RDP 9208: Credit Supply and Demand and the Australian Economy 3. Total AFI Credit and GDP
July 1992
- Download the Paper 638KB
The study by Bullock, Morris and Stevens (1989) found that all financial intermediaries (AFI) credit unambiguously lagged GDP growth. This finding was based on data to the end of 1987, and was therefore dominated by the regulated period. Total AFI credit consists of the business loans analysed in Section 2, housing loans and personal loans. However, the bulk of AFI credit is made up of lending to the business sector. The forward-looking variables and intertemporal substitution mechanisms influencing business credit are also relevant for forecasting the level of economic activity itself, and not just investment. Expectations about future activity are, after all, a key influence on investment decisions. In any case, turning points in business investment and GDP are often in line with each other (see Chart 7), and business credit was shown to lead business investment. For these reasons it might be useful to see whether more recent data has improved the indicator value of AFI credit with respect to GDP.
Chart 6 shows 12-month-ended percentage changes in AFI credit compared to nominal GDP in the top panel, and 3-month-ended percentage changes in credit in the bottom panel. Credit appears to have lagged GDP for all of the period prior to 1984Q1. The subsequent sustained strength of credit growth from 1984 to 1986 was driven by business credit. This grew strongly with investment in 1984 and 1985, and did not follow the downturn in GDP in the latter year. Business credit also grew strongly in 1986, even though GDP and investment growth declined. There was, then, some decoupling of AFI credit from the GDP cycle in these years, largely because of the behaviour of the business sector in the newly-deregulated financial system.
However, from late 1988, following a period of very rapid growth, credit began to slow continually until the end of 1991. The downturn in credit led that of GDP by about one year. The timing of the turnaround in quarterly credit growth was even earlier than this, being superficially related to the spike in borrowing immediately after the stockmarket break. But even if monthly growth rates between December 1987 and April 1988 are assumed to be the same as the average monthly growth rates for the twelve months ending in November 1987, the timing of the turnaround in 12-month-ended credit growth is not significantly affected (shaded lines in the two panels of Chart 6). Whichever way it is measured, the timing of the downturn in annual credit growth in the late 1980s significantly precedes that for GDP.
To examine these relationships more formally, total credit, GDP and the loan rate are first tested for cointegration. The left panel of Table 3 shows results for the full sample period 1976Q1 to 1991Q4. The right panel shows results for the same shorter sample period used for business credit and investment. Each of the variables individually was found to be non-stationary (integrated of order one or two), a pre-condition for testing for cointegration.[11] Details of how the tests were conducted are set out in the note to Table 3. The left hand column of each panel shows the cointegrating regression, Augmented Dickey Fuller (ADF) and Phillips-Perron (Z) statistics.
| Dependent Variables (1976Q1–1991Q4) |
Dependent Variables (1982Q2–1991Q4) |
|||||||
|---|---|---|---|---|---|---|---|---|
| lnC | ΔlnC | ΔlnY | Δi | lnC | ΔlnC | ΔlnY | Δi | |
| Const. | −4.600 |
0.000 (0.2) | 0.016 (3.8) | −1.110 (2.2) | −6.170 |
0.009 (2.8) | 0.006 (0.8) | −1.250 (3.2) |
| lnY i | 1.521 −0.009 | – – |
– – |
– – |
1.650 −0.000 | – – |
– – |
– – |
| Residual | – |
−0.066 (3.7) | −0.020 (0.6) | −0.572 (6.8) | – |
−0.292 (12.4) | −0.296 (4.0) | −2.680 (−0.4) |
| ΔlnC−1 | – |
0.847 (7.4) | – |
– |
– |
0.499 (7.4) | – |
19.970 (2.7) |
| ΔlnC−2 | – |
– |
– |
– |
– |
0.443 (5.0) | – |
– |
| ΔlnC−3 | – |
0.216 (2.0) | – |
– |
– |
0.400 (6.2) | – |
– |
| ΔlnC−4 | – |
– |
– |
40.939 (4.0) | – |
– |
– |
– |
| ΔlnC−5 | – |
−0.068 (2.4) | – |
– |
– |
– |
– |
– |
| ΔlnC−6 | – |
– |
– |
– |
– |
– |
−0.478 (3.3) | – |
| ΔlnY−1 | – |
0.127 (2.1) | – |
−11.990 (2.1) | – |
– |
0.677 (4.4) | – |
| ΔlnY−2 | – |
– |
0.369 (3.0) | – |
– |
– |
0.449 (4.0) | – |
| ΔlnY−3 | – | – | – | – | – | – | – | – |
| ΔlnY−4 | – |
−0.168 (2.4) | – |
– |
– |
−0.348 (6.5) | – |
– |
| ΔlnY−5 | – |
– |
– |
– |
– |
– |
0.459 (4.4) | – |
| ΔlnY−6 | – |
– |
– |
– |
– |
– |
– |
18.420 (1.7) |
| Δi−1 | – |
−0.005 (5.8) | – |
0.704 (5.4) | – |
0.008 (8.0) | – |
0.270 (2.0) |
| Δi−2 | – |
0.004 (3.2) | – |
– |
– |
– |
– |
– |
| Δi−3 | – |
– |
– |
0.259 (2.6) | – |
– |
– |
– |
| Δi−4 | – |
– |
– |
0.327 (4.0) | – |
– |
– |
– |
| Δi−5 | – | – | – | – | – | – | – | – |
| Δi−6 | – |
−0.002 (3.8) | – |
0.280 (2.5) | – |
−0.003 (4.5) | – |
– |
| ADF | −1.72* | – | – | – | −2.19** | – | – | – |
| Z | −2.17** | – | – | – | −2.08** | – | – | – |
| R2 | – | 0.88 | 0.13 | 0.62 | – | 0.95 | 0.60 | 0.38 |
| DW | – | 2.4 | 1.6 | 2.1 | – | 2.01 | 2.42 | 1.76 |
Note: The Augmented Dickey Fuller (ADF) tests (with the null of a unit root) are performed on each model, allowing for trend and/or drift terms, if relevant. TWO asterisks denotes that the null of no cointegration can be rejected at the 5 per cent level. One asterisk denotes rejection at the 10 per cent level. The Z test is the test proposed in Phillips (1987). This test involves making non-parametric adjustments to the ADF test. All standard errors in the error correction regressions are White (1980) – corrected for heteroskedasticity. C denotes credit, Y denotes GDP, and i denotes the loan rate. |
||||||||
Results for the longer sample period are focused upon first. These suggest that credit, GDP and the loan rate are cointegrated at the 10 per cent level using the ADF statistic, and at the 5 per cent level using the Z statistic.
The shortness of the sample period is such that the cointegration results are unlikely to be very robust. Nevertheless, the finding of any long-run relationship between these three variables suggests that the temporal ordering (or “causality”) tests should at least be conducted in an error correction (as opposed to VAR) framework. The error correction regressions with GDP, credit and the loan rate, respectively, as the dependent variables, are shown in the second, third and fourth columns of Table 3. The significance of the parameter on the residual from the long-run levels relationship in each of these cases suggest that:
- nominal credit adjusts to previous movements in the levels of nominal GDP and the loan rate;
- the loan rate also adjusts to prior movements of GDP and credit; but
- nominal GDP is not led by prior movements in the levels of nominal credit and the loan rate.
That is, the nominal activity variable appears to be weakly exogenous and leads the credit and loan rate variables, both of which are endogenous. This finding is entirely consistent with that of Bullock, Morris and Stevens.
The results for the shorter sample period shown in the second panel again suggest cointegration, this time at the 5 per cent level for both the ADF and Phillips-Peron statistics. The error correction results suggest a number of important differences compared to the results for the longer sample period. With credit as the dependent variable the parameter on the residual from the levels relationship −0.292, or speed of adjustment, is highly significant and much stronger compared to the long sample period results. It implies adjustment to the long-run equilibrium relationship of 3½ quarters. This compares with a speed of adjustment of −0.066, or 15 quarters, if data from the regulated period is included. Credit still adjusts to prior movements in the loan rate and GDP, but much more rapidly than for the full sample period. With GDP as the dependent variable, the weak exogeneity finding over the full sample period is no longer supported by the data. Nominal GDP also adjusts endogenously to prior movements in the levels of credit and the loan rate over the period 1984Q1 to 1991Q4.
Both credit and GDP, therefore, appear to be useful for forecasting each other, when used in conjunction with the lending rate. Only the loan rate appears to be weakly exogenous over the shorter sample period (the final column of Table 3). The finding of two-way causation between GDP and credit since 1984 contrasts with that of Bullock, Morris and Stevens, who found that credit unambiguously lagged GDP.
Footnote
Most of the variables are integrated of order one. Nominal credit is ambiguous being either I(1) or I(2). Tests of the order of integration for each series are presented in Appendix C. [11]