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 article presents quantitative trading as a tool increasingly used by professional and institutional market participants. It argues that rule-based models can complement fundamental and technical analysis by screening many stocks against user-defined…
This brief exchange explains how to interpret a feature-importance chart when one or more short-horizon factors receive a score of zero. The response says the chart ranks factors within the particular set used to train a particular model. A low-ranked factor…
The document gives a brief historical and conceptual overview of reinforcement learning. It starts with Markov decision processes and dynamic programming, then describes temporal-difference learning and Monte Carlo methods as ways to estimate value without…
This Chinese-language case study describes using an AI coding assistant to reproduce and modify a high-frequency volume factor for Chinese equities. The source factor uses intraday volume’s Fourier spectrum to measure energy in a short-period band, with…
This introductory overview surveys regression, neural networks, support vector machines, clustering, dimensionality reduction, and recommendation methods. It distinguishes numerical prediction from classification, explains how logistic regression turns a…
This configuration specifies a Qlib workflow for a TabNet model using the Alpha360 feature handler and CSI 300 instruments, with the index as benchmark. It sets a two-day-ahead close-price return label, applies robust normalization and missing-value filling…
The document summarizes a 2019 study that combines two representations of the same stock data to predict prices: numerical time series and images of stock charts. An LSTM extracts temporal patterns from the sequence data, while a CNN extracts visual patterns…
This research summary describes a method for studying whether public information-seeking behavior can reveal topics associated with subsequent stock market moves. The cited 2014 study analyzes historical data from 2004 through 2012, drawing on Google and…
This workflow note explains how to prepare data when training models across rolling windows. As each window advances, the training sample changes, and learned processor state—such as means and standard deviations—must be recalculated for that window. Reusing…
The document explains how to remove stocks from a model’s predicted buy signals when they fail a chosen condition. Its example filters out stocks whose return_5 value is below 1.05, so a signal containing three names could proceed with only the two that…
This overview describes quantitative investing as using mathematical models, statistical analysis, and computer algorithms to guide investment decisions systematically. It surveys time-series and regression analysis, factor models, automated execution, large…
This BigQuant question and answer explains the distinction between labels and input features in a supervised machine-learning workflow. A label supplies the target value that a model learns to predict, while the input feature list is used to produce the…
The strategy turns conventional stock-selection rules into Boolean features for a CNN classifier. Its inputs include returns and return ranks, turnover, valuation and profitability measures, money-flow signals, volatility, technical indicators, and recent…
This strategy combines machine-learning stock ranking with trend-based entry filters for a manually selected pool of China Securities 150 constituents. The pool emphasizes larger, liquid companies, often described as core or sector leaders. A model scores…
The document explains a portfolio construction approach that forecasts stock trading volume and uses the forecasts as a proxy for expected price impact. With trade size held fixed, lower volume implies a higher participation rate and greater expected impact,…
The document introduces multi-task learning (MTL), in which one model addresses several related tasks and uses shared patterns and task-specific differences. It suggests that this can improve learning efficiency and predictive accuracy compared with training…
This configuration specifies a Qlib workflow for training a CatBoost model on Alpha158 features for the CSI 500 universe. The data spans 2008 through mid-2020, with training through 2014, validation during 2015–2016, and testing from 2017 onward. The model…
This forum exchange explains how to guide an AI assistant when generating VeighNa CTA strategy code. It recommends supplying a focused set of references instead of the entire codebase: the CTA base class, the ArrayManager indicator API, and one or two…
The document describes an Expert Advisor that combines moving-average changes from hourly, four-hour, and daily charts. Each timeframe contributes three relative changes between recent moving-average values, weighted by adjustable parameters and rounded…
This exchange asks how to make a deep-learning model produce two predictions for futures trading, one for a long view and one for a short view. The response explains that producing multiple outputs for a regression or classification model may require a…
This document describes an Expert Advisor that predicts a future price from a sequence of past prices using linear prediction. Burg’s method estimates model coefficients by reducing mean-square prediction error over a training window. The strategy’s…
This forum post documents a BigQuant workflow in which LightGBM finishes running, but the subsequent score-to-position step fails because its query orders predictions by a column named score that is absent from the input dataset. The error identifies the…
This configuration describes a Qlib workflow for training a Localformer model on China A-share data and evaluating its predictions as a portfolio signal. It uses the CSI 300 universe and benchmark, an Alpha360 data handler, robust feature normalization,…
This Chinese-language support exchange asks why BigQuant’s DNN template does not show a separate standardization module and how to handle an error after adding one. The response explains that the DNN is implemented as code and that data processing, including…