Rolling Origin Forecast Backtesting

How to use Rolling Origin Forecast Backtesting

Each fold forecasts the next observation using only earlier rows. Naive, seasonal naive, drift and fixed-alpha exponential smoothing share your initial training count. Errors are actual minus forecast; scaling uses each fold’s prior seasonal-difference MAE.

1. Enter ordered numeric json array, initial training observations, seasonal lag for naive method and error scale, ses alpha.
2. Choose the stated convention and check the units and supported model.
3. Calculate, inspect the result, then copy or download it.

Good to know

• 3–1000 equally spaced ordered numeric observations, absolute value at most 1e9 and nonzero magnitude at least 1e-12. Initial training is at least 2; seasonal lag is 1–100 and smaller than that prefix. SES begins at the first observation, without fitted alpha. Mean fold-scaled error is unavailable if any training scale is zero. Approximate arithmetic only; no interval forecast, model selection guarantee, trading signal or future performance claim.
• Local calculation only; no input persistence or external lookup. Text/JSON is limited to 30000 characters and output to 1000000 characters. Numeric model inputs and precision/work limits are stated on this page. Approximate JSON numbers retain full stored values; exact arithmetic outputs retain decimal or fraction strings. Unsupported states fail explicitly.