REGULATORY SOLVENCY PREDICTION IN PROPERTY-LIABILITY INSURANCE: RISK-BASED CAPITAL, AUDIT RATIOS, AND CASH FLOW SIMULATION

Abstract

ABSTRACT This article analyzes the accuracy of the principal models used by U.S. insurance regulators to predict insolvencies in the property-liability insurance industry and compares these models with a relatively new testing approach -- cash flow simulation. Specifically, we compare the risk-based capital (RBC) system introduced by the National Association of Insurance Commissioners (NAIC) in 1994, the Financial Analysis and Surveillance Tracking (FAST) audit ratio system used by the NAIC, and a cash flow simulation model developed by the authors. Both the RBC and FAST systems are static, ratio-based approaches to testing, whereas the cash flow simulation model implements dynamic financial analysis. Logistic regression analysis is used to test the models for a large sample of solvent and insolvent property-liability insurers, using data from the years 1990 through 1992 to predict insolvencies over three-year prediction horizons. We find that the FAST system dominates RBC as a static method for pred icting insurer insolvencies. Further, we find the cash flow simulation variables add significant explanatory power to the regressions and lead to more accurate prediction than the ratio-based models taken alone. INTRODUCTION Increases in the frequency and severity of insurer insolvencies in the mid-1980s led to concern about the adequacy of state insurance regulation and the accuracy of the methods used by regulators to provide early warning of insurer insolvencies. [1] The National Association of Insurance Commissioners (NAIC) responded by adopting a solvency policing in 1989. The agenda resulted in a number of changes in state regulation including the adoption of the Financial Analysis and Surveillance Tracking (FAST) monitoring system and risk-based capital (RBC) requirements for both life and property-liability insurers. [2] FAST was implemented in 1993, and the property-liability insurance RBC system went into effect in 1994. Well-designed monitoring systems should identify a high proportion of troubled companies early enough to permit regulators to take prompt corrective action and should minimize the number of financially sound insurers that are identified as being troubled. Earlier research has called into question the effectiveness of the NAIC's RBC system in accomplishing these objectives. Grace, Harrington, and Klein (1998) (GHK) find that, although the ratio of actual capital to RBC is negatively and significantly related to the probability of subsequent failure, relatively few companies that later failed had ratios of actual capital to RBC within the NAIC's ranges for regulatory action. Cummins, Harrington, and Klein (1995) (CHK) confirm that the predictive accuracy of the RBC ratio is very low, even when the components of the ratio, rather than the overall ratio, are used as predictors. [3] The only prior tests of the FAST system were performed by GHK (1995, 1998). They tested the overall FAST score, a univariate summary statistic compiled by the NAIC based on the twenty-nine financial ratios constituting the FAST system. The NAIC assigns scores corresponding to a company's ratios based on a subjective evaluation of the importance of the ratios and their relationship to solvency, and the scores are summed to obtain the company's overall FAST score. Financial strength is considered to be inversely related to the overall FAST score. In their tests, the overall FAST score performs considerably better than RBC in predicting insolvencies, and the addition of the RBC ratio to the FAST-ratio prediction models leads to only modest improvements in predictive accuracy. A limitation of both the RBC and FAST systems is that they are based on a snapshot of the firm at a given point in time, i.e., they are static rather than dynamic approaches to testing (Cummins, Harrington, and Niehaus, 1993, 1995). …