Don’t Bet the Farm on Serial Persistence

The serial persistence of Venture Capital returns has become something of an article of faith within the limited partner community.  Serial persistence basically means that the best indicator of future VC performance is past VC performance (See Paul Kedrosky’s thoughtful posts on this topic here, here, and here.)  These days LPs talk about getting "allocations" in top quartile funds the way high school seniors talk about getting into their "first choice" college:  they seem to possess an almost blind faith that just getting accepted will guarantee them many years of above average returns thanks to the wonders of serial persistence.

If only it were true.  The reality is that making money investing in venture funds is a lot harder than simply putting as much money as possible behind the best performing funds that will take your money.  That’s because while top quartile performance is indeed a significant determinant of future performance, it’s not a sole or even a sufficient determinant.

Arguably the most definitive study on this issue comes from the University of Chicago and was authored by Steve Kaplan and Antoinette Schoar.  The study took a database of every venture and private equity fund in existence between 1980 and 2001 and analyzed it to see if, among other things, returns were serially persistent within GPs across funds.   Sure enough, the study found strong statistical evidence that past performance by a GP was a significant factor in explaining the variability of future performance, i.e., for individual GPs, returns were serially persistent across time.

For many in the VC and LP community, this study simply confirmed conventional wisdom in that the only “can’t lose” way to make above average money in venture was to “bet the farm” on top quartile firms.

Unfortunately, the actual numbers suggest that betting the farm solely on past performance may be much riskier strategy than one might imagine.  Elsewhere in the study, the authors’ analyze the probability that a GP with a top tercile fund will generate another top tercile fund.  It turns out that this probability is. at best, a coin-flip.   In other words, there’s only a 50% chance of a GP delivering back-to-back top tercile performances.  That’s hardly a "lock" and not the kind of numbers that should inspire someone to bet the farm.   In addition, even though the study did not divide performance into quartiles (as is often the custom in VC fundraising), I had a brief exchange with Steve Kaplan in which he confirmed that the same pattern not only holds for quartiles, but that the probability of a GP having back to back top quartile funds is significantly lower than 50%.

I would be willing to bet a large sum of money (perhaps even “the farm”) that if you surveyed 100 LPs today and asked them “what is the probability of a top quartile fund repeating that performance” that you would get an average result well above 50%, and probably closer to 75%, thanks to the prevailing conventional wisdom.  So what’s most interesting to me about the U of C study is how far off-base this supposed wisdom really is.  After all, according to the study, a new fund of a top tercile GP has just as much chance of becoming a 2nd or 3rd tercile fund as it does a top tercile fund. 

Of course, I should point out again that the rest of study, as well as these probabilities, makes it clear that past performance is definitely something that should be considered carefully when making new venture investments and something that it is more significant than many other factors, however it’s also clear that blind allegiance to past returns as a predictor of future performance is not much better a recipe for success than simply flipping a coin.  As Kaplan himself pointed out to me in our exchange, there are a wide range other factors that people need to consider when investing including partner turnover, motivation levels, style creep, etc. in addition to past performance.

Personally, I think that as the number of funds and average assets under management increases, achieving back-to-back top quartile performance will become harder and harder thanks to the inevitable dispersion of “home runs” across a wider base.  This will make it increasingly difficult for LPs to invest via their rear view mirror and will require much more forward looking due diligence on the prospective strategies of individual GPs.  Of course, this perspective doesn't play with the self-serving Sandhill party line that the best VCs will always be the best VCs, but I find such sentiments delusional because they in effect assume infinite economies of scope for the best VCs as the industry scales and those simply don't exist.

As for the serial persistence of GP returns, I buy it on an academic level, but given the probabilities involved, at a practical level I have a hard time seeing how anyone can reliably translate that academic insight into actionable decisions with absolute confidence.   In this light any LP that is making "top quartile performance" the sole reason for making an investment or the litmus test for even considering one is likely making a big mistake.  My bet is that in 20 years, outside of few exceptional cases, the VC industry will look a lot more like the mutual fund industry where serial returns are not believed to persist in any significant way.