Has IPO Market Structure Fundamentally Changed? Evidence from Negative Binomial Regression with Structural Breaks

This paper introduces Bai-Perron structural break detection combined with negative binomial regression to model overdispersed U.S. IPO count data. Using monthly data from 1995 to 2024, we identify five breaks that partition IPO activity into six distinct regimes, each with fundamentally different variance characteristics. We then employ negative binomial regression that incorporates these breaks. IPO data show substantial overdispersion (variance-to-mean ratios: 2.77 to 33.74). The negative binomial model reveals that market uncertainty (as measured by the VIX) and financing costs (as indicated by 10-year Treasury rates) reduce IPO activity, while lagged IPO volume drives activity in the current period. Regime-specific likelihood ratio tests reveal that statistically significant overdispersion first emerges during the 2008 financial crisis, subsides during the post-recession period, and returns with unprecedented intensity after May 2020. An OLS model without the identified structural breaks incorrectly suggests positive interest rate effects.