Time-Series Forecasting, Chatfield
A time series is anything that we can view as a function of time x(t). Often we look at discrete time, in which case, we just have a series (xt)
Given a time series, we'd like to forecast/predict what its future values will be.
If we know x1, …, xn, and want to know xn + h, h is called the horizon
We'll denote this by xn(h), to make explicit the fact that what we're finding out is the projected value at the horizon, from a fixed point n. In other words, we want to think of xn − 1(h+1) differently from xn(h), because they're working from different data sets (xn has more data).
Based on intuition ex. Delphi method
Depend only on a single time series
Depend on more than one time series
Automatic methods don't require human intervention
Also called ex-ante.
xn(h) uses data up to time n.
Some predictions influence the results.
ex. economic predictions influence people to take certain actions, which in turn have an influence on what happens in the economy
On the other hand, short-term weather forecasting doesn't actually influence the weather
Long-term weather forecasting (in terms of years) may have influence, if it causes governments/companies to alter their policies to account for, say, climate change.
Forecasting generally assumes that the future will be like the past
Forecasts are conditional statements: if things continue like A, then B will happen