Book Description

This dissertation consists of two essays on forecasting real GDP growth and predicting recessions in the United States. In the first essay, we create a new indicator of economic activity based on a business cycle pattern, able to better forecast real output changes. The second essay utilizes the same indicator with the purpose of improving recession forecast. The accurate prediction of economic activity is valuable for the business community, policymakers, and the general public because better forecasts of GDP growth have the potential to improve economic conditions. In the first essay, we create a new indicator based on the correlation of residential and non-residential marginal product of capital (MPK) estimates and use it to improve forecasts of output growth. The correlation of residential and non-residential MPK is highly negative during recessions, while in expansions the same correlation is positive. For six out of seven expansions, the correlation of the two series becomes zero between one and three quarters before the subsequent recession. This cyclical behavior allows the use of a measure based on the correlation of the MPKs to create a better forecast of GDP growth. To this end, we compare the out-of-sample predictability of the model including the indicator against a benchmark model, and strongly reject the hypothesis of no out-of-sample predictability from the newly created indicator to GDP growth. We also provide evidence in favor of highly improved in-sample fit when the new indicator is included, and conclude that it Granger-causes GDP growth. The improvement in GDP growth forecasts is greater when an oil price measure is included in the models. The second paper employs a probit model for the US to describe the probability of an economic recession during the next five quarters, using the new indicator based on the correlation of residential and non-residential marginal product of capital. We find that in every one to three quarters prior to a recession, the correlation of the two series is not significantly different from zero, with the exception of the Great Recession. We show that models including the new measure improve both in-sample fit and out-of-sample performance when compared to nested baseline alternatives, giving accurate out-of-sample forecasts for the 1990-1991 recession. We also show that forecasts including the new indicator outperform those reported in the survey of professional forecasters, suggesting that other variables would not undo the contribution of the new indicator.