“… If significant risk exists in a single transaction, overall risk should be reduced by making that purchase one of many mutually-independent commitments. Thus, you may consciously purchase a risky investment – one that indeed has a significant possibility of causing loss or injury – if you believe that your gain, weighted for probabilities, considerably exceeds your loss, comparably weighted, and if you can commit to a number of similar, but unrelated opportunities. Most venture capitalists employ this strategy. Should you choose to pursue this course, you should adopt the outlook of the casino that owns a roulette wheel, which will want to see lots of action because it is favored by probabilities, but will refuse to accept a single, huge bet.”
Buffett’s recommendation fits perfectly with Wiltbank’s suggestion:
“In any ONE investment, an angel investor is more likely than not to lose their money, i.e. to earn less than a 1X return. It is risky. However, once investors had a portfolio of at least six investments, their median return exceeded 1X. Irving Ebert, of the Ottawa Angels, has done some outstanding Monte Carlo simulation with this data, finding that making near 50 investments approximates the overall return at the 95th percentile.”
As Wiltbank correctly notes, this is not an argument for “spray and pray” Angel investing. Wiltbank: “This is critical: Each investment has to be done as though it’s your only one; the bar can’t be lowered to enable you to more quickly build a bad portfolio.”
Wiltbank should focus more on the question of why the distribution of returns is a power law. The answer, as I explained in the previous post on this blog, is that underlying the data is a product of “path dependence”/ the Matthew Effect.
Power laws appear in many different contexts, but their appearance is not random but rather a signature of path dependence. Not taking that into account in establishing an investment strategy is a huge mistake since fighting the power law is like building a home at the bottom of an avalanche prone mountainside.
As to the nature of Wiltbank’s data that produces the power law and his conclusion about overall return, it is implausible that the distribution of return data would not be an even clearer power law distribution if survivor bias were fully taken into account. As one paper by another author notes:
“… In addition to selection bias, research on angel investments is plagued by survivorship bias. The idea is that investors who have failed and left the business are not sampled. This means that surviving investors are oversampled. Investors who remain in the business are likely to have been more successful than those who have left….”
Having said that about selection and survivor bias, the data is complete enough for the power law to be identified and the optimal strategies adopted.
As for Wiltbank’s conclusion: “The best estimate of overall angel investor returns from this data is 2.5 times their investment,” it is not plausible that returns are this high given the top down constraint of GDP/Income growth. In other words, Wiltbank’s bottoms up analysis needs top down reconciliation with overall investment returns. If Angels were earning 2.5X their investment there would be a LOT more Angels that there actually are in the world. Angel investing as a market is more inefficient than public markets, but is not inefficient enough for returns to be that high overall. It is possible that top tier Angels are generating these returns but that again is section bias, survivor bias and the Matthew effect at work. I am sure this is disappointing to Kauffman, who would love to see a huge ramp up in Angel investing nationwide. That is not to say that Angel investing can’t produce high returns, only that the data set is incomplete.