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 course-lecture summary introduces the role of planning and learned models in reinforcement learning. It identifies Dyna and Monte Carlo tree search as examples of algorithms that use models, and notes that the lecture is presented by research engineer…
This alert tool monitors an indicator that compares linear, quadratic, logarithmic, and exponential regressions, then selects the model that best fits the analyzed data. It notifies the trader when that selection changes, including when the fitted regression…
This research summary examines whether mutual fund managers’ tendency to follow prior institutional trading can reveal investment skill. It describes a fund-level dynamic herding measure, FH, and tests whether funds with stronger herding subsequently…
Spectr is described as an indicator that measures the oscillation spectrum of a financial asset and normalizes it using a simple moving average. This gives a brief conceptual description of how the indicator relates price movement to a moving average, but…
This note introduces fractal dimension as a way to describe how price patterns retain similar roughness across chart timeframes. It explains the concept through covering a pattern with objects of different sizes: the relationship between the number and scale…
This note describes a Chinese equity screen requiring turnover between 3% and 12%, year-over-year growth in net profit attributable to parent-company shareholders above 20% and no more than 100%, and exclusion of STAR Market stocks. The accompanying example…
This report examines stocks that mutual funds hold at a larger portfolio weight than their benchmark index. It describes these holdings as tending toward larger, higher-valued companies and certain sectors, and reports that many outperformed during fund…
This document introduces the cyclically adjusted price-to-earnings ratio, or CAPE, as a way to compare equity prices with a decade of inflation-adjusted earnings. It summarizes historical analyses of the S&P 500 and Shanghai Composite, reporting that higher…
The document explains the cyclically adjusted price-to-earnings ratio (CAPE), calculated as an inflation-adjusted share price divided by the average inflation-adjusted earnings per share over the prior decade. It compares CAPE with future real returns for…
The article describes a stock screen that excludes Beijing-listed shares, selects stocks with a daily high-low range above one percent, and applies a filter based on the prior day's turnover and the ratio of current opening-auction volume to the previous…
This brief article presents ten questions about common assumptions in quantitative research. They concern whether researchers must be highly skilled in statistics, whether borrowed methods can be adopted without adaptation, and whether more variables, more…
The document explains a Donchian Percentage oscillator that expresses the closing price as a percentage of its position between the channel’s lower and upper bounds. The bounds are the lowest low and highest high over a configurable lookback period. A…
The stated screen looks for stocks with turnover between 3% and 12%, a listing year of 2021, and ownership concentration below 20%. The article treats lower concentration as a sign of more dispersed ownership and suggests that the turnover band may help…
The screen selects stocks with turnover between 3% and 12%, beginning with the Chinese stock-code prefix 60, and keeps companies whose 2021 revenue exceeded their 2018 revenue by more than 10%. The article describes this as a way to combine trading activity…
This stock screen ranks companies by a measure of capital strength, such as net inflows or trading value, and requires at least two limit-up sessions within a 500-day window. The article frames the screen as targeting stocks with strong buying interest and…
This Chinese-language post describes a short-term stock screen using three conditions: a 14-period RSI below 65, first-level buy volume greater than first-level sell volume, and prior-day trading value above 60 million. It frames the RSI as a price-condition…
This Chinese-language research summary examines fund performance, subscriptions and redemptions, manager concentration, and sector returns around the first half of 2020. It reports gains for three categories of active equity funds and compares the scale of…
This post summarizes a 2017 J.P. Morgan report on using large and alternative datasets in investment research. It groups alternative data into information generated by individuals, business processes, and sensors, with examples such as social media,…
This forum post reports that a strategy based on a platform template runs, but simulation fails at a dropna step because of NaN values after the user adds factors. The author’s attempted fix was to fetch 90 calendar days of earlier daily index data,…
The document explains the price-to-book ratio as a valuation measure calculated by dividing a stock’s market price by book value per share. It interprets lower ratios as possible signs of undervaluation or pessimistic expectations, while higher ratios can…
The post reports a discrepancy between a backtesting platform’s monthly return for the CSI 300 benchmark and an external reference calculation. It attributes the difference to the platform using the first day of the month’s closing price as the starting…
This short discussion outlines data-related challenges faced by quantitative investment teams as they develop and test strategies. It describes quantitative investing as turning historical market behavior into data, using statistics and programming to…
This document proposes a Chinese stock-selection rule that combines a 14-period RSI below a stated threshold, displayed first-level buy volume greater than sell volume, and year-over-year growth in net profit attributable to parent-company shareholders…
This article introduces Monte Carlo methods through random sampling examples, contrasting an approach that can return a promising answer without guaranteeing the optimum with randomized search that keeps trying until it finds a valid solution. It illustrates…