MULTIVARIATE NONLINEAR FORECASTING Using Financial Information to Forecast the Real Sector

Abstract

Previous work shows that financial series contain important information on the current state of the economy and expectations for the future. Further, numerous papers find links between the financial sectors and the real sectors of the economy. We add to those findings by exploring whether financial variables help to forecast the growth rate of industrial production. We evaluate linear and nonlinear forecasting methods using out-of-sample forecasting performance. We compare autoregressive models, error-correcting models, and multivariate nearest-neighbor regression models, and we explore the use of optimally combined forecasts. We find that no single forecasting technique appears to outperform any other method, and the evidence for persistent nonlinear patterns is weak. However, although nonparametric methods do not offer significant improvements in forecast accuracy by themselves, more accurate forecasts are obtained when the nonlinear forecasts are optimally combined. Our results indicate that financial information can statistically improve the forecasts of the real sector in these combined models, but the magnitude of the improvement in root-mean-squared error is small.