New Time Series Models for Corporate Bond Log Yields

Abstract

We propose a class of simple time series models for rates and spreads of portfolios of corporate bonds classified by ratings provided by Bank of America. We evaluate these models based on statistical analysis of innovations: Whether they are independent identically distributed. Our tests are unusually rigorous, compared with standard practice, Our novelty is taking logarithms of rates or spreads instead of rates or spreads themselves. We find out that the best option is to take logarithms of spreads of logarithms of rates. Sometimes, dividing these innovations by the volatility index for stocks makes them pass our statistical tests. It is remarkable that stock volatility can also serve as bond volatility.

7 pages, 2 figures. Keywords: stochastic volatility, autoregression, Ljung-Box test