Three Autoregressive Integrated Moving Averages (ARIMA) forecast procedures for Census Bureau X-11 concurrent seasonal adjustment were empirically tested. Forty time series from three Census Bureau economic divisions (business, construction, and industry) were analyzed. Forecasts were obtained from fitted seasonal ARIMA models augmented with regression terms for outliers, trading day effects, and Easter effects. Revisions between initial and final seasonally adjusted values were computed. Ranked analyses of variance were used on various revision measures to determine the statistical significance of the differences between the extension procedures. The main conclusion was that extending the series with enough forecasts to apply a symmetric filter reduced the revisions more effectively than not extending the series and using asymmetric filters. This result held whether the model was carefully fit by the analyst or was a simple default model. Extension of the series with only 1 year of forecasts was also examined. Two data tables are included. (Author/TJH)