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.
3,481 documents
This report develops a framework for rotating between the CSI 300 and ChiNext indices, whose valuations, market capitalizations, and industry exposures differ. It organizes potential drivers of their relative performance into factors covering real economic…
The document introduces MACD as a momentum and trend indicator derived from the difference between a short-period and a long-period exponential moving average. It describes the signal line as an EMA of MACD and the histogram as the difference between those…
This introductory overview explains machine learning as using historical data to fit a model and then applying that model to new cases. A scheduling example illustrates how prior observations can inform a prediction, while noting that a practical model may…
This beginner’s post describes an attempted reproduction of a quantitative research report’s “Water Drops Wear Away Stone” factor and a single-factor stock-selection backtest. The author says the implementation was prepared as a learning exercise with help…
The report estimates a fund’s equity exposure by regressing daily fund returns on daily returns for 29 first-level industry indices. It treats the sum of fitted industry coefficients as the estimated position. Because industry returns are correlated, it…
This weekly report describes two event-driven approaches to Chinese equities. The product-price strategy tracks listed companies’ main product prices and flags cases where a clear price increase has not yet been reflected in share prices. It is updated…
This forum post documents a failure in a stock strategy’s trailing-stop routine. The routine reads each position’s last sale date, subtracts it from the current date to estimate holding days, and uses that value as the bar count when requesting historical…
This brief Chinese-language page introduces a portfolio optimizer as a mathematical tool for selecting and weighting assets subject to constraints. It identifies three possible objectives: maximizing expected return, minimizing risk, or balancing the two.…
The document addresses a conflict between stop-loss logic and a machine-learning strategy's next-day buy list. A stock may be sold when it reaches a trailing stop, then selected again by the model on the following trading day. The proposed fix is to track…
This strategy note selects Chinese convertible bonds using low price, low conversion premium, and low remaining issue balance. It ranks eligible bonds by the sum of price and conversion premium, favors the lowest scores, and describes equal-weight holdings…
This short research note discusses how machine learning might classify stock price patterns. It uses a triple-bottom formation to show why chart labels can be ambiguous: after a decline and consolidation with three lows, an upward break may confirm a…
This document describes a Chinese equity strategy research approach that uses supervised neural networks to generate many weak alpha factors from a set of input factors. It proposes transforming those outputs into low-correlation components, then weighting…
This document describes a chart indicator that draws a price channel from the high and low of a candle on a selected higher timeframe. The user chooses the timeframe, and the indicator uses that period’s candle range as the channel boundaries on the chart.…
The author outlines an index-enhancement approach for Chinese large-cap and blue-chip equities. First, a human-selected universe emphasizes historically strong companies and stocks that held up against the broad market during declines. A machine-learning…
The response clarifies what a factor analysis does in the context of evaluating five candidate factors. It says the analysis is run on each factor individually, rather than testing whether a portfolio or combination of all five factors is effective. This is…
This forum post asks how a factor dashboard calculates win rates for low and high factor quantiles and for a long–short portfolio. It contrasts two possible interpretations: counting individual stocks with positive returns, or counting rebalancing periods in…
The document presents a stock-selection design that first ranks Chinese market sectors using a weighted combination of price momentum, money flows, and recent policy-related news. It then scores sub-sectors using reported net-profit growth and price…
This report summary compares cross-sectional, market-neutral commodity futures strategies based on inventory deviation and historical momentum. The inventory signal favors commodities with inventories below their own trend and shorts those above trend. It is…
This overview outlines a quantitative workflow: collect and clean data, develop a strategy, manage risk, backtest on historical data, and automate execution. It then sketches strategies for Chinese equities and futures, including Turtle-style breakouts,…
This forum post asks about a “Golden Cross Winner” ETF strategy that uses an unusual MACD-style calculation. The author questions why the signal line, DEA, is calculated as a 26-period exponential moving average of DIF instead of the commonly cited 9-period…
This introductory guide outlines a retail trader’s path into quantitative trading. It describes using market data and mathematical models to generate decisions, learning market concepts and a programming language, building models from historical data, and…
The article argues that a factor can be useful even when its strongest information lies in identifying poor-performing stocks. It describes ranking stocks into daily groups by factor value and interpreting unusually strong performance in the worst-ranked…
This study summary treats a company’s choice to capitalize or expense research and development costs as a signal about the expected future benefits of its projects. Because accounting criteria leave room for discretion, capitalization can also be used to…
This overview explains decision trees as nonparametric models that divide observations through sequential feature rules. Each path defines a group, and a leaf assigns a prediction based on the labels in that group. The same feature can be used at multiple…