The parameter p determines the number of past samples that we consider.
AR(0): the series is just white noise.
AR(1): the series is a random walk if
The parameter q is the number of previous white noise terms considered.
The term I refers to how many times the series has been differenced to achieve stationarity. An ARIMA model is simply an ARMA model on the differenced time series.
ARIMA that includes additional autoregressive and moving average components. These additional lags are offset by the frequency of the seasonaility (sn).
They take into account exogenous variables.
To determine if the time series is stationary. For the data to be stationary (so to reject the null hypothesis) the ADF test should have a p-value <= a significance value to be set. Null Hypothesis: The data is not stationary. Alternative Hypothesis: The data is stationary.
- The standard residual plot: should have mean zero and uniform variance
- Histogram and KDE estimate: the KDE estimate should be close to the normal distribution
- Normal q-q: most of the points should lie on the straight line
- Correlogram / ACF plot: for lag > 0, 95% of the correlations should not be significant
Useful resources: