ARIMA is the classical workhorse of time-series forecasting and the model an interviewer reaches for when they want to see whether you understand temporal structure rather than just throwing data at a neural network. The name packs in three ideas: autoregression (the series depends on its own past), integration (differencing to reach stationarity), and a moving average (the series depends on past forecast errors). This lesson builds each component, shows how the Box-Jenkins method uses the ACF and PACF from the Stationarity and Autocorrelation lesson to pick the orders, forecasts an AR(1) by hand, identifies an MA(1) from its ACF, and covers estimation, diagnostics, and the reasons ARIMA underwhelms on raw returns. It is the bridge from describing a series to predicting it.
Table of Contents
Already have an account? Log in!