X11 output B 4 - Replacement Values for Extreme S-I Differences (Ratios)

First a preliminary estimate of the X-11 seasonal component is computed by applying a weighted 5-term moving average separately to the B 3 values for each month. Then a centered 12-term moving average of the preliminary factors for the entire series is computed, and the resulting values are adjusted to sum to zero (additive model) or 12.0 (multiplicative model) within each year. Next an initial estimate of the irregular component is obtained by subtracting from the S-I differences (additive model) or dividing the S-I ratios by the initial estimate for the seasonal component. For the resulting initial estimate of the irregular component, a 5-year sliding standard deviation (s, sigma) is computed, and extreme values in the central year that are beyond 2.5*s are removed. The 5-year sliding s is then recomputed and the process repeated; however, this time a zero weight is assigned to irregular values beyond 2.5*s, a full weight is assigned for values within 1.5*s, and linearly graduated weights between zero and one are assigned for values between 1.5 and 2.5 * s. Values receiving less than full weights are then recomputed as the average of the respective value times its weight and the nearest two full- weight values preceding and following the respective value in that month. Table B 4 shows the final replaced (re-computed) values, and the sliding 5-year s's.
For more information, see X-11 Census Method II Seasonal Adjustment .