You've completely missed my point. You had to go to a 4th order polynomial to get a regression with a decent fit. Look at your own graph, the curve constantly changes (gasp - 3 times! polynomial 101!). Try getting rid of that third upswing in the data and then calculating a 3rd order regression on the truncated data set, then projecting the future using the regression. You'll see that your model has no predictive qualities, you'll be off by a million miles. Now think what adding one more 'point' is going to change? Absolutely nothing.
Linear regression works well, do a quadratic if the data looks exponential, but once you start getting into 3rd and 4th order regressions without knowing exactly what the [censored] you are doing and why, it's just stupid. Learn to math, friend!
edit: I don't mean to be rude, after having this stuff drilled into my head for years it pains me to see bad math. And using higher-order polynomials just so your model fits past data better with the benefits of hindsight is just bad math. The cumulative data looks exponential, use a quadratic and call it a day if you want to be genuine. But it will still be a bad model because Rob Ford and the news cycle isn't predictable.