0

I asked a question earlier in the forum on auto arima click here Performance evaluation of auto.arima in R and UCM on one dataset. The auto.arima provided strange forecast, upon further looking at the code I did not find anything wrong in my R code see code below. This seems a very straightforward problem. If auto.arima does not fit a simple straightforward dataset, I would be very cautious in using this function to fit more complicated datasets. I would encourage using other tools/ functions and verify forecast. 

plot(eggs)

## Hold out 10 data points - 1984 thru 1993
egg_price = ts(eggs,start = 1900, end = 1983)

## Fit arima model

fit <- auto.arima(egg_price)
fcast <- forecast(fit,h=10)
plot(fcast)
Community
  • 1
forecaster
  • 7,349
  • 9
  • 43
  • 81
  • The auto.arima function is... interesting. For one thing, if the differencing factor is over 1, it doesn't fit an intercept. Here is a good explanation of some of the problems: http://www.stat.pitt.edu/stoffer/tsa2/Rissues.htm Personally, I specified auto.arima(d=1) and it gave me much better results after that. – Olga Mu Aug 27 '13 at 01:50
  • If the differencing factor is over 1, an intercept induces a quadratic trend or worse. Who would want that? Also, the link you provide does not discuss auto.arima. – Rob Hyndman Aug 27 '13 at 03:33

0 Answers0