Do Financial Market Variables Predict Unemployment Rate Fluctuations Jingyi Chen Department of Economics College of Arts Science East Carolina University M S Research Paper June 27 2002 Abstract This paper examines empirically the Granger causal relationship between financial market variables and real economic activity as measured by the unemployment rate We find in our paper that the in sample measures of fit are largely affected by one particular influential observation 1974 12 This observation accounts for superior performance of the paper bill spread in explaining the unemployment rate We then show that none of the commonly employed measures of monetary policy contain incremental information useful in forecasting the unemployment rate A simple pure autoregressive model performs better than three variable models that contain the paper bill spread the federal funds rate or M2 in out of sample forecasting The different data vintage matters in evaluating a model The fact that the results are sensitive to the different data vintage make us suspect the robustness of the Granger Causality between financial market variables and the unemployment rate concluded in the earlier studies I thank Professor Mark Thoma for kindly providing the data used in Thoma and Gray 1998 I also thank Dr Rothman for his help in supervising this project page 1 of 35 I Introduction Monetary policy is a central bank s actions to influence the availability and cost of money and credit as a means of helping to promote national economic goals Broadly the monetary policy affects the real economy through three channels 1 through the cost of borrowing in the market 2 through the exchange rate and 3 through the prices of financial assets especially equities The primary tools of monetary policy include open market operations discount policy and reserve requirements Financial market developments and asset prices thus provide useful information for a monetary policy that focuses on price stability and form an integral part of the overall assessment of economic developments required for the successful conduct of monetary policy Over the past three decades the researchers and economists have reached the consensus that the monetary policy has effects on the real activity such as the real output growth rate and the unemployment rate This paper focuses on the predictive power of the financial market variables on real activity as measured by unemployment rate Since the 1970s there have been extensive studies and researching papers on the Granger causality of the financial market variables on the real economic activities These studies have broadened the money income causality literature spawned by Christopher Sims and encompassed a diverse set of measures of monetary policy more comprehensive and sophisticated empirical assessment strategies and a richer array of explanations for observed correlations between financial market variables and economic activity The common evidence presented in the previous literature is that particular interest rates and spreads not only dominate monetary aggregates as predictors of economic performance but also are remarkably powerful predictors The ensuing papers page 2 of 35 also found that the predictive power of the paper bill spread weakened during the second half of the 1980s and the early 1990s Thoma and Gray 1998 examined the predictive power of financial market variables on real economic activity as measured by industrial production They reviewed the important methodological pitfalls in the earlier studies that rely primarily on insample measures of fit Since test statistics are sensitive to some influential observations during the sample period in sample measures of fit will become misleading indicators of out of sample measures forecast errors with the presence of the extreme outliers Thoma and Gray raised their concern that the previous conclusions of the Grangercausality between financial market variables and the real activity need reassessment because these conclusions are potentially incorrect when relying on in sample measures of fit They made a striking illustration of the drawbacks of in sample measures of fit The technique employed in their paper is the rolling recursive regression with the goal of addressing the extreme sensitivity across sample periods of the causality statistics to assess the explanatory power of financial market variables As the empirical studies have suggested that the in sample measures of fit are sometimes heavily influenced by individual observations Thoma and Gray found that the observation of 1974 12 accounts for the uniformly superior performance of the paper bill spread reported in many studies The outliers present in the data in 1974 they argued could lead one to conclude incorrectly that the paper bill spread contains information generally useful in forecasting real activity In sample measures of fit do not provide reliable indicators of out of sample fit To make this point clearer they evaluate out of sample forecasting page 3 of 35 ability of paper bill spread federal funds rate and M2 The comparison of the RMSEs shows that the forecasting ability of three models deteriorates dramatically in late 1974 To address whether the financial market variables contain useful information to predict the industrial production they departed from the common practice of framing empirical exercises as horse races among those competing financial market variables Rather they address this question by comparing the out of sample forecasting power of a simple autoregressive model of industrial production to the predictive power of models that include the paper bill spread the federal funds rate and M2 What they found is that none of the financial market variables considered aid systematically in forecasting industrial production whether the variables are considered alone or in combination They also concluded that either monetary policy innovations have no significant real effects or we collectively have failed in our efforts to measure monetary policy Besides the fact that the individual observation might influence the in sample measures the particular data vintage used might matter for evaluating forecasts The economic data such as real output money stock the consumption spending etc that are publicized by the Fed are subject to revisions and redefinitions according to the information available The information set available at particulate date is called a