46% IRR, crowd-enhanced angel returns and the future of crowdfunding

I had a hunch some time ago, that failure in startups could be predicted with certain quantitative measures.  You get a form of "failure ESP" after you've been in enough of them.  But are there bits of information that an algorithm can digest to predict startup failure?  Apparently so.

After creating some script magic, I ran my thesis against the CrunchBase database (kind of a crowdsourced wikipedia for startup information).  The run showed a promising correlation between the failure predictor (I call it the f factor) and startups which hit the "dead pool".  It was suggested to me to run a logistical regression, to find the correlation between my failure predictor f, and the binary outputs of fail/win.  The correlation results in hand, one can then plug the predictive advantage into the Angel Investor Performance Project (AIPP) data, and simulate the returns using a healthy sized portfolio.  The improvement takes the returns from an average 2.4x payout multiple and IRR of 30% to a multiple of 3.8x and a 46% IRR!

Now that makes a bucket of assumptions, e.g. how the predictability of f is distributed across investments, etc.  At any rate, the bigger point to make here, is that even using a crowdsourced database can enhance returns.  Adding in prediction markets in a crowdfunding environment, where all the players are involved, could increase this even further.  As I mentioned in the epiologue of the book:

"Human versus the machine: will we create algorithms to make early picks of winning startups, which are better than many humans (like has been done in chess)?"
Disclosure: no positions