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 describes execution concerns for an XGBoost signal model trading MNQ micro futures, with momentum among its features and several signals per day. The author reports that market-order slippage averaged 1.07 points and says this erased the…
The document explains why extracting principal components from constituent stock returns does not by itself forecast a particular stock’s next-day return. PCA represents each asset return as a weighted combination of components estimated from historical…
The document discusses using reinforcement learning and other machine learning methods to schedule a fixed-size order over a set horizon. It contrasts strategies learned from order book or spread data with model-based optimal execution, and describes…
The document considers how to assemble observations for a logistic regression that estimates insolvency probabilities and produces credit scores. The question arises when a database contains repeated monthly records for each company, raising concerns about…
The document asks whether skilled traders with little capital can exploit unusually large inefficiencies that are inaccessible to large firms. One response argues that small opportunities are not necessarily overlooked: firms with strong research and trading…
The document examines a proposed approach to forecasting whether a future tick return will be positive, negative, or unchanged. It describes converting outcomes into class labels, fitting separate binary logistic regressions, and comparing predicted…
The document examines a coding question about simulating geometric Brownian motion with the Milstein method. Its update formula contains a Brownian increment in both the first-order stochastic term and the squared-increment correction. The author asks…
The document asks how published estimates of recession probability over coming quarters were calculated, referring to estimates reported in a 2016 article for the United States and euro area. The accepted response says there are several possible approaches…
The document explains how to construct two Brownian motions with a target correlation by combining one Brownian motion with an independent one. The second process is formed as a weighted sum: the first path receives weight rho, and the independent path…
The document explains why market efficiency cannot be tested in isolation: observed prices must be compared with a benchmark for what prices should be, and that benchmark depends on an asset-pricing model. A rejection can therefore reflect market…
The discussion considers what a trading firm might mean when it says its quantitative methods are difficult for new hires to learn. One answer points to machine-learning methods such as neural networks, support vector regression, and graphical models. It…
The discussion explains how to define training targets when using neural networks with market indicators. Rather than selecting visually appealing entry points at random, the researcher should decide what outcome the model is meant to predict and label…
The document gathers recommendations for teaching C++ to learners who want to apply it to financial mathematics and quantitative research. Contributors disagree on whether a finance-specific book is the best starting point. One instructor reports success…
The document examines how to describe signal decay when comparing lagged information ratios. Its example uses two orthogonal signals with equal autocorrelation and the same absolute decline in information ratio, but very different starting strengths.…
The document explains why market orders may not appear as resting entries in a limit order book: they execute against available limit orders and generate trades. Limit orders can also execute immediately when they cross the spread, either partially or fully.…
The document explores how discrete ARMA and seasonal autoregressive models might relate to continuous-time processes. One proposal replaces the discrete lag coefficients with continuous lag-weighting functions and represents the lagged contributions through…
The document examines whether a neural network can recover two-factor Hull–White parameters from caplet prices, and why volatility and correlation estimates may be unstable. It recommends training on implied volatilities rather than raw prices, since prices…
The document responds to a request for a step-by-step R example of fitting an ARMA-GARCH model to a stationary time series, with interest in selecting ARMA orders by AIC and forecasting future values. It points to the rugarch package for univariate models…
The document describes bid-ask bounce in high-frequency data: trades alternating between bid and ask can make observed prices jump around the underlying market value. Taking log returns of midprices and features may reduce some nonstationarity, but the…
The document explains why factor investing and machine-learning applications are more common in equities than in fixed income. Bonds differ by maturity, liquidity, optionality, and capital-structure position, so securities from one issuer are less…
The document asks why variables that appeared relevant in a regression lose apparent predictive power after first differencing to address unit roots. The response explains that differencing can remove signal as well as trend, and that apparent relationships…
The document asks whether portfolio optimization can avoid using empirical estimates such as historical returns and volatility. Its brief response points to unsupervised learning as a way to investigate relationships among assets, including correlation-like…
The discussion explains that Bayesian versions of Fama-French or Carhart factor models can be approached as Bayesian regression. It recommends learning Markov chain Monte Carlo and Gibbs sampling, while noting that Hamiltonian Monte Carlo software can reduce…
The document surveys how quantitative investors may use machine learning and data, including for estimating fair value, setting trading quotes, and researching stock signals. One response frames high-frequency firms as liquidity providers: models can…