RDP 9704: Financial Aggregates as Conditioning Information for Australian Output and Inflation 1. Introduction

This paper follows on the work in Tallman and Chandra (1996) by investigating further whether financial aggregates explain subsequent fluctuations in real output growth and inflation. In that paper, using the vector autoregression (VAR) methodology, Tallman and Chandra draw the general conclusion that financial aggregates show no exploitable correlations with output growth and inflation that are robust across various time-periods and specifications. However, the paper finds in an out-of-sample setting that some financial aggregates may help forecast inflation. Also, in-sample evidence from their variance decomposition analysis reveals that in certain specifications financial aggregates seem important for determining the forecast error variance of inflation and output growth. These findings, as well as the criticism that all the results rely on the adequacy of the VAR for the real output growth and inflation process, suggest that further research is warranted.

Here, we apply additional empirical methodologies to investigate further the usefulness of financial aggregate data for forecasting output and inflation. In the first method, we test the information value of financial aggregates by employing a VAR to generate a simple, artificial experiment that indicates whether foreknowledge of financial aggregates improves the forecasts of real output growth and inflation. In our study for Australia, the forecast-improvement statistics suggest that financial aggregates are not particularly useful for predicting either real output growth or inflation in the unrestricted VAR setting. The notable exception is the growth in credit as conditional information for improving the prediction of real output growth.

While the above VAR-based approach may uncover meaningful correlations in a relatively unrestricted setting, it has limitations. By limiting the number of included variables, it implicitly imposes exclusion restrictions that may ignore important explanatory relationships found in existing single-equation models. To address this criticism and to provide evidence from an empirical methodology other than the VAR, we analyse the effect of adding financial-aggregate variables to restricted reduced-form single-equation models of real output growth and inflation. To preview the results, we find no evidence that financial-aggregate data improve the fit for inflation. We find in one specification for real output growth that both the contemporaneous and four lags of credit growth explain a significant proportion of growth in real output. However, another specification for real output growth – that using US output instead of OECD output as a measure of world output – shows no evidence of the explanatory power of credit growth. Also, when we restrict the real interest rate coefficients to be zero and leave the financial aggregate as the only financial channel, no aggregate shows significant explanatory power. We suggest these negative results imply that there is a lack of robustness in the positive result for credit growth.

Taking the evidence from both unrestricted VAR prediction tests and the restricted specification, there appears no robust and potentially exploitable correlations between growth in any of the financial aggregates and real output growth and inflation. There is some evidence, however, that in periods of considerable financial restructuring, changes in credit growth may provide useful information regarding the future course of output. Outside such periods, credit growth appears to be simply another corroborating information variable.