The Role of Information Technology in Operating Cost and Operational Efficiency of Banks: An Application of
Frontier Efficiency Analysis
†CHU-FEN LI*
Department of Finance and Graduate Institute of Business and Management, National Formosa University, Taiwan
ABSTRACT
This study explores how information technology, operating cost, and operational efficiency are related to each other in banking. It is well known that the adoption of information technology can reduce organizational operating cost and improve operational efficiency. However, the intuitive impacts should be evaluated in greater detail. This paper proposes a framework for measuring the performance of information technology application, which provides us with empirical evidence as follows. First, low operational efficiencies exist in the banking industry during the study period. These inefficiencies are in nature ascribable to a combination of both wasteful overuse of information technology resources and inappropriate scale of information technology investments. Second, operational efficiencies measured by two frontier efficiency analyses, data envelopment analysis and stochastic frontier approach, present a significant strong relationship. Third, for an individual inefficient bank, the operational efficiency can be enhanced if the total amount of information technology investments is enlarged. Fourth, the different ownership type has a significant effect upon the performance contributions of information technology application. Fifth, to enhance performance, banks can reduce operating costs by increasing the number of financial cards issued and improve operational efficiency by installing more automated teller machines and providing customers with a wide variety of information technology services. Furthermore, the mutually-owned banks require a cutback in information technology personnel as well to enhance performance.
Key words: information technology, operating cost, operational efficiency, performance, banking, frontier efficiency analysis.
1. INTRODUCTION
Over the past few decades, modern business organizations have been investing increasingly substantial amounts of money in information technology (IT) with the objective of improving their operational efficiency and competitive ability in the industry. The important role IT plays in contemporary business is unquestionable. IT is regarded as a critical factor for business enterprises to survive and to grow further; however, empirical evidence in support of these anticipated benefits has been mixed. Some researchers asserted that IT investments can really promote the enterprises’ operational performance by reducing costs, raising profit margin, upgrading production levels, increasing service quality, advancing customer satisfaction and improving overall operations. In contrast, other researchers did not demonstrate the positive effect of IT investments and concluded
† This work was supported by the National Science Council of the Republic of China with Project No. NSC 90-2416- H-309-010.
* E-mail: cfli@nfu.edu.tw
that IT spending brought no significant contributions to the enterprises’ operations, and so the “IT productivity paradox” has been an issue of continuous debate for decades.
The differences among research objects, methodologies and performance indices result in inconsistent conclusions obtained in the literature. In this respect, the lack of good quantitative measures for the output and the value created by IT has made the studies on justifying IT investments particularly difficult. Brynjolfsson (1993) proposed the following four explanations for the IT productivity paradox: (a) mismeasurement of inputs and outputs; (b) lags between cost and benefit; (c) redistribution and dissipation of profits; and (d) mismeasurement of information and technology. Obviously, a new research method should be able to eliminate or overcome the above-mentioned defects.
To explore the effects of IT spending on organizational performance, this study provides a framework for performance evaluation using frontier efficiency analysis approaches, not only nonparametric data envelopment analysis (DEA) involving linear programming, but also parametric stochastic frontier approach (SFA) involving econometrics. The data set is collected from a sample of banks in Taiwan. The banking industry has been particularly information intensive. In history, banking has always been a crucial area for IT to be implemented. That is, an area where the advantage from using IT is so considerable that the state-of-the-art IT is developed almost as soon as it becomes available. A widely held belief is that IT is absolutely vital to a bank’s survival and growth. In this regard, it seems especially meaningful to link this issue with banking institutions.
To sum up, the purpose of this study is to assess the impacts of IT investments on bank performance by accomplishing the following five main objectives. The first is to explore the relevant theories and thereby develop more complete models for evaluating bank performance. The second is to measure operational efficiency for each individual bank and analyze the main sources of operating inefficiency. The third is to investigate the impacts of IT investments on operating cost and operational efficiency. The fourth is to compare and interpret the effects of IT investments on performance of different types of banks. The fifth is to contrast alternative approaches to the measurement of IT value.
The remainder of this paper is organized as follows. Section 2 reviews the previous empirical research at the firm level. Section 3 describes the analytical techniques employed to measure operational efficiency—DEA and SFA, as well as the empirical data. Section 4 reports and discusses the empirical results from three aspects: (a) evaluation of operational efficiency; (b) comparison of efficiency differences between various types of banks; and (c) impact assessment of IT investment on operating cost and operational efficiency. Section 5 sums up the main findings and presents the conclusions.
2. LITERATURE REVIEW
Because of the lack of relevant empirical studies in the field of banking, most studies described in this section are thus related to other service organizations,
manufacturing firms and hospital institutions. Of the studies, both the financial ratio analysis and parametric econometric approach were more widely used for evaluating performance, while nonparametric linear programming approach was scarcely used. Some econometric-based studies adopted SFA as the analytic tool and Cobb-Douglas function as the theoretical model. Here, we briefly review the relevant literature in chronological order. A summary of it is provided in Table 1.
Table 1. Previous empirical research
Authors Sample Variables Research Method
Strassman (1990) 38 service firms Spending for computers Return on investment Profits
Productivity
Correlation analysis
Weill (1992) 33 manufacturing firms during 1982-1987
IT investment types Sales growth ROA
Labor productivity
Hierarchical regression
Loveman (1994) 60 manufacturing firms during 1978-1984
Productivity Material expenditure
Non-IT purchased services expenditure Total labor compensation
IT capital Non-IT capital
OSL regression
Mitra and Chaya (1996)
448 public companies (exc.
bank and insurance firms) during 1988-1992
IT spending/sales Average total cost Average production cost Average overhead cost Average labor cost
Hypothesis testing
Rai, Patnayakuni,
& Patnayakuni (1997)
497 public companies in 1994
ROA ROE Value added Total sales Labor productivity Administrative productivity IT budget
IT capital
IT infrastructure investment Client/server expenditure Hardware expenditure Software expenditure Telecom expenditure
SFA
Devaraj and Kohli (2000)
8 hospitals Net patient revenue IT labor expenses IT support expenses IT capital expenses
Number of full-time employees
Correlation analysis
Lee and Menon (2000)
83 hospitals during 1976-1994
Productivity IT capital Non-IT capital IT labor Non-IT labor
DEA SFA
Table 1 (continued).
Authors Sample Variables Research Method
Ham, Kim and Joeng (2005)
21 hotels in 2002 Performance of lodging operations Front-office applications Back-office applications
Restaurant/banquet management system Guest-related interface applications
Hypothesis testing
Andersen and Foss (2005)
88 multinational computer products and service firms
Economic performance
Computer-mediated communication Strategic opportunity
Multinationality
Hierarchical regression
Strassman (1990) investigated the relationship between IT and return on investment in a sample of 38 service sector firms using correlation analysis. He found that some top performers invested heavily in IT, while some did not. He concluded that there was no correlation between spending for computers, profits and productivity.
Weill (1992) studied 33 medium and small-scaled valve manufacturing companies to explore the relationship between the IT investments and organizational performance using hierarchical regression. Although transactional IT investment was found to be strongly related to superior organizational performance, there was no evidence that strategic IT investment, on a long-term basis, would increase or decrease organizational performance. However, the results implied that strategic IT investment was beneficial to relatively poor-performing firms in the short run.
Loveman (1994) utilized OSL regression to assess the productivity impact of IT using a sample of 60 manufacturing firms during 1978-1984. The results showed that during the five-year period, the contribution of IT investment to the output of manufacturing firms was nearly zero. There existed no sufficient evidence to support the benefit of IT from productivity enhancement.
Mitra and Chaya (1996) used a sample of 448 large and medium-sized U.S.
corporations during 1988–1992 to analyze the performance impact of IT investment.
Using hypothesis testing, they found that higher IT investments were associated with lower average production costs, lower average total costs, and higher average overhead costs. They also found that larger companies spent more on IT as a percentage of their revenues than smaller companies. However, they did not find any evidence that IT reduced labor costs in organizations.
Rai et al. (1997) employed Cobb-Douglas function approach to probe the relationship between IT investment and business performance using a sample of 497 firms in 1994. The results suggested that IT investments could make a positive contribution to firm output and labor productivity. However, various measures of IT investment did not appear to have a positive relationship with administrative productivity. Furthermore, IT was likely to improve organizational efficiency, its effect on administrative productivity and business performance might depend on such other factors as the quality of a firm’s management processes and IT strategy links, which could vary significantly across organizations.
Devaraj and Kohli (2000) examined monthly data collected from 8 hospitals over a recent 3-year time period to study the relationship between IT and performance. Correlation analysis results provided support for the relationship between IT and performance that is observed after certain time lags. Such a relationship may not be evident in cross-sectional data analyses. Moreover, the results indicated support for the impact of technology contingent on the business process reengineering practiced by hospitals.
Lee and Menon (2000) used DEA and SFA to analyze the financial data of 83 hospitals during 1976-1994. They found that hospitals that were characterized by high technical efficiency also used a greater amount of IT capital than firms that exhibited low technical efficiency and that a group of hospitals exhibiting high technical efficiency also exhibited low allocative efficiency, indicating that, while processes might have been efficient, resource allocation and budgeting between various categories of capital and labor had not been efficient. Moreover, they found that IT labor had a negative contribution to productivity and that non-IT capital had a greater contribution to productivity than IT capital.
Ham et al. (2005) examined the effect of IT applications on performance using a sample consisting of 13 five-star hotels and 8 four-star hotels in 2002 in Korea. Results of hypothesis testing supported the relationship between IT usage and the performance of lodging operations. Furthermore, they found that front-office applications, restaurant and banquet management systems, and guest-related interface applications significantly and positively affected performance of lodging operations; however, the effect of guest-related interface applications was not significant.
Andersen and Foss (2005) investigated the role and effects of information and communication technology in multinational enterprises. They suggested that the attendant cost-benefit tradeoff could be influenced by computer-mediated communication. From a sample of 88 organizations in the computer products industries using hierarchical multiple regression analyses, they found that multinationality in itself did not guarantee a higher level of strategic opportunity.
Instead, use of IT to facilitate communication among managers across functional and geographical boundaries enhanced coordination of multinational activities in the development of strategic opportunity, which in turn was associated with superior performance.
3. METHODOLOGY
This study attempts to investigate the impact of IT on operational performance of banking firms by means of two frontier efficiency analysis approaches, namely, DEA and SFA. This section introduces the methodology employed with the research methods and data, as well as variables described respectively as follows.
3.1 The Research Methods
In the past several years, the assessment of operational performance has received much attention from both academia and business due to competitive and margin pressures in the market. While evaluating the performance of a decision-making unit (DMU), it is indispensable to use reliable approaches. Up to the present, the three prevailing techniques developed for efficiency measurements in both industrial and academic worlds are; financial ratio analysis, an econometric approach and a linear programming approach. The traditional financial ratio analysis shows the relations between two financial figures after being compared using data from financial statements. Although it has been widely used, a critical limitation is that financial ratio analysis fails to consider the multiple input-output characteristics of business enterprises and cannot give an overall clear picture of organizational operations because firm performance may exhibit considerable variation, depending on the indicator chosen. In the recent banking literature much attention has mostly been directed to the latter two techniques of frontier efficiency analysis, namely, econometric approach and linear programming approach, which can provide comprehensive insights beyond those available from financial ratio analysis for evaluating and improving banking efficiency.
Since the seminal study by Farrell (1957), methodological development in frontier efficiency analysis has been continuing at a rapid pace. To date, there are a multitude of techniques, parametric and nonparametric, stochastic and deterministic.
The essential differences among these techniques primarily reflect differing assumptions used in estimating the shape of the frontier and the distributional assumptions imposed on the random error and inefficiency.
There are at least five different types of approaches in the literature that have been employed in measuring banking efficiency. Of those, three econometric approaches, such as stochastic frontier approach (SFA), distribution-free approach (DFA) and thick frontier approach (TFA) are parametric, and two linear programming approaches are nonparametric, such as data envelopment analysis (DEA) and free disposal hull (FDH). Each of the approaches has weaknesses, as well as strengths relative to the other. The literature has not yet come to a consensus about the preferred approach for determining the best-practice frontier against which relative efficiencies are measured.
In general, parametric approaches are stochastic, and so attempt to distinguish the effects of inefficiency from the effects of noise. A key drawback of parametric approaches is that they usually specify a particular functional form that presupposes the shape of the frontier. If the functional form is misspecified, measured efficiency may be confounded with the specification errors. In sharp contrast to parametric approaches, nonparametric approaches are inherently bounding techniques, and so they impose less structure on the frontier. They are deterministic and do not allow for random error owing to luck, data problems or other measurement errors. If random errors do exist, measured efficiency may be confounded with these random deviations from the true efficiency frontier. So the former’s limitations are exactly the latter’s advantages and vice versa. Consequently, we employ both
nonparametric DEA and parametric SFA as research methods at the same time. The DEA and SFA are among the most popular approaches of frontier efficiency analysis. Thus, we not only can analyze the impact of IT investment on banking operational performance, but can also contrast the results of both approaches. DEA and SFA are described as follows.
3.1.1 Data envelopment analysis
Data envelopment analysis (DEA) is in essence a mathematical programming technique initially developed by Charnes, Cooper and Rhodes (1978) from the basic concepts of relative efficiency and nonparametric frontier of Farrell (1957).
DEA extends the notion of Farrell’s productive efficiency from single-output case to multiple-output case. Unlike parametric frontier approaches, DEA does not require any assumptions about the functional form. The DEA frontier is formed as the piecewise linear combinations that connect a set of best-practice DMUs, which is obtained from the observed sample, yielding a set of convex production possibilities. Thus, a maximal efficiency measure for each DMU relative to all other DMUs in the observed data set can be calculated only with the requirement that each DMU lies on or below the external frontier.
The most important characteristics of the DEA methodology can be presented by the CCR model. For the discussion to follow, let us suppose that there are n DMUs to be evaluated. Each DMUk (k=1, 2,…, n ) consumes varying amounts of s inputs to produce r outputs and each has at least one positive input and one positive output. The primal input-oriented CCR model proposed by Charnes et al. (1978) is formulated as follows.
1 ,
1
max =
=
=
∑
∑
r i ie i
e n
u w
j je k
u y E
w x
subject to 1
1
1 , for , , ,
=
=
≤ =
∑
∑
r i ik i
s j jk j
u y
k 1 2 n
w x
(1)
, 1, 2, ,
≥ =
ui i r
, 1, 2, ,
≥ =
wj j s
where the subscript (e) denotes the DMU being evaluated from the observed data;
Ee is the efficiency rating of the DMU being evaluated; yik and xjk represent the observed amount of output i (i =1, 2, …, r) and input j (j =1, 2, …, s), respectively, for DMUk , andyik≥0, xjk≥0; ui and wj represent the weights for output i and input j, respectively ; and is a non-Archimedean infinitesimal constant.
The CCR model is formulated under of the assumption of constant returns to scale (CRS) technology. Banker, Charnes and Cooper (1984) changed the
technology to variable returns to scale (VRS) and suggested BCC model as follows.
1 ,
1
max =
=
−
=
∑
∑
r
i ie e
i
e s
u w
j je j
u y u
E
w x
subject to 1
1
1 , for , , ,
=
=
−
≤ =
∑
∑
r
i ik e
i s
j jk j
u y u
k 1 2 n
w x
(2)
, 1, 2, ,
≥ =
ui i r
, 1, 2, ,
≥ =
wj j s
u free e
where the subscript (e), yik, xjk, ui and wj and͍are as defined earlier, and ue is free.
The existence of ue shrinks the feasible region for BCC model from the conical hull considered in the CCR model to the convex hull of the DMUs. The convexity requirement provides the basis for measuring economies of scale in DEA. It exhibits CRS if ue is equal to 0; it exhibits increasing returns to scale (IRS) if ue is greater than 0; while it exhibits decreasing returns to scale (DRS) if ue is less than 0.
3.1.2 Stochastic frontier approach
The stochastic frontier approach (SFA) specifies a functional form, such as Cobb-Douglas or translog, for the cost, profit or production relationship among inputs, outputs and environmental factors, and allows for random error. It was first proposed by Aigner, Lovell and Schmidt (1977), Battese and Corra (1977) and Meeusen and van den Broeck (1977) simultaneously from different continents. Its residuals contain two error terms, one for inefficiency that is assumed to be either nonpositive or nonnegative relying on its distributional assumption and another for noise or random error that is unrestricted to be positive or negative. The former represents factors that can be controlled by DMUs, while the latter represents those effects, which cannot be controlled by the DMUs, including quality or measurement errors. The inefficiency term is supposed to follow a one-sided distribution, usually asymmetric half-normal or truncated normal distribution;
whilst the noise component is supposed to follow a symmetric distribution, usually the standard normal.
Let us consider that a specific DMUk for k=1, 2,Ǿ, n, uses s inputs
1 2
( , , , ) +
= s ∈
x x x x R to produce scalar output y∈R+ , using the most frequently adopted Cobb-Douglas production function as a theoretical model, the production frontier can be expressed as follows.
0 1
ln β β ln
=
= +
∑
s − +k j jk k k
j
y x u v 1 (3)
where yk and xjk represent the observed amount of output and input j (j =1, 2, Ǿ, s), respectively, for DMUk ; uk denotes the measure for technical inefficiency and is assumed to be nonnegative random variables of half-normal distribution; vk denotes a random error term indicating the usual statistical noise and is assumed to be normal distributed with mean zero and constant variance. The estimate of returns to scale͏, i.e., the elasticity of scale is calculated as follows.
1
η β
=
=
∑
s jj
(4)
The͏values greater than, equal to or less than unity represent IRS, CRS or DRS, respectively.
The Cobb-Douglas production function is simple and easy to transform for estimation, but it also imposes some inherent restrictions on the production structure. To facilitate comparison and provide more convincing evidence, the study also considers the more flexible translog function as follows.
0
1 1 1
ln ln 1 ln ln
β β 2 β
= = =
= +
∑
s +∑ ∑
s s − +k j jk jq jk qk k k
j j q
y x x x u v (5)
where yk, xjk, uk and vk are as defined earlier, but uk has truncated normal distribution. The estimate of returns to scaleis calculated as follows.
1 1 1
ρ β β ln
= = =
=
∑
js j+∑∑
js qs jq xjk (6)Thevalues greater than, equal to or less than unity represent IRS, CRS or DRS, respectively.
3.2 The Data
The research target of this study includes all banks operating in Taiwan during the years 1996-2000. The definition of a bank here is adopted in a broader sense; it is referred to as a financial institution able to create deposit currency, and includes general banks, and community financial institutions, such as credit cooperatives and credit departments of farmers and fishermen’s association. The empirical data come from the following two sources. The first is primary data that mainly provide information about IT investments and spending, which is taken from the results obtained in a questionnaire survey of current officers in charge of an information unit, or center, of each bank. The second is secondary data that chiefly provide financial information, which is taken from the annual financial
1 If it is assumed that uk = 0, the problem simplifies to one of OLS estimation of the parameters of a production function with no inefficiency; while if it is assumed that vk = 0 the problem simplifies to one of estimating the parameters of a deterministic production frontier with no noise. In the former case there is no efficiency measurement problem to worry about; while in the latter case there is no decomposition problem to worry about.
statements of each bank. For general banks, the financial statements are collected from the publication “Financial Statistics” by the Bureau of Monetary Affairs of the Ministry of Finance, R.O.C. For community financial institutions, the financial statements are collected from their annual reports. To improve survey results, a pretest was carried out before the questionnaire survey, and the questionnaire was revised accordingly.
A total of 74 questionnaires were distributed to the participants through mail, email, Internet, or facsimile transmission. Of the 74 copies, 46 are sent to general banks, 22 to credit cooperatives and 6 to credit departments of farmers and fishermen’s association. The participants of the self-administered survey are the officers in charge of the information unit or center2. Finally, 41 responses were received after follow-ups. Among them, 11 were invalid, leaving 30 for analysis.
The effective response rate is up to 41%. Thus, the total sample for this study consists of all 30 individual banks, which can be categorized into three groups according to their business ownership category: 5 publicly-owned banks (17%), 12 privately-owned banks (40%) and 13 mutually-owned banks (43%).
3.3 The Variables
According to the relevant theories and literature, eight variables are used for this study; namely, number of IT personnel, number of ATMs, number of PCs and terminals, number of financial cards issued, diversification of IT services, pre-tax profits, total IT expenses and total operating costs. Among those, the output is pre-tax profits and the inputs are the first five variables for both DEA and SFA models. These input variables, coupled with the last two variables are used in both correlation analysis and regression analysis. The value of each variable is defined as an annual average value during the study period. Table 2 summarizes the arithmetic mean, standard deviation, and the minimum and maximum for each variable mentioned earlier.
Table 2. Descriptive statistics of data
Variable Mean S.D. Min. Max.
IT expenses 166.1 182.7 8.3 654.0
Operating costs 17,360.1 14,788.1 1,269.4 102,358.8
IT personnel 53.7 42.2 5.2 180.4
ATMs 117.5 140.3 12.0 483.8
PCs/terminals 955.2 1,107.5 54.2 4,251.2
Financial cards 383.2 533.9 17.4 2,186.7
IT services 180.3 47.3 98.2 245.4
Pre-tax profits 1,782.9 2,663.8 32.0 17,778.0 Note. Except that the number of financial cards issued is measured in thousands and pre-tax profits are
measured in million NT dollars, the other variables are measured in units.
2 We contacted nearly 90% of the information officers by telephone in advance. If they were not willing to fill out the questionnaire or unable to offer us the relevant information in the questionnaire, we excluded their banks from our sampling population.
Since a strong correlation between input and output variables should, in principle, be suggested, this study conducted a Pearson correlation analysis and the result demonstrates a highly significant positive correlation between both input and output variables at the 1% significance level. The Pearson correlation coefficients are between 0.749 and 0.862, implying that the input and output variables chosen by this study are suitable for our models. In addition, this study also performed collinearity analysis among input (independent) variables. The results show that the tolerance value ranges between 0.809 and 0.972, suggesting that tolerance is sufficient and collinearity does not affect the predictive ability of a regression equation.
4. RESULTS AND DISCUSSION
In the study, both DEA and SFA techniques are utilized at the same time to measure the operational efficiency using a sample of banks in Taiwan. The empirical results of this study are described and interpreted as follows from three aspects: (a) evaluation of operational efficiency; (b) comparison of efficiency differences between various types of banks; and (c) impact assessment of IT investment on operating cost and operational efficiency.
Table 3 presents the results related to various efficiency indices for each bank and also provides some summary statistics in the form of arithmetic mean, standard deviation and minimum and maximum values. The first column in the table is bank code. The second and third columns, CCRTE and BCCTPE, indicate technical efficiency and pure technical efficiency derived from CCR and BCC models of DEA, respectively. The last two columns, COBBTE and TRANTE, represent technical efficiency values using the Cobb-Douglas and translog models of SFA.
Here, we evaluate the operational performance of each bank with technical efficiency, which measures the ability of a bank in utilizing IT resources to create profits. The fourth column, SCALE, represents scale efficiency, which is calculated as a ratio of CCRTE to BCCTPE. SCALE is employed to examine whether a bank is operating at economies of scale. An efficiency value less than unity implies that a bank is not on the production frontier and thus operating inefficiently, while that equal to unity implies quite the opposite. As shown in Table 3, low operational efficiencies do exist in banking during the study period. On average, the values of technical efficiency measured according to CCR, Cobb-Douglas and translog models are 40.9%, 39.5% and 40.8%, respectively. The efficiency figures seem to be quite close. As shown in Table 4, the coefficients of Pearson correlations between DEA and SFA efficiency indices are over 0.60. We further conduct a test of hypotheses and find that the null hypothesis H0: = 0 is rejected at the 1%
significance level. The correlation results suggest that there is a significant positive relationship between both DEA and SFA indices.
Table 3. Efficiency indices with a statistical summary
Bank CCRTE BCCPTE SCALE COBBTE TRANTE
B1 0.443 0.562 0.788 0.264 0.361
B2 1.000 1.000 1.000 0.697 0.635
B3 0.601 0.714 0.841 0.321 0.326
B4 0.723 0.725 0.998 0.572 0.643
B5 0.339 0.596 0.569 0.241 0.227
B6 0.562 0.570 0.986 0.445 0.518
B7 0.118 0.467 0.252 0.153 0.146
B8 0.353 0.487 0.724 0.901 0.937
B9 0.185 0.504 0.368 0.350 0.366
B10 0.407 0.502 0.810 0.999 0.997
B11 0.017 0.400 0.043 0.080 0.090
B12 0.155 0.482 0.322 0.228 0.213
B13 0.025 0.513 0.049 0.035 0.035
B14 0.674 0.678 0.994 0.769 0.775
B15 0.218 0.552 0.394 0.204 0.173
B16 0.211 0.485 0.435 0.187 0.199
B17 0.025 0.562 0.045 0.022 0.021
B18 0.913 1.000 0.913 0.636 0.599
B19 0.465 0.736 0.631 0.461 0.496
B20 0.413 1.000 0.413 0.247 0.258
B21 0.516 1.000 0.516 0.650 0.785
B22 0.649 0.808 0.804 0.843 0.894
B23 0.549 1.000 0.549 0.501 0.553
B24 0.448 0.773 0.579 0.193 0.214
B25 0.676 1.000 0.676 0.285 0.178
B26 0.444 0.851 0.522 0.282 0.326
B27 0.274 0.685 0.400 0.263 0.276
B28 0.403 0.745 0.540 0.253 0.243
B29 0.365 0.653 0.560 0.655 0.656
B30 0.093 0.559 0.167 0.100 0.103
Mean 0.409 0.687 0.563 0.395 0.408 S.D. 0.253 0.193 0.288 0.268 0.280 Min. 0.017 0.400 0.043 0.022 0.021 Max. 1.000 1.000 1.000 0.999 0.997 Note. The study also examines the returns to scale patterns for each bank by DEA and SFA, and the results indicate that
all banks are operating at IRS except that B2 exhibits CRS.
Furthermore, in Table 3, the sources of operational efficiencies are also explored through the relationship between CCRTE, BCCPTE and SCALE. For a technically inefficient bank, both of its BCCPTE and SCALE are found to be very low at the same time apart from a few exceptions. That is, these operational efficiencies are in nature ascribable to a combination of both pure technical inefficiencies and scale inefficiencies. The former arises from wasting IT resources, while the latter results from not operating at an optimal scale. To explore the causes of scale inefficiencies, the study further examines the returns to scale patterns for each bank by both DEA and SFA. The results indicate that IRS is the exclusive cause of scale inefficiency. Specifically, all banks are operating at IRS except that
the large bank B2, which is operating efficiently, exhibits CRS. That is, for an inefficient individual bank, the operational efficiency can be further enhanced if the total amount of its IT investment can be increased.
Besides the correlations between DEA and SFA efficiency indices that have been discussed above, the results of the Pearson’s correlation analysis in Table 4 suggest that both pre-tax profits and total IT expenses appear to have a positive relationship with each of the other performance indices with only one exception.
Although not all relationships are statistically significant at the 10% test level, the traditional performance index, profits are significantly and positively related to total IT expenses, operating costs, CCRTE and SCALE. Moreover, total IT expenses are hardly related to any performance indices but significantly and positively related to scale efficiency, profits and operating expenses. These results imply that for the case of banking firms, the total spending for IT can improve profits significantly, but can neither reduce operating costs nor enhance operational efficiency significantly.
Table 4. Correlation matrix for performance indices
Profits IT exp. Op. costs CCRTE BCCPTE SCALE COBBTE
IT exp. 0.765***
Op. costs 0.948*** 0.817***
CCRTE 0.540*** 0.227 0.385**
BCCPTE 0.231 -0.163 0.068 0.740***
SCALE 0.495*** 0.332* 0.391** 0.893*** 0.410**
COBBTE 0.254 0.098 0.096 0.632*** 0.280 0.728***
TRANTE 0.204 0.088 0.058 0.601*** 0.261 0.715*** 0.987***
Note. * indicates significant at the 10% level; ** indicates significant at the 5% level; *** indicates significant at the 1% level.
As mentioned previously, the sample banks are classified into three types, publicly-owned, privately-owned and mutually-owned, according to their business ownership type. Table 5 compares the difference in efficiency between various types of banks. It can be seen in the table that, except for pure technical efficiency, publicly-owned banks have the highest efficiency values consistently, followed by mutually-owned banks; with the privately-owned banks having the lowest efficiency values. The ANOVA results do not support that there are no differences between the average efficiency values of different types of banks except COBBTE, indicating that the difference of ownership type has a significant effect upon the performance of the bank, with respect to the contribution of IT investment to operational efficiency. Moreover, the results of post-hoc tests using both Scheffé’s pairwise and Tukey’s HSD multiple comparison procedures suggest that the publicly-owned banks are significantly superior to the privately-owned banks in technical, pure technical or scale efficiency. On average, the mutually-owned banks have higher efficiency index than the privately-owned banks; however, this is true only for both technical and pure technical efficiencies obtained by DEA.
Table 5. Efficiency differences among the three types of bank ownership
CCRTE BCCPTE SCALE COBBTE TRANTE Ownership Type
Mean S.D. Mean S.D. Mean S.D. Mean S.D. Mean S.D.
[1] Public (n=5) 0.621 0.258 0.719 0.172 0.839 0.178 0.419 0.204 0.438 0.190 [2] Private (n=12) 0.246 0.213 0.517 0.069 0.452 0.350 0.364 0.342 0.373 0.351 [3] Mutual (n=13) 0.478 0.200 0.832 0.156 0.559 0.185 0.413 0.226 0.429 0.251
F-value 6.75 18.28 3.81 0.12 0.15
P-value 0.0047*** 0.0000*** 0.0350** 0.8879 0.8585
Post-hoc test [1>2] [3>2] [1>2] [3>2] [1>2] ʳ ʳ Note. ** indicates significant at the 5% level; *** indicates significant at the 1% level.
Table 6 investigates the impacts of IT investment on operational efficiency applying a Tobit regression model. In the regression model3, technical efficiency of banks obtained by DEA and SFA is the dependent variable and all input variables are regarded as the independent variables. In the first row of the table, the designation of regression models consists of the following two parts: (a) characters A, B and C represent that the bank data are taken from all banks (n = 30), privately-owned banks (n = 12) and mutually-owned banks (n = 13), respectively4; (b) Arabic numerals 1, 2, 3 and 4 indicate that CCR, BCC, Cobb-Douglas and translog models are employed to evaluate the dependent variables, i.e., efficiency indices, respectively.
Table 6. Impacts of IT adoption on operational efficiency in banking
Model Intercept IT personnel ATMs PCs/TEs Fin. cards IT services F R2 A1 0.8 *** -0.0011 0.0030 ** -0.0000 -0.0004 0.0030 ** 3.0 ** 0.383 A2 1.4 *** -0.0019 0.0016 ** 0.0000 -0.0001 0.0042 *** 12.8 *** 0.727 A3 0.3 -0.0017 0.0007 -0.0000 0.0000 0.0005 2.0 0.294 A4 0.4 -0.0000 0.0007 -0.0001 -0.0001 0.0002 2.1 0.286 B1 -1.1 0.0057 0.0041 * 0.0001 -0.0005 0.0033 3.6 * 0.749 B2 0.6 * 0.0005 0.0014 * 0.0000 -0.0002 0.0009 3.1 * 0.592 B3 -1.5 0.0152 0.0003 -0.0001 0.0001 0.0052 1.7 0.358 B4 -1.6 0.0158 0.0008 -0.0001 0.0000 0.0056 1.7 0.373 C1 0.8 -0.0191 * 0.0011 0.0011 -0.0020 0.0003 4.6 ** 0.765 C2 1.6 ** -0.0127 ** 0.0020 0.0007 -0.0007 0.0048 11.4 *** 0.890 C3 -0.3 0.0055 0.0246 ** -0.0008 0.0032 0.0097 ** 8.1 *** 0.852 C4 -0.5 0.0126 0.0245 ** -0.0014 0.0027 0.0116 ** 9.2 *** 0.868 Note. * indicates significant at the 10% level; ** indicates significant at the 5% level; *** indicates significant at the
1% level.
As exhibited in Table 6, only three variables, namely, number of IT personnel, number of ATMs and diversification of IT services contribute significantly to operational efficiency of banks, while the other independent variables fail to attain statistical significance even at a 90% confidence level. Among the three variables,
3 The study also carries out residual analysis and demonstrates that the regression model does not suffer from problems of autocorrelation, heteroskedasticity and residual non-normality, implying that the regression assumptions are not violated.
4 The regression results of the publicly-owned banks are not listed in the table, since those type of banks are not suitable for the regression model, even after adjustment.
the number of ATMs and the diversification of IT services have a positive impact on operational efficiency by DEA for all banks, and also have a positive impact on operational efficiency by SFA for the mutually-owned banks. Furthermore, the number of ATMs has a positive impact on operational efficiency by DEA for the privately-owned banks; while the number of IT personnel has a negative impact on operational efficiency by DEA only for the mutually-owned banks. The results reveal that to improve operational efficiency, banks are suggested to install more ATMs and to provide customers with convenient access to a wide variety of IT services. Moreover, the mutually-owned banks also require a cutback in IT personnel to increase operational efficiency.
Table 7 examines the effects of IT investment on operating costs using OLS regression analysis. In the regression model5, operating costs of banks are considered as the dependent variable and all input variables are viewed as the independent variables. In the table, models E, F and G indicate that the bank data are taken from all banks (n = 30), privately-owned banks (n = 12) and mutually-owned banks (n = 13), respectively6. We find that the numbers of ATMs, as well as PCs and terminals contribute significantly and positively, and the number of financial cards issued contributes significantly and negatively to operating costs for all banks. In particular, the number of ATMs has a greater potential for impacting operating costs than the number of PCs and terminals and the number of financial cards issued. As for the mutually-owned banks, the number of personnel has a statistically significant negative impact on operating costs, while the number of PCs and terminals has a statistically significant positive influence on operating costs. However, no significant relationships are found between IT input variables and operating costs for privately-owned banks. The results exhibit that an expansion of investments in ATMs, as well as PCs and terminals cannot help banks decrease their operating costs; however, banks are able to reduce their operating costs through an increase in the number of financial cards issued. Moreover, the mutually-owned banks can also save operating costs by cutting back IT staff and increasing the number of financial cards issued.
Table 7. Impacts of IT adoption on operating costs in banking
Model Intercept IT personnel ATMs PCs/TEs Fin. cards IT services F R2 E 4634.4 13.8 156.9 *** 13.3 *** -21.9 *** -59.5 91.4 *** 0.950
F 11482.0 118.3 -4.6 -2.1 15.1 -36.9 1.9 0.617
G 1696.1 ʳ -127.8 ** 48.0 ʳ 19.2 ** -9.7 *ʳ 1.6 ʳ 12.0 *** 0.896 Note. * indicates significant at the 10% level; ** indicates significant at the 5% level; *** indicates significant at the
1% level.
5. CONCLUSIONS
The study has investigated the impacts of IT investment on the operational performance of banks. By comparison with the existing literature, this study is
5 See note 3 above.
6 See note 4 above.
characterized by using the combination of nonparametric DEA and parametric SFA techniques to measure operational efficiency and explore the effects of IT investment on operating costs, which have been scarcely dealt with in the literature.
Our major findings and suggestions are outlined as follows.
First, low operational efficiencies existed in the banking industry during the study period. On average, the technical efficiencies measured by both frontier efficiency approaches are around 40%. According to the Pearson correlation results, there is a significant strong positive relationship between DEA and SFA efficiency indices. The bank management will be more confident of the evaluation results when adopting both approaches at the same time.
Second, generally speaking, the operational inefficiency may be caused mainly by a combination of both pure technical inefficiency and scale inefficiency.
Therefore, in order to enhance operational efficiency, bank management has to solve the problems of wasting IT resources and operating at an inappropriate scale.
Since IRS is the dominant source of scale inefficiency, an increase in the total amount of IT investment is a suggested tip for improving the operational efficiency for an inefficient individual bank.
Third, the Pearson correlation results indicate that total IT expenses are not significantly related to any performance indices but significantly and positively related to profits. These results suggest that for the case of banking firms, the total IT spending can probably improve profits significantly, but can neither reduce operating costs nor enhance operational efficiency significantly.
Fourth, the results from ANOVA exhibit that the difference of ownership type has a significant effect upon bank performance, with respect to the contribution of IT investment to operational efficiency. Moreover, the results of post-hoc tests suggest that the publicly-owned banks are significantly superior to the privately-owned banks in technical, pure technical or scale efficiency. A possible explanation is that the publicly-owned banks are relatively large-scale in IT investments and applications; they are hence able to achieve greater economies of scale and scope, so their operational performance is detected to be the best.
Fourth, according to the Tobit regression results, both the number of ATMs and the diversification of IT services contribute positively to the operational efficiency of banks. Consequently, banks are suggested to install more ATMs and provide customers with convenient access to a wide variety of IT services to improve their performance. Furthermore, the operational efficiency of the mutually-owned banks is negatively affected by the number of IT personnel.
Therefore, besides installation of more ATMs and provision of a wide variety of IT services, the mutually-owned banks also require a cutback in IT personnel to promote performance.
Finally, the OLS regression results reveal that the numbers of ATMs, as well as PCs and terminals contribute positively, and the number of financial cards issued contributes negatively to operating costs for all banks; and the number of personnel also contributes negatively to operating costs for the mutually-owned banks. These results can be interpreted that an expansion of investments in ATMs, as well as PCs and terminals can probably not help banks to decrease their operating costs;
however, banks are able to reduce their operating costs through an increase in the number of financial cards issued. Besides that, the mutually-owned banks can also save operating costs by cutting back IT staff.
ACKNOWLEDGEMENT
The author appreciates the financial support of the National Science Council of the Republic of China with Project No. NSC 90-2416- H-309-010. Particularly, the author would like to acknowledge many senior information officers of banks and credit cooperatives for their kind help with the collection of data, most of which are unpublished.
REFERENCES
Aigner, D. J., Lovell, C. A. K., & Schmidt, P. (1977). Formulation and Estimation of Stochastic Frontier Production Function Models. Journal of Econometrics, 6(1), 21-37.
Andersen, T. J., & Foss, N. J. (2005). Strategic Opportunity and Economic Performance in Multinational Enterprises: The Role and Effects of Information and Communication Technology. Journal of International Management, 11(2), 293-310.
Banker, R. D., Charnes, A., & Cooper, W. W. (1984). Some Models for Estimating Technical and Scale Inefficiencies in Data Envelopment Analysis.
Management Science, 30(9), 1078-1092.
Battese, G. E., & Corra, G. S. (1977). Estimation of a Production Frontier Model:
with Application to the Pastoral Zone of Eastern Australia. Australian Journal of Agricultural Economics, 21(3), 169-179.
Brynjolfsson, E. (1993). The Productivity Paradox of Information Technology.
Communications of the ACM, 36(12), 67-77.
Charnes, A., Cooper, W. W., & Rhodes, E. (1978). Measuring the Efficiency of Decision Making Units. European Journal of Operational Research, 2, 429-444.
Devaraj, S., & Kohli, R. (2000). Information Technology Payoff in the Health-Care Industry: A Longitudinal Study. Journal of Management Information Systems, 16(4), 41-68.
Farrell, M. J. (1957). The Measurement of Productive Efficiency. Journal of the Royal Statistical Society, 120, 252–281.
Ham, S., Kim, W. G., & Joeng, S. (2005). Effect of Information Technology on Performance in Upscale Hotels. Hospitality Management, 24, 281-294.
Lee, B., & Menon, N. M. (2000). Information Technology Value through Different Normative Lenses. Journal of Management Information Systems, 16(4), 99-119.
Loveman, G. W. (1994). An Assessment of the Productivity Impact of Information Technologies. In T. J. Allen & M. S. Scott Morton (Eds.), Information Technology and the Corporation of the 1990s: Research Studies (pp. 84-110).
Cambridge, Massachusetts, USA: MIT Press.
Meeusen, W., & van den Broeck, J. (1977). Efficiency Estimation from Cobb-Douglas Production Functions with Composed Error. International Economics Review, 18(2), 435-444.
Mitra, S., & Chaya, A. K. (1996). Analyzing Cost-Effectiveness of Organizations:
The Impact of Information Technology Spending. Journal of Management Information Systems, 13(2), 29-57.
Rai, A., Patnayakuni, R., & Patnayakuni, N. (1997). Technology Investment and Business Performance. Communications of the ACM, 40(7), 89-97.
Strassman, P. A. (1990). The Business Value of Computers: An Executive’s Guide.
New Canaan, Connecticut, USA: Information Economics Press.
Weill, P. (1992). The Relationship between Investment in Information Technology and Firm Performance: A Study of the Value Manufacturing Sector.
Information Systems Research, 3(4), 307-333.
Chu-Fen Li received her M.B.A. degree from National Chengchi University, Taiwan and her Ph.D.
degree in information management, finance and banking from Europa-Universitat Viadrina Frankfurt, Germany.
She is now an Assistant Professor in the Department of Finance and Graduate Institute of Business and Management, National Formosa University, Taiwan. Her major research interests include performance evaluation, Internet banking, online auctions, and Internet pricing. Her works have been published in several international refereed journals, such as European Journal of Operational Research, International Journal of Information and Management Sciences, Journal of System and Software, and Asian Journal of Management and Humanity Sciences.
She has also participated in international conferences in Germany, UK, Japan, Singapore, Hong Kong, and Taiwan.
APPENDIX: INFORMATION TECHNOLOGY QUESTIONNAIRE
Dear information officer,
We are conducting an academic research on the subject of the impact of information technology on operational performance of banks. It would be appreciated if you would complete this questionnaire. The results of the questionnaire will be used for purposes of this research only. And all information you provide will be kept strictly confidential. Thank you for participating in this study.
Graduate School of Business and Operations Management, CJCU
Instructions for Completing this Questionnaire
Information technology (IT) is comprised of computer hardware, software, information processing, communication networks, satellite communications, robotics, videotext, cable television, e-mail, databanks, optical recognition system, automatic service machines, and automated office equipment, but telephones, fax machines, money counters/counting machines, video cassette recorder, security equipment are excluded.
All branches refer to a bank’s total offices which formally open for business, including headquarters, domestic branches, simplified branches, representative offices, self-service banks, foreign branches, and offshore banking units (OBUs).
Officer’s Name: ____________________
Title: _____________________________
Department: _______________________
Telephone: (0 ) ___________________
Fax: (0 ) ________________________
I. Bank Data
Name of bank: ________________________________________________(Commercial) Bank/ Credit Cooperative/ Credit Department of Farmers and Fishermen’s Association
Date of foundationΚMonth ________Day ________Year ___________
Total number of the following branches (please fill in the blanks with “0”, if you haven’t this kind of the office.)
Year 1996 1997 1998 1999 2000
1. Domestic branches (inc. headquarters)
2. Simplified branches
3. Representative offices
4. Self-service banks
5. Foreign branches
6. OBUs
II. Full-Time Employees of All Branches
III. Quantity of IT Equipment
Automated teller machines (ATMs) include CD/ATMs and foreign exchange ATMs, excluding passbook entry machines, information kiosks, and multimedia stations.
Phone banks refer to Interactive voice response (IVR) and computer telephony integration (CTI) call centers, excluding mobile banks and Internet banks.
Year 1996 1997 1998 1999 2000
Number Number of transactions (in
thousands) 1. ATMs
Total amount of deposits, withdrawals and transfers (in millions)
Number 2. Terminals Number of transactions (in
thousands) 3. PCs & stations Number
Number 4. Phone banks Number of transactions (in
thousands)
5. Funds transfer centers Number of transactions (in thousands)
IV. Expenditure of IT Equipment
Total IT expenditure refers to incurred expenses for maintenance and repair, rent, depletion of IT equipment, and information sourcing services.
Year 1996 1997 1998 1999 2000
1. Total IT expenditure (in millions) 2. Total sourcing costs (in millions)
Year 1996 1997 1998 1999 2000
1. Number of employees 2. Number of IT employees
3. Total salary for employees (in millions) 4. Total salary for IT employees (in millions )
V. Information Services except Deposits, Withdrawals and Transfers
Information service system
Established or prepare to establish (please fill in the date of completion)
Not established (please mark with an “X”) 1. Stored-value card system Month _____ Year ______
2. Debit card system Month _____ Year ______
3. Credit card system Month _____ Year ______
4. Trust service system Month _____ Year ______
5. Foreign exchange system Month _____ Year ______
7. Financial electronic data interchange (FEDI) Month _____ Year ______
8. Website Month _____ Year ______
9. Internet banking system Month _____ Year ______
10. Mobile banking system Month _____ Year ______
11. CTI call center Month _____ Year ______
12. Dual host backup system Month _____ Year ______
Month _____ Year ______
13. Dual center backup system Please mark with an “X”:
ϭ Self-built centerʳ ϭ Sourcing backup Month _____ Year ______
14. Network backup system Please mark with an “X”:
ϭ Dial up backupʳ ϭ Always on backup
