aTSA (version 3.1.2)

# stepar: Stepwise Autoregressive Model

## Description

Fit a stepwise autoregressive model

## Usage

stepar(y, xreg = NULL, trend = c("linear", "quadratic", "constant"),
order = NULL, lead = 0, newx = NULL, output = TRUE, ...)

## Arguments

y
a numeric vector of response
xreg
a numeric vector or matrix of exogenous input variables. The default is NULL.
trend
the type of trend with respective to time. The default is linear.
order
the order to fit the AR model for residuals. The default is NULL.
lead
the number of steps ahead for which prediction is required. The default is 0.
newx
a matrix of new data of xreg for predictions. The default is NULL.
output
a logical value indicating to print the results in R console. The default is NULL.
...
additional arguments for ar function.

## Value

• A list with class "stepar" containing the following components:
• coefa estimated coefficient matrix including the t-test results.
• sigmathe square root of the estimated variance of the random error.
• R.squaredthe R^2 for fitted model in the first stage.
• predthe predictions, only available for lead > 0.

## Details

The stewise autoregressive model uses a two-stage procedure to fit time series. The first stage is to fit a (constant,linear,quadratic) model with respective to time sequence: $t = (1:n)/n$, where $n = length(y)$. If xreg is supplied, the fitted model is updated by $$y = \mu + \beta*xreg + e[t]$$ for trend = "constant", and $$y = \mu + \beta*xreg + \alpha*t + e[t]$$ for trend = "linear", and $$y = \mu + \beta*xreg + \alpha*t + \alpha*t^2 + e[t]$$ for trend = "quadratic". The second stage is to fit an autoregressive process to the residuals of the fitted model obtained in the first stage, which is accomplished by using ar function in stats package.

## Examples

x <- 5*(1:100)/100
x <- x + arima.sim(list(order = c(1,0,0),ar = 0.4),n = 100)
stepar(x)
stepar(x,order = 1)

# with xreg supplied
X <- matrix(rnorm(200),100,2)
y <- 0.1*X[,1] + 1.2*X[,2] + rnorm(100)
stepar(y,X)
# make a prediction with lead = 1; used with caution.
newdat1 <- matrix(rnorm(2),nrow = 1)
fit1 <- stepar(y,X,lead = 1,newx = newdat1,output = FALSE)
# make a prediction with lead = 2; used with caution.
newdat2 <- matrix(rnorm(4),nrow = 2)
fit2 <- stepar(y,X,lead = 2,newx = newdat2,output = FALSE)