The impact of financial development on industry growth and
reallocation: a micro-level approach
Master Thesis (6-7-2012)
D. Veenstra (s1700618) d.veenstra.4@student.rug.nl
University of Groningen Faculty of Economics and Business MSc Thesis supervisor: prof. dr. M. Koetter
Abstract
This study aims to investigate the relationship between financial development and economic growth in more detail by incorporating a specific reallocation term to account for micro-distortions. Based on approximately 1.6 million firm observations for 9 eurozone countries over the 1996-2006 period, industry-level output growth is decomposed into input factor accumulation, technical change and reallocation to analyse how different sources of growth are affected by financial deepening. Overall, the findings do not indicate a significant and robust impact of financial development on growth or its components, suggesting that the link between finance and growth may not be as strong as once thought.
2
1. Introduction
Ever since Schumpeter (1912) underlined the importance of finance for economic growth through the facilitation of innovation, the link between financial development and growth has been analysed extensively.1 Fundamentally, it has been argued that the financial system can stimulate growth by either improving the allocation of resources and/or by ensuring that the allocated resources are used more productively. Indeed, several empirical studies have already demonstrated the importance of financial development using cross-country regressions (e.g. King and Levine, 1993; Levine and Zervos, 1998; Rajan and Zingales, 1998). However, most empirical research on the relationship between finance and growth simply looks at aggregate growth measures, as opposed to examining the sources of growth. One potentially important source of growth is the reallocation of resources between firms, which could be relevant for two reasons: i) the increased focus on firm heterogeneity and ii) the role of finance in improving allocation and productivity.2
There is a growing body of literature highlighting the importance of firm-level heterogeneity and reallocation in accounting for aggregate productivity growth (e.g. Foster et al., 2001; Hsieh and Klenow, 2009; Syversen, 2011). These studies document large and persistent differences in productivity across individual producers, even within industries, indicating the potential scope for welfare enhancing reallocation towards more productive producers. Indeed, while varying per industry and period, substantial reallocation of inputs and outputs across firms has been observed and this reallocation occurs mostly within, rather than between, sectors. The considerable productivity differentials and large-scale reallocation that is taking place suggest that reallocation can play a significant role in explaining aggregate productivity growth.
Second, the financial intermediation theory postulates that the financial system is important for improving allocation and productivity (e.g. Boyd and Prescott, 1986; Greenwood and Jovanovic, 1990). The theoretical models stress that financial intermediaries may reduce the cost of acquiring and processing information for investors, allowing for a better assessment of investment opportunities and thereby improve resource allocation. By alleviating information asymmetries between potential investors and relaxing financing constraints, the financial system can thus induce a more efficient allocation of capital. Several empirical papers have confirmed the hypothesis that the financial system can improve resource allocation (e.g. Jayaratne and Strahan, 1996; Aghion et al.,
1 See Levine (2005) for a comprehensive survey.
2 Productivity is simply efficiency in production: how much output is obtained for a given set of inputs
3 2007).3 In light of the importance of within-sector reallocation for aggregate growth and the potential role of the financial system therein, it is interesting to examine whether financial deepening has a direct impact on this channel. This paper will therefore attempt to shed more light on the finance-growth nexus by focusing on the sources of growth in more detail.
1.1 Problem statement
The thesis will try to link the two branches of literature by adopting the Rajan and Zingales (1998) difference-in-difference approach, while using growth measures derived from firm-level data as opposed to aggregate industry measures.4 Doing so will allow for a better understanding of the relationship between finance and growth. Adopting a micro-approach also provides several important advantages. First, using firm-level data it is possible to account for micro-distortions in the allocation of resources across firms. Based on Basu et al. (2009), I will show that output growth can be decomposed into three different factors: total input growth, productivity growth and reallocation of resources among firms.5 This decomposition is relevant as the impact of financial development is likely to differ across these three sources. In fact, recall that one of the main channels through which finance is purported to stimulate growth is by improving the allocation of resources, as argued by, for instance, Aghion et al. (2007), who show that credit markets facilitate the reallocation of resources from low-productivity to high-productivity firms. Furthermore, another channel through which financial intermediation can affect growth is through facilitating uncertain but productivity enhancing investment projects, thereby promoting technical change. By decomposing output growth into the three different sources and focusing in more detail on the channels through which finance is supposed to stimulate growth, it will be possible to empirically examine in more detail how financial development affects growth.
Therefore, adopting this strategy allows me to evaluate how aggregate growth and its components are affected by financial development. While numerous other studies have built upon the Rajan and Zingales (1998) framework, none of them have considered the relationship between finance and growth using a more sophisticated decomposition of output growth derived directly from firm-level production functions. In light of the potential significance of micro-level distortions,
3
Aghion et al. (2007) look primarily at entry and exit of firms to analyse reallocation, whereas I also consider reallocation among existing firms. Also, see Levine (2005) for more studies investigating the finance, reallocation and growth relationships.
4
The Rajan and Zingales (1998) difference-in-difference specification attempts to address the issue of
endogeneity between finance and growth. This approach is discussed in more detail in the literature review (2) and methodology section (3).
5
4 this approach thus constitutes an important contribution to a better appreciation of one of the channels linking finance to growth.
The remainder of this paper is organized as follows. Section 2 will provide a brief overview of the relevant literature on finance and growth. Next, section 3 discusses the applied methodology, while section 4 describes the data used in the estimation. Section 5 will present the estimation results, followed by a robustness analysis in section 6. Finally, section 7 concludes.
2. Literature Review
The study of financial development and economic growth dates all the way back to Schumpeter (1912), in which he underlines the importance of finance for economic growth. Schumpeter argues that credit enables the entrepreneur to alter the economic structure through innovative activities that will subsequently generate economic growth. The academic research on the topic has come a long way since then and, from a theoretical point of view, the modern view on the role of finance contends that financial development brings improvements in the (i) information available about possible investments, (ii) monitoring of investments, (iii) trading, diversification, and management of risk, (iv) mobilization and pooling of savings and (v) exchange of goods and services (Levine, 2005). Improvements in each of these financial functions allow for better investment decisions and could in turn facilitate growth.
Turning to empirical analyses, the traditional finance-growth literature has emphasized the importance of finance by including a financial development measure in standard growth regressions (e.g. King and Levine, 1993; Levine and Zervos, 1998; Levine, 2002).6 These studies adopt long-run cross-section analyses to confirm Schumpeter’s insights that financial intermediation stimulates development. For instance, King and Levine (1993) examine the relationship between financial development and economic growth for 80 countries over the period 1960-1989, finding that financial deepening is positively and robustly associated with economic growth, capital accumulation and the efficiency of capital allocation. The financial indicators adopted in the traditional finance-growth studies measure both bank development - such as the formal financial intermediary sector relative to GDP and the ratio of credit issued to private firms to GDP – and stock market development – such as the overall size of the market (market capitalization relative to GDP), activity of the market (value of trades relative to GDP) and market liquidity (value of trades relative to market capitalization). It is worth noting, however, that in the on-going debate on the merits of market-based versus bank-based financial systems, Levine (2002) concludes that it is overall financial development that
6
5 ultimately matters for economic growth, regardless of financial structure.7 For that reason, I will use both bank- and market-based financial indicators in the analysis.
Building on this literature, the seminal contribution by Rajan and Zingales (1998) has attempted to address the issue of reverse causality between financial development and economic growth using a difference-in-difference approach. The problem in traditional growth regressions is that finance could either foster growth, but at the same time growth could also foster financial development.8 To address this, the premise in Rajan and Zingales (1998) is that financial development lowers the cost of raising funds externally and finance should therefore disproportionately help firms typically dependent on external finance for growth. By considering this specific mechanism and adopting the strategy of interacting country-specific financial development with between-industry differences it is possible to establish causality more firmly. To test their hypothesis, the authors construct a measure of dependence on external finance by industry, assuming that industries have different dependencies on external finance for structural reasons. It is then postulated that if financial development fosters growth, industries that are highly dependent on external finance should grow faster in countries with well-developed financial systems, as this lowers the cost of external funds. The evidence indeed appears to be consistent with the hypothesis that industries with greater dependencies on external finance grow faster in countries with better developed financial systems. The study thus highlights one important channel through which financial development fosters growth: by lowering the cost of external finance.
Numerous other studies have built upon the Rajan and Zingales (1998) difference-in-difference framework. For instance, Beck (2003) provides evidence that financial development can be a source of comparative advantage for industries heavily reliant on external finance. Financial development could thus underpin international trade patterns. In addition, several firm-level studies focusing on the relation between financial development and financing constraints of firms broadly confirm the results by Rajan and Zingales (e.g. Demirguç-Kunt and Maksimovic, 1998; Beck et al., 2005).
As stressed before, however, none of them have considered the relationship between finance and growth using a more sophisticated decomposition of output growth, including a reallocation term derived directly from firm-level production functions. In light of the research documenting the importance of widespread firm-level heterogeneity and reallocation in accounting for aggregate productivity growth (e.g. Foster et al., 2001; Hsieh and Klenow, 2009; Syversen, 2011), this could be a relevant contribution of this paper.
7 The analysis by Levine (2002) is on a cross-section of 48 countries over the 1980-1995 period.
8 The traditional finance-growth tries to address this issue by using initial values for financial development.
6 Finally, while traditional theory and empirics have broadly demonstrated the positive ramifications of financial development for growth, it is important to consider several recent studies that have downplayed the role of finance in growth, lending support to financial development sceptics such as Robinson (1952).9 For instance, Manning (2003) argues that financial development has a greater impact on growth in non-OECD countries and, in particular, that the results of previous studies depend heavily on the performance of the Tiger economies during the 1980s.10 Furthermore, the study also concludes that it is difficult to disentangle the effect of finance from that of other correlated factors. Additionally, Rousseau and Wachtel (2011) also report that the – in the literature firmly entrenched – relation between finance and growth has weakened from the 1990s onward. They suggest that the impact of financial deepening on growth has diminished over time due to the incidence of credit booms and widespread financial liberalizations during the 1980s and 1990s, rendering financial development less effective. Thus, the finance-growth relation appears to be more disputed recently and these studies provide another impetus to link theory more closely to empirics to examine whether financial development has an impact on different sources of growth, which may be obscured in aggregate studies.
3. Methodology
To analyse the impact of financial development on growth in more detail and evaluate how the different sources of growth are affected by financial development, the Rajan and Zingales (1998) difference-in-difference approach will be adopted. This specification is used to account for endogeneity in the finance-growth relation. More formally, the economic model of interest is as follows:
, , = + + ℎ , + , + + , ∗ + , , (1)
where , represents industry growth rates and its components derived from firm-specific data, total
input growth, factor reallocation and technical change11. The dummy variable C controls for fixed country effects, T controls for year fixed effects, Share is the initial share of the industry j in total manufacturing in country i and is included to account for convergence. is the level of financial development in country i, indicates the dependence on external finance for industry j and , is
9
Robinson (1952) has famously claimed: “where enterprise leads, finance follows”.
10 Manning (2003) examines the datasets and methodologies in Levine and Zervos (1998) and Rajan and
Zingales (1998). The Tiger economies are Hong Kong, Singapore, South Korea and Taiwan.
11
7 an error term.12 Following Rajan and Zingales (1998), it is assumed that industries have different dependencies on external finance for technological reasons and that these differences persist across countries.13 The analysis is thus carried out at the industry level, where the impact of financial development is identified by exploiting between-country differences. Indeed, the interaction term
∗ is the variable of interest and it is hypothesised that δ is positive: an industry with a higher dependence on external finance grows comparatively faster in a country with a more developed financial system.14 In particular, it is expected that the impact of financial development is especially pronounced for the reallocation component, in line with the financial intermediation theory emphasizing the importance of finance for reallocation. It will be interesting to analyse whether the traditional link between finance and growth is confirmed using micro-level data and whether the relationship alters for varying sources of output growth, while it is also important to bear in mind the studies downplaying the role of finance more recently.
As discussed above, an important focus of the study is the decomposition of output growth at the industry level into three different sources: factor accumulation, technical change at the firm level and reallocation of resources across firms. To estimate equation (1), the ultimate research objective, industry output growth will thus first have to be decomposed into the three factors. This is done following a two-stage procedure. First, industry level production functions are estimated to derive the estimates of (i) output elasticities of the production factors and (ii) total factor productivity, which is normally defined as the residual in a production function estimation. Subsequently, the industry-specific output elasticities of the production factors are used in the decomposition of output growth into the three different sources of growth. Before discussing this decomposition in more detail in section 3.2, the production function estimation technique will be examined to illustrate the estimation issues and motivate the chosen estimation methodology.
12 Note that EF is time-invariant, in line with Rajan and Zingales (1998). This is done to make the marginal effect
analysis in section 5 more intuitive, as changes in EF then represent moving from one industry to another with a different external finance structure, rather than also moving within industries over time.
13
This is a key assumption in Rajan and Zingales (1998). It is argued that if an industry A in country X has a structural higher capital requirement and longer gestation period before cash flows are generated than industry B in country X, this will also hold in country Y. Empirically, robustness checks using different
benchmark countries did not significantly alter the results in their analysis. Furthermore, it is also stressed that much of the demand for external funds is likely to arise from technological shocks increasing an industry’s investment opportunities. To the extent that these shocks are worldwide, industry dependence on external finance can reasonably be assumed to converge across countries. This assumption, however, has been challenged by Fisman and Love (2007), who argue that external finance dependence is implicitly testing for industry growth opportunities, rather than ‘inherent’ debt structure.
14
8 3.1 Production function estimation
As a starting point in the production function estimation, it is assumed that production takes place according to a Cobb-Douglass production function, which has the advantage that it is easy to work with empirically and it can also easily be verified whether the estimated coefficients and returns to scale make economic sense. Specifically, a general production function looks as follows:
, = , , , , (2)
where Yi,t represents gross output of firm i in period t, Ki,t, Li,t, Mi,t are capital, labour and intermediate
inputs, respectively, and Ai,t is a measure of total factor productivity. Taking logs results in the
following linear production function:
, = + , + , + , + , (3)
where lower case letters represent natural logarithms and ln (A,it) = β0 + εi,t. The parameter β0
indicates the mean efficiency level across firms and over time, whereas εi,t represents the producer
and time specific element, which can be further decomposed into an observable productivity term (for the firm, but not for the econometrician) and unobservable error component, resulting in the following equation to be estimated:
, = + , + , + , + , + , (4)
where ωi,t = β0 + vi,t now represents firm-level productivity and μi,t is an error term. The total factor
productivity term ωi,t is typically estimated as the residual after accounting for the impact of factor
accumulation. However, estimating equation (4) involves dealing with several methodological issues, most notably the endogeneity of input choices, caused by the fact that vi,t can be observed by the
firm when inputs are chosen and, hence, the two variables are correlated .15 Therefore, the Wooldridge (2009) GMM method of the Levinsohn and Petrin (2003) estimator to account for the simultaneity of inputs and productivity at the firm-level will be used here. The main idea of the Levinsohn and Petrin (2003) strategy is briefly outlined in appendix 1 and the Wooldridge GMM
15
9 variant of this method will ultimately be used.16 It has been demonstrated that this approach addresses identification issues associated with the Levinsohn and Petrin method in case the variable input labour is also a deterministic function of unobserved productivity and state variables; the coefficient on labour would then be unidentified (Wooldridge, 2009). Furthermore, the Wooldridge GMM approach results in a more efficient robust estimator, taking into account serial correlation and heteroskedasticity.
Having obtained industry level output elasticities and total factor productivity estimates from this estimation, it is possible to move on to the second and final step leading to the estimation of equation (1): the decomposition of industry output growth into factor accumulation, technical change and reallocation.
3.2 Output growth decomposition
The industry-specific output elasticities obtained after the production function estimation will be used in the decomposition of output growth into its three different sources based on the exposition by Basu et al. (2009). Specifically, once the output elasticities β for the inputs are known, it is possible to decompose firm output growth into a change in inputs and technical change:17
= ∑ ∆ + ∆ (7)
for x = l, k, m. Crucially though, in the presence of firm heterogeneity, at the aggregate level it not only matters that total inputs increase, but also at which firms. As demonstrated by Basu et al. (2009), the reallocation of resources to firms with a larger mark-up is also important and, denoting the firm-specific cost share of each input by c and its output elasticity by β, it is possible to rewrite output growth for firm i as follows:
∆ = ∑ ∆ + ∑ [( − )∆ ] + ∆ (8)
for x = l, k, m. The first term on the right hand side in equation (8) represents the contribution of input changes to output growth. More interestingly, the second term represents the reallocation factor, as it compares the marginal product to marginal cost for each input and moving inputs from low mark-up to high mark-up firms improves allocation. The last term in equation (8) represents the
16
A more technical exposition can be found in Wooldridge (2009).
17 Note that the industry-specific output elasticities β and the technical change term a (which is simply the
10 contribution of technical change to output growth. Furthermore, note that these growth components are now all defined at the firm-level, whereas the estimation of the main model of interest, equation (1), is specified at the industry-level. Therefore, the growth components still need to be aggregated to the industry level, which can be easily done by an appropriately weighted sum over firms i for each growth component. Following Hulten (1978), it can be shown that aggregate output growth is given by:
∆ = ∑ ( ∆ ) (9)
with the weights wit equalling the firm’s so-called Domar weight, the two-period average ratio of a
firm’s nominal output over aggregate value added.18 The Domar weight thus determines each firm’s contribution to aggregate output growth and applying this weighting methodology aggregates the decomposition of output growth at the industry level into (i) a factor accumulation factor, (ii) a reallocation term and (iii) a technical change factor:
∆ = ∑ ( ∑ ∆ ) (8i)
∆ = ∑ ( ∑ [( − )∆ ]) (8ii)
∆ = ∑ ( ∆ ) (8iii)
These variables are now all defined at the industry level and, hence, they can be used as dependent variables in the estimation of equation (1) alongside aggregate industry output growth ∆yt. As
already underlined in the introduction, this decomposition of output growth is important because it relates more directly to the specific channels through which financial development is purported to stimulate growth: reallocation of resources and/or facilitating technical change.
3.3 The main model revisited
All the relevant variables for the difference-in-difference model specified in equation (1) have now been derived and the analysis can be carried out for the different dependent variables (∆y, ∆x, ∆a, ∆r) using a panel estimator at the country-industry level, controlling for industry-country and year fixed-effects. Clustered standard errors by industry-country are used to deal with potential problems of heteroskedasticity and autocorrelation.
11
4. Data
4.1 Data sources
The baseline difference-in-difference analysis specified in equation (1) will be conducted for a panel data of 22 industries in 9 European countries (Austria, Belgium, France, Germany, The Netherlands, Finland, Spain, Italy and Ireland) representing an important part of the Eurozone economy with highly differing economic outcomes, for the period 1997-2006. The focus is on these countries as this comprises a relatively homogeneous sample of developed countries, reducing the risk that potentially unobserved institutional factors drive the results.19 The choice is also motivated by data availability concerning the firm-level variables required for the production function estimations, as only these Eurozone countries are well covered in this regard.
The data for the relevant variables is derived from several sources. First, consider the firm-specific variables required for the production function estimation in equation (4). For this, data is required on gross output Y, labour L, capital K and intermediate inputs M. Data for these variables is collected from AMADEUS, a pan-European database developed by Bureau van Dijk Electronic Publishing (BvDEP). This database contains financial and structural information on approximately 9 million private and public non-financial firms. For the present study, data on 1.614.030 firms, ranging from small private firms to large MNEs, spread over 27 industries and 9 countries over the 1996-2006 period is collected to be used in the production function estimation.20 Following this estimation, it will be motivated why the sample is eventually reduced to 22 industries in section 4.3.
From the AMADEUS database the following variables are extracted to be used in the production function estimation and output decomposition: Operating Revenue / Turnover, Fixed Assets, Employees, Working Capital and Cost of Employees. Operating Revenue is used as a measure for firm output Y, Fixed Assets is used as a measure for capital inputs K, Employees, which represents the number of employees, is used as a measure for labor inputs L, while Working Capital is used as a proxy for intermediate inputs M.21 It is assumed that working capital also responds smoothly to productivity shocks in the same way as intermediate inputs, making it a valid proxy for unobserved productivity. The Cost of Employees variable is used to calculate the cost share of labour in the
19
Some authors have suggested that financial development is only effective once complementary institutions are in place, such as the rule of law or quality of government (e.g. Rousseau and Wachtel, 2011). By using a relatively homogeneous sample of developed countries, the risk that these unobserved factors will drive the results is reduced.
20 Note that initially data is collected over the 1996-2006 period. Once growth rates are constructed the period
covered for the analysis will be 1997-2006.
21
12 output decomposition specified in equation (8).22 During the data management process, the dataset is cleaned to include only strictly positive values for the relevant financial variables, as negative, zeros and/or missing values are incompatible with the Wooldridge production function estimation technique.23 To deal with outliers, only the data between the 1st and 99th percentile will be used. Furthermore, the financial variables are deflated and standardized to 1996 Euros using industry price indices from EU KLEMS.24
Additionally, concerning the data for industry dependence on external finance, the measure for EF is based on the Rajan and Zingales (1998) methodology. They approximate an industry’s optimal capital structure using the amount of external finance for large listed U.S. firms, assuming that the U.S. financial markets are relatively frictionless, making for a good benchmark. Here, UK firm-level data from Amadeus will be used as a benchmark for industry dependence on external finance, under the assumption that the UK financial market can also be viewed as relatively frictionless. Dependence on external finance is measured as the average share of debt in total assets.25
Data on financial development will come from the Beck and Demirguç-Kunt (2009) dataset on financial indicators. The financial indicators that I will use are (i) the total liquid liabilities relative to GDP, which is a measure of overall financial depth; (ii) private credit by deposit money banks to GDP, which measures bank development; (iii) stock market total value traded relative to GDP, measuring stock market activity; and (iv) stock market turnover, which equals the ratio of value of shares traded to market capitalization and is a measure of stock market liquidity. These indicators are most commonly used to measure financial development in the literature (Levine, 2005).26
4.2 Data description production function variables
A brief overview of the variables used in the production function estimation is provided in table 1, while the mean (log) values over time for these variables are shown in figure 1. As can be observed in
22 The cost share of intermediates can be derived from the financial variables directly, whereas the capital
share is defined as the residual of labour and intermediate factor shares assuming constant returns to scale, in line with standard growth accounting. The validity of this assumption will be explored in the robustness analysis.
23 Note that this may induce a selection bias, however, it is necessary for production function estimation to be
feasible.
24
Ideally, one would deflate using firm-level prices, rather than industry-level. However, firm-level prices are unattainable, which is why industry price indices are used, under the assumption that firms within an industry face the same prices (Van Beveren, 2012).
25
Calculations are based on approximately 2.6 million firm-year observations over the 1996-2004 period.
26 While these indicators have become the standard in empirical work on financial development, it is important
13 figure 1, the mean values for all variables increase in 1997, before slightly decreasing over time. Finally, more detailed summary statistics with a decomposition per industry for the production function variables are presented in table 2.
Table 1: Overview of production function variables
Variable name Brief description Used for
Operating Revenue (Y) Operating Revenue/Turnover in 1000s of Euros
Measure for Y in production function estimation
Fixed Assets (K) Fixed Assets in 1000s of Euros Measure for K in production function
estimation
Intermediates (M) Working Capital in 1000s of Euros Proxy for intermediates M in production function estimation
Employees (L) Number of Employees Measure for L in production function
estimation 2 3 4 5 6 7 lo g s 1996 1998 2000 2002 2004 2006 year
Revenue (Y) Labour (L)
Capital (K) Intermediates (M)
14
Table 2: Summary statistics for revenue, labour, capital and intermediates
Revenue Labour Capital Intermediates
Industry mean sd mean sd mean sd mean sd N
(1) Agriculture, hunting, forestry and fishing 6.14 1.56 1.64 1.22 5.22 1.87 4.71 1.78 138806
(2) Mining and quarrying 7.09 1.70 2.41 1.29 6.44 2.11 5.63 1.83 27965
(3) Food products, beverages and tobacco 7.35 1.91 2.47 1.41 6.01 2.02 5.54 2.23 158055
(4) Textiles, textile products, leather and footwear 7.19 1.67 2.57 1.34 5.38 1.99 5.70 1.85 161052
(5) Wood and products of wood and cork 6.70 1.58 2.21 1.19 5.33 1.81 5.30 1.67 74111
(6) Pulp, paper, paper products, printing and publishing 6.78 1.77 2.20 1.43 5.19 2.19 5.14 1.87 153813
(7) Coke, refined petroleum products and nuclear fuel 9.14 2.49 3.45 1.87 8.00 2.72 6.90 2.58 2034
(8) Chemicals and chemical products 8.29 2.17 3.13 1.75 6.76 2.55 6.79 2.15 57331
(9) Rubber and plastics products 7.69 1.69 2.85 1.40 6.18 2.02 5.99 1.76 70234
(10) Other non-metallic mineral products 7.28 1.71 2.64 1.38 5.84 2.11 5.80 1.84 86460
(11) Basic metals and fabricated metal products 7.04 1.61 2.54 1.28 5.27 1.99 5.47 1.75 305678
(12) Machinery, nec 7.54 1.65 2.75 1.47 5.25 2.11 5.99 1.90 149597
(13) Electrical and optical equipment 7.54 1.80 2.67 1.58 5.46 2.18 5.88 2.06 110361
(14) Transport equipment 7.82 2.02 3.03 1.73 5.57 2.41 6.19 2.14 40733
(15) Manufacturing nec; recycling 6.82 1.62 2.28 1.29 4.82 1.90 5.35 1.71 121439
(16) Electricity, gas and water supply 8.05 2.59 2.66 2.06 7.57 3.06 6.20 2.52 19077
(17) Construction 6.21 1.41 1.93 1.16 3.75 1.81 4.73 1.81 978294
(18) Sale, maintenance and repair of motor vehicles 6.89 1.70 1.83 1.14 4.57 1.83 4.93 1.78 352908
(19) Wholesale trade and commission trade, except motor vehicles 7.23 1.68 1.81 1.25 4.64 1.99 5.55 1.78 951065
(20) Retail trade, except of motor vehicles and motorcycles 6.12 1.38 1.43 1.09 4.10 1.86 4.32 1.54 701086
(21) Hotels and restaurants 5.73 1.30 1.91 1.19 4.41 1.95 3.22 1.71 227324
(22) Transport and storage 6.64 1.68 2.14 1.39 4.89 2.02 4.55 1.80 307352
(23) Post and telecommunications 6.75 2.16 2.23 1.70 4.64 2.63 4.93 2.24 15347
(24) Financial intermediation 6.24 2.19 1.58 1.59 5.15 2.85 4.59 2.54 53833
(25) Real estate activities 5.76 1.90 1.14 1.20 5.27 2.49 5.09 2.40 337735
(26) Renting of machinery & equipment and other business activities 6.05 1.81 1.80 1.56 4.34 2.34 4.42 2.04 836547
(27) Other community, social and personal services 5.86 1.73 1.77 1.37 4.83 2.18 3.62 2.13 171428
Total 6.57 1.76 1.94 1.37 4.67 2.16 4.92 1.99 6609665
4.3 Production function estimation results
The production function specified in equation (4) is estimated using the Wooldridge variant for the Levinsohn and Petrin (2003) technique. The results for the production function estimation are provided in tables A1-3 in appendix 2. Each column reports the estimated coefficients per industry. The output elasticities can be obtained from the variables on l, k and m, for labour, capital and intermediate inputs, respectively. The remaining variables are the instruments used in the estimation and consist of lags and interactions of capital and intermediates.
A closer inspection of the output elasticities will help to evaluate which industries can be included in the estimation of equation (1). In case the coefficients are economically implausible for the three main production input variables (labour, capital and intermediates), using these parameter estimates for the decomposition of output growth may be invalid. From the estimation results in table A3, it can be inferred that the output elasticities for intermediates and capital are negative for the industries ‘Hotels and restaurants’ (industry 21) and ‘Real estate activities’ (industry 25) respectively. As it is highly unlikely in practice for output elasticities to be negative, the output decompositions on the basis of these industry coefficients may be unreliable. These industries will therefore be excluded from the subsequent analysis.
16
Table 3: Output elasticities, total returns to scale and external finance dependence per industry
Industry Labour Capital Intermediates Total EF
βl βk βm
(1) Agriculture, hunting, forestry and fishing 0.54 0.16 0.01 0.70 -0.03
(2) Mining and quarrying 0.49 0.18 0.08 0.75 0.78
(3) Food products, beverages and tobacco 0.63 0.15 0.11 0.89 -0.53
(4) Textiles, textile products, leather and footwear 0.38 0.16 0.16 0.70 -1.03
(5) Wood and products of wood and cork 0.64 0.11 0.14 0.88 -0.38
(6) Pulp, paper, paper products, printing and publishing 0.61 0.10 0.13 0.84 -0.78
(8) Chemicals and chemical products 0.47 0.10 0.26 0.83 4.42
(9) Rubber and plastics products 0.60 0.16 0.12 0.88 -0.13
(10) Other non-metallic mineral products 0.49 0.17 0.10 0.75 -0.24
(11) Basic metals and fabricated metal products 0.55 0.18 0.08 0.81 -0.25
(12) Machinery, nec 0.55 0.09 0.13 0.78 0.04
(13) Electrical and optical equipment 0.54 0.13 0.15 0.82 0.97
(14) Transport equipment 0.54 0.14 0.22 0.91 0.05
(15) Manufacturing nec; recycling 0.45 0.19 0.11 0.75 -0.53
(17) Construction 0.57 0.13 0.09 0.78 1.19
(18) Sale, maintenance and repair of motor vehicles 0.49 0.19 0.16 0.84 0.24
(19) Wholesale trade and commission trade, except motor vehicles 0.46 0.14 0.14 0.74 0.32
(20) Retail trade, except of motor vehicles and motorcycles 0.64 0.12 0.07 0.83 0.06
(22) Transport and storage 0.61 0.01 0.09 0.71 0.01
(23) Post and telecommunications 0.45 0.09 0.26 0.80 0.19
(26) Renting of machinery & equipment and other business
activities 0.53 0.09 0.18 0.79 0.72
Average 0.53 0.13 0.13 0.80 0.21
Notes: Table 3 shows the industry technology parameters for the 22 industries included in the analysis. Coefficient estimates for output elasticities are based on the Wooldridge (2009) GMM estimation and the original sample consists of 6.609.665 observations (1.614.030 firms) for 1996-2006. All coefficients are statistically significant, with the exception of βm for industry 1. External finance dependence (EF) shows the period-average share of debt in total
17
4.3 Data description for output growth decomposition and main model (1) variables
Having considered the production function variables and industry technology parameters, this section will discuss the output growth decomposition and financial development data, which will be used to estimate equation (1). First, table 4 reports firm-level descriptive statistics for the variables used in the decomposition of output growth according to equation (8). The table indicates that the dispersion in the data is considerable, with extreme minimum and maximum values. In order to deal with this problem of potential outliers, the data will be winsorized at the 1st and 99th percentile.
Table 4: Descriptive firm-level statistics 1997-2006
Variable Mean SD Min Max
Domar weight (%) 0.01 0.23 -13.00 111.93 Labour share 0.25 0.16 -0.06 0.97 Intermediates share 0.22 0.15 0.00 0.98 Capital share 0.53 0.21 0.00 1.02 Revenue growth 0.07 0.29 -8.27 9.04 Labour growth 0.04 0.37 -12.57 17.30 Capital growth 0.07 0.53 -10.25 10.63 Intermediates growth 0.09 0.70 -9.65 9.71
Notes: The Domar weight represents the mean contribution of each firm to aggregate output and is defined as the two-period average ratio of a firm’s nominal output over aggregate value added. Average shares show rolling two-year averages of each input’s cost to revenue ratio.
18
Table 5: Descriptive statistics for growth components, external finance and financial indicators
Variable Mean SD Min Max N
Output growth (%) ∆y 0.90 2.35 -12.46 17.34 1517
Input growth (%) ∆x 0.77 2.13 -9.49 14.24 1517
Technical change (%) ∆a 0.48 1.95 -10.67 13.47 1517
Reallocation term (%) ∆r -0.34 1.49 -10.32 8.89 1517
External finance dependence EF 0.21 1.08 -1.03 4.42 1517
Liquid liabilities/GDP FD1 0.78 0.20 0.47 1.16 1390
Private credit/GDP FD2 0.92 0.29 0.50 1.67 1390
Stock market capitalization/GDP FD3 0.80 0.41 0.24 2.66 1517
Stock market turnover FD4 0.97 0.50 0.21 2.26 1517
Notes: Table 5 shows the descriptive statistics for all variables used to estimate equation (1). The sample comprises 22 industries for 9 countries over the 1997-2006 period. ∆y is aggregate growth, ∆x is input growth as in equation (8i), ∆r is reallocation as in equation (8ii) and ∆a is technical change as in equation (8iii). EF represents dependence on external finance based on the average share of debt in assets for UK industries derived from Amadeus. FD1-4 are different financial development indicators from the Beck and Demirguç-Kunt
(2009) dataset.
Again, it can be observed that input growth is the main contributor to overall output growth for most industries, with the exception of sectors 16 and 23, for which technical change is the biggest driver
-. 0 05 0 .0 05 .0 1 .015 1 2 3 4 5 6 8 9 10 11 12 13 14 15 16 17 18 19 20 22 23 26
Output growth Input growth
Technical change Reallocation
19 behind output growth. Finally, it can be inferred that the reallocation term is not only negative on average overall as seen in table 5, but reallocation is also negative across all industries. While this is surprising at first glance, indicating that reallocation has a negative impact on overall output growth, several factors can help explain this finding. As shown in equation (8ii), reallocation is negative in case the cost share ci is higher than the output elasticity βi, by construction. Therefore, the
reallocation term could be affected by a systematic downward bias in the beta coefficients or an upward bias in the cost shares. Now, recall that the share of capital is simply calculated as the residual after accounting for the labour and intermediate inputs shares, assuming constant returns to scale. However, as reported in table 3, the total returns to scale in the sample are decreasing at approximately 0.80, rather than 1 for constant returns to scale. Hence, a systematic upward bias in the cost share of capital is likely to be present, resulting in the negative reallocation term across industries. Therefore, the decomposition of output growth is also performed using 0.80 as total returns to scale, as indicated by the results in table 3. The growth contributions per industry using this altered cost share of capital are presented in figure 3 in appendix 3. It can be observed that, while still negative for most industries, the reallocation component is now positive for industry 19 (Wholesale trade and commission trade). Furthermore, by construction the reallocation term has become more important at the expense of input growth.
Still, as the potential bias in the cost share does not alter the relative position of reallocation, the finding that reallocation is negative overall does not necessarily need to be problematic for the analysis, as differences in within-sector reallocation across industries can still be influenced by the interaction of financial development and external finance dependence. That is, a positive effect of the interaction term in equation (1) still represents an improvement in the allocation of resources and is thus beneficial for the economy. Nevertheless, a robustness check will investigate whether the results could be driven by the construction of the reallocation term by re-estimating equation(1) using the alternative output growth decomposition, assuming total returns to scale of 0.80.
5. Estimation results and discussion
This section will report and discuss the estimation of the baseline difference-in-difference model specified in equation (1), which is reproduced below:
, , = + + ℎ , + , + + , ∗ + , , (1)
20 focus of the analysis will be on the FD and EF variables and, crucially, the interaction between the two. Concerning the FD variable, it is expected that this variable will have a positive impact on all growth components, in line with the traditional literature arguing that financial development is beneficial for overall growth. The direct effect of dependence on external finance EF is ambiguous from a theoretical point of view, as it simply represents an industry-specific technology structure. Industries with higher dependencies on external finance could either grow faster, slower or at the same pace as other industries. More importantly, however, is the interaction term, which is the main variable of interest following Rajan and Zingales (1998). Indeed, it is hypothesised that δ coefficient for the interaction term is positive: an industry with higher dependence on external finance grows comparatively faster in a country with a more developed financial system. Moreover, it is expected that the impact of financial development is particularly pronounced for the reallocation component, as this is a key channel through which the financial system is hypothesized to enhance growth.
The baseline regression results for the estimation of equation (1) are given in tables 6 and 7 below, including conditional effects of FD on different levels of external finance dependence in the bottom panel. First, it can be inferred from the results that the direct effect of dependence on external finance does not appear to affect aggregate output growth, although it does seem to positively and significantly affect the reallocation component (table 6, columns 4 and 8), while negatively affecting input factor accumulation when private credit to GDP is used as a financial development indicator (column 6). The direct impact of financial development on aggregate growth and its components does not appear to be significantly different from zero, with a positive effect on reallocation only when stock market turnover is used as an indicator for financial development. As can be easily observed from the tables, this finding is not robust across different financial development indicators. Furthermore, the direct impact of financial deepening does not adequately control for endogeneity in the relationship between finance and growth, which is why the estimation is specified as a difference-in-difference model with an interaction term.
21 The results suggest that financial development has a significant positive impact on reallocation when stock market turnover is used as financial indicator, as can be observed from column 8 in table 7. All the other marginal effects are insignificant. Concerning the economic meaning of the direct impact of FD4, a one standard deviation (0.5) unit increase in stock market
turnover would increase reallocation by approximately 0.17%.27 While this may seem small, recall that mean output growth is only 0.90% in the sample. Indeed, moving from the mean value of stock market turnover to the maximum value in the sample would increase reallocation by 0.43%, accounting for approximately 18% of the variation in output growth. It can thus be tentatively concluded that countries with greater financial development foster reallocation and this could constitute a reasonable part of aggregate output growth.
However, this result will have to interpreted with caution, as a closer inspection of the results indicate that the positive marginal effects are primarily driven by the direct impact of FD4 on
reallocation and the results do not significantly change for the different levels of external finance dependence.28 Recall that the direct impact may not adequately control for endogeneity between finance and growth, which is precisely why equation (1) is specified as a difference-in-difference model using an interaction term, which remains insignificant. It is expected that financial development would be more important for industries with higher dependence on external finance, in line with Rajan and Zingales (1998), but this is not confirmed by the results. It can thus not be safely concluded that financial development is stimulating reallocation. The overall findings of the analysis, therefore, do not indicate that financial development has a positive impact on aggregate growth or its components. Indeed, even the influence of financial development on a key and theoretically more valid channel through which finance could foster growth, reallocation, does not appear to be significant and robust.
27 Standard deviations can be found in table 5. 28
22
Table 6: Estimation results for equation (1) using overall and bank-based financial development indicators
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent: ∆y ∆x ∆t ∆r ∆y ∆x ∆t ∆r
EF -0.0015 -0.0028 -0.0011 0.0025* -0.0015 -0.0029** -0.0005 0.0019* (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) FD1 0.0006 0.0123 -0.0087 -0.0046 (0.009) (0.010) (0.008) (0.007) FD1*EF 0.0023 0.0037 0.0019 -0.0035* (0.002) (0.002) (0.002) (0.002) FD2 0.0007 0.0062 -0.0036 -0.0009 (0.005) (0.006) (0.005) (0.004) FD2*EF 0.0020* 0.0033** 0.0010 -0.0023* (0.001) (0.002) (0.001) (0.001) Observations 1,390 1,390 1,390 1,390 1,431 1,431 1,431 1,431 R-squared 0.127 0.113 0.082 0.059 0.121 0.113 0.080 0.061
Effects of FD conditional on EF:
Percentiles of EF EF=-0.78 (5th) -0.00126 0.00939 -0.0102 -0.00183 -0.00093 0.00362 -0.00439 0.000864 se 0.00913 0.00944 0.00882 0.00731 0.00543 0.00563 0.00509 0.00391 EF=-0.25 (25th) -0.00001 0.0114 -0.00916 -0.00371 0.00015 0.00537 -0.00386 -0.000361 se 0.00909 0.00937 0.00838 0.00731 0.00554 0.00568 0.00541 0.00380 EF=0.01 (50th) 0.000604 0.0124 -0.00865 -0.00465 0.000684 0.00624 -0.00359 -0.000968 se 0.00912 0.00953 0.00841 0.00794 0.00557 0.00576 0.00500 0.00381 EF=0.32 (75th) 0.00132 0.0135 -0.00806 -0.00572 0.00130 0.00724 -0.00329 -0.00167 se 0.00921 0.00969 0.00848 0.00745 0.00545 0.00588 0.00516 0.00381 EF=1.19 (95th) 0.00335 0.0167 -0.00639 -0.00879 0.00306 0.0101 -0.00243 -0.00366 se 0.00969 0.0104 0.00839 0.00735 0.00543 0.00561 0.00505 0.00405
Notes: Table 6 shows the baseline regression results for a sample of 22 industries, 9 countries 10 years (1997-2006), using total liquid liabilities to GDP (FD1) and private credit by deposit money banks to GDP (FD2) as financial development
23
Table 7: Estimation results for equation (1) using stock market based financial development indicators
(1) (2) (3) (4) (5) (6) (7) (8)
Dependent: ∆y ∆x ∆t ∆r ∆y ∆x ∆t ∆r
EF 0.0009 0.0003 0.0009 -0.0004 0.0006 0.0002 0.0000 0.0001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) FD3 -0.0026 -0.0021 0.0012 -0.0017 (0.003) (0.003) (0.002) (0.002) FD3*EF -0.0006 -0.0002 -0.0007 0.0002 (0.001) (0.001) (0.001) (0.001) FD4 -0.0014 -0.0019 -0.0020 0.0034** (0.002) (0.002) (0.002) (0.002) FD4*EF -0.0003 -0.0001 0.0004 -0.0004 (0.001) (0.001) (0.001) (0.001) Observations 1,517 1,517 1,517 1,517 1,517 1,517 1,517 1,517 R-squared 0.126 0.115 0.079 0.062 0.125 0.115 0.079 0.064
Effects of FD conditional on EF:
Percentiles of EF EF=-0.78 (5th) -0.00209 -0.00191 0.00170 -0.00189 -0.00118 -0.00187 -0.00228 0.00372** se 0.00323 0.00266 0.00155 0.00174 0.00210 0.00206 0.00190 0.00163 EF=-0.25 (25th) -0.00243 -0.00202 0.00135 -0.00178 -0.00133 -0.00192 -0.00206 0.00354** se 0.00286 0.00247 0.00160 0.00188 0.00200 0.00202 0.00166 0.00153 EF=0.01 (50th) -0.00260 -0.00208 0.00118 -0.00172 -0.00140 -0.00194 -0.00196 0.00344** se 0.00270 0.00294 0.00159 0.00171 0.00230 0.00214 0.00172 0.00161 EF=0.32 (75th) -0.00279 -0.00214 0.000975 -0.00165 -0.00149 -0.00197 -0.00184 0.00334** se 0.00253 0.00328 0.00165 0.00170 0.00196 0.00234 0.00168 0.00156 EF=1.19 (95th) -0.00335 -0.00233 0.000405 -0.00146 -0.00173 -0.00205 -0.00148 0.00303* se 0.00220 0.00280 0.00182 0.00183 0.00207 0.00213 0.00187 0.00179
Notes: Table 7 shows the baseline regression results for a sample of 22 industries, 9 countries and 10 years (1997-2006), using stock market capitalization to GDP (FD3) and stock market turnover (FD4) as financial development indicators. EF
represents dependence on external finance. Initial share variable and fixed effects for countries and years are included but not reported. The bottom panel shows marginal effects conditional on different levels of dependence on external finance. Clustered standard errors by industry-country are in parentheses. *** p<0.01, ** p<0.05, * p<0.1 denote significance.
24
6. Robustness analysis
As an initial check on the results presented in the previous section, model (1) will be re-estimated for manufacturing industries only, in line with the analysis of Rajan and Zingales (1998), who also restrict their analysis to manufacturing industries. While the benefits of financial development need not be strictly limited to manufacturing, it may very well be that the link between finance and growth is more pronounced for conventional industrial sectors, as opposed to services or retailing. However, the results for the analysis using manufacturing industries only do not alter the main findings of section 5: financial development does not appear to foster growth in sectors with a higher dependence on external finance.29
Furthermore, equation (1) is re-estimated using the alternative cost share of capital, assuming total returns to scale of 0.80 in the decomposition of output growth presented in figure 3.30 While this does not qualitatively alter the results, the impact of the interaction terms on reallocation in table 6 do become ‘less negative’ using the alternative decomposition method, indicating that these results may be driven by the fact that reallocation is negative for all industries in the initial sample. The marginal effects analysis, however, remains unchanged.
Next, the robustness of the results will be examined by considering the production function estimation discussed in section 4.3 In case a systematic bias is present in the industry-specific output elasticities, the subsequent analysis of the reallocation term will be biased too. Indeed, from table 3, it can already be inferred that the total returns to scale are slightly below 1. Although the technology parameters appear to be plausible from an economic perspective, constant or even increasing returns to scale could be expected in a sample with many small firms. Therefore, different production function estimations will be conducted in order to examine how this affects the results.
First, as pointed out by Levinsohn and Petrin (2003), a potential problem in the Wooldridge production function estimation could arise from the proxy for intermediate inputs, for which no clearly measurable variable is available. I have used working capital as a best available approximation to intermediates, assuming that working capital responds smoothly to productivity shocks in the same way as intermediate inputs. On the other hand, Basu et al. (2009) suggest using an alternative proxy from the Amadeus database: Revenue – (Operating Profit + Labour Cost + Depreciation). They assume that Operating Profit + Labour Cost + Depreciation is a good approximation of value added in production and that Revenue is a good approximation of the total value of production. Therefore, the production function is re-estimated using this alternative proxy for intermediate inputs. The resulting output elasticities, however, lead to implausibly high beta coefficients for intermediates, whereas the coefficients for labour and capital are extremely low. Furthermore, average returns to scale
29
The unreported regression results are available upon request.
25 across all industries are approximately 0.75, which deviates further from constant returns to scale compared to the baseline Wooldridge estimation using working capital as a proxy for intermediates (overall total returns to scale were 0.80 in that case).31 For these reasons working capital is preferred over the alternative measure and the subsequent analysis is therefore not carried out using the newly estimated betas. Nevertheless, it is important to bear in mind that working capital remains only a best possible approximation to intermediate inputs in the production function specification.
Another issue in production function estimation concerns the applied methodology, as multiple research methods are available to estimate the production relation (see, for instance, Van Beveren, 2012). Here, the Wooldridge (2009) GMM variant of the Levinsohn and Petrin (2003) estimation technique has been adopted, in order to account for the unobserved simultaneity between productivity and input choices. While it is important to consider this correlation from a theoretical perspective – and this has played a significant role in the motivation for choosing the Wooldridge estimation technique – the empirical implementation of the Wooldridge technique is not without difficulties, for instance, with the choice of instruments to be included (Wooldridge, 2009). To evaluate whether the choice of methodology to estimate the industry-specific output elasticities has an impact on the findings for the estimation of equation (1), the analysis will be conducted using a general OLS production function estimation. The results for this robustness test can be found in tables A4 and A5 in appendix 3. A comparison with the baseline findings presented in section 5 shows that the changes in results are marginal: only the impact of the overall interaction term of FD2*EF on aggregate growth and the direct impact of stock market turnover (FD4) lose their
significance. The marginal effects of financial development conditional on different levels of external finance dependence, the main focus of the analysis, do not materially differ. This finding thus seems to suggest that the choice of a specific production function estimation methodology is not driving the results.
In addition to potential issues with production function estimation, the specification of equation (1) will be examined. Note that the model has been specified as a panel, whereas Rajan and Zingales (1998) adopt a cross-section analysis for 41 countries over the 1980s. The reason for this is that, due to data availability problems of production function variables, the model can only be estimated for a relatively small number of countries. Specifying a cross-section model would therefore lead to a considerable loss of degrees of freedom. Nevertheless, it is interesting to examine whether this specification may alter the result. Indeed, Bertrand et al. (2003) has argued that conventional difference-in-difference models do not adequately account for serial correlation and a proposed remedy of this issue is to collapse the data over time. The results for the cross-section
31
26 analysis of equation (1) are presented in tables A6 and A7 in appendix 3. The main marginal effects analysis remains unchanged, confirming the baseline findings, although one interesting result can be found. Stock market turnover (FD4) appears to have a positive and significant marginal impact on
aggregate industry growth, factor accumulation and technical change. The impact on reallocation, surprisingly, turns out to be negative, suggesting that while financial development may on aggregate be beneficial to growth through its impact on input growth and technical change, it may actually hamper within-industry reallocation. However, the marginal effects again do alter for different levels of EF and, as already previously discussed, the direct impact does not adequately account for endogeneity. Furthermore, it has to be stressed that this finding is not robust across financial development indicators and the conditional marginal effects remain insignificant for other measures of financial deepening.
So far, the analyses seem to confirm the baseline findings in that there appears to be no significant and robust link between financial development, growth and reallocation, conditional on the industry dependence on external finance. As a final check on the results, I replicate the baseline Rajan and Zingales (1998) estimation as closely as possible using the sample in the study, to examine whether the reported link between finance and growth can be found using aggregate industry growth collected directly from a macro database, as opposed to deriving aggregate growth from micro data. For this purpose, industry growth data will be collected directly from the EU KLEMS database. The results for this estimation are presented in table A8 in appendix 3. Again, no significant impact of financial development on aggregate industry growth is found, consistent with the baseline regressions results presented in section 5. The fact that the findings in Rajan and Zingales (1998) cannot be confirmed for the current sample using a specification closely resembling the one adopted in their study could suggest that the importance of the link between financial development and growth – as reported in the literature – may have diminished. This would be in line with the findings of Rousseau and Wachtel (2011), who show that the influence of financial sector development on economic growth has considerably weakened over time. One potential and relevant explanation they provide for the diminished importance of finance for growth is that the widespread financial expansion of the 1980s and 1990s could lead to credit booms, in effect making financial deepening less effective. The results in my analysis lend support this hypothesis, as no significant, robust impact of financial development on growth has been reported in a sample of 9 developed Eurozone countries, not even for the more theoretically motivated reallocation component.
7. Conclusion
27 By decomposing output growth into input factor accumulation, technical change and reallocation it is possible to analyse how different sources of growth are affected by financial deepening. An important advantage of this strategy is that this specification explicitly identifies a reallocation term, which is a key channel through which finance is purported to stimulate growth from a theoretical perspective.
The findings seem to indicate that the positive link between financial development and economic growth reported in the traditional literature (e.g. Rajan and Zingales, 1998; King and Levine, 1993; Levine and Zervos, 1998) is not as strong as initially thought, at least for this smaller sample of 9 Eurozone countries. No robust significant marginal effects between measures of financial deepening, conditional on external finance dependence, have been found. Rather, the analysis appears to lend support to the hypotheses of financial development ‘sceptics’, such as Rousseau and Wachtel (2011), who argue that the importance of financial deepening for economic growth has diminished over time.
However, it is important to bear in mind several caveats of my analysis. First, considering the data, production function estimations have to deal empirical with several empirical problems, such as the issue of multi-product firms and how to measure inputs (Syversen, 2011). Indeed, data availability restrictions mean that the measures used in the empirical production function estimation are only best approximations to the actual variables suggested by theory. In particular, the cost shares of each input are hard to measure empirically and more detailed data on firm expenditures on intermediates and capital services could improve the reallocation decomposition. In a similar vein, the financial development indicators are also merely imperfect measures of financial development. Furthermore, the assumption that the dependence on external finance is determined by industry-specific technology, and thus identical across countries, may be invalid. However, robustness checks in previous papers (e.g. Rajan and Zingales, 1998) using different benchmark countries for external finance dependence indicate that this is not driving the results.
In addition to the data limitations, it is also important to recall that the research in this paper addressed within-industry reallocation, as opposed to across-industry reallocation. Although the literature has reported the importance of within-sector heterogeneity and it is therefore relevant to consider whether the financial sector is able to improve the allocation of resources within industries, it might very well be that financial development still fosters reallocation across industries, by exploiting the potential in growing sectors.
29
References
Aghion, P., T. Fally and S. Scarpetta (2007), “Credit constraints as a barrier to the entry and post- entry growth of firms”, Economic Policy, 22: 731–779.
Basu, S., L. Pascali, F. Schiantarelli, and L. Serven (2009), “Productivity, welfare and reallocation: Theory and firm-level evidence”, NBER Working Paper No. 15579.
Beck, T. (2003), “Financial Dependence and International Trade”, Review of International Economics, 11(2): 296-316.
Beck, T. and A. Demirgüç-Kunt (2009), "Financial Institutions and Markets Across Countries and over Time: Data and Analysis", World Bank Policy Research Working Paper No. 4943.
Beck, T., A. Demirgüç-Kunt, and V. Maksimovic (2005), “Financial and Legal Constraints to Firm Growth: Does Size Matter?”, Journal of Finance, 60(1): 137-177.
Bertrand, M., E. Duflo and S. Mullainathan (2004), "How Much Should We Trust Differences-in- Differences Estimates?", Quarterly Journal of Economics, 119(1), 249-275.
Boyd, J. H. and E. C. Prescott (1986), "Financial Intermediary-Coalitions”, Journal of Economic Theory, 38(2): 211-232.
Claessens, S. and L. Laeven (2003), “Financial Development, Property Rights, and Growth”, Journal of Finance, 58(6): 2401-2436.
Demirgüç-Kunt, A. and V. Maksimovic (1998), "Law, Finance, and Firm Growth”, Journal of Finance, 53: 2107-2137.
Fisman, R. and I. Love (2007), “Financial Dependence and Growth Revisited”, Journal of the European Economic Association, 5(2-3): 470-479.
Foster, L., J. Haltiwanger, and C. J. Krizan (2001), “Aggregate Productivity Growth: Lessons from Microeconomic Evidence”, In New Developments in Productivity Analysis: 303–363, University of Chicago Press.
Greenwood, J. and B. Jovanovic (1990), "Financial Development, Growth, and the Distribution of Income”, Journal of Political Economy, 98(5): 1076-1107.
Hsieh, C.-T. and P. Klenow (2009), "Misallocation and Manufacturing TFP in China and India", Quarterly Journal of Economics, 124(4): 1403–1448.
Hulten, C. (1978), "Growth accounting with intermediate inputs", Review of Economic Studies, 45(3): 511-518.
Jayaratne, J. and P. E. Strahan (1996), “The Finance-Growth Nexus: Evidence from Bank Branch Deregulation”, Quarterly Journal of Economics, 111(3): 639-670.
King, R. G. and R. Levine (1993), “Finance and Growth: Schumpeter Might Be Right”, Quarterly Journal of Economics, 108(3): 717-738.
