A new study compares real venture capital (VC) portfolios to a carefully constructed random benchmark and finds that, in broad terms, VC outcomes look much like what chance would produce. The authors report that the full distribution of portfolio returns is close to what a constrained random model predicts. In particular, the frequency of very large wins — the rare deals that drive most VC gains — is statistically indistinguishable from the random benchmark.

The researchers use a Crunchbase-based dataset of publicly announced funding rounds from 2010 to 2022. It covers 26,522 startups and 8,249 investors and follows investments from Seed up through Series C. For each startup they compute a “multiple”: the ratio of the amount raised in the next round to the amount raised in the current round. If a startup does not raise a subsequent round within three years, the multiple is set to zero. For each VC, the portfolio multiple is the average multiple of the startups they backed.

To test whether VCs select better deals than random, the authors build a constrained random benchmark. For each real VC portfolio they repeatedly (1,000 Monte Carlo draws) replace each deal by a random deal drawn from the same year, sector, and region, and keep the same portfolio size. This preserves the opportunity set — timing, geography and industry — while randomizing which specific companies a VC holds. Comparing the empirical and benchmark distributions shows close agreement across funding stages. Deviations are small and appear mostly in the lower part of the return distribution. A rank-based test that compares each investor to the expected outcome for that rank under random sampling finds that even the top-performing portfolios do not surpass what chance would predict.

This result matters because VC returns are driven by a tiny number of extreme successes. When outcomes are so skewed, it becomes very hard to tell whether strong aggregate performance reflects skill or luck. The study suggests that, at the portfolio level and under the constraints examined, systematic ability to increase the chance of blockbuster outcomes is difficult to detect. The authors note a related finding for financial analysts predicting earnings: performance there is also largely consistent with random benchmarks.