[See also Description.pdf for an economical description of the methodology].
In this package one can find two popular techniques to pool different sources of information. The first relates to combining individual forecasts through simple averaging schemes (mean or median) or through a discounting weighted function, proposed by Stock and Watson (2004). The second, consistent with Neely (2011), combines the explanatory factors through a principal component regression. Many options are included to optimize the forecasts, as for instance:
- expanding or rolling window
- the number of lags in the regressions
- different distributions for the coefficient estimates (Normal, Exponential, Logit, etc...)
- different combination techniques
- manual choice of the discount factor
- the number of principal components to be included in the forecasts (see also screenshot)
- whether the eigenvectors should be obtained on the base of the correlation matrix or covariance matrix.
Both techniques are implemented through a dynamic (real-time) framework.
The package consists of the following files:
- indivfc.m: function that makes individidual forecasts for k factors
- combinefc: function that combines the given individual forecasts
- pcafc.m: function that performs a pooled regression on the base of J principal components.
- Description.pdf: full methodology described
- dataset.mat: time series obtained from Yahoo finance as an illustrative example.
- Example.m: main function which demonstrates the use of this package.
All functions are provided with a carefull and detailed description, in a similar format as the MATLAB guidelines.
J. H. Stock and M. W. Watson. Combination forecasts of output growth in a seven-country data set. Journal of Economic Literature, 23:405-430, 2004.
C. J. Neely, D. E. Rapach, J. TU, and G. Zhou. Out-of-sample equity premium prediction: Fundamental vs. technical analysis. Technical report, Singapore Management University, 2011.