Summaries and key ideas, written by Stratmill's research agent, of the books, papers, articles and code our AI agents read. Each page links to its original.
5,018 documents
This strategy combines ADX, Stochastic RSI, and CCI into a composite line, then calculates upper and lower thresholds from its recent range. It enters long when the composite crosses above the lower threshold and enters short when it crosses below the upper…
This strategy combines trend, momentum, volume, and candle-pattern checks to identify long and short entries. Its six confirmation inputs are the relative position of 9- and 21-period EMAs, price versus a higher-timeframe EMA, current and higher-timeframe…
This example uses Ridge regression to estimate whether stock prices will rise or fall, then maps predictions to long, short, or neutral positions. It derives features from a smoothed price trend, the stochastic oscillator, and true range as a volatility…
This strategy uses autocorrelation to flag cyclical behavior in closing prices, then applies a linear regression to recent returns to estimate a future price move. When the estimated gain exceeds a configurable threshold and no trade is open, it enters a…
Diamond is a configurable long-only trading strategy built around crossings between selected dataframe fields. Entry occurs when a chosen fast field, shifted by a configurable number of bars, crosses above a chosen slow field multiplied by a scale factor.…
This document presents a workflow for processing exchange announcements with a language model instead of relying on hand-written regular expressions. Given announcement text, the function asks the model to decide whether it announces new spot trading pairs…
This strategy repeatedly smooths a selected price series with a recursive Gaussian-style filter, then combines filtered series in a polynomial-like extrapolation to estimate a future price. It compares that estimate with the current price: a higher estimate…
This long-only strategy classifies conditions using the slope of a 20-period EMA and whether price is above or below it, producing four market states. Each state selects a different pair of moving average types and periods from SMA, EMA, RMA, and HMA. A…
This US equity strategy builds a machine learning feature set from fundamentals, quality measures, price and volume behavior, technical indicators, securities data, and market signals. Its target is the following week's return. Fundamental inputs are…
The document describes a strategy that combines a moving-average crossover with a binary indicator based on whether consecutive bar ranges are positively correlated. It smooths that indicator and uses a threshold to permit long entries when price crosses…
The document presents a contrarian idea: take the opposite side of a machine learning system’s probability forecast derived from financial news. If its signal indicates a likely rise, the strategy sells; if it indicates a likely decline, it buys. The example…
The article explains a workflow for training a machine-learning model in Python, exporting it through ONNX, and running it from an MQL5 trading program. Its example is a regression model using open, high, and low prices as inputs and close price as the…
The article presents an MQL5 implementation path for Extralonger, a Transformer framework that combines temporal patterns with relationships across instruments. Its architecture is described as having temporal, spatial, and mixed branches: temporal attention…
This article generalizes a position sizer originally tied to one prop-firm challenge by moving program-specific terms into a configurable rule set. The rules cover phase targets, daily and overall loss limits, whether the daily limit expands with intraday…
This article introduces the Multifractal Model of Asset Returns (MMAR) as an alternative framework for volatility modeling, contrasting its construction with standard GARCH. GARCH estimates conditional variance from past shocks and variance, capturing…
The article presents Hamiltonian Monte Carlo (HMC) as a way to sample high-dimensional probability distributions more efficiently than random-walk MCMC. It explains the Hamiltonian formulation: parameters act as positions, auxiliary momentum is repeatedly…
This concluding article presents Extralonger, a neural-network design for modeling market data across time and relationships among instruments. Its architecture processes temporal, spatial, and combined information through parallel routes, then uses global…
This article describes interface improvements for an MQL5 system that sends prompts to an AI service. It adds animated loading indicators for request preparation and processing, elapsed-time feedback, and buttons to regenerate a response or export it to a…
The article outlines a MetaTrader 5 approach that selects currency pairs using Pearson correlation and cointegration tests, then combines a negatively correlated pair with a trading system framed around Nash equilibrium. It explains Nash equilibrium as a…
The article proposes moving parameter reoptimization into an Expert Advisor so it can adapt without stopping for repeated manual runs in the Strategy Tester. A genetic algorithm searches for inputs, initially for a selected strategy and later potentially for…
The article describes an MQL5 trading system organized as an ensemble of seven agents with different intended horizons and styles, including trend following, momentum, breakout, scalping, volatility, and swing trading. Each agent interacts with market data…
The article adapts lightweight motion-prediction methods from autonomous driving to market trajectory forecasting, seeking lower inference cost without sacrificing forecast quality. It describes encoding agents’ relative historical movements with an LSTM,…
This technical article examines how Scikit-learn regression models can be exported to ONNX and executed from MQL5, with a focus on differences between single- and double-precision calculations. It explains that some ONNX machine-learning operators accept…
The article introduces Hidformer, a Transformer-style architecture designed for time-series forecasting, and describes its dual-tower encoder. One encoder analyzes temporal patterns, while another works in the frequency domain to identify dependencies and…