Eficiência Produtiva, Alterações Tecnológicas E Produtividade No Sector Bancário Português

Productive Efficiency, Technological Change and Productivity in Portuguese Banking.

Victor Mendes

Assistant Professor

CEMPRE-Faculdade de Economia da Universidade do Porto

R. Dr. Roberto Frias

4200 Porto - Portugal

João Rebelo*

Assistant Professor

Universidade de Trás-os-Montes e Alto Douro

Av. Almada Lucena

5000 Vila Real - Portugal

email:

* Corresponding author.

Abstract

In this paper we aim at studyingefficiency, productivity and technological change in Portuguese banking during 1990-95, using information on the vast majority of banks operating in Portugal during that period. We use a translog variable cost function and a stochastic frontier model to estimate inefficiency and technological change. Our results suggest that the increased competition that Portuguese banks witnessed over the last few years did not lead to a better overall performance from the standpoint of costs: on the one hand, the annual efficiency average did not clearly increase over time; on the other hand, many more banks are now less efficient (in relative terms) than they were in the early 90s. They also suggest that there is not a clear relationship between size and cost efficiency. Efficiency and scale economies also seem not to be related with size: some of the less efficient institutions (net assets bellow 50 million contos) are the ones facing global, although small, economies of scale and the largest institutions are the more efficient but face diseconomies of scale, therefore suggesting that they remain competitive via a better cost control. As for technological progress, our results suggest the existence of technological recess, along the 6 years of the sample.

Productive Efficiency, Technological Change and
Productivity in Portuguese Banking

1. Introduction

The Portuguese financial system experienced a dramatic structural change in the last twenty years. Following the “carnation revolution” (April 1974) all Portuguese private banks were nationalized in 1975 (three private foreign banks escaped the wave). In the subsequent years new entries in the banking industry were barred. The management of banks and the banking activity were heavily regulated.

In 1985 and as a result of the integration in the EEC, a period of deep economic liberalization started to unfold, along with the reopening of the banking industry to the private initiative. In 1984, only 17 banks (commercial, investment and savings) existed, that number increasing to 30 in 1990 and 45 in 1995. The number of domestic branches also climbed from 1481 in 1985 to 1990 and 3729, respectively, in 1990 and 1995[1].

As a result of the increasing number of institutions and branches, the concentration indexes (C4 and C8, net assets - see table 1) decreased, although slightly[2]. At the same time, the increased competition in both the deposit and the credit markets, along with diminishing inflation rates, caused a sharp decline in the interest margin (either in constant prices or as a percentage of total net assets). The privatization process was initiated in 1989. Meanwhile, the financial marks, with no activity after the revolution, were revamped and quite successfully restarted their operations in a new regulatory and operational framework. This, along with the liberalization of products and services, “produced” the need to divert bank production to new lines of business; therefore, the interest margin in percentage of the banking product sharply declined, representing less than 75% in 1995. On the other hand, it has become more and more difficult to do business the “old fashioned” way, that is, loaning out money to businesses and individuals: the share of interbank lending in the money market represents 45.06% of the total credit granted in 1995, and the share of credit to individuals and firms dropped from 71.72% in 1990 to 54.94% in 1995[3]. But a similar movement occurred on the liability side, with deposits from banks increasing steadily (from 10.88% in 1990 to 32.68% of total deposits in 1995).

On the cost side, the cost of borrowed funds still represents around 80% of variable costs. The price of labor has steadily increased in the six years of our sample, but labor only represents around 12% of variable costs. The decreasing number of employees by bank only means that the new entrants are basically small institutions. However, operating costs (staff expenses and other administration costs) increased less that net assets, therefore allowing for a drop on operating cost as a percentage of assets. But the difference between the interest margin and operating costs was dramatically reduced between 1990 (2.62% of net assets) and 1995 (0.6%). Thus, the sharp drop on ROA and ROE does not appear to be such a big surprise. These figures call for either an increase in the interest margin (or some other non-traditional profit-making lines of business) or a better cost control by banks, if they want to maintain their long run surviving chances.

In the past 20 years, new entrants in the banking business were typically small banks; there was a huge amount of money invested either in the distribution channels (ie, the number of branches and ATMs increased substantially) or in computer and software systems. As of now, the universal bank model is the most prominent form of banking firm organization; on the other hand, there are no geographical restrictions to the banking business activities (for UE members). In this paper we aim at studying the efficiency of Portuguese banks during a period of sharp decline on banks rentability and the (eventual) existence of technological progress, using a traditional cost function framework. The paper unfolds as follows. We describe the model, the estimation technique and the indicators we use in section two and section 3 discusses the results. The paper concludes with some final remarks in section 4.

2. Model

2.1. Stochastic Frontier Model

Assuming that banks are cost minimizing organizations, their production process can be represented by the function C = f(y, w, t); this function represents the solution of the optimization problem Min Error! Bookmark not defined.w´x: y = g(x, t)Error! Bookmark not defined., where C represents variable cost, x is the vector of variable inputs, w represents the variable input price vector, y is the output vector and t represents time.

As a result of the assumed bank behavior, the stochastic cost frontier function can be written as

(1) lnCs = lnC(y, w, t, ) + us + vswith s = 1,..., N

where s represents the bank;  is a vector of unknown coefficients; v is a two-sided classical statistical error term with a zero mean and homoskedastic variance, 2v ; u is a one-sided, non-negative stochastic element, independent from v, representing cost-inefficiency. Clearly, if v=0 the frontier stochastic model and the frontier deterministic model are the very same one.

Furthermore, the vs are assumed iid, whilst the us, besides independently distributed from the vs, are assumed to follow a one-sided probability density function (the half-normal, truncated normal, exponencial or gamma, for example) that needs to be previously specified. Traditionally, the us are assumed to be half-normal (us is the absolute value of a N(0, 2u[4]) variable). Given the above mentioned hypothesis on the composite error term, the frontier function can be estimated by maximum likelihood (Greene, 1991), and efficiency levels are estimated using the regression errors. For the case of the normal-half-normal stochastic model, a firm inefficiency index is estimated via the us conditional average (Jondrow et al., 1982), with

(2)

where Error! Bookmark not defined.s = vs + us, Error! Bookmark not defined. = Error! Bookmark not defined.u/Error! Bookmark not defined.v, Error! Bookmark not defined. = Error! Bookmark not defined.u + Error! Bookmark not defined.v, and f(.) and F(.) are, respectively, the standard normal density function and the standard normal cumulative distribution function.

From an empirical standpoint, the different assumptions on the one-sided component of the composite error term do not lead to the very same results in terms of efficiency levels. According to Schmidt e Sickles (1984), the existence of panel data alleviates this problem for the researcher is able to estimate the sectional and time-series components of the composite error term. However, the assumption that efficiency is constant over time (Maúdos, 1996) is not reasonable, particularly when technological change is predicted to have occurred; under these circumstances, efficiency could be an important source of changes in total factor productivity (Esho and Sharpe, 1995). In our case, the Portuguese banking industry was subject to some major changes during 1990-95; therefore, we will assume that inefficiency varies across observations and time.

2.2. Estimation of the frontier function

From a theoretical standpoint, there does not exist consensus as to the best way to look at the banking production process. Two approaches are at stake, the production and the intermediation approaches[5]. From an empirical standpoint, none of them seems to have a clear edge over the other (Clark, 1988). In our case it is not possible to follow the production approach for we were not able to get information on the number of credit and deposit accounts at each individual bank. Therefore, we follow the intermediation approach.

On the other hand, in this type of studies there does not exist consensus on the variables which best define the banking product. According to Humphrey (1992) there is no reason to look only at the asset side of the balance sheet (credit granted and other applications, for instance) for the production of deposits (an input) absorbs a significant amount of other resources (labor and capital) used in banking. In this paper we follow the value added approach Berger e Humphrey (1992) and Maúdos (1996), deposit being simultaneously considered an input and an output.

We use the following products: Y1 = Deposits (from other banks + clients + securities issued + other liabilities); Y2 = Loans outstanding (credit granted + loans to other financial institutions + securities). We intend to estimate productivity and scale economy indicators, therefore using variable costs (C ), including both operating and financial costs (ie, we consider deposits, labor and other materials as inputs). Input prices are defined as follows: W1 = price of labor = staff expenses/number of employees; W2 = price of deposits = (interest and similar costs + commissions paid + financial losses)/Y1; W3 = price of other materials = (fixed asset depreciation + other administration costs + other operating costs)/net assets.

We use the following multiproduct translog cost function

where C = total variable cost; Yi = output i (i=1, 2); Wk = price of input k (k=1, 2, 3); s and t represent, respectively, the bank index and the year index; Greek symbols are the unknown parameters, assumed equal for all observations during the six years of the sample. The time variable, t, represents shifts in the production technology.

Our database was built using the information on the annual report of individual banks, which includes the balance sheet and income statement as of December, 31st of each year on a non-consolidated basis. Excluded from the sample were those banks which did not operate during the full year (ie, new banks that opened up for business during the year). The full list of the 221 banks used in the study is in Annex 1 (it is an unbalanced panel data set), and in table 2 we present information on some of the data set variables. Following the standard methodology in this type of studies (Rebelo, 1992), symmetry restrictions on the second order parameters and linear homogeneity on input prices are imposed prior to the estimation of model (3). Both variable costs and deposit and labor costs are divided by the price of other materials.

2.3. Economies of scale, technological progress and total factor productivity

Global scale economies (EE) measures the relative change in a firms total cost for a given proportional change on all outputs. EE is given by

(4)

when we use the translog function (3). If EE is greater than, equal to, or smaller than one, it is said that the technology exhibits increasing returns, constant returns and decreasing returns to scale, respectively.

Technological progress (PT) can be defined as the relative cost change over time as the result of technological change of the neutral type, coeteris paribus, and is defined as

(5)

Technological progress (regress) exists when PT is negative (positive), for we are using a cost function. The first two elements on the right hand side of (5) represent pure technical change, whilst the third element is associated with scale augmenting technical change (technological change due to modifications in the scale of production). The last component of (5) measures non-neutral technical changes (Kumbhakar and Heshmati,1996).

Following Esho and Sharpe (1995), once we estimate the movements of the cost function as a result of PT, the movements along the cost function as a result of EE and the cost inefficiency indexes (u), it is possible to compute the growth rate of factor productivity (TPF), given by

(6) TPF = -PT + (1 - EE)Error! Bookmark not defined. - u

where Error! Bookmark not defined. is the growth rate of the aggregate production, weighted by the cost elasticity of output.

3. Results

Parameter estimates for the frontier model (3) are presented in Annex 2. The estimated function verifies the regularity conditions (concavity and monotonicity) other than homogeneity and symmetry (which were imposed prior to the estimation) at the arithmetic mean point of approximation.

In tables 3 and 4 we show, respectively, the annual behavior of the above mentioned indicators and the estimated indexes by class of net assets. Our results are somewhat different than those estimated by other authors. In fact, our estimated efficiency levels are greater than those presented by Mendes (1995) and Pinho (1994), and also by Canhoto (1996). These three papers use a different output definition and also a different time-span; however Canhoto (1996) uses a non-parametric deterministic model but Mendes (1995) and Pinho (1994) use a stochastic model.

The average estimated inefficiency indexes are very similar along the 6 years of our sample, reaching the minimum in 1993 (0.05) and the maximum in 1995 (0.063). The average value of 0.057 means that Portuguese banks could have annually saved about 5.7% of variable costs if they were all able to use the best practice technology. However, individual banks do not all perform in the same fashion: the increasing value of the coefficient of variation means that some banks are loosing ground vis-à-vis their best-practice competitors, jeopardizing their long-run surviving chances. Therefore, we may conclude that the increased competition that Portuguese banks witnessed over the last few years did not lead to a better overall performance from the standpoint of costs: on the one hand, the annual efficiency average did not clearly increase over time; on the other hand, many more banks are now less efficient (in relative terms) than they were in the early 90s.

Results in table 4 suggest that larger banks (those with net assets in excess of 1000 million contos) are also the more efficient (estimated inefficiency = 2.9%), followed by average sized banks (more than 100 million and less than 300 million contos). The less efficient institutions are both big banks and very small institutions. Therefore, there is not a clear relationship between size and cost efficiency. However, it is interesting to note that efficiency and scale economies also seem not to be related with size: some of the less efficient institutions (net assets bellow 50 million contos) are the ones facing global, although small, economies of scale (EE<1) and the largest institutions are the more efficient but face diseconomies of scale (EE=1.021), therefore suggesting that they remain competitive via a better cost control. The optimal size of a bank seems to be somewhere in the neighborhood of 50-150 million contos of net assets.

As for technological progress, our results suggest the existence of technological recess, along the 6 years of the sample. From 1990 to 1995, the average annual real growth rate of costs was about 6% due to technological recess. The early years were worse than the mid nineties (9.8% and 2.9% costs increase in 1990 and 1995, respectively). The existence of new technologies, the large sums invested in computer technology, the diversification of products and services, did not seem to have yet a positive impact on cost reduction, but our estimates suggest that banks may be reaching the turning point regarding cost reduction due to those big investments (in 1994 and 1995 some banks were actually able to have cost decreases: see the min values in table 3). Results in table 4 also suggest that technological progress/recess and size do not seem to be related.

Table 5 shows the annual total factor productivity growth rate. Estimated PTF[6] values are very similar to the symmetrical of (PT+u), for EE is close to unity. Therefore, it is not surprising that the total factor productivity estimates are all negative, the average for the six years reaching -11.6%.

4. Conclusions

Our results suggest a general productivity problem in Portuguese banking. The opening of the banking industry in 1984 to new entrants (both private nationals and foreigners), the deregulation movement that followed (both on the pricing of products and services and on the variety of products and services banks were allowed to produce) as a result of the integration of Portugal in the EEC, and the reopening of the capital markets in Portugal produced a whole new way of doing banking in Portugal. New banks opened up for business, old banks expanded their productive capacity (see section 1 above), and some overcapacity may have been created. This, by turn, along with fierce competition on both the deposit and credit markets, had a negative impact on the total factor (deposit, labor and other materials) productivity, in spite of the acknowledged evolution of the information and communication systems.

Our results suggest that Portugal is facing a state of over-banking and over-branching. The size of the market, the existing number of institutions and branches, the increased competition, the recent desintermediation process, suggest that Portuguese banks were not fully able to absorb the probable benefits of the large sums recently invested. On the other hand, the existing entry barriers (language, culture, image, market knowledge, and so on) may be soon vanishing with the crescent European integration and the single currency, and some increased external competition may appear. Therefore, competition could well be intensified in the near future (we should not forget new phenomena such as home banking and electronic banking). The dramatic reduction of the difference between the interest margin and operating costs calls for a better cost control. Inflation rates in Portugal are steadily decreasing, and the nominal interest rates on loan applications are predicted to fall in the near future. The difference between credit and deposit rates is also predicted to diminish, putting some additional pressure on the banking system.