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.
5,018 documents
This notebook fits a long short-term memory network to NASDAQ-100 microstructure features to predict forward returns over several minute-scale horizons. Unlike a flattened linear model, an LSTM processes each observation in sequence and learns how much past…
This chapter explains how to evaluate financial machine-learning models while respecting the information available at each decision time. It distinguishes training, validation, and a final holdout, and explains why randomly shuffled k-fold validation lets…
FlyTV maps a measured fruit fly connectome into a chart-based trading strategy. The script description says its network represents 1,329 neurons and 20,000 measured connections. It assigns price-derived stimuli to modeled neural circuits: looming detectors…
This proposed trend-following system combines price position relative to 200-period EMA and SMA filters with a directional prediction signal. It describes taking long positions when price is above both averages and the prediction is positive, and short…
This document describes an automated trading setup that presents itself as machine-learning based. Its stated components include a Hull moving average, two exponential moving averages, candle-based support and resistance levels, and a higher-timeframe…
This research prototype turns trading frameworks attributed to 99 crypto traders into a daily Bitcoin consensus signal. It builds market features from price, technical indicators, and macro series, maps those features to trader-specific capability profiles,…
The document presents an open-source Pine strategy combining two AI-labeled trend estimates with SuperTrend. Its visible settings define two K-nearest-neighbor configurations, each with neighbor and historical data counts, and separate weighted moving…
This experimental strategy combines a fast and slow volume-weighted moving average (VWMA) crossover with a k-nearest-neighbors classifier. A fast-line cross above the slow line proposes a long entry, while a cross below proposes a short entry. The classifier…
This document outlines a research prototype for estimating the terminal direction of BTC in Polymarket 15-minute Up/Down contracts. It combines Binance trade flow and Polymarket reference feeds with the official price-to-beat, time remaining, and forecast…
This workflow has multiple language models act as virtual traders, each maintaining a separate simulated account and producing standardized position actions. Their decisions use technical readings from daily, hourly, and five-minute intervals, including RSI,…
This strategy trains an XGBoost model on six price and volume features: normalized RSI, distance from a 50-period EMA, rate of change, ATR relative to price, relative volume, and the close's location within a recent price range. It periodically fits the…
This strategy combines three technical signals: fast and slow EMA crossovers for trend changes, RSI thresholds for overbought or oversold conditions, and MACD line crossovers. Its signal logic can require any enabled indicator to trigger or require all…
This example describes a monthly U.S. stock strategy that uses a machine-learning model to rank stocks by their predicted likelihood of outperforming the cross-sectional median. Its input features are past closing-price returns measured over several lookback…
The document proposes identifying stock accumulation with a combination of moving averages, price behavior, volume, and momentum. Its written rules require price above the 200-day average and the 50-day average above the 90-day average, then look for rising…
This excerpt describes a crypto trading system that combines adaptive moving-average and Laguerre filters with a Hurst-style measure, volatility compression, and a hidden Markov model regime detector. It also maps Asian, London, and New York sessions to…
This example applies a linear support vector regression model to daily closes of the CSI 300 index. It constructs features from the current close and the preceding twenty daily closes, then fits the model to predict the following day’s close. The historical…
This strategy uses a K-nearest-neighbors model to turn seven technical features into a directional prediction: price momentum, RSI, volume ratio, volatility, trend strength, MACD, and Bollinger Band position. The features are Z-score standardized so their…
This Polymarket strategy screens event contracts for liquidity, trading activity, spread, market competitiveness, and price, then scores hourly candlestick patterns such as gradual rises, growing volume, narrowing pullbacks, and breakouts. Candidates passing…
This example trains a two-layer LSTM model on the logarithm of closing prices for three named NASDAQ stocks. It labels the next observed close as up or down and uses those labels as targets in a model trained with mean squared error and an LBFGS optimizer.…
This strategy seeks to trade new Binance perpetual contracts at launch. It polls official announcements, tracks upcoming listings, and gathers market data from CoinMarketCap and four other exchanges, alongside news and token fundamentals. An AI model…
This strategy classifies market direction by comparing current indicator features with historical feature vectors, then voting across the nearest labeled observations. The description mentions RSI, ROC, CCI, and volume as possible inputs, normalized to a…
This strategy combines an SMA and standard-deviation price channel with candlestick pattern rules, volume confirmation, and directional probability estimates. It describes nine patterns, including hammer, shooting star, engulfing, and multi-candle…
This document describes an AI-assisted workflow that turns natural-language crypto factor ideas into computed signals and validation reports. A language model interprets the idea, generates a factor function, and assigns signal direction; the system then…
This document presents a directional trading strategy that applies a fixed artificial neural network to a price-change input. It uses the prior day's percentage change as its single input, passes that value through two tanh hidden layers with 5 and 33 nodes,…