Here is venture investor Steve Kim on data-driven insights from the Venture Stories podcast.
We model that 30% of the startups that we invest in will go to absolute zero. They won’t return anything. Then between 60 and 69% will have a reasonable expected value. And that reasonable expected value is going to fit a power law curve. Then we have one or two percent that will be outliers — model outliers. Then we run Monte Carlo simulation so we can have a strong intuition of what path dependency means. This gets a lot more complicated in how we think about venture which is whether something is ergodic or non-ergodic, but there is path dependency in the investment thesis.
I first learned about ergodicity from Nassim Taleb, but has a lot of applications. In general, an ergodic system will visit all of its states. Like flipping a coin: heads or tails. A non-ergodic system will not, because the space of possible outcomes is so huge. So it is really hard to make predictions based on past data.
Taleb gives the example of Broadway plays when predicting survival. The best prediction is how long the play has already existed. Who knows how long it could go. A play like Grease of Cats can be expected to last much longer. Shakespeare may exist for thousands of years. This application of non-ergodic systems is also called the Lindy Effect:
… which states that the longer a non-perishable item has been around, the longer it’s likely to persist into the future. The Lindy Effect was named after a New York deli and originally referenced the career prospects of comedians.
Steve Kim didn’t really get into what non-ergodic systems have to do with all this. But Steve said in this power-law world, where you have massive outsized impacts, you need very high exposure and to let companies play out over time.
Whereas most investors might have about 30 firms in their portfolio to achieve “diversification” under an assumed Gaussian distribution, his firm has exposure to 1,000 startups to take advantage of the power-law (or extreme logorithmic) distribution.
One thing I thought was interesting is that Steve said they use models to get an intuitive sense, rather than to generate decisions. That’s very different than how defense analysts tend to look at their models.
Of course you’re going to need investors across asset classes. Later stage companies and assets like bonds might be better modeled with a Gaussian distribution. But the generative nature of venture is an important driver of economic change.
This is perhaps the point that is largely missing in DoD acquisition. The S&T arena isn’t supposed to be creating full-up programs, equivalent to startups. S&T derisks technology and does experimentation. Someone else is supposed to scale it. So the way DoD thinks about Budget Activity 6.4 and Milestone A programs should be much more like venture. Start a lot more efforts to take advantage of the winners. Focus on winners, because those multiples create decisive military asymmetries. Perhaps at Milestone B or C do you shift to a more risk-averse mindset.
Here is Steve Kim on discerning whether a leader has the right mindset for this early-stage (pre-seed through series A) class of investment.
The first question we tried to get answered is how they think about the venture asset class. If they think it looks like a Bell shaped asset class, and most asset classes are, or do they think it isn’t. If we have a conversation with a [general partner] and the first thing that comes out of their mouth is ‘we don’t have any losses in our portfolio,’ or ‘our loss ratio is really low,’ that usually signals that they don’t think the asset class is skewed or power law driven. They think of the asset class from a private equity buy-out lens.
Leave a Reply