Knowledge library

This notebook describes a GT-GAN-inspired architecture for synthesizing time series with genuinely irregular timestamps, such as tick, volume, or dollar bars. A GRU-ODE encoder maps observations into latent states, while ODE-based generator and discriminator…

This notebook compares three ways to use market regimes in a forecasting model: omit regime information, add the filtered probability of a volatile state as a feature, or fit separate experts for hard-assigned regimes. It builds regime probabilities from a…

This notebook sets out a process for evaluating a challenger model in shadow mode before it receives capital. It compares stored predictions from a fixed incumbent and candidate, constructs equal-weight ranked long-short books, applies the same…

This case study uses linear models as a baseline for predicting forward returns across a broad US equity panel. It explains why overlapping price and volume features create collinearity, making ordinary least squares coefficients unstable, and compares Ridge…

This analysis compares registered double machine learning treatment effects from nine trading case studies. It loads current results from each study’s registry, checks registry integrity and duplicate labels, and makes coverage explicit. Effects and…

This notebook explains how model uncertainty can add information beyond a point estimate. It fits a stochastic volatility model in which latent log volatility evolves over time and returns follow a Student-t distribution. Posterior spread, interval width,…

This evaluation harness separates four kinds of failure in a financial retrieval-augmented generation pipeline: retrieving relevant evidence, grounding answers in that evidence, refusing unanswerable questions, and resisting unsafe instructions in retrieved…

This notebook presents a multi-round bull-versus-bear debate for a contested probability forecast. Each side argues from shared research-agent summaries, sees the other side's prior position, and revises its probability. The process stops when the forecast…

This notebook builds fitted-model features from foreign-exchange price histories, complementing indicators calculated directly from past prices. It describes three approaches: a state-space filter that estimates a slowly changing price level and related…

This notebook describes a workflow for turning S&P 500 companies’ quarterly MD&A disclosures into trading features. It uses filing acceptance dates as the point-in-time anchor, collapses multiple filings for a company on the same date to the latest period,…

This chapter surveys deep learning for time series through a practical question: when does observation order contain information that cross-sectional prediction misses, and which model designs can use it? It covers recurrent networks and their sequential…

This notebook applies TabM, a tabular neural architecture that shares a two-layer feature network across ensemble members. Each member scales the shared representation with its own learned vector and uses its own output layer; averaging member predictions…

This notebook uses SHAP to explain Ridge regression forecasts of ETF returns. For a linear model, feature attributions have a closed-form relationship to the fitted coefficients and deviations from background feature values, allowing the analysis to check…

This notebook presents instrumented principal component analysis (IPCA), a latent-factor approach to forecasting returns from firm characteristics. Instead of assigning each characteristic a direct return weight, IPCA models each firm’s factor exposures as a…

This analysis compares registered model families on a cross-asset ETF panel using walk-forward validation results. Models predict 21-trading-day returns and are assessed by cross-sectional information coefficients, which measure how well forecasts rank funds…

This notebook tests a two-model exit design on hourly BTC, ETH, and SOL perpetual data. An entry classifier identifies unusually strong forward returns; its predicted probability then joins technical features in a separate model that estimates whether the…

This document describes dataset utilities for preparing time-series panels for model training and rolling inference. The training dataset returns sliding sequences of features, forward returns, volatility scales, and masks, with configurable sequence length,…

This analysis compares validation predictions from several model families for perpetual futures funding. It explains that information coefficient measures the ranking of subsequent returns, AUC measures the ordering of up and down outcomes, and log loss…

This case study configures an LSTM model for forecasting FX-pair labels. It contrasts the recurrent model’s carried hidden state with fixed lookback transformations and convolutional receptive fields, while deferring comparisons with other models to a…

This case study applies gradient-boosted trees to NASDAQ-100 microstructure features to forecast returns and direction at several intraday horizons. It contrasts trees with linear models: trees can discover conditional relationships, such as order-flow…

This notebook assembles a forecasting system with four stages: a panel of research agents, probability aggregation, an adversarial debate, and a supervisor that identifies disagreements, searches for clarifying evidence, and issues a final probability. The…

This notebook compares deterministic relational retrieval with embedding search on a corpus of institutional 13F holdings. It creates a point-in-time snapshot by keeping each institution’s latest reported position before a fixed cutoff, then builds questions…

This notebook demonstrates training a generative adversarial network on engineered ETF features with differential privacy. It introduces the privacy parameters epsilon and delta, explains why DP-SGD clips each record’s gradient before aggregation and adds…

This notebook applies classical ordinary least squares inference to ETF features and forward returns. It reviews coefficient estimates, standard errors, t statistics, p values, and tests for heteroscedasticity, residual autocorrelation, non-normality, and…