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 brief Chinese-language support note addresses how to use factors produced by a genetic factor-mining process. It says the discovered factor has an expression, but that a user must convert the expression manually before sending it to a factor analysis…
This configuration defines a Qlib workflow for training a Temporal Fusion Transformer model on Alpha158 features for CSI 300 stocks. It sets Chinese market data from 2008 through mid-2020, using 2008–2014 for training, 2015–2016 for validation, and 2017–2020…
This Qlib documentation describes visual reports for evaluating intraday portfolios and prediction models. Portfolio reports display benchmark and portfolio cumulative returns, returns with and without transaction costs, turnover, drawdowns, and cumulative…
This tutorial compares three ways to train an XGBoost model for stock selection: ranking securities by a score, classifying outcomes into categories, and predicting a numeric target through regression. It frames these choices within a broader modeling…
This tutorial compares three ways to train an XGBoost model for stock selection: ranking securities by a score, classifying outcomes into categories, and predicting a numeric target through regression. It frames these choices within a broader modeling…
The document summarizes research on forecasting multiple future steps from limit order book data. Rather than predicting only one future point, the proposed approach uses sequence-to-sequence encoder-decoder networks with attention to generate a path of…
This paper description presents a learnable scheduler for sequence-learning problems with related prediction tasks, such as forecasting returns at different future horizons. During training, the scheduler chooses an auxiliary task based on the current model…
This forum post presents a workflow for combining predictions from three model outputs. It merges the datasets on instrument and date, preserves columns that are not already present, renames each model’s prediction column, and computes their arithmetic mean…
This guide describes ways to organize AI agents inside a trading strategy, from a single analyst to specialist research teams, opposing bull and bear views, and sequential debate. It distinguishes deterministic strategies, agent-led decisions, and hybrid…
The article tests whether a convolutional neural network can classify stock direction from chart-like images generated from OHLC data. Each sample uses 32 time steps, normalized to a 128-by-128 binary image: groups of columns mark open, high-low range, and…
This forum post raises an implementation question about deploying BigQuant StockRanker models for live trading through a brokerage server. The author believes StockRanker includes a gradient boosting decision tree model and asks whether deployment transfers…
This article outlines a machine-learning stock selection strategy intended to find shares that may rebound after declines while limiting drawdowns during weak market conditions. It targets China’s small and medium-sized board, chosen for its activity and…
This guide explains how to participate in a BigQuant quantitative challenge using A-share minute bars and order-book snapshots to predict future 30-minute VWAP returns. It covers the factor-mining and end-to-end modeling tracks, available templates and data…
This discussion examines whether the length of a model’s training window changes an AI strategy’s results. It describes manually rolling training for a visual template strategy, comparing longer histories of five to ten years with shorter windows ranging…
The document describes Temporal Routing Adaptor (TRA), a way to extend a stock prediction model so it can learn from different patterns in market data. It notes that momentum and reversal behavior may coexist, which challenges the assumption that…
Qlib separates forecasting signals from portfolio construction. A strategy turns prediction scores into trading decisions, while a weight-based base class lets users specify target holdings and delegates order generation to the framework. The documented…
This example describes a concentrated long-only stock portfolio built through a sequence of AI agents. A research agent ranks companies for understandable businesses, cash generation, and attractive prices. A second agent challenges each idea by examining…
This Chinese course listing outlines a study of A-share stocks that reach their daily upper price limit. Its stated sequence is to explain the limit-up mechanism, classify limit-up events, examine subsequent stock returns, and then apply a support vector…
The document describes a basic workflow for evaluating a trained quantitative model. After fitting the model on training data, apply it to a validation set, then compare its predictions with the observed values to assess performance. This separates model…
This podcast recap discusses how AI agents may interact with crypto assets and decentralized applications, alongside a vision for regulated DeFi that connects conventional banking with self-custodied digital assets. The guest describes agents as systems that…
This expert-advisor design turns four RSI readings into a single weighted perceptron score. It uses RSI periods of 12, 36, 108, and 324, rescales each indicator around zero, and combines them with weights selected through optimization. The trading threshold…
This research summary explains how to build a machine-learning stock-selection process using historical factor values to predict subsequent returns. In the training stage, a supervised model learns the relationship between inputs and returns; in the testing…
This MQL5 demonstration illustrates supervised classification with a support vector machine (SVM), using a fictional animal-recognition task to explain labeled examples and learned decision boundaries. It generates seven-feature observations with rule-based…
This helper prepares spread changes and their lagged values as inputs for a regression model. It can expand the lag features with pairwise products, split a chosen in-sample period into ordered training and test sets, and keep a separate out-of-sample…