RDP 9606: The Information Content of Financial Aggregates in Australia 1. Introduction

Policymakers base decisions on the expected behaviour of inflation and real output, using information from a wide range of macroeconomic indicators. Measures of financial aggregates often figure in discussions of monetary policy, but the usefulness of these variables as macroeconomic indicators, in Australia as in many other industrialised countries, is considered debatable. In the mid 1980s, changes in the regulation of financial intermediaries and various innovations of financial products altered the perceived relationship of financial aggregates with real output and inflation. Proponents of the importance of these aggregates argue that monetary authorities should nonetheless monitor the financial aggregates closely as they may still provide valuable forward-looking information.

This paper addresses the question of the usefulness of financial aggregate measures in policymaking by proposing a minimal standard on the information value of financial aggregates. The usefulness of monetary and financial aggregate measures can be judged by how the information contained in these data helps forecast the subsequent behaviour of output and inflation. This view is also expressed by Friedman (1996) ‘… the whole concept [using monetary aggregate information as an information variable] is senseless unless observed fluctuations in money do anticipate movements of prices, or output, or whatever constitutes the ultimate objective of monetary policy: What would it mean to exploit an information variable that contains no relevant information?’

To investigate the forecasting value of financial aggregates on output growth or inflation, we employ the vector autoregression methodology. This empirical strategy is useful in summarising the dynamics of a small economic model. In this way, we can examine the interrelationships between the financial aggregates and policy goals, as well as take into account other important variables, such as interest rates and exchange rates. The methodology allows investigation of correlations among the data without imposing strong exclusion restrictions on lags of the chosen variables. The motivation for this approach is to uncover correlations in the least restrictive setting; that is, one that does not rely on the imposition of a single theoretical structure. This has the advantage that any correlations uncovered are not dependent upon the chosen structural restrictions.

We employ four different financial aggregates to conduct this investigation: currency (CU), M3, broad money (BM), and credit of all financial intermediaries (CR). Prices and output are measured by the underlying CPI and real GDP(A). Initially, the financial aggregates are investigated in bivariate systems: that is, using a financial aggregate and either real output growth or inflation in a system. The systems are then expanded to three variables containing the growth of the financial aggregate, inflation, and the growth of real output. Subsequently, the system is expanded further to include the differenced interest rate (90 day bank bill rate) and the rate of change in the exchange rate (trade-weighted index).

The initial in-sample VAR results suggest that financial aggregates are not particularly useful for predicting either real output or inflation. Tests of exclusion restrictions (F-tests and block exogeneity tests^[1]) of lags of the financial aggregates indicate that in a reduced-form setting there are few instances where any of the financial aggregates appears useful. Evidence from variance decompositions is used to investigate further the explanatory power of financial aggregates for forecasting real output and inflation. Three different specifications are used to generate the variance decomposition evidence, varying the sample period and the identification ordering. We fail to find any results in support of an informational role for financial aggregates that are robust across all three settings.

The above in-sample results indicate the correlations in the data, and document the usefulness of financial aggregates in an artificial setting. Policymakers, however, are more interested in whether the information in financial aggregates can help forecast output and inflation in real-time settings: that is, when we forecast today values that will only be known at some time in the future. To mimic this problem faced by policymakers, we generate tests of the accuracy of out-of-sample forecasts of VAR systems that include a financial aggregate relative to the corresponding VAR system that excludes the financial aggregate.^[2]

For output growth forecasts, the results suggest that adding the financial aggregate rarely improves forecast accuracy relative to the VAR that excludes that aggregate. In some cases, the addition of the financial aggregate improves out-of-sample forecast accuracy for inflation relative to the corresponding VAR without a financial aggregate. But on closer inspection, the improvement in forecast accuracy occurs almost entirely in the latter part of the forecast sample, and appears uncharacteristic of the previous empirical relationship, suggesting that the result has not been stable over time.

Footnotes

Block exogeneity tests assess whether the addition of lagged values of a variable are important for explaining the dynamics of the other variables in the system of equations in addition to the explanatory power of the lags of those other variables. [1]

In no sense are we pursuing the optimal forecasting model for output growth and inflation. The forecasting tools employed in this paper were selected for their usefulness as criteria for comparing the respective models, as well as for inferring the marginal forecast contribution of the respective aggregates. [2]