# auto.arima picks seasonal model for non seasonal series

I have a monthly time series data of 3 years whose acf and pacf plots confirms absence of seasonality. But auto.arima picks a seasonal model by seasonal difference first and then seasonal AR component and then it gives warning that seasonal difference will not be done.

Can anyone please explain why even seasonal seasonal model is selected and the warning below which just contradicts what auto.arima did?

ACF plot: No seasonal lags significant

auto.arima(data_ts)


EDIT : Data provided

forecast package version is 8.5

 library(forecast)

data_ts <- ts(c(858202,1268645,1483015,1469438,1316090,1240012,1148112,1557396,1210394,1298259,1146520,1205847,1612222,2206315,2380862,2009639,1628334
,2084517,2672135,1706195,1726972,1749780,1859735,1727888,2745657,2383249,2969345,3106084,2441399,2069020,2602255,3016157,2137989,2129232
,2223055,2202366),frequency = 12,start = c(2016,01))

auto.arima(data_ts)

• Can you plot the series itself ? What makes you think it is not seasonal? – Skander H. Feb 6 at 18:59
• @ skander had the series been seasonal then seasonal spikes would have been significant in ACF plot – joy_1379 Feb 7 at 6:12
• This is indeed strange. I'm afraid there is little else to do than to step through the code (which you can see by typing auto.arima) and figuring out what goes wrong where. – Stephan Kolassa Feb 7 at 12:33