ARIMA Modeling on specific time series

Question

i have some time series example that i want to predict, lately i was applying ARIMA and Bayesian Forecasting method. But i really am interested in making a further best ARIMA Model (or just knowing some suggestion about the time series), so here is my time series data (it is as far only almost 2 years)

data: purchase of Alpara Kapl medicine from Feb 2015 - Oct 2017

0 2 0 0 0 1 2 3 3 0 0 2 0 1 0 0 3 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0

From these time series, i get best model of ARIMA is ARIMA(0,0,0) which is White Noise, but then there are other models (that is likely more redundant) from ARIMA(0,0,0) which is these (with the calue of AICc):

ARIMA(2,0,2) with non-zero mean : Inf

ARIMA(0,0,0) with non-zero mean : 97.09692

ARIMA(1,0,0) with non-zero mean : 98.46206

ARIMA(0,0,1) with non-zero mean : 98.42101

ARIMA(0,0,0) with zero mean : 104.5138

ARIMA(1,0,1) with non-zero mean : 100.3878

there is ARIMA(0,0,1) the best second model,

but when using at forecasting, White Noise will give a flat forecasting (because its definition is not suitable for forecasting as i known so far), so is it a good way to implement ARIMA(0,0,1) for this reason? or maybe is there some other clue to define other model? any thoughts is welcomed thankyou :)

Answer 1

Your data suggest the need for an Intermittent Demand solution ... . I use AUTOBOX to identify a useful model identifying a three period interval between demands . It uses a sophisticated i.e. robust extension of the Croston method https://stats.stackexchange.com/search?q=user%3A3382+croston which deals with anomalies (one in this case ). . The useful model is here . Note that the forecasts can be simply integerized but are shown for completeness purposes.