Forecasting With Confidence: Bayesian Structural Time Series in R With bsts

Turn a single forecast line into monthly odds of breaking even, hitting targets, and paying off a bigger ad budget

Most small business forecasts end up as a single line on a chart. Next month’s orders will be 427. Next quarter’s revenue will be $118,000. Those numbers feel precise, and that precision is the problem. Owners plan payroll, inventory, and marketing budgets around a figure that has a real chance of being wrong, with no sense of how wrong it might be.

Bayesian forecasting replaces that single line with a distribution of plausible futures. Each forecast month comes with thousands of simulated outcomes, so you can ask practical questions directly. What’s the probability we cover costs in February? How likely are we to hit our annual order target? Does a bigger ad budget actually move those odds?

The bsts package, short for Bayesian structural time series, is one of the most practical ways to build these forecasts in R. Steven Scott developed it at Google, and it powers the CausalImpact package that many analysts use for measuring campaign effects. The model breaks a time series into understandable parts, such as trend and seasonality, then uses Markov chain Monte Carlo sampling to…