Are large firms born or made? Evidence from developing countries

See Levine (2005) for a review of the finance and growth literature.

See, for example, Evans (1987) and Davis et al. (1996).

An exception is Hsieh and Klenow (2014) who contrast growth trajectories in India and Mexico with those in the US.

Studies in the USA suggest that initial conditions are strong predictors of future public firm status and hence growth and productivity (e.g., Maksimovic et al. 2013) and future skill level of establishments (e.g., Ayyagari and Maksimovic 2018).

To examine the relative influence of firm effects compared with country effects, we use variance decomposition analysis. Similar methods have been used in finance to examine the importance of country- versus firm-level effects in explaining property rights protection (Ayyagari et al. 2008). The method is also well established in corporate strategy studies in the context of decomposing profitability into corporate and industry effects (e.g., Schmalensee 1985; Rumelt 1991; McGahan and Porter 1997, 2002; Khanna and Rivkin 2001). By using this method, we can focus directly on the general importance of these effects in explaining firm size and growth over life cycle without any assumptions on structural analysis. We also complement this analysis using the decomposition of the R² into contributions of each set of variables according to the Shapley value (see Huettner and Sunder 2012) and find very similar results.

While the Enterprise Survey data surveys establishments, we use the terms firm and establishment interchangeably since we restrict our analysis to single establishment firms.

The establishments in the survey were explicitly told that if the establishment was privatized, then the establishment year provided should refer to when the original government-owned establishment began operations.

Of the explainable variation in size at the country level, 7–31% can be explained by using historical institutional variables proposed in the literature such as legal origin (La Porta et al. 1998), ethnic fractionalization (Easterly and Levine 1997; Alesina et al. 2003), and latitude (Engerman and Sokoloff 1997). In smaller subsamples where we also include settler mortality (Acemoglu et al. 2001), a proxy for the disease environment during European colonization that shaped property rights protection, the historical institutional variables together explain between 57 and 80% of the explainable variation in size for firms across different age bins. Similar results hold for growth, suggesting that the literature has been looking at first-order institutional factors.

However, we also find a great deal of variation in the size–age gap across countries. In about 22% of the sample (27 countries), the size–age ratio is as large or larger than in the US.

Our paper also contributes to the literature on the importance of life cycle explanations for many fundamental corporate finance policies, including dividends (Fama and French 2001; Grullon et al. 2002; DeAngelo et al. 2006; Denis and Osobov 2008), financing (Berger and Udell 1990), stock valuations (Pástor and Veronesi 2005), and acquisitions (Maksimovic and Phillips 2008). We also know that size and age are closely related and are the best predictor of financing constraints (Hadlock and Pierce 2010).

The Enterprise Surveys and their precursor, the World Business Environment Survey (WBES), have been used to investigate a series of questions in finance and economics including the relation between property rights and contracting institutions (e.g., Acemoglu and Johnson 2005), investment climate and business environment obstacles to growth (e.g., Beck et al. 2005; Ayyagari et al. 2008), firm financing patterns (e.g., Beck et al. 2008; Cull and Xu 2005; Ayyagari et al. 2010), dispute resolution via courts (e.g., Djankov et al. 2003), corruption (e.g., Barth et al. 2009), and tax evasion (Beck et al. 2014).

The master list of firms is sometimes obtained from other government agencies such as tax or business licensing authorities. In some cases, the sampling universe is generated from lists maintained by the Chamber of Commerce and business associations or marketing databases where registration is voluntary. In a few cases, the sample frame is created via block enumeration.

We have a few firms in the sample that report having less than 5 employees. All our results are robust to dropping these firms. Our results are also robust to dropping firms that started out as informal enterprises and were formally registered after starting operations.

Most surveys contain three sets of weights—strict, median, and weak weights depending on the eligibility criteria used to construct the sample universe. The survey implementation manual recommends the use of median weights for cross-country comparisons which is what we use in this paper.

In the Enterprise Surveys, the establishment is defined as a physical location where business is carried out and where industrial operations take place or services are provided. In addition, an establishment must make its own financial decisions, have its own financial statements separate from those of the firm, and have its own management and control over its payroll.

When we focus on multi-establishment firms, we find that they have similar size–age profiles as single-establishment firms. The average 40+ multi-establishment firm is 7.33 times larger than the average multi-establishment firm that is < 5 years of age. If we were to combine single- and multi-establishment firms, we find that the average 40+-year firm is 5.4 times larger than the average establishment that is < 5 years of age.

The Enterprise Survey database was downloaded in July 2019.

Our results are robust to using median regressions rather than OLS for the specifications with size ratios.

The 70% mean response rate across our sample of countries is superior to most other survey-based studies. For instance, Campello et al. (2011) report response rates of 3 to 7% in their survey of how companies use credit lines during a financial crisis. Other studies in corporate finance report response rates between 7 and 9% including Graham and Harvey (2001) and Lins et al. (2010).

Three countries—Ethiopia, Iran, and Yemen—were reclassified from French civil law in La Porta et al. (1998) to English common law in Djankov et al. (2007). We retain the original classification, but our results are not significantly different with the change.

A parallel literature has established the costs associated with government ownership in financial markets (Khwaja and Mian 2005; Dinc 2005; Cole 2009; Sapienza 2004; Carvalho 2014).

In unreported robustness tests, we also checked the explanatory power of more endogenous variables such as whether the firm is an exporter, years of experience of the top manager, whether the firm has an informal/unregistered competitor, whether the firm has access to a checking/savings account, whether the firm has access to an overdraft facility/line of credit, and whether the firm has an internationally recognized quality certification and found them to have much lower predictive power than the variables we consider above.

In robustness tests reported below, for a much smaller number of countries where we have panel data on firms, we are able to add firm-fixed effects to examine the upper bound that can be explained by firm-level factors.

Figure 1 shows that the firm that is 40+ years on average employs 5.3 times as many people as the firm that is less than 5 years old. The 95% confidence intervals show that this number could vary between 4.39 and 6.24.

In 10% of the sample (12 countries), the size–age ratio for the firm that is 40 years or older is 10 or over, and in 20% of the sample, the size–age ratio is over 7, which is the number by which the average 40-year-old firm in the US is larger than the average firm that is 5 years or younger in the US (Hsieh and Klenow 2014).

The first column in panel A presents the underlying data in Fig. 2. Across income groups, we find the size–age gap ranges from 5.5 times in the case of low income countries to 6.83 times in the case of high income countries. Across regions in panel B, we find that the size–age gap for firms 40 years and older ranges from 2.13 in East Asia and Pacific and North Africa to 5.64 in Sub-Saharan Africa.

We exclude countries with Socialist legal tradition because Ayyagari et al. (2008) argue that these countries are fundamentally different from others in their perceptions of property rights protection. One of the other concerns with the former Socialist economies might be that the data on firm year is noisy because of mass privatizations though the firms were queried on when the firm first started operations even if it was under state ownership. However, in Fig. 4, we find consistent results with our main finding when we drop former Socialist economies.

As shown in Table 10 in the Appendix, we get very similar results if we were to repeat this analysis in terms of log (size).

The constant shows that the average firm that is less than 5 years old is 2.67 times its size when it first started operations, whereas the coefficient of the 40+ age bin shows that the average 40+-year-old firm is 8.21 times its size when it first started operations.

See Coad (2007) for an excellent survey of the literature on firm growth drawing from multiple fields including economics, management, and sociology.

The interaction of country dummies and size at birth dummies explains 6.2% of the variation in size ratios for young firms, 4.3% of the variation in size ratios for mid-age firms, and 9.1% of the variation in size ratios for mature firms. When we look at labor productivity, country dummies × size at birth dummies explain 44.9% of the variation in labor productivity for young firms, 43.2% of the variation in labor productivity for mid-age firms, and 33.2% of the variation in labor productivity for mature firms.

Following Davies and Jeppesen (2015), we dropped Ghana and Venezuela due to unreliable sales values; dropped observations where the interviewer deemed that the respondent was untruthful or provided arbitrary or unreliable numbers; deflated total annual sales in local currencies by CPI and converted to USD; dropped sales values lower than 100 or greater than GDP in the same year; and finally in addition to winsorizing size and labor productivity at the 99^th percentile, we dropped values that are less than 25^th percentile minus 3 times the interquantile range (difference between 75 quantile and 25 quantile) or greater than 75^th percentile plus 3 times the interquantile range.

For comparison sake, the OP covariance term for labor productivity reported by Bartelsman et al. (2013) for the US is 0.51 which implies that the industry index of labor productivity in the average US manufacturing industry is 50% higher than it would be if employment shares were randomly allocated within industries.

Note that this aligns with our earlier finding that size and productivity increase with age.