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
The article compares two forecasting targets for a Bollinger Bands trading approach: whether price will rise or fall over a future horizon, and which of four zones around the bands price will occupy. It describes labeling historical GBPUSD daily data,…
This article prepares a feature correlation matrix for clustering, a prerequisite for the clustered feature-importance analysis planned in the next part of its series. It separates two problems: finite-sample estimation noise, which spreads eigenvalues even…
The article introduces self-organizing maps (SOMs), an unsupervised neural-network method that arranges high-dimensional input vectors on a two-dimensional grid. Each node has a weight vector; training repeatedly finds the best-matching node for a sample and…
The article presents bacterial chemotaxis optimization as a population based method for searching an objective function. Candidate solutions are modeled as bacteria moving through a bounded search space along straight trajectories, with random turns and…
The article compares Kelly sizing with a confidence based signal method, then proposes combining them in two stages: the signal method accounts for estimated confidence and overlapping positions, while a Kelly multiplier adjusts for asymmetric win and loss…
The article adapts the Atom-Motif Contrastive Transformer, originally developed for molecular data, to trading inputs. It treats individual candles as low-level elements and recurring candle formations as motifs, encoding each sequence through a separate…
This article explains trend scanning as an alternative to assigning every market observation a fixed prediction horizon. For each starting point, it fits linear trends over a range of forward windows, calculates a t statistic for each, and selects the window…
The article introduces machine learning for price prediction, then outlines a workflow using MetaTrader 5 hourly EUR/USD data, Pandas, scikit-learn, and neural networks. It explains basic decision trees and the GRU architecture, comparing its reset and…
The article proposes using neural networks to filter signals from an adaptive version of Thomas DeMark's Sequential strategy. It focuses on Setup and Intersection signals near potential turning points, treating the network as a decision aid rather than a…
The document introduces Bayesian inference as a way to update probabilities as new evidence arrives, then applies the idea to several components of an MQL5 trading system: signal generation, money management, and trailing stops. For signals, it uses…
The article describes adding reinforcement learning to a multilayer perceptron (MLP) in an MQL5 trading system. Q-learning is used to guide model training rather than act as the direct signal generator. The market environment is represented by combinations…
The document explains semi-supervised and active learning, then describes how a classifier can use unlabeled examples to guide which labels to request. It outlines assumptions that support learning from unlabeled data, including nearby points tending to…
The article introduces artificial neurons and multilayer networks, describing inputs, weights, weighted sums, activation functions, and how learning changes weights. It explains scaling inputs to a bounded range and compares step, sigmoid, and hyperbolic…
The article explains why neural networks used in trading may benefit from scaling their inputs. It introduces z-score normalization, which centers values around their mean and scales by standard deviation, and min-max normalization, which maps values using…
This article proposes transforming financial price changes into a lower-dimensional representation with a Restricted Boltzmann Machine (RBM), then passing that representation to a multilayer perceptron (MLP) that predicts a subsequent price change. It…
The article outlines a proposed trading system that combines a quantum-circuit state encoder, a deep Q-network, language-model analysis, and a self-adaptation mechanism. It motivates the design with familiar reinforcement-learning difficulties in markets:…
The article explains the Arithmetic Optimization Algorithm (AOA), a population-based metaheuristic that updates candidate solutions using division, multiplication, subtraction, and addition. It initializes candidates within bounded ranges, then uses…
The article describes the Backtracking Search Algorithm, an evolutionary method for real-valued optimization that uses both a current population and a historical population. Its mutation step moves candidates according to the difference between current and…
This article introduces neural network optimization as the process of updating model weights to reduce prediction loss. It compares stochastic gradient descent, which updates from individual examples, batch gradient descent, which uses the full dataset, and…
The article explains OPPO, an offline reinforcement learning approach that learns a policy directly from comparisons between trajectories. It contrasts this with the common two-stage method of first fitting a scalar reward model and then training a policy…
This article describes an Expert Advisor that classifies daily market conditions as flat, uptrend, or downtrend by comparing price changes with the Average True Range. It counts historical transitions among these states to estimate a 3-by-3 Markov transition…
The article presents Across Neighborhood Search (ANS), a population-based metaheuristic for numerical optimization. Agents explore around both a shared collection of strong solutions and their own best solutions, using Gaussian-distributed perturbations to…
The article explains how to extend a linear model into a simple artificial neuron by adding an intercept, called bias, alongside the slope, called weight. It uses mean squared error over training examples as a cost function and estimates each parameter’s…
This article explains how Exponential GARCH models address a limitation of standard GARCH estimation: positivity constraints on conditional variance can force parameters to boundaries and restrict the fit. EGARCH models the logarithm of variance, keeping…