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.
16,761 documents
This code manages backtest outputs for momentum experiments. It reads results from multiple train and test intervals, aggregates captured returns, and can rescale those returns to a target volatility. It calculates performance summaries that include return,…
This research summary addresses how to combine highly correlated stock-selection factors when there are too many inputs to orthogonalize efficiently or when orthogonalization fails to retain their shared information. It describes reducing each factor group…
This example describes a daily portfolio rebalancer for five exchange-traded funds spanning US equities, long-term Treasury bonds, and gold. It starts by checking each holding’s weight against a target allocation. A rebalance is triggered on the first…
The screen selects stocks with a five-day average turnover rate between 3% and 12%, displayed first-level bid volume greater than first-level ask volume, and a closing price above the previous session’s low. The stated rationale combines trading activity and…
The report describes applying Thompson Sampling, an online learning method, to adjust a multi-factor investment model as market leadership changes. It uses China A-share constituents from the CSI 300 and CSI 500, with valuation, profitability, growth,…
The report compares sparse covariance estimation, which sets selected off-diagonal elements to zero, with rotation-invariant methods that adjust sample covariance eigenvalues. The latter include simulation-based, linear, and random-matrix-based nonlinear…
This note describes a simple Chinese equity screen: select stocks with turnover rates between 3% and 12%, codes beginning with 60, and a positive recent return. It frames the turnover band as a liquidity filter, the code prefix as a way to restrict the stock…
This notebook explores how several market series relate to the Russian ruble exchange rate against the Australian dollar. It fits separate ordinary least squares regressions for each candidate regressor within each year, then repeats the analysis…
This article surveys Microsoft Research’s published work on AI for quantitative investing, especially Chinese equity selection. It explains methods that use graph neural networks to represent relationships among stocks, attention mechanisms to adapt to…
This Chinese-language post describes a daily stock screen for companies in the metaverse industry. It selects stocks when the KDJ indicator has just formed a bullish crossover and at least one limit-up session occurred within the prior 25 trading days. The…
This report studies trend-based allocation across CITIC Securities’ first-level industry indexes. It frames the process as choosing assets, measuring momentum, selecting industries, assigning weights, and managing portfolio tracking. The analysis describes…
This proposed equity screen looks for stocks with amplitude above one percent, a control measure above 21, and seven consecutive down days. Its refined version also requires circulating market capitalization of at least 3 billion. Example implementations are…
This stock selection proposal combines a daily amplitude threshold, a recent reversal candle pattern, and a measure described as today’s control rate. Its refined rule requires amplitude above one percent, a reversal pattern within the past three days, and a…
This brief exchange explains an apparent one-day mismatch between backtest output logs and the trade details table. The stated cause is the difference between a signal and its execution: the log records when a price bar arrives and triggers the signal, while…
The document describes an MQL5 Wizard Expert Advisor that combines MACD signals with the Parabolic SAR trend indicator. The system evaluates signals when a new bar appears and assigns the two indicators different weights: MACD receives the larger weight in…
The document summarizes a strategy that combines the relative strength index, exponential moving average, volume-weighted average price, and moving average convergence divergence to generate trading signals. It describes this as a “Weapon Candle” approach…
The document describes three ways to refine spread trading signals. A threshold filter enters or maintains a long or short spread position only when the predicted spread change crosses a chosen boundary; an asymmetric version allows different boundaries for…
The article explains why a strategy can make money without winning more often than a coin toss. It contrasts high-win-rate approaches, which seek frequent small gains, with high reward-to-risk approaches, which tolerate more losing trades in pursuit of…
This Chinese stock-selection example screens for turnover between 3% and 12%, circulating market capitalization from 1 to 55 hundred million yuan, and a KDJ K value below 20. It presents the criteria as a way to select stocks, then gives example expressions…
This short forum exchange reports a TypeError encountered when opening a backtesting module. The response attributes it to a strategy parameter-name list containing an item that is not a string, which prevents the framework from using that item as an…
A user reports that a price-ratio buy condition never triggers during a year-long backtest, although the sell condition does. The response attributes the training error to StockRanker receiving only two factors, with a Boolean factor causing the model to…
This indicator description explains a ZigZag-based method for marking potential swing highs and lows at four evaluation lengths. It places colored, nested symbols at detected turning points instead of drawing connecting lines. The four user-configurable…
The document describes an Expert Advisor that combines Bollinger Bands with the Donchian Channel. A sell setup requires a falling Donchian Channel, an open above the Bollinger upper band, and a close below it. A buy setup requires a rising channel, an open…
This Freqtrade strategy uses 15-minute candles to enter long positions when price falls below a selected lower Bollinger Band and to exit above a selected upper band. The bands use a 20-period window with four standard-deviation settings. Optional guards can…