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 document explains how Bayesian decision theory assigns an observation to the class with the largest posterior probability. It separates the posterior into a class-conditional likelihood and a prior class probability. The prior can be estimated from class…
This study summary describes using recurrent neural networks, including long short-term memory cells, to forecast stock movements separately by industry. The rationale is that firms in one sector may share common influences, while industries can differ in…
This report summary describes using gene expression programming, a heuristic inspired by biological evolution, to discover price-volume factors from short-horizon data. Because many candidate factors had weakly monotonic group returns or no excess return in…
This article explains how to turn future returns into labels for supervised trading models, arguing that the label should match the instrument, trading horizon, liquidity, and costs of the intended strategy. It first describes using an n-period return as a…
The article compares two Chinese equity approaches built around market capitalization. The traditional strategy selects the smallest-cap stocks for a portfolio. The AI version uses a StockRanker ranking model with market capitalization as its sole feature.…
The author describes an unfinished attempt to build a multiple logistic regression library in MQL5 that can handle a variable number of independent data columns. Earlier functions were hard-coded for particular numbers of predictors, which the author…
The document describes a graphical workflow for downloading historical bars, configuring a CTA strategy backtest, reviewing performance statistics, and inspecting trades on a candlestick chart. Data can come from a domestic market data service, an…
The document frames a model-selection problem for reinforcement-learning-based dynamic hedging of long-dated swaptions. The proposed application uses 2y2y and 4y2y swaptions, requiring simulated paths that update both a forward swap curve and an implied…
The document raises an econometric modeling question involving a GARCH(1,1) volatility equation and a fourth equation in a simultaneous system. A variable from the fourth equation enters the GARCH specification as an exogenous regressor, while the…
The document considers how to build a fundamental scoring model for a defined stock universe using metrics for size, growth, valuation, quality, and risk. It describes the practical challenge of collecting current, historical, and estimated Bloomberg fields,…
The document outlines ways to begin building a machine learning credit scoring model for a thesis, focusing on public datasets, example competitions, and learning materials. It points to credit default prediction tasks as sources of data and published…
The document presents a QuantLib Python calibration attempt for a time dependent Heston model that fails with a Boost assertion. The code builds a volatility surface, creates a piecewise time dependent model, attaches an analytic pricing engine, and…
The responses survey reinforcement learning (RL) applications in quantitative finance, with portfolio allocation as the main example. They describe critic-only methods, which choose actions using learned value estimates; actor-only methods, which optimize…
The document addresses how a technically capable beginner can move from trading infrastructure and market knowledge toward designing strategies. It describes strategy as a broad category, ranging from simple rules based on price gaps to models using…
The document asks why a generalized linear model’s design matrix is assumed to have full column rank and whether maximum-likelihood estimation can proceed when it does not. The response connects rank deficiency to non-identifiability: if columns are linearly…
The document asks whether an anomaly detector can be trained on normal gift-card activation transactions and then evaluated on anomalous cases. It also describes a setting with no reliable fraud labels and transaction-level fields such as merchant, location,…
The document discusses reward design for reinforcement learning applied to foreign exchange hedging. A typical objective combines portfolio gains with penalties for risk, such as variation in portfolio value, and for trading costs associated with changes in…
The document gathers advice for a mathematician with statistics and programming experience who wants to enter quantitative finance without first joining a hedge fund. It recommends exploring quant roles at banks or in academia, building relationships with…
The document considers feature choice for a multilayer perceptron that classifies financial price moves as buy, hold, or sell. The proposed inputs concatenate recent returns and transaction volumes, and the question is whether a vector containing more than…
The document clarifies how tick imbalance bars are formed from transaction data. Each tick receives a direction of plus or minus one based on the change from the prior trade price; unchanged prices inherit the previous direction. The running sum of these…
The document presents a validation question for a machine-learning portfolio strategy evaluated out of time against a strong benchmark. Its custom rolling metric reflects portfolio value and additional state variables such as inventory and time. The observed…
The document gathers suggestions for quantitative investigation of suspicious stock trading, including spoofing and possible information asymmetry. One proposed method builds behavioral attributes for each trader, such as trading frequency and volume, then…
The document asks how to make a trading bot imitate a typical nonprofessional foreign exchange trader, including whether such traders can be treated as acting randomly regardless of market conditions. The response rejects assuming random behavior as a sound…
The document asks how to evaluate predictions of asset returns when matching the sign of each forecast to the realized return matters more than explaining return magnitude. It gives a small example where ordinary least squares has low explanatory power,…