Issues

[Models] Expose and plot the inverse AR/MA roots of fitted ARIMA models

Nixtla/statsforecast#1338 · opened

on Nov 1, 2026

auto_arima_f with method="CSS" and non-normal distribution fails silently

Nixtla/statsforecast#1333 · opened

on Oct 27, 2026

1

Laplace ARIMA refit re-estimates scale instead of keeping it fixed

Nixtla/statsforecast#1332 · opened

on Oct 27, 2026

AutoARIMA.simulate ignores non-normal innovation distributions

Nixtla/statsforecast#1331 · opened

on Oct 27, 2026

arima: parscale is computed but never passed to the optimizer

Nixtla/statsforecast#1330 · opened

on Oct 27, 2026

[Core] cross_validation_fitted_values() fails with Polars inputs

Nixtla/statsforecast#1327 · opened

on Oct 27, 2026

[Core] predict() combines models from an earlier fit with data from a later forecast/cross_validation call

Nixtla/statsforecast#1326 · opened

on Oct 27, 2026

[MSTL] `stl_kwargs={"seasonal": ...}` raises TypeError, so the seasonal window cannot be set

Nixtla/statsforecast#1322 · opened

on Oct 22, 2026

1

[AutoETS] ETS optimiser stops far from the optimum, so AutoETS picks the wrong model

Nixtla/statsforecast#1321 · opened

on Oct 22, 2026

1

[Core] prediction_intervals passed to one call stay on the models and change the intervals of later calls

Nixtla/statsforecast#1301 · opened

on Oct 18, 2026

[Model] ARIMA forward refits report a likelihood for a re-estimated scale while keeping the original sigma2

Nixtla/statsforecast#1295 · opened

on Oct 16, 2026

[Model] AutoARIMA with method='CSS' and a non-normal distribution returns a broken model instead of raising

Nixtla/statsforecast#1293 · opened

on Oct 15, 2026

[Model] AutoARIMA.simulate ignores the fitted innovation distribution

Nixtla/statsforecast#1292 · opened

on Oct 15, 2026

[Model] Non-normal ARIMA likelihood diverges on constant series

Nixtla/statsforecast#1289 · opened

on Oct 15, 2026

[Model] ARIMA BIC and AICc count one distribution parameter for t, skew-normal and GED

Nixtla/statsforecast#1288 · opened

on Oct 15, 2026

[Model] Non-normal ARIMA intervals: fitted intervals use normal quantiles and skew-normal lower bounds mirror the upper bound

Nixtla/statsforecast#1287 · opened

on Oct 15, 2026

[Models] GARCH/ARCH point forecasts are a random draw, and the level intervals are centred on it

Nixtla/statsforecast#1280 · opened

on Oct 13, 2026

Distributed backends (fugue) are not supported with pandas 3

Nixtla/statsforecast#1265 · opened

on Oct 7, 2026

Float32 residual squaring collapses small prediction intervals and overflows large ones

Nixtla/statsforecast#1249 · opened

on Oct 2, 2026

AutoARIMA: drop collinear exogenous regressors instead of fitting them

Nixtla/statsforecast#1235 · opened

on Sep 9, 2026

1

Seasonality and trend extraction in the presence of level shifts

Nixtla/statsforecast#1214 · opened

on Aug 21, 2026

1

[Models] ConformalSeasonalPool: documented n_samples sufficiency rule covers only the lower bound and its example is off by one; no validation or warning when the pool is too small

Nixtla/statsforecast#1202 · opened

on Aug 4, 2026

1

[Enhancement] Support empirical bootstrap prediction intervals in `forecast_arima` and `forecast_ets`

Nixtla/statsforecast#1191 · opened

on Jul 17, 2026

Conformal intervals become extreme/null after inverse log transform (expm1) — what's the recommended approach?

Nixtla/statsforecast#1119 · opened

on Mar 23, 2026

6

docs (notebooks): list items not rendered

Nixtla/statsforecast#1089 · opened

on Feb 10, 2026

Navigation Menu

Repository navigation

Issues

[Models] Expose and plot the inverse AR/MA roots of fitted ARIMA models

auto_arima_f with method="CSS" and non-normal distribution fails silently

Laplace ARIMA refit re-estimates scale instead of keeping it fixed

AutoARIMA.simulate ignores non-normal innovation distributions

arima: parscale is computed but never passed to the optimizer

[Core] cross_validation_fitted_values() fails with Polars inputs

[Core] predict() combines models from an earlier fit with data from a later forecast/cross_validation call

[MSTL] `stl_kwargs={"seasonal": ...}` raises TypeError, so the seasonal window cannot be set

[AutoETS] ETS optimiser stops far from the optimum, so AutoETS picks the wrong model

[Core] prediction_intervals passed to one call stay on the models and change the intervals of later calls

[Model] ARIMA forward refits report a likelihood for a re-estimated scale while keeping the original sigma2

[Model] AutoARIMA with method='CSS' and a non-normal distribution returns a broken model instead of raising

[Model] AutoARIMA.simulate ignores the fitted innovation distribution

[Model] Non-normal ARIMA likelihood diverges on constant series

[Model] ARIMA BIC and AICc count one distribution parameter for t, skew-normal and GED

[Model] Non-normal ARIMA intervals: fitted intervals use normal quantiles and skew-normal lower bounds mirror the upper bound

[Models] GARCH/ARCH point forecasts are a random draw, and the level intervals are centred on it

Distributed backends (fugue) are not supported with pandas 3

Float32 residual squaring collapses small prediction intervals and overflows large ones

AutoARIMA: drop collinear exogenous regressors instead of fitting them

Seasonality and trend extraction in the presence of level shifts

[Models] ConformalSeasonalPool: documented n_samples sufficiency rule covers only the lower bound and its example is off by one; no validation or warning when the pool is too small

[Enhancement] Support empirical bootstrap prediction intervals in `forecast_arima` and `forecast_ets`

Conformal intervals become extreme/null after inverse log transform (expm1) — what's the recommended approach?

docs (notebooks): list items not rendered