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 system studies rotation within AI-related equity baskets traded through USDT perpetual contracts. It screens for equity contracts, uses an LLM to assign companies to one or more AI subsectors, and tracks relative returns and Z-scores to identify…
This document presents a proposed trend following strategy that compares a current price window with historical windows using Euclidean distance. It selects the nearest examples, averages their subsequent price changes, and combines that direction with a…
This strategy uses a neural network to turn the percentage difference between current OHLC4 and a reference-timeframe OHLC4 into a directional signal. The described network has two hidden layers, uses tanh activations, and applies fixed weights in a forward…
This strategy adapts the RSI lookback according to the rolling correlation between price and a momentum series. It maps the correlation into a configurable minimum-to-maximum period range: stronger positive correlation leads to a shorter RSI period, while…
This indicator strategy combines simple moving averages, adaptive linear moving averages, and infinite impulse response lines to display potential trend and trade conditions. Crosses among the lines can inform entries; spacing among IIR lines is used to…
This strategy combines price interactions with two unusual Fibonacci retracement levels, historical candlestick pattern matching, and distribution statistics. It calculates the levels from a recent high-low range; a matching routine compares the current…
The document describes a stock analysis system that combines RSI, Impulse MACD, VIDYA, volume measures, price-based market-depth proxies, and sector strength relative to SPY. Each component contributes a weighted score, with bullish and bearish thresholds…
This strategy combines predictions from a neural network applied to price changes on a large and a small timeframe. It opens a long or short position only when both model outputs exceed a threshold in the same direction, and closes positions when the…
This machine learning example uses a two-layer LSTM to produce daily stock weights for a NASDAQ-100 universe. For each training asset, it uses the one-period rate of change of a linearly weighted moving average of closing prices as its feature, and labels…
The document presents a crypto trading system combining Bollinger Bands, RSI, Supertrend, and ATR-based stop and target levels. Its stated entry logic looks for price outside a Bollinger Band with RSI at an extreme, then uses Supertrend as a directional…
This example uses a RidgeClassifier to predict whether each crypto futures asset’s next closing price will rise or fall. For features, it takes the logarithms of recent closing prices from three other futures contracts, then fits a separate classifier for…
This document describes a long-only Bitcoin strategy using Bollinger Bands and RSI on a 15-minute chart. It presents three variants: one using 20-period bands, another using 17-period bands with stop-loss and take-profit levels, and a third that adds MFI…
This strategy smooths four market inputs with Nadaraya–Watson kernel regression, then combines them using preset weights: closing price, RSI, ATR relative to price, and price deviation from a moving average. A crossover between the integrated signal and its…
This document describes an event-driven system that tracks posts from a selected X account and uses a large language model to extract named stocks, direction, confidence, and rationale. It filters out ambiguous or low-confidence posts, then checks whether…
This QuantConnect example trains a small TensorFlow neural network on daily SPY opening prices. It uses a rolling history of 30 prior days, pairing each opening price with the next day’s opening price, and trains a one-hidden-layer network with gradient…
This machine-learning strategy selects the 15 least volatile liquid S&P 500 stocks and uses a fully connected neural network to estimate whether each will rise over the next day. Inputs combine ATR as a share of closing price, normalized RSI readings across…
The workflow pits multiple language models against one another as virtual traders, each with a separate simulated account and tracked performance. Models receive daily, hourly, and five-minute price data with RSI, MACD histogram, ATR, and OBV, then issue…
This single-instrument framework combines technical indicators, relevant news, current positions, and stored trading rules to produce structured AI decisions. Its workflow records market snapshots, applies position and confidence limits, manages entries and…
This strategy uses a smoothed Stochastic oscillator to identify potential overbought and oversold entries, with a 200-period simple moving average serving as a trend filter for additional entry conditions. The description also presents an AI signal as a…
This workflow example combines account positions, news sentiment, and technical indicators to produce automated trade decisions for tokenized US stock markets. It collects position data, analyzes news for short- and long-term sentiment, and calculates MACD,…
The proposed system defines bullish and bearish market conditions using short- and long-period simple moving averages, RSI thresholds, and a rolling standard deviation as a volatility measure. It then applies preset transition probabilities to choose a next…
The document describes an automated approach to trading Polymarket prediction contracts. It scans popular markets by recent volume, asks an AI model to create search queries, gathers current web and news results, and then uses the model’s assessment to…
This QuantConnect example trains a small PyTorch feed-forward network each trading day using recent daily SPY opening prices. It pairs each observation with the following day’s open, optimizes the network with mean squared error, and derives upper and lower…
This strategy uses K-nearest neighbors to classify the current market state from similar historical observations. Users can choose a price-derived input, such as HL2, VWAP, or a moving average, and a target such as price action or volatility. The directions…