Abstract
This work aims to predict downturns in real economic activity, by exploring a well-described theoretical linkage between the yield spread and a coming recession. Since the Israeli data are of short span and do not provide convincing evidence of forward-looking yield spread, I introduce the "standardized" yield spread, adjusted for monetary shocks, which are not found to be connected directly to real output. This prior adjustment factor is captured as a latent variable, which can be identified by a regime of conditional volatility. For this purpose, the two-regime Markov chain with switching mean and variance is applied.
Downturns are evaluated in respect of their onsets, but not their deepness or duration. The ex ante probability of decline, calculated six months ahead, is kept conditional on the current economic growth. To account for the serial correlation effect, a dynamic ordered method has been applied. The parameters are estimated via the Gibbs sampler.
The suggested approach seems to overcome the lack of linearity between the yield spread and subsequent real output. The main recessionary episodes of 1996, 1998 and late 2000 could be anticipated, by simulation results. Since the model parameters are self-updating upon the new sample data, this method can apparently be used in a current projection.
In the upward direction, only two rightly predicted upturns have been accepted. Because of a short sample, one cannot conclude whether the well-known asymmetry of yield-spread predictive power (in favor of recessionary signals) really holds.