auto.arima
returns two different models weather I define my time series with frequency=1
(default value) or frequency=7
which is the seasonality period for my data (weekly data).
Below the forecasted values when frequency
parameter is set to 1.
And now the forecasted values when frequency
parameter is set to 7 (which is the seasonality of the data).
From the help page (?ts
) I can see that the frequency
parameter sets "the number of observations per unit of time".
I don’t quite understand what does "unit of time" refers to. Since my time series consists of daily data, my first guess would be that “the number of observations per unit of time” would be 1 since I collect one observation every day. On the other hand, when I define parameter frequency=7
, I get better forecasts from auto.arima
.
My question is:
- What is the true meanning of the
frequency
parameter? Is it the seasonality period?
Best Answer
The
frequency
parameter of the time seriests()
object the number of observations per season, as mentioned on this article - robjhyndman.com/hyndsight/seasonal-periods.As my time series is composed by daily observations and because I have weekly seasonality, the
frequency
parameter must be set to7
. That means, data tends to "repeat itself" every week.