Knowledge library

This notebook builds a monitoring workflow for a deployed financial model using real holdout artifacts. It checks that incoming feature data is populated and structurally valid before measuring drift, then compares reference and current windows. Feature…

This notebook connects position sizing with analysis of maximum adverse and favorable excursions. It demonstrates fixed-fractional sizing, where shares are determined by a per-trade risk budget and entry-to-stop distance, subject to a concentration cap. It…

This notebook explains how to interpret Ridge regression forecasts for an ETF panel with SHAP. For a linear model, each feature’s attribution is its coefficient times the feature’s deviation from its training-background mean, so the attributions can be…

This notebook presents a workflow for turning SEC annual and quarterly filings into structured text suitable for later sentiment, topic, and embedding analysis. It maps desired sections to form-specific item numbers, emphasizing that management discussion…

This notebook explains how a conditional autoencoder extends an instrumented principal component model: a neural network maps stock characteristics to factor exposures, while the model retains the same exposure-times-factor-return structure. It fits the…

This study screens financial and GJR-GARCH volatility features against forward equity returns. For each session, it calculates the rank correlation across stocks between a feature and subsequent returns, then assesses average association, consistency across…

This reference describes two financial news corpora used for sentiment analysis, text-feature development, and experiments linking news to returns. FNSPID connects headlines with stock tickers and covers a broad historical span; a smaller sample is available…

This notebook shows how to test whether four managed-portfolio strategies add explanatory power for SPY returns after controlling for broad movements captured by ten principal components. The candidate portfolios rank ETFs by recent momentum, low volatility,…

This document describes a workflow for turning quarterly SEC 13F bulk filings into an institutional holdings panel. It selects the latest filing per manager by filing date, uses CUSIP rather than filer-entered company names to identify securities, and…

The document explains instrumented principal component analysis for modeling ETF returns. Unlike ordinary PCA, which assigns each fund fixed factor exposures, IPCA makes each fund’s exposures a shared linear function of its observable features. It estimates…

This notebook uses NASDAQ ITCH trade records to examine how volume and trading activity vary through the regular session. It resamples trades into time bars, compares price and volume for representative high-, medium-, and low-activity stocks, then…

This notebook compares predictive models for ranking NASDAQ-100 stocks by their next 15-minute return using intraday microstructure information such as spreads, depth imbalance, signed volume, and price impact. It selects one representative prediction set…

This evaluation screens option-surface, price-derived, and conditional-volatility features against forward equity returns. For each session it computes a cross-sectional rank correlation between feature values and subsequent returns, then examines the time…

This module explains how to prepare historical S&P 500 prices when ticker symbols can refer to different securities over time. It applies corporate-action adjustment factors, tracks security identifiers, and marks identity boundaries. Return calculations are…

This notebook explains instrumented principal component analysis (IPCA) for an equity panel with option-surface features. Unlike ordinary PCA, which estimates each stock's exposures from its return history, IPCA models exposures as a shared linear function…

This notebook outlines a pipeline for extracting supplier, customer, and competitor relationships from company 10-K filings and storing them as a Neo4j knowledge graph. A local language model proposes structured subject-predicate-object triples, after which…

This dataset guide compares two sources of US equity tick data for order book and market microstructure research. Databento market-by-order records provide preprocessed add, modify, cancel, and trade events, while NASDAQ ITCH supplies a raw exchange feed…

This notebook demonstrates fine-tuning a transformer for financial named entity recognition, turning text spans into structured organization, person, money, date, and percentage fields. It introduces BIO boundary labels and explains how word-level…

This notebook explains three ways to turn minute-level NASDAQ-100 data into model-based features: a rolling HAR regression for variance forecasts and forecast errors, a Fourier transform for recent volume patterns, and depth-two path signatures for the order…

This notebook compares four ways to retrieve passages from a corpus of company filing text: BM25, dense embeddings, reciprocal rank fusion (RRF), and a cross-encoder reranker. RRF combines ranked lists rather than raw scores, so it can join lexical and…

This case study evaluates daily cross-sectional signals across a broad US stock universe and lays out a long research pipeline, from point-in-time data and engineered features through model comparison, portfolio construction, costs, and holdout assessment.…

This notebook compares LightGBM with an MLP, TabM, and, when available, TabPFN for predicting ETF forward returns from tabular features. Models are evaluated on shared chronological walk-forward folds, with rank information coefficient and training time as…

This notebook compares ordinary least squares with Ridge, LASSO, and Elastic Net for predicting 21-day forward returns across ETFs. The models use the same declared walk-forward folds, with a purge gap matched to the label horizon. Each fold clips feature…

This notebook explains a local linear trend state-space model for extracting financial features from price levels. The hidden state contains a price level and its slope; observations are noisy closes. At each step, the filter predicts the state, measures the…