Electronic copy of this paper is available at: http://ssrn.com/abstract=478323

INFORMATION TECHNOLOGY AND PRODUCTIVITY CHANGES IN THE BANKING INDUSTRY

di Luca Casolaro* e Giorgio Gobbi**

Abstract

This paper analyzes the effects of investment in information technologies (IT) in the financial sector using micro-data from a panel of 600 Italian banks over the period 1989-2000. Stochastic cost and profit functions are estimated allowing for individual banks displacements from the best practice frontier and for non-neutral technological change. The results show that both cost and profit frontier shifts are strongly correlated with IT capital accumulation. Banks adopting IT capital intensive techniques are also more efficient. On the whole, over the past decade IT capital deepening contribution to total factor productivity growth of the Italian banking industry can be estimated in a range between 1.3% and 1.8% per year.

JEL classification: D24, G21, O33.

Keywords: banking, productivity, efficiency, information technology.

(*) Banca dItalia, Filiale di Livorno, Piazza del Municipio 47, 57123 Livorno Italy. E-mail address: luca.casolaro@bancaditalia.it.

(**) Banca dItalia, Economic Research Department, Via Nazionale 91, 00184 Rome Italy. E-mail address: giorgio.gobbi@bancaditalia.it.

Electronic copy of this paper is available at: http://ssrn.com/abstract=478323

1

1. Introduction1

This paper examines the effects of investment in information processing,

telecommunications, and related technologies (IT) on productivity growth in the

banking industry. Several reasons make this sector a particularly interesting case for

testing the impact of IT capital deepening. First, financial services account for between

5 and 10 % of GDP in most industrialized countries. Banks contribute to a large

proportion of financial-related output and play an important role in key activities for the

functioning of the economy, such as the allocation of capital and the provision of

liquidity, payment and safekeeping services. Efficiency improvements in the production

and delivery of these services can bring widespread welfare effects.2 Second, the

acquisition and the treatment of information is a central activity in banking and the

impact of innovations in IT is likely to be larger than in other industries. Finally, the

stability of the banking system is a primary policy objective. Innovations affecting

production costs and profit opportunities can have major consequences on the industry

structure altering banks incentives to take risks and need to be monitored by regulators.

Although banking is among the most IT-intensive industries and among those that

started earlier to rely massively on computers for their operations, the large bulk of

applied literature on bank technology includes very few studies on this topic, mainly

1 We acknowledge the helpful suggestions of Allen N. Berger, Luigi Guiso, Salvatore Rossi,

Stefano Siviero, two anonymous referees and the participants at the seminars held at the Bank of Italy and the Federal Reserve Board. We are also indebted with Tim Coelli who kindly made available the econometric software FRONTIER 4.1, used in part of the estimations. Ginette Eramo and Roberto Felici provided outstanding research assistance. All the remaining errors are ours. The views expressed in this paper do not necessarily represent those of the Bank of Italy. 2 The following straight calculation provides a quantitative insight of how large these effects could be:

If a typical economy has a capital output ratio of 3:1, it follows that an increase in the efficiency with which capital is allocated of 2 percent or roughly 20 basis points has a social benefit equivalent to that of 6 percent of GNP in forgone consumption. (Summers (2000), pp 2-3).

2

because of the paucity of appropriate quantitative information. We overcome this hurdle

using a panel of more than 600 Italian banks over the period 1989-2000, which includes

information on both investment and expenses related to computers and software. To our

knowledge, no prior study on the banking industry has measured directly the impact of

IT diffusion on such a comprehensive basis.

Using stochastic cost and profit frontier techniques we are able to estimate the

relationship between total factor productivity (TFP) changes and the use of IT. We find

a positive correlation between IT capital intensity with both frontier shifts and

efficiency scores. We interpret our results as strongly supporting the hypothesis that the

diffusion of digital technologies fosters TFP growth. The economic magnitude of the

effect is also relevant: IT investment has allowed banks to reduce costs by 1.3% a year

and increase short run profits by 2.0%.

We also show that a large number of other factors influence TFP, some of them

reinforcing the IT drift, others curbing it. Most of these factors reflect restructuring and

changes in regulation over the sample period that affected the banking industry in Italy,

as well in numerous other countries. As a consequence, in the absence of detailed data,

inferences about the impact on bank costs or profits of any single factor should be taken

prudently.

The paper is organized as follows. Section 2 selectively reviews the literature

related to our paper. Section 3 highlights the key developments of the Italian banking

industry over the sample period. Section 4 lays out the methodology for the empirical

analysis and Section 5 presents the data. In section 6 we display our empirical results

and Section 7 briefly concludes.

3

2. Related literature

Over the past 20 years firms responded to the fall in the quality-adjusted price of

computers and related technologies by substituting IT capital for others production

factors. Computers share with other landmark technological innovations the feature of

being a multipurpose technology, which can be adopted into virtually every economic

activity. Historically, technological advances that falls in this class were followed by

major total factor productivity pushes although the empirical evidence varies greatly

across time periods, countries, and measurement strategies.

Studies on TFP growth in the U.S. conducted during the 1980s and the early

1990s generally found that the use of computers had a negligible impact on productivity

(Roach (1987); Berndt and Morrison (1997)). Later research found evidence of a

significant contribution of IT capital to the recent productivity growth in the US (e.g.

Jorgenson and Stiroh (2000); Oliner and Sichel (2000); Jorgenson (2001)) albeit more

skeptical scholars argued that productivity gains were largely owed to cyclical factors

(Gordon (1999)).

Most of these studies are based on aggregate industry data and their conclusions

are likely to underestimate the effect of IT on productivity for two reasons. First, the use

of computers is often associated with large changes in the quality of output that are

difficult to measure accurately (Boskin et al. (1997)), especially when aggregating

broad ranges of products and services. Second, the use of digital technologies is likely

to require time to adjust workplace organization and employees skills, and industry

data can average out large cross-firm variability due to different stages in IT adoption.

As a consequence, recent research has shifted to micro-data trying to uncover the

4

transformation that digital technologies induce on firm organization and workplace

activities.3

Bresnahan et al. (2002) find a positive correlation between the accumulation of IT

capital, innovation in workplace organization and the use of more highly skilled

workers. Brynjolfsson and Hitt (2003) estimated a production function for a panel of

600 large US firms, finding that the contribution of IT investments to output growth

significantly exceeds its factor share, implying a positive effect of computers on

productivity growth in the long-run. The results also suggest that IT capital deepening is

associated with far-reaching organizational changes within the firm.

The policy relevance of IT investment has spurred international organizations to

collect data and to sponsor research programs. OECD (2004) gathers a set of empirical

papers that offers a comprehensive overview of the impacts of IT on economic

performance in advanced countries. The nine studies based on micro-data show

significant impacts of digital technologies on firm-level performance. In most cases

there is also evidence that IT investment is associated with more rapid TFP growth and

that this effect differs across industries. Firms in the financial sector are among those

that have benefited more from the new technologies.

To our knowledge, there are only two academic papers that directly investigate the

effects of IT in the banking industry, and both of them use data from case studies4.

Parsons et al. (1993) estimate a cost function using data from a large Canadian bank

over the period 1974-1987 and find a weak but significant correlation between

3 Pilat (2004) reviews most of the relevant literature in this field.

4 By contrast, the issue is widely debated within the industry. Olazabal (2002) claims that many of the

post-1995 IT investments were aimed at rising revenue-increase activities that have yet to be repaid, leading to a decline in banking productivity due to excess capacity.

5

productivity growth and the use of computers. Autor et al. (2002) examine the

introduction of automatic image processing of checks in one of the top 20 U.S. banks.

They argue that this single technological innovation led to complex organizational

changes and distinct effects on different departments of the bank. Computers turn out as

substitute for low skilled works in standardized tasks, but complementary to computer-

skilled labor.

Leaving out of consideration technological innovations, there is a large body of

bank studies that investigate productivity growth by evaluating changes over time both

in the best practice frontier and in the average industry efficiency levels (Berger

(2003)). A common finding of these studies is that in the last quarter of a century the

banking industry experienced small, if not negative, TFP growth, owing to the

overlapping of a wide range of shocks.

Deregulation seems to have had prolonged depressing effects on productivity

growth in several countries. Using data from large samples of US banks, Berger and

Humphrey (1991), Bauer et al. (1993), Humprey and Pulley (1997) and Wheelock and

Wilson (1999) find that the deregulation of deposit rates was followed by negative

productivity growth rates for most of the 1980s. Fiercer competition among banks and

from non depositary institutions forced organizational changes, such as the opening of

more branches that, at least in the short run, raised average costs. This pattern has been

partially reversed only in the late 1990s when US banks were able the ripe the gains

from the higher quality services (Stiroh (2000), Berger and Mester (2003)).

Berg et al. (1992), argue that in the late 1970s deregulation in the Norwegian

banking sector was followed by a prolonged period of excess capacity, absorbed only in

the 1980s. Grifell-Tatje and Lovell (1996) find that deregulation induced a sharp decline

in the productivity of the top performer Spanish saving banks in the late 1980s.

6

Kumbhakar et al. (2001) using a larger sample and a longer time horizon panel show

that the decline is mostly attributable to an increase in the average industry inefficiency

level. Gobbi (1996) also finds a decline in productivity for a sample of large Italian

banks in the aftermath of the regulatory reforms of the early 1990s. Casu et al. (2003)

in a comparative study of European banks during the 1990s find a recovery in

productivity growth brought about mainly by shifts in the best practice frontier whereas

there does not appear to have been catch-up by non-best-practice institutions.

A second source of shocks is the wave of mergers and acquisitions, which in the

past 20 years reshaped the financial systems in most countries. Rebounds on efficiency

and productivity growth vary according to country and time-period, but a general

finding is that the final effects tend to materialize after several years from the deals

(Amel et al. (2004)) owing the long time required by post merger restructuring. This

implies that in the past two decades the performance of a sizeable fraction of the

banking industry has been influenced by the adjustment to larger and more complex

business organizations.

Finally, the pattern of business specialization in the banking industry has changed

dramatically. Revenues accruing from non-interest income have been on the rise

everywhere and there is some evidence that this has brought about an increase in

operational efficiency (Rogers (1998); Vander Vennet (2002)).

The Italian banking industry is a highly representative case for documenting the

effects of IT capital accumulation in business environment shaped by deep internal

restructuring and several concurrent external shocks. The following section briefly

describes the basic facts.

7

3. The Italian banking industry in the 1990s

Our sample period starts in 1989, when IT technologies had already been adopted

by the vast majority of banks, and extends up to year 2000, covering a further wave of

IT capital deepening. In the first half of the 1990s the IT-related expenses of Italian

banks accounted for about 9 % of total operating costs, peaking above 13 % in 1999

with the introduction of the single currency and the Y2K (Table 1). Annual investment

in hardware increased by six times, after adjusting for hedonic price changes, that on

software by three times (Table 2). In 2001 the total real value of hardware, software and

EDP specific plants per employee was almost six times higher than in 1989.

Over the sample period the Italian banking industry experienced a variety of

shocks ranging from severe macroeconomic fluctuations to sweeping changes in

regulation and ample demand shifts. A new banking law was passed in 1993 which

abolished a wide range of regulatory restrictions, barriers to entry and market

segmentations. It also paved the ground for the privatization of virtually all banks under

public control.

The prolonged slow down of GDP growth winding up in the 1994 recession

depressed the demand financial services and led to widespread excess capacity across

the industry. Rigidity in staff costs and massive provisions for loan losses brought down

net return on equity from 10% to 1% between 1990 and 1995, while the cost-income

ratio increased by almost seven percentage points.

Post deregulation heightened competition and low profit margins triggered off a

wave of mergers and acquisitions swept through the industry involving some 400

mergers and 500 deals in which the control of one banks changed hands.

8

Performance subsequently recovered along with a decrease in the burden of staff

costs and bad loans and with a strong increase in revenues from fees and commissions.

Measures of productivity growth in our data sample are likely to reflect the

compounded effects of all these shocks. Our empirical strategy consists in testing the

existence of a link between the usage of IT and changes in bank efficiency and

productivity in an indirect way.

4. Modeling strategy

4.1 IT investment and TFP growth

Following a well established methodology, we decompose productivity growth

into changes in the best practice and change in efficiency. Our approach is very close in

spirit to that of Berger and Mester (2003).

The best practice reflects the production choices of the best performing firms in

the industry. It identifies a frontier, but not necessarily the technological frontier

because of other factors such as (de)regulation or macroeconomic shocks, that may

prevent even the best existing bank from adopting the fully efficient technology.

Nonetheless, as long as technical progress shifts the efficient technological frontier it

should also shifts the best practice frontier. This is the first link between IT capita

deepening and TFP growth.

The second link between IT investment and TFP growth is the change in average

industry efficiency. For each bank an (in)efficiency score can be defined in terms of its

distance, expressed in a proper metric, from the best practice frontier. Idiosyncratic

shocks, poor management, and late adoption of new technologies concur to determine

an ongoing average level of inefficiency within a given industry. A well-documented

9

stylized fact is that the pattern usage of new technologies over time typically follows an

S-curve (Geroski (2000)). Firms lagging behind are likely to be less efficient just

because they use obsolete production plans. As long as the diffusion of a new

technology gains momentum we should observe the closing up of the efficiency gap.

Firms that move over time from lower to higher efficiency levels add to industry TFP

growth.

To account for the two channels linking IT and TFP growth, we apply the

stochastic frontier methodology developed by Battese and Coelli (1993), which allows

us to estimate simultaneously both a time variant best practice and the pattern of

industry inefficiency.

4.2 The empirical model

Our analysis builds on the specification of dual functions in place of the standard

primal production function owing to the multiple nature of the bank output and to the

measurement problems associated with services provided through the activities of

lending and fund-raising (Hanckock (1991)). The details are discussed in the Appendix.

First we estimate a stochastic translog cost function with a time polynomial

included as a production factor to allow for non-neutral technological progress.

(1)

where the yns are the N outputs and the pms are the prices of the M inputs. These

variables are in logs and standardized by subtracting the sample means. Standard

symmetry and linear homogeneity conditions are imposed. Bank and time subscripts i

ititttttm

M

m

tpn

N

n

ty

mn

N

n

M

m

pykm

M

m

M

kppln

N

n

N

lyym

M

m

pmn

N

n

yn

uvrGNPtttpty

pyppyypyC

mn

mnkmln

++++++++

+

++++=

==

= == == ===

212

2111

1 11 11 1110

21

10

and t are omitted for simplicity. We include the growth rate of the real GNP (GNPt) and

the real official discount rate (rt) in order to extract the business cycle effect from the

time polynomial. The cost inefficiency term uit measures how close the costs of bank i

at time t are to those of a bank on the efficient frontier, producing the same output. The

term vit stands for the usual white noise error terms.

The best practice frontier estimated through a cost function focuses on the input

side of bank production and may overlook important dimensions of banks decision

making. Berger and Mester (1997) argue that the use of a profit frontier can be more

appropriate than that of a cost frontier because it accounts for both output and input

suboptimal banks decisions. However, in standard profit analysis output prices are

taken as exogenous, considering profit inefficiency as a sub-optimal choice with respect

to input-output relative prices. This is not an accurate representation of the banking

industry because a large share of revenues originates from bilateral contracting in which

banks have some bargaining power. We therefore adopt an alternative formulation of

the profit function, introduced by Humphrey and Pulley (1997), in which output prices

are flexible while output levels are taken as given, like in the cost function. Errors in the

choice of outputs do not affect the efficiency vector, while errors in setting output

prices do.5 The estimated efficiency levels in this case measure how close a bank is to

its maximum profit given an input-output quantity mix. The price to be paid for this

gain is the assumption that profits are maximized conditional on output levels, which is

also a very demanding hypothesis. Nonetheless, we estimate an alternative profit

5 The standard profit function presents the drawback that is not defined for constant or increasing

return to scale technologies. In that case the appropriate model is the short-run profit function in which some production factors are constrained to be operated at a fixed level. These and other technical difficulties of the standard dual frontier have made the alternative profit function an increasingly popular

11

function model as a consistency check for the cost function findings because previous

research shows that the results may vary substantially according to which dual model is

specified and estimated.

The translog alternative profit frontier is specified as:

ititttttm

M

m

tpn

N

n

ty

mn

N

n

M

m

pykm

M

m

M

kppln

N

n

N

lyym

M

m

pmn

N

n

yn

uvrGNPtttpty

pyppyypy

mn

mnkmln

++++++++

+

++++=

==

= == == ===

212

2111

1 11 11 1110

21

pi (2)

where the variable uit measures inefficiency in terms of forfeited profits.

In both cost and profit analysis the inefficiency term is modeled as follows:

=

+++=J

jittitjjit wtzv

10 (3)

where 0 is a constant term, zit are individual time-varying covariates, t is a linear time

trend to account for time patterns of inefficiency, and itw a white noise stochastic

disturbance . We estimate equation (1) and equation (2) jointly with equation (3) using

maximum likelihood techniques that produce asymptotically consistent and efficient

estimates.

Among the covariates zj we include a measure of IT capital, because the pace at

which different banks have invested in the new technologies can be very different. On

the one hand, the speed of adoption of IT can be hampered by lack of information on

how to implement them or by the need for workforce skills that are obtainable only

through learning-by-doing processes (following the epidemic models of technology

model in applied banking research (see, among others, Altunbas et al. (2001) Clark and Siems (2002), Vander Vennet (2002), Berger and Mester (2003)).

12

diffusion). On the other hand, different firms may want to adopt the new technology at

different times because of idiosyncratic adjustment costs. A difference in the adoption

rate of new technologies across banks is likely to be reflected in the distribution of

efficiency levels.

Let k)

be the estimated coefficient of IT capital obtained from equation (3). If its

sign is negative (positive) a raise in IT capital reduces (increases) the extra cots or

forfeited profits due to inefficiency. For each period t, we compute the average

contribution of IT to industry inefficiency as follows:

INEFF_ITt = (100*i k)

ITit)/Ht (4)

where Ht is the number of existing banks at period t. INEFF_ITt is a measure,

expressed in percentage points, of the average cost reduction or of the profit increase

due to the impact of IT on industry inefficiency. The first difference INEFF_ITt is the

annual change in average inefficiency induced by IT capital accumulation. Averaging

over the time the INEFF_ITts we obtain a measure of the average yearly TFP change

due to the effect of IT capital accumulation on industry efficiency.

As a last step we investigate the correlation between IT capital accumulation and

shifts in the cost and profit frontiers for banks over time. The idea behind this exercise

is that frontier shifts are driven by the introduction of new best practices along with the

increase of IT capital.6

6 Salter in his seminal empirical investigation on productivity and technical change observed that

gross investment is the vehicle of new techniques, and the rate of such investment determines how rapidly new techniques are brought into general use and are effective in raising productivity. (Salter (1966), p. 17).

13

We first compute the time derivative of the frontier function for each bank in each

year, which is given by:

FSit = jM

jpti

N

iyttt pytt

fji

==

+++=

1121 2 (5)

where the f stands for the cost or the profit function. For each bank i the time derivative

is computed taking the projection on the frontier of its input-output mix. The frontier

shifts represents the variation in total cost or profit for a given level of outputs and

inputs owing to TFP changes.7

As in the case of idiosyncratic inefficiency changes, frontier shifts reflect the

sum of a variety of shocks affecting the industry, among them the effects of technical

progress. More importantly, these shocks are likely to hit each bank with different

intensity and this is reflected on the frontier, which is estimated using information on all

the production units in the sample.

To filter the effect of IT on frontier shifts, we estimate the panel of banks frontier

shifts for each year with a fixed effects regression of the kind:

itti

J

jjitj w +++=

=1itFS (6)

where the wjs are time-varying bank specific variables, the is are bank fixed effects,

the ts are time dummies and it the white noise disturbance error. This procedure is

very close to the approach proposed by Brynjolfsson and Hitt (2003). As in the case of

the inefficiency estimates among the wj variables we include a measure of IT capital. In

7 For a thorough discussion of the technical and economic meaning of these derivative in the case of

the cost function we refer to Hunter and Timme (1986).

14

a similar way we obtain an estimate of the average contribution of IT to the frontier

shift in each period as:

FS_ITt =(100*i k)

ITit)/Ht (7)

where k)

is the estimated coefficient of IT capital obtained from equation (6) and, as

before H the number of banks indexed by i. The time mean of FS_ITt is our measure of

the average yearly TFP change due to the frontier shifts induced by IT capital

accumulation.

5. Data

5.1 Sample

In our empirical analysis we employ data referring to an unbalanced panel of 618

Italian banks over the period 1989-2000. We use information from the Supervisory

Returns Database of the Bank of Italy, which includes detailed information on balance

sheet and profit and losses accounts disclosed by banks complying to prudential

regulation requirements. Data on IT expenses, investment in software and hardware,

outsourcing and the number of employees working in the computing centers are

collected yearly.

The panel has three sources of attrition. The first one is due to the demography of

the industry: some banks go out of businesses and others are established. This is

accommodated using unbalanced panel estimation techniques. The second one owes to

banks that disappear because of mergers. We adjust for that by computing pro-forma

balance sheets and profit and loss accounts. Possible productivity shifts of the banks

involved in M&A deals are accounted for by introducing the share of the target banks

15

into the joint assets of the target and the acquiring banks among the explicative

variables of both the inefficiencies and the frontier-shifts equations. Acquisitions that

maintain the charter of the target bank within a banking group are controlled for by

means of dummy variables.

The third source of attrition is the change in the information requested to the

banks for supervisory purposes. Up to 1994 the very small-sized banks were exempted

from filing in the data on IT investment. Subsequently, the more extensive information

requirement was enforced gradually in order to minimize the compliance costs of these

banks. This affects about 170 banks in our sample, although the number of those that

for which this information becomes available increases over time.

Finally we exclude from the sample the branches of foreign banks because their

balance sheets and profit and loss accounts are strongly affected by the strict

relationships they entertain with their parent companies.

In our sample there is a great variance across banks owing to size, business focus

and organizational complexity. In principle, banks of different size could have access to

different production set and this could distort the estimates of the parameters of the

frontier functions. Thus we divide the sample banks into three groups according to their

average total assets during the sample period. The first group includes the smallest two

thirds of the sample and contains almost exclusively cooperative banks: we call it small

banks. In the second group, our medium-sized banks, we include banks between the top

33 and the top 10% of total asset distribution, while the large banks are those in the top

10% of the distribution.

16

5.2 Outputs and inputs

The appropriate representation of the bank production operations and, in

particular, the definition of bank output is a long-standing and controversial issue. In

this paper we follow what has come to be known as the value added approach, in which

all the activities which produce value added are considered outputs of the bank. We

classify services produced and supplied by banks into two broad areas.

The first group includes services for which banks charge direct fees such as

payment execution, safekeeping, brokerage, and asset management. Although this type

output is usually overlooked in previous research, they represent a major source of

revenues for the banking industry.8 These outputs are likely to be standardized and in

principle it would be possible to have appropriate physical measures for them, such as

the number of transactions. However, data are collected in a rather aggregate form, and

owing to quality heterogeneity they are at best only proxies for output levels. Moreover,

banks supply a large number of different services which can hardly be accommodated

in an econometric model. We have therefore combined the information on three broad

categories of services in a single composite output using their shares in total fees and

commission income as weights. Specifically, we have the turnover of current accounts

as a proxy for payment services, the total amount of deposited securities as a proxy for

both safekeeping and asset management, and loan guarantees as a proxy for off-balance

intermediation activity.

8 Rogers (1998) and Clark and Siems (2002) are instances of the few exceptions. Both these papers find

that bank efficiency scores are affected by the off-balance-sheet activities which are one of the sources of fees and commissions bank revenue.

17

The second group of services consists of traditional intermediation activity:

raising and lending funds. According to the theoretical micro-foundations of financial

intermediation, banks produce financial contracts taking advantage of economies of

scale in transaction technologies as well as in screening and monitoring activities. The

revenue from intermediation services is given by the spread between the interest paid

on financial liabilities and the interest charged on assets. Since this kind of activity

tends to be customer-specific, there is no clear-cut distinction between prices and

quantities. Intermediation services are to a large extent not standardized and it is hard to

find compelling physical measures. In the applied literature they are usually assumed to

be proportional to the face value of assets and liabilities, meaning that a 100 loan

corresponds to two times the amount of services produced with a 50 loan. The

outstanding amounts of loans to and deposits from the non-bank sector are assumed to

be the main carriers of these kinds of services. The exclusion of other items, such as

interbank accounts and securities, can be motivated on the ground that they contribute

little to the value added once their user costs are properly taken into account (Hancock

(1991) and Fixler (1993)). The idea behind the value added approach is to identify bank

output as the repackaging of assets and liabilities not traded in organized markets. For

instance, interests received from and capital gains accrued on a portfolio of securities

add to bank revenues, but not to bank output, because there is no value added originated

by the fact that that portfolio is held by a bank instead of a private investor. On the

contrary banks repackage the funds raised from depositor in a format that make them

loanable to the borrowers adds to both revenues and output. The depositors benefit from

the insurance against liquidity shocks and a large number of borrowers would not be in

18

a position to raise funds directly on the capital market. Under this perspective interbank

loans are very close to securities and are treated in the same way.9

We consider separately short-term loans, those with original maturity up to 18

months, and medium- and long-term loans on the grounds that they are likely to differ

in terms of the amount of resources they require for their origination, screening and

monitoring. In Italy, most of the short-term loans consist of non-collateralized

commercial and industrial lending and medium and long-term loans mainly include

mortgaged lending.

On the input side we have considered labor and two types of physical capital

services: IT capital and other premises and equipment. For operating expenses other

staff and capital expenditure, since we do not have enough information to separate

prices from quantities, we assume that they enter into the production in fixed proportion

with other inputs. The IT price is calculated as a usage cost, given by the ratio between

IT expenses and the value of IT capital stock. As a proxy for the price of labor we use

the average wage, while the price of the other type of physical capital is constructed as

the ratio of the capital expenses divided by its book value. We also include estimates of

the value of leases consistently with the definition of the expenses on the left hand side.

Total operating expenses are the left-hand side variable in the frontier estimation.

Price homogeneity of the cost function has been imposed by normalizing both costs and

input prices by the price of labor. Outputs and inputs have been deflated by the

consumer price index in order to avoid the estimated coefficients of the time variable

9 An anonymous referee pointed out that this treatment of interbank transactions may underrate the

connection between IT investment and the creation of an electronic screen based market for interbank deposits (loans) in 1990. However, since over the sample period only medium and large Italian banks traded funds on the screen based market, the sample splits according to different banks size classes capture most of the innovation spurring effects of these transactions.

19

being affected by inflation. In the profit function the dependent variable is the value of

total revenue before taxes.

5.3 IT capital

The IT capital stock at the micro level has been computed by applying the

permanent inventory method to the real investment in hardware, software and premises

for computing equipment. The value of banks investment has been deflated using

hedonic price indexes of software and hardware developed by the Bureau of Economic

Analysis (BEA) and adjusted for the variation in the ITL/USD exchange rate.10 The

U.S. deflator is the only one available for each group of products; its application to

Italian data may be justified by the high share of imported IT equipment and the

existence of a global market for these products. Capital stock is obtained as the sum of

past investment flows, weighted by the relative efficiency in production of different

vintages. We assume the depreciation rate is constant in time but different across goods.

Following Seskin (1999) and Jorgenson and Stiroh (2000) software is assumed to

depreciate at a yearly rate of 44%, hardware and dedicated premises by 32 %yearly.

5.4 Explanatory variables of inefficiency levels and frontier shifts

On the right-hand side of the inefficiency and the frontier-shift regressions we

include several explanatory variables to control for environmental and bank individual

characteristics. To investigate the impact of IT capital accumulation on productivity we

introduce the variable IT_CAP, representing the amount of IT capital stock per

10 Data have been downloaded from the web site www.bea.org.

20

employees; in a more detailed analysis we split this variable in two, representing

software and hardware capital stock (respectively SOFT_CAP and HARD_CAP). The

variable IT_STAFF represents staff costs for IT personnel divided by total staff costs;

the variable OUTSOURCE is given by the fraction of IT expenses coming from

outsourcing of computing services.

The development of alternative distributive channels for bank products is

measured by the variable REMOTE, defined as the number of remote customers

standardized by the total number of current accounts, and the variable ATM_BR, given

by the number of automatic teller machines (ATM) divided by the number of branches.

We the ratio of income from services to gross income, SERVICE as a measure of

impact of banking activities. REDUNDANCE represents the fraction of staff made

redundant by banks through major restructuring programs and should proxy for the

absorption of excess capacity.

There is a widespread presumption that the impact of IT on productivity varies

according to business focus. Banks active in wholesale (i.e. securities trading) or in

markets for highly standardized financial products (i.e. consumer credit) rely largely

standardized information flows and a large fraction of their operations is screen-based

or involve automatic data processing. On the other side, bank activities like small

business lending are based on the acquisition and the processing of soft information,

which in turn require the physical proximity of the decision unit to costumers (Berger

et al. (2005)). Size is usually taken as a proxy for bank specialization under this respect:

large banks are taken on the hard information and small banks are assumed to work

mainly with soft information. Although this appears a reasonable classification, it is not

completely appropriated for our sample: in the Italian banking industry several large

banks compete with small banks in supplying finance to small business, and small

21

institutions specialize in standardized products. We therefore try to capture the effects

of business specialization regarding information processing through the variables

SMALL, which stands for the fraction of loans to small and medium-sized firms,

MANAGERS, defined as the ratio of managers costs to total staff costs, and

STAFF_BR, given by the average number of employees per branch.

The variable CAP_ASS, defined as the ratio of capital and reserves to total assets,

controls for the degree of capitalization of the bank, the variable M&A measures the

increase in assets obtained with mergers. We also include the variable AGE, given by

the expression:

foundyearyearAGE = 11

where foundyear is the year of foundation of the bank and year is the current date. In

this way, differences in the age of the banks are more important for relatively younger

banks. The dummies NW (northwestern regions), NE (northeastern regions), SOUTH

(southern regions), ISL (Sicily and Sardinia, the two main islands) account for

geographical differences, while the dummies GROUP1 and GROUP2 have value one if

the bank is part of a large group (the top 25% of the groups total assets distribution) or

of a medium or small group (the last 75%) respectively. Summary statistics of the

variables used in the empirical analysis are reported in table 3.

6. Results

6.1 Frontier estimates and catching-up effects

Table 3 summarizes the descriptive statistics of the variable used in the

regressions and table 4 shows the estimated coefficients of the traslog cost and profit

22

functions. The pattern of signs and the magnitude of the coefficients are remarkably

consistent both with the prediction of the theory and with those of similar studies. The

average inefficiency levels estimated from the cost and the profit frontiers are displayed

in figure 1 and 2, respectively. Cost inefficiency averages at 9.2% and profit

inefficiency at 8.1%, with large differences across size classes. These values are far

lower than those found in the vast majority of other studies both on Italy (Resti (1997))

and on other countries (Berger and Humphrey (1997)). We interpret this result as a

consequence of accounting for IT capital as an input and of nontraditional activities as

an output. This is consistent with the findings of Rogers (1998): Models that ignore

nontraditional outputs are likely to underestimate bank efficiency.

Robustness checks, not reported for the sake of brevity, show that average

inefficiency levels remain comparatively small also if the status of deposits is changed

from output to input and considering different time windows.11

The time pattern of inefficiency levels is the same for all bank classes, displaying

an increasing inefficiency over the decade, although small banks are always more

efficient than the other groups. This can be attributed to the different magnitude of the

shocks and the different speed at which banks responded to them.

The parameters of the inefficiency equations, jointly estimated with the translog

cost and profit functions, are reported in table 4. The effect of IT capital stock on

inefficiency is negative and statistically significant in both equations. This result

confirms that the most IT-capitalized are more likely to be closer to the efficient cost

and profit frontiers. Specifically, for the median bank, an increase of one standard

11 In particular we have estimated the model separately for the periods 1989-1995 and 1996-2000. The

entire set of results is qualitatively unchanged, albeit reducing the variability along the time dimension inflates somewhat the standard errors.

23

deviation in the level of IT capital per employee will generate a 3.7% cut in costs and a

7.5% profit boost due to decreased inefficiency. The average contribution of IT capital

to TFP growth via the catching up effect estimated from the cost function is as large as

0.5% per year. The magnitude of the same contribution estimated through the

alternative profit function is higher and amount to 1.1%. One possible explanation of

the difference between the estimates obtained from the two frontier models is that banks

that have invested more in the new technologies have also improved the quality of the

services they supply and this is reflected on higher prices and short run profits, as

stressed by Berger and Mester (2004). If the production of higher quality services

requires also higher cost, it is then plausible the productivity gains measured in terms of

short-run profit efficiency are higher than those measured in terms of cost efficiency.

This explanation based on an improvement of the quality of services is supported

by the estimated coefficients of other variables in the inefficiency regression.

SERVICE, which represents the fraction of earnings coming nontraditional activities, is

negatively correlated with the degree of efficiency. REMOTE, an indicator of

innovation in the delivery channels of banking services, has a negative and statically

significant effect only on inefficiency measured through the alternative profit function.

On the contrary, banks with a larger number of ATM per branch display greater

inefficiency. The interpretation is that, in the majority of cases, the staff downsizing

required by large-scale automation of some basic activities takes time or it is not fully

adjusted, leading to some excess capacity. The existence of frictions in the adjustment

of labor, which is supported by the estimated negative correlation between inefficiency

and the fraction of staff made redundant, is a consequence of major restructuring plans

(REDUNDANCE).

24

Among the other factors affecting inefficiency, the ratio of capital to total assets

(CAP_ASS) has a sizeable impact, although the causality link is ambiguous. Large

capital ratios may follow from excess capacity or may be due to agency problems as

long as more efficient banks can signal their quality through a high leverage as in

Berger and Bonaccorsi di Patti (2006).

Mergers and acquisitions are negatively correlated with inefficiency. Since we use

pro-forma data, this result simply means that banks which buy other banks have above-

average efficiency scores, a well-established result in the M&A literature (Amel et al.

2002). The dummy variables identifying the partnership in a large (GROUP1) or small

(GROUP2) banking group have estimated coefficients with positive signs in the cost

function and negative signs in the profit regression. Most of these banks have been

acquired by a bank holding company during the sample period. Target banks were

usually poorly performing, and restructuring within the banking group has initially

focused on the revenue side and only later gradually extended to the cost side (Focarelli

et al. (2002)).

The AGE variable is negatively correlated with inefficiency: this may be due to

the time needed by the de novo banks to reach their optimal level of capacity utilization.

Finally, the geographical dummies show that banks with headquarters in the richest

regions of the country (NE and NW) are closer to the efficient frontier than banks

located elsewhere.

6.2 Frontier shift regressions

Tables 5 and 6 report the results of the panel data estimation of cost and profit

frontier shifts. A negative sign for a coefficient in Table 5 implies an inward shift of the

industry best practice cost function, while a positive sign of a coefficient in Table 6

25

implies an outward shift of the industry best practice alternative profit function. The

first three columns report the results of different specifications of the IT variables for

the entire sample. In the last three columns we check for differences owing to bank size.

In columns A, D, E and F, IT capital deepening is measured simply by total IT

capital per employee. The results are consistent with the hypothesis that computers and

related technologies lead to sizeable increases in productivity, here measured by frontier

shifts. The coefficient of the IT capital variable is different from zero at high

significance levels, and the sign is negative for the cost frontier regression and positive

for the profit frontier one. The result holds true across all bank size classes (columns D,

E, and F) and the two the type of frontier.

On average the shifts in best practice due to IT capital accumulation can be

quantified in a reduction of costs of 0.8% per year and in an increase of short run profits

of 0.7%.

In column B we have split IT capital into two broad categories: hardware and

software. The coefficient of the HARD_CAP variable is negative in the cost function

and positive in the profit function and statistically significant, indicating a favorable

effect of computers on productivity. The coefficients of SOFT_CAP are estimated less

precisely and have an opposite pattern of signs. A possible explanation is that software

capital stock is poorly measured because hedonic prices do not account properly for

quality improvements (Jorgenson and Stiroh (2000)).

In columns C we introduce a variable aiming to capture the impact of IT on labor

organization: the fraction of salary expenses devoted to IT staff. This variable is

available only for a sub-sample of banks, about half of the total number. The sign of its

the coefficient is positive in the cost frontier shift regression and negative in the profit

frontier one. Within-firm production of IT services is detrimental to TFP growth. This

26

finding is confirmed by the productivity enhancing effects of the variable

OUTSOURCE, which measures the amount of IT-related expenses for outsourced

activities. The result is consistent with theories of the firm boundaries, which predict

that firms allocate to external suppliers activities for which they do not have

comparative advantages (Heshmati (2003)).

Among the other regressors, SERVICE has the same economic effect as in the

inefficiency regressions: shifting from traditional to nontraditional banking activities

shift also the best practice frontier. The same holds, at least partially, for the dummies

that identify membership of a banking (GROUP1 and GROUP2). On the other hand,

the influence on TFP growth of other variables, like CAP_ASS, ATM_BR, and

REDUNDANCE is of opposite sign when measured by frontier shifts and by changes in

the level of industry inefficiency. This can be interpreted in terms of equilibrium and

out of equilibrium effects. For instance, introducing network of cash dispensers (ATMs)

boosts productivity only after that the activities branch staff have been reorganized and

the idle capacity has been cut. Before a new optimal allocation of resources within the

bank it is likely to observe a deterioration of performance. This pattern is fully

consistent with our estimates.

The number of staff per branch (STAFF_BR), the share of managers salaries in

total salaries (MANAGERS) and that of small business loans in total loans (SMALL)

are negatively correlated with TFP growth. All these three variables capture the

information specialization of banks and show that the route to innovation has been less

straightforward for those institutions which rely more on soft information..

7. Conclusions

27

Several studies have emphasized the role of technological innovation as a primary

source of structural changes within the financial industry. However, empirical evidence

on the effects of technical progress in increasing TFP and abating unit costs is still

scarce. This paper contributes in building up evidence by investigating the effect of

information technologies on the Italian banking industry over the period 1989-2000. We

have estimated stochastic cost and profit functions allowing for individual banks

displacements from the efficient frontier and non-neutral technological change. Data on

IT capital stock for individual banks enabled us to distinguish between movements

along the We found that both cost and profit frontier shifts are strongly correlated with

IT capital accumulation. Banks adopting IT capital intensive techniques are also closer

in average to the best practice of the banking industry, implying ceteris paribus a higher

level of efficiency. We interpret this last result as evidence of a catching-up effect

consistent with the usual pattern of diffusion of the new technologies.

Compounding the estimates from the inefficiency regressions with those coming

to the analysis of frontiers, the TFP growth associated with IT investment is of 1.3 % if

estimated from the cost function and 1.8 % if estimated from the profit function. These

figures appear quite large if compared with the estimated TFP growth for the banking

industry of other studies. But comparisons should be made with homogenous figures

restricted to IT contribution, referring to the financial sector as well to other industries,

which are not available. Nonetheless, our results fall in line with those obtained by

other studies based on micro-data: they support the hypothesis that computers are a

general purpose technology which contribute to output growth not only through the

process of capital deepening, but also because of the changes it induces in workplace

organization reflected in higher TFP growth. Skeptical views about the existence and

28

the magnitude of productivity gains brought by digital technologies in the financial

sector are unwarranted.

29

Appendix: Stochastic frontiers and inefficiency

Considering a generic cost (profit) frontier function, we estimate the following

system:

Fit itititit vutPYf ++= ][ ,, (A1)

ititit wzu += (A2)

where Fit stand for the costs (profits) of bank i in period t, Yit is an n-dimensional

vector of output quantities (prices), and Pit is the m-dimensional vector of the m input

prices. The residual of equation (A1) is a composite term formed by a standard white

noise residual vit plus an asymmetric distribution term uit, accounting for inefficiency. In

equation (A2) the uit term is specified as a stochastic process with mean dependent from

a vector of covariates zit, and a random component wit defined as the truncation of an

independent normal distribution with mean zero and variance 2U

, such that uit is a

non-negative truncation of the normal distribution N(zit, 2U) 12.

This general specification partially mitigates the usual problems the stochastic

frontier methodology related with the ad hoc assumptions imposed to the inefficiency

distribution (Battese and Coelli (1995)).13 The model allows also inefficiency levels to

vary over time without imposing the same ordering of firms in terms of efficiency for

12 The usual procedure in the past literature was first to estimate the stochastic frontier and predict

inefficiency under the assumption of identical distributions, then to estimate an inefficiency equation using the predicted uit term, in contradiction with the prior distributional assumption (see Khumbhakar, Ghosh and McGukin,(1991)).

13 The assumption that the uit are independent distributed i and t, made to simplify the model, is a

restriction because it does not account for heteroskedasticity or for possible correlation structures among the residuals. In our estimates, however, we control for robustness to different inefficiency estimation methods.

30

each period, as in Battese and Coelli (1993).14 In our setting a bank i can be relatively

more efficient than bank j in year t and become less efficient in year t+1.

The ui term can be identified, following Jondrow et al. (1982), conditional on the

estimated residual ei = vi + ui. Battese and Coelli (1988) prove that the best predictor of

cost inefficiency for firm i is the expression eui, for which the following expression

holds (Battese and Coelli (1993)):

[ ] [ ][ ]

]2/2[

)1(

)1(| aiee

ez

ez

eu

eEiit

a

aiita

ii

+

=

(3)

where a is equal to the square root of the expression (1-)(2U +2V), is the density

function of a standard normal random variable = 2U/(2U+2V), 2U and 2V being

respectively the estimated variance of the inefficiency term uit and of the white noise vit.

The parameter estimated in the regression represents the fraction of total

variance around the frontier due to inefficiency. A value of equal to zero means that

the distance to the frontier is entirely due to noise, while a value equal to one indicates

that the entire deviation is due to inefficiency. The expectation of inefficiency uit

conditional to the whole residual eit measures the ratio between the costs (profits) of a

generic firm i and those of the most efficient firm i*, which is on the frontier with an

inefficiency level equal to zero (eui* = 1).

14 The relation zit + wit > zit + wit for i i does not imply zit + wit > zit + wit for t t .

31

References

D. Amel C. Barnes F. Panetta C. Salleo (2004) Consolidation and Efficiency in the Financial Sector: A Review of the International Evidence, Journal of Banking and Finance, 28: 24932519.

D. H. Autor - F. Levy - R.J. Murnane (2002), Upstairs, Downstairs: Computers and Skills on Two Floors of a Large Bank, Industrial Relations Review, 55: 432-447.

G.E. Battese - T.J. Coelli (1988), Predictions of Firm Level Technical Efficiencies with a Generalised Frontier Production Function and Panel Data, Journal of Econometrics, 38: 387-400.

G.E. Battese - T.J. Coelli (1993), A Stochastic Frontier Production Function Incorporating a Model for Technical Inefficiency Effects, Working Papers in Econometrics and Applied Statistics, 69, Department of Econometrics, University of New England, Armindale.

G.E. Battese - T.J. Coelli (1995), A Model for Technical Inefficiency Effects in a Stochastic Frontier Production Function with Panel Data, Empirical Economics, 20: 325-332.

P.W. Bauer - A.N. Berger - D.B. Humphrey (1993), Efficiency and Productivity Growth in US Banking in Fried, H.O. et al. (ed.) The Measurement of Productive Efficiency: Techniques and Applications, Oxford University Press: Oxford.

S.A. Berg - F. Finn - E.S. Jansen (1992), Malmquist Indices of Productivity Growth During the Deregulation of Norwegian Banking 1980-89, Scandinavian Journal of Economics, 94 Supplement: 211-228.

A.N. Berger (2003), The Economic Effect of Technological Progress: Evidence from the Banking Industry, Journal of Money, Credit and Banking, 35: 141-176.

A.N. Berger - E. Bonaccorsi di Patti (2006), Capital Structure and Firm Performance: A New Approach to Testing Agency Theory and an Application to the Banking Industry, Journal of Banking & Finance, 30, Issue 4, April 2006: 1065-1102.

A.N. Berger - D.B. Humphrey (1991), The Dominance of Inefficiency over Scale and Product Mix Economies in Banking, Journal of Monetary Economics, 28: 117-148.

32

A.N. Berger - D. B. Humphrey (1997) Efficiency of financial institutions: International survey and directions for future research, European Journal of Operational Research 98:175212.

A.N. Berger - L. Mester (1997), Beyond the Black Box: What Explains Differences in the Efficiency of Financial Institutions?, Journal of Banking and Finance, 21: 895-947.

A.N. Berger - L. Mester (2003), Explaining the Dramatic Changes in the Performances of US Banks: Technological Change, Deregulation and Dynamic Changes in Competition, Journal of Financial Intermediation 12: 57- 95

A.N. Berger - N. H. Miller - M. A. Petersen - R. G. Rajan J. C. Stein (2005), Does function follow organizational form? Evidence from the lending practices of large and smallbanks, Journal of Financial Economics, 76, 237-269.

E.R. Berndt - C.J. Morrison (1997), Assessing the Productivity of Information Technology Equipment in US Manufacturing Industries, Review of Economics and Statistics 79: 110-125.

E.R. Boskin E. R. Dulberger R. J. Gordon Z. Griliches - D. W. Jorgenson (1997), The CPI Commission: Finding and Recommendations, American Economic Review, 87 (Paper and Proceedings): 78-83.

T. Bresnaham - E. Brynjolfsson - L. Hitt (2002), Information Technology, Workplace Organization and the Demand for Skilled Labor: Firm Level Evidence, Quarterly Journal of Economics, 17: 339-376.

E. Brynjolfsson - L.M. Hitt (2003), Computing Productivity: Firm-Level Evidence, Review of Economics and Statistics, 85: 793-808.

J.A. Clark - T.F. Siemes (2002), X-efficiency in Banking: Looking Beyond the Balance Sheet, Journal of Money, Credit and Banking, 34: 987-1013.

D.J. Fixler (1993), Measuring Financial Service Output and Prices in Commercial Banking, Applied Economics, 25: 983-993.

D. Focarelli - F. Panetta - C. Salleo (2002), Why Do Banks Merge?, Journal of Money, Credit and Banking, 34: 1047-1066.

33

P.A. Geroski (2000), Models of Technology Diffusion, Research Policy, 29: 603-625.

G. Gobbi (1996) Efficiency and Productivity Growth of Italian Banks over the Liberalisation Period, Research Monographs in Banking and Finance, Institute of European Finance, Bangor University.

E. Grifell-Tatje - C.A.K. Lovell (1996), Deregulation and Productivity Decline: The Case of Spanish Savings Banks, European Economic Review, 40: 1281-1303.

R.J. Gordon (1999), Has the New Economy Rendered the Productivity Slowdown Obsolete?, Working Paper, Northwestern University.

D. Hancock (1991), A Theory of Production for the Financial Firm, Boston, Kluwer Academy Publisher.

A. Heshmati (2003), Productivity Growth, Efficiency, and Outsourcing in Manufacturing and Service Industries, Journal of Economic Surveys, 17: 79-112.

W.C. Hunter - S.G. Timme (1986), Technical Change, Organizational Form, and the Structure of Bank Production, Journal of Money, Credit and Banking, 18: 151-166.

D.B. Humphrey (1992), Flow Versus Stock Indicators of Banking Output: Effects on Productivity and Scale Economies Measurement, Journal of Financial Service Research, 6: 117-130.

D.B. Humphrey - L.B. Pulley (1997), Banks Responses to Deregulation: Profits, Technology and Efficiency, Journal of Money, Credit and Banking, 29: 73-93.

J. Jondrow - C.A.K. Lovell - I.S. Materov - P. Schmidt (1982), On the Estimation of the Technical Inefficiency in the Stochastic Frontier Production Model, Journal of Econometrics, 19: 233-238.

D.W. Jorgenson (2001), Information Technology and the US Economy, American Economic Review, 91: 1-32.

D.W. Jorgenson - K.J. Stiroh (2000), Raising the Speed Limit: US Economic Growth in the Information Age, Brooking Papers on Economic Activity, 1: 125-211.

34

S.C. Khumbhakar - S. Ghosh - J.T. McGukin (1991), A Generalized Production Function Approach for Estimating Determinants of Inefficiency in US Dairy Firm, Journal of Business and Economics Statistics, 9: 279-286.

S. C. Kumbhakar - A. Lozano-Vivas - C.A.K. Lovell - I. Hasan (2001), The Effects of Deregulation on the Performance of Financial Institutions: The Case of Spanish Savings Banks, Journal of Money, Credit and Banking, 33: 101-120.

OECD (2004), The Economic Impact of ICT Measurement, Evidence and Implications, OECD, Paris.

N. G. Olazabal (2002), Banking: The IT Paradox, McKinsey Quarterly, n.1.

S. D. Oliner - D.E. Sichel (2000), The Resurgence of Growth in the Late 1990s: Is Information Technology the Story?, Journal of Economic Perspectives, 14: 3-22.

D. Parsons - C.C. Gotlieb - M. Denny (1993), Productivity and Computers in Canadian Banking, Journal of Productivity Analysis, 4: 95-114.

D. Pilat (2004), The ICT Productivity paradox: Insights from Micro Data, OECD Economic Studies, 38: 37-64.

A. Resti (1997) Evaluating the cost-efficiency of the Italian banking system: What can be learned from the joint application of parametric and non-parametric techniques, Journal of Banking and Finance 21: 221250.

S.S. Roach, (1987), Americas Technology Dilemma: A Profile of the Information Technology, Morgan Stanley Special Economic Study.

K. E. Rogers (1998), activities and the efficiency of US commercial banks, Journal of Banking and Finance 22: 467482.

W.E.G. Salter (1966), Productivity and Technical Change, 2nd edition, Cambridge, Cambridge University Press.

E.P. Seskin (1999), Improved Estimates of the National Income and Product Accounts for 1959-98: Results of the Comprehensive Revision, Survey of Current Business, BEA.

K.J. Stiroh (2000), How Did Bank Holding Companies Prosper in the 1990s, Journal of Banking and Finance, 24: 1703-1745.

35

L.H. Summers (2000), International Financial Crises: Causes, Prevention, and Cures, American Economic Review, 90 (Paper and Proceedings): pp. 1-16.

R. Vander Vennet (2002), Cost and Profit Efficiency of Financial Conglomerates and Universal Banks in Europe, Journal of Money, Credit and Banking, 34: 254-282.

D. C. Wheelock P. W. Wilson (1999), Technical Progress, Inefficiency, and Productivity Change in U.S. Banking, 1984-1993, Journal of Money, Credit and Banking, 34: 212-234.

36

Table 1

ITALIAN BANKS OUTALY RELATED TO IT

Year IT expenses as a percentage of Composition of IT expenses (percentages)

Gross income

Operating costs Hardware Software Staff Outsourcing Others Total

1989 5.6 9.0 26.1 9.9 32.2 13.9 17.9 100.0

1990 5.8 9.4 27.2 10.8 30.3 14.4 17.3 100.0

1991 6.1 9.5 27.1 10.7 29.8 14.8 17.7 100.0

1992 6.0 9.2 24.4 12.4 28.9 14.9 19.5 100.0

1993 5.5 9.1 23.4 13.2 26.5 16.4 20.5 100.0

1994 6.6 9.8 22.0 13.4 26.8 17.7 20.0 100.0

1995 6.7 9.9 19.2 13.6 25.1 22.3 19.8 100.0

1996 6.4 9.6 16.6 15.2 21.7 25.3 21.2 100.0

1997 6.4 9.4 16.9 16.7 21.2 25.6 19.7 100.0

1998 7.4 12.2 21.3 20.7 16.5 17.8 23.7 100.0

1999 7.6 12.8 19.7 19.7 15.0 26.6 19.0 100.0

2000 7.6 13.5 14.9 18.3 12.4 38.7 15.8 100.0

2001 8.0 14.5 15.6 20.8 10.4 34.1 19.2 100.0 Sources: Bank Supervisory Reports.

37

Table 2

ITALIAN BANKS IT INVESTMENT AND CAPITAL STOCK* (1989=100)

Nominal IT investments Real IT investments Real IT Capital

Year Hardware Software Hardware Software Total Per

employee

1989 100 100 100 100 100 100 1990 101 126 135 154 129 125

1991 114 149 154 166 155 148

1992 112 179 158 188 176 167

1993 98 191 133 164 179 168

1994 101 222 164 205 193 182

1995 96 295 185 270 215 202

1996 97 268 258 264 254 243

1997 99 337 302 302 300 292

1998 123 430 523 405 421 414

1999 91 370 430 300 458 455

2000 105 393 535 289 522 514

2001 107 453 656 312 612 599

Sources: Bank Supervisory Reports, ISTAT and BEA.

38

Table 3

DESCRIPTIVE STATISTICS

Symbol Description Obs. Mean Min Max St.dev

COST Log of total cost (divided by the price of labor) 5231 5.077 1.676 10.696 2.006 PROF Log of gross profit (divided by the price of labor) 5231 8.2889 2.480 10.279 .316 Y1 Log of short-term loans 5231 4.341 -5.276 10.902 1.932 Y2 Log of long-term loans 5231 4.000 -1.440 10.952 1.744 Y3 Log of services weighted commission income. 5231 5.167 -2.624 10.786 1.611 Y4 Log of deposits 5231 -0.181 -5.291 7.151 2.107 P1 Log of IT capital price (divided by the labor price) 5231 2.069 0.253 3.744 0.620 P2 Log of branches price (divided by the labor price) 5231 -0.137 -1.266 1.833 0.530 CAP_ASS Capital divided by total assets 5231 0.093 0.017 0.396 0.034 AGE agebank 11 5231 0.842 0.000 0.875 0.082 SERVICE Net services value revenue divided by gross income 5231 0.060 -0.983 0.556 0.060 SMALL Value of small loans (under 5 billion lire) over total loans 5231 0.675 0.023 1.000 0.197 GROUP1 Part of a big group (top 25 % of the groups total asset

distribution) 5231 0.059 0.000 1.000 0.236

GROUP2 Part of a medium or small group (last 75 % of the groups total asset distribution) 5231 0.088 0.000 1.000 0.283

MERACQ Relative increase in total assets of the bank due to M&A 5231 0.011 0.000 1.536 0.066 STAFF_BR Number of employees divided by number of branches 5231 11.200 2.000 242.00 10.057 MANAGERS Wages paid to managers divided by total wages. 5213 36.189 0.520 331.46 26.645 REDUNDANCE Number of redundant employees divided by total staff 5231 0.001 0.000 0.519 0.012 IT_CAP Stock of real IT capital per employee (billion lire) 5231 0.056 0.001 0.482 0.044 HARD_CAP Stock of real hardware capital per employee (billion lire) 5231 0.051 0.000 0.467 0.042 SOFT_CAP Stock of real software capital per employee (billion lire) 5231 0.004 0.000 0.060 0.004 OUTSOURCE Share of IT expenses due to outsourcing of services 5100 .3689 0.000 1.000 .2996 IT_STAFF Wages paid to IT personnel divided by total wages 3031 0.055 0.000 1.000 0.070 REMOTE Number of phone and electronic banking accounts of

households divided by total number of current accounts. 5231 0.018 0.000 1.000 0.091

ATM_BR Number of ATM divided by number of branches. 5231 0.526 0.000 20.00 0.750 NW Headquarters in the North-West 5231 0.201 0.000 1.000 0.401 NE Headquarters in the North-Est 5231 0.408 0.000 1.000 0.492 CE Headquarters in the Centre 5231 0.211 0.000 1.000 0.408 SOUTH Headquarters in the South or islands 5231 0.180 0.000 1.000 0.384

Sources: Bank Supervisory Reports. All financial variables are measured in ITL million and are adjusted for inflation. Statistics refer to the panel of 618 banks for the period 1989-2000 with the exception of internet banking accounts, available from 1995.

39

Table 4

COST AND PROFIT FRONTIERS The dependent variables are the total operating expenses and the total revenue before taxes, both normalized by the price of labor. The independent output variables are the amount of short term loans (Y1) medium and long term loans (Y2), a composed measure of bank services (Y3) and the amount of deposits (Y4). The input variables are the estimated price of IT (P1) and of the other physical capital (P2), both normalized by the price of labor. The parameters 1 and the 2 refer to growth rate of Italian GNP and to the Italian real interest rate level. All variables, except the interest rate, are in log and are computed as differences from the sample mean.

Variables COST FRONTIER PROFIT FRONTIER Variables COST FRONTIER PROFIT FRONTIER

CONST -0.8337 *** 0.7211 *** Y1P2 -0.0843 *** -0.0175 *

0.0648

0.0420

0.0147

0.0098 Y1 0.3547 *** 0.0182 * Y1T -0.0243 *** 0.0006

0.0211

0.0109

0.0028

0.0018 Y2 0.1574 *** 0.0286 *** Y2Y3 -0.0304 *** 0.0118 ***

0.0127

0.0086

0.0044

0.0027 Y3 0.0780 *** 0.0448 *** Y2Y4 -0.0334 *** 0.0022

0.0167

0.0112

0.0045

0.0027 Y4 0.3765 *** -0.0965 *** Y2P1 0.0000 -0.0090 *

0.0226

0.0149

0.0075

0.0050 P1 0.0550 *** 0.0027 Y2P2 0.0599 *** 0.0538 ***

0.0155

0.0104

0.0102

0.0071 P2 0.1069 *** 0.1227 *** Y2T 0.0032 * -0.0008

0.0190

0.0128

0.0018

0.0012 T1 0.1462 *** -0.1473 *** Y3Y4 0.0172 ** 0.0135 **

0.0093

0.0060

0.0081

0.0054 Y12 0.0039

0.0079 *** Y3P1 -0.0240 *** -0.0101 * 0.0030

0.0021

0.0082

0.0055 Y22 0.0323 *** 0.0059 *** Y3P2 0.0637 *** -0.0071

0.0019

0.0013

0.0107

0.0072 Y32 -0.0050

0.0019

Y3T 0.0154 *** -0.0023 0.0042

0.0028

0.0023

0.0015 Y42 0.0379 *** 0.0033 Y4P1 0.0082 -0.0148 **

0.0035

0.0023

0.0098

0.0068 P12 0.0073

0.0014

Y4P2 -0.0365 ** -0.0093 0.0053

0.0034

0.0170

0.0116 P22 -0.0228 *** -0.0443 *** Y4P2 0.0057 ** 0.0108 ***

0.0083

0.0056

0.0029

0.0019 PT2 -0.0097 *** 0.0069 *** P1P2 -0.0104 0.0051

0.0005

0.0003

0.0099

0.0066 Y1Y2 -0.0303 *** -0.0003 P1T -0.0046 ** -0.0001

0.0036

0.0025

0.0021

0.0014 Y1Y3 0.0231 *** -0.0222 *** P2T -0.0129 *** -0.0059 ***

0.0078

0.0049

0.0026

0.0017 Y1Y4 -0.0144 * 0.0155 *** 1 0.0213 *** -0.0092 ***

0.0076

0.0046

0.0029

0.0020 Y1P1 0.0171 * 0.0189 ** 2 0.0348 *** -0.0159 ***

0.0112

0.0075

0.0039

0.0026

Observations = 5231

Cross-sections

= 618

Time periods

= 12

Note: Statistically different from zero, respectively, at: *** 99%, **95% and *90% significance level.

40

Table 5

COST AND PROFIT INEFFICIENCY CORRELATES

The dependent variables are respectively the cost and the profit inefficiencies jointly estimated from the translog specification of Table 4. A complete description of the independent variables is given in Table 3.

Note: Statistically different from zero, respectively, at: *** 99%, **95% and *90% significance level.

Variables COST INEFFICIENCY PROFIT INEFFICIENCY

CONSTANT 0.2126 ***

0.8587 ***

0.0568

0.0651

TREND 0.0117 ***

0.0352 ***

0.0038

0.0068

IT_CAP -0.9100 ***

-1.7889 ***

0.1939

0.3683

SERVICE -1.5195 ***

-1.2455 ***

0.2018

0.0482

REMOTE -0.0294

-0.0592 *

0.0551

0.0347

ATM_BR 0.0192 ***

0.0417 ***

0.0078

0.0077

REDUNDANCE -0.0148

-0.8690 ***

0.2841

0.2184

CAP_ASS 0.0139

3.9501 ***

0.1510

0.4436

MERACQ -0.0293

-0.2367 ***

0.0721

0.0422

GROUP1 0.0661 ***

-0.7449 ***

0.0210

0.0961

GROUP2 0.1665 ***

-0.6888 ***

0.0288

0.0877

AGE -0.0095

-0.6347 ***

0.0569

0.0266

NW -0.3572 ***

-0.3838 ***

0.0609

0.0212

NE -0.4983 ***

-0.0438 *

0.0943

0.0228

CE -0.2293 ***

-0.0156

0.0384

0.0266

2 0.0521 *** 0.0669 *** 0.0039

0.0086 0.4419

*** 0.8320

*** 0.0511

0.0282

Observations

= 5231 Cross-sections

= 618 Time periods

= 12

41

Table 6

COST FUNCTION SHIFTS

Variables (A)

FULL SAMPLE

(B) FULL

SAMPLE

(C) FULL

SAMPLE

(D) LARGE BANKS

(E) MEDIUM-

SIZED BANKS

(F) SMALL BANKS

CONSTANT 0.1296*** 0.1297*** 0.0826 *** 0.0399 0.0746*** 0.2700 ***

0.0119

0.0119

0.0129

0.0281 0.0163 0.0211

IT_CAP -0.1597*** -0.1412 *** -0.1022** -0.0747*** -0.1953 ***

0.0131

0.0198

0.0501 0.0216 0.0163

HARDW_CAP -0.1753***

0.0142

SOFTW_CAP 0.2759*

0.1486

IT_STAFF 0.0851 ***

0.0173

OUTSOURCING -0.0109 -0.0098*** -0.0220 *** -0.0345*** -0.0252*** -0.0018

0.0024 0.0024

0.0041

0.0105 0.0042 0.0029

ATM_BR -0.0362*** -0.0364*** -0.0368 *** -0.0355*** -0.0385*** -0.0326 ***

0.0009

0.0009

0.0012

0.0030 0.0016 0.0011

REMOTE -0.0007 0.0002 -0.0014 0.0092 -0.0172*** 0.0265 ***

0.0058

0.0058

0.0080

0.0332 0.0062 0.0099

CAP_ASS -0.7754*** -0.7739*** -0.7348 *** -0.2278** -0.6437*** -0.7691 ***

0.0250

0.0250

0.0300

0.0973 0.0440 0.0318

SMALL 0.0761*** 0.0744*** 0.0949 *** 0.0789*** 0.0672*** 0.0706 ***

0.0063

0.0063

0.0089

0.0259 0.0124 0.0074

SERVICE -0.2267*** -0.2229*** -0.1980 *** -0.1207*** -0.2288*** -0.2819 ***

0.0117

0.0118

0.0149

0.0303 0.0176 0.0185

STAFF_DEN 0.0023*** 0.0023*** 0.0024 *** 0.0022*** 0.0028*** 0.0025 ***

0.0001

0.0001

0.0002

0.0002 0.0003 0.0002

REDUNDANCE 0.0838* 0.0777* 0.0592 -0.0934 -0.1494 0.1383 **

0.0446

0.0446

0.0619

0.1041 0.1109 0.0552

MANAGERS 0.0002*** 0.0002*** 0.0001 *** 0.0000 0.0000 0.0004 ***

0.0000

0.0000

0.0000

0.0001 0.0000 0.0000

MERACQ 0.0133** 0.0134** 0.0159 * -0.0289 0.0096 0.0216 ***

0.0066

0.0066

0.0084

0.0429 0.0129 0.0074

GROUP2 -0.0231*** -0.0234*** -0.0251 *** -0.0324*** -0.0241*** -0.0263

0.0026

0.0026

0.0027

0.0114 0.0025 0.0244

GROUP1 -0.0162*** -0.0167*** -0.0174 *** -0.0151*** -0.0205*** - -

0.0026

0.0026

0.0027

0.0053 0.0029

AGE -0.0957*** -0.0966*** -0.0607 *** -0.0363 -0.0140 -0.2764 *** 0.0125 0.0125 0.0124

0.0272 0.0156 0.0239

Observations 5082 5082 2921 435 1318 3329 Banks 612 612 396 49 146 417 R-squared (within) 0.756 0.757 0.787 0.698 0.836 0.760 R-squared (between) 0.001 0.001 0.014 0.049 0.329 0.016 R-squared (overall) 0.440 0.439 0.451 0.405 0.688 0.375 The dependent variable is the time derivative of the cost function estimated in Table 4. A complete description of the independent variables is given in Table 3. The full sample regression refers to the sample employed for estimations in Tables 4 and 5, reduced by the missing values in the OUTSOURCING and the IT_STAFF variables. The Large banks sample refers to the top 10 % of banks in total asset distribution, the Medium-sized banks sample include banks from the 33 to the 10 % and the Small banks the last 66 % of banks. For the estimated coefficients, a positive sign implies an increasing effect on costs and a correspondent decreasing effect on productivity.

Note: Statistically different from zero, respectively, at: *** 99%, **95% and *90% significance level.

42

Table 7

PROFIT FUNCTION SHIFTS

Variables

(A) FULL

SAMPLE

(B) FULL

SAMPLE

(C) FULL

SAMPLE

(D) LARGE BANKS

(E) MEDIUM-

SIZED BANKS

(F) SMALL BANKS

CONSTANT -0.1376*** -0.1379*** -0.0940 *** -0.0507** -0.0825*** -0.2575*** 0.0086 0.0086 0.0092 0.0203 0.0119 0.0150 IT_CAP 0.1351*** 0.1362 *** 0.1271*** 0.0699*** 0.1545*** 0.0094 0.0141 0.0363 0.0158 0.0116 HARDW_CAP 0.1419***

0.0102 SOFTW_CAP -0.0706

0.1071 IT_STAFF -0.0705 ***

0.0124 OUTSOURCING 0.0124*** 0.0118*** 0.0205 *** 0.0258*** 0.0222*** 0.0060*** 0.0017 0.0017 0.0029 0.0076 0.0031 0.0021 ATM_BR 0.0258*** 0.0259*** 0.0262 *** 0.0278*** 0.0287*** 0.0221*** 0.0006 0.0006 0.0008 0.0022 0.0011 0.0008 REMOTE 0.0001 -0.0003 0.0067 0.0255 0.0134*** -0.0238*** 0.0042 0.0042 0.0057 0.0240 0.0046 0.0070 CAP_ASS 0.5722*** 0.5721*** 0.5461 *** 0.2003*** 0.4855*** 0.5625*** 0.0180 0.0180 0.0214 0.0705 0.0322 0.0225 SMALL -0.0518*** -0.0509*** -0.0634 *** -0.0480** -0.0440*** -0.0473*** 0.0045 0.0045 0.0064 0.0188 0.0091 0.0052 SERVICE 0.2333*** 0.2317*** 0.2019 *** 0.1234*** 0.2293*** 0.2899*** 0.0084 0.0085 0.0107 0.0220 0.0129 0.0131 STAFF_DEN -0.0020*** -0.0020*** -0.0021 *** -0.0018*** -0.0022*** -0.0023*** 0.0001 0.0001 0.0001 0.0002 0.0002 0.0002 REDUNDANCE -0.0579* -0.0549* -0.0222 0.0888 0.1299* -0.1162*** 0.0321 0.0321 0.0442 0.0754 0.0812 0.0391 MANAGERS -0.0003*** -0.0003*** -0.0002 *** -0.0002*** -0.0001*** -0.0005*** 0.0000 0.0000 0.0000 0.0001 0.0000 0.0000 MERACQ -0.0026 -0.0026 -0.0018 0.0264 -0.0025 -0.0081 0.0048 0.0048 0.0060 0.0311 0.0094 0.0052 GROUP2 0.0190*** 0.0192*** 0.0195 *** 0.0259*** 0.0194*** 0.0409***

0.0019

0.0019

0.0019 0.0083 0.0018 0.0173 GROUP1 0.0152*** 0.0155*** 0.0160 *** 0.0157*** 0.0178*** - -

0.0019

0.0019

0.0019 0.0038 0.0021 AGE 0.0694*** 0.0700*** 0.0393 *** 0.0197 0.0074 0.2134***

0.0090

0.0090

0.0089 0.0197 0.0114 0.0169

Observations 5082 5082 2921 435 1318 3329Banks 612 612 396 49 146 417R-squared (within) 0.801 0.801 0.833 0.785 0.871 0.800R-squared (between) 0.046 0.044 0.152 0.067 0.307 0.033R-squared (overall) 0.493 0.490 0.551 0.484 0.706 0.433The dependent variable is the time derivative of the profit function estimated in Table 4. A complete description of the independent variables is given in Table 3. The full sample regression refers to the sample employed for estimations in Tables 4 and 5, reduced by the missing values in the OUTSOURCING and the IT_STAFF variables. The Large banks sample refers to the top 10 % of banks in total asset distribution, the Medium-sized banks sample include banks from the 33 to the 10 % and the Small banks the last 66 % of banks. For the estimated coefficients, a positive sign implies an increasing effect on profits and a correspondent increasing effect on productivity.

Note: Statistically different from zero, respectively, at: *** 99%, **95% and *90% significance level.

43

Figure 1

INEFFICIENCY ESTIMATES FOR THE COST FUNCTION MODEL

Figure 2

INEFFICIENCY ESTIMATES FOR THE PROFIT FUNCTION MODEL

1

1 .0 2

1 .0 4

1 .0 6

1 .0 8

1 .1

1 .1 2

1 .1 4

1 .1 6

1 .1 8

1 9 8 9 1 9 9 0 1 9 9 1 1 9 9 2 1 9 9 3 1 9 9 4 1 9 9 5 1 9 9 6 1 9 9 7 1 9 9 8 1 9 9 9 2 0 0 0

A l l s a m p le la rg e b a n k s m e d iu m b a n k s s m a ll b a n k s

0 .9 5

1

1 .0 5

1 .1

1 .1 5

1 .2

1 .2 5

1 9 8 9 1 9 9 0 1 9 9 1 1 9 9 2 1 9 9 3 1 9 9 4 1 9 9 5 1 9 9 6 1 9 9 7 1 9 9 8 1 9 9 9 2 0 0 0

a ll s a m p le la rg e b a n k s m e d iu m b a n k s s m a ll b a n k s