Usually windowing is done to smooth your time series and thus reduce noise and let you see trends more clearly in your data. A larger window gives more smoothing but obscures high frequency features. If your interest is in predicting beyond the end of the time series, there is no advantage to windowing other than perhaps letting you choose a good fitting function. Just do a least-squares regression fit to the raw data and extrapolate. Beware, accurate prediction is difficult, especially of the future.

Thank you so much, Sir, Julien Clinton Sprott , .Agreed but i want a reference article for this what you stated. because I'm writing my article and need a reference for to defend my statement '' short time window does not have not as much of a significant linearity which thus, does not lead to an improvement in prediction''

I shall be thankful if you can refer me some reading with your generous experience.

Try W. H. Press, et al., Numerical Recipes: The Art of Scientific Computing (Cambridge Univ. Press) and references therein.