Data & decisions · focused workspace

Forecast Backtest Lab

Compare naive, moving-average and exponential-smoothing forecasts against held-out observations, with errors shown for every prediction.

Model comparisonCompare scenariosCSV & printable report

Make the next step clearer

Open CSV / project

Preparing the worked example…

Make it yours

Settings & assumptions

Observations

Your result, explained

02 / RESULTS

Read the result in context

Rolling one-step backtests reveal each actual observation before predicting the next. Dates must be unique; spacing is not modeled. Historical accuracy is not a promise of future performance.

How to use Forecast Backtest Lab

Explore the worked example. Hold out the last three values of a dated series and compare three one-step prediction methods on identical test observations.
Replace the observations with your own entries. Set moving-average window, smoothing alpha, held-out observations.
Choose “Compare rolling forecasts”. Review model comparison and rolling predictions, then compare the visual overview with the detailed tables.
Capture alternative scenarios, download a reusable project and export the selected table or a report when you are ready.

A starting point

Hold out the last three values of a dated series and compare three one-step prediction methods on identical test observations. The example is illustrative and can be replaced.

Know what you are comparing

Does the backtest look ahead at the value it predicts?

No. Each prediction uses only earlier observations. After scoring a held-out point, its actual value becomes available for the next rolling prediction.

Files, privacy & saving

Calculations run locally in a browser worker. Input CSV files use the column names in the downloadable template and keep your current settings. Projects restore inputs and up to four scenarios. Imports are limited to 1 MiB and 100 entries. There is no automatic saving. Public pages include advertising and optional analytics; see our privacy policy.

CSV exports contain the entire selected table. JSON and printable HTML reports include all result tables; open the report and use your browser’s Print command for PDF.

Using Forecast Backtest Lab

Compare naive, moving-average and exponential-smoothing forecasts against held-out observations, with errors shown for every prediction.

Explore the worked example. Hold out the last three values of a dated series and compare three one-step prediction methods on identical test observations.
Replace the observations with your own entries. Set moving-average window, smoothing alpha, held-out observations.
Choose “Compare rolling forecasts”. Review model comparison and rolling predictions, then compare the visual overview with the detailed tables.
Capture alternative scenarios, download a reusable project and export the selected table or a report when you are ready.

Example

Hold out the last three values of a dated series and compare three one-step prediction methods on identical test observations.

Questions & answers

Does the backtest look ahead at the value it predicts?

No. Each prediction uses only earlier observations. After scoring a held-out point, its actual value becomes available for the next rolling prediction.

Can I compare and save alternative plans?

Yes. Capture up to four calculated scenarios, restore their inputs and save them in a project JSON file. There is no automatic saving.

How do I bring in my own entries?

Download the CSV template, retain its column names and replace the example rows. Imports are limited to 100 entries and 1 MiB. A saved project also restores settings.

What should I check before using the result?

Rolling one-step backtests reveal each actual observation before predicting the next. Dates must be unique; spacing is not modeled. Historical accuracy is not a promise of future performance.