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 tutorial describes using a recurrent LSTM network to forecast stock closing prices from recent price history. It forms rolling sequences from historical data, reshapes them into the three-dimensional input expected by the network, and separates each…
The author outlines a China A-share swing-trading strategy built around a manually selected pool of companies considered strong performers or quality blue chips. Relative performance against the broad market is used to identify stocks that remain resilient…
This post outlines a study of autoregressive (AR) and mean-reverting (MR) models applied to CSI 300 prices and returns. It proposes using the models to fit and forecast the index series, and says the article will explain the modeling workflow and compare…
This overview introduces machine learning as a way to learn patterns from data and use them to make predictions. It outlines a typical workflow: divide data into training, validation, and test sets, fit a model, evaluate it, then tune it before applying it…
The article introduces attention as a mechanism for focusing on relevant parts of an input sequence, then motivates Transformers by contrasting them with recurrent networks such as RNNs, LSTMs, and GRUs. Recurrent processing is described as limiting parallel…
This tutorial outlines a general PyTorch workflow for building and training neural networks. It explains how tabular or other input data can be converted to tensors, wrapped in a Dataset, and supplied in batches through a DataLoader. Custom datasets define…
This note discusses the relative roles of predictive factors and model design in quantitative strategies. It argues that factor quality usually sets the ceiling for potential performance, while model quality determines how effectively a strategy can use the…
The article describes an unsupervised approach to identifying market regimes from the returns of 17 US factors beginning in 1970. A Gaussian mixture model clusters observations by their return distributions; cross-validation is used to choose four clusters.…
The document summarizes a study applying convolutional neural networks (CNNs) to multi-factor stock selection in the Chinese A-share universe. It proposes arranging factor observations as two-dimensional inputs, treating convolution as factor combination,…
The report describes a daily Chinese equity selection model built with a fully connected deep neural network. It derives 98 features from seven basic price and volume inputs using rolling averages, extrema, dispersion, ranks, weighted averages, correlations,…
The overview presents FinRL-Meta as a framework for data-driven reinforcement learning in finance. It separates the workflow into data, market-environment, and agent layers, with interfaces that allow components to be replaced or customized. It also…
The document outlines a lesson comparing autoregressive (AR), moving-average (MA), and ARIMA time-series models using CSI 300 index prices and returns. Its stated aim is to explain the modeling workflow and try each approach for fitting and forecasting…
This platform support exchange distinguishes continuing training from a saved StockRanker model from saving a model for a full training run. The user wants incremental training, such as adding trees or increasing the number of leaf nodes, but reports an…
This overview introduces several major machine-learning approaches and explains their basic purposes. It contrasts linear regression for numeric prediction with logistic regression for classification, then describes neural networks as layered processing…
The document explains FreqAI’s main software components and how they support model development. A persistent model object handles data collection, feature engineering, training, and inference; a per-asset data kitchen provides processing tools and metadata;…
The notebook describes an evaluation workflow for stock-return prediction models, including linear and neural models as well as a Transformer and an approach combining predictors with a learned router. It ranks predictions cross-sectionally each day,…
The document presents factor-category weighting as a simple neural-network problem. Gradient-based optimization can learn both the weights within categories and the weights across categories from cross-sectional observations. Under certain conditions,…
This Qlib example demonstrates an end-to-end simulation of rolling online model workflows. It initializes Chinese-market data and experiment settings, generates rolling tasks at a configurable step, and connects those tasks to an online manager and a…
This overview explains how neural networks combine neuron types and connection patterns, then surveys architectures including feed-forward networks, Hopfield networks, Boltzmann machines, autoencoders, recurrent networks, convolutional networks, and others.…
This overview surveys a platform’s machine learning tools across classification, regression, ranking, and clustering. It names representative methods such as support vector machines, neural networks, gradient boosting, random forests, nearest neighbors, and…
The environment layer in FinRL-Meta uses cleaned data to create market simulations with a shared, Gym-style interface. Users can build on these environments and compare strategies across a common framework. The document describes account options for margin…
This module describes a candidate-selection process for pairs trading based on dimensionality reduction and clustering. It starts from a panel of asset prices, converts prices to returns, standardizes them, and applies principal component analysis to create…
The document explains how to inspect trained parameters in a Keras model that combines an LSTM layer, an auxiliary input, and several dense layers. It distinguishes trainable variables from the arrays returned by the model’s weight retrieval method, then…
This short forum exchange addresses an error in a regression-evaluation workflow that reports a missing label. The response says the problem cannot be diagnosed precisely without details about the task, then recommends first defining what quantity the model…