I would argue that using 20+ years of data to extrapolate a mere 3-4 years in the future is perfectly acceptable. His coefficient of determination is quite hig

I would argue that using 20+ years of data to extrapolate a mere 3-4 years in the future is perfectly acceptable. His coefficient of determination is quite high so I'm pretty comfortable using that model. Of course you have to be cautious when making predictions with any cyclical data, but I believe his methods are adequate.