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 guide explains how to obtain and read monthly partitioned stock bar data at four intraday frequencies, then align the local compressed tables with corresponding cloud tables used for prediction. It describes the Feather file layout, recommends loading…
This summary describes a market-timing approach that uses a convolutional neural network to extract patterns from lower-dimensional index features, including technical indicators, and classify expected returns. Predicted return classes are then used to drive…
The document outlines a proposed system for automating A-share investment research reports. It describes combining company and industry information, financial analysis, valuation, price-chart analysis, and analysis of news and announcements. A LangGraph…
This QlibRL example describes how to configure and run a reinforcement learning workflow for executing orders in one asset. The training setup defines a simulator with 30-minute steps, a categorical action space, a full-history state representation, a…
This weekly Chinese equity research note reviews long-short screens built from high-frequency and technical factors. The listed signals include return skewness, downside volatility share, opening-period buying interest and large-order flows, reversal,…
This document describes a Chinese stock screen using amplitude above 1, a stated K-line value below 20, and prior-day actual turnover between 3% and 28%. It presents these conditions as a way to find stocks with notable price movement and moderate trading…
This Expert Advisor starts with a simple CCI rule: positive readings indicate buys and negative readings indicate sells. It adds three linear perceptrons—one for sell decisions, one for buy decisions, and one that combines their outputs—to override the base…
This forum post argues that AI should be treated as a tool rather than an autonomous source of trading intelligence. It emphasizes the difficulty of learning financial patterns from noisy data, the challenge of achieving stable profits even with…
This assignment describes building a stock strategy from four factors, including small market capitalization and turnover. The author used a provided template and AI assistance to implement a linear regression strategy, then packaged three models as options…
The document poses a quantitative modeling question: how to turn continuous data into discrete categories, then encode those categories as model features. It briefly explains one-hot encoding as representing each category with a vector that has a single…
This note outlines a workflow for creating an AI-assisted stock selection strategy. It recommends combining signals about the broad market, industry groups, and individual stocks rather than relying on a single factor. Example inputs include index returns…
The response explains how to structure a multi-stock, multi-date dataset for a multiple regression of stock returns on factors. Each stock on each date forms one observation: the return is the target variable, and that stock-date’s factor values are the…
This account describes a BigQuant assignment that applied linear regression and XGBoost within a rolling-training strategy, then compared their backtest returns. The author reports annual returns of 30% for linear regression and 52% for XGBoost, but provides…
The author discusses whether AI stock selection discovers repeatable market patterns or earns returns because its chosen style happens to suit current conditions. The reported live and backtest observations suggest that alpha varies with market regime:…
This FAQ explains practical design and troubleshooting points for FMZ Quant Workflow strategies. It covers host-version requirements, JavaScript-only code nodes, sequential execution, trigger behavior, reading data from connected parent nodes, and sharing…
This brief troubleshooting post presents a ValueError raised while a BigQuant workflow runs an automatic labeling module. The traceback ends in pandas' binning routine and reports that the generated bin edges are not unique: after negative and positive…
The post asks how to persist and reload a model produced by the newer BigQuant DAI stock-ranking workflow. It describes reading a model data source, saving it with pandas pickle, then loading it in a custom Python module and writing it back as a data source.…
This Expert Advisor uses four Accelerator Oscillator readings, sampled at the selected bar shift and three successive seven-bar intervals, as inputs to a weighted perceptron. It multiplies each reading by a weight derived from an optimized parameter, sums…
This stock screen combines three filters: daily amplitude above a threshold, a shrinking negative MACD histogram on a 15-minute interval, and positive indicators associated with beverage and alcohol imports and exports. The stated rationale is to seek…
This tutorial walks through a small Python implementation of the skip-gram model, a method for learning word vectors by predicting surrounding tokens from a center token. It represents input and output embeddings as separate matrices and uses a toy sentence…
This beginner’s reflection defines quantitative investing as identifying relationships between inputs, or factors, and future returns, checking those relationships against historical data, then using current observations to estimate future opportunities. It…
This article introduces convolutional neural networks and explains how one-dimensional convolutions can extract local patterns from financial time series. It describes convolution as applying learned weights across sequence windows, while pooling summarizes…
This introductory article explains the basic objects and notation of linear algebra used in machine learning and deep learning. It defines scalars, vectors, matrices, and higher-order tensors, with examples such as feature vectors, neural-network weights,…
This configuration describes a Qlib workflow for training a binary LightGBM model on one-minute CSI 300 data from China. It uses the Alpha158 feature handler, robust feature normalization, missing-value filling, and cross-sectional label ranking. The label…