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 discusses why fixed-interval log returns are common and whether researchers can model returns sampled with volume bars or generated by technical signals. It lists common motivations for log returns: approximate correspondence with simple returns…
The response surveys several problems facing rates desks: systems for managing multiple curves and their risks, convexity adjustments for CMS and cash settled swaptions, forward volatility smiles, and finding opportunities in credit, structured products, or…
The discussion considers alternatives to estimating daily stock volatility with the standard deviation of close-to-close log returns, which can be strongly affected by outliers in a small sample. Suggested robust dispersion measures include the interquartile…
The document outlines a workflow for consumer credit default scoring and discusses when expert judgment might supplement a statistical model. It recommends checking whether a sample represents the target population, measuring variable stability with the…
The document outlines a research project on whether weather, agricultural statistics, and other information can help explain or predict US soybean futures. It proposes narrowing the study to Illinois and examining how local conditions affect yields and…
The discussion surveys reported tools used to price and manage electricity and natural gas risk. One contributor cites neural networks in corporate trading technology. Another describes a broader toolkit spanning futures, forwards, swaps, and options,…
This discussion explains how to scale new market features for a predictive trading model after fitting a scaler on historical data. Its central point is to treat scaling as a fitted transformation: estimate its parameters from history, then apply those same…
The document collects suggestions for learning about machine learning and data mining in quantitative investing, including a standard statistical learning textbook. One response treats data mining and machine learning as closely related terms. Another…
The document addresses how to choose a historical sample when calibrating a Heston volatility model. The question reports accurate parameter estimates alongside wide confidence intervals for samples around a decade long, and asks whether a longer sample is…
The document considers how to classify rare observations in a high-dimensional dataset when labeled historical events are available. Its proposed approach assumes a multivariate normal distribution and uses Mahalanobis distance to measure how far a vector…
The document is a discussion about turning formal statistical machine learning coursework into practical finance projects. The original questioner has studied regression, classification, neural networks, support vector machines, and principal component…
The document describes the central idea behind the Universal Model associated with Justin Sirignano and Rama Cont: train a forecasting model on pooled data from many stocks, then apply the shared model to forecast individual stocks, including stocks outside…
The document asks whether continuous-state space or dynamic linear models can forecast security-level equity risk premia over medium-term horizons. It contrasts these models with discrete-state Markov and hidden Markov models, suggesting that continuous…
The document considers a classifier that predicts whether a fixed target or stop is reached first. Since many trades may offer little predictive edge, it questions whether ordinary binary cross-entropy is well suited to selecting only cases with favorable…
This document raises the question of whether state-space models can improve forecasts of equity returns from five-minute price data. The author describes working with several years of observations across multiple equities, comparing linear regression with an…
The document outlines several ways a physicist can build the mathematical and financial background needed for quantitative finance. It distinguishes a PDE-centered route, which can introduce financial derivatives with limited stochastic calculus, from a…
The discussion considers how to split tick data when tuning a fast trading strategy with machine learning. The main recommendation is to preserve chronological order: train and validate on earlier observations, then reserve a later period for testing.…
The author describes a binary stock-direction classifier built with a random forest. Historical open, high, low, close, and volume data are used to derive technical features, including momentum, oscillator, volume, and moving-average indicators. The target…
The discussion explains ways to turn visual chart patterns into rules a program can evaluate. It presents a head-and-shoulders definition based on five consecutive local extrema: the middle peak is higher than the outer peaks, while the two shoulder values…
The question concerns forecasting sales for a large panel of firms using sector industrial production and lagged financial statement variables. The response suggests using principal component analysis (PCA) to reduce the many firm-level series to a smaller…
The document discusses how to allocate capital across strategies that span assets and holding periods. It frames the challenge as controlling shared market exposures while diversifying risks unique to individual strategies, such as a single-name failure or a…
The document describes calibrating a one-factor mean-reverting log-price model for oil using futures prices at multiple maturities as observations. It specifies a latent spot-price state, a measurement equation linking that state to log futures prices, and a…
The document distinguishes data snooping from ordinary in-sample versus out-of-sample model selection. Data snooping arises when researchers reuse the same observations across many strategy or hypothesis tests; searching enough candidates can produce an…
The document asks how to estimate a country’s default probability when its historical record contains no defaults. The proposed starting point is a logit regression using macroeconomic variables, but the absence of default observations leaves the model…