The Fast and the Curious: VC Drift

Abstract

We develop a measure of a VC firm’s investment style and its change over time (drift). While drift can be beneficial for responding to new market conditions, it reduces the ability to develop style expertise. We document evidence of drift among VCs and find that it is more prevalent among VCs who are less experienced and face pressure to invest their funds. We also find a negative relation between drift and performance, with stronger effects for VCs who herd and are seasoned. Overall, our results are consistent with the hypothesis that drift is detrimental to VC performance.

Appendices

Appendix A: Style drift examples

We present some examples that illustrate and provide intuition for the various approaches that might be taken in determining style drift, and thereby explain why the approach selected in the paper is preferred.

We begin by examining why cumulative investments are better than simply accounting for the actual investment in each year. This is best understood by assuming a simple setting with just two hypothetical VC styles, and a total investment of 100 across two years. There are three options that might be pursued. One, style vectors comprise the actual investment made each year. Two, style vectors comprise the proportions invested in each style each year. Three, our chosen one, implements style vectors as the proportions of cumulative investments made by a fund in each year. In order to set ideas, assume that two VC funds make the following investments (a total amount of 50 across years) in each of two years:

	Style 1	Style 2
VC Fund 1
Year 1	48	1
Year 2	0	1
VC Fund 2
Year 1	29	1
Year 2	1	19

If we treat these templates as the style vectors in each year and compute style drifts $d_{jt} \in (0,1)$ using Eq. 1 in Section 2.1, we get the style drift of VC Fund 1 as $d_{12}= 0.9792$, and that of Fund 2 as $d_{22}= 0.9131$. Common sense dictates that Fund 2 changes its policy more than Fund 1, yet the drift measure is higher for Fund 1. Moreover, the measure is impacted by the size and not the proportion of investments.

What if the metric for drift is modified to be computed from the style proportions rather than the absolute investment amounts? The new tables appear as follows:

	Style 1	Style 2
VC Fund 1
Year 1	48/49	1/49
Year 2	0	1
VC Fund 2
Year 1	29/30	1/30
Year 2	1/20	19/20

We get the style drift of VC Fund 1 as $d_{12}= 0.9792$, and that of Fund 2 as $d_{22}= 0.9131$. Therefore, we see that using absolute dollar amounts or proportions does not change the results, Fund 2 has a lower style drift, even though it experiences a bigger reallocation of its portfolio weights across the two styles.

A final approach is to use cumulative proportions instead. We adopt this approach for this paper. The table is as follows:

	Style 1	Style 2
VC Fund 1
Year 1	48/49	1/49
Year 2	48/50	2/50
VC Fund 2
Year 1	29/30	1/30
Year 2	30/50	20/50

We get the style drift of VC Fund 1 as $d_{12}= 0.0002$, and that of Fund 2 as $d_{22}= 0.1493$. Here we see that Fund 2 now has a greater style drift than Fund 1, as intuitively desired. The drift of the portfolio is minimal as expected in the case of Fund 1 and it is reasonable as in the case of Fund 2.

We now examine a few more tableaus to gain an understanding of more complicated cases. Consider the following two VC funds with five years of absolute value investments.

	Style 1	Style 2
VC Fund 1
Year 1	98	0
Year 2	0	1
Year 3	0	0
Year 4	0	0
Year 5	0	1
VC Fund 2
Year 1	98	0
Year 2	0	0
Year 3	0	1
Year 4	0	0
Year 5	0	1

We see here that the sequence of financing differs. Converting these investments into cumulative proportions and computing their drifts results in the following tables:

	Style 1	Style 2	Drift
VC Fund 1
Year 1	1	0	–
Year 2	98/99	1/99	$5.206 \times 10^{-5}$
Year 3	98/99	1/99	0
Year 4	98/99	1/99	0
Year 5	0.98	0.02	$5.204 \times 10^{-5}$
VC Fund 2
Year 1	1	0	–
Year 2	1	0	0
Year 3	98/99	1/99	$5.206 \times 10^{-5}$
Year 4	98/99	1/99	0
Year 5	0.98	0.02	$5.204 \times 10^{-5}$

Hence, the average drift for both funds across these years is the same, as it should be. What if the rate at which investments are made differs? Take as an example investments in the following two funds:

	Style 1	Style 2
VC Fund 1
Year 1	90	1
Year 2	0	2
Year 3	0	2
Year 4	5	0
VC Fund 2
Year 1	90	1
Year 2	0	0
Year 3	0	0
Year 4	0	2
Year 5	0	0
Year 6	0	0
Year 7	0	2
Year 8	5	0

Without detailed calculations, we can see that the average drift of Fund 2 will be smaller than that of Fund 1 because it has years of zero drift that are more numerous than in the case of Fund 1. Clearly the speed at which investments are made will be related to the drift, again, as is intuitively desired.

In our model implementation we assume that funds live for 10 years on average, and the example above will result in an aggregate cumulative funding at the VC firm level across both Fund 1 and Fund 2 as follows:

	VC Firm
	Style 1	Style 2
Year 1	180	2
Year 2	180	3
Year 3	180	5
Year 4	185	8
Year 5	185	8
Year 6	185	8
Year 7	185	10
Year 8	190	10
Year 9	190	10
Year 10	190	10

The style drift is then computed for all 10 years off the aggregate proportion values. In the case when the two funds begin in different years, then the aggregate cumulative proportions will extend up to 10 years from the inception of the last fund to start.

Appendix B: Variable definitions

Variable	Description
Time-varying VC characteristics
Drift	VC’s annual drift
5-year Drift Qtle	VC’s drift quartile, using annual drift averaged over five-year window, with zero-drift VCs in a separate category.
3-year Drift Qtle	VC’s drift quartile, using annual drift averaged over three-year window, with zero-drift VCs in a separate category.
VC Age	Natural log of one plus the VC’s one-year lagged age, in years, where age is from its founding until the year of the financing round.
Synd experience	Natural log of one plus proportion of cumulative rounds that the VC has syndicated as of the year prior to the financing round.
Early stage focus	Natural log of one plus the proportion of the VC’s cumulative companies that received early stage financing, as of the year prior to the financing round.
IPO rate	Natural log of one plus the VC’s ratio of IPOs to number of portfolio companies in the last three years, as of the year prior to the financing round.
Style HHI	Natural log of one plus the VC’s style HHI, based on the number of investments in different styles as of the year prior to the financing round.
New Fund Yr	Equals 1.0 if VC raised a new fund in the prior year.
% Funds invested	Natural log of one plus the proportion of VC’s all active funds invested cumulatively as of the year prior to the financing round.
Seasoned (Young) VC	Equals 1.0 if VC’s age is at least (less than) 11 years (0 otherwise).
Herder (Contrarian)	Equals 1.0 VC firm whose style drift vector is positively (negatively) correlated with the average style drift vector across VCs (0 otherwise).
VC AUM	Natural log of one plus the sum of the VC’s all active funds under management in the prior year.
Early stage (Dummy)	Equals 1.0 if the round is an early or seed stage financing and zero otherwise.
Syndication (Dummy)	Equals 1.0 if the round is syndicated, zero otherwise.
Portfolio Age	number of years the company has been in the VC’s portfolio.
Time-invariant VC characteristics
Independent VC	Equals 1.0 is the VC is an independent VC.
Fin Inst VC	Equals 1.0 is the VC is a financial institution VC.
VC firm U.S./non-U.S.	Equals 1.0 if the VC is in the USA.
VC firm CA/MA	Equals 1.0 if the VC is in the state of CA or MA.