Comparing the Test and Training for the "UNDER 18 YEARS" group

We can also examine the plots for the training and test groups for this sub segment to see if anything is really going on:

par(mfrow = c(1, 2))
plot(forecast(fit))
plot(forecast(fit2))

From the following plots we can see that although there is a decline in the not-covered percentage for this group, there is probably not enough data to discern trend so that the projections are set at the mean values.

For the test data plot we can also see why the absolute residual plot dropped in xxx as it transitioned from a local peak in 2010 (when the Affordable Care Act was enacted) to the start of a decline in 2011: