4. Analysis
We begin our analysis by looking at the correlations across performance metrics. Table 7 shows (Pearson & Spearman rank) correlations across our three core levels measures of profitability, labour productivity & MFP. As we would expect, the rank correlations are positive & significantly different from zero.29 Presumably the high correlation between labour productivity & MFP is partly a reflection of any inadequacy in our Cobb-Douglas model and/or the high degree of correlation between capital and labour. Table 8 shows correlations of annual growth rates, as opposed to levels and expands the performance measures to include sales & employment.30 Again, most measures are positively correlated, with the main exception being the relationship between employment growth and productivity growth. This negative correlation is perhaps to be expected over the short-run, consistent with an economic model with adjustment costs (ie, as firms scale up this requires changes in structure, learning, etc, which impose costs). Such a model is also consistent with our finding that sales and employment growth are positively correlated (ie, a scale effect).
Figures 1, 2 & 3 present industry-level distributions of labour productivity, MFP and profitability (respectively) in 2005.31 Looking at the productivity distributions first, it appears that some industry differences are partially explained by differing average capital intensities. For example, the mining industry (ANZSIC B) productivity distribution sits to the right in Figure 1. Conversely the accommodation, cafés & restaurants industry (ANZSIC H) sits to the left of the labour productivity distribution. After controlling for capital intensity the apparent productivity differences between these two industries is diminished (Figure 2). Most industry profitability distributions are centred close to zero (Figure 3), with communication services and health and community services being the most noticeably right-skewed.32 Conversely, decent returns in 2005 seemed hardest to come by in the accommodation, cafés & restaurants industry.33
Despite the high degree of churn in the underlying population of firms suggested by Table 1, distributions of performance levels and growth rates are remarkably stable over time. For example, Figure 4 shows the distribution of profitability for three years (2001, 2003 & 2005). The distribution across years is very similar for negative profits, with moderate year-on-year variations in the proportions of firms that record positive profitability. Similarly Figure 5 shows that annual growth rate distributions are also quite stable across years. Having said that, the range of annual growth outcomes is broad with log differences in labour productivity across years often exceeding one.
Attempting to reconcile this vibrant firm-level dynamism with the seemingly structural stability in aggregate distributions has a long history in the economics literature (see Sutton 1997 for an excellent review). Part of the reconciliation has to do with the persistence of performance in incumbent firms, remembering that these firms account for almost two fifths of observations. Table 9 summarises the transitions of firms between 2000 and 2005 labour productivity deciles.34 The main point to note is that a large proportion of surviving firms maintain their relative productivity level. Overall, 23% stay in same decile while a further 30% move only one decile up or down the distribution. Another way to think about this persistence of performance is to look at the autocorrelation of various performance metrics across various lag lengths. Table 10 demonstrates the strength of the temporal relationship between the levels of our three key performance metrics, labour productivity, MFP & profitability. Looking at the autocorrelations in annual growth rates it appears that there is some short-term reversion in performance (ie, "good" years are followed immediately by "bad" years). Over the longer term, annual growth rates do not appear to be correlated at all (except in the case of employment growth, where a weak negative correlation persists). These results concur with work done previously (see, eg, Law & McLellan 2005 & Law et al. 2006).
This is not to undermine the important contribution to the economy that comes from entering and exiting firms. As Table 11 and previous analyses (eg, Carroll et al. 2002) have demonstrated, entering and exiting firms account for a large proportion of net job creation. Using the LBD we can describe the productivity impact of firm turnover. Standard methods of decomposing the contribution of firm turnover to productivity usually consider impacts over a five year period. Since this would leave us with a single observation, we instead, in Figures 6 & 7, present cohort analyses for entering and exiting firms from which it may be inferred that entry and exit probably make the expected long-term positive contribution to aggregate productivity growth. Entrants have lower productivity initially and then tend to move to or past the average productivity level of full-period incumbents by their second or third year of operation (conditional on survival).35 Conversely, exiting firms have below average productivity throughout the last few years of their existence.
Two further groups of firms have a particular interest for innovation, trade & competition policy: exporters & foreign-owned firms. We turn now to a brief discussion of each of these groups. Very little is known about the microdynamics of New Zealand's export sector. Because our measure of exporting comes purely from Customs merchandise trade data, this section of the paper focuses exclusively on manufacturing firms. We do this to reduce the potential of misestimating the correct denominator in our calculation of the proportion of firms that export (an issue that is sometimes overlooked when very low rates of exporting are reported in New Zealand), and to control for broad industry in our discussion of relative productivity performance.
Previous research suggests that exporting is concentrated in a small number of firms, and that an even smaller number of firms generate a large proportion of sales from exports. For example, Simmons (2002) reports the proportion of all firms that export between 4-5% (a figure susceptible to the "appropriate denominator" criticism). In our data we find that 11.4% of manufacturers exported goods in 2005. Figure 8 summarises export intensity deciles in 2005 after the 88.6% of firms that have zero exports have been removed.36 The data strongly supports the idea that exporting is a sideline activity for most firms, with the mean (median) firm exports constituting 17.9% (5.9%) of total sales.
Figure 9 demonstrates that, while relatively scarce, manufacturing exporters punch above their weight displaying higher average labour productivity. Simple tests of differences in means (1% significance level) suggest that both incumbent & new manufacturing exporters have higher labour productivity levels than non-exporters (and incumbent exporters are significantly higher than entrants also), and that entering exporters have higher annual employment growth than non-exporters (starting from a higher average total employment). The theoretical (and some empirical) literature suggests that exporters may experience faster productivity growth through, for example, learning effects. These effects might be expected to be more prominent in market entrants, rather than incumbent exporters. To investigate this possibility, Figure 10 breaks the labour productivity growth distributions into entering exporters, incumbent exporters, and non-exporters. Labour productivity growth in exporters (neither incumbent nor entering) is no higher than in non-exporters.37 Potentially, we should be looking for longer-run effects from foreign market participation. However, these preliminary findings are consistent with the international literature, for example, Bernard & Jensen (1999) who find that employment growth is higher in US exporters, but not labour productivity growth.
Figures 11 & 12 compare the labour productivity distributions of foreign-owned and domestic firms using the IR4 foreign-ownership indicator as the basis of splitting the sample. This analysis is restricted to limited liability companies, since only these firms file IR4s. Focussing first on Figure 11, it is apparent from the data that foreign-owned companies are more productive than domestically-owned equivalents, with the difference in productivity levels quite startling. Perhaps this is a consequence of the simple univariate breakdown in the data? In particular, the foreign-owned firms are concentrated mainly in five industries that also tend to have higher mean labour productivity levels: mining; construction; wholesale trade; communication services; and finance & insurance. Another possibility is that foreign-owned firms are also exporters. Foreign-owned manufacturers are roughly four times more likely to be goods exporters than domestically-owned manufacturers.
Figure 12 breaks the labour productivity distribution of manufacturing companies down by both ownership and export status. Apparently, exporting behaviour cannot explain the size of the productivity gap between foreign-owned & domestic-owned manufacturers. All this suggests that careful econometric analysis is required to disentangle the underlying causes of higher firm productivity. For example Figure 12 is consistent with a model where FDI has no impact on productivity, but rather that foreign firms have better access to capital and can buy out more productive NZ firms; or, alternatively, that the effect of FDI is on organisational behaviour rather than access to foreign markets. Before turning to a discussion of future work (in our concluding section), we quickly outline data issues that also need to be factored into our interpretation of these results.
Back to Top