RDP 9606: The Information Content of Financial Aggregates in Australia 2. Literature Survey

Not surprisingly, there have been a sizeable number of research papers on the explanatory power that financial aggregates have for real output and the price level. These studies employ a variety of research methodologies and there is some variation in the data sets with respect to both the component data series and the sample period. The description below draws some overall conclusions from this set of research studies and emphasises how the current research relates to the existing literature.

There has been considerable interest in Australia about the relationship between financial aggregates and price and output variables. Orden and Fisher (1993) examine the relationship between money, prices and output for New Zealand and Australia using a VAR methodology. Of the existing literature, it is among the most similar to the present study. Variance decomposition results for the period 1965:Q2 to 1982:Q4 suggest that in Australia money shocks contributed significantly to subsequent variations in prices (5 to 30 per cent of forecast-error variances), but have contributed little to subsequent variations in output. This result is notable because the implied causal ordering chosen to generate the variance decompositions and impulse response functions places prices and output before money, and therefore restricts the contemporaneous influence of financial aggregate innovations on output growth and inflation to zero. These results are in direct contrast with the variance decomposition results reported below. We note, however, that there are some important differences between their study and ours. For example, their sample period ends before the major financial deregulation, their data set employs the GDP deflator as the price level measure and uses only M3 as the financial aggregate, and the estimation is conducted in an error correction framework rather than a VAR in levels or differences.^[3]

Several studies from the Reserve Bank of Australia investigate the correlations among financial aggregates, inflation and real output growth. Bullock, Morris and Stevens (1989) employ correlation analysis on a data set of financial aggregates and output and inflation over the period 1968 to 1987. The results in their study lead them to conclude that M1 and short-term interest rates are the most useful financial indicators because they have a consistent, leading relationship to real private demand.^[4] In a follow up to that study, Stevens and Thorp (1989) employ VAR methods to detect the leading and lagging relationships among the data over the sample period 1969 to 1988. They find that GDP tends to lead broader financial aggregates such as credit of all financial intermediaries (credit) and M3, consistent with the idea that the broader financial aggregates are endogenous to the movements in real output. In addition, the study refines the results in Bullock, Morris, and Stevens by showing that M1 does not have a strong leading relationship with real output.^[5]

Blundell-Wignall and Gizycki (1992) estimate a VAR with credit data from 1976 to 1991; they find that in the post-deregulation period total credit and nominal GDP have been useful for forecasting each other, while business credit (a sub-component of total credit) has been a strong leading indicator for nominal investment. A difficulty in assessing the results of Blundell-Wignall and Gizycki relative to the ones cited above is that they fail to disentangle real output from the price level, so that we cannot infer whether the credit measure can predict real output and inflation separately, as policymakers would like.

Rather than examining the relationship between financial aggregates, real output and the price level in an unrestricted reduced form, recent research by de Brouwer, Ng and Subbaraman (1993) focused on a more narrow question, namely whether the standard money demand specification is stable. A stable money demand would imply cointegration among the variables, provided the data series are integrated of the same order. Thus, the researchers test for cointegration among candidate measures of prices, output, financial aggregates and interest rates. The study examines a wide assortment of alternative data measures to investigate the sensitivity of the inferences to modest alterations in the specification. The results suggest that the empirical estimates of the function are not in general cointegrated over the sample, suggesting that money demand was unstable over the period. De Brouwer et al. argue that this finding supports the view that monetary aggregates may have limited indicator properties in the long run.

In addition, Fahrer and Myatt (1991), Coelli and Fahrer (1992) and de Brouwer and Ericsson (1995) investigate models to forecast inflation, and find weak to nonexistent support for financial aggregate measures as predictors of inflation. The results of the empirical studies on the financial aggregates in Australia suggest that evidence in support of their usefulness for predicting (as well as inferring monetary policy effects on) real output growth and inflation is weak, and has weakened as the data sample has grown.

In other related literature, several studies employ US data and the VAR methodology as the main method of inquiry, highlighting the diversity of results obtained using the general VAR techniques, and noting the sensitivity of the results to changes in the chosen variables, sample period, and identification method.

Friedman (1996), in one of the most recent examples of the US literature, notes that regardless of whether money growth acts as an intermediate target or simply as an information variable, it needs to anticipate movements in prices and/or output to fulfil either of these roles. Friedman uses US data on the log-level of output, the price level, and a monetary aggregate in a three-variable VAR as well as a four-variable VAR that includes the interest rate. He imposes a recursive causal ordering that places money last in order to generate variance decompositions to investigate money growth's contribution in explaining subsequent output and price fluctuations. The results indicate that the predictive role of US monetary aggregates (M1 and M2) declined in the 1990s to the point where it is virtually nonexistent.^[6]

There have also been several studies in the US literature that focus entirely on the predictive power of monetary aggregates for real output, searching for a non-neutrality of money. Stock and Watson (1989) provided evidence from three and four-variable VARs, in differences as well as in levels, that a narrow monetary aggregate (M1) was a statistically significant predictor of real output (as proxied by industrial production). Friedman and Kuttner (1993) examine the robustness of this finding by extending Stock and Watson's sample period and using a different interest rate measure. In-sample causality tests show that the Stock and Watson results are not robust to these changes. In addition, Friedman and Kuttner show that in the United States the spread between commercial paper interest rates and the Treasury bill rate was superior to monetary aggregates at forecasting real activity in a VAR.

Thoma and Gray (1994) point out that Friedman and Kuttner fail to confirm the forecasting power of the spread variable by performing out-of-sample forecasting tests. Thoma and Gray find that there is little difference in the forecasting power of the paper-bill spread and M2 in an out-of-sample setting. This argument is relevant because real-time forecasting is an essential element to policymaking. For Australian data, Trevor and Thorp (1988) investigate out-of-sample properties of simple VAR models for forecasting the Australian economy. Their concern is to emphasise the difficulty of the real-time forecasting problem for policymakers, an issue dealt with more extensively below.

This paper extends the literature by presenting a comprehensive analysis of the information value of financial aggregates by examining both in-sample and out-of-sample tests of a set of financial aggregates for predicting prices and output. The data and methodology adopted are discussed below.

Footnotes

The vector error correction (VECM) framework differs from the VAR in that the VECM implies cointegration of the data series. [3]

The sample period in the study ends in 1987 and the sample therefore provides only a partial reflection of the major changes that took place following deregulation of the financial system. [4]

Weber (1994) finds evidence that innovations associated with M1 had a significant impact on real output in an historical decomposition of the 1990–1992 recession in Australia. One criticism of the paper is that the VAR does not include the cash rate, which is the operational instrument of monetary policy. The innovations in M1 may only be proxying for interest rate innovations. [5]

We note, however, that these results may be sensitive to the choice of causal ordering. [6]