How Many Parameters Are Necessary--or Sufficient? A Comparison of the Brass and Lee-Carter Mortality Models

Andrew Noymer, University of California at Berkeley

This paper compares the Brass model lifetable system and the Lee-Carter system. It is a "comparison" in a literal sense, because the effort is not to determine, e.g., which one minimizes squared error for some arbitrarily-chosen modeling task. Rather, we determine how the two systems' parameter spaces relate to each other. An easily-visualized graphical relationship of one model to the other is used to bring down-to-earth the complex high-dimensional problem of modeling human mortality. The innovation is to compare these two systems graphically. At low levels of mortality (life expectancy above 80), the Lee-Carter projection can be viewed as having an equivalent projection in the (ostensibly simpler) Brass system. However, no single model in the Brass framework can represent the changes the Lee-Carter projection undergoes as it progresses through lower values of e(0). This is clearly illustrated in the graphs accompanying the extended abstract.