Forecasting in least absolute value regression with autocorrelated errors: a small-sample study