TStoX {regtools}

Transform Time Series to Rectangular Form

Description

Input a time series and transform it to a form suitable for prediction using lm etc.

Usage

TStoX(x,lg)
TStoXmv(xmat,lg,y)

Arguments

x

A vector.

lg

Lag, a positive integer.

xmat

A matrix, data frame etc., a multivariate time series. Each column is a time series, over a common time period.

y

A time series, again on that common time period. If NULL in TStoXmv, then y is set to x (i.e. for a univariate time series in which older values predict newer ones).

Details

Similar to stats::embed, but in lagged form, with applications such as lm in mind.

TStoX is for transforming vectors, while TStoXmv handles the multivariate time series case. Intended for use with lm or other regression/machine learning model, predicting y[i] from observations i-lg, i-lg+1,...,i-1.

Value

As noted, the idea is to set up something like lm(Y ~ X). Let m denote length of x, and in the matrix input case, the number of rows in xmat. Let p be 1 in the vector case, ncol(xmat) in the matrix case. The return value is a matrix with m-lg rows. There will be p*lg+1 columns, with "Y," the numbers to be predicted in the last column.

In the output in the multivariate case, let k denote ncol(xmat). Then the first k columns of the output will be the k series at lag lg, the second k columns will be the k series at lag lg-1, ..., and the lg-th set of k columns will be the k series at lag 1,

Author(s)

Norm Matloff

Examples

x1 <- c(5,12,13,8,88,6) 
x2 <- c(5,4,3,18,168,0) 
y <- 1:6 
xmat <- cbind(x1,x2) 

TStoX(x1,2)
#      [,1] [,2] [,3]
# [1,]    5   12   13
# [2,]   12   13    8
# [3,]   13    8   88
# [4,]    8   88    6

xy <- TStoXmv(xmat,2,y)
xy
#      [,1] [,2] [,3] [,4] [,5]
# [1,]    5    5   12    4    3
# [2,]   12    4   13    3    4
# [3,]   13    3    8   18    5
# [4,]    8   18   88  168    6

lm(xy[,5] ~ xy[,-5])
# Coefficients:
# (Intercept)    xy[, -5]1    xy[, -5]2    xy[, -5]3    xy[, -5]4
#       -65.6          3.2         18.2         -3.2           NA
# need n > 7 here for useful lm() call, but this illustrates the idea