MittnikReduction: Balance and Reduce a Model

Description

Balance and reduce the state dimension of a state space model a la Mittnik.

Usage

MittnikReduction(model, data=NULL, criterion=NULL, verbose=TRUE,warn=TRUE)
    MittnikReduction.from.Hankel(M, data=NULL, nMax=NULL, 
       criterion=NULL, verbose=TRUE, warn=TRUE)

Arguments

model

An object of class TSmodel or TSestModel.

data

If the supplied model is of class TSestModel and data is not supplied then it is taken from the model. If the model is of class TSmodel then data must be supplied.

criterion

Criterion to be used for model selection. see informationTestsCalculations.

verbose

logical indicating if information should be printed during estimation.

warn

logical indicating if some warning messages should be suppressed.

a matrix. See details.

nMax

integer indicating the state dimension of the largest model considered.

Value

A state space model balance a la Mittnik in an object of class TSestModel.

concept

DSE

Details

MittnikReduction gives nested-balanced state space model using reduction by svd of the Hankel matrix generated from a model. If a state space model is supplied the max. state dimension for the result is taken from the model. If an ARMA model is supplied then singular values will be printed and the program prompts for the max. state dimension. criterion should be the name of one of the values returned by informationTests, that is, one of ("port","like","aic","bic","gvc","rice","fpe","taic", "tbic","tgvc","trice","tfpe"). If criteria is not specified then the program prompts for the state dimension (n) to use for the returned model. The program requires data to calculate selection criteria. (The program balanceMittnik calculates svd criteria only and can be used for reduction without data.)

The function MittnikReduction.from.Hankel is called by MittnikReduction and typically not by the user, but there are situations when the former might be called directly. It selects a reduced state space model by svd a la Mittnik. Models and several criteria for all state dimensions up to the max. state dim. specified are calculated. (If nMax is not supplied then svd criteria are printed and the program prompts for nMax). The output dimension p is taken from nrow(M). M is a matrix with p x (m+p) blocks giving the markov parameters, that is, the first row of the Hankel matrix. It can be generated from the model as in the function markovParms, or from the data, as in the function estSSMittnik.

data is necessary only if criteria (AIC,etc) are to be calculated.

References

Gilbert, P. D. (1993) State space and ARMA models: An overview of the equivalence. Working paper 93-4, Bank of Canada. Available at http://www.bankofcanada.ca/1993/03/publications/research/working-paper-199/.

Gilbert, P. D. (1995) Combining VAR Estimation and State Space Model Reduction for Simple Good Predictions. J. of Forecasting: Special Issue on VAR Modelling, 14, 229-250.

Mittnik, S. (1989), Multivariate Time Series Analysis With State Space Models. Computers Math Appl. 17, 1189--1201. Mittnik, S. (1990), Macroeconomic Forecasting Experience With Balance State Space Models. International Journal Of Forecasting, 6, 337--348. Mittnik, S. (1990), Forecasting With Balanced State Space Representations of Multivariate Distributed Lag Models. J. of Forecasting, 9, 207--218.

Examples

Run this code

data("egJofF.1dec93.data", package="dse")
    model <- toSS(estVARXls(egJofF.1dec93.data))
    newmodel <-MittnikReduction(model, criterion="taic")

Run the code above in your browser using DataLab