Knowledge library

This notebook uses a synthetic asset panel to demonstrate Instrumented PCA, where factor loadings depend linearly on characteristics observed before returns. Alternating least squares estimates the characteristic-to-loading map and realized factors. Because…

This notebook explores ARIMA as a feature generator for models that may use its output alongside other predictors. It explains how ACF and PACF plots can suggest model orders, how information criteria can compare candidate orders on training data, and why…

This document outlines a shared data system for quantitative trading research, cataloging datasets across equities, options, futures, crypto, foreign exchange, factors, macroeconomics, filings, positioning, news, and prediction markets. It describes the…

This case study describes producing an out-of-sample prediction set for an already selected S&P 500 options model. The holdout configuration is fixed using validation results, then fitted again on data ending before the holdout window. A label buffer…

This notebook explains how double machine learning (DML) estimates whether FX momentum affects future returns after adjusting for configured confounders. It distinguishes this intervention question from prediction: predictive models are compared by…

This notebook presents lightweight falsification diagnostics for feature triage, explicitly distinguishing mechanism consistency from causal identification. It first scans ETF features across forward-return horizons with multiple-testing correction, then…

This notebook surveys supervised-learning labels using ETF price data. It covers fixed-horizon forward returns for regression or direction classification, time-series rolling percentiles, cross-sectional percentile labels, triple-barrier labels with fixed or…

This notebook aggregates registry results for OLS, Ridge, Lasso, and ElasticNet across nine case studies. It selects complete validation results at each study's primary label, compares mean daily cross-sectional rank information coefficients and HAC…

This notebook trains Skip-gram Word2Vec on labeled financial news sentences and explains how its vectors represent words that occur in similar contexts. It describes the roles of vector size, context window, minimum frequency, prediction mode, negative…

This notebook compares learned and analytical hedges for a short European call when rebalancing is discrete and trading incurs proportional costs. It defines the self-financing terminal P&L from hedge gains, turnover costs, and the option payoff, then trains…

This notebook evaluates ETF features one at a time by calculating a date-by-date rank information coefficient between feature values and subsequent monthly returns. It screens financial and model-derived features over walk-forward validation dates, then…

This notebook evaluates candidate features for ranking US equities by subsequent returns. It computes daily cross-sectional information coefficients, summarizes their average with uncertainty estimates that account for serial dependence, and adjusts…

This notebook explains PCA as a baseline latent-factor model for forecasting ETF returns. It decomposes a training panel of forward returns into leading directions and estimates each fund’s exposure and factor premia. The estimator ignores the available…

This notebook demonstrates fine-tuning three transformer checkpoints for three-class financial sentence sentiment and evaluating them with accuracy, macro F1, and confusion matrices. It uses a stratified train, validation, and test split so class imbalance…

This notebook estimates the adjusted effect of a continuous ETF momentum measure on forward returns using double machine learning. It contrasts an unadjusted regression with DML estimates that control for recent and longer-term volatility, market regime, and…

This notebook explains how a conditional autoencoder extends instrumented principal component analysis (IPCA): it retains the two-stage structure in which fund features map to latent factor exposures and those exposures combine with factor returns, but uses…

This analysis compares predictive models for ranking NASDAQ-100 stocks by their next 15-minute return using intraday microstructure features such as spreads, depth imbalance, signed volume, and volatility. It emphasizes selecting comparable prediction sets…

This notebook rehearses a deployment cycle for a crypto funding-rate direction model. It trains a LightGBM classifier on historical Binance-derived perpetual data, fetches live hourly bars and funding rates from OKX, computes the model's features, and…

This notebook compares sklearn HistGradientBoosting, XGBoost, LightGBM, and CatBoost for predicting forward ETF returns. It measures cross-sectional information coefficient, training time, and memory use across model-complexity presets, with GPU runs…

This notebook maps a family of ETF return models that infer common latent directions in a panel, with features used to estimate fund exposures. It distinguishes five approaches: unconditional principal components; instrumented PCA with a linear…

This notebook explains how to design a search tool for a forecasting agent so evidence has a consistent structure, a traceable origin, and an auditable path into the model. A shared client protocol returns typed results across providers and includes a cutoff…

This notebook shows why searching across many signals or strategies makes the top observed result look stronger than its underlying predictive value. A simulation uses factors with no true information to illustrate how selecting the largest information…

The document describes how a trading research pipeline assesses whether latent-factor model fits completed in a usable state. For models trained by gradient descent, it checks that the final recorded training objective is finite; for the stochastic discount…

This notebook applies TSMixer to ETF sequences using one globally shared function across funds. Each example contains an individual fund's history and covariates; temporal and feature interactions are learned across the panel, but one fund's observations are…