ENH: Add a method to append new data to an existing model to simplify updating out-of-sample observations

[ddtpower3]

Hi, proffessor.

Is there a functon that can make arch model fitted with histroy data do forecasts with new observations? For now, once new observations come, the model must be rebuilt with data frame including new obs to make latest forecasts.

Thanks

[bashtage]

There is no explicitt append feature. This would be good. For now, the "solution" is to create a new model and then to use forecast from the new model passing in the old parameters. You may need to manually set the model's backcast value as well to ensure it is identical.

[quantvol]

Hi, I'm trying to do this manually. When you say set the model's backcast value, is that done via volatility.forecast() with the backcast parameter? And if one wants to forecast an arch_model object, then pass the volatility forecasts through the mean model? My goal is to update the model with new data and pre-fit parameters like OP, but I'd also like have a way to fully reproduce a model's state with minimal inputs.

Thanks!