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 summary examines ways to improve measures used to monitor crowding in Chinese equity factors. It focuses on pair correlation and factor volatility, two of four measure families previously tested alongside valuation spreads and long-term return…
The post proposes screening Chinese metaverse-related stocks using a price condition and historical revenue growth. It seeks shares whose opening price is near the 10-day moving average and whose 2021 revenue is more than 1.1 times 2018 revenue. It supplies…
This overview introduces SciPy as a Python library built on NumPy for scientific and numerical computing. It surveys modules used for linear algebra, optimization, integration, interpolation, probability and statistical analysis, signal processing, and…
This BigQuant overview summarizes strategy and implementation templates for Chinese equities and futures. Examples include moving-average pullbacks and crossovers, turtle breakouts, valuation and event-driven stock selection, a cointegration-based pairs…
This note outlines a Chinese stock screen based on price amplitude above 1, a rounded or arc-shaped price pattern, and an upward-sloping 30-day moving average. The author describes the amplitude filter as a way to seek stocks with trading activity, the…
This research proposes choosing between industry-level timing and within-industry stock selection based on how dispersed individual stocks are and how much the industry moves relative to the market. It classifies industries into timing-oriented and…
This brief platform support note explains a data limitation affecting backtests for Beijing Stock Exchange equities. It says the platform provides daily price data for these shares but has no minute-level data. As a result, a backtest that requests VWAP or…
This brief troubleshooting note addresses an index-out-of-bounds error during a BigQuant backtest. The explanation says the sorted input data begins on November 28, 2019, while the backtest requests stock data starting October 26, 2019. The proposed remedy…
The document describes a market-level stop rule that uses the Shanghai Composite Index as a risk signal. It proposes tracking the index’s cumulative return over five days and liquidating holdings if that return falls below −4%. The rule is intended to reduce…
This article explains how residual networks address degradation in very deep neural networks. In a plain network, adding layers can worsen accuracy even when gradients are managed through initialization and other techniques. A residual block instead learns a…
This overview introduces the Transformer architecture from the paper “Attention Is All You Need.” Earlier sequence-to-sequence systems commonly used recurrent or convolutional networks, often combined with attention. The Transformer instead relies on…
The document raises a question about two percentile-rank filters used to select stocks by total market capitalization in a dividend-yield strategy. One expression applies a threshold above the default percentile rank, while the alternative explicitly…
The post documents an attempt to move daily Chinese stock bars into a custom data source and use them as input to an automated labeling module. Its example queries a date-bounded range of stock records, adjusts instrument identifiers, writes the resulting…
The discussion explains a timing mismatch between backtests and live simulation when a strategy uses the current day's opening price. In the platform's backtest premarket handler, the user says same-day data can be accessed, allowing an opening-price rule to…
This article presents nine intraday chart patterns for interpreting a stock’s opening move and choosing whether to exit, reduce a position, or keep holding. Exit warnings include a weak prior-day limit-up attempt followed by another failed push, a new…
This report proposes treating effective factors and style factors differently when allocating across a multi-factor equity strategy. Effective factors rotate according to their relative strength, which the report seeks to explain using macroeconomic…
This China A-share strategy uses the stocks underlying all listed convertible bonds as its universe, on the premise that eligibility to issue a convertible bond helps screen for established companies. It ranks candidates equally on three cross-sectional…
This document summarizes a 2012 study that trained a deep convolutional neural network to classify high-resolution images into 1,000 categories. The model used convolutional and fully connected layers, GPU-based computation to speed training, and dropout to…
The document raises a data construction question for a rolling-trained deep learning model that predicts a stock’s next-day outcome from a fixed-length history. The dataset concatenates each stock’s chronological observations, then randomly selects training…
The document explains a convertible-bond rotation strategy that ranks bonds by absolute price and holds a selected group of the cheapest issues. Unlike a double-low approach, it does not use conversion premium as a selection variable. Its thesis depends on…
This article describes an A-share stock selection approach that uses XGBoost to combine Barra style factors, then examines whether selecting a smaller set of factors according to market conditions can improve the strategy. It compares three selection…
This Chinese-language support note explains why a BigQuant feature such as close_0 may not match the stock’s quoted historical closing price. The example compares a reported feature value of 134.5385 for 600030.SHA on 2021-04-01 with a stated actual close of…
This platform discussion addresses why repeated calculations can produce slightly different results despite using the same dates and data. It identifies missing values in the input table during neutralization as a possible cause: the neutralization routine…
The document briefly explains how to inspect a trained StockRanker model in BigQuant's strategy generator. After training, the platform can display feature importance and the model's tree branch structure. These views can help users see which inputs…