The least absolute shrinkage and selection operator ('lasso') has been widely used in regression shrinkage and selection. We extend its application to the regression model with autoregressive errors.