Firm Survival and Exit Analysis

A logistic model on a survived-or-not indicator ignores when firms failed and mishandles firms still alive at the end of observation, both of which bias the results. The hazard model uses the full timing information and contributes the survivor probability for censored firms, so it estimates the risk of exit without discarding the survivors. Audretsch and Mahmood adopt the hazard approach precisely because it captures how the risk of failure changes with firm age across a founding cohort, which a single binary outcome cannot represent.

A declining hazard means that, conditional on having survived so far, a firm's instantaneous risk of exit falls as it gets older, consistent with the liability of newness in which young firms are most vulnerable. However, an aggregate decline can also reflect selection: frailer firms exit early, leaving a hardier surviving population. Distinguishing genuine age dependence from this compositional effect requires modeling unobserved heterogeneity, which is why frailty terms and careful interpretation matter in firm survival analysis.

Exit from the data is not the same as failure. A firm that is acquired or merges leaves the population for reasons very different from bankruptcy, and lumping these together inflates measured mortality and confounds the analysis. Geroski's industry-dynamics perspective stresses that entry and exit are heterogeneous processes, so the cleanest practice is to define exit modes explicitly and, where the question concerns failure specifically, use competing-risks models that let acquisition and dissolution have separate hazards.