RDP 2012-02: The Role of Credit Supply in the Australian Economy 6. Robustness

The results from structural VAR estimation can be sensitive to the model's specification. Our main finding – that a shock to the difficulty of obtaining finance has a significant effect on Australian GDP – changes very little in response to six different types of alternative specifications described below (Figure 13):

1. Number of lags. The baseline specification is estimated using two lags. We also estimate the baseline specification using one and three lags (Figure 13, panel i).
2. Identification. A number of alternative identification schemes for contemporaneous relationships are considered (Figure 13, panel ii):
   • The credit spread responds to difficulty obtaining finance only with a lag, while the exchange rate is assumed to have a contemporaneous effect on inflation (‘Ident 1’).
   • GNE is assumed to respond contemporaneously to credit, reflecting a quick pass-through of credit to aggregate demand. GDP is assumed to respond to the exchange rate only with a lag (‘Ident 2’).
   • Monetary policy can respond to GNE and GDP contemporaneously reflecting the partial information available to policymakers through the quarter. Inflation doesn't respond to GDP and GNE contemporaneously, only with a lag (‘Ident 3’).
3. Omitted variables:
   • It is likely that the confidence of lenders is closely related to the confidence of businesses. To cover the possibility that a ‘credit shock’ might be capturing a ‘confidence shock’, we include a measure of business confidence from the ACCI-Westpac survey in the model (Figure 13, panel iii).
   • To control for the foreign component of credit shocks originating from global capital markets, we include an international measure of financing conditions – the Senior Loan Officer Opinion Survey (SLOOS) of loan officers' willingness to make commercial and industrial loans (Figure 13, panel iii). However, this series only starts in 1990, so prior to this we splice on the SLOOS series for consumer instalment loans (not secured by real estate, e.g. auto and personal loans). This survey question has a longer history and generally moves in line with the SLOOS series for business lending conditions.
4. Sample period. Our baseline model is estimated from December 1983 until December 2011. Given the large movements in both financial and real variables during the financial crisis, we also estimate the baseline model over a shorter sample ending before the crisis, in December 2006 (Figure 13, panel iv).
5. Treatment of structural break in survey variable. Instead of using a Hodrick-Prescott filter to remove the trend from the difficulty obtaining finance, we also try demeaning and detrending the data from September 2002 onwards (Figure 13, panel iv).
6. Definition of variables. Various alternative variables were substituted into the model, one at a time, as detailed in Table 2 (Figure 13, panels v and vi).

The historical decomposition of credit supply shocks on GDP is also robust to alternative models that include net worth and business confidence (Figure 14).

As an aside, we also estimate the model using household credit data as it is possible that the ACCI-Westpac series on the difficulty of obtaining finance captures an economy-wide ‘credit shock’. In particular, we use household credit and the spread between banks' variable mortgage indicator rate and the cash rate. In this model, a credit supply shock has a statistically significant effect on household credit and GDP, but the effect is smaller than in the baseline model. This might suggest that the impact of credit shocks is stronger through the business sector than the household sector, or that credit cycles differ somewhat between these sectors.