On Rebel Theorem and YC batch quality
A couple of years ago, I published a post On YC batch quality at scale that showed the upward trend in Y Combinator startup outcomes despite a massive increase in YC batch sizes over time. While it’s counter-intuitive to many investors, YC has managed to only maintain but improve the quality of startups it admits into the program even as batch sizes have increased from a dozen or so companies in the early years to 200+ today.
My previous post measured YC batch quality simply by the percentage of startups from each batch that made it onto YC’s Top Companies list ($150M+ valuations) or became unicorns ($1B+ valuations). While these are great measures of YC startup success, they are retrospective in nature, since it takes several years for YC startup valuations to mature.
Since Rebel Fund maintains the most comprehensive database that exists of YC startups, founders, and outcomes outside of YC itself, we’re in an excellent position to not only look at retrospective measures of YC startup quality like valuation growth, but also prospective measures like overall company and founder quality.
A few weeks ago, we announced a new and highly-advanced Rebel Theorem 3.0 machine learning algorithm that accurately scores and ranks YC startups according to how likely they are to achieve certain outcomes. While we use this algorithm primarily to screen individual startups for further diligence by the fund, it can also be used to score the quality of entire YC batches, which will be the focus of this post.
Before getting into our findings, I should caveat that Rebel Theorem scores are not a perfect predictor of YC startup or batch outcomes. While I don’t believe any better predictor exists, there are many micro and macro factors that influence startup outcomes and our algorithm can’t capture them all.
As I discussed in On Rebel Theorem 3.0, our algorithm scores startups according to how likely they are to end up in each of the following buckets:
“Success” — $60M+ valuation and operating or exited
“Zombie” — Under $60M valuation and still operating
“Dead” — No longer operating nor exited
I also showed that the algorithm is particularly good at predicting which startups will end up successful or dead — more than twice as good as random — which is much better than most human investors.
So, we decided to focus on average YC startup “Success” and “Dead” scores on a batch year level as a current predictor of YC startup quality. Here’s what we found in terms of average Success scores by YC batch year:
The clear trend here is that YC batch quality has not only trended higher historically (which makes sense given the upward trend we see in Top Companies and unicorn rates) but also in recent vintages that are too young to measure by valuation outcomes. This means that future YC startup outcomes are likely to continue trending in the right direction.
Here’s what we found in terms of average Dead scores by YC batch year:
Not only are the Dead scores trending downwards (which is good) but they’re doing so even more steeply than Success scores are trending upwards. This tells me that while YC partners are definitely getting better at selecting startups more likely to be successful, what they’re really getting good at is screening out the startups that are more likely to end up failing.
I think the reason some investors think YC’s best days are behind it is when you hear about a YC startup going public or getting acquired, it‘s often a startup that went through YC many years ago. This isn’t because YC startup quality has gone downhill, but simply because it takes ~7 years on average for a YC startup to achieve an exit, and the best exits often take the longest.
The implication here for investors is that despite all the panic about increasing YC batch sizes, AI hype, etc, if you look just at the fundamental trend in YC startup and founder quality over time, now is a better time than ever to invest in YC companies.
