Various funds claim to have good predictive models for startups. Smart people who have looked at this problem are skeptical.
- It is very hard to get your hands on solid historical data. The publicly available data is incomplete and often just wrong. Some big VC firms have pretty large & accurate data but even those datasets are likely way too small (see below).
- There is so much randomness in startup returns that non-massive datasets have more noise than signal.
- The returns are driven by a few hits, meaning the models end up just saying “invest in companies like those couple of winners”. Whatever that means.
- The world keeps changing, screwing up any model you might have built from historical data.