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 introductory course overview presents cryptocurrency quantitative trading as an application of the same mathematical, statistical, and computational methods used in equities and futures. It describes building systematic models from historical data to…
This practitioner note argues that both feature selection and model choice shape quantitative strategy performance. It compares Random Forest, scikit-learn gradient boosting and random forests, and XGBoost, describing tradeoffs in robustness, speed,…
This brief project note describes using small market capitalization as a factor and training a model to relate that factor to subsequent 10-day returns. It presents an initial quantitative research idea: use a size characteristic as an input and forward…
This research tests whether lagged returns in economically related industries can help predict future industry returns, based on the idea that information and shocks may diffuse across sectors over time. To limit overfitting in a large predictor set, its…
This tutorial explains how a cross-sectional equity model’s predictions become portfolio decisions in VeighNa’s AlphaStrategy workflow. It distinguishes model evaluation from strategy design, then describes storing predictions in a dated signal table,…
The article describes selecting Chinese “monster stock” reversal candidates with a changing, multi-factor score. It examines recent profitable picks by industry and a normalized factor, then uses those observations to infer which sectors and factor ranges…
This research summary describes applying XGBoost, a gradient-boosted tree method, to Chinese stock selection and index enhancement. The model uses CART trees and classifies each stock into one of three groups formed by sorting market returns. This turns…
This research summary describes a method for extracting nonlinear relationships between equity style factors and returns. It keeps an interpretable linear factor model as a baseline, then trains machine learning models on the linear model’s residuals. The…
The article introduces K-nearest neighbors (KNN) classification using Euclidean distance: a sample is assigned the class most common among its nearest training examples. It then outlines a stock-selection application in which financial and technical factors…
This article introduces graph neural networks and focuses on graph attention networks (GATs) for financial time-series prediction. It explains that GATs use attention to weight neighboring nodes' features, allowing the weights to depend on node data rather…
The article discusses design choices for a hidden Markov model used to time exposure to a Chinese equity index. It proposes screening a broad set of candidate features, including technical indicators, with feature-selection or extraction methods such as PCA,…
The tutorial presents a workflow for modeling single-stock trading as a Markov decision process and training an agent with deep reinforcement learning. In its example, the agent trades Apple shares using an action space that permits buying, holding, or…
The discussion addresses how to feed minute-frequency market data into an AI stock-ranking workflow. Its answer is to replace the stock ranker template’s basic feature extraction step with direct data-source extraction, because minute data does not come with…
This research tests whether machine learning can estimate and reduce the optimism bias in analyst earnings forecasts. It uses 27 variables drawn from research reports, analyst characteristics, company fundamentals, and market information. LASSO provides a…
The article explains how classification trees make decisions by splitting observations according to features. It describes information gain and Gini impurity as criteria for choosing splits, and notes that information gain can favor features with many…
The document explains how cross-validation estimates whether a statistical or machine-learning model can predict data beyond its training sample. It describes ordinary k-fold validation, in which data is split into folds and each fold is used in turn for…
This configuration specifies a Qlib experiment using Alpha158 features for CSI 300 equities, with the Shanghai Composite 300 index as benchmark. It divides the history into training, validation, and test periods, applies robust feature normalization and…
This brief event description introduces a machine-learning strategy that updates as new data arrives. It focuses on rolling-window training: repeatedly using a moving segment of historical observations to refresh and optimize a model, with the aim of…
This guide explains how to configure a genetic programming module to search for trading factors from supplied market data and input features. Users choose a market and data frequency, return field and horizon, time splits, fitness measure and thresholds,…
The article describes how analytics, artificial intelligence, and machine learning may support cryptocurrency trading. It highlights searching large datasets for patterns, comparing markets, forecasting potential changes, and automating trading decisions. It…
A user reported a training error despite having a dataset described as containing 115,478 rows. The discussion identifies a possible cause: at least one factor had the same value across all samples. According to the reply, the older version of the stock…
This overview introduces LumiBot as a Python framework for rule-based, AI-assisted, and hybrid trading strategies. It describes a shared strategy lifecycle for historical backtests and broker runs, while emphasizing that the startup configuration must match…
A forum exchange discusses using a model built in MATLAB alongside VeighNa trading software. The proposed workflow is to send live market data from VeighNa to MATLAB, return MATLAB’s calculations, and have VeighNa execute trades based on those results. A…
This example describes embedding a pretrained ONNX classifier in an Expert Advisor to recognize digits drawn by a user. The model is trained for the MNIST task, which uses 28-by-28 grayscale images. A user draws a digit in an on-screen grid and activates…