shwinters - best approach assessing fit

[Graham Roberts]

My data set is 208 weeks (4 years) of admission data - there is both an underlying trend and definite seasonality. I have run shwinters to forecast 52 weeks forward from the data set. Checking visually the fit of the prediction from the first 3 years of data fits is a very good fit for the fourth year.

Going back to my model using first 4 years to forecast the following year, I cannot find any information about which of the measure of fit variables I am best to use to assess the fit of the model.

Thanks, Graham.

STATA v16.1 output

. tssmooth shwinters smooth_206=EpisodeID if Arrivalweek<207, forecast(52) period(52) normalize additive from(.1 .1 .1) iterate
> (100)
computing optimal weights

Iteration 0: penalized RSS = -354976.5 (not concave)
Iteration 1: penalized RSS = -303451.52
Iteration 2: penalized RSS = -300545.35
Iteration 3: penalized RSS = -299716.72
Iteration 4: penalized RSS = -299539.3
Iteration 5: penalized RSS = -299496.85
Iteration 6: penalized RSS = -299486.81
Iteration 7: penalized RSS = -299485.88
Iteration 8: penalized RSS = -299485.42
Iteration 9: penalized RSS = -299485.39
Iteration 10: penalized RSS = -299484.91
Iteration 11: penalized RSS = -299484.91 (backed up)
Iteration 12: penalized RSS = -299484.91 (backed up)

Iteration 98: penalized RSS = -299484.91 (backed up)
Iteration 99: penalized RSS = -299484.91 (backed up)
Iteration 100: penalized RSS = -299484.91 (backed up)
convergence not achieved

Optimal weights:
alpha = 0.2977
beta = 0.0000
gamma = 0.0000
penalized sum-of-squared residuals = 299484.9
sum-of-squared residuals = 299484.9
root mean squared error = 38.12886