Productivity growth, technological progress and efficiencychanges in vietnamese high tech industries

VIETNAM THE NETHERLANDS VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS PRODUCTIVITY GROWTH, TECHNOLOGICAL PROGRESS AND EFFICIENCY CHANGES IN VIETNAMESE HIGH-TECH IND

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

PRODUCTIVITY GROWTH, TECHNOLOGICAL PROGRESS AND

EFFICIENCY CHANGES IN VIETNAMESE

HIGH-TECH INDUSTRIES

BY

DAO HOANG BINH THIEN

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

HO CHI MINH CITY, January 2015

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

PRODUCTIVITY GROWTH, TECHNOLOGICAL PROGRESS AND

EFFICIENCY CHANGES IN VIETNAMESE

HIGH-TECH INDUSTRIES

A thesis submitted in partial fulfilment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

By

DAO HOANG BINH THIEN

Academic Supervisor:

Dr TRUONG DANG THUY

HO CHI MINH CITY, January 2015

ABSTRACT

Recently, Vietnamese high-tech industries have been receiving attention from both the government, foreign companies, as well as the private sector due to the notable figures of export values (Ministry of Trade and Industry [MoIT] & United Nations Industrial Development Organization [UNIDO], 2011) This thesis attempts

to estimate the productivity growth of Vietnamese high-tech manufacturers and its sources of growth Stochastic Production Frontier (SPF) approach is applied to the 2000-2012 panel dataset of Vietnamese high-tech manufactures, which are divided

in 5 sub-industries Total Factor Productivity (TFP) is then measured and decomposed to three sources, namely Technological progress (TP), Technical efficiency changes (TEC), and Scale change effects (SCE) Three different technical inefficiency effects models are also applied to investigate the determinants of technical efficiency The empirical results show considerable controversy in both signs and magnitudes of TFP and its components, TE and its determinants across models However, in general, maximum likelihood estimates show that TFP is not the main source of output increase Furthermore, the productivity and efficiency of Vietnamese high-tech manufacturers are unlikely to change largely over time Nevertheless, there are differences of technical inefficiency effects across regions, sub-industries, firm sizes, and type of ownerships

Keywords: Vietnam, High-tech, manufacturing, productivity, TFP,

Technological progress, Technical efficiency, Scale change effects

ACKNOWLEDGEMENT

I have taken efforts in this thesis However, it would not have been completed without supports of many individuals and organizations I would like to express my appreciation to all of them

I would like to give special thanks to my academic supervisor, Dr Truong Dang Thuy, whose comments and encouragement helped me to write this thesis

Furthermore, I would also like to acknowledge the Scientific Committee and the staff of Vietnam-Netherlands Programme for their guidance and support as well

as for providing necessary information regarding the thesis

Lastly, my thanks also go to my family and my classmates for their precious support which help me completing this thesis

TABLE OF CONTENTS

ABSTRACT iii

ACKNOWLEDGEMENT iv

LIST OF FIGURES viii

LIST OF TABLES viii

LIST OF APPENDICES ix

ABBREVIATIONS x

CHAPTER 1 INTRODUCTION 1

1.1 Problem statement 1

1.2 Research objectives and hypotheses 5

1.3 Scope of study 5

1.4 Structure of thesis 6

CHAPTER 2 LITERATURE REVIEW 7

2.1 Concepts 7

2.1.1 Total factor productivity (TFP) 7

2.1.2 Technical change or Technological progress (TP) 8

2.1.3 Technical efficiency (TE) and Technical efficiency change (TEC) 8

2.1.4 Scale economies and Scale change effects (SCE) 9

2.1.5 Allocative efficiency (AE) 11

2.2 Approaches to measure and decompose TFP growth 13

2.2.1 Primal or dual approach with production, cost, or profit function 13

2.2.2 Stochastic and deterministic approaches 14

2.2.3 Parametric and non-parametric methods 14

2.3 A review of alternative Stochastic Production Frontier (SPF) models 15

2.3.1 Time-invariant models 16

2.3.2 Time-varying models 17

2.3.3 Exogenous inefficiency determinants 19

2.3.4 TFP growth decomposition 22

CHAPTER 3 OVERVIEW OF VIETNAMESE HIGH-TECHNOLOGY MANUFACTURING SECTOR 25

3.1 High-technology (HT) 25

3.2 Overview of Vietnamese HT manufacturing sector 26

CHAPTER 4 METHODOLOGY 29

4.1 Empirical models 29

4.2 Functional form 30

4.3 Estimation method 34

4.4 Hypotheses and testing 34

4.5 Variable measurement 34

4.5.1 Variables in the frontier model 35

4.5.2 Determinants of Technical inefficiency 36

4.6 Data source and filter process 39

CHAPTER 5 EMPIRICAL RESULTS 40

5.1 Data description 40

5.2 Maximum likelihood estimates 43

5.3 Results of hypothesis testing 47

5.4 Results of TFP decomposition 49

CHAPTER 6 CONCLUSIONS 54

6.1 Findings 54

6.2 Policy implications 55

6.3 Limitations and future research 55

REFERENCES 57

APPENDICES 64

LIST OF FIGURES

Figure 1: Value added of HT manufacturing industries of the world and selected

regions during 1997–2012 (in billions of current dollars) 2

Figure 2: High-tech exports of Vietnam & other countries in Asia (1997-2012) 3

Figure 3: Production frontier, Technological progress, Technical efficiency, and optimal Scale of production 10

Figure 4: Technical efficiency and Allocative efficiency 12

Figure 5: World exports & value-added of HT manufacturing sector (2001-2012) 27 Figure 6: Exports of Vietnamese HT manufacturing sub-industries 28

Figure 7: Proportions of HT firms operating in five sub-industries 41

Figure 8: Percentage of HT firms divided by regions 41

Figure 9: Number of firms of different sizes, during 2000-2012 42

Figure 10: Change of HT WFOEs and SOEs during 2000-2012 42

Figure 11: Kernel density of TE (3 models BC92, BC95, and HL94) 43

Figure 12: List of major obstacles chosen by Vietnamese manufacturing firms (in 2009) 50

LIST OF TABLES Table 1: Contribution of Vietnamese HT in value added of manufacturing sector during 2000–2012 (in percentage) 27

Table 2: Some main characteristics of three models 33

Table 3: Criterion to divide HT firms into three kind of sizes 36

Table 4: Definition and measurement of all variables in the study 38

Table 5: Descriptive statistics of production function variables 40

Table 6: Descriptive statistics of TI effects mean variables 40

Table 7: Maximum Likelihood estimates of translog production frontier 45

Table 8: Maximum Likelihood estimates of technical inefficiency effects model (Model BC95 and Model HL94) 46

Table 9: LR Tests of hypotheses 48

Table 10: TFP & its decomposition in five HT sub-industries (model HL94) 49

Table 11: TFP change & its sources of change over time (model HL94) 51

Table 12: Growth rate of production inputs across HT sub-industries 52

Table 13: Returns to scale across HT sub-industries during 2000-2012 52

LIST OF APPENDICES

Appendix 1: HT manufacturing industries in International and Vietnamese Standard

Industrial Classification 64 Appendix 2: Provinces and Cities of Vietnam divided by regions 65

ABBREVIATIONS

AE Allocative efficiency AEC Allocative efficiency change CRS Constant returns to scale DEA Data envelopment analysis DRS Decreasing returns to scale GSO General Statistics Office

HT High-technology IRS Increasing returns to scale

LR Likelihood-ratio

LS Least Squares MFP Multi-factor Productivity

ML Maximum likelihood MLDV Maximum Likelihood Dummy Variable OLS Ordinary Least Squares

PIM Perpetual Inventory Method SCE Scale change effects

SF Stochastic frontiers SFA Stochastic Frontier Analysis SOE State-owned enterprises SPF Stochastic Production Frontier

TE Technical efficiency TEC Technical efficiency change TFP Total Factor Productivity

TI Technical inefficiency

TP Technological progress VEC Vietnam Enterprise Census VND Vietnam Dong

WFOE Wholly foreign owned enterprises

CHAPTER 1 INTRODUCTION

Since last decades of 20th century, the world has experienced the unexampled evolution of advanced technology-intensive manufacturing industries such as pharmaceuticals, computers, telecommunications, precision engineering, or aircraft Those high-technology (HT) industries have contributed considerably in promoting human beings’ health and longevity, extending the ability of communication, and improving the knowledge accessibility (Hamburg Institute for Economic Research [HWWA], Kiel Institute for World Economics [IfW] & National Research Council [NRC], 1996) Moreover, people are convinced that these HT industries will bring the bright future of remarkable economic growth, including high value-added, high wage employment From the microeconomic perspective, HT firms are believed to spend a large amount in R&D and innovation, which can lead to inventing new products, gaining more market shares, using resources more productively, and creating positive social returns that benefit other sectors (HWWA et al., 1996) Figures of global value added of HT sector during recent years show clearly the

promising trend of its growth, especially in the dynamic Asia region (see Figure 1)

Figure 1: Value added of HT manufacturing industries of the world and selected regions during 1997–2012 (in billions of current dollars)

Source: Appendix table 6-7 of Science and Engineering Indicators 2014

(National Science Board, 2014)

The power of nations is also believed to not influenced by heavy industries like steels but the role is now played by HT manufacturing industries and knowledge-based services, which means that the national autonomy can be improved by developing these industries (HWWA et al., 1996) Indeed, the aging industrial economy has bowed out to give way to the promising knowledge-based and technology-intensive economy Due to that importance, HT manufacturing industries are the target of industrial policies in many countries and regions, including Vietnam

HT manufacturing industries have been paid more attention in Vietnam recently with many high-tech FDI projects built up (MoIT & UNIDO, 2011), together with new Laws and Decisions approved to facilitate the science and technology activities Interestingly, even during the time of global crisis (2008-2009), this sector still had

an increase of export values with about USD 593 million, while low-tech and medium-tech sectors experienced a reduction in exports (MoIT & UNIDO, 2011) Indeed, during the period 2000 to 2009, over half of total exports are HT products

(MoIT & UNIDO, 2011) Although Vietnamese HT manufactures account for only a small proportion of the world market share, its annual growth rate shows a potential

of development for this sector (see Figure 2) Moreover, the government expects HT

industries to play a key role in helping Vietnam economy develop and gain a higher

position in the global value chain (Strategy on exports and imports for 2011-2020,

with visions to 2030, 2011)

Figure 2: High-tech exports of Vietnam & other countries in Asia (1997-2012)

Bubble indicates HT exports in 2012 Annual growth rate is the geometric average annual growth rate of exports during 1997-2012

Source: Appendix table 6-21 of Science and Engineering Indicators 2014 (National Science Board, 2014)

It is obvious that the growth of Vietnamese HT manufacturing industries is remarkable and it seems to be consistent with expectations However, most of HT firms in Vietnam are known to be operating as assembly lines rather than concentrating on R&D and inventing new products From another aspect, the large

proportion of imported components in HT products exported from Vietnam may affect considerably on the production of HT firms Thus, the question is whether Vietnamese HT manufacturing industries perform as well as they look like or not To answer it, productivity, which indicates how well the firms perform using given factor inputs, should be investigated carefully Indeed, productivity is considered an important indicator of development, playing a key role to firms’ survival (Duong, Lai, Nguyen, Le, & Hua, 2014; National Science Board, 2014; Syverson, 2011) More specifically, empirical researchers often estimate total factor productivity (TFP), a measures of firms’ overall productivity, in their analysis Moreover, they do not stop

at measuring TFP and its growth only, some authors try to examine what drives TFP growth Theoretical literature indicates that TFP growth of HT firms stem mainly from the progressive technological change (Sun & Kalirajan, 2005) However, if governments only focus on attracting investments to enhance technological progress

of HT sector, they may ignore the contribution of other important sources such as effects from changes in scale of production (Hamit-Haggar, 2011; Kim & Han, 2001) Besides, empirical studies show evidence that firms can also obtain higher TFP gains

if they apply best practice methods of the given technology, such firms are considered

“technically efficient” (Kalirajan, Obwona, & Zhao, 1996) In this circumstance, technological progress may be absent; instead, effects from improving technical efficiency are the key source contributing to TFP growth Thus, such components should be taken into account when modelling the production function and measuring TFP They will provide more comprehensive insights of HT sector’s status for policy makers in taking HT development policies in consideration

Nevertheless, there are very few papers analyzing the status of TFP change of Vietnamese HT manufacturing industries as well as its decomposition The study of Nguyen, Pham, Nguyen, and Nguyen (2012), which can be the only paper touching that field of TFP growth’s decomposition for Vietnamese manufacturing sector until now, is not focused on HT manufacturing industries Other studies, if conducted in analysis of HT sector, stop at measuring TFP (Newman & Narciso, 2009), or

investigate only one source of TFP change, namely technical efficiency with analyses

on its various determinants (firm size, firm location, ownership…) (Le & Harvie, 2010; Nguyen, Giang, & Bach, 2007) Obviously, the literature of empirical researches on TFP of Vietnamese high-tech manufacturing industries and its sources

of change is rather poor

Thus, with longer timespan (2000-2012) and narrower research object tech industries), besides estimating determinants of technical inefficiency, this paper attempts to measure TFP growth of Vietnamese HT manufacturers as well as its decomposition The results of the study may provide some information to understand the performance of Vietnamese HT sector and be helpful for HT sector development policies

This paper aims to investigate the productivity and efficiency of Vietnamese high-tech manufacturing sector, namely three objectives to attain:

 To measure TFP growth of Vietnamese high-tech manufacturers

 To decompose TFP growth into Technological progress, Scale change effects, and Technical efficiency change

 To examine determinants of technical inefficiency

1.3 Scope of study

The unbalanced panel data in this research includes 5822 observations of 2403 Vietnamese high-tech manufacturing firms through 13 years from 2000 to 2012 The selected sector includes five sub-industries:

(i) Pharmaceuticals;

(ii) Computers and peripherals;

(iii) Radios, TVs, and communication equipment;

(iv) Precision instruments;

(v) Aircrafts

Firms in the sample include various sizes from small to large, different ownerships from state owned, foreign owned, to private owned, with their headquarters located nationwide in six regions of Vietnam

1.4 Structure of thesis

The thesis is divided in six chapters with the following structures:

Chapter 2 presents the literature of productivity and efficiency measurement and decomposition Starting with the definitions of key concepts such as high-technology, productivity, and efficiency, various approaches dealing with the productivity measurement are then reviewed Moreover, different models of productivity decomposition and efficiency estimation are also discussed with advantages and disadvantages of each own

Chapter 3 provides a brief overview of Vietnamese HT manufacturing sector after discussing about definitions and classifications of high technology firms and industries

Chapter 4 describes the specific research methodology, in which the parametric approach and regression technique are expressed in details This chapter also discusses the seven hypotheses mentioned in the second part of chapter Introduction more clearly with the testing methods

Chapter 5 presents the empirical results in two parts, namely descriptive statistics of the data and results of the regression Based on empirical evidence from econometric models, the inference and analysis is then drawn and discussed about productivity and efficiency of Vietnamese high-tech sector

Chapter 6 concludes main findings of the study as well as policy and managerial implications stemmed from the results presented in Chapter 5 This chapter also point out limitations of the thesis and then refer to directions for researches in the future

CHAPTER 2 LITERATURE REVIEW

This chapter provides some definitions of key concepts such as total factor productivity and kinds of efficiency In addition, various approaches measuring and decomposing productivity change are also discussed in this chapter Especially, stochastic production frontier analysis (SPF) is the main focus of this chapter

2.1.1 Total factor productivity (TFP)

Productivity of a firm implies the ratio of outputs over inputs in production (Coelli, Rao, O’Donnell, and Battese (2005) In other words, it shows how well the outputs can be produced from given amounts of inputs Productivity is often used to compare performance between firms or industries: the larger the ratio is, the better the firm (or industry) performs In case there are multiple outputs and multiple inputs involving the production, partial productivity measures, which only take one factor

of production into account, may be selected to estimate to simplify the estimation process There are many partial measures of productivity such as labor productivity, land productivity, or fuel productivity Meanwhile, Total Factor Productivity (TFP)

is the measure of overall productivity, which involves all factors of production In this case, TFP is defined as a ratio of aggregate output produced over aggregate input used (Coelli et al., 2005) It is a better choice of performance measurement than partial measures because partial productivity measures can misrepresent the performance of a firm (Coelli et al., 2005) Because we may never take into account all the factors affecting the output level, Multi-factor Productivity (MFP) is the more precise term that should be used in empirical calculation However, researches tend

to use those two terms interchangeably in their studies, which is also applied in this thesis Over time, TFP tends to change, usually positively, which is believed to be an important factor contributing to the survival of firms (micro perspective) and economic growth (macro perspective) in the long-run

2.1.2 Technical change or Technological progress (TP)

According to neo-classical economists, due to the law of diminishing returns, the firm cannot increase its output levels forever if it keeps accumulating factors of production, given the current technology (Sharma, Sylwester, & Margono, 2007) Thus, when a firm is observed to increase its TFP in the long-run, they argue that the only reason for TFP growth is that the firm has adopted more advanced technology, implying that there is technological progress (TP)

In Solow (1957)’s model, positive technical change (or TP as in some reviews), which is exogenous and unexplainable by the model, is the only source of long-run growth of per capita income Graphically, TP is expressed as the upward shift of the production frontier In other words, with the presence of TP, a firm can

increase its potential productivity beyond previous limits (see Figure 3 for

illustration)

However, arguing against Solow (1957)’s, later studies of other authors have proved that not only TP is the main source of TFP growth, the improvement of technical efficiency, the exploitation of scale economies or allocative efficiency also drive TFP growth (Coelli et al., 2005)

2.1.3 Technical efficiency (TE) and Technical efficiency change (TEC)

A producer is considered as technically efficient “if and only if it is impossible

to produce more of any output without producing less of some other output or using more of some input” (Koopmans, cited in Kumbhakar & Lovell, 2000) Despite the popularity of Solow (1957)’s, this model has a critical weakness when assuming that the firms are operating with full efficiency, i.e the firms are operating along with the

production frontier (see Figure 4) If ignoring the potential contribution of efficiency

changes to TFP growth, the estimate of productivity may be biased and misleading (Hamit-Haggar, 2011) Nishimizu and Page (1982) were the pioneers in introducing efficiency change as a source of productivity growth The assumption of full efficiency is also unrealistic while it is likely that many firms’ productions are inefficient, which means that there is the gap between the production frontier and the

firm’s actual production level Kim and Han (2001) argued that improvements in technical efficiency (TE) can cause TFP growth for firms that are not fully utilize existing technology due to some restraints such as organizational factors The literature also shows more evidence that positive (negative) technical efficiency change (TEC) can contribute to progressive (regressive) TFP change For instance, Nguyen et al (2012) and Kim and Han (2001), after measuring and decomposing TFP change, drew a conclusion about the positive contribution of TEC into TFP growth, whereas findings of Kim and Shafi'i (2009) and Hamit-Haggar (2011) confirmed that TFP of manufacturing industries can be hurt with worsen technical efficiency

2.1.4 Scale economies and Scale change effects (SCE)

According to theoretical background, which are clearly reviewed in Coelli et

al (2005) and Kumbhakar and Lovell (2000), a firm is exploiting scale economies when the ray from the origin is at a tangent to the production frontier and thus defines the point of maximum possible productivity, i.e the point of optimal scale (see Figure 3) They also indicate that when a firm production is technically efficient, it can still increase productivity by exploiting scale economies, which is called scale change effects (SCE) More exactly, when the production function exhibits increasing returns

to scale (IRS), the contribution of SCE to TFP growth will be positive, whereas the decreasing returns to scale (DRS) will worsen the TFP growth Obviously, if constant returns to scale (CRS) exist in the production, there will be no scale effects on the improvement (decline) of TFP

Estimating sources of TFP growth in manufacturing industries of Korea during 1980-1994, Kim and Han (2001) show that the almost scale components are negative

or close to zero, which results in a decrease in TFP growth In other words, Korean manufacturers were operating at DRS or CRS during the period of study Kim and Shafi'i (2009) when estimating TFP growth for the case of Malaysian producers also confirmed that SCE influence significantly on the overall productivity; however, the impact is different across industries

Figure 3: Production frontier, Technological progress, Technical efficiency,

and optimal Scale of production

Source: Coelli et al (2005) Where

F’0: Production frontier at time 0 F’1: Production frontier at time 1 x: input y: output

A: the firm has technical efficiency at time 0 B: the firm has technical inefficiency at time 1 C: the optimal scale at time 0

From F’0 to F’1: technological progress from time 0 to time 1

A

Technological progress

2.1.5 Allocative efficiency (AE)

Allocative efficiency, which involves the mix of input selection (capital and labor, for example), is also another component of TFP growth Indeed, while the factor prices and the production technology hold constant, selecting the mix of inputs

in optimal proportions such that costs of production reach minimum can also increase the productivity of the firm (Coelli et al., 2005) According to Schmidt and Lovell (1979), allocative inefficiency exists when the factor price, i.e the marginal cost of

an input is not equal to its marginal revenue product This leads to the inefficient production process Thus, a firm is called “allocatively efficient” when it can minimize costs of production by selecting the right proportions of inputs The decrease in costs of production may result in the increase in output levels (while the input levels and costs unchanged) or the decrease in input levels (while the output level is fixed); both cases lead to the improvement in productivity Graphically, there are circumstances that firms are fully technically efficient (the output value lies on the efficient isoquant) or allocatively efficient (output value lies on the isocost line) Combining AE with TE, they create a new measure called economic efficiency, or overall efficiency Economic efficiency occurs when the output value is the point of contact between isoquant curve and isocost line (see Figure 4)

Empirically, most of Canadian manufacturers benefited from AE according to the research of Hamit-Haggar (2011) On the other hand, estimation of Kim and Han (2001) for Korean firms expressed that AE had a negative impact on TFP growth In other words, there were an inefficient allocation of inputs in production They also implied that the degree of capital market distortion might be the cause of AE difference across industries: The allocative inefficiency was more clearly observed in industries supported by the government However, allocative efficiency is almost unlikely to be estimated in empirical studies when there is the unavailability of data

on costs and prices (Sharma et al., 2007)

Figure 4: Technical efficiency and Allocative efficiency

Source: Coelli et al (2005)

where

SS’: The isoquant curve, representing the possible input

combinations to produce a given output level when the firm

is fully efficient AA’: The isocost line, representing the combinations of inputs

that minimize the costs

x1, x2: input q: output R: The firm is allocative efficient but not technically efficient Q: The firm is technically efficient but not allocative efficient Q’: The firm is both technically and allocative efficient,

indicating that it reaches full economic efficiency

2.2 Approaches to measure and decompose TFP growth

2.2.1 Primal or dual approach with production, cost, or profit function

From neo-classical perspective, TFP growth can be measured as the residual factor by deducting input growth (labor and capital) from output growth, expressed

in the famous model of Solow (1957), known as “growth accounting” The original Solow model is also called the “primal residual” approach Studies adopting primal approach often use product functions to measure TFP The production function of a firm is considered the technological possibilities of that firm to produce an output using some amounts of inputs A production function should have several properties such as non-negativity, weak essentiality, monotonicity, and concavity (Coelli et al., 2005)

Meanwhile, the cost function is set out to find the input quantities that minimize costs from the set of all technically feasible input-output combinations, given the current technology Similarly, the profit function solves the problem of profit maximization obtained from given amounts of input The methods adopting cost frontiers or revenue (profit) frontiers can be more useful because they can measure economic efficiency, not only technical efficiency (Coelli et al., 2005) Using cost or profit function also means adopting the “dual approach” For example, Hsieh (2002) used the “dual residual” approach including both quantities and costs

of factors of production to calculate TFP growth of four East Asian countries during 1966-1991 His estimates seem to explain more exactly the economic growth of these Asian countries because costs of factors reflect actual market conditions better than quantities However, the dual approach is data demanding; it requires the information

of factor prices of production, which are often difficult to obtain in Vietnamese young industries like high-tech For that reason, primal approach using production function

is preferred when studying TFP growth of Vietnamese high-tech manufacturer because it only needs data on quantities of inputs and outputs

2.2.2 Stochastic and deterministic approaches

Deterministic production functions do not allow random events or other factors to impact on output, i.e any deviation from the production frontier will be considered as an inefficiency (Coelli et al., 2005) Obviously, this approach will overestimate the technical inefficiency (TI), leading to underestimating the contribution of technical efficiency change into TFP growths (or shrinks) To solve the problem, another random variable is introduced to represent statistical noise That kind of frontier is called stochastic frontier Indeed, stochastic frontier models take both inefficiency and statistical noise in account to explain why the production (or cost or profit) is not going along with the frontier (Sharma et al., 2007) In other words, it allows for calculating TFP changes and its component sources in a stochastic environment

2.2.3 Parametric and non-parametric methods

TFP improvements (recession) and its decomposition can be calculated by adopting parametric or non-parametric methods Popular non-parametric approaches that can be mentioned are index number techniques and data envelopment analysis (DEA) Regarding index number techniques such as Fisher’s (1922) or Törnqvist’s (1936) indices, the main advantages are that they are easy to calculate and need only two observations (two firms/industries or two periods of time of the same firm) (Kumbhakar & Lovell, 2000) However, those techniques cannot answer the question about sources of TFP change To solve this, one can conduct DEA, which has strong points that it does not need the specific functional form of the production (or cost or profit) function Nevertheless, DEA estimators cannot separate the impacts of random shocks and inefficiency from the change of TFP and also not applicable for time series dataset (Coelli et al., 2005)

Meanwhile, parametric approaches need the distributional form of the inefficiency term and statistical noise as well as the restrictions on the underlying technology (Coelli et al., 2005) There are several parametric methods that researchers often apply, which are Least Squares (LS) econometric production

models, or Stochastic frontiers (SF) Each of them has its own strengths and drawbacks For instance, LS models are simple but they are built based on the deterministic approach, which means all variation in output not associated with variation in inputs will be considered TI (Kumbhakar & Lovell, 2000) Meanwhile, SFs can separate that variation to random events and technical efficiency

SF estimators including exogenous inefficiency determinants can be obtained

by using Ordinary Least Squares (OLS) or Maximum likelihood (ML) SFs can also

be estimated in two-step procedure (OLS at first to estimate the slope parameters then

ML at second to estimate the intercept and the variances of error terms) or one-step procedure (ML only)

OLS is the easier method to calculate than the latter There is, however, the downward bias in the normal OLS estimators of intercept coefficients, Modified Least Squares or Corrected Least Squares should be applied to shift up the biased OLS intercept parameter (Kumbhakar & Lovell, 2000) Meanwhile, despite the requirements of distributional forms and computational issues in reaching convergence, ML is argued to be the better choice when analyzing large sample (Coelli et al., 2005)

2.3 A review of alternative Stochastic Production Frontier (SPF) models

It can be easily recognized that panel data is more informative than sectional data One can adapt panel data estimation techniques to relax some strong assumptions of cross-sectional data such as the distribution or independence of the TI error terms (Kumbhakar & Lovell, 2000) The literature on panel data stochastic production frontiers can be divided into two main categories based on the assumption that TI effects change or unchanged over time Representatives of time-invariant models are Pitt and Lee (1981), Schmidt and Sickles (1984) and Battese and Coelli (1988), whereas papers written by Cornwell, Schmidt, and Sickles (1990), Kumbhakar (1990), Battese and Coelli (1992), Lee and Schmidt (1993), and Greene (2005) can exemplify time-varying models Besides, TI can be specified as a function

cross-of exogenous factors such as the models cross-of Battese and Coelli (1995) and Huang and

Liu (1994) to permit the pattern of TI effects to not only change over time but also vary across firms

2.3.1 Time-invariant models

Pitt and Lee (1981) are considered the pioneers in extending SPF to panel data Allowing the TI vary across firms but unchanged through time, they estimated parameters using ML and assumed the TI error component to follow normal-half normal distribution, namely:

Afterward, this model is generalized to normal-truncated normal case in Battese and Coelli (1988) with (4) is replaced by 𝑢𝑖~𝑁+(𝜇, 𝜎𝑢2) (𝜇 = 0 is the special case of the distribution function, which makes it become Pitt and Lee’s (1981) model)

SPF estimators can also be performed using fixed-effects techniques like the model in Schmidt and Sickles (1984):

After estimating the parameters, the estimators of the fixed-effects 𝑢𝑖 can be obtained from:

If N is large and T is small, GLS can be used to estimate the regressors, then

𝑢𝑖 can be obtained from:

2.3.2 Time-varying models

Cornwell et al (1990) replaced the firm-specific effects 𝛼𝑖 in (5) by a function

of time allowing parameters to vary across firms:

𝛼𝑖𝑡 = 𝜃𝑖1+ 𝜃𝑖2𝑡 + 𝜃𝑖3𝑡2,

(13)

This specification allows the TI to change through time while still having specific characteristics However, this pattern requires Nx3 parameters to estimate firm effects, which leads to the impact on degree of freedom (Belotti, Daidone, Ilardi,

firm-& Atella, 2012)

To reduce the numbers of parameters, alternative of time-varying TI effects were proposed by Kumbhakar (1990), Battese and Coelli (1992), and Lee and Schmidt (1993) with the general function of 𝑢𝑖𝑡 as following:

where θ𝑡 is a set of time dummy variables

(17)

In (15), there are only two parameters b and c to estimate It allows 0 ≤𝑔(𝑡) ≤ 1 and 𝑔(𝑡) can be concave or convex, monotonically increasing or decreasing However, the estimation and inference of technical efficiency change can

be complicated

In (16), 𝜂 is the unknown scalar parameter representing the rate of change in

TI If 𝜂 > 0, 𝑢𝑖𝑡 decreases as t increases, and vice versa; while 𝜂 = 0 means

time-invariant TI, i.e the firm-specific effects This model has the advantage that there is only one parameter 𝜂 that needs to be estimated and it is easy to conclude about the technical efficiency as well as its change over time However, this specification only allows TE to increase at a decreasing rate (𝜂 < 0) or decrease at an increasing rate (𝜂 > 0)

In (17), the parameters can be easily obtained because it does not need any parametric form of 𝑢𝑖𝑡 Nevertheless, this model should not be applied in cases of large T because the number of parameters to be estimated is also a large figure

The three above models (15) - (17) can solve the problem of reducing the parameters but the changes of 𝑢𝑖𝑡 over time are the same between producers, which

is a restriction that makes it less flexible than Cornwell et al.’s (1990) model It also means the bias in 𝑢𝑖𝑡 estimation (Belotti et al., 2012) Arguing that taking account of time-invariant latent variables into the inefficiency effects while those factors are not associated with the production process can create a misspecification bias, Greene (2005) introduced the “true” fixed-effects and “true” random-effects model to separate the heterogeneity from efficiency:

“True” fixed-effects model: 𝑦𝑖𝑡 = α𝑖+ 𝑥𝑖𝑡β + 𝑣𝑖𝑡 − 𝑢𝑖𝑡, (18)

where α𝑖 is the group specific constant

Belotti et al (2012) gave advice that model (18) requires a long panel (𝑇 ≥ 0)

to make the estimated parameters α𝑖 consistent Moreover, α𝑖 in (18) creates a large dimension of parameters if the number of firms is large, which leads to a computational issue Thus, a Maximum Likelihood Dummy Variable (MLDV) approach should be used with caution (Belotti et al., 2012; Greene, 2005)

Despite its flexibility among other time-varying models, one can argue against model (19) that the some portions of latent heterogeneity do impact on inefficiency (Belotti et al., 2012) Therefore, the decision to disentangle those two parts or not depends on the characteristics of data

2.3.3 Exogenous inefficiency determinants

One can argue that time-varying technical efficiency may not only stem from the passage of time but also from other variables, which may be or may be not production inputs and outputs but can still affect the productivity of the firm (Kumbhakar & Lovell, 2000) Examples of those exogenous variables are the firm’s

characteristics (ownership, firm age, firm size, location), or policy variables (government regulations) If that observable heterogeneity is not controlled in the SPF, the estimated parameters of technical efficiency can be biased and affect the post-estimation inference (Belotti et al., 2012) Nevertheless, there is no common agreement about the impacts of those determinants For instance, Nguyen et al (2007) showed that there was a positive relationship between firm size and technical efficiency of small and medium manufacturing firms in Vietnam, whereas estimation results in the study of Badunenko, Fritsch, and Stephan (2006) about German firms indicated that larger firms tend to be less technically efficient than small ones Another example of the controversy in effects of exogenous variables is findings about effects of foreign ownership on TE: negative in findings of Pham, Dao, and Reilly (2009) but positive in estimates of Zhou (2014) Thus, inference about exogenous sources of TE should be made with caution

Regarding the estimation methods, from the early approach, the two-step procedure was used, which means inefficiency then exogenous effects are estimated

in sequence However, as pointed out in Wang and Schmidt (2002), there is strong evidence in favor of the one-step procedure (exogenous effects are estimated simultaneously with other parameters of the model)

The models of Huang and Liu (1994) and Battese and Coelli (1995) are considered the most well-known ones that apply one-step approach to incorporate exogenous effects into efficiency by parameterizing the mean 𝜇𝑖𝑡 of the pre-truncated distribution, i.e 𝑢𝑖𝑡 𝑖𝑖𝑑 𝑁+(𝑧𝑖𝑡𝛿 + 𝑧𝑖𝑡∗𝛿∗, 𝜎𝑢2) The general function of technical efficiency effects is defined as:

𝑢𝑖𝑡 = 𝑧𝑖𝑡𝛿 + 𝑧𝑖𝑡∗𝛿∗+ 𝑤𝑖𝑡,

(20)

where 𝑧𝑖𝑡 is a 1xM vector of exogenous determinants

𝛿 is an Mx1 vector of unknown parameters that need to be estimated

𝑧𝑖𝑡∗ is a vector of interaction between exogenous variables 𝑧𝑖𝑡 and input variables 𝑥𝑖𝑡

𝛿∗ is a vector of unknown parameters to be estimated

𝑤𝑖𝑡 𝑖𝑖𝑑 𝑁+(0, 𝜎𝑤2) is the random unobserved component such that 𝑢𝑖𝑡 ≥ 0

In the study of Battese and Coelli (1995), there is no inputs variables in the specification of 𝑢𝑖𝑡 Therefore, all estimators of 𝛿∗ in (20) are equal to zero Meanwhile, Huang and Liu (1994) indicated that TI comes from two main parts: firm-specific characteristics and the intensity of input use They argued that producers may gain “more information, knowledge, and experience with respect to one input productivity than another” Thus, the coefficients of 𝛿 and 𝛿∗ in (20) are different from zero The model of Huang and Liu (1994) is called non-neutral stochastic because the effects of TI effects on productivity of firms will be biased toward some inputs other than others

Instead of the mean parameterization, parameterizing the variance 𝜎𝑖𝑡2 of the pre-truncated distribution, i.e 𝑢𝑖𝑡 𝑖𝑖𝑑 𝑁+(𝜇, 𝜎𝑖𝑡2) where 𝜎𝑖𝑡2 = exp(𝑧𝑖𝑡𝛾) , is also a way to analyze the effects of exogenous determinants on inefficiency as well as solve the problem of heteroscedasticity (Kumbhakar & Lovell, 2000) It can be understood that those determinants affect on the production risks, representing by the variance

𝜎𝑖𝑡2 (Coelli et al., 2005; Kumbhakar & Lovell, 2000) Caudill, Ford, and Gropper (1995) and Hadri, Guermat, and Whittaker (2003) are among the authors supporting this approach

As stated in Wang (2002), the approach of Battese and Coelli (1995) and Caudill et al (1995) can be combined, which implies that

𝑢𝑖𝑡 𝑖𝑖𝑑 𝑁+(𝜇𝑖𝑡, 𝜎𝑖𝑡2), 𝑤ℎ𝑒𝑟𝑒 𝜇𝑖𝑡 = 𝑧𝑖𝑡𝛿 + 𝑧𝑖𝑡∗𝛿∗ 𝑎𝑛𝑑 𝜎𝑖𝑡2 = exp(𝑧𝑖𝑡1𝛾) 𝑧𝑖𝑡 and 𝑧𝑖𝑡1 are not necessarily the same vector of variables

Caudill et al (1995) also propose 𝜎𝑣2 = exp(𝑧𝑖𝑡𝜆) to solve the problem of heteroscedasticity of random error term component 𝑣𝑖𝑡 However, as clearly stated in Kumbhakar and Lovell (2000), the heteroscedasticity of 𝑣𝑖𝑡 does not cause severe consequences on the estimators Thus, the parameterization of 𝜎𝑣2 seems to be neglected in the literature of efficiency measurement

2.3.4 TFP growth decomposition

As mentioned in the concepts part, TFP growth can stem from many sources such as Technological progress (TP), Technical efficiency change (TEC), Scale change effects (SCE), and Allocative efficiency change (AEC)

The decomposition process starts with the general production frontier:

𝑦𝑖𝑡 = 𝑓(𝑥𝑖𝑡, 𝑡) exp(−𝑢𝑖𝑡) ,

(21)

From now on, the ‘it’ subscripts are omitted for simplicity

Taking logarithm of both sides of (21) and totally differentiating with respect

to time, the output growth rate is yielded as:

where a dot over a variable means the time rate of change of that variable

𝑇𝑃 = 𝜕𝑙𝑛𝑓( )/𝜕𝑡 is the primal rate of technological change 𝜀𝑗 = 𝜕𝑙𝑛𝑓( )/𝜕𝑥𝑗 is the

output elasticity of the j th input 𝑇𝐸𝐶 = −𝑑𝑢/𝑑𝑡 is change in technical efficiency

Then, TFP change is defined as output growth unexplained by input growth:

where 𝑅𝑇𝑆 = ∑ 𝜀𝑗 𝑗 is the estimates of returns to scale, 𝜆𝑗 = 𝜀𝑗/𝑅𝑇𝑆; 𝑆𝐶𝐸 =(𝑅𝑇𝑆 − 1) ∑ 𝜆𝑗 𝑗𝑥̇𝑗 measures the change in scale of production; and 𝐴𝐸𝐶 =

∑ (𝜆𝑗 𝑗 − 𝑆𝑗)𝑥̇𝑗 is the change in allocative efficiency However, when it is impossible

to collect the price data, allocative efficiency cannot be estimated In that case, according to Kumbhakar and Lovell (2000), it is assumed that 𝑆𝑗 = 𝜀𝑗

𝑅𝑇𝑆= 𝜆𝑗, then (24) becomes:

TP can be positive (negative) under progressive (regressive) technological change It will be eliminated if the technology does not change over time This term can be decomposed to neutral (autonomous change) and biased components (depending on the intensity of input use)

Regarding the effects of changing scale of production, the sign of SCE depends on signs and magnitude of both output elasticity and rates of input change over time It is positive (negative) if there is IRS (DRS) and increasing (decreasing) use of inputs If the assumption of CRS holds (RTS=1), SCE equals zero, then the formula will be the same as the one in Nishimizu and Page (1982)

The estimation of changes in technical efficiency is more complicated, as mentioned in previous parts Various specifications of 𝑢𝑖𝑡 will produce different estimates of TEC For example, if 𝑢𝑖𝑡 is defined as in (14) and (16), TEC can be estimated using:

𝑇𝐸𝐶 = −𝑑𝑢

𝑑𝑡 = 𝜂 × 𝑢,

(26)

If 𝑢𝑖𝑡 is specified as in (20), then TEC will be obtained from:

𝑇𝐸𝐶 = −𝑑𝑢

𝑑𝑡 =

𝑑𝑙𝑛𝑇𝐸(𝑥, 𝑧, 𝑡)𝑑𝑡

𝑗

+ ∑𝜕𝑙𝑛𝑇𝐸(𝑥, 𝑧, 𝑡)

𝜕𝑧𝑚

𝑑𝑧𝑚𝑑𝑡

The right-hand side of (27) includes three parts: the first part of is the passage

of time, the second part is the effects due to changes in input use, and the last part is the effects due to changes in the firm-specific characteristics If Battese and Coelli’s (1995) model is applied, then the second part is omitted If the firm-specific characteristics do not change over time, the last part equals zero If both second and third parts equal zero, it means that the technical efficiency change is neutral

According to Wang (2002), marginal effects of exogenous factors can be monotonic, which implies that exogenous determinants can impact positively and negatively on the technical efficiency of the firm For instance, when the firm is young, its performance may be inefficient due to lacking of experience Thus, it can improve the TI over time through learning by doing Afterward, the increase rate of efficiency will gradually reduce, especially when the firm exists for a long time that

non-it becomes less flexible to unexpected shocks The models of Caudill et al (1995), Hadri et al (2003), or the combined model in Wang (2002) allow for the non-monotonic efficiency effects Nevertheless, the non-monotonicity does not exist in Huang and Liu’s (1994) and Battese and Coelli’s (1995) models because the specification of 𝑢𝑖𝑡 restricts that all of the sample’s observations have either efficiency-improving or efficiency-reducing (Wang, 2002)

CHAPTER 3 OVERVIEW OF VIETNAMESE HIGH-TECHNOLOGY

MANUFACTURING SECTOR

This chapters attempts to define and classify HT manufacturing industries in Vietnam Afterwards, the current status of Vietnamese HT manufacturing sector, including contribution, pattern, and trends, will be described briefly

Organization for Economic Co-operation and Development [OECD] (2005) identifies that “technology is a stock of (physical or managerial) knowledge which makes it possible to make new products or new processes” Then, it implied that high technology is the most state-of-the-art knowledge available, which make more rapid progresses than other and requires continuous huge efforts in R&D with a strong technological base Law on High technologies, approved by the National Assembly

of Vietnam in 2008, also mentions high technology as “high scientific research and technological development content” and this kind of technology can make

“environmentally friendly products of superior quality and utilities and high added value”

Defining a HT firm, Hecker (1999) used the definition of the U.S Congressional Office of Technology Assessment in his paper: Firms that design, develop, and introduce new products and innovate new production processes by applying scientific and technical knowledge as well as advanced techniques are considered HT firms He also indicated that R&D budgets of HT firms are usually large, and many experts in science, technology, or engineering are hired in those firms Meanwhile, Vietnamese Law on High technologies (2008) provides a simpler definition: an enterprise is classified as HT when it produces HT products, provides

HT services and conducts R&D activities relating to HT

Regarding the definition of HT industries, according to Vietnamese Law on High technologies (2008), which industries manufacturing HT products and providing HT services will be considered as HT OECD (2005) makes it clearer when

stating that both intensively producing and utilizing technology belong to high-tech sectors Using graphics to demonstrate, the current state of technology in the industry can be represented by the production frontier, which is a set of maximum outputs

attainable from each input level (Coelli et al., 2005) (see Figure 3)

Despite the fact that “high-technology” or “high-tech” has been agreed in general definition and used popularly, it is difficult to reach a common agreement on the classification of industries, products, or even employment into low-tech, medium-tech, and high-tech sectors There is still a controversy in this field regarding the approaches and criteria Some classifications are based on input-based criteria such

as the embodied R&D expenditures relative to the goods value (Davis, cited in Mani, 2000) or R&D intensity, i.e R&D spending in relation to value added (OECD, cited

in Kask & Sieber, 2002) Other researchers used the products’ natures (output-based criteria) for their HT classification (Kask & Sieber, 2002) Each approach has its own strengths and limitations For instance, Mani (2000) criticized that using R&D expenditure as a criteria to consider some products as HT may be reasonable in 1960s but not in 1980s because the kind of technology adopted at that time would be not advanced anymore However, OECD (2005) also stated that R&D intensity is the only criterion to conduct quantitative research due to the issues of data availability when one attempts to apply other criteria to classify industries into high-tech sector Thus, this study adopts the classification of OECD (2005) to determine the HT industries Specifically, based on OECD (2001) and GSO (2007), HT industries includes (i) Pharmaceuticals; (ii) Computers and peripherals; (iii) Radios, TVs, and communication equipment; (iv) Precision instruments; and (v) Aircrafts (see Appendix 1 for more details)

HT manufacturing sector in Vietnam has a bright prospect due to high demand

in the world market Not only the demand comes from consumption of individual customers, products of this sector also benefit other manufacturing and services industries such as online commerce and internet banking Indeed, world exports of

HT products, which can be considered as a measure of global demand for HT

products, keeps increasing considerably over the recent decade (see Figure 5)

Following that trend of the world, Vietnamese HT manufacturing sector also contributes to the enlarging proportion in the industrial structure over years (see

Table 1)

Figure 5: World exports & value-added of HT manufacturing sector

(2001-2012)

Source: Appendix tables 6-7 and 6-21 of Science and Engineering Indicators

2014 (National Science Board, 2014)

Table 1: Contribution of Vietnamese HT in value added of manufacturing

sector during 2000–2012 (in percentage)

HT manufacturing 4.22 3.68 3.99 4.38 5.04 4.92 4.77 4.91 5.10 5.37 6.49 7.44 6.03

Source: Appendix tables 6-7 and 6-14 of Science and Engineering Indicators

2014 (National Science Board, 2014)