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 surveys neural network choices for return forecasting and asks whether they can support high-frequency trading. Responses emphasize that model choice depends on the data, horizon, inputs, and underlying dynamics. One answer presents recurrent…
The document describes a practical issue in implementing Meucci’s entropy pooling model. A user reports that imposing two inequality views on the expected return of the same asset causes an overflow in the SciPy optimization routine, although applying either…
The discussion addresses whether a long historical price series is suitable for forecasting a stock whose behavior has changed over time. It notes that an LSTM can be configured with a chosen number of time steps, so each training sample can use a recent…
This answer outlines a visual approach to choosing autoregressive and moving-average orders in an ARIMA model. It recommends using the autocorrelation function (ACF) and partial autocorrelation function (PACF) to identify candidate AR and MA terms, with…
The document explores ways to estimate daily or weekly security prices when only monthly security observations are available, alongside higher-frequency index and sub-index data. One proposal applies expectation-maximization to log returns, treating them as…
The document considers how to build a machine-learning model to predict mutual fund ratings when funds have different operating histories. The dataset includes returns over several lookback periods, while some newer funds lack the longer-period observations.…
The document asks whether daily return skewness and kurtosis can be forecast in a manner similar to variance models such as GARCH, using only daily S&P return data. It raises the presence of fat tails and negative skewness as motivations for studying higher…
The document explains how the return forecasting model relates to a GARCH volatility model. It distinguishes the conditional mean of returns from the residual process: returns equal their model-implied conditional mean plus an error, and the error is…
The document collects explanations for C++’s continued use in quantitative finance and weighs technical advantages against development costs. Reasons offered include performance, control over memory and processor instructions, parallel computing tools,…
The document examines a workflow that estimates a covariance matrix from stock log returns, extracts principal components, and then projects log prices onto those components. Its central concern is that price levels are nonstationary, so the covariance…
The document frames a risk-management question: whether models can improve on Filtered Historical Simulation (FHS) when generating many realistic financial return paths. It describes a proposed classical workflow: select an ARMA-GARCH specification and lags,…
The document describes a feature-engineering problem: turning a bank’s customer transaction history into inputs for a model that predicts loan default. It gives examples based on account balances, gambling expenditure, declined payments, ATM withdrawals…
The document explains how a GARCH model represents returns and their time-varying conditional variance, addressing confusion about whether to fit the model to returns or squared returns. In the stated setup, returns equal a conditional mean plus a…
The document considers how to choose training and testing windows for walk-forward strategy optimization, how far to shift the windows, and how to select parameters across repeated fits. The response says there is no universally best window length: the…
The document discusses whether machine learning models built from historical stock prices should use raw closing prices or adjusted closing prices. Adjusted closes account for corporate actions such as splits and dividends, which can make historical returns…
The document points readers to approximate dynamic programming as a way to solve finance problems involving decisions over time, focusing on dynamic portfolio choice. Its central method is value function approximation: simulate possible paths and approximate…
The document considers how to identify orderly segments in market prices using net movement, path linearity, and duration, with the aim of informing machine-learning models or projecting prices. The answer challenges the idea that observed behavior can be…
The document asks how to determine whether a one-minute return forecasting model trained on trade prices has predictive value beyond bid-ask bounce. The proposed benchmark simulates normally distributed mid-price returns using observed volatility, adds an…
The document raises the possibility that annual report commentary and disclosures contain latent information that could be converted into quantitative signals. It asks how much time analysts spend reading company reports and which parts of that work might…
The discussion compares the daily high–low difference with the high-to-low ratio as a predictor in a model of whether the market will rise or fall. The difference is expressed in price units, so its scale can change over time; the ratio is more comparable…
The document recommends resources for learning how to apply machine learning and deep learning to finance. It distinguishes between financial research practice, neural-network implementation, and data-science programming. The central recommendation is a…
The document surveys sources of historical stock market data for research and model development, emphasizing coverage, cost, format, and service quality. It describes a paid vendor option with broad consolidated, OTC, and exchange coverage and API access,…
The document asks whether a training set spanning different market regimes should use one global class weight or separate weights for each period. Its example is a two-year sample with a bullish first year and a bearish second year: a global adjustment for…
The document asks how to cluster roughly one million time series using statistical features while keeping memory use near linear. It notes that many familiar clustering methods rely on an affinity matrix, which can become a memory bottleneck, and raises…