PitchBook is Right: Networks Matter

At Level, we believe that network theory captures the underlying dynamics of the venture capital ecosystem. It should be no surprise, then, that we're delighted to see a core private market data provider (and one that is an integral component to our own research) highlight the importance of network effects. PitchBook, in their latest post Network effects: Well-connected VCs see lower failure rates, better returns, describes the importance of network connectivity in startup performance.

One of our primary applications of network theory is our Fund Manager Model (FMM). You can read about the model here, read how we handle statistical uncertainty here, see slides from a talk about it here, read our papers about networks here, and read about our training setup here. FMM is a network-centric model that predicts venture fund performance based on the relationships that a fund manager has within an ecosystem. Like PitchBook's analysis, we incorporate connectivity (i.e. how many relationships a fund manager has within the ecosystem), but we also extract: influence, connection quality, relationship to talent, relationship dynamics (how relationships change over time), and relationship directionality (if funds tend to follow or lead other funds).

Using all of these network signals, we make a robust and granular prediction of the performance of a fund with reasonable accuracy. You can see a summary of how actual fund performance compares to our predictions in the chart below.

We report performance in vintage-adjusted percentile; a predicted score of 0.85 means that we predict a fund would have 85th percentile returns with respect to other funds in the same vintage year. We run rigorous backtesting (after ensuring there is no data leakage) and we can see that our model's predictions strongly correlate with actual performance. For example, if we predict a fund is 95th percentile or above, we see that the median actual performance is 80th percentile. It's also worth noting that 25% of these funds were ~90th percentile or above.

We're excited to see that network-based approaches are becoming more and more prominent in venture and we're grateful to have comprehensive PitchBook data to power our own network science research.