I feel like venture capital is a major choke point in our current implementation of capitalism, where "the market will decide" is really a euphemism for "a small handful of extremely wealthy investors will decide." We're paying a massive opportunity cost on all of those ideas that don't ever reach the market because VC's don't think they're easy money.<p>A good way to resolve this would be to reform the accredited investor laws into something more meritocratic. Instead of needing to own one million dollars in assets, there ought to be some sort of knowledge-based competency exam so that regular people can invest in ideas they think are worthwhile.
This analysis is a little too "content free" for my tastes. (Edit: I don't mean there is nothing to it, I mean the analysis is largely formal as opposed to the content of the decision)<p>But this problem hasn't gone unnoticed and there are some ideas around how to solve the "pick the rare winners" problem:<p>1. Andreessen: Don't pick winners. Invest in the startup after it has already demonstrated itself to be a winner but before it goes public. This is the safe growing area.<p>2. McClure: Invest small amounts, early so that you can afford to spread the investment over many companies.<p>3. Thiel, Gurley: Be right<p>4. Graham: combination of #2 and #3<p>5. Doerr: Network like crazy to have a shot of being in the few good ones (this assumes you will recognize them)<p>#3 is not necessarily something we can reproduce
The model seems to correlate what worked previously to what will work next. However, in a field that is supposed to lead to new product funding - past industries may steer funding in the wrong direction from what will work next. Two issues:<p>1. The purple cow effect - the opportunities for highest growth may be in underserved segments which are best addressed by founders less represented in Silicon Valley. (the next big thing may be a farming startup in India, but the founder won't fit the well connected or well advised by people with startup exits model this framework uses, and thus will be missed by SV investors, which leads to problem 2. )<p>2. The money trumps product problem - whatever you can't beat with a solid product, you could always hammer with more money in the bank. Instead of hearing about a farming innovation in India, American farmers could be getting FarmVille ads on TV instead and tuning out. Since VCs invest locally, even if a startup starts picking up steam in Chicago, SF investors who don't have a toe in that pond, pick the local fish that eats the same algae and fund it far better than the Chicago company, which might have a better product. In a land-grab industry that money may be enough to gain adoption to the SF pond-dweller, but returns for the entire market will be lower, due to lower product quality and tendency of big firms here to pick only one company per industry.<p>So the framework, biased by past data, may skew future data away from results that would be optimal without a framework.
There's a few VC firms (Correlation Ventures leaps to mind, as noted in the article) that invest solely on the basis of a quantitative model that looks at the various features of the business (market, founders, etc.) and then doing some kind of neural network/similarity scoring analysis on it. Of course, a big feature that I presume their models have which this paper does not is the understanding that an IPO is worth tens of small wins, if not more.<p>The real optimal setup here would be to pair that kind of mathematical rigor with the dealflow of an a16z or KP. I would suspect that both of those two would say that a similar model exists in the heads of their partners so far as pattern recognition, but..
With zero information, "picking winners" is like buying lottery tickets.<p>It's obvious that you can win at lottery if the cost of buying all the tickets is less than the total payout. It's usually not.<p>In the case of startups, it may still be the case. For this study they examined 24,000 companies over 16 years (1500 companies per year). 24k companies x $10k = "only" $240 millions (or, over 16 years, $15 millions / year on average).<p>So, if one invested $10k in each of these companies, as early as possible (to get the maximum equity in exchange for these $10k) then they would probably have come out ahead (hard to verify as I couldn't find the total exits in $$ in the study).<p>I think this is fundamentally, and ultimately, YC's business model: root out the obvious hacks and cracks, and accept everyone else.
Despite it being so obvious, I really like their analogy,<p><pre><code> pharma::drug
studio::movie
vc::startup
</code></pre>
"Just" need one to work out!
So I briefly scanned this paper.<p>"The results show that our modeling and portfolio construction method are effective and they provide a quantitative methodology for venture capital investment."<p>I really can't find evidence for this statement anywhere in the paper.<p>This paper seems to conflate a prescriptive model with a descriptive model. A prescriptive model would be more focused on the data and less on the formal mathematical model. It would also be actually useful. It looks like they came up with an interesting way to model the portfolio selection problem, but it will simply not provide a "quantitative methodology" for VC investment as they claim. It's perplexing why that sentence was allowed in this paper.<p>If you are a VC could you extract anything that is useful and non-trivial from this paper? I doubt it.
If you look at figure 4 on page 24, it basically says that the VCs picking ability is about good as random chance. They have some really smart people working for them and its still hard to pick winners.
Recapping: top drivers roughly in rank order were: Great market, previous founder, founder worked at a company that IPOed or acquired, founder came from a top school, the quality of current investors (do they have exits). Seems like the standard things most investors look for, so the question is does it come down to deal flow.
One of the challenges here is that "sector" is strongly predictive, but the data analyzed are historical. It's easy to know which sector one should have invested in historically -- and hard to know which sector to invest in going forward.