|
The U.S. mortgage market is a significant market with more than $13.7 trillion in outstanding debt. Despite this significance, many mortgage risk evaluation models fail to identify the extent of potential losses. These Value-at-Risk models generally appear to be biased downward. Such biases may be introduced when random variables are replaced estimates of their expected values in models. Thereby understating variances and tail lengths of estimated loss distributions. Other credit loss model failures include too much reliance on estimated variance measures and reliance on over-aggregation of data which shorten loss distribution tails by overstating portfolio diversification effects.
By focusing on the loan-level, joint estimation of mortgage payment states (including prepayments and defaults), less biased estimates of the tails of a loss distribution can be obtained. Combined with simulation, the distribution of extreme losses can be modeled for further inference and analysis. This approach may be termed Distributional Value-at-Risk and is similar to Conditional Value-at-Risk but supports calculation of higher level moments in the tails of the loss distribution.
|
