Optimization criteria for R's ets function

0
votes

When using the ets function in R's forecast package, what objective function is being optimized when opt.crit="amse"? (I'm fitting a linear additive model.)

The documentation mentions "average MSE over first nmse forecast horizons", so would that be 

(MSE_1 + MSE_2 + ... + MSE_nmse)/nmse

where MSE_i is the mean squared error associated with i-step forecasts? If so, is there a way to configure the ets function to just optimize with respect to MSE_nmse?

1
rtime-seriesforecasting

1 Answers

0
votes

sorry for the late answer but could be helpful for others. Your understanding of "amse" and nmse is correct. But nmse is only used if you use opt.crit = "amse". So if you simply want to optimize with respect to MSE_nmse then you should not use "amse" but opt.crit = "mse".

Hope it helps :)