I. Introduction to forecasting
If you missed the first part, I suggest you read it before going through this article. It gives a good introduction as well as an overview of traditional risk management and big data simulation. This article is instead more focused on big data forecasting.
There are nowadays several new techniques or methods borrowed from other disciplines which are used by financial practitioners with the aim of predicting future market outcomes.
Eklund and Kapetanios (2008) provided a good review of all the new predictive methods and I am borrowing here their classification of forecasting methods, which divides techniques into four groups: single equation models that use the whole datasets; models that use only a subset of the whole database, even though a complete set is provided; models that use partial datasets to estimate multiple forecasts averaged later on; and finally, multivariate models that use the whole datasets.
II. Single equation models
The first group is quite wide and includes common techniques used differently, such as ordinary least square (OLS) regression or Bayesian regression, as well as new advancements in the field, as in the case of factor models.
In the OLS model, when the time series dimension exceeds the number of observations, the generalized inverse has to be used in order to estimate the parameters.
Bayesian regression (De Mol, Giannone, …