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 Chinese research summary explores using data-mining models to time factor exposure and rotate among equity factors. It describes random forests conditioned on external market variables to assess when factors may be effective, alongside approaches that…
The note compares two strategy versions and argues that both feature quality and model choice matter. It uses a cooking analogy to convey that useful factors cannot fully compensate for a weak model, and a capable model still depends on informative inputs.…
The document introduces two regression methods used in financial analysis. Logistic regression models the probability of a binary outcome with values constrained between zero and one, then assigns a class by comparing that probability with a threshold. It…
This brief educational overview introduces machine learning and deep learning as tools for quantitative investing. It contrasts their potential use with conventional approaches built around financial time-series analysis, statistics, and economic principles.…
This Qlib guide explains how to connect a user-defined forecast model to the framework. A custom class subclasses Qlib’s model base and implements initialization, fitting, and prediction; the fit and prediction methods receive a dataset through the expected…
This document summarizes research on estimating KOSDAQ initial public offering stock values with artificial neural networks. The stated motivation is that multiple-based IPO valuation reflects sentiment in the market environment and may not reliably predict…
This platform support discussion explains a LightGBM training failure in a stock-selection workflow. The intended factor marks a day when the previous six closes were all below the five-day moving average and the current close moves above it. Inspection of…
This configuration sets up a Qlib experiment that uses a double-ensemble model built from gradient-boosted trees to rank CSI 300 stocks. The dataset uses Alpha158 features and divides the history into training, validation, and test segments. Model settings…
This document describes an MQL5 script and library workflow for importing third-party tick archives into MetaTrader 5 custom symbols. The script can refresh a local archive, synchronize data for one selected symbol or all available symbols, and create…
This Freqtrade strategy combines WaveTrend line crosses with Stochastic K and a difference filter. It enters long when the first WaveTrend line crosses above its signal line and all three indicator values fall within tunable ranges. It exits on the reverse…
This tutorial explains how a naive Bayes classifier uses observed features and prior class frequencies to estimate which category is most likely. It works through examples of classifying illness, online accounts, and gender, multiplying each feature’s…
This documentation describes a basic Transformer model for financial time-series prediction and how to configure and train it in a quantitative research framework. Inputs are represented as batches, time steps, and features. An embedding layer transforms…
This example explains a simple single-layer perceptron that combines several time-lagged readings of a smoothed RSI. Each selected reading is multiplied by a parameter-specific weight, and the weighted values are added to produce a raw score. The inputs can…
A user asks about an error in a shared quantitative strategy. The response points to a mismatch between the data universe and the requested features: index data does not provide certain return and price fields that are available for stocks, futures, or…
This brief BigQuant post reports a test combining rolling model training with LightGBM and an experimental factor described as “暗流涌动” (roughly, hidden currents). The author says the factor's standalone effect is not prominent, but it may contribute something…
This short BigQuant support note addresses a rolling-training workflow failure. It explains that a custom module was continuing to read data from an earlier module instead of receiving data that changed with the rolling time window. As a result, a later…
This research summary describes a method for grouping stocks with similar price behavior so that an investor can substitute one holding for another within a group. It builds on a grouping genetic algorithm, representing groups, stocks, and portfolio…
The article describes a workflow for finding S&P 500 healthcare companies that develop or use AI. It filters the index by industry, gathers recent company news, and uses a large language model to summarize products, technologies, and research activity.…
The article describes training BigQuant’s StockRanker machine-learning model using an overnight-return factor, following earlier factor research that links this signal to informed trading. Training uses all A-share stocks from 2013 through the start of 2021.…
This brief strategy entry describes an A-share stock-selection example built around a k-nearest neighbors classifier. The accompanying discussion clarifies that the target is binary: a stock is labeled positive when its five-day return is above zero and…
This brief discussion shows how to retrieve financial factor data through a data-access module using an SQL query. Its example selects dates, instrument identifiers, and a trailing-twelve-month operating-profit growth field from a Chinese stock…
This research overview explains how knowledge graphs represent entities and their relationships, then outlines a construction process: gather data, preprocess it, identify entities and links, reconcile information from different sources, form relationship…
The document describes an Expert Advisor that uses a single-layer neural network, identified as a perceptron, to classify the direction of quote movement. The implementation draws on the iAC indicator and bar opening-time history. It also reports a testing…
This configuration describes a Qlib workflow that trains a gated recurrent unit (GRU) model on Alpha158 features to rank CSI 300 stocks. It uses 20-step time-series samples, robust feature normalization, missing-value filling, and cross-sectional label…