In few cases, the auto modeling procedure in X-12-ARIMA failed to converge during the estimation of AR and MA coefficients' values. The user needs to (1) - Rerun the model with initial values set from the last iteration, or set a higher iteration number (or lower tolerance).
Reference: X-12-ARIMA reference manual (http://www.census.gov/ts/x12a/v03/x12adocV03.pdf), page 40
" When X-12-ARIMA's iterative estimation scheme fails to converge, several remedies are available. If the program stopped short of convergence because it reached the maximum number of iterations (indicated by a warning message to this eect and the printing of parameter values at the last iteration), then rerunning the program with initial parameter values set at the values obtained at the last iteration may produce convergence.
An easier, though computationally slower, alternative is to simply increase the number of iterations allowed and rerun the program. If the program crashed before converging or reaching the maximum number of iterations, then it may help to rst t the model by conditional likelihood, and then use the resulting parameter estimates as initial values for exact maximum likelihood estimation."