RDP 2005-06: Credit and Monetary Policy: An Australian SVAR 2. The Set-up of the SVAR

2.1 Variables Included in the SVAR

The form of the structural vector autoregression (SVAR) used in this paper reflects the fact that Australia is a small, relatively open, economy for which external shocks can be an important driver. A key decision is how many variables to include in the model. This paper follows Brischetto and Voss (1999) in using a small-scale model, including two variables for the external sector, and five for the domestic sector. While a larger SVAR, such as the 11-variable model in Dungey and Pagan (2000), would allow for richer interactions, a more parsimonious model with more degrees of freedom is likely to be easier to estimate and more stable. The 7-variable SVAR used here seems to provide a good compromise between these trade-offs and is still capable of capturing the key macroeconomic interactions. Dungey and Pagan (2000) provide a survey of other VAR studies of the Australian economy. A brief description of the SVAR methodology, which may be of particular interest to readers not familiar with this technique, is contained in Appendix A.

The role of the external sector is captured by real commodity prices (commodity prices in US dollars deflated by the US CPI, comm) and real US GDP (usgdp). The domestic sector is captured by: real Australian GDP (gdp), quarterly inflation (π), real credit (nominal credit deflated by the CPI, cred), the cash rate (i), and the real trade-weighted exchange rate index (twi).^[2]

Commodity prices are included because they contain information about the world business cycle and are likely to be particularly relevant to a commodity-exporting country such as Australia. The inclusion of commodity prices has also been found to help resolve the ‘price puzzle’ in VARs, that is, the finding that the price level tends to increase in response to a contractionary monetary policy shock. Commodity prices are thought to control for policy-makers' expectations of future inflation, seemingly the missing factor responsible for the ‘price puzzle’ (for a survey, see Christiano, Eichenbaum and Evans 1998). Australian VAR studies capture commodity prices in a number of different ways. Suzuki (2004) includes commodity prices, while in Dungey and Pagan (2000) the terms of trade play a similar role. Brischetto and Voss (1999) include world oil prices.

Many papers have found that the global business cycle is an important driver of domestic activity. US GDP is included here as it has been shown to have a particularly strong relationship with Australian activity (see, for example, Gruen and Shuetrim 1994; de Roos and Russell 1996; Beechey et al 2000). While US GDP has been used by other recent Australian VAR studies to represent world economic activity, such as Dungey and Pagan (2000) and Suzuki (2004), the US federal funds interest rate has also been used (Brischetto and Voss 1999).

The inclusion of GDP to represent domestic activity is standard. Inflation is included, following Dungey and Pagan (2000), rather than the price level as in Brischetto and Voss (1999). There are no nominal level variables in the model and so the rate of change of prices seems to be a more logical variable to interact with real variables and a nominal interest rate. In particular, for over half of the sample the objective of monetary policy has been an inflation target.

No Australian VAR studies have included total credit. Suzuki (2004) is closest with the inclusion of bank lending. Brischetto and Voss (1999) include a monetary aggregate, while Dungey and Pagan (2000) include neither monetary nor credit aggregates.

The overnight cash rate is included as this has been the chief instrument of monetary policy since the float of the dollar in December 1983 (Grenville 1997 and Dotsey 1987). Brischetto and Voss (1999), Dungey and Pagan (2000) and Suzuki (2004) also include the cash rate, but Haug, Karagedikli and Ranchhod (2003) use a 90-day interest rate to be consistent with New Zealand, the other country in their study.

The real exchange rate is also viewed as an important measure of Australian economic conditions. Sims (1992) suggested that the inclusion of the exchange rate can also help to resolve the price puzzle. Dungey and Pagan (2000) also use this variable, while Brischetto and Voss (1999) and Suzuki (2004) both use the US dollar bilateral exchange rate. The real trade-weighted exchange rate used in this paper captures a broader relationship and is more appropriate to interact with other real variables. Throughout the paper, the ‘exchange rate’ and ‘trade-weighted index’ will be taken to mean the real trade-weighted exchange rate index.

All data are quarterly; the sources and further details are given in Appendix B.

2.2 Identification

Structural shocks in a SVAR can be identified by placing some restrictions on contemporaneous relationships. There are few simple theoretical macroeconomic models that explicitly include credit, and seemingly none that determine the timing of effects needed for identification in a SVAR. Therefore previous studies and stylised facts are used to determine the identification restrictions outlined in this section.

The restrictions placed on the contemporaneous relationships among the variables are characterised by Equation (1), which is the left-hand side of the standard SVAR representation (Equation (A2) in Appendix A).

The (non-zero) coefficients b_ij in Equation (1) indicate that variable j affects variable i instantaneously (for example, b₂₁ is the instantaneous impact of commodity prices on US GDP). The coefficients on the diagonal are normalised to one, while the blank entries indicate that those entries in the matrix are constrained to be zero. The assumptions embodied in Equation (1) exactly identify the system.

The transmission of international shocks to the domestic economy can be very rapid. For example, an increase in commodity prices results in an immediate increase in the value of Australian exports, and hence domestic income. Therefore, apart from two exceptions, it is assumed that all foreign variables affect all domestic variables contemporaneously. The first exception prevents an immediate effect of US GDP on monetary policy (that is, the cash rate). This assumption reflects the informational lags faced by policy-makers and is also employed in an open-economy SVAR by Kim and Roubini (2000). The second exception prevents an immediate effect of US GDP on inflation, since the domestic inflationary consequences of world economic activity would normally be thought to be transmitted indirectly through domestic activity.

The domestic variables are deemed not to affect the international variables, reflecting the relatively small size of Australia's economy.

Australian real GDP is assumed to be affected contemporaneously by inflation and credit. Output might respond contemporaneously to inflation because nominal incomes, and so spending, may be fixed in the short term. Alternatively, this assumption can be motivated by the Lucas-Phelps imperfect information model, in which producers face a signal extraction problem. Contemporaneously, producers only observe their own price, and so are unsure whether an increase in their price reflects inflationary pressures or an increase in demand. As a result, they increase production, even if the price increase is purely inflationary. This increase in production could occur quite quickly.^[3] The contemporaneous response of output to credit follows Safaei and Cameron (2003) and reflects a quick pass-through of credit to aggregate demand. Given the cost of borrowing, credit will typically be spent as soon as the funds are obtained, immediately adding to aggregate demand.

Equation (1) allows for the possibility of a contemporaneous response of inflation to output. This assumption is common in domestic (Brischetto and Voss 1999; Dungey and Pagan 2000) and international (Bernanke and Blinder 1992) studies.^[4] Other domestic variables are assumed to affect inflation only with a lag.

Credit is assumed to respond to output, inflation and the overnight cash rate, contemporaneously. The expectation of future activity is an important determinant of credit demand, as noted by Blundell-Wignall and Gizycki (1992). Current activity, as observed by individual agents, and interest rates, should give some indication of what future conditions hold. The contemporaneous interaction of credit with the interest rate and inflation is justified by the perception that borrowers and potential borrowers will respond quickly to the real cost of credit (the difference between the interest rate and the inflation rate). Note that these assumptions are in contrast to Safaei and Cameron (2003).

The overnight cash rate is assumed to respond contemporaneously only to commodity prices, credit, and the exchange rate. This is justified by information lags. Direct information on these variables is available within the quarter, unlike the other domestic variables. The exchange rate is assumed to respond contemporaneously to all variables, as is common in SVAR studies.

While this study uses the same number of variables as Brischetto and Voss (1999), and attempts to capture the same key macroeconomic interactions, as noted above, it uses a different set of variables to do so. Of particular note is that the SVAR in this paper includes credit, rather than a monetary aggregate as in Brischetto and Voss, in order to specifically understand the interaction of credit with other key macroeconomic variables.

This paper also imposes an important restriction on the lagged structure of the model. Given that the Australian economy is small relative to the global economy, it is assumed that lags of the Australian variables have no effect on the international variables. This restriction was not imposed by Brischetto and Voss (1999) but has been used in other studies (for example, Dungey and Pagan 2000). Lags of all variables are included in the equations for the five domestic variables.

2.3 Non-stationarity

Unit root tests suggest that most, if not all, of the variables included in the model are non-stationary, I(1), processes. These tests are available as an unpublished appendix upon request. This raises the issue of the appropriate estimation methodology. This paper follows the existing literature which typically estimates VARs in levels even when the variables are I(1). Indeed, of the VAR studies referenced in this paper, only Haug et al (2003) use a vector error-correction model. The preference for VARs in levels can be explained, at least in part, by a reluctance to impose possibly incorrect restrictions on the model.^[5] Even with I(1) variables, the residuals will be stationary because of the inclusion of lagged levels of the variables in the VAR. Nevertheless, the possibility of spurious relationships between the I(1) variables remains. Ensuring this is not the case is perhaps best achieved by confirming that the relationships summarised by the SVAR are plausible on economic grounds.

Footnotes

This study uses the credit series revised by the Reserve Bank in 2004 to ensure better coverage of the securitisation of loans, which has been a major innovation in the credit market over the past few years (RBA 2004). [2]

See Romer (2001) for a discussion of the Lucas-Phelps model. [3]

Brischetto and Voss and Bernanke and Blinder use the price level rather than inflation. [4]

See the discussion in Hamilton (1994), p 652. [5]