VC Fund Performance - Selection Bias

Some of the comments to my first post on VC Fund Performance talked about "selection bias". Selection bias means that the data is skewed because some funds do not report their data. It could be that the worst funds don’t report. Or it could be the best funds don’t report. Venture Economics, which is the source of the data I published yesterday, relies on funds to voluntarily report their data and so it could be subject to selection bias.

On the other hand, Cambridge Associates compiles benchmark data based on the funds their clients invest in. If you take money from an investor that uses Cambridge Associates as an advisor, your fund performance will be in the Cambridge Associates benchmark data.

So I thought it would be interesting to look at Venture Economics data versus Cambridge data to see if we can see the selection bias. Here is a chart of total value divided by paid in capital (TVPI) for funds that were raised from 1981 to 1997 (again I stopped at 1997  so we can focus on fully realized funds).

Tvpi_comparison

What you see from this data is that the two benchmarks are very similar (with the exception of ’94 and ’95 when Cambridge’s benchmark was much higher).

This says to me that selection bias is not really all that big of a deal in these benchmarks. What do you all think?

#VC & Technology

Comments (Archived):

  1. VC Data Junky

    Fred, I believe you have some mislabeling. The Venture Economics data appears to be the pooled average (not far off the top quartile in most vintages), not median. LPs would be thrilled if that was the median since the median has never in recorded history exceeded 4x. If you posted the median the Venture Economics numbers would be a fair bit lower from about 1986 on.Cambridge is an interesting comparison, and I think your note about the selection bias due to it being their client base is valid. If we assume they would only put their clients in funds that are generally above the median in previous funds (would be very poor advisers otherwise – unless they are big gamblers) then, if we believe the serial persistance argument, their funds are more likely to be above the median in the future producing an upward bias. If you plotted against the median I think it would be more visable in the graph.

    1. fredwilson

      vc data junky – you are right. this is pooled average data. i started with mean and then changed the numbers and forgot to change the labels. my bad. i am sorry about that. but since it’s apples to apples, it doesn’t impact the point i am trying to make.you may be right that cambridge data is subject to selection bias since they tend to recommend better managers. but that would suggest that the VE data tends to reflect the better funds too.fred

  2. Steven Kane

    I agree with a comment on your previous post — that, separate from “selection bias”, the real issue isn’t VC returns, but VC returns compared to other investing strategies (assuming, with common sense, that dollars must get invested somewhere, not nowhere.)So for me, the fact that top decile or quatrile funds outperform the herd is interesting but irrelevent, because even the top decile firms only manage so much money in the end, and the vast, vast majority of LPs simply can’t get into those funds and never will.It’s like saying, my baseball team could win if we only had $350 million to sign A-Rod.Well, we don’t have $350 million to sign A-Rod so we better come up with another strategy.Likewise for LPs. Even including the top performers, VC as an asset class underperforms essentially even the most conservative investing strategies. Its not that risk-taking isn’t amply rewarded; its that risk-taking isn’t rewarded at all.So IMHO the average LP is like the average mutual fund investor and similarly should listen to John Bogle’s always frustrating but always wise advice — a strategy of hunting for the best managers almost always results in huge fees and little if no market beating returns. There’s a kind of scary mad rush going on in the LP world these days, as everyone careens away from traditional strategies and tries to mimic Harvard and Yale. But unless you live in Lake Wobegone, everyone can’t be above average and so pretty simple math tells us that a bunch of pensions and endowments will probably get bruised.

    1. Koz

      Is it a reasonable assumption that the fund’s lifecycle is always 10 years?If so the annualized return for the asset class seems to almost always be between 7.5 and 12%? That’s lower than I thought it’d be based on the volatility and risk profiles… Do you have any references to the long-term average annualized returns for VCs?

      1. fredwilson

        the fund life is 10 years but the cash comes in over 5-7 years and comes out usually in years 5-10 so the average “duration” of a venture fund is more like 5 years than 10 years. meaning the IRRs are closer to 15% to 20% net of fees and carry.fred

        1. Koz

          Thanks for that fred, I figured I was missing something 🙂

  3. Steven Kane

    Fred, is the data shown here (and the other data discussed in this meme) before fees or after fees? Typical VC fund takes 2% committed capital per annum, or 14%-20% off the top just in management fees, so if the data is not after fees performance, the story is actually pretty bleak…

    1. fredwilson

      steve, this is net to the limited partners, after fees and carry are taken out

  4. Ggekko

    Considering these figures, why should LP’s invest in VC when others investments such as buy out make average return between 15 and 20 % for the median ones ?

    1. fredwilson

      the median returns on VC are pretty similar to the media returns on buyouts.

  5. LP

    The old numbers are fine but as you get into the more recent vintages (99 and later) which I would argue are more relevant these days when evaluating managers, ‘n’ drops DRAMATICALLY and the VE numbers become completely useless. For ’01 they have 54 funds, ’02-19 funds, ’03-15 funds, ’04-19 funds, ’05-10 funds, ’06-16 funds.

  6. JayR

    I would think the selection bias will be more significant when you look at the funds after the Class of 1997, simply because of the proliferation of new funds that came in over the next three or four years, many of whom did not do well enough to raise second funds.