正在加载视频...
视频加载失败
Recently, Garry Tan revealed some pretty incredible data on the returns for investors at YC demo day. If you invested in 3+ startups every demo day over a two-year period (2018-20), the returns you would've seen: Top 10%: 16x Top 25%: 8x Median: 5x Bottom 25%: 3.3x Compare this... show more
9 条评论

What's the optimal number of startups to invest in at YC demo day? Is it 3? Or is it the Yuri Milner / SV Angel playbook when they offered every YC startup $150k?

Over the weekend, I built an app to help model this out for Orange Collective. We call it The Unicorn Ratio. The code is open sourced. Try it out: Repo:

One takeaway: this challenges the conventional VC wisdom of making a few, highly concentrated bets as the only winning strategy. Instead, the data suggests a broader, more systematic approach to YC investing could also be a viable strategy (and possibly beating most VC funds).

The Unicorn Ratio is based on two mathematical frameworks: 1) Kelly Criterion for optimizing portfolio construction and capital allocation 2) Binomial Distribution for modeling how the probability of picking a unicorn changes based on portfolio size More here on how it works:

I learned of the Kelly model last week on a call that YC organized for some of its demo day investors. Super grateful to YC for the openness and collaboration, and for giving me an excuse to build and open-source something :) Hopefully others find it useful too.

Also, an excuse to share one of my favorite movie clips of all time:

Assuming this data is for more recent vintages of startups, the caveat is IRR paper returns vs capital returns. It seems pretty clear many vintages are stuck in valuation traps and likely will die or have down exits (often with the earliest investors getting little to nothing as they are at the bottom of the payout stack)… YC companies are still a better bet than average and I’ve invested in multiple ones myself pre/post YC, but the size of returns here for the low end seems overly optimistic?

@garrytan It’s 2018-20 vintages so avoids ZIRP craziness from Covid (which would’ve skewed the numbers even higher), but agree would be curious to see a wider sample size

@garrytan No fucking way this is true

