He isn't using it to predict far outside the data, for which a polynomial will capture the important trends. 2013 is pretty close to the data, and to 4th order, it does match a sinusoid. So, I'd say it's your fault for trying to apply his predictions outside of the bounds that he did both implicitly (the graph) and explicitly (the 2013 quote).

There's no need to be angry just because he didn't do it the way you would have.

But can you really just use curve-fitting to extrapolate, without any theory of why the data should follow a curve? Would the data follow the same curve if you had 100 years of data? 200?

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.

It seems to be the sort of mindless numerology that social science types seems to think makes them "scientific"; business and management are to social science what engineering is to physics.