In this study, we propose a two-step method to circumvent the “announcement lag” problem of the National Bureau of Economic Research recession dummy, which may not be suitably dealt with by existing studies on the dynamic Probit model. Specifically, we first identify the sub-sample of the recession dummy, which is unobserved because of this problem, using the information contained in certain macroeconomic indices. Then, we estimate the parameters of the dynamic Probit model based on the interest rate spread tween ten-year and three-month using the identified recession probabilities, and make the out-of-sample predictions accordingly. We show that our model is not only useful for circumventing the announcement lag problem but also for refining the performance of some alternative models.