FORECAST.ETS.STAT Function in Microsoft Excel: Syntax, Examples & Troubleshooting

=FORECAST.ETS.STAT(values, timeline, statistic_type, [seasonality], [data_completion], [aggregation])

Values, timeline and statistic_type are required. Historical values must align with equally sized numeric timestamps on a regular, nonzero timeline. Excel can sort timestamps implicitly. There is no target-date argument: the third argument is the statistic selector, not the date to forecast.

Use one of the documented selectors 1–8. Microsoft’s current page mistakenly repeats the SMAPE description under its MAE heading; that copied sentence does not make MAE and SMAPE interchangeable. This guide does not infer undocumented metric normalization or exact formulas for reproducing every diagnostic.

For seasonality, 1 or omitted means automatic detection, 0 disables seasonality, and another positive whole number sets the pattern length in observations. The maximum is 8760; unsupported values produce #NUM!. A value of 1 is the automatic-setting flag, not a request for a fixed one-observation cycle. Turning seasonality off does not make this function an ordinary least-squares regression formula.

The optional data_completion setting distinguishes missing observations from real zeros: 1 or omitted fills missing points using the neighboring average; 0 treats missing points as zero. Microsoft documents support for up to 30% missing data. A recorded zero already present in your dataset is a value, not an absent timestamp.

aggregation selects how equal timestamps are combined. The individual worksheet reference describes 0 as the default average. Microsoft’s forecast aggregation enumeration, checked October 17, 2026, lists 1 Average, 2 Count, 3 CountA, 4 Max, 5 Median, 6 Min, 7 Sum. These sources do not express the default consistently. The same worksheet page also mentions duplicate-timestamp #VALUE! errors while describing aggregation of duplicates. These documentation inconsistencies were not resolved by native testing. For predictable preparation, aggregate raw records yourself into one value per timestamp. The source workbook used here has unique timestamps, so its examples do not depend on choosing among duplicate-handling methods.

Download Microsoft’s original forecast sample workbook, linked from the official function reference and checked October 17, 2026. On Forecast.ets sample, use all 57 historical rows, with dates in A4:A60 and airport-passenger counts in B4:B60. The history runs from January 2009 through September 2013. A4 is January 1, 2009 (date serial 39814), B4 is 2644539; A60 is September 1, 2013 (41518), B60 is 3781168.

Keep the entire dataset. These endpoint values are an identity check, not enough data to reproduce the fit by themselves. The workbook contains actual stored formulas and cached outputs. Results below are Microsoft-workbook cached results, read without recalculation, not fresh tests in Excel or independently reconstructed ETS estimates.

On Forecast.ets sample, I75 contains the detected period 12. The actual formula stored in I76 is:

=FORECAST.ETS.STAT($B$4:$B$60,$A$4:$A$60,1,I75,1,1)

Its cached result is 0.251. The third argument 1 requests alpha. The next argument, I75, supplies a seasonal length of 12. These two positions serve different purposes, so moving I75 into the statistic position would not reproduce the example.

The workbook also contains a nearby text label with an outdated reference. Use the formula actually stored in I76, shown above, rather than copying that explanatory label as if it were the active formula.

Since I75 is 12, substituting 12 for that reference preserves the vendor sample’s inputs:

=FORECAST.ETS.STAT($B$4:$B$60,$A$4:$A$60,1,12,1,1)

Its expected value follows from that identical input substitution, 0.251; it is not a separately executed test. With the cached I76 value, =ROUND(I76,2) independently gives 0.25 for display. Retain full precision in calculations and label the figure “alpha,” rather than calling it a 25.1% forecast-accuracy score.

Alpha is a smoothing parameter. A larger value means recent observations influence the level more strongly; it is not the probability that the forecast is correct. Likewise, beta and gamma describe different model components and should not be interpreted as the same metric.

Create a column of selector values 1–8 and place each beside its diagnostic name. A copied STAT formula can reference that selector column while keeping history and model settings absolute. Preserve the units and labels: RMSE and a percentage-based error measure are not directly comparable numbers merely because both are small.

The verified source fixture provides the alpha result above. No specific beta, gamma, error metric or step-size value is claimed from that fixture without a corresponding verified saved result. A diagnostic panel is a useful application pattern, not evidence that all eight outputs were tested here.

A timeline with no identifiable regular step produces #NUM!; mismatched values and timeline sizes produce #N/A. Invalid seasonality also produces #NUM!. Only selectors 1–8 are documented; avoid relying on an exact unverified error code for other selectors.

Resolve repeated timestamps deliberately before analysis because the documentation contains conflicting duplicate-handling descriptions. Keep the history and optional settings identical when comparing diagnostics with a forecast. Changing a model’s inputs while comparing its old statistics can create a misleading impression of improvement.

Microsoft’s reference checked on October 17, 2026 lists: Excel for Microsoft 365 Excel for Microsoft 365 for Mac Excel 2024 Excel 2024 for Mac Excel 2021 Excel 2021 for Mac Excel 2019 Excel 2016. Microsoft’s alphabetical catalog, checked October 17, 2026, identifies Excel 2016 as the introduction-version marker for FORECAST.ETS.STAT. This perpetual-release marker does not establish an earlier Microsoft 365 preview or rollout date. Platform restrictions remain separate. The last functioning version is unverified; no retirement is asserted.

The individual reference explicitly excludes Excel for the web, iOS and Android. Use a supported desktop build. The sample workbook’s saved outputs do not establish which current build will reproduce every last decimal. Example verification: ETS fitted results are explicitly Microsoft-workbook cached results, not independently fitted predictions. Only identified arithmetic, reference substitution and display checks are independently derived. These examples have not been tested in Microsoft Excel. No screenshot or native execution is claimed.

ROUND in Excel | COUNT in Excel | IF in Excel | AVERAGE in Excel

• Microsoft FORECAST.ETS.STAT reference, retrieved October 17, 2026.