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.
5,018 documents
This index introduces latent factors as shared patterns extracted from the way stocks’ returns move together, with each stock’s loading indicating its exposure to a pattern. It contrasts principal component analysis, which estimates factors and stock…
This analysis explains how to read information coefficients for a catalog of CME futures prediction models. It defines each date’s IC as the rank correlation between predicted and realized returns across products, then averages those daily values across…
This notebook runs three otherwise identical research agents on a recession question, either by replaying a saved capture or by running live models and search in parallel. It records each agent’s probability, confidence, evidence-gathering activity, and…
This notebook uses TreeSHAP to explain LightGBM return predictions, with global feature importance, individual prediction explanations, dependence plots, and pairwise interaction values. It compares SHAP rankings with permutation importance and tree-based…
This notebook implements DeePM, an end-to-end portfolio policy for a diversified ETF universe. Its model combines temporal features, asset metadata, cross-sectional attention, and a macro graph prior that permits attention within asset classes and across…
This notebook turns financial headlines into stock-level signals and evaluates them against forward returns. It embeds headlines, measures news surprise as semantic distance from a rolling embedding baseline, and combines surprise with sentiment direction to…
The notebook demonstrates a pipeline for extracting supplier, customer, and competitor relationships from company annual filings and storing them as a knowledge graph. A local language model generates candidate subject–relationship–entity triples from filing…
This notebook asks whether sequence models can improve on hand-built summaries of perpetual funding premium history. It compares NLinear, a simple linear model that reads an ordered window, with a two-layer LSTM. Both use the same feature order, walk-forward…
This document presents a staged method for matching company names from alternative data, filings, news, and price sources to securities. It first joins on available identifiers in a deliberate trust order, preserving unmatched records and checking that…
This notebook runs or replays a panel of identical research agents answering a shared forecasting question, then inspects their forecasts, search activity, and conversation traces. Separate per-agent tracing preserves attribution during parallel calls. The…
This feature-engineering notebook constructs variables that require information beyond one asset’s price history. For futures, it computes annualized roll yield from contemporaneous near and deferred contract levels, plus term-structure slope and curvature…
This notebook defines monthly total-return labels for a cross-sectional US firm-characteristics study and checks how those outcomes align with the provider’s rows. The panel pairs characteristics from the prior month with the return earned in the month…
This notebook explains how a temporal convolutional network (TCN) can forecast returns from ordered financial data. Causal convolutions ensure that an output at a given time depends only on current and earlier inputs. Dilations expand the receptive field…
This case study turns stored model predictions for US stocks into long-short, equal-weight portfolios. It sweeps multiple entry schemes, each selecting the highest-ranked names for long positions and the lowest-ranked names for short positions, to examine…
This document describes several ways to convert prediction scores into trading signals: fixed cutoffs, rolling per-asset percentiles, and cross-sectional percentiles. Fixed thresholds use a chosen score level; rolling thresholds adapt to an asset’s recent…
This notebook builds descriptive market regimes from monthly Federal Reserve economic data: unemployment, the federal funds rate, the Treasury yield curve, and inflation. It explains how to align series with different release frequencies, transform trending…
This notebook builds a simulated market-making task in which a proximal policy optimization agent chooses quote skew and spread width. The environment uses a GARCH volatility process calibrated to hourly crypto perpetual-futures returns, a reservation price…
This chapter surveys methods for recovering common structure from return and characteristic panels, from PCA and eigenportfolios to IPCA, risk-premium PCA, conditional autoencoders, stochastic discount factor estimation, and supervised autoencoders. Its…
This notebook compares sklearn HistGradientBoosting, XGBoost, LightGBM, and CatBoost on an ETF return-prediction task. It measures cross-sectional information coefficient, training time, and process memory across CPU and, where supported, GPU runs, using…
This notebook turns cross-sectional stock predictions into long-short portfolios by ranking stocks at each rebalance, buying the top k and shorting the bottom k with equal capital per position. Equal weighting provides a baseline that isolates differences in…
The notebook previews S&P 100 10-K and 8-K filing data intended for financial knowledge graph construction. It checks schema completeness, unique company and accession keys, form consistency, filing year, and recorded text length. Annual reports supply…
This notebook presents a feature-selection workflow for ETF research, intended to narrow many candidate transforms and lookbacks to a smaller set for later modeling. It computes cross-sectional Spearman information coefficients between features and forward…
The document describes fitting a simple linear sequence model to NASDAQ-100 one-minute features to forecast forward returns at several horizons. Each input contains the preceding hour for one stock, expressed as changes from the latest observation, and the…
The document presents a machine-learning trading research workflow that carries ideas from data and feature construction through model training, backtesting, transaction costs, portfolio and risk decisions, deployment, and monitoring. It emphasizes an…