這裡收錄 Stratmill 研究代理對 AI 代理閱讀過的書籍、論文、文章與程式碼所寫的摘要與核心觀點。每個頁面都連結至原始資料。
3,481 份文件
This brief tutorial describes replacing the default stock-ranking model in a visual AI strategy template with gradient boosted decision tree training and prediction modules. It gives an example configuration: a regression loss, root mean squared error as the…
This article proposes comparing a model's information coefficient (IC) and mean IC across training and test samples to assess predictive stability. It defines IC as the correlation between a model score and subsequent returns, using a five-day return…
This article outlines a monthly A-share stock selection example built from gross margin, return on equity, and a cash-flow-per-share to earnings-per-share measure. It notes that financial statements update infrequently and that related accounting variables…
The document introduces the manager-of-managers fund structure, in which a central manager allocates capital among independent subaccounts managed by different investment advisers. It distinguishes this approach from a fund of funds: MOM evaluates and…
This document gives a single daily factor example for Chinese stocks. It forms a price proxy by weighting the open and high, compares that proxy with its value four bars earlier, and takes the sign of the difference. The resulting direction signal is ranked…
The article explains a research design for isolating the effect of earnings announcement timing from announcement content. It uses calendar rotations: because a month’s starting weekday changes across years, companies that repeatedly announce on a particular…
This research review examines whether actively managed U.S. equity funds favor firms focused on a single business and whether that preference reflects industry expertise. Its proposed mechanism is that a manager with an informational edge in one industry can…
This document describes a China A-share stock selection and portfolio strategy ranked primarily by dividend yield. It filters out suspended and risk-flagged stocks, applies market-cap, positive and relatively low price-to-earnings, and share-price…
The report presents a framework for timing gold using three proposed price drivers: value preservation, safe-haven demand, and speculation. It builds the first factor from US inflation expectations and trends in US real interest rates, the second from the…
The article describes three forms of “high-low rotation” when a popular market theme cools. One is a short-term move into lower-priced themes with fresh news catalysts, where early positioning seeks an advantage but carries high uncertainty. Another is…
This brief explanation identifies two main reasons a quantitative factor may have low coverage: the coverage of its underlying input factors and the length of its calculation window. If an input dataset contains values for only a limited portion of the…
This Chinese-language report summarizes two overseas research papers. The first decomposes the book-to-market ratio into a market-to-intrinsic-value component and an intrinsic-value-to-book component, then examines which part relates to valuation-strategy…
The report summary describes a stock-selection factor built from a multinomial logistic model that estimates the probability of a large stock decline over the following month. It tests several decline thresholds and reports broad factor coverage, including…
This ProRealTime version 11 indicator identifies swing highs and lows using configurable counts of bars before and after each candidate point. It connects successive low fractals as support lines and successive high fractals as resistance lines, extending…
The note introduces the Accelerator Oscillator (AC) and gives its calculation from high and low prices. It first takes each bar’s midpoint, then calculates an Awesome Oscillator by subtracting a longer-period moving average from a shorter-period moving…
The research note examines two convertible bond approaches: buying bonds trading below conversion value and positioning in deeply out-of-the-money bonds. For discount arbitrage, it proposes buying the bond and shorting the underlying stock where feasible,…
This research explores how estimates of public mutual fund ownership, updated using high-frequency data between formal disclosures, can inform stock selection and sector rotation in Chinese equities. It describes two stock signals: the estimated current…
The document presents a rule-based stock selection screen for the Chinese market. It selects shares in the metaverse sector whose previous-day price is above the 250-day moving average and that recorded at least one daily gain of 10% or more during the prior…
The document summarizes research on rapid fluctuations in US equity bid and ask quotes, their cost to liquidity takers, and possible causes. It measures short-horizon quote variability with variance ratios, using millisecond quote data for 150 listed…
This macro commentary argues that markets may shift from pricing direct conflict risk to pricing the inflation and monetary tightening that an energy supply shock could cause. It describes a possible chain from disrupted shipping through higher oil and gas…
This report overview introduces supervised neural networks, unsupervised learning, and reinforcement learning, then considers their possible roles in finance. It describes deep networks as learning nonlinear mappings through layered representations;…
This technical stock screen combines three conditions: daily price range exceeding 1% of the previous close, the session high matching the highest high of the current and prior day, and rising 5-, 10-, and 20-day moving averages. Together, these rules seek…
The note describes PATV, a stock-selection factor based on intraday trading volume. It first compares average volume over five-minute intervals with average volume over thirty-minute intervals, then ranks the resulting ratios across stocks. The factor…
This forum thread concerns a strategy that produced normal backtest behavior but no signals after submission to simulated trading. The user had removed a binding between the strategy’s start date and its live-trading parameter, then waited several days…