Recursive estimation and forecasting

How can i improve my forecasting?

Each time I estimate a model I should do a one-step ahead prediction, .e.g, predict first 21th observation then, re-estimate the model inlcuding this time period, and do a new one-step ahead prediction,now for the 22th observation etc up to 100

How can i improve this script to achieve my job?

Mdl = arima(1,0,0);

Mdl = estimate(Mdl,data);

res = infer(Mdl,data); % Retrieve inferred residuals

foreValues = forecast(Mdl,1,'Y0',data','E0',res) % forecast

thank you

Answer 1

[Pavl M.]

OK. Good. There are really quite many quite very interesting questions here.

In order to answer in full to the question please provide what is the full historical data of the model,plant?

In your rolling forecaster, from where actual new data about the model and underlying plant will come each forecasting iteration?

Whether from the residuals?

Or each time you modify your model?

Are you going to forecast in each iteration from the model produced from the same initial input data or how you augment your model with new data for forecasting coming from the same initial input data will not improve your forecast.

The model estimate just interpolates. The extrapolation accuracy depends on interpolation ultimately for same extrapolation scheme used in each iteration.

On same data the forecasting will not be more accurate than if you right from the start catch as much data about the model with better than arima model (try ssest) and forecast till 100.

Answer: make your model as accurate and as full right from the start