Zansa › Answers › Forecasting and data science
What is backtesting a forecast and why should I do it?
Backtesting means pretending to forecast in the past: you train on data up to a certain date, forecast the following period, compare with what actually happened, then repeat over several earlier dates. It shows how a method would have performed without waiting months, and it prevents you trusting a model that only fits history.
The key rule is that the model may only use information that was available at the time of each forecast. Using future data, even accidentally, makes results look better than reality.
Good practice:
- Test over several periods, including busy and quiet seasons.
- Match the forecast horizon you actually need.
- Compare against a simple baseline.
- Look at bias as well as accuracy.
Zansa treats backtesting against a baseline as a standard step before a forecast goes into a dashboard.
Want this applied to your business? Zansa builds dashboards and forecasts for small and mid-sized companies worldwide. Start with a free intro call.
Book a call