Why this decision matters
Almost every operational decision is a forecast in disguise — how much inventory to order, how many call-center agents to staff, how much cash to load into ATMs. Get the forecast wrong and you either over-spend (excess inventory, idle staff) or under-deliver (stockouts, queues, complaints). And unlike other ML tasks, you can't shuffle time-series data — past predicts future, never the reverse.
By the end of this topic you'll be able to
Decompose a series into trend, seasonality, and residuals; forecast with classical methods (moving average, exponential smoothing, ARIMA-family) and modern ones; respect time-aware train/test splits; quantify forecast uncertainty in a way the business can plan around.
Materials
Key concepts to know
- Trend, seasonality, residual — every series can be decomposed into these three components.
- Stationarity — many forecasting methods assume the statistical properties don't change over time. Often they do; you have to handle it.
- Lag features — yesterday's value is the best predictor of today's. Almost always.
- Moving average & exponential smoothing — simple, robust, hard to beat for short-horizon forecasts.
- ARIMA family — the classical forecasting workhorse. Auto-regressive integrated moving average.
- Time-aware validation — never random-split a time series. Always train on past, test on future.
- Forecast intervals — a single number is a guess; the business needs a range.
Readings & class notes
- Time Series Analysis — ReferenceComprehensive treatment of the methods covered in lecture.
- Time Series in SAS Model StudioStep-by-step guide to building forecasts in the enterprise tool.
- Week 14 Class Notes — Student VersionSelf-contained notes from the in-class lecture.
- Self-Study Notes (Text + Time Series)Combined notes for the last two modules.
Hands-on: forecasting in Excel
Excel has surprisingly capable forecasting built in. Start here to build intuition before moving to Python.
- Excel Forecasting ExamplesWalked-through examples of moving average, exponential smoothing, and FORECAST.ETS.
- FRED Economic Data — Excel DemoPull macroeconomic series from FRED into Excel and forecast them.
Case kits — three operational forecasts
Three real operational forecasting scenarios, each with data, code, and instructions.
- ATM Cash Demand ForecastingHow much cash should we load into each ATM, given day-of-week / holiday patterns?
- Equipment MonitoringPredict when machinery is drifting toward failure based on sensor history.
- TrendCart — E-Commerce TrendForecast e-commerce sales trends with seasonality and promotional effects.
Standalone notebooks
- ATM & Equipment — Combined Code ReferenceReference notebook with the techniques used in the case kits.
- Economic Data AnalysisForecast macroeconomic indicators pulled from FRED.