Venture Capital and Regression to the Mean

Several articles have been written since the end of Q1 2017 that have circled around one of three themes: 1) the decline in venture investments; 2) a pause in allegedly frothy valuations; and 3) the colossal new funds that VC firms have raised. As frequently tends to happen when contradictory statistics like these are published, followers of the asset class initially wonder whether the bubble is beginning to burst, with the 3rd theme of massive new funds being regarded as a lagging indicator. The speculation was taken to new heights, however, with a recent VentureBeat post raising the specter of whether the entire VC asset class is evaporating before our very eyes.

The author of that article may in fact be right, yet for the wrong reasons (in a decade VC may become what we currently call “AI-powered,” with firms effectively cannibalizing themselves by funding AI companies today and ultimately metamorphosing into firms like Correlation that replace human-selected investments with artificial ones). Automation notwithstanding however, the author’s errors lie in his prognosis of a decline in early-stage deal funding, new fund size increases, and large late-stage venture rounds as symptoms of the VC asset class’s inevitable dissolution. The author is not the first to commit the frequent Texas sharpshooter fallacy, and since his article encapsulates the three themes I cited in the first sentence of this post, it’s worth examining the evidence to see why it’s wrong.

First, the decline in early-stage deal funding. The lower quantity of completed deals from recent highs is often described as “investments continued down” or “volume drops for the fourth straight quarter.” These observations, while empirically true, are nevertheless misleading. Primarily, they conflate bearish sentiments with the statistical phenomenon of regression to the mean. Put more plainly, these statements interpret a decline in VC deal volume as the beginning of the tech bust, rather than a simple return to the average number of VC deals completed in a quarter.

Let’s take the latest quarter for which we have data, Q1 2017, as an example. In the US, 1,800 companies received venture financing, according to PitchBook. That’s down nearly 20% from Q1 2016 (2,200) and about 28% from Q1 2015 ~2,500). But looking back at 2011–2012 financings (hardly “bust” years), deals completed in Q1 of each year were about 1,800 and 2,000, respectively, making last quarter’s volume appear far less ominous. Instead, venture deal activity has simply returned to its mean, whereby the heightened activity of the 2013–2015 years has been followed by the more moderate activity of 2016-present day.

Although the VentureBeat article doesn’t address startup valuations, current valuation discussions reflect a similar instance of widespread ignorance of the regression to the mean phenomenon that has been a hallmark trait of this cycle. Back in 2011, when deal volumes and valuations were on the upswing, even Steve Blank committed the fallacy, creating a 180 slide SlideShare presentation preparing entrepreneurs for “the New Bubble” and warning against “frothy and wacky” valuations. Yet Steve was comparing valuations in 2011 against recent lows during the Great Recession years of 2008–2009, which also was a time when seed rounds were not yet institutionalized and when Series A pre-money valuations were close to $6M. Thus rather than being “frothy,” these valuations were merely regressing back up to their pre-financial crisis norm.

Back to the VB article, the author additionally references the $50B in new venture funds that have been raised since 2016 as the highest amount since the 2000 tech bubble. Here, it’s inherently implied that there is some sort of causal link between the amount of new venture fundraising and an impending crash. However, the net amount of capital that’s being committed — i.e. the new funds raised less liquidated funds — is ignored. While I don’t have figures for 2016, total AUM for US VC in 2015 was $165B according to NVCA. In 2000, it was roughly $230B, again indicating that there is little irrational exuberance in the cited $50B gross number and that it is rather a normalized figure.

Finally, the article inaccurately paints the increased size and number of later-stage venture deals as fatalistic for the industry. While this trend undoubtedly raises legitimate concerns around liquidity for not only VCs and their LPs, but also for founders, early employees, and critical new hires, again there is no causal relationship between this development and a tech bust. According to PitchBook and NVCA, $9.4B was invested across 432 financings in Q1 2017, figures that are both higher than the previous two quarters. Yet as you may be able to guess by now, those 432 financings are roughly the average number of deals closed per quarter since 2010, while the $9.4B deal value follows the Q3 and Q4 2016 trends of receding from 2014–2015 highs (average deal value = ~$11.25B). As the IPO window widens in 2017 after the initial successful debuts of companies like Snap, Okta, and Cloudera, I’d expect the late-stage deal value trendline to continue regressing to 2012–2013 mean of roughly $6.3B.

Of course, this all begs a pressing question: how do we tell the difference between regression to the mean and a bust?

Specific to venture, I’d argue that a bust is defined by three characteristics: 1) VC firms struggling to raise new funds because LPs aren’t receiving satisfactory returns; 2) a widespread drop in revenue and/or EBITDA multiples of public comparables; and 3) slack in the hiring market. Ostensibly these three factors are all interrelated — VCs are unlikely to raise capital from LPs when public valuations are dropping and startups can’t hire if they don’t have money — yet the existence of one does not necessarily mean the other two are imminent. Fred Wilson has a good post that dives deeper into knock-on effects of the second characteristic. Although it’s written in 2014 and US interest rates have since risen above 0%, Fred’s logic still holds because the Federal Reserve has appropriately managed their rate increases (in my view).

Therefore without these three “bust” characteristics, declines in deal volume, isolated valuation movements, new fund sizes, and late-stage financing sizes do not independently portend a downturn. And more importantly, before rushing to shout that the sky is falling, we first ought to determine where the mean is. With the Q2 2017 venture landscape reports out in about a month, we’ll thus be able to read between the lines more accurately.