Tracking Angel Returns - data biases?

IRRs of 20%+ sound unlikely. If angels were really doing that, wouldn’t they have followed hedge fund managers to Puerto Rico or the Caymans?

Maybe - although a 3x return on a $10,000 investment, which is the median investment made by an investor in VentureSouth groups, might not get you all the way there.

But it does seem unlikely that the returns are truly this good. This isn’t because angels are (or aren’t) “beating the market.” The efficient market hypothesis says beating the market is impossible, as prices reflect known information (all public, or all including inside information, depending on your strength). On this, EMH is definitely correct – which is why Buffett is winning his hedge fund bet so easily. But angels aren’t beating the market – they’re investing in a different market.

Still, if 22% was a consistent return, everyone would pile into that asset class, drive up deal valuations, and therefore eliminate the “excess returns.” (Or invest in more “marginal” (or nutty) companies and therefore bring the median return down.) So can it be true?

There are two approaches here: either the 22% is wrong, or we need to explain why it can be true. This post tries the first; a follow-up post with tackle the latter.

So let’s try to knock some holes in the 22%. There are likely several sources of exaggeration, because the data is (generally) provided voluntarily. Two are:

  • Self-selection bias: in general, people only report information if they think it’s notable (which would suggest greater reporting of big wins or big losses, less reporting of 0.5x-2x returns) and much prefer to report their winners than their losers.
  • Survivor(ship) bias: similarly, it’s likely that angel groups or funds that did poorly weren’t around to report their data, while successful and sustainable groups did.

How much of an impact do these biases have? In a 2012 study of hedge funds, funds that did not report to commercial databases had alpha (outperformance vs. the market) 60% lower than funds that did report – from a “truncated left tail” in the returns distribution. Commercial databases of hedge funds likely miss the worst performers. That is true for mutual funds and VC and PE (though perhaps less so as public pension fund investors have to report their data – good or bad); and it likely applies to angel too.

The authors of the angel studies are obviously aware of these biases, and try to control for, or at least analyze, them. Their tests suggested angels reported honestly (the data for syndicated deals was generally consistent between the different syndicate respondents, for example); sampling through groups makes it more likely the “failures” are included; and comparing “high response” groups with “low response” groups showed very little difference between the returns declared, suggesting “low response” groups were not just supplying their best results. 

And it may also be that as positive results take longer to achieve the “achieved” returns underestimate the eventual return – this “locked up” capital bias possibly offsetting the others. (It’s the opposite of “instant history bias.”)

So, yes, 22% is probably not exactly right. The lack of data relative to hedge funds, PE or VC makes definitive conclusion impossible. My view is that 22% is probably a touch high to be the median result, but I don’t think they are much lower than that – which is why I’m an investor in the Palmetto Angel Fund (VentureSouth's first “index fund”) and will be in the next one too.