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.
742 documents
This document demonstrates connecting a five-day ETF momentum strategy to Alpaca through a shared strategy interface. The strategy tracks momentum across SPY, QQQ, and IWM and generates signals when it crosses a threshold. Broker-specific adapters handle…
This document compares alternative position-sizing methods for CME futures strategies selected from a baseline ranked by equal-weight validation Sharpe. Equal weighting treats every selected product alike, although futures contracts can have very different…
This notebook turns model prediction sets into comparable S&P 500 options backtests. On each weekly decision date, it ranks predicted returns, selects the highest-ranked symbols in the liquid universe, and sells equally weighted at-the-money straddles.…
This document explains how to compare an incumbent and a candidate model in shadow evaluation before allocating capital. It fixes their roles and promotion thresholds in advance, then assesses the candidate on five dimensions: improvement in annualized…
This notebook demonstrates how to configure Feast with Parquet sources, entities, feature views, timestamps, and a time-to-live policy, then retrieve features for training examples and a live-style as-of query. It compares Feast’s output feature by feature…
The notebook introduces a supervised autoencoder for equity factor modeling. Unlike unsupervised autoencoders that learn a bottleneck by reconstructing the cross-section, this model also includes forward returns in its training objective. The target…
This notebook applies a neural stochastic discount-factor model to CME futures. In asset-pricing terms, a stochastic discount factor is a random variable whose product with each asset's return has the same expected value under no arbitrage. The model…
This configuration note distinguishes settings that appear in a backtest identity from settings that actually affect simulated trading costs. In the described case study, all registered runs use a vectorized, return-to-expiry path. The configured…
The document demonstrates a broker wrapper that checks orders and portfolio state before forwarding trades. Its controls include per-order share and value caps, position exposure limits, order-rate limits, asset allow and block lists, and an emergency halt.…
This analysis checks whether settlement data can support a weekly, cross-sectional CME futures strategy before any model is fit. It tests whether enough products are quoted on rebalance dates to fill the intended long and short portfolios, estimates spread…
This case study evaluates monthly long-short decile strategies built from US firm characteristics, comparing linear models, gradient boosting, and latent-factor approaches under point-in-time accounting lags, walk-forward validation, and era-dependent…
This notebook compares embedded, server-based, general-purpose, and trading-focused databases on financial time-series tasks: bulk writes, full scans, time-window queries, OHLCV aggregation, and trade-quote alignment. Its main lesson is that storage layout…
This educational treatment compares parametric stochastic models and bootstrap methods for generating synthetic financial paths. It covers geometric Brownian motion, jump diffusion, mean reversion, Heston stochastic volatility, and GARCH, alongside…
This analysis compares three ways to convert signed model return forecasts into binary positions: a fixed zero cutoff, a trailing percentile for each symbol, and a cross-sectional percentile across symbols. It applies the rules to registered validation…
This notebook configures Feast as a feature store for financial and model-based data, then compares its historical and point-in-time retrievals with a transparent, hand-built join. It declares entities, timestamped sources, feature views, and a time to live,…
This notebook compares gradient boosting objectives on monthly US stock returns while holding features, target, and walk-forward folds fixed. It contrasts squared error with absolute error and Huber loss, which reduces the influence of large residuals. The…
This module coordinates walk-forward fitting and evaluation for several latent factor models, including principal components, instrumented principal components, and neural autoencoders. It prepares panel data and folds, invokes model-specific runners, and…
This notebook measures how changing proportional transaction costs affects one validation-selected FX strategy per return label. It first selects a parent from signal, allocation, and risk-overlay candidates, then varies only the aggregate cost per traded…
This analysis explains how to read cross-sectional information coefficients (ICs) for CME futures models and how they relate to later portfolio tests. It computes each date’s rank correlation between predicted and realized returns, then averages those…
This notebook assesses whether historical ETF data can support a monthly ranking strategy before any model or forecast is built. It checks point-in-time eligibility using prior-year liquidity, counts available funds on rebalance dates, converts per-share…
The document explains how the Deflated Sharpe Ratio (DSR) adjusts a selected strategy’s Sharpe for the number of strategies tested, variation across their Sharpe estimates, and non-normal return distributions. It contrasts this with an uncorrected…
This notebook investigates why two backtesting engines can produce different results from identical signals. It holds a precomputed ETF strategy fixed, forms a deterministic portfolio from its predictions, and changes execution settings one at a time. The…
This notebook compares LightGBM models on a US equities panel across three return horizons. Its grid varies tree leaf counts and loss functions, then scores each configuration at multiple tree-count checkpoints. The discussion explains how shallow trees can…
This notebook evaluates validation predictions derived from equity and options research by translating them into long-only equity portfolios. It uses an equal-weight top-K baseline, compares several portfolio concentrations, and follows the label-specific…