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 research overview presents a workflow for building stock-selection signals from daily price and volume data. Genetic programming generates candidate factors, using rank information coefficients to target linear relationships or mutual information to…
This Chinese-language forum post raises practical questions about what factor weights from a labeling or training process represent. It distinguishes possible interpretations, including neural-network parameters, regression coefficients, and tree-based…
This overview introduces autoregressive and moving-average models for representing serial dependence in financial time series. An ARMA model combines lagged observations with lagged errors and assumes stationarity; the article notes that differencing or…
The article describes a supervised learning approach that classifies whether a stock’s closing price will be higher or lower after a chosen horizon. It smooths historical prices with exponential weighting, then uses six technical indicators—including RSI,…
This tutorial explains XGBoost through the broader framework of supervised learning and gradient-boosted decision trees. It introduces loss functions for regression and classification, then describes an ensemble as the sum of tree outputs. Training adds one…
This Chinese-language listing introduces a talk by the CEO of quantitative investment firm High-Flyer. The described presentation surveys what artificial intelligence is, how it has developed and been applied, examples of AI in investment, the firm’s work in…
This introductory overview explains neural networks through the progression from a single perceptron to deeper architectures. It describes weighted inputs, activation functions, loss minimization, and gradient descent, noting how learning rate affects…
This short note raises a basic question about what a probability estimated by logistic regression represents. It contrasts interpreting a prediction as the chance that one particular person succeeds with interpreting it as the expected success rate among…
This article presents Almanax’s approach to preventing blockchain exploits through AI-assisted code review and continuous security monitoring. The described system uses language models to examine code logic and potential attack surfaces, and is intended to…
This overview explains feature engineering as the work of turning raw market and company data into useful inputs for quantitative strategies and machine-learning models. It uses a dual moving-average crossover to show that a trading rule can be understood…
The document introduces the challenge of forecasting intraday volatility across financial assets. It notes that researchers and practitioners have developed many approaches, including intraday estimators, GARCH models, and continuous-time models. Volatility…
This brief discussion asks whether a strategy’s return calculation horizon should match its holding period. The response says the answer depends on the trading logic, but that the two are generally kept consistent so a backtest represents the intended trade.…
The page introduces FinRL as a framework for applying deep reinforcement learning to automated stock trading. It explains that reinforcement learning agents learn through interaction and trial and error, while deep neural networks approximate the functions…
The document describes preparing supervised-learning labels from Chinese daily stock bars. It calculates a forward price ratio using lead values, clips observations to the cross-sectional 1st and 99th percentile boundaries, and assigns the result as the…
This report frames next-day direction in China’s Wind All A-share index as a binary classification task. It trains a decision-tree-based AdaBoost model on 51 daily features drawn from several markets and asset classes, including repo rates, credit spreads,…
This example introduces support vector machines through a hypothetical animal-classification task. It defines seven numeric features and labels an observation positive when every feature falls within a specified range. Randomly generated observations and…
This configuration defines a Qlib experiment using Alpha158 features for the CSI 300 and the Shanghai 300 index as benchmark. It normalizes features with robust z-scores, fills missing values, filters a selected subset of columns, and applies cross-sectional…
These release notes describe changes across versions of the VeighNa trading framework. For quantitative research, notable updates include a cross-sectional percentile ranking function, revised factor and signal performance analysis, an added VWAP matching…
This shared open-source example builds a daily classifier for Chinese stocks using price and volume data, technical indicators, and three supervised models: a random forest, logistic regression, and a linear support vector classifier. Features include…
The report describes applying OpenFE, an enumerate-and-filter framework, to discover interpretable fundamental factors from Chinese companies’ balance sheets, income statements, and cash-flow statements. It first removes fields with substantial missing data,…
This study proposes a market mechanism for trading data products by separating data ownership from the right to use the data. Ownership is divided into tradable units to support liquidity, while use rights remain indivisible and go to an end user. The…
The document introduces Zhice, a strategy research workspace within VeighNa Fusion. It describes a staged workflow that takes a user's trading idea through clarification of the logic, code drafting and review, backtesting, parameter optimization, and results…
This Chinese-language post presents a stock screen combining at least five overlapping moving averages, a 2021 date condition, and a shortening MACD histogram on a 15-minute chart. It interprets converging averages as a sign of price stability and the…
This script creates order labels from per-stock historical backtest data, then saves the resulting records in training, validation, test, and combined datasets. It rejects a stock’s data if it is empty, contains missing values, or has very small minimum…