Forecasting Recession and Slow-Down Probabilities with Markov Switching Probabilities as Right-Hand-Side Variables
Levanon, Gad, Business Economics
Forecasting the likelihood of recessions and slowdowns is an important issue for many who use economics in business. While there has been a large and growing body of literature on the likelihood or recessions, there has been little on slow-downs.. This paper presents a new method that uses Markov switching probabilities as right-hand-side variables. The advantages of doing so are described, as are the results of empirical work. These results suggest that the approach presented here is worth adding to economists' tool kits.
Business Economics (2011) 46, 99-110.
Keywords: forecasting, recessions, leading indicators, Markov switching, business cycles
Over the past two decades, there has been a large and growing body of literature that tries to develop methods for forecasting recession probabilities. In this paper, I am proposing a new, detailed, multistep method for forecasting recessions in which some of the steps are drawn from the literature and others are new.
As in many other papers, I use a probit model with a binary variable getting the value of one during recessions as the dependent variable. One of the original contributions of this paper is using Markov switching probabilities for individual indicators as right-hand side variables in the probit model. The main reason for doing so is that the Markov switching probability summarizes quite well the weakness of an indicator during a certain month in one number. This allows me to limit the number of variables and parameters in the forecasting equations, and therefore reduce the risk of over-fitting.
There are two underlying features in the methods I am using to forecast recession probabilities. One is to allow a regression to determine the weights of each individual indicator. The second is to avoid over-fitting in the forecasting equations by limiting the number of parameters being estimated.
In this method, I first estimate recession probabilities forecasts using one indicator at a time, and then I pool these forecasts into an aggregate forecast. The pooling is done in such a way that the weights of each indicator are determined by a regression rather than by predetermined weights. The pooling is done in two steps. I first aggregate the individual indicator forecasts into forecasts of three subgroups: (1) economic activity indicators, (2) sentiment indicators, and (3) financial indicators. In the second step, I pool the subgroup forecasts into one aggregate forecast. I am using two steps in order to reduce the number of parameters in a single equation, and thus reduce the risk of over-fitting. (1) In addition, there is value in observing the forecasts of the subgroups themselves.
While the literature on recession probabilities is extensive, very little has been written about forecasting other periods of economic activity. In addition to forecasting recession probabilities, I am attempting to forecast periods of economic slow-down, defining slow-down as periods of rising unemployment rates.
The first step in this method is to select indicators for forecasting current month and future recession probabilities. Relatively few, hand-picked variables were used. One of the purposes of this paper is to include in the model variables that were first released in the 1980s and have rarely been included in other studies related to recession forecasting. The inclusion of these new variables significantly improved the forecasting results reported here. Among the new variables, two in particular stood out: the two-year interest rate swap spread and the number of employees in the temporary-help industry.
In estimating the forecasts from individual indicators, I considered three different specifications. The first is a simple one, with just the Markov switching probability as a right-hand-side variable, together with a constant. In the second specification, I wanted to capture the serial correlation nature of the recession variable by allowing the current state of the economy to impact the recession probability forecast. …