4.2  Firm dynamics and productivity change

Another line of evidence on the sources of firm productivity improvements, given new technology and skill levels, comes from studies of firm dynamics. These provide a more detailed description of what is going on at the firm level, and allow a clearer separation of the contribution to productivity of worker and firm characteristics. For our purposes, this evidence highlights the considerable heterogeneity in productivity levels that exists across firms in an industry, the effects on aggregate productivity growth of resource reallocation across plants, and the influence of skill levels on firm productivity.

Several authors (Bartlesman and Doms (2000), Haltiwanger, Lane and Spletzer (2000)), summarise the evidence on productivity differentials between firms within an industry:

• The amount of productivity dispersion is very large. Some firms are much more productive than others.
• Highly productive firms today are more than likely to be highly productive tomorrow, although there are some changes in the productivity distribution over time.

Dunne, Foster, Haltiwanger and Troske (2002), again utilising establishment level data, summarise the changes that have been occurring in productivity and wage dispersion:

• The between plant component of wage dispersion is an important part of total wage dispersion, and is occurring within industries.
• The between plant measures of wage and productivity dispersion have increased substantially over the past few decades.
• A significant fraction of rising wage dispersion, and to a lesser extent productivity, is accounted for by changes in the distribution of computer investment across plants.

Foster, Haltiwanger and Krizan (1998) provide a summary of recent (mainly US) research on the resource reallocation occurring amongst manufacturing plants over time:

• There is a large level of reallocation of inputs and outputs across plants, and much of this reallocation occurs within rather than between sectors.
• Resource re-allocation from less to more productive plants plays an important role in aggregate productivity growth, and here firm entry/exit has a dominant effect.
• The pace of reallocation varies secularly, cyclically, and by industry, and affects aggregate rates of growth.
• Plant level heterogeneity can be accounted for by uncertainty, differences in managerial ability, capital vintage, location and disturbances, and the slow diffusion of knowledge between firms.

The OECD (2003b, Chapter 4) has compared the productivity implications of firm dynamics across a number of OECD countries. In relation to higher labour productivity, the evidence suggests that the resource changes within each firm play the dominant role, and that resource reallocation between firms is typically small. The entry and exit of firms had a variable effect, accounting for 20-40% of total labour productivity growth. An exception here is in industries that are experiencing technological change, where firm entry is more important. The within firm component contributed less to MFP growth, and resource reallocation between firms and firm entry and exit contributed more. It found that start-ups in the US were smaller and less productive that for Europe, but those which survived grew faster. The OECD (2003b: Chapter 4) also discusses the effects of product and labour market regulations on this resource re-allocation. This suggests that in countries with more restrictive regulatory regimes, labour productivity increased through capital deepening and improved production processes, while in more open economies faster MFP growth occurred through new firms entering with more appropriate input and technology mixes.

Bartlesman and Doms (2000) and Foster, Haltiwanger and Krizan (1998) also review the factors that may be influencing the patterns of resource re-allocation amongst firms in an industry. They note that differences in managerial and entrepreneurial ability seem to be important, as they make choices about technology and work practices. For instance, changing ownership (eg, through mergers) provides the opportunity to leverage off managerial quality. There is some evidence that the productivity of plants is positively related to the productivity of the firm to which it belongs, and that above-average productivity growth has followed changes of ownership. Productivity dispersion may also be due to slow diffusion of knowledge about new technologies and market opportunities. They note that productivity growth is positively linked to the uptake of technology, although this can occur by existing firms re-tooling as well as by new firms entering.

Mills and Timmins (2004) show that once comparable data is used by excluding zero-employee firms, New Zealand’s size distribution and firm dynamics are not outliers amongst OECD countries. Its distribution of firm size is similar to many other OECD countries. Firms also enter and exit at a similar or larger size compared with other OECD countries. Again, while New Zealand also has a relatively high firm turnover rate (annual entries and exits as % of total firm population), Mills and Timmins find that the survival and growth rates of New Zealand firms are not out of line with other OECD countries. Nevertheless, job turnover rates in entering and exiting firms is at the high end of the OECD range. Carroll et al (2002) used a wider measure of job turnover (job creation plus destruction as % of total workforce). This measure would give a clearer overall picture of labour market patterns.

4.2.1Skill-based effects 

Haltiwanger, Lane and Spletzer (2000) use a matched longitudinal US database to look at the relationship between worker skills, wages and firm productivity, both when firms enter and then as learning/selection processes operate over time. Skills are measured by education level. They find that firms make very different choices about the key inputs – technology, capital, organisational structure and worker skill mix – and that these are quite persistent over time. They also note that new businesses exhibit greater heterogeneity of earnings and productivity than do mature businesses. This reflects the effects of both selection (business failure) and learning (firms adjusting factor mix). They find that differences in productivity, skills and wages are highly correlated – that high productivity workplaces have highly skilled workers and high earnings/worker, and others exhibit low productivity, wages and skills. This suggests that the choice of worker mix is likely to be complementary to the other choices of the firm. They conclude that more mature businesses locate themselves along an upward sloping productivity/skill profile by adjusting their workforce and other input mixes. They also adopt an upward sloping earnings/skills profile. A caveat here is whether sufficient controls have been included to account for unobserved firm heterogeneity. Other factors may be influencing productivity differences.

A firm’s earlier choice of skill levels appears to be important for ongoing technology adoption, with the growing dispersion of wages and productivity being linked to differential rates of technological adoption between plants in the same industry. Models of these effects have been formulated by Kremer and Maskin (1996) and Caselli (1999). Kremer and Maskin’s model allows increases in plant level segregation of workers by skill. It allows the supply of skills to influence the skill-technology matches within firms, which in turn accentuates wage inequality. It proposes that workers of different skill levels are imperfect substitutes, that different tasks within a firm are complementary, but that different tasks within the firm are differentially sensitive to skill. The model implies that there are several competing forces determining the equilibrium patterns of matching skills and tasks in plants. If a wider range of production tasks is required by the firm, this favours less skill segregation within individual firms, while greater complementarity between tasks favours greater skill segregation. If the overall distribution of skills is compressed, high and low skilled workers will be matched in the same plant, but as the distribution of skills widens this will be less so.

Caselli (1999) also models the effect of technical change on the dispersion of firm wages and productivity. However, he focuses more on the costs faced by high and low skilled workers in re-training when adjusting to the introduction of new technologies. New technologies require different types of skills, and so segregation of skills across plants based on technology use is assumed. The model matches machines and worker skill levels to achieve higher productivity. Since retraining for more skilled workers is less costly, when technological change occurs the demand for more skilled workers increases. Correspondingly, the demand for low skilled workers decreases. They will continue to use older machines, with lower levels of productivity and wages. In this sense, technical change would be skill biased. In the technology diffusion process, capital flows towards the new technologies, and pressures exist for widening wage differentials for skill. This, in turn, increases the opportunities for re-training, and slows the rising wage dispersion.

Kremer and Maskin point to US evidence of economic activity shifting from firms in industries that typically use both high an low skills (eg, General Motors), to industries that used either high skills (eg, Microsoft) or low skills (eg, McDonalds). Perhaps more pertinent evidence is the increasing correlation of wages among workers in establishments, and amongst US states. Those with a wider dispersion of educational attainment are also more segregated by education amongst firms. Caselli provides a range of evidence for the past two decades that is consistent with his model. First, the wages and education of workers are higher at plants with high measures of R&D and technology adoption, plants using more advanced technologies are investing more and skill upgrading across plants and industries is positively correlated with capital-output ratios, and (as with Kremer and Maskin) there has been a substantial increase in between-plant wage inequality. He also cites evidence that US industries experiencing higher capital to labour ratios have had relatively large increases in wages and the employment of skilled workers.

Dunne, Foster, Haltiwanger and Troske (2002) also explore the US evidence for these models. Kremer and Maskin’s model implies that a greater dispersion of skills will yield a wider segregation of workers by skill across plants, and wider cross-plant dispersion of wages and productivity. Caselli’s model has similar implications. Since more skilled workers use better machines relative to less skilled workers, skill biased technical change leads to a greater dispersion of labour productivity and increased wage dispersion between plants. Dunne et al note that these mechanisms give rise to two testable hypotheses: that the dispersion of wages and productivity occurring across plants are linked; and that this dispersion of wages and productivity is linked to differential rates of skill biased technology adoption between plants. They find, along with earlier studies, that much of the increase in industry level productivity is associated with resource re-allocation from less to more productive plants within an industry. They also find that between plant wage and productivity dispersion over the 1975-92 period is an important and growing of total dispersion, that much of this between plant dispersion is within industries, and that a substantial fraction of the rising dispersion of wages, and to a lesser extent productivity, is accounted for by the distribution of computer investment across plants.

The work by Abowd, Haltiwanger, Julia Lane and others (2002) links skill levels, firm productivity and the market value of firms. Their approach is demanding of data on firms and workers, and utilises the matched employee-employer and firm market value data sets available for seven US states. This allows the effects of unobserved skills to be calculated as firm market value less tangible assets. They note that these intangible assets valued by the capital market could include knowledge assets, organisational structure, human capital complementarities, although little progress is made on distinguishing which were the important elements. As with other firm dynamics studies, they find a strong positive relationship between skills and firm productivity. They note that the unobserved component is much more important, and is more correlated with wages and market value than is the observed component of skill levels. The data also allows the effects on firm productivity of different distributions of skills in firms to be explored. They find that the most skilled and least skilled workers have a disproportionately positive and negative effect respectively on productivity.