We study the effects of outliers on short-term forecasting errors and on autoregressive-integrated moving-average (ARIMA) model characteristics such as the Ljung-Box statistics and estimates of the seasonal moving-average parameter. We have fitted sixty Census Bureau monthly time series with ARIMA models, identified additive point outliers, and sought their external causes. Modification of outliers was found not increase the mean absolute forecasting error (of one, two, and three steps-ahead forecasts over the last three years of the data) in 31 out of 44 series with identified outliers. We also discuss consequences of different methods of outlier modification, choice of outlier identification threshold, and effects on the seasonal adjustment of time series.