This paper examines the predictive power of credit spreads from the corporate bond market. The high-yield bond spread and investment-grade spread can explain 68 per cent and 42 per cent of output variations one year ahead, while the term spread based on government debts can explain only 12 per cent of them. For output forecasts up to one year ahead, the corporate bond spreads also outperform popular indicators such as the paper-bill spread, federal funds rate, consumer sentiment index, Conference Board leading indicator, and the Standard & Poor's index both in-sample and out-of-sample. The forecasts from the high-yield spread are more accurate than those from the investment-grade spreads. For forecasts beyond the one-year horizon, the term spread and the federal funds rate dominate the corporate spreads. The author finds that linear models based on stock market movements, the risk-free short rate, and the term spread can explain only 7 per cent of the variations in the high-yield spread. The credit channel theory in monetary economics suggests that the functional form should be non-linear. Statistical tests reject the linearity assumption for both corporate spreads in favour of a threshold non-linear specification that is consistent with the credit channel theory. The threshold models explain 63 per cent of the variations in the high-yield spread and 75 per cent of the variations in the investment-grade spread.