RDP 2003-10: Productivity and Inflation 5. Results

Table 4 presents our core results for the effect of growth in the IPDs on productivity growth. The results in the opposite direction are not the focus of this paper but are reported and discussed in Appendix E. Here we have suppressed the results for the output gap – the coefficients are uniformly significant and of the expected (positive) sign. While our interest lies in the sign and significance of any relationship, not the numerical value of the actual coefficients, an example of the output from a complete system of equations is reproduced in Appendix C.[21] Table 4 also reports the R2 for each equation within the SURs. A plot of residuals from the IPDs to labour productivity regressions are presented in Figure D1, the residuals from the MFP regressions were not appreciably different.

Table 4: Productivity Growth and Inflation Model
IPDs causal effect on industry productivity growth, with two lags on price and productivity variables: 1967–2002
  From IPDs to labour productivity   From IPDs to multifactor productivity
Coefficient R2 Coefficient R2
Mining ** 0.45   ** 0.41
Manufacturing 0 0.08   0 0.25
Utilities ** 0.20   0 0.39
Construction +** 0.26   +* 0.20
Wholesale & retail trade ** 0.46   ** 0.47
Transport, storage & communications 0 0.21   0 0.30
Average relationships(a) ** 0.88   * 0.95
Aggregate relationship(b) 0 0.42   0 0.37

Notes: Number of observations = 34; number of parameters = 5.
(a) We calculate this ‘average’ result for the industry-level productivity growth and prices models using a cross-sectional estimate, described in Pesaran and Smith (1995).
(b) The ‘aggregate’ result is a straightforward Ordinary Least Squares estimate of the model with the changes of the GDP deflator and aggregate labour and multifactor productivity substituted for the industry price and productivity measures.

Table 4 presents the results of two tests, as do all the following tables reporting results from the various models. The first test asks whether the lagged independent coefficients sum to a sign significant at the 10 per cent confidence level. The signs report these results, with a zero indicating the coefficients have no significant sign.[22] The second test is the traditional Granger causality test of the joint significance of the coefficients on the lags of the explanatory variable, with * and ** indicating significance at the 10 per cent and 5 per cent levels, respectively.[23]

The first group of results in Table 4 describes the SUR estimates of the industry-level relationship. These results are followed by the cross-sectional estimate, which is an unbiased means of observing the ‘average’ relationship across our six industries. The aggregate (i.e., whole economy) relationship is the final reported result. We discuss these aggregate models and their results further in Section 7.3.

It is clear from these results that the aggregate pattern hides some divergent industry level results. We find many significant results with the majority showing a negative relationship, although the construction industry is an exception. The results for MFP confirm this pattern in labour productivity with some marginal differences.

Results for causal flow in the ‘reverse’ causal direction differ between labour productivity and MFP. With labour productivity there is little evidence of any significant ‘causation’ from productivity to inflation and, thus, we can be confident of the results. For MFP there is a more perplexing relationship – faster MFP growth now appears to cause higher prices in the next two years. This is not the expected relationship and certainly doesn't square with the predominantly negative relationship seen in Table 4 or the theoretical priors. We examine this in more detail in Appendix E but, for now, leave its interpretation open. Given this, one may want to treat the MFP results with more caution.[24]

The results for the industry group of transport, storage & communications are insignificant for the causal flow from prices to productivity growth. The parameter estimates for this case are consistently insignificant in the various regressions we run. While we do not fully understand why this is the case, the explanation may partly lie in this industry group being composed of two very different sectors – communications (controlled by a government-owned monopoly for most of our sample) and transport & storage. Table 6, which summarises the industries' structures, gives some indication of the extent of these differences. The productivity-inflation relationship may behave differently in these sectors, leading us to find no significant relationship across the industry group as a whole. Disaggregated series for these sectors are only available from 1981, meaning there is little scope to get statistically meaningful results from the disaggregated series.

5.1 Considering the Results in More Depth

While we have used industry inflation in our regressions it may be that each industry is merely responding to aggregate inflation as proxied by the industry inflation series. Our industry-level price measures must all contain some element of aggregate inflation. Table 5 reports the IPDs' contemporaneous correlation with the aggregate price measures of the GDP deflator and CPI inflation through the full sample.

Table 5: Correlation of the IPDs and Aggregate Inflation Measures
IPD GDP deflator CPI inflation
Mining 0.55 0.49
Manufacturing 0.83 0.77
Utilities 0.54 0.57
Construction 0.74 0.76
Wholesale & retail trade 0.63 0.68
Transport, storage & communications 0.76 0.65
Note: ‘GDP deflator’ is annual percentage change in the non-farm GDP deflator.

As expected, most of the IPDs' movements are common across the economy. Further testing, however, indicates that industry deflators are to be preferred to the aggregate deflator. Aggregate inflation (measured by the GDP deflator) is uniformly insignificant when added to our equations. When we substitute CPI inflation or changes in the GDP deflator for the IPDs in our model, neither performs at all well at predicting productivity growth.[25] This is useful information. These results indicate that the industry-specific component of the IPDs seems to matter more for an industry's productivity growth than aggregate inflation.

Returning to the issue of why the sign of causation from inflation to productivity growth diverges between industries, we look more closely at how the industries differ. One of the simplest means of understanding how industries differ is asking how production in that industry is organised – in particular, whether it is dominated by small or large firms. Of course there are other ways in which the industries differ, but given limited data this seems to best help us understand our results.

Table 6 summarises the most relevant measures. We present the N-firm concentration ratio for an industry – the proportion of gross-value-added produced by the largest N firms – and information on the 1st and 9th deciles for firm income and sales. Note, these measures do not impose any priors about how these firms competitively interact in the market; rather they all observe the industry structure at the level of the firm qua autonomous economic agent.

Table 6: Industry Structure
Firm size characteristics by industry
  4-firm GVA concentration ratio(a) 20-firm GVA concentration ratio(a) Firm income ($'000)   Firm sales ($'000)
  1st decile firm 9th decile firm 1st decile firm 9th decile firm
Communications 0.95 0.96 na na   na na
Mining 0.32 0.56 150 25,126   130 25,753
Utilities 0.18 0.56 na na   na na
Transport & storage 0.28 0.46 47 1,422   48 1,503
Wholesale & retail trade 0.14 0.21 100 4,484   89 4,194
Manufacturing 0.07 0.19 66 4,064   60 3,540
Construction 0.04 0.17 37 1,222   43 1,282

Notes: (a) A similar measure of concentration for the largest 12 and 25 firms in an industry is available for the start of our sample. It suggests that the rankings in 1969 were not too different from those reported here.

Sources: ABS; Industry Commission and Department of Industry, Science and Tourism (1997). Concentration ratios are an average of the ratios for 1998/99 and 1999/00 published by the ABS. The income and sales distribution data was collected in the 1995 wave of the Business Longitudinal Survey.

We see a consistent picture here. Large firms appear to be more important in mining, utilities and trade than in manufacturing and construction. Both the smallest and the largest firms are smaller in manufacturing and construction than in wholesale and retail trade, utilities and mining. Likewise, the largest 4, or 20, firms are responsible for a smaller proportion of an industry's output in these least concentrated industries. This pattern of industrial structure correlates with the signs in our results for the inflation-productivity growth relationship.[26]

Our results show a break in the sign of the inflation-productivity growth relationship, from the negative relationship observed in the mining, utilities and wholesale & retail trade industries, to the insignificant relationship observed in the manufacturing industry, to the positive relationship observed in construction. At the same point, there is an observable shift in the industry structure, albeit not a clear break. Those industries where large firms play a more important role reported a negative relationship. Contrast construction where the relationship was positive: small firms appear to be relatively more important in the construction industry. In between lies manufacturing, where no significant relationship was observed, and where small firms play an intermediate role in producing the industry's output.

We hypothesise that observed sign differences may reflect compositional effects within the industries. Specifically, in the concentrated industries we would not expect much change amongst the firms in business. Thus, inflation's observed effect on productivity growth probably reflects within-firm effects. On the other hand, we would expect much higher rates of firm bankruptcy and formation in industries where small firms play a larger role. So the differences in results may reflect compositional differences between the industries. We explore this idea further in Section 7.2.

This section presented our core results. We analyse them in Section 7 after considering their robustness in the next section; the reader less concerned with iterations of the results may proceed directly to Section 7.

Footnotes

Results from the complete estimates of the other equations are available upon request from the authors. [21]

The simple t-test is H0: β1 + β2 = 0, where β1 and β2 are the coefficients on the lags of the explanatory variable (e.g., in the equations using the IPDs to explain productivity growth, the first and second lags of the IPDs). [22]

This F-test's nul hypothesis is: H0: β1 = β2 = 0 . [23]

However, if anything, this confirms the finding that lower inflation leads to higher productivity – reverse causality is not the reason for our inflation to productivity story because it is of the wrong sign. [24]

The GDP deflator performs less poorly, as is expected given its slightly higher correlation with the IPDs. Reassuringly, both are good predictors of the IPDs. [25]

Note that transport & storage is disaggregated from communications in Table 6. This is intended to illustrate the differences in these industries, and particularly to highlight that while communications is dominated by one firm, transport & storage is one of the less concentrated industries. Its firms are similar in size (as measured by the 1st and 9th deciles) to those in construction, another low-concentration industry. This is likely to indicate why we do not observe a strongly significant inflation-productivity growth relationship when we aggregate these industries together. [26]