Forecasting Workers Compensation Severities and Frequency Using the Kalman Filter

Estimating trend rates of growth of frequency and severity is crucial to workers compensation ratemaking. Trend growth rates can be estimated using unobserved components models and structural time series models. These two types of models derive from parsimonious and transparent data-generating processes and, in the case of structural time series models, allow the researcher to incorporate economically meaningful explanatory variables into a time series framework. When specified in state-space form, unobserved components and structural time series models become available to the Kalman filter. The Kalman filter is an estimation technique that explicitly accounts for possible measurement (reporting) errors in the frequency and severity data and, hence, is of critical import at NCCI.