Time Series Analysis and Forecasting

A three-lecture course for the DHL/CPP programme, now in its third year, moving from time-series fundamentals through stationarity and baseline forecasting to automated ARIMA modelling and forecast evaluation, with companion Jupyter notebooks for each lecture, a hands-on lab, and a graded practical exam.

Lecture 1 loads and visualizes a real-world time series (an Arctic climate dataset) and decomposes it into trend, seasonal, and residual components. Lecture 2 covers stationarity (ADF testing), transformations (differencing, log, Box-Cox), autocorrelation (ACF/PACF), and baseline forecasting. Lecture 3 covers Auto-ARIMA, Prophet, and residual diagnostics, comparing models by MAPE. The lab has students build a complete end-to-end forecasting pipeline, from raw data to a tuned ARIMA model with forecast intervals, on a new dataset.