Problem statement
According to Decree No 56/2009/ND-CP, small and medium-sized enterprises (SMEs) in Vietnam are defined as businesses that employ between 10 and 300 individuals and possess total equity of less than a specified amount.
As of March 2015, small and medium-sized enterprises (SMEs) in Vietnam represent over 90% of all businesses, generating more than 500,000 jobs annually and contributing approximately 40% to the country's overall GDP.
Small and medium-sized enterprises (SMEs) are vital for the sustainable growth of the economy, making their productivity enhancement a pressing priority for the Vietnamese Government The economic growth of Vietnam heavily relies on the productivity levels of these firms, with innovation and technological advancement identified as key drivers (Bartelsman & Doms, 2000) Despite this, many SMEs in Vietnam continue to face operational challenges that result in inefficiencies A significant barrier is the lack of recognition among SMEs regarding the importance of innovation and the adoption of new technologies in boosting productivity, despite the well-documented benefits these changes can bring to their operations.
The common measurement for innovation in empirical studies is R&D expenditure of a particular firm Various empirical studies have been conducted to quantify the relationship between R&D expenditure and firm’s performance
Research indicates a strong correlation between various factors (Siedschlag, Zhang and Cahill, 2010; Belderbos, Carree and Lokshin, 2004; Crespi and Pianta, 2009) However, in Vietnam, small and medium-sized enterprises have not extensively documented their investments in research and development activities.
Innovation activities in SMEs tend to be less formal and encompass a variety of different practices compared to larger companies Consequently, studying the effects of innovation on the productivity of SMEs in Vietnam presents several challenges.
The relationship between innovation and the productivity of SMEs in Vietnam has been underexplored, highlighting the need for in-depth investigation This is particularly crucial in an economy where SMEs play a dominant role and technological advancement remains limited Enhancing understanding in this area is vital for policymakers to effectively guide the development and implementation of innovative activities that can drive both firm and national growth.
Research objectives
This study aims to explore the connection between innovation and productivity among Vietnamese SMEs, focusing on defining and quantifying this relationship at the firm level.
Research questions
The study aims to provide empirical evidence for the main questions emerged:
Is there any relationship between innovation and productivity in the context of SMEs in Vietnam? If yes, then how does innovation can affect SMEs productivity?
Research motivations
This study examines the productivity of Vietnamese SMEs through the lens of Levinsohn and Petrin's (2003) approach, focusing on how productivity is influenced by levels of innovation While innovation is vital for development, the outcomes of such activities remain uncertain, leaving firms unsure if these efforts will generate added value The research findings offer policymakers evidence to effectively allocate resources aimed at enhancing productivity This topic is particularly relevant in developing countries like Vietnam for two key reasons: first, the potential benefits of innovation may not be fully realized, and second, national resources for fostering new innovations are often limited, despite the critical role of innovation in driving global growth.
Research scope and data
The study aims to determine the relationship between innovation and productivity in Vietnam SMEs from 2005 to 2013 in six selected industries include:
The study focuses on six key industries: foods, wood and wood-related products, rubber and plastic products, non-metallic mineral products, fabricated metal products, and furniture These industries were chosen due to their significant representation, accounting for nearly 70% of total SMEs in a five-round survey, making them a reliable sample for the entire dataset It's important to note that the data utilized in this study was collected before the complete dataset from the 2015 survey was available.
As such, data used in this study only ends in 2013.
The structure of this study
This study contains five chapters which can be presented as follow:
Chapter 2 provides theoretical and empirical studies on the relationship between innovation and productivity Chapter 2 begins with Schumpeter Theory of Innovation that explains the role of innovation to economic growth Then this chapter reviews the concept of productivity and the methods of how productivity can be estimated as well as its determinants In addition, the definition of innovation and how it is measured are discussed in the chapter The relationship between these two concepts has been reviewed through literature
Chapter 3 presents the methodology which is utilised in the study An overview of Vietnam SMEs is discussed On the ground of literature review in Chapter 2, the conceptual framework is constructed The measurement of relevant variables and regression techniques are described In addition, this section also includes the process of how to filter data
Chapter 4 presents empirical results Statistical descriptive of data is presented in this chapter Then, the findings on Vietnam SMEs’ productivity are described and discussed The results of regression in relation to the relationship between innovation and productivity are presented in this chapter
Chapter 5 provides the summary of the main results and proposes some policy implications based on the results described in Chapter 4 This Chapter also includes research limitation and suggests some further research direction in the future.
Schumpeter Theory of Innovation – How does Innovation play its role in
Schumpeter was seen as a person who built very first basic foundation to the theory of innovation and economic development In his famous book The Theory of
In his 1912 work on Economic Development, Schumpeter emphasized the significance of technological innovations in driving the business cycle He argued that when new technology is introduced and the economy is prepared to adapt, it can lead to an upward trend in economic activity However, if such innovations occur during periods of economic saturation, the economy may become susceptible to negative shocks and potential depression, rendering new technology less effective Schumpeter also highlighted the importance of firms being willing to take risks and invest in emerging technologies to capitalize on early profit opportunities before competitors adopt them.
Joseph Schumpeter's influential works, including "The Theory of Economic Development" (1934), "Capitalism, Socialism and Democracy" (1942), and "Business Cycles" (1939), highlight the critical roles of innovation and entrepreneurship in economic development He posited that innovation serves as the fundamental catalyst for growth, with entrepreneurs facilitating transformative technological changes that propel the economy beyond its equilibrium state.
Schumpeter explained the development of the economy is mainly driven by innovation which he categorized into five types:
(i) launching new products, whether these are about improving a part of products or totally new to the market,
Introducing innovative production methods, exploring untapped markets, and discovering new sources of raw materials and inputs are essential strategies for business growth and sustainability.
(v) acquiring new market structures in any industry (i.e changing the monopoly position)
Innovation plays a crucial role in driving economic growth, as outlined by Schumpeter's four-step process: invention, innovation, diffusion, and imitation While the initial stages—invention and innovation—may have limited immediate impact, the latter stages—diffusion and imitation—significantly contribute to economic development As economies recognize the potential for increased sales and cost reductions during diffusion and imitation, they tend to invest more in innovations However, mere ideas are insufficient; they require effective implementation, which is where entrepreneurs come into play Entrepreneurs allocate resources to replace outdated technologies with new ones, a process Schumpeter termed "creative destruction." Thus, economic development is fundamentally driven by innovations that lead to creative destruction.
Productivity: concept and measurements
Productivity measures how efficiently a firm, industry, or country transforms input factors into output, defined as the ratio of output to inputs in the manufacturing process As a key indicator, productivity reflects the economic performance of an organization or nation.
Productivity can be influenced by the availability of input resources and the value added during the production process A firm's productivity may decline due to insufficient inputs or inefficient use of resources However, enhancing productivity can be achieved by creating value with available inputs and optimizing specific activities in the manufacturing process.
Productivity can be assessed through various methods, primarily categorized into two types: single factor productivity measures, which evaluate the output relative to a single input, and multifactor productivity measures, also known as total factor productivity, which assess output in relation to multiple inputs.
In the realm of single factor productivity, productivity can be measured through labor productivity and capital productivity, both expressed as a ratio of labor or capital input to gross output or value added While these measures are straightforward to calculate, they only capture partial productivity, failing to reflect the efficiency of combining various input factors in the production process To achieve a more comprehensive assessment of productivity that considers multiple inputs, total factor productivity (TFP) is utilized as a more effective metric Consequently, this research employs total factor productivity to evaluate the productivity of firms.
Estimating total factor productivity using production function estimators has been crucial in examining key issues in the literature, such as the relationship between foreign direct investment and the productivity of domestic firms (Javorcik, 2004), the impact of research and development (Hall et al., 2009), and the effects of information technology (Chun et al., 2015) These analyses are predominantly conducted through simple Cobb-Douglas production function regression.
Where 𝑌 𝑗 represents firm j’s output, 𝐾 𝑗 is physical capital stock, 𝐿 𝑗 is labor input and 𝐴 𝑗 denotes for firm’s level of efficiency, 𝛽 𝑘 and 𝛽 𝑙 are output elasticities with respect to capital and labor
Based on the definition of productivity above, 𝐴 𝑗 is referred to Total Factor Productivity and could be derived by taking natural logs of (1):
In the context of time series analysis, Total Factor Productivity (TFP) is represented by two key components: the average productivity across all firms, denoted as 𝛽₀, and the firm-specific deviation from this average, represented by 𝜖𝑗𝑡 The latter reflects the impact of unobserved factors on a firm's output beyond its inputs Furthermore, the deviation 𝜖𝑗𝑡 can be decomposed into firm-level productivity 𝑤𝑗𝑡 and an independent and identically distributed (i.i.d) component 𝑣𝑗𝑡.
Therefore researchers can get firm’s productivity from estimating (3) and solving for 𝑤 𝑗𝑡 :
Then, the exponential of 𝑤̂ 𝑗𝑡 is the result of firm-level productivity
Research on calculating total factor productivity primarily follows two approaches: non-parametric and parametric methods The non-parametric technique, notably growth accounting, was popularized by Robert Solow's 1957 paper on technical changes and production functions This method, based on the assumptions of constant returns to scale and competitive factor markets, illustrates how changes in output growth can be attributed to variations in input types and total factor productivity However, growth accounting has limitations, particularly in addressing causality, as it cannot determine whether investments in technological changes drive productivity growth or vice versa In contrast, the parametric approach utilizes econometric methods to estimate total factor productivity by analyzing the relationship between production inputs and output through production function estimators Econometric techniques offer several advantages, including the ability to test parameter significance and address endogeneity issues.
The non-parametric growth accounting method, introduced by Robert Solow in his 1957 paper on technical change, analyzes aggregate production functions to assess the sources of economic growth This approach distinguishes between growth attributable to input contributions, which occurs through movement along the production function, and growth resulting from technological advancements that shift the production function It operates under the assumption of constant returns to scale, where the total elasticities of all input factors sum to one Typically, input factors are weighted by their income shares for national productivity calculations or by their cost shares for firm-level productivity assessments.
Productivity is calculated by solving equation (4) without econometric sense
The "Solow residual," denoted as 𝑤̂ 𝑗𝑡, represents the portion of economic growth that occurs when the output growth rate surpasses the growth rate of input factors This metric not only captures growth driven by technological advancements but also encompasses various external factors that influence overall efficiency beyond just input contributions (Schreyer, 2001).
Endogeneity may arise from the relationship between input decisions and unobserved productivity shocks (𝑤 𝑗𝑡), leading firms to adjust their input levels based on these shocks For instance, firms are likely to increase investments in response to positive productivity shocks, while they may reduce their workforce during unfavorable shocks Consequently, the input coefficients obtained from OLS regression could be biased and inconsistent (Eberhardt and Helmers).
In addressing the issue of endogeneity in production function estimation, several solutions have been proposed in the literature, including Instrumental Variables (IV) regression and dynamic panel estimators Notably, the Generalized Method of Moments (GMM) approach, developed by Arellano and Bond (1991) and Blundell and Bond (1998), along with the methodologies introduced by Olley and Pakes, have gained significant traction in this field.
(1996) which is categorized as ‘structural estimators’, then been further developed by Levinsohn and Petrin (2003)
In standard IV regression, to achieve consistent and unbiased coefficients, it is essential to instrument independent variables causing endogeneity, specifically input quantities (K and L), with variables that are related to them but exogenous to unobserved productivity Input prices (r, w) are often introduced as instruments under the assumption of perfectly competitive markets; however, their effectiveness is questioned by Eberhardt and Helmers (2010) and Van Beveren (2012) for several reasons Firstly, many datasets lack comprehensive information on input prices, and when available, these prices do not vary sufficiently across firms to estimate the production function accurately Secondly, the assumption of perfectly competitive input markets is problematic, as productivity shocks may grant firms market power, thereby influencing input prices and creating a correlation between the instrument variables and the error term Lastly, even if the assumption holds, changes in input prices, such as wages, may reflect unobserved labor quality, which is also tied to unobserved productivity, rendering wages invalid as instruments for labor input in production function estimation A similar relationship exists for rental rates and capital stock concerning unobserved productivity Consequently, standard IV regression using input prices as instruments for input quantities fails to produce consistent results.
Finding a robust instrument for input quantity in production function regression can be challenging Arellano and Bond (1991) and Blundell and Bond (1998) advanced the field by introducing the Generalized Method of Moments (GMM) estimator This method utilizes past values of both dependent and independent variables as instruments to address endogeneity issues The validity of these instruments is supported by the premise that input decisions made prior to time t are uncorrelated with productivity shocks occurring at time t, denoted as 𝑤𝑗𝑡.
The GMM approach effectively addresses endogeneity issues and produces satisfactory results; however, it is not built upon a structural model that reflects firm behaviors (Eberhardt and Helmers, 2010) Olley and Pakes also highlight this limitation.
In 1996, OP introduced a novel method for understanding a firm's production function by analyzing observed behaviors, addressing the endogeneity issue in production functions They utilized firm investment decisions as a proxy for unobserved productivity, denoted as 𝑤𝑗𝑡 Two key assumptions underpin this approach: the "monotonicity assumption," which posits a strong positive correlation between firm investment, capital stock, and unobserved productivity, allowing for the determination of productivity through the inversion of the investment function, 𝑤𝑗𝑡 = 𝑓𝑡(𝑖𝑗𝑡, 𝑘𝑗𝑡), and the assumption that labor is fully flexible and can adjust immediately upon observing 𝑤𝑗𝑡 The second assumption, known as the "scalar unobservable" condition, asserts that only unobserved productivity influences a firm's investment decisions OP's "structural estimator" is derived through a two-step process.
Firstly, from (3) output 𝑦 𝑗𝑡 has been regressed on labor input 𝑙 𝑗𝑡 and a proxy of firm-specific productivity:
Innovation: concept and measurements
The Oslo Manual (OECD 2005, p.46) defines innovation as the implementation of a new or significantly improved product, process, marketing method, or organizational practice This widely accepted definition is frequently referenced in studies and surveys conducted by various institutions regarding different aspects of innovation.
There are four type of innovations which proposed in Oslo Manual (OECD 2005):
Product innovation involves the introduction of new goods or services that significantly enhance existing offerings These improvements may include advancements in appearance, technical specifications, functionality, or user-friendliness, setting them apart from current market options.
Process innovation refers to enhancements in the production and distribution methods of goods and services, leading to reduced costs and improved product quality It is closely linked to the techniques utilized, as well as the supporting equipment and software involved in the production process.
Marketing innovation involves enhancing product design, pricing strategies, and promotional campaigns by employing new marketing methods aimed at better meeting client needs, capturing market share, and boosting sales.
Organizational innovation encompasses changes in structure and business environment that aim to reduce administrative and transaction costs while enhancing work efficiency This type of innovation not only focuses on internal processes but also seeks to improve external relationships with suppliers, clients, and state agencies.
Clearly distinguishing between different types of innovation is essential, as many innovations exhibit characteristics of more than one category For example, a new product that requires a novel production process can be classified as both product and process innovation The Oslo Manual (2005) offers comprehensive guidelines for differentiating these types of innovation.
Different types of innovation can be easily represented by dummy variables which applied by many studies such as Hall, Lotti, & Mairesse (2008), Griffith, Huergo, Harrison, & Mairesse (2006), Mairesse, Mohnen, & Kremp (2005), Mairesse
Research by Robin (2009) and Polder, Van Leeuwen et al (2009) highlights that traditional measurements of innovation intensity may not accurately reflect differences among firms, as noted by Mohnen and Hall (2013) This can lead to misleading conclusions, particularly when comparing innovation activities across firms of varying sizes Large firms often engage in a wider array of activities that may not necessarily indicate greater innovation Consequently, using innovation dummy variables may not be appropriate for concluding that large firms are inherently more innovative than their smaller counterparts.
Innovation can be assessed through both input and output measures Input approaches focus on a company's initiatives to launch new products, enhance production processes, explore new markets, and increase operational efficiency Conversely, output measures are evident in the introduction of new products, improvements in production processes, cost reductions, and overall gains in efficiency (Mohnen and Hall, 2013).
Innovation is often quantified through Research and Development (R&D) expenditures, which focus on the creation of new products or production methods However, many non-R&D activities also contribute significantly to innovation, as highlighted in the Oslo Manual (2005) These activities include acquiring patents, paying royalties, enhancing employee skills through internal training, purchasing new equipment or software, and improving management structures or product introduction methods Together with R&D efforts, these non-R&D activities aim to enhance a firm's efficiency, making it essential to consider their expenditures alongside R&D costs when measuring innovation Additionally, the presence of skilled employees is crucial, as they serve as key assets in facilitating the adoption of new technologies, managing operations, and addressing potential technological challenges.
Innovation is typically assessed by its quality, which can be reflected in revenue generation A key metric for evaluating innovation's impact on a company's performance is the percentage revenue share from new products or existing products enhanced through the innovation process This indicator has been widely utilized in research, including studies by Miguel Benavente, to represent innovation effectively.
(2006), Jefferson et al (2006), Siedschlag et al (2010) Another measurement of innovation effect on firm’s performance is cost deduction due to process innovation (Peters, 2008).
How has the relationship between innovation and firms’ performance been
performance been analysed in the literature?
The relationship between innovation and a firm's performance has been extensively studied, yet there remains no consensus on its impact On one hand, innovation is often seen to positively influence firm performance, measurable through various indicators Conversely, numerous studies have highlighted the potential negative effects that innovation can also bring.
Innovation significantly impacts a firm's performance by transforming innovation inputs into outputs that enhance overall results (Crépon, Duguet, and Mairessec, 1998) The CDM model, developed by these researchers, estimates the relationship between innovation and performance and has been utilized by various scholars, including Miguel Benavente (2006) and Mairesse et al (2005) In a study by Siedschlag, Zhang, and Cahill (2010), the CDM model was applied to panel data from 723 firms in Ireland, focusing on foreign ownership and international trade This model involves three stages: (i) the firm's decision to invest in innovation, (ii) determining innovation output from inputs, and (iii) analyzing the relationship between innovation output and final production The findings revealed that foreign-owned firms and those engaged in export activities are more likely to invest in innovation and achieve higher innovation outputs, while innovation expenditure showed no significant effect on output Additionally, there is a positive correlation between innovation outputs and labor productivity.
The relationship between innovation and firm performance is often viewed as causal, with studies by Belderbos et al (2004), Lokshin et al (2008), Parisi et al (2006), and Santos et al (2014) supporting this notion Innovation leads to the development of new products and enhancements in production processes and business practices, resulting in improved efficiency and overall performance for firms Conversely, firms that exhibit better performance are more likely to invest in innovation Due to this causal link, many studies have employed Instrumental Variable (IV) or Generalized Method of Moments (GMM) estimation techniques to address the endogeneity issues that may arise.
Belderbos, Carree, and Lokshin (2004) analyzed data from the Community Innovation Survey for 2,056 manufacturing firms to explore the relationship between innovation activities and external collaboration They assessed innovation through two variables: internal innovation expenditure relative to sales and external collaboration in R&D with competitors, suppliers, customers, and research institutions Firm performance was measured by labor productivity growth and sales of new market products Using IV regression, the study controlled for factors such as firm size, industry classification, ownership status, and past productivity levels The findings revealed that various forms of R&D collaboration and the intensity of innovation positively influence productivity growth, although innovation intensity did not significantly impact sales growth from new products.
Lokshin, Belderbos, and Carree (2008) discovered a significant positive correlation between internal innovation and labor productivity, emphasizing the importance of a firm's R&D expenditures Internal innovation refers to a firm's own R&D spending, while external innovation involves contracting R&D services from other firms Their study utilized GMM estimation for a dynamic panel analysis based on the augmented Cobb-Douglas production function, examining 304 manufacturing firms in the Netherlands from 1996 to 2001 The model incorporated various innovation variables, including quadratic and interaction forms of internal and external R&D expenditures The findings revealed that internal and external R&D complement each other in enhancing productivity, though with diminishing returns Internal R&D is crucial for a firm's productivity, while external R&D significantly impacts productivity only when a firm has sufficiently invested in internal R&D.
Parisi, Schiantarelli, and Sembenelli (2006) conducted a study on 941 manufacturing firms in Italy, utilizing data from surveys in 1995 and 1998 to explore the relationship between innovation and firm performance Their research focused on both product and process innovation efforts, employing two analytical approaches to assess the impact of innovation on performance The first approach applied the Cobb-Douglas production function to output growth, using variables such as lagged output per labor, material per labor, capital per labor, R&D intensity, and firm size The second approach regressed Total Factor Productivity (TFP), calculated using the Levinsohn and Petrin method, against the same innovation variables The findings indicated a positive effect of both process and product innovation on productivity, with process innovation demonstrating a greater impact than product innovation, and these results remained consistent across both analytical methods.
Numerous studies indicate that innovation does not significantly enhance productivity For instance, Santos et al (2014) found that innovative investments do not meaningfully influence a firm's performance Similarly, Li and Atuahene-Gima (2001) attributed the negligible impact of innovation on firm performance to the inherent uncertainties and risks associated with innovation activities, which often require substantial resource allocation without guaranteed value creation Additionally, Branzei and Vertinsky (2006) emphasized the necessity of specialized resources and organizational capabilities to ensure that innovation efforts yield positive outcomes for firms.
In Vietnam, there is a growing interest in exploring the concepts of innovation and productivity, particularly among SMEs Nguyen et al (2007) examined the link between innovation and exports in Vietnamese SMEs, using indicators such as the introduction of new products, new processes, or improvements to existing products to represent innovation Similarly, Vu and Doan (2015) investigated the relationship between innovation and SME performance, incorporating marketing changes as additional proxies for innovation, with performance measured by annual gross profit Their study identified endogeneity issues in the innovation-performance relationship and employed a 2SLS model to address this The findings revealed that innovation efforts in product development, production processes, and marketing positively influence firm performance.
Productivity for Vietnamese firms have been analysed widely in the literature
Ha and Kiyota (2014) used data from Annual Survey on Enterprises from 2000 to
In 2009, a non-parametric method was introduced to estimate a firm's Total Factor Productivity (TFP) while examining the link between productivity and turnover Yang and Huang (2012) utilized the Levinsohn and Petrin approach to assess TFP in Vietnamese SMEs, although their research concentrated on the impact of trade liberalization on productivity, setting it apart from the current study.
This study aims to examine the significance of innovation on the productivity of SMEs in Vietnam, a transitioning economy It assesses innovation through various metrics, including innovation expenditure intensity, a binary indicator for innovation, and the proportion of high-quality employees within the total workforce Productivity is evaluated using the Levinsohn and Petrin approach, which addresses endogeneity issues that may arise from the relationship between input decisions and productivity shocks This method of estimating total factor productivity (TFP) is seldom utilized in the context of Vietnam's literature.
Appendix 2 provides the summaries on the related empirical studies on the relationship between innovation and firm’s performance.
An overview of Vietnamese Small and Medium-sized Enterprises
Small and Medium-sized Enterprises (SMEs) are typically classified by international institutions and countries based on criteria such as the number of employees, total revenues, and total assets or equity, in accordance with established regulations.
According to No 56/2009/ND-CP, small and medium-sized enterprises (SMEs) in Vietnam are defined as businesses that employ between 10 and 300 individuals and possess total equity below a specified threshold.
100 billion dong The details of the classification following Decree No 56/2009/ND-
CP is presented in Table 3.1 below:
Table 3.1: Classification of SMEs in Vietnam
Micro enterprises Small enterprises Medium enterprises
Industry Average no of employee
Average no of employee Total assets Average no of employee Total assets Agriculture, forestry and fishery
Industry and construction < 10 10-200 < VND 20 bil 200-300 VND 20–100 bil
Services < 10 10-50 < VND 10 bil 50-100 VND 10–50 bil
Source: Government’s decree No 56/2009/ND-CP
Since the implementation of the Enterprises Law in 2005, Vietnam has seen a significant rise in the number of businesses Notably, the total number of small and medium-sized enterprises (SMEs) doubled from 120,074 in 2006 to 367,300 in 2013, highlighting the rapid growth of the SME sector during this period.
In Vietnam, micro, small, and medium-sized enterprises (MSMEs) dominate the business landscape, comprising approximately 96% to 98% of all firms from 2006 to 2013 While the total number of enterprises has increased annually, their size has been gradually shrinking Specifically, micro firms (with 1-10 employees) constituted 61% of all firms in 2006, a slight decrease from 66% in 2006 Small firms accounted for 35% in 2006 but fell to 32% by 2013, while large firms saw a significant decline from 4% in 2006 to just 1.6% in 2013 This trend of downsizing among Vietnamese enterprises results in a lack of medium and large-sized firms, which are crucial for steering the economy towards successful international integration.
Figure 3.1: Number of enterprises at 31/12 (by size of employees)
Small and Medium-sized Enterprises (SMEs) have experienced substantial growth, representing the largest segment of total enterprises and making significant contributions to the economy The Mid-term Evaluation Report on the Implementation of the Developing Small and Medium-sized Enterprises Plan (2011-2015) highlights four key areas where SMEs impact the economy: Gross Domestic Product (GDP), government revenue, overall investment, and job creation.
From 2009 to 2012, non-state owned enterprises, predominantly small and medium-sized enterprises (SMEs) which make up 98.6% of this sector, contributed significantly to the GDP, accounting for approximately 48-49% of the total economy In contrast, the contribution from state-owned enterprises, with 59.3% classified as SMEs, experienced a decline, dropping from 37.32%.
From 2009 to 2012, the government's privatization plan significantly impacted the economy, with the contribution of state-owned enterprises to GDP dropping from 32.57% to 32.57% Meanwhile, foreign direct investment, predominantly from small and medium-sized enterprises (SMEs), consistently contributed around 17-18% to the total GDP during the same period.
In 2012, the structure of GDP and the ongoing privatization process indicate that small and medium-sized enterprises (SMEs) in the non-state owned sector are poised to play a crucial role in driving future GDP growth.
Micro Small and Medium Large
The total amount of taxes and fees that SMEs has contributed accounting for a large part in government budget since they are the key sector in the economy In
2010, SMEs contributed over VND 181,060 billion dong to the state budget, accounting for 41% total collected taxes and fees from all enterprises In 2011 and
2012, these figures are VND 181,210 billion dong (34%) and VND 205,260 billion dong (34%)
Small and medium-sized enterprises (SMEs) play a significant role in national investment, contributing a substantial portion to the overall economy In 2010, SMEs accounted for VND 236,119 billion, representing 32% of total enterprise sector investment This figure saw a remarkable increase in 2011, reaching VND 699,690 billion (57%), primarily from micro and small-sized enterprises, before dropping to VND 235,463 billion (29%) in 2012 Notably, small-sized firms consistently held the largest share of total investment within the enterprise sector, contributing between 61% and 68% from 2010 to 2012.
Small and medium-sized enterprises (SMEs) significantly contribute to Vietnam's economy by creating and maintaining jobs In 2010, Vietnamese SMEs provided over 4.35 million jobs, representing nearly 45% of total employment in the enterprise sector By 2012, this number rose by 16.83% to 5.09 million jobs, accounting for 47% of total employment Additionally, worker income in the SME sector has improved over time, highlighting the positive impact of these enterprises on the livelihoods of employees.
42 million dong/employee/year and this figure reaches to VND 46 million in 2011 and continuous increases to VND 61 million in 2012, approximately 90% average income of employee in the whole enterprise sector
Small and medium-sized enterprises (SMEs) are crucial to Vietnam's economic growth, significantly contributing to national income, government revenue, total investment, and job creation They play a vital role in stabilizing workers' incomes and enhancing living standards Despite their importance, Vietnamese SMEs continue to encounter numerous challenges.
Small and medium-sized firms often struggle with limited internal capital, making it difficult for them to access external funding sources such as banks and financial institutions This challenge arises from a lack of collateral, untrustworthy financial practices, complex application procedures, and insufficient information As a result, these businesses frequently find themselves unable to secure the necessary capital for market expansion, technology upgrades, or new project investments.
Vietnamese SMEs are lagging in technology adoption compared to global standards, with only 10% of their total exports and imports being technology imports, significantly lower than the 40% average of other countries This limited technological advancement can be attributed to three main factors: lower profit margins compared to larger firms, restricted access to government support programs, and a lack of integration into supply chains and developed sub-industries Consequently, the inadequate technology in these SMEs adversely impacts labor productivity, product quality, and overall competitiveness in the market.
In 2012, a significant 75% of workers in Vietnamese SMEs lacked access to official technical training, while only 40% of SME owners held a university degree Additionally, many managers were not adequately trained in essential areas such as economic knowledge, business management, and law, resulting in management decisions primarily relying on personal experience Consequently, the operational efficiency of Vietnamese SMEs remains lower compared to their regional and global counterparts.
Methodology
From the theories and empirical studies, the conceptual framework for this study is built and illustrated as Figure 3.2 below:
Innovation significantly influences productivity, which can be analyzed in two key stages The first stage involves estimating total factor productivity for each firm, utilizing the Levinsohn and Petrin (2003) approach within the production function framework In this analysis, labor, capital, and intermediate inputs serve as input variables, while value added is considered the output variable.
The examination of the relationship between innovation and productivity involves regressing total factor productivity for each firm against various innovation proxies, including a dummy variable for investment in innovation activities, innovation expenditure intensity, and the proportion of high-quality employees in the total labor force This analysis is grounded in Schumpeter's Theory of Innovation and supported by empirical studies conducted by Siedschlag, Zhang, and Cahill (2010); Belderbos, Carree, and Lokshin (2004); Crespi and Pianta (2009); Santos, Basso, Kimura, and Kayo (2014); and Lokshin, Belderbos, and Carree (2008).
In addition to innovation variables, the regression analysis will incorporate control variables such as firm age, firm size, capital structure, and previous productivity levels Numerous empirical studies suggest that these factors significantly influence a firm's productivity.
Output (Value added) Total factor productivity
- Share of high-quality labor in total labor force
Jaumandreu (2004); Dhawan (2001); Margaritis and Psillaki (2010) and
The detail methodology for above two stages are presented in the next sections
3.2.2.1 First stage: Total factor productivity estimation using Levinsohn and Petrin (2003) approach
Levinsohn and Petrin (2003) developed a novel method for calculating Total Factor Productivity (TFP) within the framework established by Olley and Pakes (1996) Unlike previous approaches that relied on investment, they utilized intermediate inputs to account for unobserved productivity, which is correlated with the error term Their findings demonstrated that this alternative proxy leads to more reliable and consistent estimates of the coefficients.
This study utilizes the LP approach to estimate firm productivity, focusing on the value-added case The method begins with the Cobb-Douglas production function in logarithmic form, incorporating two inputs: labor (l), which is freely variable, and capital (k), treated as a state variable The output measured is the value added (y) of firm i in year t.
𝑦𝑗𝑡 = 𝛽0+ 𝛽𝑘𝑘𝑗𝑡+ 𝛽𝑙𝑙𝑗𝑡+ 𝜀𝑗𝑡 (7) Error term 𝜀 𝑗𝑡 can be divided by two components: unobservable productivity (productivity shocks that correlated to inputs) 𝑤𝑗𝑡 and an i.i.d component which does not affect inputs choice decision 𝑣𝑗𝑡 Rewrite (7) we have:
In a scenario of perfect market competition, all firms experience identical prices for inputs and outputs Consequently, the demand for intermediate inputs can be represented through capital and unobservable productivity, without needing to specify input and output prices This relationship is denoted as 𝑖 𝑗𝑡 = 𝑖 𝑡 (𝑘 𝑗𝑡 , 𝑤 𝑗𝑡), indicating that intermediate inputs are influenced by capital stock and unobservable productivity shocks The function is indexed by time t to account for variations in input and output prices over time.
The monotonicity condition assumes that, given a fixed level of capital, an increase in productivity or a positive productivity shock will result in greater use of intermediate inputs This is because, with higher productivity, firms can increase output as the marginal revenue from these inputs rises In a perfectly competitive market, this relationship is clear; however, in an oligopolistic market, firms may refrain from increasing output in response to a positive productivity shock, as doing so could lower prices and ultimately reduce their sales.
Under the assumption of monotonicity, the demand function for intermediate inputs can be inverted, allowing productivity shocks \( w_t \) to be expressed as a function of both intermediate inputs and capital: \( w_{jt} = w_t(i_{jt}, k_{jt}) \) In this context, productivity shocks are effectively proxied by intermediate inputs.
Substituting productivity shock function into (8) get:
To estimate the coefficients of input variables in equation (7), LP suggested a process of two stages:
The first stage: Estimating coefficient of freely variable labor:
Equation (9) represents a partially linear model, exhibiting linearity in labor while being non-linear concerning intermediate inputs and capital According to Robinson (1988), this type of equation can be estimated through a semiparametric approach In this method, the expected values of the dependent variables \( y_{jt} \) and \( l_{ji} \) are predicted based on given values of \( i_{jt} \) and \( k_{jt} \) using weighted least squares, applying a second-order approximation for \( (i_{jt}, k_{jt}) \).
Equation (9) then can be rewritten as:
𝐸(𝑦 𝑗𝑡 |𝑖 𝑗𝑡 , 𝑘 𝑗𝑡 ) = 𝛽 𝑙 𝐸(𝑙 𝑗𝑖 |𝑖 𝑗𝑡 , 𝑘 𝑗𝑡 ) + 𝜑 𝑡 (𝑖 𝑗𝑡 , 𝑘 𝑗𝑡 ) (10) (it is noted that: 𝐸(𝑣 𝑗𝑡 |𝑖 𝑗𝑡 , 𝑘 𝑗𝑡 ) = 0 and: 𝐸(𝜑 𝑡 (𝑖 𝑗𝑡 , 𝑘 𝑗𝑡 )|𝑖 𝑗𝑡 , 𝑘 𝑗𝑡 ) = 𝜑 𝑡 (𝑖 𝑗𝑡 , 𝑘 𝑗𝑡 )) Subtracting (10) from (9) yields:
𝑦𝑗𝑡− 𝐸(𝑦𝑗𝑡|𝑖 𝑗𝑡 , 𝑘𝑗𝑡) = 𝛽 𝑙 (𝑙𝑗𝑡− 𝐸(𝑙𝑗𝑖|𝑖 𝑗𝑡 , 𝑘𝑗𝑡)) + 𝑣 𝑗𝑡 (11) Labor coefficient 𝛽 𝑙 in equation (11) can be estimated using no-intercept Ordinary Least Squares and yield unbiased and consistent result since 𝑣 𝑗𝑡 is uncorrelated with inputs by assumption
The second stage: Estimation of capital coefficient:
State variable capital is assumed to respond gradually to productivity shocks When these shocks are divided into forecastable and non-forecastable components, capital tends to adjust in response to the forecastable part but not the non-forecastable part Additionally, the LP model assumes that unobservable productivity follows a first-order Markov process.
𝑤 𝑗𝑡 = 𝐸(𝑤 𝑗𝑡 |𝑤 𝑗𝑡−1 ) + 𝜂 𝑗𝑡 (12) where: 𝜂𝑗𝑡 is the “news” in unobservable productivity 𝑤
Two moment conditions to follows:
𝐸(𝜂 𝑗𝑡 + 𝑣 𝑗𝑡 |𝑘 𝑗𝑡 ) = 𝐸(𝜂 𝑗𝑡 |𝑘 𝑗𝑡 ) + 𝐸(𝑣 𝑗𝑡 |𝑘 𝑗𝑡 ) = 0 which means capital is mean independent of unpredictable productivity and:
𝐸(𝜂 𝑗𝑡 + 𝑣 𝑗𝑡 |𝑖 𝑗𝑡−1 ) = 𝐸(𝜂 𝑗𝑡 |𝑖 𝑗𝑡−1 ) + 𝐸(𝑣 𝑗𝑡 |𝑖 𝑗𝑡−1 ) = 0, referring that decision in intermediate input choices in previous period do not have relationship with the unpredictable productivity in in the following period
Next step would be choosing candidate value for (𝛽 𝑘 ), let denotes- (𝛽 𝑘 ∗ ), equation (8) then can be rewritten as:
𝑦 𝑗𝑡 − 𝛽 𝑙 𝑙 𝑗𝑡 − 𝛽 𝑘 ∗ 𝑘 𝑗𝑡 = 𝑤 𝑗𝑡 + 𝑣 𝑗𝑡 (13) Together with 𝛽 𝑙 estimated from first stage, computing 𝑤 𝑗𝑡 ̂+ 𝑣 𝑗𝑡 as:
𝑤 𝑗𝑡 ̂+ 𝑣 𝑗𝑡 = 𝑦 𝑗𝑡 − 𝛽̂ 𝑙 𝑙 𝑗𝑡 − 𝛽 𝑘 ∗ 𝑘 𝑗𝑡 (14) Estimation of 𝜑𝑡 got from first stage then can be applied to the following function:
𝑤̂ = 𝜑 𝑗𝑡−1 ̂ − 𝛽 𝑡−1 𝑘 ∗ 𝑘 𝑗𝑡−1 (15) The result of (14) then be regressed against result got from (15) using quadratic least square to get the estimation of 𝐸(𝑤𝑗𝑡|𝑤 𝑗𝑡−1 )
Substituting (12) on (13) then re-arranging the equation yields:
The equation \( y_{jt} - \beta_l l_{jt} - \beta_k k_{jt} - E(w_{jt} | w_{jt-1}) = v_{jt} + \eta_{jt} \) indicates that all variables on the left side have been estimated through the aforementioned steps, allowing for straightforward computation of \( \hat{v}_{jt} + \eta_{jt} \) This leads to an algorithm for selecting \( \beta_k^* \) aimed at minimizing \( \hat{v}_{jt} + \eta_{jt} \) The two-stage estimation process developed by Levinsohn and Petrin effectively addresses the endogeneity issue caused by the correlation between input choices and unobservable productivity, yielding consistent input coefficient results for calculating Total Factor Productivity (TFP).
The levpet command in Stata simplifies the calculation of Total Factor Productivity (TFP), making it easier for users to implement this process This study utilized the levpet command for accurate TFP calculations.
3.2.2.2 Second stage: Regression model - How does innovation affect firms’ productivity?
According to the related concepts and empirical studies about innovation- productivity relationship at firm level, the dynamic panel regression including lagged effect of dependent variables is constructed as below:
The set of innovation variables (𝑋 𝑖𝑡) includes dummy variables for innovation, intensity of innovation expenditure, and the share of high-quality labor in a firm's total workforce, alongside control variables such as the firm's age, size, and capital structure Incorporating a lagged dependent variable (𝑇𝐹𝑃𝑗𝑡−1) into the model may result in biased estimations, as Achen (2000) suggests that this can lead to an overestimation of the lagged variable and an underestimation of other explanatory variables due to the combined effects of serial correlation and significant trends in exogenous variables.
Ordinary Least Squares (OLS) estimation can produce biased estimators (Nickell, 1981), and Fixed Effects (FE) estimation may have similar issues (Bond, 2002) To address these challenges and obtain unbiased and consistent estimators, alternative estimation methods are recommended In such cases, Instrumental Variables (IV) or Generalized Method of Moments (GMM) estimation techniques are suggested (Keele and Kelly, 2006).
This model may face potential endogeneity issues, as innovation activities and firm productivity can have a simultaneous relationship (Chun, Kim, and Lee, 2015; Santos et al., 2014; Lokshin et al., 2008) To address the bias from endogeneity, Generalized Method of Moments (GMM) estimation will be utilized instead of Instrumental Variable (IV) estimation for two key reasons: first, GMM can effectively correct for heteroscedasticity and autocorrelation, while IV estimation is less efficient in the presence of heteroscedasticity; second, finding a valid external instrument variable for IV estimation poses significant challenges.
Research hypotheses and concept measurements
As mentioned in the previous chapters, innovation could have positive impact on firm’s productivity (Belderbos, Carree and Lokshin (2004), Crespi and Pianta
This study explores the positive relationship between innovation and firm productivity within the context of Vietnamese SMEs It analyzes innovation through two key components: the intensity of innovation expenditure and the proportion of high-quality labor in the total workforce.
In a further details, two hypotheses about the relationship between innovation and firm’s productivity are tested in this study are illustrated as below:
Hypothesis H 1 : Innovation expenditure intensity have the relationship with firm’s productivity in positive direction
Hypothesis H 2 : High-quality labor in total firm’s labor force have positive impact on firm’s productivity
This study employs two-stage estimation to determine the innovation – firm’s total factor productivity Variables used include:
Stage 1: three input variables (labor, intermediate inputs and physical capital) and output (value added)
Stage 2: dependent variable (total factor productivity), innovation variables (innovation expenditure intensity, dummy for innovation, share of high-quality employee in labor force), control variables (firm’s age, firm’s size and firm’s capital structure) and past level of productivity
Table 3.2 below provides the summary of the concepts and measurements for the above variables
Table 3.2: Concepts and measurements of variables used in the study
L Labor used in firm's production process in survey year
Number of regular employees at the end of survey year
C Value of physical capital used in firm's production process in survey year
Average value of physical assets at two point time: at the beginning and at the end of survey year
I Value of intermediate inputs used in firm's production process in survey year
Average value of intermediate inputs used in the production process at the beginning and at the end of survey year
Intermediate inputs include raw material and indirect inputs
Y Value added generated by firm within year
Value added generated by firm within year
TFP Total Factor Productivity Total factor productivity determined by
LP approach obtained from stage 1
Inv_exp Innovation expenditure intensity
Share of innovation expenditures in total firm's investment in survey year
Innovation expenditures include: investment in R&D, investment in human upgrading, investment in buying patents and investment in equipment and machinery
D_Inv Dummy on innovation Coded 1 if firm have invested in innovation activities and 0 otherwise
Share_L Share of high-quality employee in total labor force
Share of professional employee in total labor force
Size Firm's size Ln of total assets
The age of a firm is calculated by subtracting the established year from the year of the survey If a firm provides varying information across different surveys, the established year used will be the one reported in the first survey round.
Cap_struc Firm's capital structure Ratio of outstanding debt over total assets
L.TFP Lag one time of firm’s total factor productivity
Lag one time of firm’s total factor productivity
Data sources
This study analyzes survey data from Small and Medium-sized Enterprises (SMEs) in Vietnam, collected in 2005, 2007, 2009, 2011, and 2013, through collaboration between four organizations: the Institute of Labor Studies and Social Affairs (ILSSA), the Stockholm School of Economics (SSE), and the Department of Economics at the University of Copenhagen The surveys were conducted across ten provinces and cities, including Ha Noi, Ho Chi Minh, and Hai Phong, involving a sample of 2,500 firms, of which 80% were repeated from previous surveys The remaining 20% were replaced due to reasons such as firms ceasing operations or not responding, ensuring that the data represents the Vietnamese SMEs population effectively.
This study analyzes six major industries—foods, wood and wood-related products, rubber and plastic products, non-metallic mineral products, fabricated metal products, and furniture—which together represent nearly 70% of total SMEs surveyed across five rounds These industries are indicative of the overall dataset, highlighting their significance within the broader economic landscape.
Table 3.3: Number of observation in selected industries in dataset
Year No of obs in selected industries Total obs in raw data %
The study employs some basic filters to remove missing data and outliers The details of filter process are described below:
In Stage 1 of estimating Total Factor Productivity (TFP) using the LP approach, it is essential to have complete information on output (revenue) and input factors, including labor, intermediate inputs, and capital Consequently, firms lacking any of this critical data are excluded from the dataset The summary descriptive statistics for the remaining data are displayed in Table 3.3 below.
Table 3.4: Number of observation after filtering
In Stage 2 of the analysis, variables such as firm age and total assets, which serve as a proxy for firm size, exhibit a wide range of values To avoid biased results caused by outliers, a basic filtering process was implemented to eliminate any observations exceeding the mean plus two standard deviations This filtering was applied to both firm age and total assets across the selected industries The summary statistics after this filtering process are detailed in Table 3.5.
Table 3.5: Number of observation after filtering in stage 2
Industry Total obs No of outliers
Total Factor Productivity of Vietnamese SMEs
In the initial phase of the study, a Linear Programming (LP) approach is utilized to calculate Total Factor Productivity (TFP) for each firm over the reported years, using data on output and various input factors as detailed in Chapter 3 The output is represented by the firm's net revenue from sales for the fiscal year, while input factors consist of Labor, Intermediate Inputs, and Capital Labor is quantified by the total number of employees at the end of the fiscal year, Intermediate Inputs encompass the total expenditure on raw materials and indirect production factors, and Capital is measured by the average value of physical assets at both the beginning and end of the fiscal year Although this capital measurement may not perfectly reflect the capital used in production, it serves as a suitable proxy A summary of the data variables—firm’s net revenue, labor, capital, and intermediate inputs—is provided in Table 4.1.
Industry Foods Woods Rubber and Plastic
Year Variables Mean Std Dev Max Min Mean Std Dev Max Min Mean Std Dev Max Min
Industry Non-metallic mineral Fabricated metal Furniture
Year Variables Mean Std Dev Max Min Mean Std Dev Max Min Mean Std Dev Max Min
4.1.2 Total factor productivity from production function estimation of
Tables 4.2 and 4.3 present a comparison of production function parameters estimated through OLS, Fixed Effect, and LP regression methods across six industries: Foods, Woods, Rubber and Plastics, Non-metallic Minerals, Fabricated Metal, and Furniture, covering the period from 2005 to 2013.
The study’s results of higher estimation on free labor input in OLS regression as compared with LP regression confirmed the concern of Marschak and Andrews
In the presence of a correlation between input factors and unobserved productivity shocks, estimations of input factors can be biased upward, while capital input estimations may vary between upward and downward bias depending on their correlation with these shocks When there is weak or no correlation, particularly with labor as a free variable, Ordinary Least Squares (OLS) estimates tend to be biased downward The study found that OLS capital estimations were consistently lower compared to Levinsohn-Petrin (LP) regression across six selected industries Furthermore, fixed effects (FE) estimations differed significantly from both OLS and LP results, likely due to the fluctuating fixed effects of unobserved productivity shocks unique to each firm Additionally, FE estimation proves inefficient when there is a correlation between input factors and unobserved productivity.
Foods Woods Rubber and plastics
OLS FE LP OLS FE LP OLS FE LP lnL 0.766*** 0.470*** 0.379*** 0.680*** 0.424*** 0.361*** 0.721*** 0.527*** 0.382***
Non-metallic mineral Fabricated metal Furniture
OLS FE LP OLS FE LP OLS FE LP lnL 0.727*** 0.357*** 0.348*** 0.790*** 0.599*** 0.368*** 0.561*** 0.456*** 0.310***
Total factor productivity then be calculated and stored for each firm using the following equation:
𝑤̂ = 𝑦𝑗𝑡 𝑗𝑡− 𝛽̂𝑘𝑘 𝑗𝑡− 𝛽̂ 𝑙𝑙 𝑗𝑡 (4) Then the exponential of 𝑤̂ 𝑗𝑡 is the result of firm-level productivity.
Innovation – Firm’s productivity relationship
In this phase, a firm's productivity is analyzed in relation to its historical performance, alongside innovation factors—such as the presence of innovation, intensity of innovation spending, and the proportion of high-quality labor within the total workforce—as well as control variables including the firm's age, size, and capital structure Summary statistics for these variables are presented in Table 4.4.
Table 4.4: Descriptive statistics of TFP and its determinants
No of obs Average Std Dev Max Min
The majority of SMEs in the selected sample are micro and small-sized enterprises, with an average log of total assets at 13.12, equivalent to nearly 2 billion dong This prevalence aligns with the General Statistics Office (GSO) data on Vietnamese SMEs The average operational duration for these firms is 14 years, indicating a relatively long time horizon Out of 8,374 observations, 4,509 reported expenditures on innovation activities, representing 54% of the total sample Notably, 65% of those who reported innovation expenditures indicated non-zero spending.
4.2.2 The relationship between innovation expenditure intensity and firm’s productivity
This section examines Hypothesis 1, which posits a positive relationship between innovation expenditure intensity and firm productivity Table 4.5 presents the results from estimating model (16) using three methodologies: Pooled OLS, Fixed Effects (FE), and System-GMM The GMM approach incorporates lagged endogenous variables as instruments to enhance the analysis.
Table 4.5: Regression results of innovation expenditure intensity and firm’s productivity
Testing for autocorrelation and validity of instruments in GMM results:
The study addresses the potential endogeneity problem between innovation expenditure intensity and firm productivity, highlighting a simultaneous relationship To investigate this, the Durbin-Wu Hausman test is employed, with results detailed in Appendix 3, confirming that the innovation expenditure intensity variable is indeed endogenous Consequently, the use of OLS and FE estimators is deemed biased and inconsistent Therefore, the Generalized Method of Moments (GMM) approach is recommended to effectively address the endogeneity issue, utilizing lagged endogenous variables, specifically innovation expenditure intensity, as instruments.
The last column of Table 4.5 presents the estimation results of equation (16) using the system-GMM approach The analysis focuses on the innovation expenditure intensity (Inv_exp) as the primary independent variable, while controlling for firm age (Age), firm size (Size), and capital structure (Cap_struc) Additionally, the model incorporates the lagged value of total factor productivity (L.TFP) as an independent variable, with the dependent variable being the firm's total factor productivity (TFP).
The coefficient of innovation expenditure intensity (Inv_exp) demonstrates a significant positive effect on a firm's total factor productivity (TFP), indicating that a 1% increase in innovation spending can result in a 0.79 increase in TFP, all else being equal This positive correlation aligns with Schumpeter's Theory of Innovation and is supported by various empirical studies, including those by Belderbos, Carree, and Lokshin (2004), Crespi and Pianta (2009), Rosenbusch, Brinckmann, and Bausch (2011), and Lokshin, Belderbos, and Carree (2008).
Testing for autocorrelation in AR(1) and AR(2) yielded p-values of 0.027 and 0.924, respectively The presence of autocorrelation in the first-order AR(1) model aligns with expectations discussed in Chapter 3 More critically, testing for correlation in the first difference of AR(2) is essential, as finding autocorrelation in AR(2) would indicate the invalidity of instrumental variables Thus, a higher p-value in the AR(2) test is preferable The study results confirm the existence of autocorrelation in AR(1), while AR(2) supports the argument for valid instrumental variables.
Larger firms, indicated by the logarithm of total assets, tend to exhibit higher Total Factor Productivity (TFP) Conversely, a firm's capital structure negatively impacts its TFP Additionally, the previous period's TFP significantly influences the current TFP level, supporting the notion that past productivity levels affect present performance.
4.2.3 The relationship between high-quality labor share in total firm’s labor force and their productivity
The second hypothesis will be examined by incorporating the high-quality labor share (Share_L) into the total labor force of the firm within the innovation variables outlined in equation (16) from the previous section.
Table 4.6 below presents the comparison among OLS, FE and GMM estimations when firm’s TFP is regressed against innovation variables (Inv_exp,
Share_L), control variables (Age, Size, Cap_struc) and lagged one time of TFP
Table 4.6: Regression results of high-quality labor share in Total labor force and Firm’s productivity
Testing for autocorrelation and validity of instruments in GMM results:
The study employs the Durbin-Wu Hausman test to assess the endogeneity of the variable representing the share of high-quality labor within the total labor force (Share_L) The findings indicate that Share_L is indeed endogenous, with further details on the Durbin-Wu Hausman test provided in the study.
Appendix 4) Therefore GMM estimation would become appropriate solution to the endogeneity issue In the results showed in Table 4.6, lag of endogenous variables (in this case are Inv_exp and Share_L) are used as instruments to deal with the problem of endogeneity in GMM estimation
Both Share_L and Inv_exp demonstrate a significant positive relationship with Total Factor Productivity (TFP) Specifically, a one percentage point increase in the share of high-quality labor within the total labor force correlates with a 5.54 unit rise in a firm's TFP, assuming other factors remain constant Additionally, an increase of one percentage point in innovation expenditure intensity results in a 1.06 unit increase in the firm's TFP, ceteris paribus.
The autocorrelation test results for AR(1) and AR(2) using GMM estimation indicate p-values of 0.093 and 0.881, respectively, leading to the acceptance of the null hypothesis of no autocorrelation in AR(2), which supports the validity of the instrumental variables used Additionally, the study reveals a significant inverse relationship between a firm's leverage and its productivity, with a one percentage point increase in the debt-to-total-assets ratio resulting in a 0.5 decrease in total factor productivity Conversely, firm size positively influences productivity, suggesting that larger firms tend to exhibit higher productivity levels, all else being equal However, the age of the firm does not significantly affect productivity outcomes.
The positive and significant coefficient of lag one time Total Factor Productivity (TFP) indicates that the current level of TFP is likely influenced by its past values, aligning with findings from the previous section.