Performance, Firm Size and the Heterogeneity of Competitive Strategy for Longlived Small Firms: A Simultaneous Equations Analysis
Bernadette Power ^{+} and Gavin C Reid∗
Abstract
This paper examines the relationship between firm size, competitive strategy and performance, for the longlived small firm in Scotland. It uses structural modelling to test the hypothesis that small firms need to remain small if they are to be longlived. In a three equation simultaneous model, performance, size and the dimensions of the competitive strategy of the firm are jointly determined. Econometric estimates of the three equations are reported, using 2SLS and iterated 3SLS. A tradeoff is found to exist between firm size and performance. Further, we find that to attain higher equilibrium values of performance, a varied competitive strategy needs to be adopted. Our prescription is that small firms need to adjust downwards in size, and to cultivate a more varied competitive strategy, if there the entrepreneurs are to have a positive influence on performance, thus promoting longevity of their firms.
Keywords: 
Performance, Small Firms, Size, Competitive Strategy, Simultaneity 
JEL: 
C42, D21, G33, L2, M13, M21 
* Author for Correspondence: Professor of Economics, and Director, Center for Research into Industry, Enterprise, Finance and the Firm (CRIEFF), Department of Economics, University of St. Andrews, St. Salvator’s College, St. Andrews, Fife, Scotland, KY16 9AL, UK
email:
Phone/Fax: (+44) (0) 1334 462431 (personal); (+44) (0) 1334 462438 (CRIEFF) www: http://www.stand.ac.uk/~www_crieff/CRIEFF.html
gcr@standrews.ac.uk
^{+} Lecture in Economics, Dept. of Economics, University College Cork, Cork, Ireland
email:
Phone: (+353) (0) 21 4902986
Fax:
www: http://www.ucc.ie/ucc/depts/economics/staff/power.html
b.power@ucc.ie
(+353) (0) 21 4273920
1: INTRODUCTION The paper explains the tendency of small firms to remain small. It is a micro econometric analysis of the relationship between size, competitive strategy and performance for longlived small firms. We use a simultaneous equations model to estimate three endogenous relationships: performance, size, and heterogeneity of competitive strategy. Our evidence suggests that a tradeoff exists between the firm’s size and performance. Increases (decreases) in firm size reduce (raise) the performance of the firm, and it is this which tends to limit the size of the firm. This tradeoff encourages ownermanagers of small firms to reduce ‘headcount’ to achieve greater efficiency. Such efficiency gains arise from greater labour productivity, often with an increase in human capital of the ‘core’ workforce. The substitution of capital (including human capital) for standard labour inputs is another potential source of efficiency gain. To achieve higher levels of performance, the heterogeneity of the competitive strategy of the small firm was also found to be important. Based on this evidence, our prescription is that small firms need to cultivate more varied competitive strategies, in niche or localised markets, to improve their longrun performance prospects. The study is fieldwork based and uses evidence from facetoface interviews with ownermanagers of mature small firms in Scotland. Performance is measured by a likert scale over 28 distinct attributes. The latter incorporated aspects of: competitive environment; financial management; organisational structure; and business strategy. Size is measured by fulltime equivalent employees, and the heterogeneity of competitive strategy by a count variable of strategies pursued by the small firm in their principal market. Section 3 provides a detailed account of how we measured these variables. Our econometric estimates used 2SLS and I3SLS. The three equations were estimated using data collected on 63 longlived small firms. We define longlived small firms as businesses that have been trading for more than 10 years. They were classified as small firms at inception if they employed less than 100 people. In fact, the small firms in this study were often much smaller, typically having 10 employees at inception. Today, our longlived small firms had 13 employees, on average, indicating some, but not substantial growth since inception. Some small firms enjoy high performance, and growth. Storey (1994) describes them as ‘ten per centers’ because they are few, and Birch (1996) describes them as ‘gazelles’ because of their apparently effortless higher performance. However there is a tendency for most small firms to remain small. In becoming longlived, small firms in our sample have passed the
2
long run test of economic survival. In surviving the market selection process, they are not necessarily laggards, unless the markets in which they were operating were uncontested. Indeed, we would expect that only the fittest firms survive in a competitive market, facing a market selection test every period (Gould, 1993). However, given that the average age of the longlived small firm in the sample (26 years, over a full generation) it is difficult to imagine that inefficiencies in the market could have allowed these firms to clear the first hurdle of survival for this length of time. This suggests that there are other reasons preventing these firms from clearing the second hurdle of growth in firm size. Evidence of a negative relationship between growth and profitability was found by Cubbin and Leech (1986) and Dobson and Gerrard (1989). Reid (1993, 1995) confirmed this for a sample of small business startups within a simultaneous framework. This evidence suggests that there are diminishing returns to increasing the size of the firm and that a negative relationship is expected between firm size and performance. This study seeks to embed this tradeoff in a larger model; and to extend the analysis to incorporate: (a) the heterogeneity of the firm’s competitive strategy; and (b) performance over the long term. Thus it is recognised that other inherent attributes of the firm could explain the tendency of the small firm to remain small. Candidates for this include: the size of the market for its product (e.g. local service); the firm’s organisational capability; the level of differentiation of the product (e.g. extent of customisation); and potential risks to the income of the owner manager (e.g. as a result of cashflow difficulties, overinvestment). Briefly the development of our ideas is as follows. Section 2 examines the simultaneous equations model to be estimated. Section 3 discusses the primary source data on which this study is based, and describes the variables used in estimation. Section 4 reports on Durbin WuHausman type tests of endogeneity, employed to examine whether simultaneities exist between firm size, the heterogeneity of the firm’s competitive strategy and small firm performance. In this Section we also discuss appropriate techniques for system estimation. Our position is that it is often hard to disentangle one relationship from the other using single equation estimators as they are often dogged by "lack of identification" which may not even have been investigated. Section 5 reports the results of two appropriate system estimation techniques (two stage least squares, 2SLS, and iterated three stage least squares, 3ISLS), which are known to be relatively robust in the face of specification error. Using these techniques we report on estimates of the behavioural relations between firm size, competitive strategy and performance. This Section also examines behavioural patterns in the size
3
adjustment, heterogeneity of competitive strategy, and performance within the context of the simultaneous system. Finally, Section 6 summarises our principal results.
2: THE MODEL
We adopt a type of simultaneous equations model with industrial economics applications. Other recent examples include Jans and Rosenbaum (1997), Beccarello (1996), and McDonald and Bloch (1999). Jans and Rosenbaum (1996) use nonlinear three stage least squares to estimate quantity and price equations. Beccarello (1996) used three stage least squares to estimate markup equations. McDonald and Bloch (1999) use instrumental variables to estimate spillover effects of industrial growth/performance tradeoff model. We use 2SLS and I3SLS to estimate a threeequation growth/performance tradeoff model. This section explains how our key hypotheses are addressed. The central hypothesis examines behavioural relations between size, competitive strategy and the performance of the longlived small firm. A number of papers examine the relation between growth and profitability or size and growth. However, none explicitly examine a model where firm size, the heterogeneity of the competitive strategy and performance are jointly determined. Early work on this relation by Penrose (1959) examined the influence of external competitive pressure on the growth/profitability tradeoff. However this was not examined empirically. Penrose (1959) stated that external constraints to growth arise from a combination of increasing market saturation and more intensive competitive pressure. She stated that as a result of heightened competitive pressure higher growth can only be achieved through higher advertising expenditure and/or lower prices. This will result in a negative relationship between growth and profitability (or performance). In this case, the financial cost of these types of competitive strategies is a ‘hidden cost’ to growth. Our analysis differs in that rather than measuring competitive pressure we examine the heterogeneity of the competitive strategy of the firm. There are a number of competing models of the performance of the firm. Variants of the structureconduct and performance paradigm in industrial economics are examined by Reid (1987). According to this reasoning industries have structural characteristics (e.g. entry barriers, product differentiation), which suppress rivalry and raise the profitability of incumbent firms (Bain, 1956). More recently, the market efficiency school proposed that the firm’s stock of resources enable it to extract above normal profits as oppose to structural barriers or market power (Demsetz, 1973; Lippman and Rumelt, 1982). These works spawned a number of other studies examining the persistence of industry versus firm effects
4
(Schmalensee, 1985; Dierickx and Cool, 1989; Rumelt, 1991). Much of this effort focussed more on the performance of large firms rather than on small firm performance. Earlier work on small firms concentrated on firm size, particularly on the relationship between growth and firm size originating in the Law of Proportionate Effect (or Gibrat’s Law). This states that growth rates are independent of firm size and it’s past growth history. All firms grow at the same rate over an interval of time, regardless of their initial size. Conflicting evidence exists on the relationship between growth and size. A number of empirical studies suggest a negative relationship between growth and size, indicating that smaller firms have higher and more variable growth rates which reduce their survival rate (Mansfield, 1962; Hall 1987, Mata, 1994) while other studies (Singh and Whittington, 1975) have found a positive relationship. These models are mechanistic methods of explaining the relationship between firm size and growth. Managerial theories of the firm explicitly examined tradeoffs between the growth of the firm and performance (Penrose, 1959; Marris, 1964; Richardson, 1964; and Slater 1980). Hidden costs to growth are identified in the literature by Penrose (1959), where she describes the nature of the managerial limit. This has become known as the ‘Penrose effect’ (e.g. new managers are drawn in, who require training to be integrated into the existing framework of the firm). If the Penrose effect is indeed a source of dynamic scale economies, one would expect the underlying constraints on growth to induce a negative relationship between growth and profitability. Slater (1980) captured some key features of her approach in a formal model, which shows that the rapid recruitment of management, which accompanies faster growth, leads to an increase in marginal cost. Richardson (1964) also discussed this functional relationship between the organisational efficiency of a firm and its rate of growth. He said that the former would decline after a point as the latter rises (Richardson 1964:11), that is, there is ultimately a growth efficiency tradeoff. Evidence of a negative relationship between growth and profitability was found by Cubbin and Leech (1986), and Dobson and Gerrard (1989), and this was confirmed by Reid (1993, 1995) for a sample of small business startups, within a simultaneous framework. The latter work examined the relations between growth and performance, rather than that of size and performance, the novel concern of this paper. The second innovation of this paper is to test explicitly for simultaneities between competitive strategy, firm size and performance. We report on the results of the size/performance tradeoff, using a new behavioural relation, which is expanded to account for the intensity of the firms competitive strategy. In general terms, our threeequation model is specified as follows:
5
P = f (S, C, X _{1} )
(1)
S= g (P, X _{2} ) 
(2) 
C= h (P, X _{3} ) 
(3) 
Here, P is a measure of performance, S is a measure of size, C is an index of the heterogeneity of the firm’s competitive strategy and the X _{i} represent exogenous variables in the system (discussed explicitly in Section 3). In equation (1) size is expected to affect performance negatively (assuming a tradeoff exists between increasing firm size and performance), whereas C is expected to influence performance positively. The latter arises from Porter’s (1985) ideas on ‘good competitors’: these are competitors who, by engaging in sharp and challenging rivalry, actually promote the efficiency and innovativeness of incumbent firms, and hence improve their prospects of staying in business. The overall effect on performance depends on the relative sizes of these effects. X _{1} here represents attributes of the financial structure of the firm, and aspects of its market and age. Equation (2) represents size as a function of performance, and other exogenous variables. A negative relationship is expected here between size and performance. X _{2} incorporates lagged performance variables and other variables like the resources of the firm (hidden costs to increasing firm size). Equation (3) represents competitive scope as a function of performance and other exogenous variables. The sign of the performance effect upon competitive scope is unknown. X _{3} incorporates market structural variables to approximate the extent of external competitive pressure in the firm’s principal market. Essentially this threeequation model allows us to examine: (a) whether a tradeoff exists between the size of the firm and its performance; and (b) the influence which the heterogeneity of the firm’s competitive strategy has on this tradeoff. In general, it is expected that the greater the diversity of the firm’s competitive strategy, the higher the firm’s performance. To survive, the mature small firm becomes leaner, more efficient and provides a more customised service. The consequence of this is that there is tendency for the firm to remain small, using differentiated strategies to target localised or niche markets.
3. DATA AND VARIABLES
This section presents information on the database and the variables used in econometric estimation. It also provides summary statistics on the key variables used in equations (1), (2)
6
and (3) above (see Table III below). An account is provided of how these variables are defined, and an explanation is given of how these variables were measured in the survey instrument. We also provide some amplification of the specification of each of the structural equations, considered as part of the overall simultaneous system.
3.1 Database Briefly, the data set that we used for econometric estimation was based on interview evidence obtained from 63 ownermanagers of longlived small firms in Scotland. They were obtained from a sample frame of 90 mature small firms (a response rate of 70%). The latter were extracted from three “parent” samples of Scottish small business enterprises. These will be described as the Leverhulme (19851988), Telephone Survey 1991 and Leverhulme (19941979) ‘parent samples’, for exposition purposes (see Table II). The owner managers of businesses in these three parent samples were interviewed by one of the authors in the 1980s and 1990s. The activity fieldwork behind the Leverhulme (19851988) sample involved gathering data by facetoface interviews with the owner managers of 86 new business starts in the late 1980’s. Of these 86 firms, 25 (29%) survived and 20 of these agreed to be reinterviewed for this study. ^{i} Data on the second sample frame of 160 mature firms, were attained from the list of members of the Federation of Small Business (FSB) in Scotland. These data were collected by structured interviews over the telephone in 1991. At that time, 113 firms agreed to be interviewed. Fifty, of the original 113 firms from this parent sample were still in business in 2001 (a survival rate of 44%) ^{i}^{i} . Thirty of these firms agreed to be reinterviewed. From the Leverhulme (19941997) sample, our third parent sample, this time of 150 firms, we found that just 20 were longlived small firms aged 10 years or more. In the last case this original sample was intended to be of new business starts. These were interviewed originally using facetoface interviews from 1994–1997. Fifteen out of twenty firms aged 10 or more were still trading (a survival rate of 75%) ^{i}^{i}^{i} . Thirteen of these agreed to be reinterviewed.
[INSERT TABLE I HERE] The three parent samples are known to be fairly representative of the relevant populations of small firms in Scotland at the time of selection. They provided a secure set of known sources upon which further fieldwork could be built. Considerable benefit was derived from previous contact with entrepreneurs, in terms of access to the field. Generally, ownermanagers were happy to be looked up again, after a long lapse of time.
7
In total we gathered data on 63 longlived small firms, by facetoface interviews, between October 2001 and February 2002. The administered questionnaire we created for this purpose examined the following characteristics of the longlived small firm: changes in its scale and scope; pivotal changes in its running since startup; factors which fostered its survival; and its level of innovation and technical change. General statistical features of the database were as follows.
3.2 General Characteristics of Longlived Small Firms The firms examined were mature (25 ½ years on average; median age of 22). Almost all sectors, by SIC code, were represented in the sample, running through from agriculture (01) to domestic services (99). The main sectors, by SIC codes were: 32, mechanical engineering (4.8%); 43, textile industry (4.8%); 61, wholesale distribution (4.8%); 64, retail distribution (23%); 66, hotels and catering (4.8%), 67 repair of consumer goods and vehicles (6.3%); and 83 business services (9.5%). Thus, the modal firm was a retailer. The sample proportions between extractive/manufacturers (SIC 0160) and services (SIC 6199) were 40% and 60% respectively. These proportions were similar across the extracted and interviewed ‘parent’ samples, although they differed across the three parent samples (see Table II).
[INSERT TABLE II NEAR HERE] Of the 219 firms in the three parent samples 84 (38%) were in manufacturing (SIC 0160) and 135 (62%) were in services (SIC 6199). Figures from the Department of Trade and Industry, for the UK as a whole, indicate that 27% were in extractive/manufacturing and 73% were in services. Thus there is a slight bias towards extractive/manufacturing firms in our sample. This is partly explained by the slower progression of the Scottish economy to becoming service based, compared to the UK as a whole. It is also probably partly explained by the fact that part of the parent samples were drawn from the caseloads of Enterprise Trusts, which, at the time of early enterprise policies tended to favour manufacturing enterprise. This proactive behaviour, indeed positive discrimination towards manufacturers, disappeared from enterprise policy, as fears of deindustrialisation abated and, were replaced by a new enthusiasm for knowledge based enterprise. The following regions were represented: Aberdeen, Argyll, Aryshire, Banff, Caithneas, Cumnock, Dundee, Fife, Glasgow, Inverness, Isle of Skye, Lanarkshire, Lothian and Edinburgh, Midlothian, Moray, Orkney, Perth, Renfrewshire, Ross and Stirling. These represent well the locational diversity of longlived small firms in Scotland.
8
Of the sample of 63 longlived small firms, one (1.6%) was a sole trader operating from home, fifteen (23.8%) were sole traders operating from business premises, nineteen (30.2%) were partnerships and twentyfive (44.4%) were private limited companies. This pattern is very different from what you would expect early in the lifecycle. Then, sole proprietors would dominate. At the other end of the life cycle, private companies predominate. Eighteen (28.6%) firms changed their legal form during the life of the business. There is general evidence of changes in organisational form, from the sole proprietorship form, to the partnership and private limited company forms, over the lifetimes of the firms, cf. Reid (1998). The number of fulltime equivalent (FTEs) employees, which is one indicator of the size of these small business enterprises, varied from 1 to 130 with the average and mode being 13.55 and 6 respectively. The average size of firms (and the corresponding standard deviation) in terms of full time equivalent employees were as follows: 5.94 (5.85), sole proprietorship; 7.91(4.08), partnership; and 22.19 (27.69), private company. Size, measured by turnover for the last trading year, also varied widely by business type. Average turnover (and its standard deviation) was: £219,813 (£143,025) for sole proprietorships; £557,526 (£455,994) for partnerships; and £1,372,821 (£1,885,391) for private companies (all figures in 2001 prices). Thus changes from sole proprietorship, to partnership, to private company, are generally associated with increases in size.
3.3 Variables This subsection is concerned with the key variables used in the system estimation. We provide a detailed explanation of how they are defined, and of how the questionnaire design was used to generate these variables. The endogenous variables in the system are examined initially, namely, performance, P (measured in three alternative ways), firm size, S, and the scope of the firm’s competitive strategy, C. Exogenous variables within the system are then examined for each structural equation in the system. These exogenous variables can broadly be grouped into market and strategy variables. The variables are then used in the estimation reported upon in Section 5. Table III lists these key variables, and their summary statistics. [INSERT TABLE III NEAR HERE]
3.3.1 Performance Several approaches to measuring performance in small firms are possible. For example Smith (1997) and Reid and Smith (2000) identify three. In particular, they contrast an objective measure (e.g. quantitative measures like profitability and rate of return) with a subjective measure (e.g. a judgmental evaluation of performance, drawing on both
9
quantitative and qualitative evidence). In this paper we adopt the latter approach. It is both more comprehensive, and more compatible with our evidence base. The requirement for a comprehensive measure of performance is consistent with the literature on entrepreneurship and management accounting as applied to the small firm e.g. Wickham, (2001, Ch. 20). Essentially, it recognises that the proper control of the firm requires a comparison of current performance to a predetermined plan or objective. This is the basis for the so called ‘variance analysis’ in management accounting. This approach would see there being an indissoluble link between the setting of performance standards, and the control of the firm by the owner manager. The most commonly conceived performance standards relate to budgets. However there are many other forms, including those relating to human factors, like responsibility, and to technological ones, like hitting research milestones. As regards the compatibility of the evidence base, our own subjective measure of performance evaluation allows us to undertake modelling which would otherwise be impossible given our complex sample construction. The problem is that each parent sample typically offers distinct objective performance measures gathered at different points in time. There is an intrinsic lack of comparability of these measures across our sample. Using a new performance measurement approach breaks this impasse. Our measure is common to the three parent samples, which allows us to proceed with empirical work on a uniform basis. Our quantitative indicator of performance was multidimensional, involving 28 items, each of which was calibrated on a 100point scale. We argue that naturally there are many dimensions to performance. Our indicator examined these dimensions of performance under main headings like strategic (9 items), financial (4 items), and organisational (4 items) and environmental forces (11 items). ^{i}^{v} We hold that our approach has advantages over the use of conventional financial data. These are limited by accounting conventions (e.g. the reporting protocol). Further, lifecycle effects may make them difficult to interpret in sensible economic terms. For example up to three years of losses may be assumed early in the life cycle. Further, accounting profit is not readily related to economic profit. Thus rate of return, or profitability, which may both seem suitable quantitative indicators for assessing the performance of the mature small firm, may fail to grapple with quite simple aspects of reality. For example, profit itself may be illdefined in many small firms, as owner managers do not always make a clear distinction between profit and income. We could, of course, have substituted a simple, single question on selfappraisal of performance, for the more conventional type of question on rate of return. However, we would argue that our multidimensional approach has two main advantages over the single question approach.
10
First, it produces detailed measurement across wide spectrum performancerelevant variables, rather than a single variable. Second, by diluting variable specific effects, it produces a more comprehensive (and stable) measure of what is meant by performance, allowing common influences to come through (DeVellis, 1991). The key performance question put to ownermanagers was as follows: “We’d like to know what has kept you in business down the years. Some things are good for business and some things are bad. What effect have the following had?”. The ownermanagers were asked to rate ^{v} each of the 28 items on a scale of 0 to 100 where 100 is good, and 0 is bad and 50 is neutral. They did so by placing a cross on a line, of length 100 units. In this way, we calibrated the influence they judged this item to have had, based on their actual experience of running the business. If an item was not applicable they were asked to say so. An extract of the scale is reproduced in Figure I.
[INSERT FIGURE I HERE]
We found that ownermanagers of our longlived small firms were readily able to draw on their experience of running their businesses, in selfappraising the influence that each of these items had on their performance. In doing so, the ownermanagers had in their minds a large body of qualitative and quantitative evidence, on which they could base their judgements of performance. To illustrate, over time they had learned how best to combine their factors of production to exploit market opportunities; and they had learned how to respond to threats in a way that improved their performance, and enhanced their survival. Given that ownermanagers comfortably juggle these various performance measures in their own minds, we consider it logical to seek explicit measures of how this juggling act is sustained. Thus our measuring exercise provides us with a new form of empirical evidence, based on judgements, which nevertheless is useful in econometric estimation. From the self assessment of each item’s influence on the performance of the firm, we obtained a measure of overall performance, by summing the individual item scores. Thus an overall score for performance (Perform) was calculated for each firm, based on the summation of ratings for factors, normalised to take account of those items that were not applicable. This measure is not age related, as each dimension may assume a greater or lesser importance at any point in the lifecycle. It is envisaged that changes in this judgement of firm performance will lead the firm to modify factors like its size and its management processes to enhance performance. The
11
consequences of this will in turn modify performance judgements. Thus this variable is expected to be endogenous within our simultaneous equations framework.
3.3.2 Firm Size
The second endogenous variable in this framework is firm size, S. Here, it is approximated by the number of fulltime equivalent employees. As with performance, there are several ways approaches to measure firm size. Our analysis was repeated using assets and sales as size measures, and similar results were found to those reported in Section 5. As measured in 2001, these mature small firms were, on average, just above the microfirm upper range of 10 employees. They generated £835,000stg in turnover and had using assets valued at £330,000stg. The predominant firm type was still the microfirm, and the average size was somewhat raised by the existence of a few large firms in the sample. Essentially, the size distribution observed for these small firms is something like a Pareto distribution: that is, one branch of an hyperbola in the first quadrant, with unity as the lower bound. When investigating the early life of the small firm, in a single equation model, Reid (1993) found that size measures like assets and number of full time employees had less clear consequences for survival than the employment measure. Here, Reid’s (1993) approach is extended to examine the interrelationship between size, the heterogeneity of the firm’s competitive strategy and long run performance, all in a larger simultaneous equations system. In this way, we hope to gain a better understanding of interrelationships between firm size and performance, and the tendency for new business startups to remain small.
3.3.3 Competitive Strategy Space
The third endogenous variable in our model measured the size of the competitive strategy space of the firm, C. This variable is calibrated by a count of the number of forms of competition used by the firm This variable may range from 1 to 8 where ‘1’ indicates that the small firm competes on just one dimension of the competitive strategy space (e.g. price alone), and ‘8’ indicates that it competes across many dimensions (e.g. price, quality, delivery). To measure this C variable, the owner managers were asked: What form of competition is used in your principal market? Options included price, quality, volume, after sales service, new product development, advertising, tying up suppliers, delivery and marketing. On average our small firms competed on 4.5 dimensions. Over three quarters of them competed on price, 87% on quality, 58% after sales service and 63.5% on delivery. It was less common to compete on advertising (28.6%), tying up suppliers (25.4%) and volume
12
(23.8%).
compete. We would expect a firm, which actively competes in this way to improve its
Essentially, higher values of this variable C represent a greater willingness to
performance.
3.3.4 Exogenous Variables The model has thirteen variables, three of which are endogenous, performance (measured in three alternative ways), size and what we call competitive strategy space. There are 8 exogenous variables in the model, to improve specification, goodness of fit, and to help identify the system, in a statistical sense. They will be examined in turn.
3.3.4.1 Equation (1) P = f (S, C, X _{1} ) The exogenous variable (X _{1} ) in equation (1), the performance equation, represents
attributes of the financial structure of the firm, and aspects of its market and age. Liabilities, D, are approximated by a count variable of the number of forms of debt that the firm had incurred. This variable ranges from 0 to 4, where ‘0’ indicates that the firm has no debt, and
is financed by personal equity and retained earnings with an implied gearing ratio of zero. For
a higher level of external liabilities, a lower level of performance is expected. Over two
thirds (68.3%) of the mature small firms interviewed were being financed through some form of debt. On average, these firms had taken on two forms of debt. Nearly a half (44.4%) had a bank overdraft; over onethird (39.7%) had a bank loan, nearly a quarter (22.2%) used hire purchase and about an eighth (12.7%) had used other forms of debt finance, including, notably, leasing agreements. These were the main source of finance capital, apart from private equity. The latter was uncommon, and only three firms out of the sixtythree interviewed had any outside equity finance. The level of technical change in the industry, T, is a dummy variable which takes on the value of one when the industry in which the firm operated experienced technical change over the life of the firm and zero otherwise. Fiftytwo (82.5%) of the firms interviewed had experienced technical change in their industry over the life of their firm. Industrywide technical change has a potentially important effect on the long run survival and performance of the small firm. Age, as a variable, has an important interpretation. The success of the small firm may be attributed to the experience of the owner manager. This experience is accumulated by the practice of running the firm on a daytoday basis. The ownermanager may initially be ignorant of his ability, but, over time his ability is revealed, at the same time as his skill is acquired, Jovanovic (1982). This introduces time dependence into the
13
performance growth relationship: the age of the mature small firm is also a determinant of its growth rate as well as its size. The average age is about 26 years, (roughly one generation) and no firm was younger than 10 years old. The maximum age in the sample was 90 years (over three generations). M, is a categorical variable, which identifies the main market for the principal product group. It ranges from 1 to 4, where 1 denotes more local markets and 4 denotes national or international markets. Typically, our mature small firm operated in localised markets. Nearly a half (46%) operated in local markets, few (3.2%) in just regional markets, over a quarter (28.6%) in Scottish markets, about one fifth (19%) in the UK and few (3.2%) internationally. More than 50% operated outside local markets, with typically selling to the Scottish market. Over a quarter (47.6%) stated that their main market had changed since startup. Typically firms expanded their market extent. Initially, other variables were considered in the performance equation including institutional variables like industrial sector and legal status, and structural variables like market share, number of major rivals, and degree of competition in the principal market. These were dropped if they were insignificant.
3.3.4.2 Equation (2) S= g (P, X _{2} )
The exogenous variables (X _{2} ) in equation (2), the size equation, represent attributes like: the level of technical change in the industry, T; the organising capability of the firm, OC; and the labour productivity at startup, LP _{S}_{t} . Labour productivity early in the life of the firm (LP _{S}_{t} ), is a ratio of sales to employees in the first interview at constant 2001 prices. It is predetermined and assumed to be exogenous. It is a measure of the operational efficiency of the firm earlier in its lifecycle. Firms, which generate more sales per fulltime equivalent employees, are assumed to be more operationally efficient. Greater operationally efficiency, early in the lifecycle of the firm, indicates a superior performer at this stage. Superior performance would often be expected to lead to growth in size. This may not be the case, however, as firms grow faster earlier in their lifecycle, compared to later. On average, firms generated sales per fulltime equivalent employees of £113,489stg. at this stage in their life in 2001 constant prices. Our measure of the organising capability of the mature small firm (OC) is a count variable of its functional activities (e.g. production, accounting, it support, sales, marketing, product innovation, strategic planning etc.). According to Ghoshal, Hahn, Moran (2000), administrative reorganisation is often necessary if the firm is to grow over time, and to retain
14
its market share in a dynamic environment. Reorganisation often involves rebundling activities, and may also lead to internalising the various stages of the value chains of goods and services. To undertake such activities, a firm requires organising capability. Our variable increases with the level of organising capability. On average, our firms engaged in seven functional activities. Typically the larger number of these functional activities, the greater is division of labour required, and the larger is the expected firm size.
3.3.4.3 Equation (3) C= h (P, X _{3} ) The exogenous variables (X _{3} ) in equation (3), our competitive strategy equation, represent attributes of the firm’s principal market. The number of major rivals, Mriv, is an index of competitive pressure. In general, the greater the number of major rivals, the greater the competitive pressure. Even with an average of just 26 major rivals this pressure can be intense. Indeed, at extremes it can be destructive (Reid, Jacobsen & Andersen, 1993). Product differentiation, Diff, is a selfappraised measure of product variety. It represents a directly measured alternative to tricky measures like the cross elasticity of demand, which require evidence from estimated demand functions. It was scaled to be greater, the greater the product heterogeneity. In this sample the mature small firm typically sold similar but not identical products to its competitors. They tried to differentiate their products. Out of the total sample, only an eighth of firms (17.5%) sold ‘identical’ products or services to competitors. Over a half (52.4%) sold ‘similar’ products and just over a quarter (25.4%) produced ‘different’ products. The question seems to have been well understood, as less than five per cent (4.8%) 'could not say'.
4. ESTIMATION OF SIMULTANEOUS EQUATION MODEL
The principal relations between size, competitive strategy and performance examined in our threeequation model are outlined in equations (1) to (3) above. This outline model is amplified here in equations (4) to (6). This now explicitly incorporates the exogenous variables discussed above. In (4) to (6), the functions f(.), g(.) and h(.) of (1) to (3) are expressed in linear forms with additive disturbance terms u _{i} (i =1, 3). The structural equations to be estimated were as follows:
P _{t} =
S _{t} =
_{β} _{0}
_{α} _{0}
+β _{1} S _{t}
+β _{2} T _{t} +β _{3} C _{t} +β _{4} D _{t} +β _{4} M _{t} +β _{5} A _{t}
+β _{6} A _{t} ^2 +u _{1}
(4) 

+α _{1} P _{t} 
+α _{3} LP _{S}_{t} 
+α _{4} OC _{t} 
+α _{5} T _{t} 
+u _{2} 

(5) 
15
C t =
_{γ} _{0}
+γ _{1} P _{t}
+γ _{2} Mriv _{t} +γ _{3} Diff _{t}
+u _{3}
(6)
Initially, a priori considerations were used to classify variables as either exogenous or
endogenous. Indeed the selection of P, S and C as endogenous arises as much from the extant literature, as from statistical considerations. However, once a priori knowledge has been incorporated in the model, its legitimacy needs to be explored econometrically. Therefore, formal tests for endogeneity within the system are examined immediately below, in Section
4.1. 

4.1 
Tests of Endogeneity 
This Section reports on two matters: (a) DurbinWuHausman type tests of exogeneity, to examine whether simultaneities exist between firm size, competitive strategy and small firm performance; and (b) ways in which a system of equations can be estimated. Concerning the later, one method would account for contemporaneous correlations between the errors of the structural equations in the system, using seemingly unrelated regression or SUR estimation. Another method would account for simultaneities between two or more endogenous variables, which are determined jointly within the system. We prefer the latter.
4.1 Tests of Endogeneity
Here DurbinWuHausman type tests are applied to investigate whether the set of estimates of the structural equations obtained by least squares are consistent or not. ^{v}^{i} If the null hypothesis that OLS estimates are consistent is rejected, endogeneity (not every regressor is asymptotically independent of the disturbances) is present and the IV estimator is preferred to the least squares estimator ^{v}^{i}^{i} (Davidson and MacKinnon, 1993 p237). Failure to reject the null hypothesis suggests that there is no need for structural modelling, but failure to reject it may or may not imply endogeneity. Only under very special conditions (see Geroski, 1982 p.58, for example) in industrial economics will failure to reject be compatible with exogeneity. Hence it is convenient to regard failure to reject as only indicative of exogeneity. Using DurbinWuHausman Tests some evidence of endogeneity was found between S=g(P) and C=h(P). The F statistics (and associated probability values in parentheses) for testing the null hypothesis that the coefficients on the fitted values of relevant test variable (obtained from regressions against all the exogenous variables in the system are zero) were as follows:
16
Test for Evidence of Endogeneity EQN (4)
Test Variable Fvalue _{(}_{1}_{,} _{5}_{3}_{)} Prob >F
C _{t}
S _{t}
2.12
(0.1511)
0.99
(0.3236)
F _{(}_{1}_{,} _{5}_{7}_{)}
EQN (5)
P _{t}
8.2
(0.0058)
F _{(}_{1}_{,} _{5}_{8}_{)}
EQN (6)
P _{t}
3.35
(0.0722)
Test for Evidence of Exogeneity Given that S _{t}_{,} C _{t} and P _{t} are assumed to be endogenous.
Test Variable
Fvalue
Prob >F
EQN (4) 

D _{t} 
M _{t} 
T _{t} 
1.58 
1.12 
0.02 
0.214 
0.2954 
0.8794 
EQN (5)
T _{t}
0.41
0.5228
OC _{t}
6.671
0.0119
EQN (6) Diff _{t} .
3.30
0.0746
Mriv _{t}
1.51
0.2241
This statistical evidence confirms our a priori reasoning. As a result, theoretical and statistical criteria can be invoked to support system estimation. Labour productivity earlier in the life of the firm, LP _{s}_{t} , and age, A _{t} , are predetermined within the system and are known to be exogenous. Tests were performed for the exogeneity of the other variables in the system, namely, the level of liabilities of the firm, D _{t} , the main market of the firm, M _{t} , the organising capability of the firm, OC _{t} , the number of major rivals, Mriv _{t} , and the level of differentiation, Diff _{t} . These tests lead us to regard D _{t}_{,} M _{t} and Mriv _{t} variables as clearly exogenous. There is some evidence of endogeneity of OC _{t} at the 5% level ^{v}^{i}^{i}^{i} and Diff _{t} at the 10% level ^{i}^{x} but this seems to be unidirectional in nature and thus is not explicitly modelled here. Sample size prohibits us from examining all sources of endogeneity in this system.
4.2 System Estimation The available methods of estimating simultaneous equations vary, based on their treatment of information, and their use of different estimators (maximum likelihood ^{x} versus instrumental variables). Single equation methods, like two stage least squares (2SLS), and limited information maximum likelihood (LIML), estimate the model parameters of each equation at a time, whereas fullsystem estimators, like three stage least squares (3SLS) and full information maximum likelihood (FIML), estimate all the parameters at once. Two system estimation techniques are adopted in this paper, namely two stage least squares (2SLS) and three stage least squares (3SLS), here for comparative purposes given the evidence of endogeneity. In the presence of endogeneity 3SLS, a full system estimator, is likely to have an efficiency advantage over the single equation methods, such as 2SLS. We report the results of iterative 3SLS because these results converge to those of FIML estimation for all parameters. The work of Jans and Rosenbaum (1996) illustrates the use of a 3SLS (in nonlinear form) estimation to estimate another industrial model in which endogeneity is intrinsic.
17
Davidson and MacKinnon (1993, p234) state that it is generally desirable for a model to be somewhat over identified in order to ensure good finite sample properties, for the use of instrumental variable (IV) estimation, of which 2SLS and 3SLS are examples. For equation (4), the number of excluded exogenous variables and included endogenous regresses are four and two, respectively; for equation (5) they are six and one, respectively and in equation (6) they are seven and one, respectively. The ‘order condition’ for identification therefore suggests that each of the three equations is over identified, and that there are thirteen over identifying restrictions in total within the system.
5. RESULTS To examine whether there are tradeoffs between firm size, competitive strategy and performance for the sample of 63 longlived small firms we estimated the system of structural equations (4)(6) using 2SLS and the I3SLS ^{x}^{i} estimation techniques. These two estimation
techniques were adopted for reasons of statistical efficiency, and to examine the robustness of the results. Each of these techniques place different restrictions on the data and have different merits, which overcome the failings of other techniques particularly in finite samples. Efficiency is not the overriding concern, as we also wish to limit finite sample bias. The results are reported in Table IV and V below. The estimates of the behavioural relation between firm size, competitive strategy and performance are discussed below in subsection
5.1. Finally, patterns of adjustment of size, competitive strategy and performance within the
simultaneous systems are examined in subsection 5.2.
5.1 Equation Estimates
[INSERT TABLE IV NEAR HERE]
An initial examination of the estimates suggest that the results are robust across the performance and size equations, under both estimation techniques, (see Table IV). In Table IV, the three equations are set out under each other with performance at the top, size in the middle, and competitive strategy at the bottom. These are the estimated versions of equations (1), (2) and (3) above. Coefficients are as in equations (4), (5) and (6) above. The tvalues are shown under each coefficient. These estimates indicate that a trade off indeed exists between firm size and performance. For the I3SLS estimates, size has a significant negative influence on performance, P=f(S), f’<0 and performance has a significant negative influence size, S=g(P), g’<0. For the 2SLS estimates, size is not significant in determining performance,
18
though its sign suggests a tradeoff. We regard the I3SLS estimates as being superior, in terms of econometric properties. We need to ask why an increase in size, here measured by fulltime equivalent employees reduces the performance of the small firm. Perhaps these reductions in performance arise from reduction in labour productivity. This could be because these are increased costs associated with hiring more employees, with rising effort being expended in recruiting and training new staff. Thus the small firm sacrifices increases in profits or performance to grow. Viewed the opposite way, the performance equations suggests that reductions in firm size lead to increases in performance. As the small firm reduces in size performance increases, due to increased efficiencies (i.e. increased labour productivity caused by the substitution of labour for capital and a leaner cost base), and a relative increase in the human capital at work in the small firm. Thus by becoming a leaner organisation, the survival and the long run prospects of the small firm are promoted. However, one would expect there to be diminishing returns to a survival strategy of this sort. At the limit, one man outfits will find it difficult to compete in the same league as dominant players in the market unless their goods are very specialised i.e. niche products. Certainly, if we assume that the goal of the entrepreneur is to raise his firm’s performance our tradeoff relationship suggests that downsizing may be the principal way to gain improvements in performance. The properties of the behavioural relation between the strategy space of the firm and performance are less clear. Estimation by both 2SLS and I3SLS techniques indicates that the competitive strategy space has a positive and significant effect on performance, in the empirical version of equation (1) in Table IV. In the empirical versions of equation (3), the coefficient on performance is negative in sign (suggesting a tradeoff), but it is insignificant. Causality is perhaps unidirectional, i.e. P=f(S,C) but C≠g(P). That is, the competitive strategy of the small firm has a significant positive influence on performance, but its performance does not significantly influence the heterogeneity of the competitive strategy, C=g(P). If again here we assume that the goal of the owner manager is to raise performance, the model points to improvements in performance if the small firm reduces its size and competes strongly, using a wide variety of strategies.
5.2 Elasticities
[INSERT TABLE V NEAR HERE]
We now turn to examine each equation, by reference to elasticities at the mean, so presented in Table V. Our goal now is to focus more on the quantitative impact of variables, rather than on their significance per se. In the performance equation a 1% increase in firm
19
size leads to a 0.03% fall in performance and visa versa. Thus, increases in the size of the small firm reduce its performance, but the impact is small. By contrast, the competitive strategy, C _{t} , has the largest impact on performance. A 1% increase in the strategy space of the firm leads to a 0.15% increase in performance. Should a small firm be operating in an industry which is subject to technical change, (T _{t} ) this experience has a significant positive effect (see Table IV) on performance and its impact is a quite large (elasticity is 0.12%). Thus industry level technical change seems to have an important effect on promoting the long run survival (and prospects) of the small firm. The more liabilities (D _{t} ) the small firm is exposed to, the lower its performance, though the size of this effect is not significant in the I3SLS estimation. A 1% increase in the forms of liabilities of the firm lowers performance by 0.3%. Reid (1993) shows that gearing has a crucial impact on many key aspects of the young small firm’s existence: medium term viability, growth and profitability. Excessive levels of liabilities may be detrimental in the medium term. Here (Table IV) the effect has a negative sign, but it is not significant predictor of long run survival in the full system estimation (I3SLS). The effect of greater exposure to external liabilities in the latter part of the firm’s life is uncertain. If equity finance is a cheaper source of finance capital, the optimal strategy for highly geared small firms is to retire debt early in its lifecycle (Reid, 2003). However later in its lifecycle other structures of finance capital could be appropriate. Thus Power and Reid (2003) found that gearing was insignificant in explaining long run survival. The geographic extent of the small firm’s main market (M _{t} ) had a positive effect on firm performance, but this effect was not significant. Less dependence on local markets, or put another way greater the market’s extent, the greater the performance of the small firm, other things being equal. However Reid (2001) found that unless firms start with marketing intentions, which are explicitly aimed at national or international markets, the small firm would never make this their main market, which perhaps explains why this effect is insignificant. The effect of age (A _{t} ) on performance was insignificant. Age squared was significant at the 10% level using 2SLS but not significant under I3SLS. There is therefore a weak suggestion that as the small firm gets older its performance falls but at a decreasing rate. Performance is a convex function of age. This is a plausible result, in that, if performance fell at an increasing rate the long run survival of these mature small firms would be very fragile. Such a result would not encourage the continuous investment in these firms which we have observed.
20
In the size equation, (middle equation in Table V) performance had the largest impact on size. A 1% increase in performance leads to as much as an 11% fall in size (measured by fulltime equivalent employment). This effect is of considerable magnitude compared to the inverse influence of size on performance (elasticity of –0.03%) in the first equation. This finding demonstrates why there is a tendency for these mature small firms to reduce their headcount and to become leaner, to gain further improvements in performance. Here also, a firm operating in an industry, which is subject to technical change, experiences a significant positive effect on firm size (elasticity is 1.4%). In response to technical change in the industry the firm grows in size. Greater headcount is required to respond to technical changes (i.e. the firm may require to increase the human capital of the firm etc.) The effect of labour productivity (LP _{s}_{t} ) earlier in the life of the firm on size was negative, but not significant in the I3SLS estimation. Firms, which generate more sales per fulltime equivalent employees, are more operationally efficient. Greater operationally efficiency earlier in the life of the firm indicates a superior performer at this stage. Superior performers would be expected to grow in size. This may not be the case however, as firms grow faster in size earlier in their lifecycle, than in the latter part of their life. Thus this finding is indicative of early lifecycle effects in labour productivity. Younger small firms grow faster, in response to increases in labour productivity than mature small firms. It had the lowest influence on size (elasticity is 0.09%). The organising capability of the firm, OC, had a positive and significant effect on the size of the firm at the 10% level using I3SLS. A 1% increase in organising capability of the firm raises firm size by 0.72%. This result is consistent with the discussion of Ghoshal, Hahn, Moran (2000) on administrative reorganisation. To engage in a larger number of functional activities, a greater division of labour is required. In the competitive strategy equation (equation 3) performance had a negative and insignificant effect on the heterogeneity of the firm’s competitive strategy. This is not surprising, as many factors other than performance may be determining the size, or scope, of the firm’s competitive strategy. Thus we are not concerned about this. Higher levels of product differentiation had a positive and significant effect on the dimensions of the competitive strategy of the firm. The size of the elasticity of this effect was 0.29 and 0.48 for the I3SLS and the 2SLS estimation, respectively. To the extent that product heterogeneity (Diff) confers local monopolistic advantages on the small firm, it increases the dimensions on which the firm competes to protect these advantages. This finding supports evidence that small firms usually seek to cultivate mild forms of product differentiation, especially by
21
customer service and delivery (Reid, 1993). However, it must be borne in mind that strongly
differentiated products can only be sold in very limited niche markets, especially if they are
constructed on a customer specified basis. The greater the competitive pressure in the
market, as measured by the count of the number of rivals (Mriv), the smaller the number of
dimensions of competitive strategy of the mature small firm. Thus as the market approaches a
perfectly competitive market the mature small firm competes on less dimensions. The size of
this impact is small relative to that of the level of product differentiation (elasticity = 0.04).
5.3 Diagnostics
Overall, reflecting on the full set of estimates displayed in Tables IV and the
elasticities reported in Table V, we find them to be broadly complementary. The I3SLS
results are arguably the most satisfactory set of estimates, in terms of statistically efficiency,
since this is a full system estimator. The overall significance and individual coefficient
significance were acceptable. The system R ^{2} measure,
the generalised variance in Y explained by variation in the right hand variables in the system
equations was high at 0.8325 for the I3SLS estimates. ^{x}^{i}^{i} A likelihood ratio test of the
hypothesis that all the slope coefficients were equal zero in the I3SLS regression was
rejected. The Chisquare statistic ^{x}^{i}^{i}^{i} of 112.56 was greater than 31.32 critical value of the chi
square distribution, with degrees of freedom equal to the number of variable coefficients in
2 measure are vindicated. A
the system, 14, at an α level of 5%. Thus the indications of R
BreuschPagan Lagrange multiplier test for diagonality of the covariance matrix was rejected
[Chisquare _{(}_{3}_{d}_{f}_{,} _{α} _{=} _{5}_{%}_{.}_{)} of 23.045 > 12.84]. ^{x}^{i}^{v} This confirms that an equationbyequation
application of least squares estimation would have been inappropriate, and confirms the value
of our seeking full system methods to estimate the structural equations of the model. These
should take account of contemporaneous crossequation correlation of disturbances, by using
full system estimators, such as the I3SLS that we employed.
A test of the number of over identifying restrictions in our model can be conducted
using the following likelihood ratio test statistic 2*[Ln L _{r} – Ln L _{u} ] where Ln L _{r} is the log
likelihood from the restricted reduced form system of equations obtained from an ML
estimation of a set of nonlinearly restricted equations in which each of the regressors is either
exogenous or predetermined and Ln L _{u} is the loglikelihood from the unrestricted reduced
form system of equations obtained from the SUR regression using ML estimation. This
statistic has a limiting chisquared distribution with degrees of freedom equal to the total
. The test for the 13 over
number of overidentifying restrictions
~
R
2 , which indicates the proportion of
~
d
22
k
j
*
j
identifying restrictions is equal to 2*[558.894(616.1646)] = 2*(57.2706)= 114.5212, which
is much larger than the 0.05 (0.01) chisquare critical value of 22.36 (27.688), suggesting that
the overidentifying restrictions are not consistent with the data (see Berndt, 1991, p.554). As the results of the system estimation seem robust across the two system estimation techniques the next subsection analyses the adjustment path of the behavioural relation between size, dimensions of competitive strategy and the multidimensional measure of performance.
5.4 Adjustment Paths
A final interpretation of the model can be undertaken by examining the relationships between jointly determined variables in the system; firm size, dimensions of competitive strategy and performance. Suppose all the exogenous variables in equations (4) and (5) are assigned to their mean values, the functions for the performance equation P=f(S, C) and for the size equation S=g(P) can be linearly approximated and examined in a two dimensional graph. The stability of these behavioural relations can then be examined. Using the estimated coefficients of Table IV, and the mean values for exogenous variables, these functions are approximated as follows;
P _{t} = 
59.938 
 0.17544S _{t} 
+ 2.1594 C _{t} 

(7) 

P _{t} = 
69.742 
 0.17544S _{t} 

(7a) 

S _{t} = 
160.668 
 2.1831P _{t} 

(8) 
It is informative to graph equations 7(a) and 8, as in Figure II. Solving out these two
expressions, the equilibrium values (S*, P*), which denote the equilibrium point E is (13.639, 67.349) are close the mean values for firm size and the multidimensional measures of performance in the sample which were 13.6508 and 67.3467 respectively (see Table III). It is
to be further noted that the equations indicate a stable equilibrium point. Thus starting from a
performance level of 69.742 on the horizontal axis a convergent path to the equilibrium point
E can be traced. Similarly starting from a size of 160 fulltime equivalent employees on the
vertical axis another convergent path to E can be traced. ^{x}^{v} As E is close to the relevant mean size and performance values in the sample, the typical mature firm in the sample has reached this equilibrium point. The relative size of the adjustments for S=g(P) is much larger that the relative magnitude of the adjustments for P=f(S). In response to increases in performance, there is a strong tendency for the small firm to adjust downwards in size. Therefore, to
23
improve the long run survival prospects of the small firm they need to become leaner, and more efficient in size. Thus the tradeoff between firm size and performance implies that
there is a strong performance driven effect to remain small or to reduce in size. We now turn
to the effect of the dimensions of competitive strategy on this tradeoff.
[INSERT FIGURE II NEAR HERE]
Figure III illustrates shifts in the performance function as a result of a change in the dimensions competitive strategy of the firm. If the firm increases the heterogeneity of the competitive strategy it undertakes the performance function shifts to the right and a new equilibrium point E* is reached where E* represents higher values of performance and lower values of firm size. The magnitude of this increase in performance could be reduced if improvements in performance feedback into the heterogeneity of the firm’s competitive strategy. However this effect was insignificant across both system estimation techniques and thus is not given much emphasis here. In essence this figure is suggesting that the long run prospects of these small firms can be promoted through further specialisation of the dimensions by which they compete. Competing on a diverse range of attributes of the firm’s product, price, service enables the firm to achieve a higher equilibrium performance. This perhaps explains why small firms usually seek to cultivate mild forms of product differentiation, especially by customer service and delivery etc. Strongly differentiated
products can only be sold in very limited niche markets, especially if they are constructed on
a customer specified (i.e. bespoke) basis. However economies of scope perhaps exist for
these small firms, in the pursuit of these strategies in more localised or niche markets, Reid
(1993).
[INSERT FIGURE III NEAR HERE]
6. CONCLUSIONS
This paper examines behavioural relations between firm size, the heterogeneity of the firms competitive strategy and the performance of the longlived small firm in Scotland. The latter two variables are measured in novel ways. In this work we find there is a strong tendency for the small firm to remain small on two fronts. First as a tradeoff exists between firm size and performance, the mature small firm must become leaner and more efficient over time if it is to survive. The small firm adjusts downwards in size by a considerable amount to achieve further increases in performance. Second to attain higher equilibrium values of performance a varied competitive strategy needs to be adopted. This can be achieved through producing customised or specialist products but also through increasing the aggressiveness of its competitive strategy to defend market niches such as raising advertising and marketing
24
efforts. Survival of small firms is linked to product differentiation, as typically small firms are niche players. The tendency in this instance is to become more specialised and localised and to seek economies of scope, to improve the long run prospects of the firm. The firm must also be proactive in defending its niche in the market. Acting in these ways, entrepreneurs can have a positive influence on the long run performance of the small firm. Two system estimation methods were employed to estimate the behavioural relation between firm size, the heterogeneity of the firm’s competitive strategy and performance; two stage least squares and iterated three stage least squares. The similarity of the results across these two estimation techniques suggests the robustness of the results.
25
Appendix
Definition of variables used in main text
Variable
P
LP _{S}_{t}
OC
S
A
T
C
D
M
Diff
Mriv
= ∑f _{i} /n where f _{i} is the self appraised score between 0100 for each factor averaged overall factors 1 to n which were applicable. = [Sales at first interview (1985 for SBE, 1991 for telephone, 1994 for Leverhulme) at 2001 prices]/ [Employees at first interview (1985 for SBE, 1991 for telephone, 1994 for Leverhulme) at 2001 prices]
= ∑f _{i} where f _{i} are forms of competition used by the firm in their principal markets
Number of fulltime equivalent employees in 2001.
Age of firm, in years.
=1 a lot of technical change in your industry over the life of the business, = 0 otherwise.
= ∑f _{i} where f _{i} are forms of competition used by the firm in their principal markets
= ∑f _{i} where f _{i} are forms of debt used by the firm in their principal markets
=1 (Local), =2 (Regional), =3 (Scottish), =4(British/International)
=1 (Identical), =2 (Similar), =3 (Different), =4 (Cannot say)
= The number of major rivals
Table I The Extraction of the Sample
Sample 
Parent 
Extracted 
Survivors 

Survivors 
Non 
Total 
Non 
Interviewed 

survivors 
response 

Leverhulme (19851988) 
86 
25 
61 
86 
5 
20 
Telephone Survey (1991) 
160 
50 
63 
113 
20 
30 
Leverhulme (19941997) 
150 
15 
5 
20 
2 
13 
Total 
396 
90 
129 
219 
27 
63 
Table II The Composition of the Sample
Sample 
Extracted 
Interviewed 

Sample 
Survivors 

Manuf. 
Services 
Manuf. 
Services 

Leverhulme (19851988) 
53 
33 
12 
8 
(61%) 
(39%) 
(60%) 
(40%) 

Telephone Survey (1991) 
23 
90 
9 
21 
(20%) 
(80%) 
(30%) 
(70%) 

Leverhulme (19941997) 
8 
12 
4 
9 
(40% 
(60%) 
(31%) 
(69%) 

Total 
84 
135 
25 
38 
(38%) 
(62%) 
(40%) 
(60%) 
26
Endogenou
s
Exogenous
Figure I Response Format for Performance Indicator
_{4}_{.}_{1} We'd like to know what has kept you in business down the years. Some things are good for business and some things are bad. What effect have the following had?
[Show with a cross whether the effect was good or bad.]
Technology
N/A
Bad
Neutral
Good
0 
25 
50 
75 
100 

Rival's Innovation 
N/A 
Bad 
Neutral 
Good 



0 
25 
50 
75 
100 

Regulation 
N/A 
Bad 
Neutral 
Good 



0 
25 
50 
75 
100 
Table III
Endogenous and Exogenous Variables
Variable
N
Mean
Std. Dev.
Min.
Max.
{
{
P
S
C
LP _{S}_{t}
OC
S
A
_{S}_{e}_{c}_{t}_{o}_{r}
LegalStatus
Mriv
M
D
Diff
C
T
63 
67.3467 
8.1036 
49.11 
90.43 
63 
13.6508 
19.8488 
1 
130 
63 
4.5397 
1.8035 
1 
8 
63 
113489 
125103 
1780 
549577 
63 
7.2381 
2.1381 
3 
11 
63 
13.6508 
19.8488 
1 
130 
63 
25.5397 
15.7284 
10 
90 
_{6}_{3} 
1.6349 
0.48532 
_{1} 
_{2} 
63 
2.1905 
0.8203 
1 
3 
63 
26.0318 
126.1867 
0 
1000 
63 
2.2698 
1.2599 
1 
4 
63 
1.8254 
1.4429 
0 
4 
63 
2.1746 
0.7733 
1 
4 
63 
4.5397 
1.8035 
1 
8 
63 
0.8254 
0.3827 
0 
1 
27
Table IV Results of System Estimation
Estimation 
2SLS 
I3SLS 

Equations 
Coefficient 
Coefficient 

(t stat) 
(t stat) 

P _{t} = 

β 
0 
53.113 
54.996 

(8.059) 
(10.65) 

+β _{1} S _{t} 
4.65E03 
0.17544 

(0.6040E01) 
(2.87) 

+β _{2} T _{t} 
8.6902 
10.09 

(3.706) 
(4.46) 

+β _{3} C _{t} 
2.8866 
2.1594 

(2.015) 
(1.83) 

+β _{4} D _{t} 
1.9571 
0.91728 

(2.31) 
(1.402) 

+β _{5} M _{t} 
0.61294 
0.14136 

(0.7171) 
(0.2378) 

+β _{6} A _{t} 
0.28142 
0.14165 

(1.536) 
(1.099) 

+β _{7} A _{t} ^2 
3.78E03 
1.77E03 

(1.874) 
(1.204) 

S _{t} = 

α 
0 
146.71 
133.08 

(2.758) 
(2.71) 

+α _{1} P _{t} 
2.4254 
2.1831 

(3.023) 
(2.896) 

+α _{2} T _{t} 
25.869 
23.049 

(2.742) 
(2.738) 

+α _{3} LP _{S}_{t} 
5.44E05 
1.14E05 

(2.433) 
(0.7666) 

+α _{4} OC _{t} 
2.0858 
1.3618 

(1.615) 
(1.747) 

C _{t} = 

γ 0 
4.6964 
7.3857 

(1.393) 
(2.122) 

+γ _{1} P _{t} 
3.01E02 
5.95E02 

(0.5928) 
(1.139) 

+γ _{2} Mriv _{t} 
3.56E03 
7.54E03 

(1.951) 
(4.361) 

+γ _{3} Diff _{t} 
0.90142 
0.62544 

(3.2) 
(2.463) 
Table V Elasticities at Mean
28
Estimation 
2SLS 
I3SLS 

P _{t} = 

β 
0 
0.7886 
0.8166 
+β _{1} S _{t} +β _{2} T _{t} +β _{3} C _{t} +β _{4} D _{t} +β _{5} M _{t} +β _{6} A _{t} +β _{7} A _{t} ^2 
0.0009 
0.0356 

0.1065 
0.1237 

0.1946 
0.1456 

0.0530 
0.0249 

0.0207 
0.0048 

0.1067 
0.0537 

0.0503 
0.0235 

S _{t} = 

α 
0 
10.7475 
9.7491 
+α _{1} P _{t} +α _{2} T _{t} +α _{3} LP _{S}_{t} +α _{4} OC _{t} 
11.9657 
10.7703 

1.5642 
1.3936 

0.4519 
0.0945 

1.1060 
0.7221 

C _{t} = 

γ 0 
1.0345 
1.6269 

+γ _{1} P _{t} 
0.4459 
0.8833 

+γ _{2} Mriv _{t} +γ _{3} Diff _{t} 
0.0204 
0.0432 

0.4318 
0.2996 
Figure II Size Performance Tradeoff
Figure III Impact of an Increase in the Diversity of Competitive Strategy
29
Endnotes This research has been undertaken with the generous support of Enterprise Ireland, to whom the authors make grateful acknowledgement. We should also like to thank many owner managers of small firms in Scotland. They gave generously of their time, over the period 20012, allowing us to collect high quality data in the field.
^{i} See Jacobsen (1986).
^{i}^{i} See Reid and Andersen (1992).
^{i}^{i}^{i} See Smith (1997).
^{i}^{v} The factors were generated from theory and empirical evidence from studies examining differences in the performance of longlived small firms.
^{v} Rating factors along a continuum is a much easier task than ranking the list of factors from top to bottom especially for long lists of factors. The ranks can be tied when the factors are rated. The consistency which ownermanagers rate factors on each scale item is also improved by defining the meaning respondents should assign to middle alternatives using adjectival labeling of points which is undertaken here.
^{v}^{i} In practice the test is implemented as follows: Suppose a structural equation is
y
1
= δ
X
1
+ β
y
2
+
u
where y _{1} and y _{2} are vectors of suspected endogenous variables, X _{1} is a matrix of exogenous and predetermined
variables, and u a vector of error terms. Let
regression of y _{2} against all the exogenous and predetermined variables in the system. The DWH test is simply
yˆ is equal to zero (i.e. test πˆ = 0 ) in an estimation of the following
regression _{y}
^{v}^{i}^{i} As it happens least squares is the preferred estimator as the asymptotic covariance matrix of the least squares estimator is never larger than that of IV estimator, it will actually be smaller unless endogeneity exists (see Greene, 2000 p.383).
^{v}^{i}^{i}^{i} F _{(}_{1}_{,}_{6}_{0}_{)} statistic = 0.17 for Ho
could not be rejected
an F test that the coefficient π on
yˆ be the vector of fitted values of y _{2} from a reduced form
2
2
1
= δ
_{X}
1
+ β
_{y}
2
_{y}_{ˆ}
+ π
2
+
_{u}
.
ˆ
^{i}^{x} F _{(}_{1}_{,}_{6}_{0}_{)} statistic = 0.57 for Ho
could not be rejected
ˆ
^{x} Maximum likelihood methods are invariant to reparametrisation whereas instrumental variables are not.
30
^{x}^{i} When higher that 2 iterations are used the significance of the performance and size variables increase rapidly due to rounding errors. ^{x}^{i}^{i} Single R ^{2} measures are not appropriate in an equation system. The R ^{2} from a particular equation computed could be negative since with system estimation in general it is not the case within each equation the sum of the residuals is zero. The numerator could be larger than the denominator that is the unexplained variation can be larger than the total variation implying a negative R ^{2} . This is because single equation systems minimises e’e and therefore maximises the R ^{2} in general. System estimation methods do not minimise e’e. The maximum likelihood estimator minimises the determinant of the residual cross products matrix; that is ML minimises det E’E. Hence ML does not maximise the individual equation R ^{2} values. Since single equation R ^{2} measures are flawed in the equation system context a different goodness of fit measure should be employed.
^{~} ^{2} =
R
1
−
~ 

The system 
R 
2 reported in Shazam is defined as 

~ 
ˆ
)(Y '
Y )
∑
/ (Y −
Y
−


R = 
1 − 
where Y is an n x k matrix andY contains the sample means.
^{x}^{i}^{i}^{i} The Chisquare statistic is
χ
2
=
~
− N log 1 − R
(
(
2
))
^{x}^{i}^{v} The Lagrange Multiplier statistic reported on SHAZAM is computed as
λ
=
N
k
i
− 1
∑ ∑
i = 2
j
= 1
r
2 with
ij
squared correlation coefficient of residuals given by
r ^{2} =
ij
σ ˆ ^{2}
ij
ˆ
σ
ii
ˆ
σ
jj
. Under the null hypothesis of a
diagonal covariance structure the statistic has an asymptotic
^{x}^{v} This stability condition can be expressed:
(dP/dS) _{7}_{a} = 0.17544 > 0.45806 = (dP/dS) _{8}
2
χ (M (M −1)/ 2)
distribution.
REFERENCES
Bain J., (1956), Barriers to New Competition, Cambridge: Harvard University Press.
Birch, D., (1996), Paper presented to the Jönköping International Business Conference on Enterpreneurship, SME’s and the Macro Economy, 1314 ^{t}^{h} , June (mimeo).
Beccarello, M., (1997), Time Series Analysis of Market Power: Evidence from G7 Manufacturing. International Journal of Industrial Organisation, Vol.15: 123136.
Brock, W. A. and Evans, S. E., (1989), Small Business Economics, Small Business Economics, 1(1): 720.
Berndt, E. R., (1991), The Practice of Econometrics: Classic and Contemporary, Addison Wesley Publishing Company; United States.
Cubbin, J., and Leech, D., (1986), Growth versus Profit Maximisation: a Simultaneous Equations Approach to Testing the Marris Model, Managerial and Decision Economics, Vol. 7: 12331.
Davidson R. and MacKinnon, J. G., (1993), Estimation and Inference in Econometrics, Oxford University Press; Oxford.
Demsetz, H., (1973), Industry Structure, Market Rivalry, and Public Policy, Journal of Law and Economics, Vol. 16:19
31
DeVellis, R., (1991), Scale Development: Theory and Applications. Vol. 26. Applied Social Research Methods Series. Sage Publications; London
Dierickx, I. And Cool, K., (1989), Asset Accumulation and Sustainability of Competitive Advantage, Management Science, Vol. 35: 15041511.
Dobson, S. and Gerrard B., (1989), Growth and Profitability in the Leeds Engineering Sector, Scottish Journal of Political Economy, Vol. 36 (4): 33452
Geroski, P. A., (1982), Simultaneous Equation Models of the Structure Performance Paradigm, European Economic Review, Vol.19: 145158.
Ghoshal S., Hahn, M. and Moran, P., (2000), Organising for Firm Growth: The Interaction between Resource Accumulating and Organising Process, In Foss, N., and Volker M., (eds.) Competence, Governance and Entrepreneurship: Advances in Economic Strategy Research, Oxford: Oxford University Press.
Gould, S. J., (1993), Eight little Piggies: Reflections in Natural History, New York, Norton.
Greene, W. E., (2000), Econometric Analysis, 4 ^{t}^{h} Edition, New Jersey: Prentice Hall International.
Hall, B., (1987), The Relationship Between Firm Size and Firm growth in US Manufacturing Sector, Journal of Industrial Economics, Vol. 35: 583605.
Jacobsen, L. R., (1986), Entrepreneurship and Competitive Strategy in the New Small Firm:
An Empirical Investigation, University of Edinburgh, Department of Economics, Ph.D. Thesis.
Jans, I., and Rosenbaum, D. I., (1996), Multimarket Contact and Pricing: Evidence From the US Cement Industry, International Journal of Industrial of Organisation, Vol. 15: 391412.
Jovanovic, B., (1982), Selection and Evolution of an Industry, Econometrica, Vol. 50: 649
670.
Lippman, S., and Rumelt, R., (1982), Uncertain Imitability: An Analysis of InterFirm Differences in Efficiency Under Competition, Bell Journal of Economics, Vol. 13: 418438.
Mansfield E., (1962), Entry, Gibrat's Law, Innovation and the Growth of Firms, American Economic Review, Vol 52(5): 10231051
Marris,
Publications.
R.,
(1964),
The
Economic
Theory
of
Managerial
Capitalism,
London;
Sage
Mata, J., (1994), Firm Growth During Infancy. Small Business Economic, Vol. 6: 27 – 93.
McDonald, J. T. and Bloch H., (1999), The Spillover Effects of Industrial Action on Firm Profitability, Review of industrial Organisation, Vol. 15: 183200
32
Penrose, E. T., (1959), The Theory of Growth of the Firm, Oxford; Basil Blackwell.
Porter, M., (1985), Competitive Advantage, New York; Free Press.
Power, B. and Reid G. C., (2002), Flexibility, FirmSpecific Turbulence and the Performance of the Longlived Small Firm, CRIEFF Discussion Paper, Dept. of Economics, University of St. Andrews, No. 0207
Reid, G. C., (2003), Trajectories of Small Business Financial Structure, Small Business Economics, forthcoming.
Reid, G. C. and Smith J. A., (2000), What Makes a New Business Startup Successful? Small Business Economics, 14(3): 165182.
Reid, G. C., (1998), Limits to a Firm’s Rate of Growth: The Richardsonian View and its
Contemporary Empirical Significance, Ch.3 in B.J. Loasby and N.J. Foss (eds.) Capabilities and Coordination: Essays in Honour of G.B. Richardson; Routledge; London, 1998, 243
260.
Reid, G. C., (1995), Early LifeCycle Behaviour of MicroFirms in Scotland, Small Business Economics Vol. 7: p8995.
Reid, G. C., (1993), Small Business Enterprise: An Economic Analysis. London; Routledge.
Reid, G. C., Jacobsen L.R. and Andersen, M., (1993) Profiles in Small Business: A Competitive Strategy Approach. London; Routledge.
Reid, G. C
Instrumentation and Summary Statistics, CRIEFF Discussion Paper, Dept. of Economics, University of St. Andrews, No. 9207.
and Andersen M. E. (1992), A New Small Firms Database: Sample Design,
Reid, G. C., (1987), Theories of Industrial Organisation, Oxford: Basil Blackwell.
Richardson, G. B., (1964), The Limits to a Firm’s Rate of Growth, Oxford Economic Papers, Vol. 16 (1): 923.
Rumelt, R., (1991), How much does Industry Matter, Strategic Management Journal, Vol. 12: 167186.
Schmalensee, R., (1985), Do markets differ much? American Economic Review, Vol. 75:
341351.
Singh, A. and Whittington, G., The Size and Growth of Firms, The Review of Economic Studies, Vol. 42(1): 1526.
Slater, M., (1980), The Managerial limitation to the Growth of Firms, Economic Journal, Vol. 90: 520528.
Smith,
J. A., (1997) Small Business Strategy: An Empirical Analysis of the Experience of
New Scottish Firms, University of Abertay, Dundee, Scotland, Ph.D. Thesis.
33
Storey, D. J., (1994) Understanding the Small Business Sector. London; Routledge.
Wickham, P. A., (1998) Strategic Entrepreneurship. Pearson Education; London.
34
Гораздо больше, чем просто документы.
Откройте для себя все, что может предложить Scribd, включая книги и аудиокниги от крупных издательств.
Отменить можно в любой момент.