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.
21,023 documents
This post describes a short-term stock screen combining three prior-day conditions: price amplitude above 1%, appearance on the market’s named trading list, and turnover above 8%. The rationale is to find shares with notable movement and active trading,…
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 Chinese-language course description presents limit-up stock counts as a way to identify active areas of the China equity market. It proposes comparing counts across market capitalization, sectors, and industries, using recent industry counts to locate…
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 note describes a screen for A-shares whose codes begin with 60, whose turnover is between 3% and 12%, and whose top 20% of shareholders collectively hold less than 70% of shares. It presents these conditions as filters for liquidity and ownership…
This report reviews Chinese public quantitative funds and private quantitative market-neutral strategies using observations available through March 2022. It groups public funds into market-neutral products, CSI 300 enhanced funds, and CSI 500 enhanced funds,…
The document introduces the DecisionPoint Price Momentum Oscillator, originally developed by Carl Swenlin, and describes a comparison approach attributed to Vitali Apirine. The method applies the oscillator to compare the relative momentum of two securities.…
The document presents Benford’s law as a quantitative screening method for assessing whether company financial figures may have been manipulated. It explains that in many naturally occurring datasets, the first nonzero digit appears with a nonuniform…
This page identifies itself as a reproduction of a Chinese multi-factor stock-selection research report about information contained in moments of unusually high trading volume. It also references related research on converting intraday minute-level spike…
This study examines whether Chinese fund management companies’ visits to listed firms are associated with the performance of the funds they manage. It proposes that visits can improve investment decisions by giving institutions access to operational details…
This stock-selection proposal screens listed Shenzhen stocks using a turnover range of 3% to 12%, positive price-to-earnings ratios below 29.01, price-to-book ratios below 3.11, and a reversal-style price shape. Its reference implementation represents that…
Standard Error Bands form a channel around a smoothed linear-regression estimate of price. The center is a linear regression calculated over a chosen period and then smoothed with a simple moving average. Upper and lower bands are placed at a configurable…
This forum post describes an attempt to compare a platform’s single-factor analysis module with the open-source Alphalens package. The author uses the CSI 300 universe and the PE_TTM_0 factor, with a 22-day rebalancing interval. They aim to prepare…
The Efficiency Ratio measures how directly price moved over a period by comparing the net price change with the total absolute movement along the way. Its values range from zero to one: higher readings indicate a more efficient, directional path, while lower…
This introductory survey organizes multi-agent reinforcement learning (MARL) by the relationships among agents’ objectives: fully cooperative, fully competitive, and mixed cooperation and competition. It explains that cooperative agents can be modeled…
This document describes a channel indicator for identifying potential countertrend entries. It calculates the highest and lowest prices over a configurable lookback range, uses their midpoint as the channel center, and plots a simple moving average of…
This forum exchange addresses a question about changing historical EMA readings and repeated signals in a VeighNa CTA strategy using ArrayManager. The answer explains EMA as a recursive indicator: each new bar updates the current value using the latest price…
This equity-factor study compares several ways to measure illiquidity and examines whether less liquid stocks earn higher subsequent returns. Its microstructure measures are quoted spread, effective spread, and price-impact lambda; it also tests Amihud’s…
The document explains how to construct a DMA feature as the difference between a short-period and a long-period simple moving average. It identifies three inputs: the underlying price or feature, the short window, and the long window. Its worked example uses…
This short data-support exchange explains a discrepancy between stock prices shown by two Chinese market data services. The response attributes the difference to BigQuant providing backward-adjusted prices and notes that providers can use different…
This indicator plots recent fractal high and low levels on a price chart. It tracks candidate highs and lows, then updates the plotted level after subsequent price movements confirm the candidate. These levels can help traders identify local turning points…
The webinar summary addresses how to estimate option values in newer crypto markets where liquid options data may be sparse or absent. It describes Amberdata’s approach as using liquid equity options as a reference for generating implied volatility surfaces,…
A trader asks why MACD golden-cross detection sometimes disagrees with a charting exchange during five-minute backtests. The comparison covers both simulated and live tickers over a two-day period. The trader reports that some cross signals were incorrectly…
This page introduces the exploration–exploitation problem in reinforcement learning: a learning agent must decide when to test actions that may reveal useful information and when to choose actions based on what it has learned so far. The lecture is…