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.
715 documents
This notebook evaluates TSMixer as a global sequence model for an ETF panel. The model shares parameters across funds while using each fund’s own history and covariates; its mixing layers learn temporal and within-series relationships without combining one…
This notebook defines forward-return targets for a cross-sectional futures strategy that ranks products by term structure, going long those with stronger carry and short those with weaker carry. It distinguishes roll-adjusted prices, appropriate for…
This notebook explains how to interpret Kalshi’s federal funds rate contracts and prepare their prices for quantitative research. A binary contract price represents an implied event probability, but the feed contains the highest standing YES bid rather than…
This notebook describes using TSMixer to predict stock returns from ordered windows of each stock’s features. Each example contains 60 consecutive sessions; windows that cross gaps in a stock’s history are excluded, so the number of usable examples varies…
This document presents a staged process for linking company names from filings, news, and alternative data to securities. It distinguishes entity identifiers such as CIK and LEI from security identifiers such as FIGI, CUSIP, and ISIN, and explains why…
This document explains a supervised autoencoder factor model for predicting stock returns in an equity option analytics research setting. Unlike PCA, IPCA, and an unsupervised conditional autoencoder, its training objective combines reconstruction of the…
This document describes a fixed out-of-sample backtest for a selected CME futures strategy. The configuration, predictions, allocator, rebalance cadence, concentration, and transaction-cost assumptions are inherited from earlier research steps and applied…
This notebook studies linear prediction models for a cross-section of CME futures products using feature columns grouped into related families, including carry, momentum, volatility, and rolling risk measures. Because columns within a family are often…
This educational analysis compares four neural network designs on the same task: predicting the next daily return from a recent window of returns. A pooled set of eight exchange-traded funds provides varied market exposures. The models differ in how they…
This notebook audits the complete set of canonical validation predictions for an FX pairs study. It checks model identity, artifact availability, completeness, configured-label coverage, and whether the assembled configurations match the declared menu.…
This notebook audits the registered validation predictions for an FX pairs study. It checks that the catalog includes the configured model population, complete prediction sets, available artifacts, and current model identities. It then summarizes predictive…
This notebook measures how well the Lee-Ready method infers trade aggressor direction using Nasdaq order-by-order messages with venue-provided aggressor labels as ground truth. It reconstructs the limit order book from adds, modifications, cancellations,…
This notebook evaluates NLinear, a simple sequence model that subtracts the latest observed level from a lookback window and maps the resulting sequence to a forecast with a linear transformation. It fits the ETF model population on walk-forward folds and…
This case study fits linear models to predict returns and direction in crypto perpetual futures. Its feature matrix is dominated by variations of the perpetual premium, the gap between the contract and spot price that drives funding payments. It contrasts…
This notebook presents a conditional autoencoder for equity returns in which a neural network maps stock characteristics to nonlinear factor loadings, while another network extracts contemporaneous latent factors from characteristic-managed portfolio…
This notebook explains how to rebuild a limit order book from DataBento market-by-order messages for a single NASDAQ symbol and trading day. It models each order, aggregates orders into price levels, and maintains separate bid and ask sides. Correct message…
This notebook compares single-objective hyperparameter tuning with a multiobjective search for LightGBM prediction models. The baseline maximizes validation information coefficient (IC). The NSGA-II search instead maximizes IC while minimizing normalized…
This assessment traces a single US equities strategy from a frozen validation backtest set to its holdout evaluation. It applies a deterministic rule: choose the candidate with the highest validation Sharpe, breaking ties by backtest hash. Registry records…
This notebook explains how to select one strategy from a fixed set of US equity backtests and assess the resulting strategy on a separate holdout period. It validates that candidates share the required data and protocol identities, then ranks them by…
This notebook compares three linear time-series models, a Transformer encoder, and parameter-free forecasts on daily SPY returns. The linear approaches map a historical window to a multi-day forecast, with variants that separate a smooth component from its…
This notebook presents a multi-round forecasting debate between bull and bear roles. Each side argues for a higher or lower probability of an event, sees the other side’s prior argument in later rounds, and reports a probability and supporting evidence. The…
This notebook examines whether gradient-boosted trees can find nonlinear relationships in NASDAQ-100 microstructure features that a linear model may miss. It focuses on the possibility that order-flow imbalance predicts returns differently depending on the…
This notebook explains an operator-style research agent for iterating on quantitative case studies. Instead of limiting the model to predefined forecasting actions, it gives it general tools for files, shell commands, registry queries, and data inspection.…
This feasibility analysis checks whether a daily long-short equity ranking strategy can be researched with the available US stock panel. It examines point-in-time universe construction using price and trailing turnover thresholds, compares proportional…