fDMA (version 2.2.8)

tvp: Computes Time-Varying Parameters Regression.

Description

This function computes Time-Varying Parameters Regression (TVP) with the updating procedure as in Raftery et. al (2010).

Usage

tvp(y,x,V,lambda,W=NULL,kappa=NULL,c=NULL)

Value

class tvp object, list of

$y.hat

fitted (forecasted) values

$thetas

estimated regression coefficients

$pred.dens.

predicitive densities from each period

$y

y, forecasted time-series

Arguments

y

numeric or a column matrix of a dependent variable

x

matrix of independent variables, different columns should correspond to different variables

V

numeric, initial variance in the state space equation for the recursive moment estimator updating method, as in Raftery et al. (2010)

lambda

numeric, a forgetting factor between 0 and 1 used in variance approximations

W

optional, numeric, initial value of variance for the model equations, if not specified the method based on the linear regression, as in Raftery et al. (2010) is used

kappa

optional, numeric, a parameter in the exponentially weighted moving average in variance updating (see also fDMA), between 0 and 1, if not specified the method as in Raftery et al. (2010) is used

c

optional, logical, a parameter indicating whether constant is included, if not specified c=TRUE is used, i.e., constant is included

Details

It is not possible to set c=FALSE if ncol(x)=0. In such a case the function will automatically reset c=TRUE inside the code.

References

Raftery, A. E., Karny, M., Ettler, P., 2010. Online prediction under model uncertainty via Dynamic Model Averaging: Application to a cold rolling mill. Technometrics 52, 52--66.

Sanderson, C., Curtin, R., 2016. Armadillo: A template-based C++ library for linear algebra. Journal of Open Source Software 1, https://arma.sourceforge.net/armadillo_joss_2016.pdf.

See Also

grid.tvp, print.tvp, summary.tvp, plot.tvp.

Examples

wti <- crudeoil[-1,1]
drivers <- (lag(crudeoil[,-1],k=1))[-1,]
ld.wti <- (diff(log(wti)))[-1,]
ld.drivers <- drivers[-1,]
ld.drivers[,c(4,6)] <- (diff(drivers[,c(4,6)]))[-1,]
ld.drivers[,c(1:2,5,7)] <- (diff(log(drivers[,c(1:2,5,7)])))[-1,]
ld.drivers[,c(3,6)] <- ld.drivers[,c(3,6)]/100
# \donttest{
t1 <- tvp(y=ld.wti,x=ld.drivers,V=1,lambda=0.99)
# }
# \dontshow{
tt1 <- tvp(y=ld.wti[1:20,1],x=ld.drivers[1:20,1:4],V=1,lambda=0.99)
# }
# \donttest{
t2 <- tvp(y=ld.wti,x=ld.drivers,V=1,lambda=0.99,W=1)
# }
# \dontshow{
tt2 <- tvp(y=ld.wti[1:20,1],x=ld.drivers[1:20,1:4],V=1,lambda=0.99,W=1)
# }
# \donttest{
t3 <- tvp(y=ld.wti,x=ld.drivers,V=1,lambda=0.99,W=1,kappa=0.75)
# }
# \dontshow{
tt3 <- tvp(y=ld.wti[1:20,1],x=ld.drivers[1:20,1:4],V=1,lambda=0.99,W=1,kappa=0.75)
# }
# Model with constant only
empty <- matrix(,nrow=nrow(ld.drivers),ncol=0)
t4 <- tvp(y=ld.wti,x=empty,lambda=0.99,V=1)