# forecast

## Syntax

## Description

forecasts the response at each horizon step in `forecastY`

= forecast(`Mdl`

,`PastTbl`

)`Mdl.Horizon`

beyond the
time step of the latest observation in the past data `PastTbl`

. Before
forecasting, the function uses the data `PastTbl`

to prepare lagged
predictors. Then, for each horizon step in the direct forecasting model
`Mdl`

, the function uses the corresponding model in
`Mdl.Learners`

to forecast the response.

This syntax is appropriate when `Mdl`

does not use leading exogenous
predictors. That is, `Mdl.LeadingPredictors`

is empty.

returns forecast responses using the past exogenous predictor data
`forecastY`

= forecast(`Mdl`

,`PastX`

,`PastY`

)`PastX`

and the past response data `PastY`

. This
syntax is appropriate when `Mdl`

uses nonleading exogenous predictors and
lagged response variables as predictors, but does not use leading exogenous predictors. That
is, `Mdl.PredictorNames`

and `Mdl.ResponseLags`

are
nonempty, and `Mdl.LeadingPredictors`

is empty.

specifies leading predictor data at time steps beyond the past data, in addition to any of
the input argument combinations in previous syntaxes. This syntax assumes that
`forecastY`

= forecast(___,LeadingData=`leadingData`

)`Mdl`

uses leading exogenous predictors, and that
`max(Mdl.Horizon)`

is greater than or equal to
`min(Mdl.LeadingPredictorLags) + 1`

. To identify the variables that
`leadingData`

must include, use
`Mdl.PredictorNames(Mdl.LeadingPredictors)`

.

## Examples

## Input Arguments

## Output Arguments

## Limitations

When you use the

`forecast`

object function, the past data must contain at least`Mdl.MaxLag`

observations. The software requires these observations for creating lagged and leading predictors.

## Tips

When

`Mdl.LeadingPredictors`

is nonempty and`max(Mdl.Horizon)`

is less than`min(Mdl.LeadingPredictorLags) + 1`

, you do not have to specify`leadingData`

. In this case, consider whether to specify the leading exogenous predictors as nonleading exogenous predictors when training the direct forecasting model.

## Version History

**Introduced in R2023b**