Professional Documents
Culture Documents
Data Summary
• the results of the data indicate that the classical no-change forecast still
outperforms the other forecasts
• the autoregressive forecast performed best, using the root-mean-square
error, with a 24-month rolling training period
• the probabilistic forecasts , also using the root-mean-square error, per-
formed best with a rolling training period of 12 months
• the no-change and constant rate of growth forecasts deviate largely
from their probabilistic counterparts, and it is still unclear to me whether
this makes sense or is due to an error
• given the predictive distributions of the probabilistic forecasts, I still need
to determine the best function for the average directional average and the
prot rules under the Bayes rule for the results to be meaningful
Root- Root-
Average Average Mean Mean mean- mean-
directional directional average average square square
accuracy (1) accuracy (2) error (1) error (2) error (1) error (2)
Constant rate of
growth 0.61 0.61 1.53 1.56 2.45 2.29
Probabilistic no-
change (12 months) 0.43 0.43 1.55 1.55 1.97 1.97
Probabilistic constant
rate (12 months) 0.65 0.65 3.84 3.85 5.80 5.65
No-change $0 -$636,025 $0 $0
Constant rate of
growth -$1,027,250 -$1,202,083 -$1,196,842 -$828,217
Probabilistic no-
change (12 months) $94,808 -$83,658 -$83,658 -$271,408
Probabilistic constant
rate (12 months) -$869,544 -$845,491 -$613,943 -$512,345