharpe into an honest confidence statement. The Probabilistic Sharpe Ratio corrects it for sample length and for skew and kurtosis; the Deflated Sharpe Ratio adds the correction almost nobody applies, for the number of variants you tried before keeping the best. Everything is from scratch, the sample moments, the normal CDF and its inverse included, so there is no Python, no DLL and no library. On a real sweep of 56 moving-average variants on XAUUSD the winner looked significant at 98.5 percent by PSR, then fell to 90.6 percent once the 56 trials were admitted, below the usual bar. That gap is the selection bias, made measurable.

We implement Symbolic Fourier Approximation in MQL5 and compare it to SAX under a shared harness on identical price windows. SFA keeps low‑frequency Fourier coefficients and learns per‑position bins (MCB), with a proven, sound lower bound. The measurements show how the same bit budget behaves under different splits of word length and alphabet, and give a practical rule for choosing settings for your symbol.

This work adds an explainable validation module to the recording toolkit for double bottoms. It finds and filters swing lows, selects the neckline, quantifies depth, low similarity, symmetry and recovery, and assigns a rule-based status and scores accepted pairs. The settings are configurable, the process is reproducible, and the original sample data remains intact for comparison across rule sets.

Feature importance often understates correlated predictors by spreading one signal across many engineered copies, while unrelated noise can appear higher. We measure this effect against a known ground truth and compare four remedies: permutation importance with purged cross-validation, single-feature models, and clustered impurity versus clustered accuracy. The results include per-method rankings and a per-cluster dilution ratio that help identify true signals and avoid deleting valuable features.

We implement three MQL5 classes that replace fixed 1R/2R/3R targets with data-driven scale-out levels. CExcursionTracker records each closed trade's maximum favorable excursion in R, CExitLadderCalibrator derives runs from distribution percentiles with lookback and minimum-sample controls, and CLadderExecutor executes them on open positions. The ladder recalibrates as trades accumulate and uses a fallback until enough samples exist.

The article describes the transition from four-voice debates to the Council of Fifteen: ten analysts, four independent risk managers, and a Chair, with strict voting rules. It examines the roles of the participants, the three-phase architecture, and the parallel execution of a full cycle in 10–15 seconds. It shows the operational log, the risk-gate rules, and backward compatibility so you can quickly connect the system to an Expert Advisor.

This article extends the candlestick encoding framework by separating previously unclassified bullish and bearish candlesticks into distinct symbols. Using an MQL5 script, historical data were automatically encoded, sequential patterns were extracted, and frequency statistics were generated. The results show that these newly identified candles represent a significant portion of market activity, revealing structural information previously hidden by the generic underscore notation. The expanded taxonomy improves the completeness of the symbolic representation while preserving an objective, quantitative framework for large-scale market structure analysis.

The article presents a compact dashboard that relates trade duration to net profit using MQL5 and CCanvas. It pulls closed deals, derives duration in minutes, and renders a log‑scaled scatter by symbol, with an overlaid least‑squares line and R². A bucketed duration view identifies which hold‑time range produced the highest average result, helping assess exit timing.

In this article, we will implement a number of improvements to ensure that the position indicator accurately reflects the actual state on the trading server in terms of open positions and their current state. I should point out that the applications shown here are in no way intended to replace any of the elements available in MetaTrader 5. They should also not be used without due caution and a balanced approach, since their purpose is to provide educational code—that is, code intended solely for learning how the system works. The reason I call this code “educational” is that, in some cases, using messages is not the best way to implement certain functions.