ARIMA is a family of models used to analyze and characterize time series/temporal data. Although rudimentary compared to more modern methods, ARIMA is comparably simpler and an easier introduction to time series analysis.
One method of dimension reduction is that of matrix factorization. Just as the name implies, a matrix is 'factored' so that the resulting factors approximate the original matrix when multiplied together.
While there are many methods of optimizing a stock portfolio, one such method is to maximize its excess return to volatility ratio. Such a model aims to find a middle ground between profitability and stability.
Online updates for Bayesian regression allow it to scale better with data dimension as well as reduces the amount of running memory required for loading data or data streams.
Bayesian regression solves for the optimal distribution of regression coefficients rather than a deterministic value. The choice of prior can lead to regularization, and the probabilistic framework introduces a predictive distribution for unobserved data.