Difficulty increases aren't just an assumption, look at the graph of the last year. What I am saying is that as long as difficulty increases are growing at the current rate, mining will never be as profitable as investing directly. At the current rate, you have to have an increase in bitcoin value to compensate for the increase in difficulty to just break even. So indirectly, you are speculating on the increasing value of bitcoins when you mine.
Now, if the growth of computing horsepower slows down to 2% or less, then mining obviously is the better option. Like for the first year of bitcoin's existence when the growth was flat.
Mining is lower profit, but it's also lower risk. You can sell your rig, you can use it on other projects, or you could play games with it. Mining can also remain profitable even if the price goes down, especially if, as the price goes down, network growth rate decreases.
Your extrapolation of the difficulty growth is still just a guess. There is no reason why difficulty has to keep growing at that rate. It could easily fall to 2%.
In conclusion, mining makes as much sense now as it did on May 1st. Which is to say, it's anyone's guess what will happen.
You are absolutely correct that in the worst case, you still have hardware you can sell or use. In the worst case of investing directly you have nothing.
As far as network growth, I wouldn't say it
easily go down to 2%/day because it hasn't in the last year. That being said, it will have to slow down at some point because there are only so many GPUs out there. When that happens no one knows, but betting it will happen next week is a risky bet. Of course, once someone makes some custom ASIC arrays the game is over for everyone.
When it comes abstract mathematical metrics derived from the interaction of complex human behaviour, past performance does not predict future performance. This is economics 101.
Difficulty is not a physical object. It does not obey Newton's first law. Abstract, arbitrary metrics generally don't. (And when they look like they do, it's usually just because they haven't changed yet. See: LTCM. See: Housing prices. See: Tulips. When stuff looks predictable, it's usually because you're suffering from selection bias.)