Better to be Approximately Right than Precisely Wrong: Empirical Insights into Parameter Choice and Industry Effects in the KMV-Merton Corporate Default Risk Model
Abstract
We examine the choice of strike price and time to maturity in the standard academic application of the KMV-Merton model based on Merton’s (1974) structural credit risk model. We define the standard academic application utilises a fixed proportion of debt for strike price, and maturity is assumed in 1 year. The proxy values for each parameter we analyse are combinations of both random guesses and more sophisticated choices. We extend the thesis by introducing a conditional test on aggregated industry categories. Results are compared to (1) the standard academic parameters, (2) a naive alternative which applies the functional form of Merton, and (3) an experimental approach where we match expected default frequencies to respective maturities.
We find that the standard strike price performs better than the alternative parameters, and we fail to find a combination of strike price and time to maturity that overall yields increased explanatory and predictive power. However, when conditioning the analysis on aggregated industries, we find increased explanatory and predictive power. The best performing parameter combinations will for some industries perform better than the naive model, which the standard model could not. The functional form of the KMV model proves relevant, as it manages to capture variation in corporate bond spreads. Interestingly, the naive model proposed by Bharath & Shumway (2008) performs surprisingly well compared to more complex approaches. We believe this stems from estimation errors, particularly in volatility of asset.