# Stationarity in ARIMA modeling

I am working on a problem that I think ARIMA modeling could be useful for, and am researching the theory behind ARIMA. I came across this website that says:

ARIMA(p,d,q) forecasting equation: ARIMA models are, in theory, the most general class of models for forecasting a time series which can be made to be “stationary” by differencing (if necessary).

My question is, what does this mean? When and why is it needed to make a time series ¨"stationary", and what is differencing?

Differencing means taking differences between observations one period apart, i.e. working with $y_2-y_1, y_3-y_2, ...$ instead of $y_1, y_2, y_3, ...$. This addresses some kinds of non-stationarity. See here for other ways to address non-stationarity.