Time Series Forecasting

Programme overview

Every business that holds stock or plans staff needs forecasts, and most still use a guess. In this course you build the forecasting system for Kolanut's Lagos depot from four years of daily sales of six products, in Python. You'll frame the forecast around the weekly ordering decision, measure the weekday, December, payday, Eid and promotion patterns, test baselines on a period they never saw, and build a calendar regression that halves the best baseline's error. You'll measure what promotions really add once the dip afterwards is counted, handle the January 2026 price rise as a structural break, backtest over many origins, check whether prediction intervals are honest, and turn forecasts into order quantities with safety stock for a chosen service level.

Curriculum

1. 01Forecasting for Decisions20 min

   • Forecasting for decisionsWhat a forecast is for, how the decision sets the horizon and the level of detail, and a first look at four years of daily demand at Kolanut's Lagos depot.
2. 02Trend, Seasonality and Events15 min

   • Trend, seasonality and eventsBreak a series into its building blocks (trend, weekly and yearly seasonality, paydays, holidays and promotions) by measuring each one directly from the data.
3. 03Baselines and Honest Testing15 min

   • Baselines and honest testingTest forecasts the only honest way (on a later period the method never saw), measure them with WAPE, and set the simple baselines any real method has to beat.
4. 04Regression with Calendar Features20 min

   • Regression with calendar featuresForecast with a regression model built from what you know about future dates (weekday, month, paydays, Eid, promotions), and learn why lags shorter than the horizon are off-limits.
5. 05Promotions, Events and Structural Breaks25 min

   • Promotions, events and structural breaksMeasure what promotions and Eid really add (including the dip after a promotion), and handle a structural break like a price rise so the forecast doesn't keep predicting a world that's gone.
6. 06Backtesting over Time15 min

   • Backtesting over timeTest a forecasting method on many past periods with rolling-origin backtesting, so one lucky or unlucky test period doesn't decide which method you trust.
7. 07Uncertainty and Ordering25 min

   • Uncertainty and orderingPut a range around a forecast, check whether the range is honest, and turn forecast plus uncertainty into an order quantity with safety stock for the service level the business wants.
8. 08Forecasting Many Series15 min

   • Forecasting many seriesApply one tested method to every product, compare accuracy across them, and decide where human review is still needed.
9. 09Final Project20 min

   • Final project: the depot's ordering forecastPlan your final project, a tested forecasting and ordering system for all six products at Kolanut's Lagos depot, and start by backtesting a product other than the malt drink.

Assessment and certificate

You are assessed with a final assessment (pass mark 60%) and a final project. Your certificate is awarded when you meet those requirements. Paying for the programme does not award it on its own.