Abstract

In trying to explain the “Credit Spread Puzzle”, we empirically examine two competing structural models: the Leland (1994b) constant volatility model and the Perrakis and Zhong (2013) Constant Elasticity of Variance (CEV) model. We use the Leland model as our benchmark and hypothesize that the CEV model under state-dependent volatility will outperform it. For our estimation, we incorporate firm level time series data from different markets. Our sample covers the period from 2001 to 2011. We apply the General Method of Moment (GMM) for our estimation of the parameters of the diffusion process for the Leland and CEV models respectively. In our results, we document on average a significantly negative beta, the elasticity parameter in the Perrakis and Zhong CEV model. More importantly, we find that the CEV model can fit the historical data much better than the constant volatility Leland (1994b) model across all maturities, suggesting that the state-dependent volatility can explain the “Credit Spread Puzzle” to some extent.