A study on financial performance of IT companies in japan using data envelopment analysis (DEA) model

MASTER IN FINANCIAL MANAGEMENT A study on financial performance of IT companies in Japan using Data Envelopment Analysis (DEA) Model Graduate student Tran Van Minh Supervisor Dr Mai Anh HANOI, 2021 I[.]

MASTER IN FINANCIAL MANAGEMENT

A study on financial performance of IT companies in Japan using Data Envelopment

Analysis (DEA) Model

Graduate student: Tran Van Minh Supervisor: Dr Mai Anh

HANOI, 2021

ABSTRACT

Thesis Title: A study on financial performance of IT companies in Japan using

Data Envelopment Analysis (DEA) Model

Pages: 66

University: Vietnam National University, Hanoi

Graduate School: International School

Date: October 2020 Degree: Master

Graduate Student: Tran Van Minh Supervisor: Dr Mai Anh

Keywords: Data Envelopment Analysis, DEA models, CCR, BCC, Efficiency

analysis, OTE, PTE, SE, IT industry, Japanese

Japan IT industry has achieved exponential growth during the last one decade mainly due to government policy support, availability of trained manpower and high demand of IT products and services in the international and domestic markets However, during this period, the industry also witnessed several ups and downs, including the recent slowdown Rise and fall in the economic activities in domestic and foreign economies may pose greater risk and challenges to the industry In this context, constant monitoring and improving performance of individual companies and setting benchmarking for relatively inefficient firms become crucial for the growth and sustenance of the industry For this, measurement of the relative performance of individual IT firms and setting best practice benchmark for the relatively under-performed company is quite relevant Although, there is no dearth

of studies on the Japanese IT industry, but the studies related to these aspects are, of course, scant

This thesis is the outcome of my MFM Master project research carried out

at International School of Vietnam National University, Hanoi over the course period In these preliminary lines, I would like to express my great gratitude

to those who have contributed to the thesis for their understanding, encouragement, and support

First of all, my special words of appreciation and deepest recognition go to my thesis mentor/supervisor, Doctor Mai Anh I am sincerely grateful to him for his support and effort spent on my research during the past months I benefited enormously from his invaluable expertise, insightful comments, and excellent supervision In addition, he was continuously providing me numerous valuable comments and suggestions

I would also like to express my gratitude to lecturer Tâm for providing me a solid base of financial knowledge, which is not only useful for this research but also for future development in social life and career

Finally, a thank you to Định, Giang, Yến and all classmates who gave their time and comments The contributions that were made proved to be very valuable in conducting this research study

Finally, it would be impossible to say enough about my dear family and beloved friends, always trying to support and giving important methodology advices to overcome the most difficult time in order to complete it on time

Thank you!

Author Tran Van Minh

TABLE OF CONTENTS

ABSTRACT i

ACKNOWLEDGEMENT ii

LIST OF TABLES iv

LIST OF FIGURES/ GRAPHS v

LIST OF ABBREVIATIONS vi

CHAPTER 1 INTRODUCTION 1

1.1 Research background 1

1.2 Research Objectives 2

1.3 Hypotheses 2

1.4 Research Scope 2

CHAPTER 2 LITERATURE REVIEW 5

2.1 Performance Measurement 5

2.2 Theoretical Studies on DEA 6

2.3 Performance and Efficiency Studies on IT sector 8

2.4 Summing Up 13

CHAPTER 3 RESEARCH METHODOLOGY 14

3.1 Data Collection 14

3.1.1 Data and Variables 14

3.1.1.1 Output variable 15

3.1.1.2 Input variables 15

3.1.2 The Sampling Method and Size 15

3.2 Data Envelopment Analysis Concepts 17

3.3 Growth of DEA 17

3.4 Advantages of DEA 20

3.5 Limitations of DEA 20

3.6 DEA Models 21

CHAPTER 4 RESULTS AND DISCUSSIONS 26

4.1 IT Industry in Japan 26

4.2 Results and Discussions 31

4.3 Suggestion for Efficiency improvement of IT companies in Vietnam 45

CHAPTER 5 CONCLUSIONS 47

5.1 Conclusions 47

5.2 Limitations 48

5.3 Future Research 48

REFERENCES 49

APPENDICES 55

LIST OF TABLES

Table 3.1 Descriptive Statistics of Output and Input Variables 43

Table 3.2 Correlation Matrix of Output and Input Variables 44

Table 4.1 Descriptive Statistics of Efficiency Scores (2018-19) 45

Table 4.2 Analysis of OTE, PTE and Scale Efficient Companies (2018-19) 46

Table 4.3 OTE Scores of IT Companies 48

Table 4.4 PTE Scores of IT Companies 49

Table 4.5 Company-wise Scale Efficiency and RTS in the IT Industry 50

Table 4.6 Company-wise Slacks in the Input Variables 53

Table 4.7 Comparison of Efficiency of IT Companies by Scale-Size 55

LIST OF FIGURES/ GRAPHS

Figure 4.1 Frequency Distribution of OTE, PTE and SE of IT Companies in 2018-19 51

LIST OF ABBREVIATIONS

to the Japanese companies In order to maintain a global leading position in the current liberalized regime, Japanese IT industry has to maintain its competitive edge by persistently improving its performance

One of a company's key performance indexes is Financial performance Financial performance is the main performance category used in assessment of business performance It

is the measure of a business organization’s capacity of using its resources to create profit It is the achievement of the company's financial performance for a certain period covering the collection and allocation of finance measured by capital adequacy, liquidity, solvency, efficiency, leverage and profitability Financial performance shows the company's ability to manage and control its own resources Any change of finance can be the basis of information for corporate managers to make decisions

Performance evaluation plays a strategic role in IT companies, in order to address the best use of resources and rationing of demand The evaluation of technical efficiencies of existing companies is necessary to improve the companies' financial performance, so as to employ human resources effectively and make the finance more efficient and sustainable

Performance of IT industry, among others, depends on the overall growth of domestic economy and the economies of those countries where its products and services are exported Therefore, rise and fall in the economic activities in domestic and foreign economies may pose

greater risk and challenges to this industry In this context, constant monitoring and improving performance of individual companies and setting benchmarking for relatively inefficient ones become crucial for growth and sustenance of the industry Keeping this in view, the present study examines the relative performance of individual IT companies and set the best-practice benchmark for the relatively under-performed companies For this purpose, technical and scale efficiencies have been measured through a non-parametric method, known as Data Envelopment Analysis (DEA)

1.2 Research Objectives

The study carries following objectives:

- Measure the technical efficiencies of the IT companies in Japan to evaluate financial performance of the companies

- Suggest improvement of the efficiency in the IT industry, and particularly some lessons for the IT companies in Vietnam, like Hybrid Technologies

1.3 Hypotheses

The following hypotheses are tested:

- H1: Company size is positively associated with the technical efficiencies of the IT companies

- H2: A decrease in the employee cost has positive impact on the efficiency of the IT companies

- H3: An increase in the sales turnover to NFA ratio reduces the inefficiency level in the Industry

1.4 Research Scope

Efficiency, effectiveness, productivity, profitability, quality etc., are the different kinds of performance measures applied by decision-making units Each measure indicates to the level of performance of different activities

IT industry comprises computer hardware, software and IT and IT-enabled Services (ITES) such as IT-BPOs The scope of this study is limited to software companies only

The study analyses the data collected from the Japanese Software companies The focus is on assessment of technical efficiency, proper utilization of the resources and benchmarking of the individual companies through applying DEA Models

Apart from these, the study also examines technical efficiency change, technological change, and productivity improvement in the software companies and identifies the various factors that are accountable for the inefficiency in the IT companies As the term “performance”

is interpreted in different manner and used in different connotations; this study measures the performance of individual IT companies only in terms of technical efficiency Other aspects related to performance of the companies are beyond the scope of the present study

1.5 Research Methodology

This study is based on the data collected from various published sources A list of all the IT companies available in the database is prepared along with their annual sales turnover Initially data from 85 IT companies were collected However, all the companies could not be considered for the study because of two reasons First most of the companies did not report data

on all the required input and output variables Therefore, companies having missing data were dropped from the analysis Second, as the IT industry is a very dynamic industry, new and new companies were added during the study period Therefore, recently incorporated companies did not have the past data After taking care of these issues, finally 85 IT companies were selected for the study for the financial year 2018-19

However, since DEA measures the relative efficiency of individual companies and estimated efficiency scores are sensitive to the outlier companies, we identified the outlier companies through applying the super-efficiency DEA model After excluding the outlier companies from the DEA analysis, our dataset reduced to 75 companies Thus, analysis is based

on the input-output data collected from the 75 companies for the year 2018-19 These companies are classified as small, medium and large according to their annual sales turnover Companies having annual sales turnover less than less than JP 100 billion yen are termed as small, between

JP 100 billion – 1,000 billion yen as medium and above JP 1,000 billion yen as large

In order to neutralize the inflation effect and make the data comparable; time series data are transformed at constant prices The study applies an approach for performance assessment of individual companies Relative efficiency of individual companies is estimated using DEA models Three types of efficiencies are estimated, namely overall technical efficiency, pure technical efficiency and scale efficiency These efficiencies are estimated taking key output and inputs variables

1.6 Research Structure

This research consists of five chapters including an introduction, literature review,

research methods and methodology, data analysis and research findings, conclusions and

discussion They can be described as below:

1 Chapter 1 – Introduction:

The introduction is designed to provide the general structure of the dissertation such

as background, aim of the research, the significance of this study, brief limitations and how the

data will be analyzed

2 Chapter 2 – Literature Review:

The chapter presents relevant background information to form an academic base and

summarizes prominent theories and previous researches on measuring the performance and

efficiency of IT companies

3 Chapter 3 – Research Methodology:

This chapter presents Data collection and the DEA methodology, its origin,

advantages and limitations

4 Chapter 4 – Results and Discussions:

The chapter interprets data and provides detailed efficiency analysis, benchmarks of

efficient companies based on the cross-sectional data collected from 75 IT companies for the

financial year 2018-19

5 Chapter 5 – Conclusions:

Based on the results attained from the previous chapter, appropriate recommendations

for listed company, the chapter presents conclusion and suggests improvement for the industry

CHAPTER 2 LITERATURE REVIEW

2.1 Performance Measurement

Performance measurement is critical component in the general management process Reliable measurement system constitutes a sound basis for continuous monitoring and control of organizational performance Efficiency, effectiveness, productivity, profitability, quality etc., are the different kinds of performance measures applied by decision-making units (Ray, 2004) Each measure indicates to the level of performance of different activities

Cinca CS et al (2005) stated that efficiency analysis is vital managerial control tool for assessing the degree to which inputs are utilized in the process of obtaining desired output Economic approach assumes the existence of specific input-output relations, which can be identified by the analysis of large body of data Efficiency due to this approach is evaluated through production function, which formulates the assumed relations An engineering approach is one where efficiency is measured by comparing performance to suitable set of engineering standards There are some limitations of these efficiency measurement approaches These techniques cannot handle multiple output and input variables together

Moreover, it is usually extremely difficult to assign proper relative weights to inputs and outputs With fixed weights, it is very difficult to formulate an explicit functional relationship between inputs and outputs Another problem related to Least Square regression methods is that they are based on measurement of central tendency and failed to explain the behavior of individual Decision-Making Units (DMUs)

Therefore, this study applies DEA technique to measure the efficiencies and set benchmarking for monitoring the performance of inefficient IT companies

DEA was developed by Charnes et al (1978) and extended by Banker et al (1984) The basic DEA model defines efficiency as ratio of the weighted sum of output variables to weighted sum of input variables It has an inherent advantage in its ability to separate DMUs that define production frontier from those which are below the frontier and its ability to relate original outcomes to the sources utilized It is a non-parametric approach and can accommodate wide- ranging behavior in application

Unlike parametric approaches, it does not assume specific functional form of the distribution of the underlying data All deviations are assumed to be due to inefficiency

(Banker et al., 1984) It provides a measure of efficiency that allows inter-company comparison The DEA method has certain advantages over the stochastic frontier approach, namely, it does not impose any functional form on production or technology In case of the stochastic frontier approach, the parameter estimates are sensitive to the choice of the probability distributions specified for the disturbance terms

2.2 Theoretical Studies on DEA

Since the publication of seminal papers by Charnes, Cooper and Rhodes (1978) and Bankar, Charnes and Cooper (1984), many theoretical and applied studies have been published on DEA Although DEA was developed to measure the efficiency of public sector non-profit decision-making units (DMUs) such as educational institutions, healthcare sector, public utilities, government departments, etc.; it is now widely being used to evaluate the efficiency of both profit and non-profit organizations in public and private sectors in all sectors of the economy (service, industry and agriculture)

The CCR model is based on the constant returns to scale (CRS) technology assumption It does not take into consideration the effect of scale-size on the efficiency of a DMU It implies that when a DMU is operating at CRS, output will increase at the same rate

as inputs are increased in the production process This happens when a DMU is operating at the optimum scale However, a DMU may be too small or too big relative to the optimum size and therefore may be in a disadvantageous position vis-à-vis to those that are operating

at the optimum scale

Keeping this problem in view, Banker, Charnes and Cooper (1984) extended the CCR model by adding the convexity assumption and measured the pure technical efficiency (PTE)

by neutralizing the effect of scale-size on the efficiency The technical efficiency measured

by this model is actually the managerial efficiency, which is pure conversion of inputs into outputs, irrespective of the size of the DMU The efficiency measured through the BCC model helps in decomposing the overall technical efficiency (OTE) into pure technical efficiency (PTE) and scale efficiency (SE) by dividing the CCR efficiency from the BCC efficiency Thus, PTE will always be ≥ the OTE These two basic DEA models have been further modified and extended by the researchers

On theoretical aspects, weight restriction, categorical inputs and outputs, non- discretionary inputs and outputs, negative outputs, super efficiency, input congestion, sensitivity analysis, slack-based models are the major extension of DEA Some of the major

studies on the theoretical advancement in the DEA are: Banker et al (1984), Charnes et al (1997), Banker and Morey (1986a and 1986b), Charnes et al (1985), Thompson et al (1990), Wong and Beasley (1990), Ali and Seiford (1993, Andersen and Petersen (1993), Thompson

et al (1990), Zhu (1996), Seiford and Zhu (1998), Pondinovski and Athanssopoulas (1998), Dyson and Thanassoulis (1988), Hibki and Sueyoshi (1999), Sueyoshi, et al.:1999, Tone (2001 and 2002), and Rouse and Lovell (2003)

CCR and BCC models assume that all the inputs and outputs are discretionary However, some inputs may not be freely disposable and cannot be reduced without cost if a company is to improve efficiency For example, a company cannot retrench the workers without paying compensations, as it must follow the prevailing labor laws To solve this problem, researchers have introduced non-discretionary variables in the DEA model (for instance see, Bankar and Morey: 1986a)

The key to the proper mathematical treatment of a nondiscretionary variable lies in the observation that information about the extent to which a nondiscretionary input may be reduced is beyond the discretion of the decision makers The model incorporating non- discretionary variables are unique in the sense that the radial contraction of inputs or redial expansion of output cannot be applied to them The model eliminates the slacks of non- discretionary variables from the objective function since the decision maker does not have any control over them

Perhaps the most significant extension to DEA is the concept of restricting the possible range for the multipliers The basic DEA models have positivity restriction on the multiplier Since a priori specification of the multipliers is not required, DEA maximizes the weights for various inputs and outputs for evaluating the efficiency of each DMU While doing so, there may be possibility that one input may be assigned unreasonably less weight and the other high weight to evaluate efficiency of a DMU To avoid this problem, Assurance Region (AR) models or restricted multiplier models are developed (see Thompson et al.: 1986; Dyson and Thanassoulis: 1988; Charnes et al.: 1989; Roll, Wong and Beasley: 1990)

Slacks are also considered in the CCR and BCC models with the coefficient of (non- Archimedean) constant in their objective function However, the numeric value for in computations should be chosen to be much smaller than input and output values so that they will not affect optimization After it, many researchers (Ali and Seiford: 1993; Charnes et al.:1999; Sueyoshi, et al.:1999; Tone: 2001) gave the different slack-based DEA models to accumulate the impact of slacks on the efficiency scores Ali and Seiford (1993) suggested

a two-stage optimization procedure to avoid the use of in computations and also considered the slacks in the efficiency scores Tone (2001) provides a new measure of efficiency which deals directly with the slacks His model is known as non-redial SBM model It deals directly with input accesses and the output shortfalls of the DMUs The SBM model is unit-invariant The model by Sueyoshi et al (1999) suggests a slack adjusted model which can directly incorporate the influence of slacks into the measurement of a DEA efficiency score Liu and Tone (2008) proposed a three-stage method to measure DEA efficiency while controlling for the impacts of both statistical noise and environmental factors

Super efficiency DEA model was developed by Andersen and Petersen in 1993 When this model is applied, efficiency of a DMU is estimated excluding it from the reference-set Therefore, efficiency score estimated through super efficiency model may be greater than one This model is used to rank the efficient DMUs and also to identify the outlier DMUs The super efficiency model was further extended by Tone (2002) which is non-redial and directly deals with the slacks

Another extension in the DEA methodology is the window analysis model which was developed by Charnes et al (1985) The window analysis model is used on a panel data using

a moving average analogue, where a DMU in each period is treated as if it were a different DMU The window analysis has several advantages It is useful when sample size is small and DEA does not have sufficient discriminatory power Moreover, it tracks dynamic efficiency trends through successive overlapping windows and thus allows monitoring the performance of DMUs overtime

2.3 Performance and Efficiency Studies on IT sector

There is no dearth of studies on the software and IT industry; however, studies on efficiency and productivity measurement in the IT industry are, of course, scant In this section we review the main studies conducted on efficiency and productivity in the IT industry

Roman (undated) measures technical efficiency and competitiveness in IT industry through DEA method in 9 European countries, including Romania He uses average number

of employees per company, average salary per employee, average GFC per employee and number of companies as input variables and average production per employee and average value added per employee as output variables The study finds Germany and Netherlands as most efficient and Bulgaria and Sweden as less efficient countries Estimated efficiency

scores indicate the possibility to increase output level with given inputs For instance, Germany can produce 37.3% additional output with the use of existing inputs if it operates

at the frontier The findings further reveal that Romania and Bulgaria are operating at increasing returns to scale (IRS) that implies that the IT companies in these countries can increase their future output if their scale size is enhanced

Shao and Lin (2002) examine the effects of information technology on technical efficiency in companies, production process through a two-stage analysis, using data collected from Fortune 500 companies for the period 1988 to 1992 In the first stage, technical efficiency scores are estimated by output-oriented BCC-DEA model, using capital and labour as inputs and value added as output In the second stage, Tobit regression model

is applied to examine the determinants of technical efficiency The empirical results show that the average PTE and SE are 0.735 and 0.915, respectively which imply that average company has the scope of increasing the output with the existing level of capital and labour Further, the study confirms that IT has a significant positive impact on technical efficiency and resultantly gives rise to the productivity growth in the companies

Shao and Shu (2003) estimate the TFP growth in the IT industry of 14 OECD countries for the 13-year period of 1978 through 1990 using DEA-based MPI approach The country-wise data of IT industry were collected from the two databases, namely, OECD Stan Database for Industrial Analysis and the OECD International Sectoral Database Labour and capital are taken as input variables and IT product as output variable to measure the TFP In order to know the sources of TFP growth in the IT sector, the paper also decomposes the TFP growth into technical efficiency change and technological change The results show that among the 14 countries examined, 10 had observed TFP growth in their IT industries It is also found that IT industries in these countries are more productive than other industries when compared with previous studies The study observes that technical change (innovation)

in the IT industry of these countries contributes more to the TFP growth than the technical efficiency change (catch) Moreover, scale efficiency change unfavorably affects productivity for most countries The paper suggests that for the plan-directed countries, it is necessary to first analyse their performances in terms of technological progress, efficiency change, scale change, and productivity and then draw up their specific National IT Policy that is feasible to implement For the market-driven countries, the findings of the study can provide great insights for the IT companies that try to compete in the global market and consider investing and relocating their operations into other countries

Mathur (2007) applies input-oriented DEA model to evaluate the technical efficiency

of 92 Indian Software companies for the year 2005-06 The data for the study was collected from the prowess database of CMIE The study also examines the impact of various determinants on technical efficiency of software companies through Tobit regression analysis The study also applies MPI to estimate TFP change for the common set of software companies for the period 1996 to 2006 The estimated TFP growth is further decomposed into efficiency change (catching up) and technical change (frontier shift) to identify the sources of the TFP growth in the IT companies He also makes a comparative study of the performance of the Indian ICT sector vis-à-vis other frontrunner countries in the IT industry The findings of the study show that there have been some improvements in the TFP, technical efficiency, and technical change in the Indian Software industry during the period under study The results of Tobit regression analysis reveal that company size and net export has positive and significant impact on the technical efficiency, while total cost had the statistically significant negative impact on it Further, number of employees and age of the company do not have any significant impact on the efficiency The study further reveals that size of the company is important for exports, not for the technical efficiency

Lam and Shiu (2008) measure technical efficiency in telecommunications sector of

31 provinces of China through DEA method Each province is considered a DMU The technical efficiency scores of different DMUs are estimated for the period 2003 to 2005 Two models are used in the TFP measurement Model 1 uses business revenue as output variable, while Model 2 uses the total number of subscribers as the output variable Both models, however, use the same input variables, i.e capital and labour Apart from comparing the efficiency among different areas and regions, the paper also examines the presence of input slacks and the reasons for the differences in efficiency scores across provinces The results indicate that the performance of China’s telecommunications sector varies significantly across different regions In general, provinces and municipalities in the eastern region have achieved higher levels of technical efficiency than those in the central and western regions The differences in efficiency scores are mainly due to the differences in the operating environments of different provinces, rather than the efficiency performance of the enterprises The paper finds that labour redundancy and excess capacity of long-distance optical cable lines are major problems in the Chinese telecommunications sector The paper suggests that after a period of rapid growth in investment and number of subscribers, it is

time for the sector to put more emphasis on TFP growth in order to meet the challenges posed by WTO commitments

Mathur (undated) applies DEA and MPI to measure the technical efficiencies and total TFP in the ICT sector Technical efficiency is measured for 52 countries in 2006-07 and for 45 countries in 2002-03 TFP is measured for 45 countries between two periods TFP

is further decomposed into technical efficiency change and technical change to identify the sources of productivity growth in the ICT sector across countries Determinants of TFP growth are also identified through conducting regression analysis The study finds that efficiency scores varied significantly across countries and years For instance, ICT sector in South Korea and Argentina were relatively efficient in 2002-03, while in 2006-07, Bahrain, Brazil and Sweden turned out to be relatively efficient The findings show that the PFP growth in the ICT sector in developing and newly industrialized countries is slightly lower than that in the developed and transition countries, suggesting the catching-up of developing and newly industrialized countries The study suggests that since inefficiency in the ICT sector exists across countries; ICT regulators have to work out the orientations needed in inputs and outputs to adopt the best practices by converting relatively good ‘ICT Environment’ and “ICT Readiness” of countries into high ICT usage in their respective countries The paper also provides some policy directions for reforms in ICT policy, particularly in India, with focus on cooperating with other countries

Sahoo (2011) measures the technical efficiency in the Indian IT industry and also the contribution of the IT industry to the economic development of the country The paper is based on company-wise data collected from prowess database of CMIE DEA is applied to estimate the overall technical efficiency (OTE), pure technical efficiency (PTE) and scale efficiency (SE) of individual companies, using total sales as output variable, and employment, expenditure on computers and electronics equipments, operating expenditure, and power, fuel & water charges as the input variables The results suggest that the mean PTE score was about 0.6, indicating thereby that the companies are 40% pure technical inefficient Inefficiency in the Indian software industry may be attributed to the poor infrastructure, mainly power supply Percentage of efficient companies in terms of PTE scores was lowest in the year 2003-04 (20.83%) and highest in the year 2005-06 (40.28%) Percentage of efficient companies in PTE score followed a cyclical pattern The study further reveals that Indian software companies are becoming more scale inefficient over the years The empirical findings show that the Private Indian (PI) companies have relatively high OTE

score as compared to Private Foreign (PF) and Group Owned (GO) companies The efficiency scores highlight that the inefficiency levels in the PI, PF and GO companies are 51.8%, 62.8%, and 66.4% respectively This implies that although the levels of inefficiencies are very high across all companies, yet PI companies appear to be performing better than the other two Similar pattern is also observed in case OTE scores

Chen et al (2011) measures the technical, scale and managerial efficiencies in the Chinese IT companies using a non-parametric DEA approach The paper also estimates the TFP and its sources in the IT companies using the MPI approach The paper is based on panel data collected from a sample of 73 listed IT companies for the period 2005 to 2007 Fixed assets, intangible assets, the number of employees, and operating costs are taken as input variables and annual revenue, net profit and market value are considered as output variables Findings of the study reveal that on an average, the IT industry has 6.8 percent technical inefficiency and 5.1 percent managerial inefficiency Only four and six companies have obtained 100 percent technical and managerial efficiency, respectively The MPI results show that the IT companies do not have any substantial improvement in TFP during the period of 2005 to 2007 The convergence analysis show that there is significant technical diffusion in the Chinese IT industry from 2005 to 2006 and 2006 to 2007 separately, but the technical convergence tends to decline in the whole period of 2005 to 2007 The study further finds that the IT companies face the problems of shortage in intellectual capital and suggests that they should investment in R&D activities and make constant efforts to improve intellectual capital

A study conducted by PcW (2011) reveals that many of the large size companies show the growth similar to the growth witnessed by them prior to the slowdown However, in the case of mid-size companies, it is a mixed bag The smaller companies were impacted more

by the slower Euro zone recovery, currency fluctuations, employee attrition and rising employee costs Increasing cost of operations is seen as the major challenge facing the mid-tier IT/ ITES service providers High attrition and employability are the other challenges being faced by them The study further reveals that the IT/ITES service providers are looking

to move or expand to tier 2 cities for perceived benefits like availability of low-cost skilled resources, lower real estate cost and lower attrition However, the study finds that the IT companies feel that lack of connectivity to major cities, below par business infrastructure and sourcing of talent are the challenges they face while consolidating their IT operations to tier 2 cities The study further finds that in order to retain talent, the companies focused on

three key areas-imparting professional training, offering competitive reward and employee empowerment The study suggests that the IT industry should focus on diversification of destinations and activities; cloud computing; growth of domestic market; from service to high-end software product market through broadening the R&D base; from employment to employability; inclusivity in the IT development; and retention of talent

2.4 Summing Up

In this chapter, we have reviewed the relevant studies on the theme The reviewed studies show that most of the research work on the IT industry is related to the growth, employment and HR related issues There are only a few studies on measuring the productivity and efficiency of IT companies, especially using DEA methodology Studies conducted by Mathur (2007) on productivity and efficiency of Software and IT companies are quite relevant for the present research study However, the main focus of these studies is

on inter-country comparison of performance of IT Industry

Moreover, dataset used in these studies is up to 2005-06 Japanese software and IT sector is very dynamic and growing rapidly during the last decade, but at the same time it is facing number of challenges and opportunities owing to the emergence of some other competitor countries and discriminatory policies being adopted by the USA and other developed countries In this context, it is essential to study the recent trends in the efficiency and productivity of the IT Industry so that corrective policy measures may be adopted for its growth and sustenance The present study is an attempt toward this direction The study not only reviews the income and employment trends in the industry but also examines the technical, managerial and scale efficiencies and total factor productivity of individual IT companies It also identifies the key determinants of these efficiencies

CHAPTER 3 RESEARCH METHODOLOGY

3.1 Data Collection

Main data for the study comes from 75 IT companies for the cross-sectional analysis

of financial year 2018-19 The main source of the data is database which consists of the balance sheet-based financial data of large number of companies under IT industry, data related to income, export, employment, growth and structure of the IT industry, etc are collected from published sources Technical efficiencies of individual companies are estimated using CCR and BCC DEA models The study uses an approach for measuring the relative performance of individual companies Efficiencies of individual companies are estimated using DEA models Three types of efficiencies are estimated, namely, overall technical efficiency (OTE), pure technical efficiency (PTE) and scale efficiency (SE)

3.1.1 Data and Variables

Identification of input-output variables for the estimation of technical efficiency is the first and the most important stage in the DEA The efficiency scores depend on the choice made regarding inputs and outputs Potential input-output variables can be identified through statistical analyses, such as, correlation and regression Generally, to measure efficiency and productivity in the industrial sector, gross value of output (VOP) or net value added (NVA) are taken as output variables and capital, labour, energy & fuels and other intermediary inputs are taken as input variables However, if NVA is taken as output variable, only capital and labour are taken as input variables, as intermediary inputs, raw materials and energy & fuels are already deducted: from the VOP to get NVA Computer software and IT companies provide IT products and services which do not require physical raw material and apply capital, manpower, energy & fuels and operating expenses, etc to generate sales revenue Therefore, input-output variables for the IT companies may differ from that for the manufacturing companies, especially when VOP or sales turnover is taken as output variable The main sources of data on the IT companies in Japan are public resources and stock market databases These databases provide financial data, such as, capital assets & liabilities, input costs, sales, profits/losses, taxes, financial ratios, etc., for the individual companies

It is generally suggested in the literature that input-output variables in the DEA model should be as few as possible in order to retain discriminatory power on the comparative efficiencies of the units being assessed (Thanassoulis, 2001) With the addition of new

variable(s) to an already existing model increases the proportion of companies that are rated

as efficient as there are more facets on which a given set of companies can now be efficient (Majumdar, 1994) The rule of thumb is that number of companies in the dataset must be three times greater than the sum of input and output variables The choice of input-output variables is vital issue in DEA because distribution of the efficiency scores is influenced by

it In this study, sales turnover is taken as an output and net fixed asset (NFA), operating expenses (OE), employee cost (EC) are considered as input variables Since number of companies for the study is much higher than the three times of sum of inputs and output, the discriminatory power of DEA model in assessing the efficiency of individual companies is maintained The output and input variables are defined as:

3.1.1.1 Output variable

Sales Revenue: It refers to the gross revenue received by an IT (software) company from selling its products and services

3.1.1.2 Input variables

Net Fixed Assets: It is estimated by deducting accumulated depreciation from the

gross fixed assets It comprises capital invested in fixed assets, such as computers, buildings, vehicles and equipment, etc that stay in the business almost for more than one accounting period

Operating Expenses: Operating expenses consist of expenditure on job work,

employees’ training, software packages, project charges, technical fees paid, stores consumed, courseware manuals and transports charges, etc

Employee Cost: It includes all expenses incurred on employees, including wages and

salaries

Values of all inputs and output are taken in billion Japanese yen

3.1.2 The Sampling Method and Size

Table 3.1 shows the descriptive statistics of selected output and input variables As is obvious from the mean, standard deviation (SD), minimum and maximum values, the IT companies in the dataset vary significantly in terms of their annual sales turnover, NFA,

OE, and other inputs The average sales per company are JP 1303.46 billion yen, while

SD is of JP 4417.82 billion yen The minimum output per company is JP 12.16 billion yen and the maximum is of JP 26,935.40 billion yen, which indicates the range in size of

companies in the IT industry Similarly, magnitude of SD is higher than the average in case of all the input variables, further revealing that the companies are of diverse sizes However, all the companies are in the same business, i.e., production of IT products and services and therefore DEA approach is suitable for measuring their relative performance

Statistics Sales NFA Operating Expenses Employee Cost

Table 3.1: Descriptive Statistics of Output and Input Variables (billions Japanese yen)

Correlation matrix of output-input variables shown in Table 3.2 provides the sound basis for the selection of input-output variables for estimation of the technical and scales efficiency of IT companies The correlation coefficients between output and inputs turn out statistically significant at one percent level of significance, indicating appropriateness

of selection of variables for the study Furthermore, when output variable (sales) is regressed on the three input variables, the value of adjusted R2 turns out to be 0.998 which implies that 99.8 percent variation in the sales are explained by these three input variables Estimated F-value (magnitude of value being 9411.63) is also found statistically significant at less than one percent level of significance

Statistics Sales NFA Operating Expenses Employee Cost

Table 3.2: Correlation Matrix of Output and Input Variables

*Significant at 1% level of significance

3.2 Data Envelopment Analysis Concepts

Data Envelopment Analysis (DEA) initially developed by Charnes, Cooper and Rhodes (1978) and further extended by Bankar, Charnes and Cooper (1984) is a linear programming-based technique for evaluating the relative efficiency of homogenous set of companies It considers each individual observation and calculates a discrete piecewise frontier determined by the set of efficient companies It compares the companies that use multiple inputs to produce multiple outputs The technique is most suitable for measuring the technical efficiencies of those decision-making units (DMUs) which are homogenous and are in the same line of business For instance, if performance of the IT sector is assessed, all the companies should be in the same business that is software and IT services, not those companies which are in the business of production of hardware However, companies in the same business may be diverse in their size, location, age and other attributes The idea of DEA is based on the concept of Pareto Optimality, which states that, within the given limitations of resources and technology, there is no way of producing more of desired commodity without reducing output of some other desired commodity

3.3 Growth of DEA

The DEA approach gets its origin from the seminal paper of Farrell (1957), which proposed to measure productive efficiency Farrell defined the economic efficiency that could account for multiple inputs Further, he decomposed the productive efficiency into technical and allocative efficiency In this approach, the technical efficiency is an ability of

a company to maximize output from a set of given inputs and allocative efficiency is an ability of a company to use these inputs in optimal proportions, given their respective prices However, the pioneers of the DEA are Charnes, Cooper and Rhodes who introduced the first DEA model in 1978, based on the concept of Farrell’s efficiency Their model is known

as CCR model named after their names The model is based on the CRS technology assumption The subsequent advancement in DEA was proposed by Banker, Charnes and Cooper in 1984 The second DEA model is known as BCC model named after the names of Banker, Charnes and Cooper Coelli et al (1998) observe that the CRS technology assumption may not always be appropriate in some real-life contexts and therefore cannot

be applied in variety of circumstances To overcome this difficulty, Banker et al (1984) developed the BCC model which is based on VRS technology assumption

Since the beginning of CCR and BCC models, there has been rapid and continuous growth in the DEA field (see chapter 2) As a result, a considerable amount of published research has appeared, with a significant portion focused on DEA applications of efficiency and productivity in both public and private sector activities Several bibliographic collections have provided a comprehensive listing and analysis of DEA research In this context, Emrouznejad et al (2008) provide a survey of the first 30 years of scholarly literature on DEA It can be found at http://www.deazone.com Several other bibliographies have also been reported in the literature, including those of Emrouznejad and Thanassoulis (1997), Seiford (1997), Tavares (2002) By the year 2002, there were 3202 publications written by 2152 distinct authors characterized by 1242 keywords (Tavares, 2002) As well, Emrouznejad et al (2008) have identified more than 4000 research articles published in journals or book chapters up to the year 2007, written by 2500 distinct authors with more than 2000 keywords

Interestingly, 22 percent of all papers were written by the ‘‘top’’ 12 authors William

W Cooper, one of the founders of DEA, is the author with highest references in the DEA database, followed by Fare Rolf, Grosskopf Shawna, Sengupta Jati K., Charnes Abraham, Lovell C A Knox, Thanassoulis Emmanuel, Banker R D, Sueyoshi Toshiyuki, Zhu Joe, Cook Wade D and Seiford Lawrence M One of the factors behind this large growth of DEA literature is that DEA brings theory and practice in a mutually reinforcing and beneficial dynamics Practical applications of DEA follow theoretical developments in the field, while at the same time, the applications highlight aspects of practical importance which research must address (Emrouznejad et al., 2008)

Many researchers have done the theoretical development of DEA with its applications

in many areas In the DEA literature, DMU term is used for the company, unit or factory such as manufacturing companies, shops, banks, tax offices, hospitals, schools, universities, libraries, police stations, courts, countries, regions or individual medical practitioner (Cooper et al., 2004) Among the application areas, banking, education, healthcare, and hospital are found to be the most popular areas (Emrouznejad et al., 2008) DEA has been applied to estimate the technical efficiency of Energy (Ramanathan, 2005; Chien et al., 2007), Public utilities like Transport services (Agarwal et al., 2010,), Educational institutions (Tyagi et al., 2009), Banks (Kumar and Verma, 2002; Chatterjee and Sinha, 2006), IT sector (Mathur, 2007a and 2007b; Sahoo:2011) manufacturing industries (Yao, 2003; Singh 2006; Joshi and Singh: 2012; Mukharjee and Ray: 2004; Rao and Ray:2003,

Saranga:2009) Furthermore, development of DEA based MPI for measuring TFP growth and its decomposition into EFFCH and TECHCH is the significant achievement in the field

of productivity analysis (Balk, 2001; Kumar and Russell: 2002; Cheng and Lo, 2004; Kardag et al., 2005; Singh and Agrawal: 2006; Joshi and Singh: 2010)

We have two basic DEA models: CCR model, developed by Charnes, Cooper and Rhodes in 1978 and BCC model, developed by Banker, Charnes, and Cooper in 1984 CCR model defines the relative efficiency for any DMU as a weighted sum of outputs divided by

a weighted sum of inputs where all efficiency scores are restricted to lay between zero and one An efficiency scores less than one for a company means that it is operating below the frontier The efficiency score reflects the radial distance from the estimated production frontier to the DMU under consideration The DEA model uses input-output weights as variables and the LP solution produces the weights most favorable to the unit under reference In order to calculate efficiency scores, fractional linear programming (FLP) is converted into linear programming (LP) formulation by normalizing either the numerator

or the denominator of the fractional programming objective function

In case of output-maximization DEA program, the weighted sum of inputs is constrained to be unity to maximize weighted sum of outputs, while in input-minimization DEA program, the weighted sum of outputs is constrained to be unity to minimize weighted sum of inputs CCR model is based on CRS technology assumption which implies that if the input levels of a feasible input-output correspondence are scaled up or down, then another feasible input-output correspondence is obtained in which the output levels are scaled by the same factor as the input levels (Thanassoulis, 2001) Thus, under the CRS technology assumption, constructed production frontier is linear, revealing the output will increase at the same rate as inputs are increased Further, when CCR model is used, estimated efficiency score of a company remains same whether input-oriented or output-oriented DEA model is applied

The BCC DEA model is based on the VRS technology assumption and it measures the pure technical efficiency, i.e., conversion of inputs into output Bankar, Charness and Cooper added a convexity constraint in the CCR model The CCR model assumes that constant return to scale exists at the efficient frontiers whereas BCC assumes variable retunes to scale frontiers CCR model measures the overall technical efficiency (OTE), while BCC model measures the pure technical efficiency (PTE), net of scale-effect PTE is also known as managerial efficiency If a company scores value of both CCR-efficiency and

BCC-efficiency equal to one, it is said to operate at the most productive scale Size (MPSS) Actually, BCC model helps to decompose the OTE into PTE and the scale efficiency (SE)

SE of a company is measured by dividing the OTE from PTE For example, if PTE of a company is equal to 1 and its OTE is less than 1, it implies that the company is able to converts efficiently its inputs into output, however, it is OTE-inefficient because its size is either too big or too small related to the optimum size Thus, inefficiency in any company may occur due to its inefficient operations or due to the disadvantageous conditions under which it operates

3.4 Advantages of DEA

A DEA technique has several merits over the conventional regression-based production function approach A few of them are as follows:

- DEA can handle multiple inputs and outputs with different units (Pastor JT, et al., 1999)

- It does not require any assumption of a functional form relating inputs to outputs (Dyson, R.G and Thanassoulis, E., 1988)

- It can be applied to profit as well as non-profit making entities (Allen, R., 1997)

- It sets targets for inefficient DMUs to make them efficient (Andersen, P and Petersen, N.C., 1993)

- It also identifies slacks in inputs and outputs (Emrouznejad, A., 2009)

- It estimates a single efficiency score, identifies input excesses and output and provides benchmarks to monitor the performance of inefficient companies (Bogetoft, P and Otto, L., 2012)

3.5 Limitations of DEA

DEA also has some limitations, which are as follows:

- It is a non-parametric technique in which statistical hypothesis testing is difficult (M Fukushigea and I Miyarab, 2013)

- It is an extreme point technique because of which the measurement error cannot be measured (Cooper at al., 2000)

- Its measured efficiency is relative one and therefore comparison of the performance of a company can be made with only those companies which are in the reference set (Li, X.B and Reeves, G.R., 1999)

- It is sensitive to the choice of the input-output variables and number of companies Its results are also influenced by the size of sample If sample size is small, discretionary power of model reduces Therefore, thumb rule is that the number of companies in the dataset should be more than three times of sum of input and output variables (Khalili, M

et al., 2010)

- It is a computationally intensive method; however, availability of DEA software has made the computation of efficiency scores easy (Podinovski, 2001)

Before applying DEA, some precautions are to be taken, such as:

- Since no hypothesis testing is possible, data accuracy must be given priority

- In order to make enough discrimination between DMUs, sample-size should be adequate

It should be at least three times greater than the sum of input-output variables

- Most important exercise in DEA is the identification of input-output variables Regression analysis can be conducted to identify the best-fit in output and input variables

- Zero and negative values of any input or output should be avoided Variables in the model should be as few as possible

- Data scaling should be done before applying DEA so that input-output variables do not

have excessively large values

3.6 DEA Models

Since the origin of the DEA methodology, several DEA models have been developed

by the researchers The advancement in the DEA theory is a continuous process In this section, we describe only CCR and BCC DEA models which are applied in this research study

3.6.1 CCR Model

Since the first DEA model was suggested by Charnes, Cooper and Rhodes (1978), it

is named as CCR model after their names This model is based on the constant return to scale (CRS) technology assumption The efficiency score in the presence of multiple input-output variables is defined as:

Efficiency = weighted sum of outputs/weighted sum of inputs

Mathematically, the relative efficiency of the kth company given by:

urk = the weight given to the rth output of the kth company

vik = the weight given to the ith input of the kth company

n = No of companies

s = No of outputs

m = No of inputs; and

= a non-Archimedean (infinitesimal) constant

The above FLP problem is reformulated in LP problem as follows:

Subjected to:

Since the number of companies is larger than the total number of inputs and outputs, solving the dual of the model can reduce the computational burden Mathematically, the dual formulation of the above model is:

Subjected to:

Where:

= Slacks in the ith input of the kth company

= Slacks in the rth output of the kth company

= non-negative dual variables

Where:

(scalar) is the (proportional) reduction applied to all inputs of DMU k to impose efficiency If for DMU k, k =1 and all slacks are zero, it is Pareto efficient The non-zero slacks and (or) k <=1 identify the sources and amount of any inefficiency that may exist

in the DMU under reference

3.6.2 BCC Model

The CRS assumption is only appropriate when all companies are operating at an optimal scale Imperfect competition and constraints on finances, etc., may cause a company to be operating away from optimal scale Banker, Charnes and Cooper (1984) suggested an extension of CRS model to account for variable returns to scale (VRS) The primary difference between BCC and CCR models is the convexity constraint In the

BCC model are restricted to summing to one (i.e ) If we impose

instead of = 1, then the model is known as Non-Decreasing Returns to Scale (NDRS) model (Ramanathan, 2003)

The BCC model measures the pure technical efficiency, net of scale effect It captures the pure resource-conversion efficiencies, irrespective of whether the DMUs operate at IRS, CRS or DRS Scale efficiency of a DMU is estimated dividing the CCR efficiency score by the BCC efficiency score As BCC efficiency score is more than or equal to the CCR efficiency score, value of scale efficiency score will be less than or equal to one

We can explain the difference between these two models through the following graph: The CCR model is based on CRS technology assumption and the BCC model is based

on VRS technology assumption The CRS surface is the straight-line oicm and the VRS surface is abcde

Figure 3.1: Comparison of CRS and VRS Frontiers

Efficiency of any interior point (such as ‘k’) is intuitively given by the distance between the envelope and itself Typically, such a distance may be measured either horizontally along the x-axis or vertically along the y-axis, providing an input-oriented

or output-oriented measure, respectively For example, using an input orientated measure, technical efficiency of DMU ‘k’ will the measured by hi/hk under the CRS technology assumption and by hj/hk under the VRS technology assumption A measure

of scale efficiency is providing by the ratio hi/hj A DMU at point ‘c’ is operating at most productive scale size (MPSS), as at this point all the three efficiency measures are equal

to one (OTE=PTE =SE=1)

CHAPTER 4 RESULTS AND DISCUSSIONS

4.1 IT Industry in Japan

An overview of the status of industry is necessary before we analyze efficiency and productivity of the IT companies Therefore, this chapter discusses evolution of IT industry, its growth, structure, pattern and role in the economy first

4.1.1 Evolution

A growth-accounting exercise comparing Japan and the United States from 1975 to

2000 is provided to show the impact of IT on economic growth and productivity at the macro level The macro view is supplemented by industry and company-level analysis

IT is a typical general technology purpose, which means it diffuses widely into an economy and has heterogeneous effects on the various aspects of company activities IT changes business practices and decision-making systems, as well as relationships between suppliers and customers Thus, it is important to look at what is going on at the company level to achieve deeper understanding of the economic impact of the IT revolution

4.1.2 Role in the Japan Economy

As a determinant of productivity, use of IT has been extensively examined An IT revolution, with rapid technological progress in computers and the spread of the Internet

in the 1990s, coincides with a kink in the trend line of U.S labor productivity That is, after a slowdown in the 1980s, it regained speed in the late 1990s Oliner and Sichel (2000) show that about two-thirds of the 1.5% annual productivity revival after 1995 can

be attributed to the growth in IT investment Even after the so-called IT bubble burst in

2001, U.S labor productivity as measured by the U.S Bureau of Labor Statistics (BLS) shows strong performance Thus, it is fair to say that the IT investment surge can explain

a significant portion of the U.S productivity revival after the mid-1990s (Bailiy 2002) Jorgenson and Motohashi (2005) extend such analysis to Japan and compare the role of

IT in economic growth in the two countries The growth rate can be decomposed into contributions from factor inputs: labor, capital, and total factor productivity (TFP) Capital inputs can be decomposed into IT capital and non-IT capital