comparing var models using measures of forecast accuracy

[newsagent]

In a previous question (link), the comparison of the two VAR models
gave

#> # A tibble: 2 × 3
#>   .model  MASE RMSSE
#>   <chr>  <dbl> <dbl>
#> 1 m1     0.564 0.450
#> 2 m2     0.728 0.625

Does this mean that overall model 1 has better forecasting accuracy in total (on average) than the model 2?

When does it make sense to compare models as a whole instead of particular variables (like Income or consumption) from these models?

[robjhyndman]

On these measures, on this test set, m1 is better on average across the variables. You would need to consider the statistical significance of the difference to address the question of whether the result would probably hold across other test sets. Whether it makes sense to average accuracy across two or more variables depends on why you are doing the comparison, and what decisions will flow from it.