YC Bench: a Live Benchmark for Forecasting Startup Outperformance in Y Combinator Batches

Battle Count: 71

The paper is primarily about startup success forecasting and benchmark design, not quantitative trading. However, there is tangential relevance: (1) the concept of using short-term proxy signals as surrogate endpoints for long-term outcomes parallels factor investing and alpha signal research; (2) the ranking/selection methodology could inform early-stage venture portfolio construction; (3) the live benchmark paradigm with forward-looking evaluation is relevant to any predictive system. The connection to traditional quantitative trading (equities, derivatives, FX) is minimal.

The baseline implementation is very simple: rank startups by Google mention counts and evaluate against Pre-Demo Day Scores. The Pre-Demo Day Score itself uses straightforward max-of-weighted-metrics aggregation. Main complexity lies in data collection (SerpAPI queries, traction data sourcing from public disclosures) and the heuristic weight calibration. No ML model training is involved in the baseline. The benchmark framework is lightweight and accessible.

Code and data are available on GitHub (https://github.com/benstaf/ycbench). However, traction data is sourced from a publicly circulated analysis by Lobster Capital and LinkedIn self-disclosures for only 11 startups, making full reproduction of the traction component dependent on external sources. The Google mention data collection via SerpAPI is reproducible but requires API access. The heuristic weights are not empirically calibrated, limiting reproducibility of the exact scoring.