Abstract

High growth firms have been attracting increasing attention from policymakers interested in promoting job growth. Researchers have responded by exploiting newly available large firm-level datasets to study the role of high growth firms in the dynamics of job creation and destruction. This paper uses the recently agreed OECD definition of a high growth firm - requiring sustained growth in employment, averaging 20 per cent per year over three years - to provide benchmark numbers on high growth firms for the UK compiled from the ONS Business Structure Database over the period from 1997 to 2010. It reports counts of high growth firms and measures of incidence (high growth firm numbers relative to the size of the business population) across a set of characteristics: age; size; industrial sector; and location. It concludes with an investigation of job growth which confirms that in the UK high growth firms are, indeed, disproportionately prolific job creators.

JEL Classification: F23, L25, L26, L53, R12 Keywords: High growth firms, job creation

We thank Albert Bravo-Biosca for his support and encouragement, an anonymous referee for comments which improved the paper, and the ONS VML team for their assistance. Corresponding author: Michael Anyadike-Danes. Economics and Strategy Group. Aston Business School. Aston University, Birmingham B1 7ET. Email: [email protected].

The Nesta Working Paper Series is intended to make available early results of research undertaken or supported by Nesta and its partners in order to elicit comments and suggestions for revisions and to encourage discussion and further debate prior to publication (ISSN 2050-9820). © 2012 by the author(s). Short sections of text, tables and figures may be reproduced without explicit permission provided that full credit is given to the source. The views expressed in this working paper are those of the author(s) and do not necessarily represent those of Nesta.

Introduction

Context and motivation

1Over the last thirty years considerable evidence has accumulated (albeit of widely varying quality) which supports a 'pareto-type' conjecture that a relatively small proportion of firms – disproportionately small firms – account for a relatively large proportion of job creation. David Birch is generally credited with having first formulated this conjecture (for an accessible summary see Birch [1981]), and although his claim about the extent of the small firm contribution proved controversial (see for example Davis et al. [1996b]), the conjecture itself became widely accepted quite quickly (see for example the discussion in Storey and Johnson [1987]), and interest in it continues (for two recent contributions see Haltiwanger et al. [2012] and Neumark et al. [2011] ¹).

2Systematic work on the conjecture was long hindered by a paucity of appropriate firm-level data but, particularly since the mid-1990s, as the data deficiency was made good, policy makers began to take an increasingly active interest in this research,

"[We have] the empirical observation that there is typically a small group of firms that are responsible for a large share of new jobs created. These rapidly expanding firms, by way of their supposed or actual potential to generate jobs, have attracted the attention of policy makers, eager to reduce unemployment."Schreyer [2000, p. 6]

This growing interest, in turn, motivated the OECD to initiate a programme of work which aimed both to measure the contribution to job creation of these 'rapidly expanding firms' – christened high growth firms (HGFs) – and to investigate their differentiating characteristics (see Schreyer [2000], OECD [2002]). One of the by-products of this work was an internationally agreed definition of an HGF (set out in the next section) and a chapter dedicated to HGFs in the Manual of Business Demography (see EUROSTAT-OECD [2007, Chapter 8]). Although measuring the contribution to job creation (i.e. calibrating the pareto-type conjecture) played a role in the choice of HGF definition, its potential for use in international comparisons appears to have been decisive in preferring it to the alternative high growth metric proposed by Birch (see the discussion in Ahmad [2006, p. 57], and for Birch's "growth index”, see Birch [1987, pp.36-38]) ².

3In 2008, a year after the publication of the Manual of Business Demography, the OECD began publishing data on HGFs (see OECD [2008, Section B]), though not for the UK. However, there have as yet been relatively few studies of HGF incidence which make use of the OECD definition (amongst them are: Anyadike-Danes et al. [2009], Bravo-Biosca [2011] and Teruel and de Wit [2011]). Of course, there had been studies of HGFs in the period before the OECD definition was agreed (for a comprehensive survey see Henrekson and Johansson [2010]), nonetheless it seems that with respect to HGFs, policy makers have been running somewhat ahead of the evidence – HGF-oriented policy has been enthusiastically promoted even though it is accepted that the evidence base is very weak (for a policy-oriented overview of the evidence see Lilischkis [2011]).

4Of course the newness of the definition may in part explain the scarcity of research. ³ Certainly there has been virtually no systematic investigation of HGFs over time, ⁴ and this is the first gap in the evidence base which we address in the first chapter of the report, using a times series of cross-sections to track the evolution of HGF incidence over time by age, size and sector and the distribution of HGFs across these characteristics. The aim is an improved understanding of the 'nature' of HGFs and to provide 'benchmark' numbers for UK policy makers with an interest in the HGF agenda. The two further chapters of the report address equally under-researched topics. Chapter two examines the distribution of HGF incidence across the 380 Unitary Authorities and Local Authority Districts of Great Britain, whilst chapter three tackles the geography of HGFs and, finally, perhaps a rather more fundamental issue, the contribution of HGFs to job creation. A fourth chapter sums up and a substantial appendix investigates foreign firms, their characteristics and the incidence of high growth amongst them.

Data sources & definitions

1We use the recently released UK Business Structures Database ⁵ (compiled by the Office for National Statistics) ⁶ which records annual data on employees for the entire population of firms in the UK. This data is compiled from a series of annual 'snapshots' of the Inter-Departmental Business Register, an administrative database which captures information from a range of sources, amongst them VAT returns and employer Pay As You Earn (PAYE) tax and social security records. The unit of analysis is an "employer enterprise” – a business with at least one employee ⁷ – which we refer to as a firm. Firms may comprise a number of distinct local units (establishments or plants) but our data refer to firm-level employee numbers.

2We have linked together the annual 'snapshots' from the BSD using firm-level identifiers to form a longitudinal firm-level database (LFLD) for the UK and have devised algorithms to produce firm-level demographic markers for 'birth' and 'death'. The birth of a firm is dated by the first appearance of non-zero employment and its death is treated symmetrically and dated by the disappearance of the last employee. The data do not distinguish between de novo births and those which result from the break-up of an existing firm, similarly the data do not distinguish between the closure of a firm and its disappearance due to merger. Although the data start in 1997, firms alive in 1997 could have been born in any previous year, so the first birth year we can identify with certainty is 1998.

3Firms are classified as either 'private' or 'public' sectors and we make this split using the classification by industrial sector. All employees in: public administration and defence; education; and health and social work (SIC92 ⁸ sections L, M, N); are classified as public sector. Of course, some firms in these sectors (in health and/or education for example) are private, and some firms in our private sector are public, but ours is a reasonable approximation and ensures that most typically longer lived public entities (like schools and hospitals) do not distort our calculations.

4The first stage in the OECD metric for identifying an HGF (see EUROSTAT-OECD [2007, Chapter 8]) requires that we consider only firms which,

• are born before the beginning of the period
• are alive at the end of the period

These two requirements imply that in each period we will have a 'balanced panel' of firms – the same firms are always present throughout the period (often referred to as 'continuing firms').

An HGF is a firm in the balanced panel which,

• has at least 10 employees at the beginning of the period
• records an annual average growth of 20% in employment over the period ⁹

5Finally, we define HGF incidence ¹⁰ and the 'incidence rate' as the number of HGFs divided by the number of firms (in the balanced panel) with 10+ employees ¹¹ We use three years as our 'period': so, starting with 1998, there are ten 3-year periods: from 1998/2001 to 2007/2010 – is is the 'rolling balanced panel' (RBP) used throughout this report.

1 HGF incidence and characteristics

1.1 HGFs: age, size & sector

1.1.1 Headline figures

1The basic data on HGFs for periods 1998/2001 to 2007/2010 are displayed on Figure 5.1. The HGF numbers are plotted by bars (scale on the left hand side) and the incidence rate measured as a percentage is plotted by the line (scale on the right hand side).

2There is quite a clear pattern in the HGF numbers a 'bulge' beginning in 1999/02 and lasting for three periods with close to 13,000 HGFs in each period – then, for the six periods 2002/2005 onwards, numbers have been around 10,000 per period. The incidence rate followed a similar course to the numbers – during the bulge the incidence rate was close to 9%, since 2002/2005 it has fluctuated within a narrow range, just below 7.5% ¹².

1.1.2 Age

3Each period includes firms born a varying number of years previously and we can classify them ¹³, using their date of birth, into 'birth cohorts': cohort98, those born in 1998; cohort99, those born in 1999; and so on. There is one important exception: the firms alive in 1997 (the first year of our dataset), are a unknown mixture of firms born in every preceding year and, although these firms are not a 'true' birth cohort, for convenience we refer to them as cohort97* ¹⁴.

4From Figure 5.2, where the incidence rate is plotted against period as before, we can see the effect of distinguishing incidence by date of birth. Evidently the incidence rate for cohort97* is very much lower than that for the other cohorts (and the average of the other cohorts). Moreover we can see, as might have been anticipated, that the overall 'average' incidence rate we saw in Figure 5.1 is largely driven by cohort97*.

5The display in Figure 5.2 also puts the 'bulge' in interesting perspective. Whilst it appears that the cohort97* rate of incidence is its (proximate) cause, the incidence rate excluding 1998/2001 'slopes' downward (albeit from a much lower level) and looks rather more similar to that of other cohorts. It appears that the regular flow of new cohorts – each of which records initially a relatively high incidence rate – serves to offset the slow decline of cohort97* incidence, and keeps the all-age incidence rate broadly flat.

6It is easier to visualise the relationship between age and HGF incidence if we re-plot the data with age rather than period on the horizontal axis. On Figure 5.3 cohort97* now looks even more similar to the subsequent cohorts. Although the cohort97* incidence rate is always very much lower, with the exception of the 'jump' between age 1 and age 2, incidence declines steadily with age. All the other cohorts record incidence rates between 14% and 16% at age 1, with the earlier cohorts (98, 99, 00) rather closer to 16%, later cohorts are typically closer to 14%. On average, the incidence rate declines by about 0.5 percentage points per year of age ¹⁵.

1.1.3 Size

7We have classified HGFs into size-bands using size in the year preceding the beginning of the period (i.e. the same year used to determine whether a firm has 10 or more employees as required by the HGF definition). So, for example, the allocation into size-bands for the 1998/2001 period is based on firm size in 1997. The size-bands are: 10 to 19; 20 to 49; 50 to 99; 100 to 249; and 250+.

8It is difficult to discern any simple pattern in the display of incidence over periods by size-band in Figure 5.4. It is clear that the steep rise between 1998/2001 and 1999/2002 noted earlier is proximately due to the two smallest size-bands – 10 to 19 and 20 to 49 - most of the others either decline or are flat between the first two periods.

9After 2002/05 the incidence rate curves all flatten, and there is no stable ranking by size, although the two smallest size-bands seem to be consistently above the others – with the 20 to 49 size-band almost always recording the largest incidence rate.

10In order to probe the incidence/age relationship we need to split off cohort97*, because of its weight it dominates the overall incidence figure. However it is not possible to distinguish each of the birth cohorts because some of the age/size-band combinations fall below the permitted disclosure limit, ¹⁶ So we have an all cohort – except cohort97* – incidence rate series by size-band, which is plotted against age in Figure 5.5.

11We can now see a relatively clear pattern – most size-bands decline with age – except 250+ which is essentially flat. The 20 to 49 size-band has the largest incidence rate, but its incidence rate becomes more similar to those of the other size-bands as the cohorts age. By age 9 (by which date, admittedly, we have only observations from cohort98) all the size-bands, including 250+, are recording about the same rates of incidence.

12Evidently size effects on HGF incidence are second order (leaving the 250+ size-band to one side) when compared to age effects. ¹⁷ Of course, we might have inferred that size was unimportant from the overall incidence figure alone, but we can now see its relative unimportance more clearly.

1.1.4 Sector

13We have computed incidence for a reasonably fine-grained industrial classification – 47 categories of the 2-digit UK standard industrial classification(SIC92 ¹⁸). Although firms are very unevenly distributed across sectors, in only one sector does the number of 10+ firms fall below the lower disclosure limit (the denominator for the HGF incidence rate calculation). Incidence rates vary widely, by a factor of ten, from a minimum under 2% ¹⁹ to a maximum of 25%.

14The simplest way to approach detecting a pattern in incidence across sectors is to rank them. We have ten sets of ranks, one for each period (1998/2001 to 2007/2010) and we then compute an average rank for each sector. The 'top 10' sectors turn out to form quite a compact group: there are just ten sectors with an average rank of ten or higher, and all of them are typically ranked ten or higher in each period. Just five sectors outside this top 10 are ever in the top 10. For example, the sector at 11th place in the list of average ranks, records a top 10 rank in only two periods.

15The top 10 sectors by average rank are plotted on Figure 5, from 1 to 10 they are,

• 64 – Posts and telecommunications (KIS-HT)
• 72 – Computer and related activities (KIS–HT)
• 66 – Insurance and pension funding (KIS)
• 73 – Research and development (KIS–HT)
• 90 – Sewage and refuse disposal (LKIS)
• 67 – Activities auxiliary to financial intermediation (KIS)
• 65 – Financial intermediation (KIS)
• 74 – Other business activities (KIS)
• 71 - Renting of machinery and equipment and household goods (KIS)
• 93 – Other service activities (e.g. dry cleaning, hairdressing, funerals, fitness) (LKIS)

Three of the top four places on the list are occupied by the (only) three 2-digit sectors which EUROSTAT classify as "High-tech Knowledge Intensive services” (KIS-HT), another five are also "Knowledge Instensive Services”(KIS), whilst the final two are "Less Knowledge Intensive Services" (LKIS). So 'the top 10' are all services, and nearly all are knowledge intensive services. ²⁰

16It is worth noticing too, that the bulge between 1999/2002 and 2002/2005 is very pronounced in only a handful of the top ten. But the two sectors most affected – 'posts and telecommunications' and 'computers and related services' (both of them hi-tech knowledge intensive services) are very strongly affected. ²¹ In both cases the incidence rate is between 10 and 15 percentage points higher in the three bulge periods than it is later on. For most of the other sectors the differential is much closer to five percentage points.

17To confirm that focusing on the 'average' top 10 is not too misleading we can look at the other five sectors which are ever ranked in the top 10: two are knowledge intensive services (Recreational, cultural and sporting activities (92), Real Estate Activities (70)); one is a less knowledge-intensive service (Travel Support Consumer Services (63)); one is hi-tech manufacturing (Manufacturing of office machinery and computers (30)); and finally Construction (45), which (oddly) seems not to appear at all in the EUROSTAT classification.

18It is difficult to investigate the sectoral story in much more detail. We know already that the combined sample is dominated by cohort97*, especially in the early years and it is not possible to compute reliable estimates of incidence (even leaving disclosure issues aside) for a classification as fine-grained as 2-digit sectors if cohort97* is excluded. However, if we apply the same (admittedly rough-and-ready) ranking procedure to the data – ranks of incidence by 2-digit sector averaged for periods 2002/2005 to 2007/2010 – we get broadly similar results for the top 10 for cohort97* and for not-cohort97 as we had for the pooled data (though with not precisely the same ordering).

1.2 The distribution of HGFs across characteristics

1Using the OECD definition of an HGF we have focused on the incidence rate. This is a measure which answers a question like: what proportion of one year old 10+ firms record an instance of high growth? However there is an alternative measure for summarising the importance of HGFs that answers a different question, one like:

what proportion of HGFs are one year old?[^1] Here we will examine the distribution of HGFs across age, size and sector.

2As Table 5.1 confirms most HGFs are relatively old (it is a pattern might have been anticipated given what we know about the relative size of cohort97*) – in 2007/2010 almost 50% are more than 10 years old. Otherwise the distribution looks relatively flat, with shares between 5% and 7% at each age, but with a slight tendency to decline past age 6. Of course, data for previous periods becomes less and less informative the further we go back in time because we are unable to distinguish the ages of firms born before 1998, and so we have displayed just the 2007/2010 period. [^2] Evidently, although the incidence rate declines with age, the 'weight of the past' ensures that most HGFs in any particular period are quite elderly.

3Figure 5.7 displays the distribution of HGFs by size-band. The most striking feature of the data is that half of all HGFs are found in the smallest, 10 to 19 size-band, and very, very, few – less than 5% – are in the largest, 250+ size-band. The bulge in the first few years is once again clearly evident and it is concentrated in the 10 to 19 category.

[^1]These two measures are, in fact, quite closely (algebraically) related and the link between them is a "location quotient"(LQ)-type statistic. To see this, first write the incidence rate (inc) for an instance i of some characteristic (like age) of an HGF as the number of HGFs divided by the number of firms with 10 employees or more (f¹⁰⁺),

incᵢ = HGFᵢ / f¹⁰⁺ᵢ

Now 'standardise' the incidence rate for characteristic i by expressing it as a ratio to the average incidence rate (inc),

incᵢ / inc = (HGFᵢ / f¹⁰⁺ᵢ) / (ΣHGFᵢ / Σf¹⁰⁺ᵢ)

We can re-arrange this expression into a LQ-type statistic for an incidence rate i of some characteristic (like age) of an HGF relative to the population of f¹⁰⁺ with that same characteristic as,

(incᵢ / inc) = LQi = (HGFᵢ / ΣHGFᵢ) / (f¹⁰⁺ᵢ / Σf¹⁰⁺ᵢ)

The numerator of the LQ is the distribution of HGFs across different values of the characteristic (e.g. the ratio of one year old HGFs to all HGFs) and the denominator is the analogous distribution for 10+ firms.

[^2]For example, in 1999/2002 the split is 5.5% one year old and 94.5% greater than one year old; in 2000/2003 it is 4.9% one year old, 6.0% two years old, and 89.1% more than two years old; and so on.

4HGFs are quite heavily concentrated by sector: together the top five sectors – whose shares are displayed on Figure 5.8 – account for more than half of all HGFs in each period. Remarkably, around one fifth of all HGFs are in business services (sic74), with construction (sic45), second on the list, having just a 10% share. Notice too that the share of business services grows quite markedly in the post-bulge period: in 2001/2004 it was 18%, by 2007/2010 it was seven percentage points larger at 25%.

5Finally, it is worth noting that there is no overlap between the top five on the shares list and the top five on the incidence rate list. Indeed, of the top five by share only one, business services, is in the incidence rate top ten (it is ranked seven). Moreover, the other sectors with shares in the top five – construction (sic45), wholesale distribution (sic51); and retail distribution (sic52); hotels and restaurants (sic55) taken together account for about one third of all HGFs, even though the HGF incidence rate in most of these sectors is typically quite modest. Evidently, with the partial exception of business services, we may infer that the distribution of 10+ firms by sector looks quite different to the sectoral distribution of HGFs: the other members of the top five must account for a disproportionate share of larger businesses.

1.3 Discussion

1As mentioned earlier, research on HGF incidence using the OECD definition is relatively scarce. However, at roughly the same time as the OECD definition was agreed, Henrekson and Johansson 2010 surveyed the empirical literature on HGFs, and their findings provide a natural context for our results on the UK. After a systematic search of bibliographic databases (from 1990 to 2008) H&J compiled a list of 20 studies which analysed data from a range of countries (though not the UK). They organised their results about the characteristics of HGFs – which they refer to as "Gazelles”[^3] – around three propositions (Henrekson and Johansson [2010, p. 228]),

• "On average, Gazelles are younger"
• "On average, Gazelles are smaller than other firms"
• "Gazelles are over-represented in high-technology industries"

2As H&J recognise there is considerable variation across the studies they survey in just about every dimension (definition of HGFs,[^4] measurement of growth, choice of time period) and in the classification of HGF characteristics (sectors, size-bands, age range). However,

Sometimes this [heterogeneity] is a drawback since comparability is impaired. However in this case the large variation should be seen as an advantage, since the results regarding the importance of Gazelles turn out to be quite robust. Regardless of method, definition, time period etc. some findings emerge.” Henrekson and Johansson [2010, p. 240]

3With respect to age, H&J's answer was clear,

“The results regarding age are unambiguous. All studies reporting on age find that Gazelles tend to be younger on average.”Henrekson and Johansson [2010, p. 240]

We found that the incidence rate declined with age – a larger proportion of younger firms are HGFs; but we also found that in 2007/2010 almost half HGFs were ten or more years old. Since H&J seem to be referring to the first, the greater propensity of young firms to grow, the UK evidence appears to be consistent with their proposition.

4H&J's conclusion about size is more nuanced:

“... the results are ambiguous. Gazelles can be of all sizes... It appears that newness is a more important factor than small size.” Henrekson and Johansson [2010, p. 240]

This also fits with our findings: incidence by size-band showed little variation by period, and when incidence by size-band is displayed against age the picture is dominated by the decline with age: size plays only a secondary role.[^5]

5H&J find the proposition about the distribution of HGFs by sector to be decisively rejected:

"There is no evidence that Gazelles are overrepresented in high-technology industries. Gazelles exist in all industries... [However] they appear to be overrepresented in services." Henrekson and Johansson [2010, p. 240]

If "overrepresented” is interpreted as an LQ-type proposition: that HGFs are a larger proportion of 10+ firms in hi-tech sectors than they are in non-hi-tech sectors (see footnote 12), then the UK is different. However, if "overrepresented” means concentrated, then our evidence is consistent with H&J's proposition: since a relatively small proportion of the UK's HGFs are found in hi-tech sectors.

6There are two more recent studies which relied on the OECD definition: Bravo-Biosca [2011] and Teruel and de Wit [2011]). The second of these – a cross country investigation of the determinants of incidence used the turnover variant of the HGF definition and was restricted to firms with more than 50 employees (see Teruel and de Wit [2011, p. 9]) – and does not report results which are comparable with ours. Bravo-Biosca [2011] is also a cross-country study which covers (amongst other things) HGF incidence rates and, like us, uses the employment variant of the OECD definition, in his case computed for the three year period 2002/2005. A section on HGFs records some headline results Bravo-Biosca [2011, pp. 18–20] for a varying selection of countries,

• the incidence rate[^6]is higher for younger firms, but most high-growth firms are older (5 countries: Norway, Austria, Netherlands, Italy, Finland)
• most high-growth firms are small, but large firms can achieve high growth (9 countries: Norway, Austria, Netherlands, Denmark, Italy, Finland, Spain, United States, UK)
• high-growth firms are everywhere, not only in hi-tech or "innovative” sectors (9 countries, Norway, Austria, Netherlands, Denmark, Italy, Finland, Spain, United States, UK)

7The findings for age and size match ours, on both the incidence rate and the distribution measure, however the findings on sectors do not. Certainly we found HGFs to be ubiquitous, and also agree that "A majority of high growth firms are found in the service sectors..."Bravo-Biosca [2011, p. 19] but, using data for 2-digit sectors we found that the incidence rate was highest in hi-tech and knowledge intensive services. Bravo-Biosca's data was more highly aggregated – 4 broad sectors – and he found no detectable cross-sector pattern in incidence.

Chapter 2 hgf geography

2.1 Introduction and summary statistics

1This chapter describes the spatial distribution of hgf incidence rates across the Unitary Authorities and Local Authority Districts (UALADs)[^7] in Great Britain. It is organised into two sections. This first section provides an introduction to the distribution and its evolution over time and then introduces the summary statistic we use here: the distribution of the median HGF incidence rate of each UALAD. The second section maps the data for six broad regions of GB and provides a running commentary on the pattern in the location of places in the top 25% of the distribution.

2Our starting point is the 380 local authorities on which we have HGF incidence rate observations for each of the ten three year periods in the 'rolling balanced panel'[^8] – 2002/05 to 2007/10. A set of period-by-period boxplots displayed on Figure 5.9 provides a useful summary of the data. Evidently the dataset is reasonably compact – there are relatively few outliers (given that there are almost 400 observations per period). It is also worth noting that the median of the UALADs incidence rates, period-by-period, tracks the national HGF incidence rate quite closely, as you will see from Figure 5.10.

3Obviously since we wish to investigate the spatial pattern of the HGF incidence rate it is impractical to work with 380 observations for each of ten periods, we need a summary statistic. As HGF incidence rates in the early periods – from 1999/02 to 2001/04 – look atypically high (and this equally true of the national figure as of the median UALAD rate), we have focused on the later period 2002/05 to 2007/10. So we represent each UALAD by the median of its incidence rates in the six periods between 2002/05 to 2007/10.

4Our basic dataset – an HGF incidence rate for each UALAD in GB – is plotted on Figure 5.11. The 380 observations have been organised in ascending order, from the minimum (3.35%, Conwy in Wales) to the maximum (12.2%, Rushmoor, South East England). The x-axis of the figure is labelled: "SQ1", "SQ2”, through to "SQ8”. These labels indicate the upper boundaries of the semi-quartiles of the data – as quartiles divide the data into quarters, so semi quartiles divide it into eighths – quartiles refer to the 25th, 50th, and 75th percentiles, by analogy semi-quartiles refer to 12.5th, the 25th, the 37.5th etc percentiles. SQ1 is the bottom semi-quartile and SQ8 is the top and the HGF incidence rates at the semi-quartile boundaries have been recorded on the plot just above the x-axis tick marks. The number of UALADs in each semi-quartile varies slightly because there are tied values, and because 380 is not exactly divisible by eight.

5The most immediately striking feature of Figure 5.11 is the limited extent of variation across a very large proportion of the local authorities. Across the middle half of the distribution – 213 observations – from the upper bound of SQ2 to the upper bound of SQ6, the increase is from 5.95% to 7.66%, just 1.71 percentage points and the difference between successive semi-quartiles is typically around half a percentage point. After SQ6 the 'slope' of the curve increases and the difference between the SQ6 upper boundary and the SQ7 upper boundary is twice as large, and beyond SQ7 the rate moves up sharply: the difference between the SQ7 and SQ8 upper boundaries is almost four percentage points.

6Our focus in what follows is on the top end of the distribution: locating the UALADs which record HGF incidence rates in the top quartile of the distribution and, because the top semi-quartile stands out so clearly, we also distinguish SQ8 from SQ7.

2.2 Spatial distribution of HGFs

7With a relatively large number of datapoints – 380 in all a practicable approach to investigating the spatial pattern incidence requires that we group the the UALADs into broader spatial aggregates. Here we use regions as building blocks: the nine (formerly government office) regions of England; and the two devolved administrations, Scotland and Wales. The first step is to tabulate the frequency distribution of HGF incidence rates by semi-quartile across these areas, the counts are displayed in Table 5.2.

8If the distribution of incidence rates were uniform across regions then, of course, the frequency in each cell across a region's row would be the same. So for example the East Midland (EM) in the first row with 40 local authoritiess would record five local authorities in each semi-quartile. Evidently it does not – the count for SQ3 (at the lower end of the distribution) is 11, over 25% of all its local authorities. By contrast, the count for SQ8 is two, half as many as might be 'expected' in the top semi-quartile.

9Whilst there is no particular reason to expect each region's UALADS to be distributed uniformity across the semi-quartiles, the 'independence of rows and columns' hypothesis (although it is a traditional "null" for a table of counts like this one with fixed margins), does provide a benchmark from which to start. The 'excess' recorded as the last column of the table is the difference between the SQ8 column and 12.5% of the regional total (for example -2 for the East Midlands) and the most striking, and extraordinary, departure from this benchmark is London: 21 of its 33 UALADs are in the top semi-quartile, where we might have expected about 4, so the 'excess' is +17. London has 64% of its UALADs in SQ8 rather than 12.5%

10Necessarily if London is over-represented – it alone accounts for almost half of SQ8 local authorities – then at least one other region is under-represented and in fact most are, as we can see from the last column of the table. However two other regions still do stand out as relatively well-endowed: Scotland exceeds the benchmark 12.5%, whilst the South East matches it. Nonetheless there is a spatial rationale for including the second top semi-quartile (SQ7): London so dominates SQ8 that we need to consider SQ7 too if we are to be able to differentiate between places in regions outside London.

2.2.1 London

11In the following commentary, organised by region and groups of regions, we look at the places which populate the top end of the distribution of HGF incidence rates. We start with London itself and then consider the Greater South East, which comprises the two regions – the South East and East of England – which surround London. Next we move north, to Scotland and then discuss in turn: the North of England – comprising the North East, the North West, and Yorkshire and the Humber; followed by Wales and South West England; and finish with the Midlands – the East Midlands and the West Midlands taken together.

12In each case we plot local authorities at their centroids using the Ordnance Survey National Grid (1 unit of easting or northing on the scale is equal to 1 km).[^10] All but one of the grids are on much the same scale and display an area of about 300 × 300 kilometres. The London grid is much more detailed covering just 55 × 55 kilometres. Distinguishing symbols are used to differentiate SQ8 places and SQ7 places from each other, and from the places with rates lower down the incidence rate distribution.

13The location of London's 33 local authorities are plotted on Figure 5.12 which is (approximately) centred on the City of London (grid reference (E532, N181)). Indeed, using a euclidean distance[^11] measure, all UALADs are less than 25km from the City of London.

14We saw on Table 5.2 that 21 of the 33 London UALADS are in SQ8 (and one SQ7) and these places are noticeably more concentrated relative to the City of London – quite close to the centre and stretching to its west, whilst there are rather fewer towards the east and the south. Figure 5.13, a circular plot[^12] – with places represented by their angle (relative to a vertical axis centred on the City of London) – provides a graphical method for visualising data on orientation.

15Figure 5.13 incorporates an arrow – the 'mean direction' of the points this is a summary measure of orientation. It is very slightly north of west (at 285 ° from due north). Extending the data to include the single SQ7 point which, as we can see, is to the south of the City of London in the calculation of the mean direction changes the result, but only very slightly (to 279 ° from due north). So this summary measure further confirms the visual impression about the preponderantly westerly location of London places with high incidence rates.

2.2.2 The Greater South East

16Next we widen our focus to consider the Greater South East (GSE) (comprising the English regions South East and East of England) which surrounds London, and Figure 5.14 displays the location of these 114 places. Most of them are within 100km (euclidean distance) of the City of London, and only seven all in northern Norfolk – are further than 125km. More notably, eight of the 39 places in the GSE within 50km of the City of London are SQ8 (one of the two SQ8 'outliers', Rushmoor in Surrey, is just 53km away the other, West Berkshire is rather further). So the places in the GSE with the highest incidence rates are disproportionately concentrated relatively close to London, or to put it conversely: almost no SQ8 places in the GSE are very far from London. Moreover, as we can also see from Figure 5.14, the SQ8 places are typically to the west of the City of London, sharing a similar orientation as the SQ8 places in London (indeed the mean direction, including all SQ8 places in the GSE, is 249 ° from due north).

17The overall visual impression of the pattern changes very little when we include the 18 SQ7 places in the GSE, because about half of them (eight of 18) are within 50km of the City of London, scattered across the 'home counties' lying around the borders of London, particularly Surrey, Berkshire, Middlesex, Buckinghamshire and Hertfordshire. However the more remote locations (beyond 50km) also form an interesting group of (mainly) small cities. To the north of London: Cambridge, and more remotely the city of Norfolk (and close by Great Yarmouth); to the north west, we have Oxford and (and nearby Cherwell); and in the south a string of towns, Brighton, Horsham, Chichester and Winchester. [^13]

18Turning now from those places which are represented on either the SQ8 and SQ7 lists and looking for those places which are not on the list, even though they are relatively close to London: we have, almost entirely missing, local authorities located in Kent, Essex and Suffolk. Evidently London casts a 'shadow', particularly towards the east, in which relatively few places from the top end of the HGF incidence rate distribution are to be found.

2.2.3 Scotland

19Of course the further we move away from London the less likely is distance or direction from central London to be an influence on the concentration of the top semi-quartile of HGF incidence rates. Indeed, as we saw earlier, Scotland's share of SQ8 locations is slightly larger than the 'norm'. There are 32 UALADs in Scotland so four would have been expected in SQ8, and there are six, however the pattern in their location is easier to appreciate if we consider the five SQ7 places too.

20Figure 5.15 displays the location of 29 of Scotland's UALADs (the three missing are 'off the grid' and very remote: Eilean Siar; the Orkney Islands; and the Shetland Islands). The plot is centred approximately on Stirling, one of Scotland's SQ8 locations. Just to the south of Stirling we can see a 'line' of SQ7/SQ8 locations lying along the 'central belt' of Scotland – seven contiguous UALADs stretching from Renfrewshire in the east to the City of Edinburgh in the west, and comprising: East Renfrewshire; City of Glasgow; North Lanarkshire; South Lanarkshire; and West Lothian.

21There is another group of three SQ8 locations, more northerly, and much further from central Scotland: Aberdeen City, Aberdeenshire and Moray UALADs. Activities in the first two of these at least, it seems reasonable to suppose, are connected with the oil industry.

2.2.4 The north of England

22The north of England – taking together the North East, the North West, and Yorkshire and the Humber – includes 72 local authorities, and with just five SQ8 locations between them, they are clearly under-represented at the top end of the incidence rate distribution.[^14] The local authorities are plotted on Figure 5.16.

23Most of the places in the top two categories are either urban places, or peripheral to them: in the top right hand we have North and South Tyneside (neighbouring Newcastle upon Tyne) and Middlesbrough; towards the middle of the map we have Leeds; and in the lower left of the grid a Manchester group of six local authorities (four of these local authorities are in Greater Manchester Metropolitan County), with Manchester towards the south and South Ribble and Hyndburn to the north; and finally, to the west, Liverpool. Further south, and a little remote, but possibly associated with Manchester/Liverpool, is Cheshire West and Chester. The only real 'outlier' in this pattern is, right in the middle of the map, Richmondshire in the Yorkshire Dales

2.2.5 South West England and Wales

24South West England and Wales taken together, have 59 local authorities but just four SQ8 locations between them, but including the eight SQ7 places gives a 20% share for the top two semi-quartiles. The locations are plotted on Figure 5.17. It may appear initially that urban places, or near-urban places are rather less important than in the north of England. However in Wales we have the capital city of Cardiff (with Blaenau Gwent and Rhondda, Cynon Taff nearby), though the other two are quite remote, relatively rural, places (Gwynedd and Pembrokeshire); whilst in the South West, Bristol, Bath and South Gloucestershire form a contiguous 'cluster', with Gloucester not far away. There are also two substantial towns: Torbay (essentially Torquay) and Plymouth, leaving West Somerset as the South West's only remote and sparsely populated 'outlier'.

2.2.6 The Midlands

25The Midlands is the most under-represented of all the areas: there are 70 local authorities, and only two of them are SQ8 places. Even more strikingly, neither of those SQ8 places are in the 30 UALADs in the West Midlands – the West Midlands might have been expected to have four places at the top end of the incidence rate distribution, but it has none. Nor does taking SQ8 and SQ7 together improve the picture very much: the combined Midlands share is 13%, nine locations about half the 'expected' 18.

26The locations of local authorities in the Midlands are plotted on Figure 5.18. Almost half of the SQ7/SQ8 places are relatively rural places on the northern periphery of the grid. Of the two more central places: Ashfield is a suburban extension of nearby Nottingham; whilst North West Leicestershire, it may be worth noting, contains the East Midlands airport. Of the other three places, lower down the map, Warwick and Rugby UALADs are contiguous, but North Warwickshire is better regarded as a rather more rural 'outlier'.

Chapter 3

Job creation by HGFs

3.1 Background

1Whilst the OECD definition of HGFs does seem broadly accepted there has been little discussion of how to measure the HGFs contribution to job creation, and certainly there is no agreed methodology. This lack of agreement is all the more puzzling because the initial rationale for the identification of HGFs was, in fact, their role as prolific job creators. So our motivation here is simple, to investigate how best to answer the question: "what proportion of job creation is contributed by high growth firms?"

2Right at the start we face a difficulty. The number of firms is a stock – measured at a single time point, whereas job creation is a flow – the difference between the stock of jobs at two different time points. Consequently the relationship between the job creation flow and the stock of firms (and by extension the stock of high growth firms) depends on the length of the measurement period. This dependence is important because many firms have relatively short lives and so, as the measurement period lengthens, larger numbers of firms do not survive. Equally, as the measurement period lengthens, larger numbers of new firms are born within the period (indeed firms may be born and die within the measurement period). These side-effects of a lengthening measurement period render the short period dynamics of labour market flows increasingly invisible and serve to blur the distinction 'new' and 'continuing' firms and their relative contributions to job creation

3.2 The accounting framework

3Using the three year variant of the OECD definition of HGFs effectively commits us to a three year measurement period (t to (t+3)) for our investigation of job creation. An obvious starting point is to distinguish between job creation by HGFs from t to (t+3) and job creation by not-high growth firms (nonHGFs) from t to (t+3). However, there is a complication the OECD definition of HGFs covers only firms which are at least one year old (that is born in (t-1) or earlier), so the OECD HGF definition does not cover:

• any firms born in period t and alive in period (t+3)
• any firm born after period t up to and including period (t+3)

Firms in the first category may have jobs at time t and (t+3), whilst those in the second category may only have jobs at (t+3). So if we are to have a complete accounting for all jobs created between t and (t+3) consistency requires that we include these firms which may create jobs but are not classified as either HGFs or nonHGFs.

4For these two reasons the three year measurement period and the precise character of the HGF definition – we need to adapt the conventional (annual) job creation and destruction accounts.[^15] Here we focus on job creating firms only, and we distinguish four categories,

1. HGFs, born before t, and alive (t+3), at least 10 jobs in t and 20% average annual growth between t and (t+3) – HGF
2. nonHGFs, born before t and alive in (t+3) with more jobs in (t+3) than t, but not HGF – nonHGF
3. firms born in period t and alive (t+3) with more jobs in (t+3) than t – young
4. firms born after period t and alive (t+3) with jobs in (t+3) – new

5We can lay out the accounting framework more formally starting with the relationship between the stock of all job creating firms alive at time (t+3) (firmst+3) and its components, where the first in the pair of subscripts refer to the year of birth,

---

Footnotes

firmst+3 = HGF(beforet),t+3 + nonHGF(beforet),t+3+ + youngt,t+3+ new(aftert),t+3 (3.1)

6HGFs and nonHGFs will be referred to below (as elsewhere) as members of the 'balanced panel' of firms which comprises all firms born before period t and surviving to (t+3)). It is also helpful, again as we shall see below, to distinguish between those relatively large non-HGFs which (like HGFs) have 10 or more employees (nonHGFla) and those that do not, that is relatively small nonHGFs (nonHGFsm) – the larger nonHGF category is a useful comparator for the HGF category because it is so similar (by construction).

7This framework has been designed to account for job creation, and firms which do not create jobs have simply been left out of the picture – some of these will have the same number of jobs at period (t+3) as they had at t, some will have fewer, others will have died (so no jobs at (t+3)).

3.3 Job creating firms, 1998/2001 – 2007/10

8The first three columns of Table 5.3 provides some context for our discussion of job creating firms. Column (1) records the overall number of firms at the end of each three year period, it starts at 1.34 million in 1998/01 and is 200,000 larger by 2007/10, whilst the second column reports the number of job creating firms, it also expanded and by 2007/10 it was up by 150,000 from 0.69 million in 1998/01. However after 2005/08 both series drop, and the fall of 100,000 in job creating firms by 2007/10 matching the fall in the all firm total. The change in both series is broadly in parallel and the ratio between them recorded in column (3), which fluctuates (except for 1998/01) in quite a narrow range between 55% and 60% shows no definite trend.

9The right hand side of 5.3 displays data on the numbers in each category of job creating firm, and the numbers of new firms – in the range 500 to 650,000 – considerably exceed all the other categories yaken together. There are about half as many nonHGFs as there are new firms, but the small nonHGFs are clearly the more numerous.

The number of larger nonHGFs – typically in the 40 to 50,000 range – are similar to those of the young, whilst HGFs – generally around 10,000 – are the smallest of the categories of job creating firms by a very large margin.

10As we saw above the two most notable features of the job creating firm numbers are the rise of 250,000 between 1998/01 and 2005/08 and the subsequent fall of 100,000 from 2005/08 to 2007/10. Whilst the rise is almost equally shared by smaller nonHGFs and new firms, the fall is entirely accounted for by the number of new job creating firms. The numbers (in column (8) dropped by 50,000 per period in both 2006/09 and 2007/10. It seems reasonable to suppose that this pattern may well be cyclical, during an expansion phase (1999/02 to 2005/08) the numbers of new firms are likely to rise but with the onset of recession the number of new job creating firms is likely to be depressed. The mechanics here are in part a by-product of the definition: even if the number of new firms being born in each year of the period (remember new firms are born after period t) remain relatively unchanged, an increase in deaths of the new-born over the period (between t and t+3) will reduce numbers in the new category. The numbers of small nonHGFs are not affected by the downturn, and the other three categories – HGFs, large nonHGFs and young – have similar numbers in 2007/10 as they had in 1998/01.

11Figure 5.19 displays the data on the different categories of job creating firms as shares of the total, and serves to reinforce what has just been learned from Table 5.3. Notice that in order to plot the new firm share on the same chart as the other categories it has its own (right hand) scale, however the scales have been arranged so that the tick marks for both the left and right hand scales are the same distance apart (five percentage points) to make change comparisons easier. The share of new firms fluctuates quite widely, but averages about two thirds and we can see that fluctuations in new firms is almost exactly offset by fluctuations in the share of the smaller non-HGFs about a much smaller average share, around one fifth. The conjunctural interpretation looks even more plausible. The shares of the other three categories are essentially constant – young firms and the larger nonHGFs at 5%, and HGFs at a little above 1%.

12The choice of denominators used in HGF job creation comparisons is not covered by any agreed standard (not surprising perhaps since, as mentioned earlier, the subject seems to be discussed so rarely).

In an earlier study of HGF numbers (Anyadike-Danes et al. [2009, p. 18]) we used the ratio of HGFs to all larger firms in the balanced panel to measure the contribution of HGFs to job creation, that is the number of firms with 10 or more employees whether or not they were job creating firms – this yields an HGF ratio between 6% and 7%. The OECD seem to suggest a slightly different denominator: "the population of enterprises with ten or more employees”, OECD [2011, p. 74] – apparently including young firms with 10 or more employees as well, and this is of course all firms not just job creating firms, and using this denominator produces an HGF ratio less than half a percentage points smaller than 6%.

3.4 Job creation, 1998/01 – 2007/10

13The first two columns of Table 5.4 record the total number of jobs in the initial and terminal years of each three year period. We can see that employment has generally been rising over the period, with an overall average level around 18.5 million. From column (3), which records the difference between them – the net increase in jobs in each period – it is clear that the pace of change was typically quite slow, and that in 2007/10 total private sector jobs actually fell. As we can see from columns (4) to (6) the comparable data for job creating firms is strikingly different – the initial stock of jobs in each period is around 6.5 million (about one third of all firm jobs, from column (7)), but by the terminal period it is around 12 million (two thirds of all firm jobs, from column (8)), so between five and six million jobs are being created each period by job creating firms (and remember job creating firms are just half of all firms in the balanced panel). The job creation series peaks at just over 6 million in 1999/02, and then falls fairly steadily, on average by around 200,000 per period, until 2007/10 when it drops more steeply, by 400,000 to 4.7 million (this is the period when the number of jobs actually fell).

14Our particular interest here is the role of different categories of job creating firms. The data on job numbers are displayed in Figure 5.20 and there is a very clear hierarchy,

• new firms are at the top, in slow decline from about 2.25 million in 2002/05 to 2.1 million in 2005/08 and then a steeper drop to 1.7 million in 2007/10
• HGFs are next, again in slow, uneven, decline from 1.5 million to 1.4 million 2004/07, then a steeper drop to 1 million in 2007/10
• the larger nonHGFs are virtually constant at around 1 million per period
• the smaller nonHGF series is more volatile but typically around 0.75 million
• finally, the young firms job creation rate is more or less constant but just 250,000 per period

15New firms are the principal contributors to job creation, but they share the responsibility for the overall decline in the rate of job creation with HGFs. The pattern of change is easier to judge if we transform the data into shares, and these are plotted on Figure 5.21. The disproportionate fall in jobs created by new firms and HGFs translates into a fall of around three percentage points in each of their shares in the last few periods, and the matching rise shows up in the two categories of nonHGFs, each of which rise by three percentage points. Interestingly, from 2005/08 onwards, the larger nonHGFs accounted for the same proportion of job creation as did HGFs.

16What proportion of job creation is contributed by high growth firms? If we assume a three year measurement period there are four plausible alternatives, following from different choices of denominator,

• all job creating firms
• all job creating firms alive in period t
• all job creating firms alive in period (t-1) (the OECD balanced panel)
• all job creating firms alive in period (t-1) with 10 or more employees in period t (10+ members of the OECD balanced panel)

17These four measures are plotted on Figure 5.22. We have seen the first, broadest, measure before (see Figure 5.21) – the HGF contribution averages around 27% from 1998/01 to 2004/07, and from 2005/08 to 2007/10 the average is 22%, five percentage points lower. The second measure, which excludes new firms, follows a similar path over time, with a 44% average in the early period and in the later period almost 10 percentage points down, at around 35%. The time path for the share of HGFs in job creation by the OECD balanced pane is very similar, essentially parallel to the second measure, and the HGF share drops from an average of 47% in the early period to 38%. Finally we have HGF job creation as a share of jobs created by 10+ members of the balanced panel, and again the share is down ten percentage points, from 60% in the years up to 2002/05 and 50% in more recent years.²

18In brief, across a range of plausible alternative denominators, the contribution of HGFs to job creation varies by a factor of two – 60% versus 30% in the early period, 50% versus 25% in the last few years. It is also clear that the contribution of HGFs to to job creation has fallen irrespective of the measure, though the extent of the fall does depend on the measure.

Chapter 4

Drawing the threads together

4.1 UK HGFs: incidence and characteristics

1we have some reasonably clear findings about HGFs in the UK over the last decade. In summary,

• About 12,500 HGFs were identified in each of the six three year periods from 2003/2007 to 2007/2010 and the incidence rate was about 7%.
• The incidence rate declines as firms age, at one year old it is about 15% and then falls at around 0.5 percentage points each year. In 2007/2010 almost half of all HGFs were more than 10 years old, and it is the 'weight' of the old firms (with their lower incidence) which contribute largely to producing an average incidence rate of 7.5%
• The incidence/age relationship is largely invariant to size, most size-bands decline at the all size average of 0.5 percentage points per year. Firms with more than 250 employees are different, their incidence rate – about 10% – is essentially independent of age. Most HGFs are relatively small – more than half have less than 20 employees, 80% have less than 50 employees
• The (2-digit) sectors which record the highest incidence rates are all services, some hi-tech, all knowledge intensive. Most HGFs are found in service sectors too, in 2007/2010 one quarter were in business services; the other (2-digit) sectors with the most HGFs (each between 5% and 10%) were construction, wholesale and retail trade and hotels and restaurants

2The incidence and distribution of HGFs in the UK by age, size and sector are largely consistent with the general pattern found elsewhere in the (relatively sparse) previous studies. Having been derived from a much more comprehensive dataset – ten successive cross-sections – our results seem rather more clear cut (and likely more robust). The one area where our results differ most markedly is with respect to sectors. We have such a large dataset and are able to work at a much finer grain and have detected, contrary to what seems to have been reported previously, that although HGFs are sectorally ubiquitous, incidence rates are detectably higher in a number of hi-tech and knowledge intensive services.

4.2 Geography

3Some potentially useful generalisations about the variation of HGF incidence across UALADs are worth re-iterating.

• the first, and most striking, finding is the extraordinarily dominant position of parts of London and the home counties at the top end of the distribution of HGF incidence rates: taken together these UALADS account for 60% of the top semi-quartile (eightth) of the distribution
• not all of London's UALADs are at the top end of the distribution, the performance of outer London to the east and south is less impressive, equally the 'halo' effect on the UALADs bordering London is weakest in these same directions
• outside London cities are important: in Scotland a major role is played by the its two principal cities – Edinburgh and Glasgow – and the places between them. In the north of England cities also stand out: Newcastle (or at least nearby North and South Tyneside); Middlesbrough; Leeds; Manchester; and Liverpool; and in Wales, Cardiff (and areas close to it); and the South West of England, Bristol and Bath all appear in the top quarter of the distribution of HGF incidence
• there is, though, one very large and important exception to the prominent role of cities at the upper end of the HGF incidence rate distribution – Birmingham. Moreover this city is one of seven UALADs in the West Midlands metropolitan county (Coventry and Wolverhampton are also amongst the seven),
• and none of these urban locations appear in the top half of the distribution. But Birmingham is not the only large 'northern' city that is 'missing' – in the North, Sheffield and Bradford are as well; as Derby and Nottingham in the Midlands; and in Wales, Swansea is a notable omission
• of course, the top end of the HGF incidence rate distribution it is not just about large cities. There is a substantial collection of towns (amongst them, Chichester, Winchester, Plymouth) at the top end of the distribution. There are also a number of relatively remote, relatively sparsely populated, largely rural places that record impressively large HGF incidence rates (for example, Richmondshire and Pembrokeshire), but these perhaps might be regarded as 'outliers'

4.3 HGFs contribution to job creation

4The proposition which originally motivated interest in HGFs is the comparison between the proportion of job creating firms and the proportion of job creation they contribute. Focusing on the broadest measure (all job creating firms) and using 2007/10 data we have,

• new (firms born between 2008 and 2010): 61.2% of job creating firms and 35.5% of job creation
• young (firms born in 2007): 5.1% of job creating firms and 5.4% of job creation
• small nonHGF (firms born before 2007 with less than 10 employees in 2007): 26.6% of job creating firms and 14.8% of job creation
• large nonHGF (firms born before 2007 with 10 or more employees in 2007): 5.9% of job creating firms and 22.5% of job creation
• HGF: 1.2% of job creating firms and 21.9% of job creation

5Clearly, of the five categories distinguished here, HGFs are relatively the most prolific category of job creating firms, however their closest comparators – the larger nonHGFs – are quite prolific too. The point is, surely, that definitions are important, and that summary statements which gloss over the detail of the definitions may seriously mislead.

4.4 The bigger picture

6In a deliberately provocative paper: "Why encouraging more people to become entrepreneurs is bad public policy", Shane argued forcefully for a shift in policy priorities: "It is about encouraging the formation of high quality, high growth companies. Policy makers should stop subsidizing the formation of the typical start-up and focus on the subset of businesses with growth potential."Shane [2009, p. 141] However, even if the negative argument is accepted (stop subsidizing start-ups), it still not at all clear what the positive argument (encouraging formation of high growth companies) entails by way of policy (see Mason and Brown [2013]). Indeed, a recent policy brief for the European Commission listed as one of its policy implications: "Since substantial evaluations of policies are apparently missing so far, it remains unclear what instruments of policies for innovative high-growth SMEs are particularly successful or unsuccessful."Lilischkis [2011, p. 94]

7Indeed it may seem somewhat ironic, especially given the stimulus that David Birch's work has given to the high growth 'agenda', that he was himself somewhat sceptical about the practical policy implications of his 'discovery' of the prolific job creation performance of HGFs,

"We know that smaller, volatile firms are the major replacers of lost jobs, but we have no experience in identifying and assisting them in large numbers. Because they are small, we must reach many of them to have a measureable effect. Because they are volatile, we must monitor each individual firm's performance carefully if we are to gain maximum benefit from our invested dollars (on the high side) and avoid scandal (on the low side). From this researcher's viewpoint it seems like a very difficult problem to solve administratively. A massive bureaucracy would be required to monitor individual small businesses on the scale required ..." Birch [1979, p. 4p]

8It has become commonplace to suggest that a researcher's answer to most questions is to call for more research, but it may nonetheless be appropriate in respect of HGFs. Whilst there is widespread acceptance of the proposition that a relatively small proportion of firms are responsible for a disproportionate share of job creation, there is not yet complete agreement, despite the efforts of the OECD and EU-ROSTAT, about how such firms might be identified. Whilst evidence about the basic characteristics of these firms is still quite scarce, the findings reported here provide a solid base for an improved understanding of HGFs in the UK.

Bibliography

• Acs, Zoltan, William Parsons, and Spencer Tracy (2008) "High-Impact Firms: Gazelles Revisited,” Research Report 328, SBA Office of Advocacy.
• Ahmad, Nadim (2006) “A Proposed Framework For business Demography Statistics,” OECD Statistics Working Papers 2006/3, OECD Publishing.
• Anyadike-Danes, Michael, Karen Bonner, Mark Hart, and Colin Mason (2009) "Measuring Business Growth: High-growth firms and their contribution to employment in the UK,” research report, NESTA.
• Birch, David L (1979) “The Job Generation Process,” research report, MIT Program on Neighborhood and Regional Change, Cambridge, MA.
• (1981) “Who Creates Jobs?” Public Interest, Vol. 65, pp. 3–14.
• (1987) Job Creation in America, New York: New York: Free Press.
• Bravo-Biosca, Albert (2011) “A look at business growth and contraction in Europe," October, presented 3rd European Conference on Corporate R & D and Innovation, CONCORD-2011, Seville (Spain).
• Davis, S J, J C Haltiwanger, and S Schuh (1996a) Job Creation and Destruction, Cambridge, MA: MIT Press.
• Davis, Steven J, John Haltiwanger, and Scott Schuh (1996b) “Small Business and Job Creation: Dissecting the Myth and Reassessing the Facts," Small Business Economics, Vol. 8, No. 4, pp. 297–315.
• EUROSTAT-OECD (2007) EUROSTAT_OECD Manual on Business Demography Statistics, Luxembourg: EUROSTAT.
• Evans, Peter and Richard Welpton (2009) “Methods explained - Business Structure Database,” Economic and Labour Market Review, Vol. 3, No. 6, pp. 71–75.
• Haltiwanger, John, Ron Jarmin, and Javier Miranda (2012) “Who creates jobs? Small vs Large vs Young,” Review of Economics and Statistics, pp. 1-52, just accepted ms, on website.
• Henrekson, Magnus and Dan Johansson (2010) “Gazelles as job creators: a survey and interpretation of the evidence,” Small Business Economics, Vol. 35, No. 2, pp. 227–244.
• Hoffmann, Anders N. and Martin Junge (2006) “Documenting Data on High-growth Firms and Entrepreneurs across 17 Countries,” working paper, FORA.
• Hölzl, Werner (2011) “Persistence, Survival and Growth: A Closer Look at 20 Years of High-Growth Firms in Austria,” WIFO Working Papers 403, WIFO.
• Lilischkis, Stefan (2011) “Policies in support of high-growth innovative SMEs,” An INNO-Grips Policy Brief 2, European Commission.
• Mardia, Kanti and P. Jupp (2000) Directional Statistics, Chichester, England: Wiley.
• Mason, Colin and Ross Brown (2013) “Creating good public policy to support high-gowth firms,” Small Business Economics, Vol. forthcoming, just accepted ms, on website.
• Murray, Gordon, Ari Hyytinen, and Markku MaulaIn (2009) “Growth Entrepreneurship and Finance," in Ministry of Employment and the Economy (Finland) eds. Evaluation of the Finnish National Innovation System Full Report, Helsinki University: Helsinki University Print.
• Neumark, David, Brandon Wall, and Junfu Zhang (2011) “Do Small Businesses Create More Jobs? New Evidence for the United States from the National Establishment Time Series,” The Review of Economics and Statistics, Vol. 93, No. 1, pp. 16–29.
• OECD (2002) “High-Growth SMEs and Employment," oecd report, OECD.
• (2008) “Measuring Entrepreneurship A Collection of Indicators 2008 Edition,” OECD-EUROSTAT Entrepreneurship Indicators Programme Publication, OECD Statistics Directorate.
• (2011) “Entrepreneurship at a Glance, 2011,” OECD-EUROSTAT Entrepreneurship Indicators Programme Publication, OECD.
• ONS (2011) “Foreign Ownership of Businesses in the United Kingdom," briefing note, Office of National Statistics.
• Schreyer, Paul (2000) “High-Growth Firms and Employment,” OECD Science, Technology and Industry Working Papers 2000/3, OECD Publishing.
• Shane, Scott (2009) “Why encouraging more people to become entrepreneurs is bad public policy,” Small Business Economics, Vol. 33, No. 2, pp. 141–149.
• Storey, David and Steven Johnson (1987) Job Generation and Labour Market Change, Basingstoke, Hants: Macmillan.
• Teruel, Mercedes and Gerrit de Wit (2011) “Determinants of high-growth firms," Scales Research Reports H201107, EIM Business and Policy Research.

Chapter 5

Tables and Figures

Note: age, years since birth

Table 5.2: distribution of HGF incidence rates by semi-quartile (SQ) across UK UALADS by region: counts

Notes:

1. rows are regions: EM, East Midlands; EN, East of England; GL, London; NE, North East; NW, North West; SC, Scotland; SE, South East; South West; WA, Wales; WM, West Midlands; YH, Yorkshire and the Humber
2. columns are semi-quartiles (SQ1 lowest incidence rate, SQ8 highest incidence rate)
3. the 'excess' column is the difference between the SQ8 column and 12.5% of the regional total, see text

Table 5.3: firms and job creating firms by category, 1998/01 – 2007/10

Notes:

1. for definitions see text
2. all firms, col (1), at terminal year of period, sum of cols (4) to (8), may differ due to rounding
3. all HGF and new are job creating firms

Table 5.4: job creation, 1998/01 – 2007/10

Figure 5.1: HGF numbers ('000) and HGF incidence (%) vs period

Figure 5.2: HGF incidence vs period by cohort, cohort97* to cohort07 (%)

Figure 5.3: HGF incidence vs age by cohort, cohort97* to cohort07 (%)

Figure 5.4: HGF incidence vs period, by size-band (%)

Figure 5.5: HGF incidence vs age, all cohorts (excl. cohort97*), by size-band (%)

Figure 5.6: HGF incidence vs period, by 2-digit sic (%)

Figure 5.7: HGF distribution by size-band vs period (%)

Figure 5.8: HGF distribution by sector vs period, top 5 sectors (%)

Figure 5.9: HGF incidence rate for UALADs, boxplots by period, 1998/01 to 2007/10

Figure 5.10: HGF incidence rate: national and median of UALADS 2002/051 to 2007/10

Figure 5.11: HGF incidence rate for UALADs, median of 2002/051 to 2007/10, ordered plot with semi-quartile boundaries

Figure 5.12: London (GL) UALADs, plotted on the Ordnance Survey National grid

Appendices

Appendix A

Foreign firms in the UK: headline numbers, the incidence of high growth and its dynamic

1This appendix explores the place of foreign firms in the UK economy and, more particularly, aims to illuminate the role of foreign high growth firms. It is organised into five sections. The first section provides some summary information on the number of foreign firms and jobs by size, country and sector. This serves as a background to the second section which reports findings on the incidence and characteristics of foreign high growth firms. Sections three and four look in more detail at a single cohort of firms – the quarter of a million private sector firms born in 1998. Using cohort data allows us to abstract from the confounding effect of age (because firm mortality rates are large and strongly age-related), and so allows us to chart more clearly the evolution of the stock of firms. The discussion starts with a demographic account of foreign firms and then un-picks the inter-related sequences of entry into foreign-ownership and the experience of high growth. The brief final section records some caveats and conjectures.

2The foreign ownership information is provided to the ONS by Dun and Bradstreet. There are two variables "immediate" and "ultimate" ownership, and we have combined these two. We give primacy to the ultimate foreign ownership variable, and then, if there is a firm which has immediate foreign ownership but the ultimate foreign ownership field is blank (i.e. not the UK) then it is also labelled as foreign, and owned by the country of immediate ownership. This pattern – immediate but not ultimate – is quite rare. In 1997 and 1998 less than one per cent, in later years less than 25 cases per year. The two ownership categories are defined in an ONS briefing note,

" The IDBR holds information on the immediate and ultimate country of ownership of an enterprise. An enterprise or group of enterprises may be owned at the enterprise or group level or by a separate holding company or legal entity. This entity may in turn be owned by a series of further companies or legal entities, the highest level of which is the ultimate owner. The immediate and ultimate owner may be located inside or outside the UK. The ultimate is the enterprise or enterprise group that is the final owner in the chain which could either be inside or outside of the UK. If there is only one owner of the enterprise or enterprise group, then this would be both the immediate and ultimate owner, because there is no further ownership in the chain. The country of ultimate and immediate foreign ownership may therefore differ ... Only the first immediate foreign owner and the highest level of foreign ownership or ultimate owner are used in this analysis.”ONS [2011, p. 2]

3Our foreign ownership variable is available for every year from 1997 to 2010 and each firm has a 'string' of foreign ownership flags associated with it. So, for example, the count of foreign firms in a particular year (and the associated jobs) include only firms which are foreign-owned in that year. Some firms change backwards and forwards between domestic and foreign owners (and some foreign-owned firms change their country of ownership). It should also be noted that not all foreign firms are flagged, so the ONS recommends that the foreign figures be regarded as minimum estimates.

A.1 Headline numbers

A.1.1 Firms

Between 1997 and 2010 the stock of domestically-owned firms rose by 25% from 1.2m to 1.5m (Table A.1), but the stock of foreign-owned firms grew much more it almost tripled from just 8,500 in 1997 to 23,600 by 2010.¹ As a result the share of foreign firms in the UK stock more than doubled from 0.71% in 1997 to 1.55% in 2010 (see Figure A.1).

A.1.2 Jobs

The number of private sector jobs in the UK grew by about 7% between 1997 and 2010, from 17.39m to 18.64m, an increase of 1.24m (see Table A.1). However, the number of jobs in domestically-owned firms actually fell by 350,000 whilst, in contrast, foreign-owned firms added 1.6m jobs over the same thirteen years – a rise of 83%. This shift in the composition of the stock of jobs began in the year 2000, and in the subsequent decade the share of jobs in foreign-owned firms doubled from 10% to about 20% in 2010 (see Figure A.1).

A.1.3 Average size

The foreign share of the stock of jobs is larger by a factor of ten than the foreign share of the stock of firms. This implies that the average number of jobs per firm in foreign firms is about ten times the UK average number of jobs per firm. If we turn this into a comparison between foreign and domestic firms, the ratio (in 2010) is about 15 (Figure A.2): an average of 150 jobs per firm in foreign firms; an average of ten in domestic firms. We can also see from Figure A.2 that the average size of foreign firms, from 1999 at least, changes very little. By contrast, the average job per firm figure for domestic firms drifts slowly down, as we know it must, since the number of domestic firms is growing and the number of jobs in domestic firms is shrinking.

A.1.4 Net job creation

'Net job creation' from the job creation and destruction accounts provides an alternative perspective on the contrasting contributions of domestic and foreign firms to the evolution of the stock of jobs (Figure A.3). Both the domestic and foreign rates displayed here have been expressed as ratios to the same, UK, employment denominator.² so the UK figure for net job creation can be computed as the sum of the domestic and foreign components displayed on the chart. Since 2001 foreign firms have made often large, and always (with the exception of 2010), positive contributions to job creation, whilst over the same decade the contribution of (the much larger) stock of domestic firms was quite small, and typically negative. In 2010 domestic firms accounted for almost three quarters of the 3 percentage point contraction in UK net job creation. It is important to recognise that, here, net job creation is equal by construction to the change in job numbers. So foreign net job creation in a period includes jobs in firms which have become foreign-owned in the period as well as jobs added by firms which were foreign-owned throughout the period. Of course it is also true, again by construction, that domestic net job creation is equal to negative foreign net job creation.

A.1.5 Firms by country

In every year (since 1997) firms from more than 50 countries had employees in the UK (in 2002 more than 100 countries). However the bulk of foreign firms and foreign jobs are accounted for by a relatively small group of countries. The distribution of firms across countries has a long thin tail with (typically) two thirds of the countries with less than ten firms. Here we focus on the top end of the distribution: the 13 countries which in 2010 had more than 500 firms in the UK. Collectively they accounted for 17,706 firms (column (2) of Table A.2), which represented 75% (column(3)) of all foreign firms in the UK. The United States is by far the most heavily represented with 5,994 firms, 25% of the total. The next three taken together – Germany, France and the Netherlands – account for a fifth. At eighth place in the list (perhaps a little surprisingly) is Jersey with a share of 3.5%. Indeed there are two other countries in the top 13, perhaps not typically thought of as major foreign investors, Luxembourg and the British Virgin Islands.

A.1.6 Jobs by country

The top 13 countries also account for the bulk of jobs in foreign firms – 2.82 million (Table A.2, column (5)), almost 80% of the foreign total (column (6)). Moreover, the ranking of countries by jobs coincides quite closely with the ranking of countries by firms (column (4)). Again the United States stands out at the top of the list with one third of all foreign jobs, and the next three – Germany, France and the Netherlands– account for one quarter between them.

A.1.7 Firms by sector

Although foreign firms are distributed across all 2-digit sectors, as with the distribution across countries, a relatively small group of industries account for a large proportion of firms. If we (again) take a 500 firm cut-off in 2010, there are 11 sectors (see Table A.3, column (1)) which collectively account for 17,300 firms, 74% of all foreign firms. At the top of the list is business services which alone contributes more than 20%, and if we add in wholesale distribution, the second sector on the list, these two account for 40% of all foreign firms. We also report (column (4)), the 'lq' ³, a measure of the similarity between the distribution of foreign firms by sector and the sectoral distribution of all UK firms. Evidently foreign firms are much more heavily represented in some sectors – notably financial services, manufacture of machinery, and wholesale distribution – than are domestic firms.

A.1.8 Jobs by sector

Since we might expect that the average number of jobs per firm could vary quite systematically by sector, it is hardly surprising that the ranking of sectors by number of jobs (Table A.3 column (5)) is a little different from the ranking for firms. Nevertheless, the 11 sectors we have identified in the firm ranking account for almost two million jobs (column (6)), more than half of all jobs in foreign firms (the sectors in the top 11 by jobs but not in the top 11 by firms have been recorded in the 'memo' item at the bottom of the table). The three sectors with the largest share of jobs are (in descending order) – retail, business services and wholesale – together contributing more than one third of all jobs in foreign firms.

A.2 Foreign high growth firms: incidence and characteristics

4The base dataset here is drawn from the 'rolling balanced panel' (see Chapter 1, data sources and methods) and is further restricted to include only firms which were foreign-owned in the year immediately preceding the three year period in which high growth 'episodes' are counted. The foreign high growth firm (HGF) incidence rate for each period is defined as the ratio of foreign HGFs to the number of foreign firms with 10 or more employees.

5Section A.3 below provides a more fine-grained treatment of foreign HGFs, examining the relationship between foreign ownership and high growth episodes for the 250,000 UK firms which were born in 1998. Here we look at data on overall numbers, and variations in the incidence rate by age and size. Unfortunately there are too few foreign HGFs to allow us to analyse their sectoral distribution since it is necessary to work at the SIC 2-digit level to draw any meaningful conclusions.

A.2.1 numbers and incidence

6From the bars on Figure A.4 we can see that the number of foreign HGFs has fluctuated quite widely over the ten three year periods 1998/01 to 2007/10, but exhibits a rising trend, at least since 2002/05. Numbers in 2007/10 are almost double those in 1998/01, they rise from around 400 to almost 700. A rather more informative guide to the relative importance of HGFs in the population of foreign firms is the incidence rate, the line on Figure A.4, which after a 'bulge' in the first few periods, seems to have varied within quite a narrow range, between 7% and 8% since 2002/05, albeit with a very slightly rising trend.

7An obvious benchmark for the HGF incidence rate for foreign firms is the comparable figure for domestically-owned firms, and the two are plotted together on Figure A.5. Whilst the two rates are rather different in the early periods – the 'bulge' in foreign is rather more pronounced than it is in domestic ⁴ – from 2002/2005 onwards the rates are less than half a percentage point apart until 2007/2010, when their paths diverge, the domestic rate falling by about a percentage point and the foreign rate rising by a similar amount.

8An alternative means of providing some additional perspective on the number of foreign high growth firms is to decompose the numerator and the denominator of the HGF incidence rate: the ratio of incidence rates can be written as the foreign share of all HGFs to the foreign share of all 10+ firms. Taking the ratio of the foreign rate (HGFrate for) to the UK rate (HGFrateuk) we can re-arrange terms as follows,

HGFrate for   HGF for / HGF uk
----------- = ----------------
HGFrate uk    firm10+ for / firm10+ uk
(A.1)

9The ratio of incidence rates and these two components are plotted on Figure A.6. Leaving aside 1998/01 which, as we have seen, looks to be something of an outlier, the ratio of incidence rates (right hand scale) fluctuates within quite narrow bounds, as we might have anticipated from Figure A.5. What this relative stability disguises is the trend-like increase (described in Section A.1) for the proportion of firms in the population of UK firms to rise. We can see that both the shares of HGFs and of 10+ firms rose steadily: between 1999/02 and 2006/09 both shares roughly doubled. The widening of the gap between the HGF rates in 2007/10 can now be seen as (proximately) caused by the foreign HGF share continuing to rise whilst the foreign share of the 10+ population did not.

10In Chapter 1 Figure 5.1, we saw that since 2002/05 the number of HGFs in each period has varied within quite a narrow range between 10 and 11,000. What we now know is that the foreign share of those numbers increased substantially as the number of foreign HGFs rose from about 450 to over 700.

A.2.2 Age

11The HGF incidence rate for the UK declines with the age of the firm: at one year old it is typically around 15%, by age nine it is lower by five percentage points: high growth is a 'game for young players'. Of course, there is no a priori reason to expect the incidence/age relationship to have the same shape for foreign-owned firms since it will be affected by the (as yet unknown) relationship between age and foreign ownership.

12Figure A.7 displays the HGF incidence rate for foreign and domestic firms plotted against the age of the firm, where the rates by age are averaged over firms born between 1998 and 2006. We can see that the foreign and domestic paths do look a little different. In particular, incidence rates for foreign firms are lower immediately after birth, but they exceed those of domestic firms by about two percentage points from age three onwards, after which both rates fall more or less in parallel. There is a more detailed investigation of the interrelationship between foreign ownership and high growth over time using cohort data in the next section.

A.2.3 Size

13The incidence rate of high growth in the UK is typically inversely related to size (see Chapter 1, Section 1.1.3) – the smaller the firm the larger incidence rate – and the relationship is generally similar for foreign firms as we can see from Figure A.8 (although the 50 to 99 and the 100 to 249 size-bands do occasionally switch rank order). The average (all age) relationship between foreign and domestic incidence rates – foreign exceeds domestic – also holds typically, size-band by size-band. Table A.4 records data on rates (averaged over the periods 2002/05 to 2007/10), and the foreign/domestic differential declines with size: it is largest for firms in the smallest, 10 to 19 size-band. Exceptionally, the HGF incidence rate for foreign firms in the largest 250+ size-band is considerably smaller than that for domestic firms in the same size-band. ⁵

A.3 The turbulent history of the UK's foreign firms born into cohort98

14The most important fact of firm demography is that firms die, most of them quite soon after birth – in the UK the 10 year survival rate is less than 20% ⁶. Whilst the survival rate does vary a little by size, the overwhelming force of mortality guarantees that firm age will be a hugely significant conditioning factor when judging firm performance. One relatively simple way to ensure that empirical conclusions are appropriately conditioned on age is to work with birth cohort data as we do here.

A.3.1 Becoming foreign

15We start with the 239,649 firms born in 1998 ⁷ – cohort98. Of these, just 327 – that is, 0.14% were born foreign. By 2010 85% of cohort98 were dead, just 33,673 firms survived, and of them 771 – 2.3% were foreign (and a further 300 of those alive had, at some stage, been foreign-owned). So whilst the number of foreign firms roughly doubled between birth and 2010, the foreign-owned share of the stock of firms rose hugely more, by a factor of 16.

16It might seem reasonable to expect that the stock of foreign firms simply grew, more or less steadily, over the life of the cohort, but that is entirely wrong. In the 12 years 1998 to 2010 a considerable number – 2,716 firms (1.13% of cohort98 at birth) – were foreign-owned for at least one year. A summary of the ownership history of these 2,716 ever foreign firms is set out in Table A.5. Of the 327 which were born foreign, just 166 remained foreign (and only 16 of the 166 survived until 2010), the other 161 became domestically-owned. However, 22 of the 161 subsequently returned to foreign ownership on more than one occasion. The 2,389 firms which were born domestic, but became foreign, display a similarly turbulent history: whilst 1,460 remained foreign after having become so, 929 did not, and almost 100 of them had more than one further period of domestic ownership.

17Clearly we cannot track the evolution of the stock of foreign firms by simply labelling a firm 'foreign' on the first occasion it becomes foreign-owned. The demographic accounts displayed in Table A.6 decompose the change into components which can inform our understanding of the process of change from birth to 2010. It displays the stock of foreign firms 'alive' in the middle row, with the constituent inflows above the 'alive' row and constituent outflows below the 'alive' row. The entries connect the 327 firms 'born' foreign (in the first row) through to the final, 2010, stock of 771.

18The accounting framework which links the years from 1999 onwards can be summarised in a pair of equations. The first starts with firms which were foreign at the end of the previous year (alivet-1) from which are subtracted the outflow to domestic (transout) and death (death) to arrive at the number which remain foreign at the end of the current year (remain). The other equation adds to those remaining the two categories of inflows from the stock of domestic firms, those becoming foreign for the first time (becoming) and those previously foreign, now coming back to foreign after a period in domestic ownership (reverting), yielding the overall end-period stock (alivet).

alivet-1 - transoutt – deatht = remaint

and,

remaint + becomingt + revertingt = alivet

19One way to understand the time-path of the foreign stock is by plotting its three components from the second (inflow) equation, and the data are displayed on Figure A.9. You will notice immediately that numbers peak in 2002 when there were more than 1,200 foreign firms (from the 'alive' row of Table A.6 we can see the exact number is 1,257), that is about four times the 1998 number and one and a half times the number in 2010. Indeed in 2002 about half of all ever foreign firms were alive and foreign. We can also see clearly that the inflow of firms 'becoming' foreign was particularly strong in the first years of the cohort's life, but after 2003 the inflow largely dried up and most foreign firms were those which simply remained foreign.

A.3.2 Understanding the evolution of the foreign share

20Using a decomposition based on the first (outflow) equation, we can represent the numbers as proportions of the foreign stock, scaling them by the numbers 'alive' at the end of the previous year. Figure A.10 displays the time series of these three proportions: remaining foreign (remaining); moving from foreign to domestic (transout); and the proportion dying (death). The most striking feature of the data is the steady, and parallel, decline in both the proportion transferring back into domestic ownership and in the proportion dying – at age 2, for example, about 40% died or reverted to domestic ownership (in roughly equal proportions), with the other 60% of foreign firms remaining foreign. But by age 12 both death and outward transfer rates are less than 10% each (death, 7.2%, transout, 5.9%) and 85% of those foreign at the beginning of the year remained foreign at the end. By 2010 the transfers within the stock of firms – two categories of inflow (becoming and reverting) and one category of outflow (transout) – more or less match, leaving mortality as the principal (proximate) influence on the size of the stock of foreign firms.

21Even though the number of foreign firms peaked in 2002 and then fell or less continuously, the share of foreign firms in the stock of all firms did not fall – as we can see from the 'alive share' row of Table A.6 – it continued to rise until 2004 when it paused, after which it continued to rise (albeit at a slower pace). To better understand the factors at work here we can use a simple decomposition of the growth in the foreign firm share. In obvious notation, we can rearrange it as follows,

forsharet / forsharet-1 = (fort ÷ fort-1) / (allt ÷ allt-1)

22In Table A.7 the period has been divided into two: 1998 to 2003; and 2004 to 2010. In the first period the foreign share increases by a factor of about 10, in the second just over 1.5 ⁸. The decomposition shows that in the first period (the 'growth' period) the foreign stock itself expanded by a factor of 3.6, whilst the overall stock shrunk by a factor of one third, producing the 11-fold increase in the foreign share. In the second period, as we know, the stock of foreign firms declined, the ratio of 0.79% shown in the decomposition table implies a rate of about 20%. Here again though the continuing contraction of the stock of firms, by one half, produces an expansion in the foreign share, here by 1.5. The key finding, then, is that foreign ownership share is driven early in the life of the cohort by acquisitions, and when that slows, the rise in the foreign share continues but is driven most importantly by the continuing (and substantial) contraction in the overall stock of firms.

A.3.3 Is foreign ownership protective?

23As we have just seen the firm mortality rate plays an important role in interpreting the data on firm performance (indeed this is the rationale for studying cohort data), so it is worth investigating whether foreign owned firms have a better chance of survival – in epidemiological terms, whether it is 'protective'. The simplest way to answer this question is to compare the death rate of foreign firms (the death proportion from Figure A.10), with the corresponding death rate for domestic firms. Our death rate is, in fact, equivalent to a conventionally defined 'hazard rate' (here the hazard of mortality) familiar from event history analysis. The two death rates are plotted on Figure A.11 and it appears that the domestic rate is larger (or about the same) at every age, but with both declining from 20% in the first year to just below 10% at year 12.

24A striking feature of the plot is the pronounced 'spike' in the hazard for domestic firms at age two which foreign firms do not share, but this datapoint, like the drop in the foreign hazard at age 4, may just represent idiosyncratic departures from the general shape of the two hazards ⁹. Certainly, after the surge in the number of domestic firms becoming foreign passed, the two hazard rates become more similar. We can see even more clearly now that although the domestic mortality rate is a huge influence on the increase in the foreign firm share of the stock of firms it was not mainly attributable to foreign ownership being protective. Rather it seems that the impact of mortality on the foreign stock was offset, initially to a great extent, later to a lesser extent, by firms transferring from foreign to domestic ownership.

A.4 Foreign firms and high growth

A.4.1 Foreign firms in the rolling balanced panel

25If we are to uncover the incidence of high growth amongst foreign firms in cohort98 we need an appropriate dataset, again it is a rolling balanced panel (RBP) but restricted to the firms born into cohort98 and it covers the nine three year periods from 1999/02 to 2007/10. At the outset it is worth checking the extent to which the RBP resembles the cohort as a whole, at least in respect of foreign ownership, and the obvious place to start is with a summary of the ownership history of foreign firms in the RBP. The bottom panel of Table A.7 provides a display of RBP data, below the cohort98 data we looked at earlier. A side-effect of the selection criteria is that there are many fewer firms in the RBP: the first period is 1999/02 (all firms must be at least one year old) and we exclude from each three year period firms which do not survive that period. So, for example, the 92,000 firms in the first period are, by construction, firms that have survived until at least 2002. Overall the RBP is about 40% of cohort98 and the number of foreign firms is smaller in much the same proportion, so the proportion of 'ever foreign' is similar in both panels of the table. Importantly, the shares of the 'dynamic' components ("remained”, "turned" etc) are virtually identical: it appears that the properties of the RBP resemble those of cohort98 – at least in respect of foreign ownership and its turbulent history.

26We can check this impression in more detail using the demographic accounts for foreign firms in the RBP which are set out in Table A.8. A casual inspection of the table suggests that most of the entries in the RBP accounts are about half the size of those of the corresponding items of the cohort98 accounts, and the dynamics look rather similar. However the timing does look a little different and these differences may be, to some extent, artefacts, by-products of the RBP selection criteria: first, for the RBP we assign foreign ownership in the year preceding the growth period; and second it is not entirely clear how to align the three year periods of the RBP with the annual cohort98 data. More specifically, the peak figure for the number of foreign firms occurs in 2003/06, after which the stock declines (though not smoothly), and here too the decline coincides with the tailing off in the number of firms 'becoming' foreign.

27Finally we can look at the evolution of the foreign share. First notice from the last row of Table A.7 that over the periods covered by the RBP, there is a ten-fold rise in the foreign share of firms 'alive': from 0.18% in 1999/2002 to 1.82% in 2007/10, roughly matching the rise in the alive share for cohort98. A decomposition of this rise is provided in the lower panel of Table A.7 and we can see a pattern which resembles the cohort98 case. The contraction of the overall stock is a major factor in both periods (a 50% fall in both), what differentiates the periods is the fall in the growth in foreign firm numbers.

A.4.2 Foreign firms and the incidence of high growth

28The data displayed in Table A.9 report the frequency distribution of HGF instances for the RBP as a whole, as well as for foreign firms. Of the 2,999 firms in the RBP which recorded an instance of high growth 5% – 155 firms – were foreign-owned at the time of their high growth episode, five times the proportion of 'ever foreign' firms in the RBP population. However, this disproportion is quite readily explicable. Firstly, as we shall very soon see, the HGF incidence rate (the ratio of HGF instances to the number of firms with 10 or more employees) for foreign and domestic firms is relatively similar. Secondly, foreign firms are, on average, considerably larger than firms overall (see Section 2.3), in particular in the RBP about 50% of foreign firms have more than 10 employees, whilst in the general firm population the figure is closer to 5%. So if the ratio of HGFs to 10+ firms is similar, but the proportion of 10+ firms in the population differs by a factor of five, the share of foreign HGFs in the all HGF total will also be larger by a factor of five.

29The table also shows the frequency distribution of HGF instances across firms. From column (3), we see that one and two instances are more common than average for firms whilst foreign-owned, whilst foreign firms are relatively under-represented (just 2%) amongst firms that record three or more instances of high growth. But as we can see from the 'memo' item at the foot of the table, foreign firms recorded 1.8 instances per firm (275 in total), around 20% less than the corresponding all firm average of 2.0 (total, 6097). As we shall see below this appearance has to be quite carefully interpreted.

30Figure A.12 displays the HGF incidence rate for domestic and foreign firms and there are some features of this plot which are already familiar,

• the two incidence rate curves are (generally) downward sloping we saw earlier that the incidence rate declines with age (see above Figure A.7 and Chapter 1, section 1.1.3)
• there is a an early 'bulge' in the foreign rate, but after 2002/05 it becomes more similar to the domestic rate, diverging slightly in 2007/10 (see Figure A.12)

Whilst foreign firms are disproportionately represented amongst HGFs in cohort98, nonetheless foreign HGFs appear to be rather less likely to record multiple instances of high growth. This may account for our finding that the incidence rate for foreign firms seems quite similar, much of the time, to the domestic rate. Investigating further requires us to dig a little below the surface and to consider in more detail the timing of movements in and out of foreign ownership and of instances of high growth.

A.4.3 Foreign ownership and the timing of high growth instances

31We start our more detailed examination of the timing of events with a sequence index plot for the 155 foreign HGF firms in the RBP: this provides a visual impression of the inter-relationships over time between ownership and high growth. The data are displayed on Figure A.13. Each firm is represented by a horizontal 'strip' allocating each of the nine periods into one of a five-fold classification of states – two ownership categories (foreign and domestic) cross-classified by two growth categories (HGF and nonHGF), plus a the fifth 'dead' category and where each state has been differently coloured. The firm-level sequences of states ('strips') have been organised so that those with an early experience of high growth whilst foreign-owned (coloured blue) appear towards the bottom, whilst those with an early experience of high growth whilst domestically-owned (coloured red) appear towards the top. In between we have the foreign (but nonHGF) at birth (coloured pink) and the domestic (but nonHGF) at birth (coloured white). The dashed horizontal lines divide the vertical axis into groups of 25 firms.

32A striking feature of the chart is the complexity of the sequences: not one of the firms remains in the same state throughout its life ¹⁰. Our intention here is to find some order in these sequences, which means finding a meaningful summary of the patterns whose only obvious common denominator is that (by virtue of the selection criterion) they experienced an instance of high growth whilst foreign-owned. As we know already from Table A.5 most 'ever foreign' firms in cohort98 are domestic earlier in life, and from the sequence plot we can see that the same is true of the foreign HGFs too – 134 firms (86%) were domestic in the first period, either nonHGFs (white) or HGFs (red) – so only 21 were born foreign, almost equal numbers of nonHGF (pink) and HGF (blue).

33Rather more surprising is that there are 26 domestic HGFs in the first period, twice the number of foreign HGFs. Indeed as we look down the plot we find that for 60 of the firms the first instance of high growth is recorded under domestic ownership (26 in the first period, the other 34 after a varying number of periods as domestic nonHGFs). In many cases the instance of high growth under foreign ownership follows immediately the domestic instance. The reverse is clearly much rarer, comparatively few instances of domestic high growth are recorded immediately subsequent to foreign high growth.

34We can also see that for firms starting with periods as a domestic nonHGF not followed by a domestic high growth instance – the 74 firms in the lower middle of the plot – it is around 50% more likely that the firm will become foreign (white to pink, 45 firms) before it becomes a foreign HGF, rather than moving directly from domestic nonHGF to foreign HGF (white to blue, 29 firms). This evolution – a move from foreign nonHGF to foreign HGF (pink to blue) – is also the path followed by virtually all of the 11 firms which were foreign nonHGFs in the first period.

35To investigate further the relationship between domestic and foreign instances of high growth in foreign high growth firms we display in Table A.10 a cross-classification of the firm-level distribution of instances: the rows record the HGF foreign instances, and the 'all' column reproduces the 'foreign' column familiar from Table A.9; and the columns record the domestic instances. We can see from the 'all' row that more than half foreign HGF firms (155 – 74 = 81) report at least one instance of high growth when the firm was domestically-owned and half report multiple instances. Moreover, a considerable number report more domestic than foreign instances, in the first row 24 firms recorded two or more instances of high growth whilst domestically-owned and just one whilst foreign-owned. If we include the domestic instances at least 81 (more than half) of these firms record more instances per firm than our original calculation included. We saw earlier that the 155 firms recorded 275 instances between them, it turns out that in total there are 125 instances recorded whilst the firms were domestically-owned, so the average per firm for firms which are ever foreign HGFs becomes 2.7 (=420/155) rather than 1.8 – one third larger than overall cohort98 average.

36The (somewhat) unexpected importance of domestic HGF instances to foreign HGFs suggests it might be worth investigating the HGF experience whilst domestically-owned of other foreign but not foreign HGF firms ¹¹. The numbers have been recorded in the 'Memo' row of Table A.10 and there are 118 firms in all. Almost half, 55 of them, foreign at some stage, recorded just a single instance of high growth, and did so whilst they were domestically-owned; similarly, 63 'ever' foreign firms recorded two or more instances of high growth but did so whilst they were domestically-owned. In summary, 118 firms which were at some stage foreign-owned recorded instances of high growth whilst they were not foreign owned. Evidently many high growth firms become foreign owned only after their experience of high growth is behind them.

A.5 Caveats, conjectures and future work

37This is the first study to have added foreign ownership markers to the firm-level records from the BSD, so the findings should be treated cautiously. One particular (and slightly worrying) feature of the data is that there are some indications that the coverage of the ownership marker might be less complete in the first two years (1997 and 1998) than it is later on. For example, the 1997 and 1998 datapoints on Figure A.1 for the foreign share of the stock of firms, and on Figure A.2 for the average number of jobs per firm for foreign firms, look a little out of line. Conversely, though, there appears to be no reason to suspect any discontinuous variation in coverage from 1999 onwards.

38Another feature of the data which might be regarded as a little unexpected is the extent of the turbulence in the stock of foreign firms. In particular, the scale of the movement out of foreign ownership and then back into it is a characteristic which does not previously seemed to have been reported. But there is no reason to suppose, other than its novelty, that this is an artefact or a by-product of measurement error. Needless to say, it might worth taking a sample of cases from an alternative, independent, dataset like FAME (even though it is much less comprehensive) to investigate this phenomenon.

39A larger issue worthy of further investigation is the dynamics of the relationship between foreign ownership and instances of high growth. First, we know that the incidence of high growth declines with age. Second, it seems plausible that the chance of a firm becoming foreign-owned may increase with age – in epidemiological terms, the older a firm the more years it is 'at risk' of foreign-ownership – as we found for cohort98. Taken together these two propositions imply (as we found) that some of the (relatively numerous) firms with an early high growth experience will become foreign-owned. What we have not yet determined is whether early high growth plays a causal role in this context. More work is required to first of all ensure that the cohort98 findings do generalise and are not cohort-specific or period-specific, and then, with some more broadly based results it would be worth attempting to untangle the sequence of events more definitively.

Table A.1: Firms and jobs: 1997 & 2010

---

Footnotes

Table A.2: Foreign firms and jobs by country, 2010 (more than 500 firms)

Table A.3: Foreign firms and jobs by 2-digit sector, 2010 (more than 500 firms)

Memo:

Note: 'lq' in columns (4) and (8) are sector shares for foreign firms divided by the corresponding sector shares for all firms.

Table A.4: HGF incidence rate by size-band, average 2002/05 to 2007/10, foreign and domestic, %

Table A.5: cohort98, summary of firm ownership histories

Table A.6: cohort98 firms, foreign, demographic accounts

Notes:

1. see text for definitions
2. denominator for shares is all firms (foreign plus domestic) alive

Table A.7: cohort98 and cohort98 RBP firms, foreign share of alive, decom-position

(a) cohort98

(b) RBP

Table A.8: cohort98 RBP firms, foreign, demographic accounts

Notes:

1. see text for definitions
2. * count less than 10, below disclosure threshold, combined with 'remaining'
3. denominator for shares is all firms (foreign plus domestic) alive

Table A.9: cohort98 firms: HGF, all and ever foreign instance, frequency of HGF instances per firm, 19998/02 to 2007/10

Memo:

instances: all, 6097; foreign, 275 average per firm: all, 2.0; foreign, 1.8

Table A.10: cohort98 firms, ever foreign HGF: frequency of foreign vs domestic instances per firm, 1999/02 to 2007/10

Memo:

Note: foreign HGFs instances: 275 whilst foreign-owned; 145 whilst domestically-owned; total 420

Figure A.7: HGF: foreign and domestic, incidence rate by age, average over cohorts, 1998 to 2006, %

Figure A.8: HGF: foreign by size-band by period, 1998/01 to 2007/10, %

Figure A.9: cohort98 firms, foreign alive, 1998-2010, by inflow component, number

Notes:

1. in 1998 "becoming” is firms born foreign
2. for data see Table A.6

Figure A.10: cohort98 firms, foreign alive, 1998-2010, by outflow component, ratio to alive$_{t-1}$

Note: for data see Table A.6

Figure A.11: cohort98, deaths, foreign and domestic, years since birth, calculated hazard rates

Notes:

1. ratio of death to alive$_{t-1}$
2. for data see Table A.6

Figure A.12: cohort98 firms, HGF incidence rate, domestic and foreign, %, 1999/02 to 2007/10

Note: foreign-owned HGF, foreign in the year before the high growth episode

Figure A.13: cohort98 firms, ever foreign HGF, sequence index plot

Key: red, domestic HGF; blue, foreign HGF; white, domestic nonHGF; pink, foreign nonHGF; black, dead