An Empirical Examination of Sample Selection Methods in the Context of Life Insurer Financial Distress

This study empirically examines properties of matched-pair versus non matched- pair sampling methods in the context of financial distress for the U.S. life insurance industry. While the majority of prior insurer insolvency studies employed matched-pair sampling techniques to identify important variables and to classify and predict firms likely to become financially distressed, we provide empirical evidence that three solvency-related items are sample dependent: variables identified as important measures of insolvency, coefficients, and classification rates. Thus, empirical studies employing relatively small matched-pair samples are likely to yield sample specific results that are not fully generalizable to the relevant population of firms. Results apply directly to financial distress models and also extend to other research employing choice-based sampling methods that involve binary state models with skewed distribution of the two states of interest.