Firm dynamics and job creation: revisiting the perpetual motion machine
4. Revisiting the 2001 cohort
Search publication by keyword.
The literature suggests that analysing a cohort of firms, over a decade since it was born, would show significantly less dynamism and lower growth rates compared to analysis of growth during the early years of firms’ lives. This is precisely what we find when we analyse rates of growth for the 2001 cohort of firms in New Zealand between 2011 and 2016.
Data sources and definitions
Data sources and definitions used in this analysis are the same as in Meehan & Zheng (2015), with a few exceptions that are noted in this report.
Data on firm births and deaths are taken from the Longitudinal Business Database (LBD). The population of interest is economically active firms operating in the private sector. Data is firm level rather than plant-level. Changes in firms’ legal status, which can confound analysis of firms over time, has been addressed using “permanent” enterprise identifiers that are based on Fabling (2011).
An entering firm, or firm birth, is a firm that has administrative data in year t but not in year t-1. It is possible that some entering firms identified in this way are short-lived firms that enter at time t and exit again in the same year.
Firms are assumed to have exited if they have zero sales or zero employees in all future periods, for which we have data.
Meehan and Zheng also distinguish between permanent and temporary firm exits. A firm that has zero employment or sales in a given period but positive employees or sales in some future period is a temporary exit or “inactivity” rather than exit.
Firm size and growth rates are measured using firm employment: Firm employment is measured using rolling-mean-employee (RME) counts from the Linked Employer-Employee dataset (LEED), which is derived from IRD’s Employer’s Monthly Schedule tax form. The RME count is a 12-month average of employee counts for the year ending March. This employment measure simply uses employee head counts, with no distinction between full-time and part-time employment. Non-employing firms (i.e. working proprietor) firms are included in the analysis, but working proprietors are not included in the employment counts.
Firm activity is not adjusted for part-years. If, for example, a firm is born in July with 1 employee the data will present the firm as being born with half an employee. Future research could usefully address this aggregation problem by analysing firm dynamics by age measured in months rather than calendar years.
The biggest changes in the firm size distribution occur early in the life of a cohort of firms. This can be seen in Figure 2 which charts the distribution of employee numbers for the 2001 cohort in its birth year and again in 2011 and 2016.
The significant change in firm size distribution between 2001 and 2011 – with the entire distribution shifting out to the right – was an effect that occurred very early in the life of the cohort. The change in the distribution between 2011 and 2016 is more representative of how the distribution of firm sizes evolves over time, with comparatively minor contractions in the number of small firms and comparatively minor expansions of the number of larger firms.
Figure 2 Firm size distribution for the 2001 cohort in 2001, 2011 and 2016
In the first two years of life there is significant growth in the number of employees in firms that are born small – with less than 1 employee initially (see Figure 16 in the Chart and data Appendix). This is as much a matter of arithmetic as it is a meaningful dynamic. When small firms grow they cannot help but grow by large amounts, in terms of percentage changes.
As shown in Meehan and Zheng (2015), the first year or two of establishment is also associated with acquiring capital. Capital stocks per employee jump significantly after the first year of activity.
At the same time, a substantial number (20%) of very small firms are inactive one year after birth. Figure 3 shows the share of firms that are active in each year, defined by a firm having non-zero rolling mean annual employee count or non-zero sales in that year.
Taken together, high mortality rates and high growth rates create significant changes in the distribution of firm sizes early in the life of the cohort.
Very small firms – with less than 1 employee – make up a significant proportion of the number of firms (85% in 2001 – see Figure 15 in the Chart and data Appendix). As such, they can significantly affect measurements of firm dynamics even though their shares of employment are very small (less than 3% in 2001).
Figure 3 Activity rates by birth size, 2001 cohort
Share of firms active in the current year, by birth size
Firms in the 2001 cohort have continued to grow reasonably quickly, on average, between 2011 and 2016 (despite the impression given by Figure 2, where the use of natural logarithms means that significant changes are emphasised and more modest changes are harder to perceive). As shown in Figure 4, average number of employees per firm grew from 2.6 to 3.1 between 2011 and 2016 – an average growth rate of 3.6% per year.
Figure 4 Average firm size for the 2001 cohort, 2001-2016
The 2001 cohort exhibits substantial variation in the longevity and growth of firms born small. Firms that are born very small (with less than 1 employee) have the lowest survival rates by a considerable margin – only 32% of such firms survive to age 10 (in 2011) and 27% survive to age 14 (see Table 1). This compares to survival rates that are approximately 20 percentage points higher than for firms in the next birth size group, with at least 1 but less than 6 employees at birth.
Survival rates shown here are the proportion of firms in a cohort that are still alive (still active) at a given date (e.g. a 33% survival rate indicates that 2 out of 3 firms in a cohort are no longer active).
The probability of surviving one more year, for firms born very small, is also comparatively low (hazard rates of 32% in 2011 and 26% in 2015). This means that low survival rates are not simply a matter of many very small firms exiting early in their life cycle. Rather, firms that are born very small tend to have higher rates of mortality throughout their lifecycle. This also suggests there is something peculiar about the nature of firms that are born very small.
It may be that firms born very small are firms that are primarily working-proprietor firms, with owners that do not aspire to, or do not have the capacity to, hire staff and to grow. Or this group of firms may include a substantial number of first-time entrepreneurs who attempt to grow but tend to fail at a higher rate than other firms. In the latter case, high fail rates by small firms may include a substantial amount of learning. Future research is needed to analyse such dynamics and to assess the effectiveness of this kind of learning-by-doing.
Such research could also analyse the extent to which owners of very small firms, that exit, subsequently become owners of businesses that are larger at birth. One hypothesis is differences in survival rates by firm size reflects learning that effects the growth or success of larger firms (rather than simply size at birth).
Note that the survival rates in Table 1 are shown for the year 2015, rather than 2016. This is because the calculation of hazard rates requires knowledge of survival beyond the current year. The LBD is currently limited to observations up to the year ended December 2016. So, with a single cohort, survival for firms beyond 2016 cannot be estimated.
Table 1 Firm size transitions and survival
Number of continuing firms, by number of employees, in 2011 and 2015.
Even as very small firms exit at much higher rates than all other firms, there are a small number of firms that grow significantly. In 2011, 90 firms had grown to have 20 or more employees from less than 1 employee in 2001. Furthermore, this cohort of very small firms continued to grow, with an additional 6 firms entering the size group of 20 or more employees between 2011 and 2015. This indicates that growth of very small firms can be persistent and is not solely a feature of early stage growth.
In the 2001 cohort, for firms born with one or more employees, hazard rates do not vary much in relation to birth size. The range of hazard rates is generally between 6% and 9%. One minor exception to this observation is for firms born with 20 or more employees, where hazard rates were only 2% in 2015 compared to hazard rates of 6%-8% for other firms born with one or more employees. This observation should not be overemphasised, as the number of firms involved is relatively small (150 firms) so modest changes in the number of surviving firms can have large effects on observed hazard rates.
Table 1 shows that a number of firms decline in size quite gradually. In 2011, at age 10, only 51% of firms born with 20 or more employees still had 20 or more employees while 14% of these firms were still operating but were smaller than they were at birth. In 2015, 15% of the 2001 cohort, born with 20 more employees, had employment levels that were lower than in 2001. More broadly, amongst firms born with 6 employees or more, 20% had fewer employees in 2015 than in 2001.
Of the jobs created between 2001 and 2016, by the 2001 cohort, almost half (45%) were created by small firms that grew from less than 1 employee in 2001 to 20 or more employees in 2016 (see Table 2). This mirrors the results of Meehan and Zheng’s (2015) analysis for job creation between 2001 and 2011 (see Table 6 in the Chart and data appendix for updated results for 2011).
In contrast, job destruction is concentrated amongst both small firms with at least 1 but less than 6 employees and large firms of 20 or more employees in 2001. In 2016 these groups accounted for 29% and 47% of jobs destroyed, respectively.
In a departure from Meehan and Zheng, the revised firm data for this study indicates that firms of all birth sizes had positive job creation rates between 2001 and 2011. The previous study found that firms with 20 or more employees in 2001 had negative net job creation rates, where:
- creation rates are the sum of jobs created divided by the average of the number of employees in these firms in the base year (or birth year) and the number of employees in these firms in the year at the end of the evaluation period;
- destruction rates are the sum of jobs destroyed divided by the average of the number of employees in these firms in the base year (or birth year) and the number of employees in these firms in the year at the end of the evaluation period; and
- net job creation rates are the difference between creation rates and destruction rates.
During the 5 years between 2011 and 2016 firms that were born small were also the ones to contribute the most to net job creation. As shown in Table 3, firms that began with less than one employee and grew to 20 or more employees by 2016, created an additional net 1,900 jobs between 2011 and 2016.
In contrast, the largest source of net job destruction was firms born with 10 to 19 employees in 2001 and who had grown to have 20 or more employees in 2011, but then shed workers by 2016. Amongst these firms a net 1,400 jobs were destroyed between 2011 and 2016.
Table 2 Job creation and destruction between 2001 and 2016
 Firms with no employees have been removed from the data, for this example, before transforming the data using natural logarithms. The bandwidth used to construct these kernel densities has been chosen to produce a relatively smooth distribution to assist with visual inspection of changes in the distribution. Default parameters for density estimation in statistical software would usually produce smaller bandwidths and less smooth densities for this data.
 This is the definition used in Meehan and Zheng (2015) for calculating firm survival rates. Here, this definition is referred to as ‘activity rates’. This note differs in the definition of survival with survival rates defined by the firm never again being active – within the data available for this analysis. Classifying temporarily inactive firms as having exited is troublesome because survival rates can increase – if enough firms return to activity. This is counterintuitive. The definition of survival used here is more intuitive, but suffers from bias due to the limited sample length – a bias that is increasing later in more recent years.
 These numbers differ from those in Meehan and Zheng (2015). The data used here has approximately 4,700 (10%) more firms in it than were in the previous analysis. This does affect comparability of some data points but turns out not to affect the qualitative conclusions.
 Hazard rates are the difference between survival rates in the current year and survival rates in the next year, as a proportion of survival rates in the current year.
 Fabling (2018) suggests a substantial proportion of firms are working-proprietor firms. The number of these firms increases during periods of negative labour market conditions. So, some individuals choose to be working proprietors out of necessity and therefore may not be intending to grow their businesses but rather re-enter an employment relationship when conditions allow. >/p>
 This data is restricted to firms that are still active in the next year, so that firms that exit in 2011 or 2015 have been excluded from the counts in the table.
 High survival rates (low hazard rates) for larger firms raises questions about whether these firms take “too long” to exit the market. If a large firm’s performance is lagging but the firm remains rather than exit the market, this could mean efficient resource reallocation is restricted, potentially impeding productivity growth. This analysis, being descriptive, cannot pronounce on whether firms are too slow to exit the market. It can observe that firms that are born large, with 20 or more employees, are less likely to exit (die) than other firms even as they decline in size (with negative growth in numbers of employees).
 As in Meehan and Zheng (2015, p17) “…job creation (destruction) is the number of jobs the 2001 cohort of firm creates (destroys) through firm expansion (contraction) between 2001 and 2011” or between 2001 and 2016.
 This feature also holds for the extended sample from 2001 to 2016 (see Figure 18 in the Chart and data appendix).
 For an algebraic representation of the calculation of these rates see Meehan and Zheng (2015) p.17.