An Empirical Evaluation - CiteSeerX

The Economic Value of Network Externalities in an Electronic Inter-Bank Payment Network: An Empirical Evaluation

Eduardo S. Jallath-Coria Banco de Mexico 5 de Mayo 1, cuarto piso Mexico 06059, D.F. [email protected]

Tridas Mukhopadhyay Sandra Slaughter* Graduate School of Industrial Administration Carnegie Mellon University Pittsburgh, PA 15213 [email protected] [email protected]

Amir Yaron The Wharton School University of Pennsylvania and NBER [email protected]

August 1, 2001 * Contact Author for this paper Acknowledgements: We thank Alberto Espinosa, Mark Fichman, Margarita Moleres, Jose Luis Negrin and participants in seminars in Carnegie Mellon University, Banco de Mexico and the Workshop for Information Systems and Economics in Brisbane, Australia for helpful comments and discussions. Any remaining errors are our responsibility.

*** Do not cite, quote, or distribute without permission of the authors ***

i

The Economic Value of Network Externalities in an Electronic Inter-Bank Payment Network: An Empirical Evaluation

Abstract Theoretical work ascribes a positive value to the externalities arising from the adoption of network technologies. However, few studies have attempted to quantify the business value of these externalities. As a result, relatively little is known about the economic impact of externalities on firms that adopt network technologies. In this study, we investigate the value of an electronic inter-bank payment network in terms of its effect on the reserve management performance of commercial banks. Performance is measured by the opportunity and penalty costs generated by balances on the reserve account that commercial banks hold at the central bank. We propose that the electronic inter-bank payment network enhances banks’ reserve management performance (i.e., reducing opportunity and penalty costs) by providing more timely information on deposits and withdrawals affecting the banks’ reserve accounts. Technology impact is characterized as an initial stand-alone effect and a network externalities effect as more banks join the network. To evaluate the economic value of the technology impact, we analyze data from the implementation of an electronic inter-bank payment network adopted by all Mexican commercial banks. We find that early adopters of the electronic network, with a low ratio of electronic to overall operations, experience increasing opportunity and penalty costs. However, as additional banks join the network, the ratio of electronic operations increases, and costs decrease. After all banks have adopted the electronic network, we observe a reduction equivalent to 9.9% of each bank’s opportunity and penalty costs. The aggregated savings for all banks equal $5.3 million dollars for the 6 months following the technology adoption. Overall, the electronic inter-bank payment network project provides a significant positive net present value. Our findings provide support for the network externalities hypothesis. They also suggest an initial negative performance impact when there is concurrent use of old and new network technolo gies.

Keywords : Network Externalities; Business Value of Information Technology; Electronic Banking; Information Technology Investment; Economic Analysis.

ii

The Economic Value of Network Externalities in an Electronic Inter-Bank Payment Network: An Empirical Evaluation 1. Introduction Telecommunication technologies and networked applications have become increasingly prevalent. For instance, firms develop Electronic Data Interchange (EDI) links to share information with their network of customers and suppliers. In the financial services sector, banks offer direct connections to their corporate customers through PC banking applications and, more recently, via Internet banking services. Similarly, the World Wide Web is encouraging the creation of applications that support information sharing among different firms. These technologies have implications beyond the traditional transaction-processing gains. More timely and accurate information can enable a firm to improve its decision making process. Moreover, the benefits of the system are often positively related to the number of firms using the system, because an increase in the number of users can increase the available information, generating network externalities. Externalities are an important issue in the adoption of standards and network technologies (Katz and Shapiro 1986). For instance, word processor applications, such as Microsoft Word, derive a great part of their value from the number of compatible users of the product (Brynjolfsson and Kemerer 1996). Similarly, network applications such as electronic mail are more valuable when the number of users increases. Some studies have theoretically analyzed network externalities (Farrell and Saloner 1986, Saloner and Shepard 1995); others have empirically exa mined the likelihood of network adoption (Kauffman et al. 2000). However, few studies have attempted to empirically quantify the economic value of network externalities. Thus, it is not possible to address important questions concerning the strength of the network effect and whether the externalities effect is uniform across all users in the network. Our objectives in this study are to measure the performance impact on a firm when it adopts a network technology, to determine whether this impact can be attributed to network externalities, and to assess whether the externalities effect is uniform across all firms in the network. We hypothesize that, ceteris paribus, the adoption of a network technology and the ensuing network externalities increase the information set and thereby improve the performance of each firm in the network. To test this hypothesis we analyze detailed data on operations in an

1

inter-bank payment network operated by the Central Bank of Mexico (Banco de Mexico) and used by all commercial banks in the country. The Central Bank provides services to commercial banks similar to those provided by commercial banks to their corporate customers. Commercial banks have a reserve account that is used not only to fulfill reserve requirements but also to perform inter-bank transactions. Positive balances in this account pay no interest, generating an opportunity cost for the bank. Negative balances are charged a high penalty rate, generating an overdraft cost. Thus, excess balances may be necessary to avoid expensive overdrafts generated by uncertain operations. During the period of analysis, the banks adopted an electronic interbank payment network that allowed them to perform on- line fund transfers and queries on the reserve account balance, thereby improving the information used to forecast uncertain transactions affecting reserve balances. A unique feature of this setting is that the adoption of the network technology by the commercial banks occurred sequentially permitting us to directly analyze the costs, benefits, and externalities derived from network adoption for each bank. We find that early adopters of the electronic network, with a low ratio of electronic to overall operations, experience an initial decrease in performance, i.e., their opportunity and penalty costs rise. However, as additional banks join the network, the ratio of electronic operations increases, and costs decrease. Once all banks have adopted the electronic network, we observe a reduction equivalent to 9.9% of each bank’s opportunity and penalty costs. The aggregated savings for all banks are equivalent to $5.3 million dollars for the 6 months following system adoption. Thus, our results provide support for the network externalities hypothesis. We also find an initial negative performance impact when there is concurrent use of old and new network technologies. This paper is organized as follows. The second section reviews the related literature on network adoption and externalities. Section three develops the research model for the study. The fourth section describes the general characteristics of the Mexican financial system, the interaction of financial institutions with the central bank, the implementation of the inter-bank payment network, and the dataset. Section five describes the estimation of the model, and presents and discusses the results. The last section summarizes the contributions of this work and suggests opportunities for further research.

2

2. Related Literature The literature on externalities has characterized the role of network externalities in standards (Farrell and Saloner 1985, 1986) and in network adoption (Katz and Shapiro 1985, 1986). In the theoretical work on standards, Farrell and Saloner analyzed whether standardization benefits could lead to the adoptio n of an inferior standard when a better alternative was available. They found that the adoption of an inferior standard could not occur in the presence of complete information and also derived the conditions under which users would switch from an existing technology to a new one. The work of Farrell and Saloner is relevant to our study because it shows that users can find it optimal to adopt a new network technology, despite inertia in the use of an old technology. In the empirical work on standards, Gandal (1994) found support for the externalities hypothesis by estimating a hedonic price equation for spreadsheet programs. He showed that customers are willing to pay a significant premium for spreadsheets that are compatible with the Lotus 1-2-3 platform. Similar findings were reported by Brynjolfsson and Kemerer (1996) who measured externalities in terms of a product’s installed base. These studies attempt to characterize the economic value of externalities; however, the focus is on standards, not on value. In the research on network adoption, the general finding is that network externalities serve to promote the rate and likelihood of technology adoption. For example, using data on the estimated fax installed base, Economides and Himmelberg (1995) showed tha t the speed of adoption increases in the presence of network externalities. Similarly, Saloner and Shepard (1995) examined data on banks’ adoption of ATMs and found that the likelihood of adoption increases with the number of branches served (network effect) and the number of users (scale economies effect). Using a hazard model to evaluate the likelihood of adoption of a shared ATM network, Kauffman et al. (2000) found that banks that can generate a larger network size and a higher level of externalities tend to adopt ATM networks earlier than banks that own a large branch network. Finally, Gowrisankaran and Stavins (1999), examining data on Automated Clearing House (ACH) payments, found that network externalities increased the rate of ACH adoption. To summarize this work, there are but a few studies that measure the economic value of externalities, and these studies have focused on standards. Research on externalities for physical

3

networks has concentrated on the theoretical aspects of the problem, leaving aside the empirical measurement of the externalities. The few empirical studies in this area have examined the influence of network externalities on network adoption without attempting to assess the economic value of the externalities. As a result, relatively little is known about the economic impact of externalities on firms that adopt network technologies. A major contribution of our work to the literature in this area is to develop a general approach to modeling and evaluating the value of externalities ensuing from network technology adoption. 3. Theoretical Basis and Model In this section we define a model to evaluate the impact of network technology adoption on firm performance. Specifically, we model the impact of adoption of an inter-bank payment network on the opportunity and penalty costs (OPC) that banks incur in managing balances on their reserve account. The model proposes that, controlling for the main factors affecting performance, a bank can improve its performance (i.e., reduce its OPC) conditional on improvements in the information generated by adopting the network technology. We start this section by describing a general model to account for technological impact. The model is then detailed to address the specific technology examined in our study – an inter-bank payment network - and the implications of network adoption. The model is further developed to characterize the effect of both the adoption process and the network externalities. Finally, we enhance the model to control for changes not attributed to technological impact. 3.1. General Model of Technological Impact To account for the impact generated by the introduction of technology, we propose an augmented version of the classical growth-accounting framework developed by Solow (1957). The central equation of the Solow model is as follows: Y=A(•) f(K,L)

(1)

where Y is output, A(•) measures the impact of technology, and K and L are capital and labor. In this paper, we tailor the Solow model in two ways. First, we make explicit the functional form of the technological impact A(•). That is, we characterize the impact of technology adoption not by portraying a time trend but by depicting an initial stand-alone effect and a subsequent network

4

externalities effect as more users adopt the network. Second, we describe the input factors that affect the cost minimization process f(•) performed by a firm. The central hypothesis of our study is that, ceteris paribus, improvements in the information set generated by the adoption of network technology improve firm performance (reducing costs). This implies that, for a given set of inputs, the adoption of an electronic network allows a firm to improve its cost minimization process. Considering the structure suggested by the Solow model (1) and assuming a neutral technological change, 1 we can represent the effect of network technology adoption by a function A(•). This function depends on an observable vector of variables Z and a time invariant vector of parameters θ 1 . Similarly, we can characterize the initial OPC (i.e., before the network technology adoption) as a function f(•) that depends on an observable vector of time varying variables X, and a time invariant vector of parameters θ 2 . Thus, we can propose that, assuming no effects for technological change, the opportunity and penalty costs follow a normal distribution with conditional mean β'X and conditional variance σ2 , where {β ,σ2 }∈ θ2 . Consequently, if we put together the factors that account for technological impact A(•) and the functional form of the cost minimization process f(•) we have: OPC = A(Z; θ 1 ) f(X; θ 2 )

(2)

A graphical description of the role of the technological impact A(•), generated by the adoption of an electronic payment network, and the cost minimization process f(•) can be seen in Figure 1. The functional details of each factor are described in the following two sections. Figure 1 Factors Characterizing the Opportunity and Penalty Cost

Cost Minimization Process f( X ,θ 2 )

Opportunity and Penalty Costs OPC

Payment Network Adoption Effect A( Z , θ 1)

3.2. Technology Adoption In this section we depict the details of the technological impact A(•). We start by describing the general characteristics of a two-way payment network and how its users move from paperbased payments to electronic payments. Then we describe how adding new banks to the network creates network externalities for the banks that have already subscribed. Finally, we adapt the

5

functional form of the network externalities model proposed by Farrell and Saloner (1986) to incorporate it into the functional form of equation (2). 3.2.1. Payment Networks. Payment networks facilitate payments by concentrating the payment process in a switch. The switch has one account for every subscriber so that payments can take place by transferring funds from one account to the other. When a user wants to perform a payment, he or she refers the payment to the switch. The switch debits the user’s account and credits the account of the beneficiary. Then, the switch notifies the beneficiary about the amount of the payment. The above interaction can be framed in the context of a two-way network in which there is an interchange of information flows sent and received by users using a switch (Economides 1996). From a user stand point, there are two information flows: 1) the flow of the user sending the payment request to the switch (outgoing payment flow), and 2) the flow of the payment notification coming from the switch (incoming payment flow). In an inter-bank payment network, commercial banks are the users of the network, and the role of the switch is played by the central bank. Figure 2 provides an illustration of a two-way inter-bank payment network in which bank 1 sends a payment to bank 2. Figure 2 Two-way Network

bank 1 Outgoing payment flow

bank 2

bank 3

bank 4

Incoming payment flow

switch (Central Bank) The flows between the banks and the central bank can take place by sending paper-based documents or by sending electronic messages. In the paper-based mode, payments are performed by banks completing forms (similar to checks) that are sent by courier to the central bank (outgoing payment flow). The central bank processes each form by debiting the account of the payer and crediting the account of the payee. Once all forms have been processed, the central

6

bank produces a printed statement with the details of the operations of the day. The beneficiary bank receives by courier the printed statement with the payment information (incoming payment flow). In the electronic mode, banks have computer clients connected to the computer server of the central bank. The computer client allows every commercial bank to key in the payment requests into the system (outgoing payment flow). Once a payment has been keyed into the system, the computer server at the central bank credits the account of the payer, debits the account of the payee and updates the respective balances in real time. The beneficiary bank, using the computer client, can inquire on- line about the updated details of payment information (incoming payment flow). We can describe the adoption of an electronic inter-bank payment network in the context of a set of banks switching from a network of paper-based flows to a network of electronic flows. We expect that banks find it optimal to switch to the electronic network because of improvements in the level of aggregation and timeliness of their information flows (Ahituv 1989; Barua et al. 1989). The electronic network improves the level of aggregation of information by providing an on- line facility with a summary of incoming and outgoing payment flows. That is, the system provides a list of the operations generated by all of the bank’s departments as well as a list of the operations generated by other banks. Similarly, it improves the timeliness of information by speeding up information about incoming flows generated by the banks adopting the electronic network. Thus, a bank switching to an electronic network of size N experiences two effects. The first occurs when all the departments of a bank adopt the electronic network, and as a consequence all outgoing payment flows become electronic and are summarized into the computer system (improved level of aggregation). The second is a cumulative effect of speeding up incoming flows that depends on other banks adopting the electronic network (improved timeliness). Since we have two separate effects and the second effect depends on other banks joining the network, the effects for banks adopting the network will vary through time. The initial sequence of adoption is as follows: •

The first bank that switches to the electronic network will have all of its outgoing

payment flows electronically sent to the central bank. The system at the central bank will

7

summarize these operations (improved level of aggregation). However, the first bank will receive the information about incoming flows using the next period printed statement because the other N-1 banks have not yet joined the electronic network. Similarly, the N-1 banks experience no changes in their paper-based payments flows with the central bank. •

The second bank that joins the electronic network will send all of its outgoing flows to

the central bank electronically (improved level of aggregation) and it will receive electronic payment flows generated by the first bank that adopted the network (improved timeliness). The first bank that adopted the network will receive electronic incoming payment flows from the second bank (improved timeliness). However, the first and the second banks will receive the remaining incoming flows using the paper-based procedures because the other N-2 banks have not joined the network. The N-2 banks keep operating their paper-based flows with the central bank without changes. The adoption of the electronic network continues until all N banks adopt the electronic network and all incoming and outgoing flows with the central bank are performed using the new technology. An important point to highlight is that the summary of outgoing payments is complete from the first day of network adoption but the incoming transactions are not electronic until all banks have adopted the network. This gradual adoption of the electronic network forces banks to use two technologies. That is, during the transition period, banks must deal with incoming electronic transactions coming from the banks that have joined the network and with incoming paper-based transactions coming from the other banks. The concurrent existence of the two network technologies contrasts with the model presented by Farrell and Saloner (1986) in which the adopters of the new technology generate negative externalities to the old network installed base. In this context, the adopters of the new technology do not affect the utility of the users of the old technology but instead must deal with a transition period in which they concurrently use both old and new technologies. The problem of the dual use of technologies is eliminated once all banks have subscribed to the electronic network. 3.2.2. Network Externalities. Theoretical literature on network externalities suggests that the value of network adoption derives from two sources: the initial adoption effect and the

8

externalities effect based on the number of subscribers in the network (e.g. Farrell and Saloner 1986; Kauffman et al. 2000). The basic representation is given by: sa+ne(n)

(3)

where sa is the stand-alone effect that a firm derives from adopting the technology and ne(n) represent the network externalities effect derived when n firms have joined the network. 2 In the context of an electronic inter-bank payment network, the stand-alone effect sa is characterized by the switch to electronic mode of the outgoing payment flows, whereas the network externalities effect ne(n) signifies the automation of the incoming payment flows. If the focus of the analysis is on the information set of a bank, the benefits of the stand-alone effect are characterized by the improvements in the level of aggregation of outgoing payment flows, whereas the benefits of the network externalities derive from improvements in the level of aggregation and timeliness of the incoming payment flows. In an inter-bank payment network, the improved level of aggregation and timeliness directly relate to the amount of information generated in the system (i.e., the number of payment flows). Expression (3) assumes that all n banks generate a homogeneous number of payments. When we relax this assumption, and allow for heterogeneity in the number of payments per bank, the adoption benefits will depend on the relative number of payment transactions rather than on the number of banks. Thus, the adoption effect for a given bank can be expressed as: sa+ne(tr/TR)

(4)

where tr is the number of incoming transactions (incoming payment flows) generated by banks that have switched to the new network and TR is the total number of incoming transactions for a given bank. Therefore, we hypothesize that the improvement in the information set available to each bank is a function of the transactions generated by the initial stand-alone adoption effect and the network externalities generated by other banks in the network. We assume that a bank will use the new information to improve the forecast used in its cost minimization process. In this context, we can express the stand-alone and network externalities effects as the factors that constitute the technological impact on the OPC process, or A(•). The symbolic representation is expressed in (5) as follows:

9

OPC=A(sa+ne(tr/TR);θ 1 ) f(X;θ 2 )

(5)

3.3. Opportunity and Penalty Costs (OPC) In this section we depict the details of the cost minimization process f(•). We start by describing the general characteristics that generate a bank’s OPC. Then, we define the set of input factors that affect the cost minimization process. Reserve requirements and the provision of inter-bank payments require that commercial banks keep a reserve account at the central bank. As with any account, the reserve balance can be positive or negative. A positive balance may pay a low or zero interest rate for the bank, thus generating an opportunity cost. A negative balance results in a high interest rate charged on the overdraft amount creating a penalty cost. It would be optimal for a bank to keep the exact amount of money needed to pay for the transactions that take place during the day; that is, to keep a zero end-of-day balance. However, to avoid expensive overdrafts, a bank must keep some excess reserves to allow for uncertain operations that occur during the day and in particular, overnight. 3 Uncertain operations generate informational constraints under which banks try to forecast a balance to minimize the opportunity and penalty costs for the reserve account. In other words, banks try to minimize costs conditional on the available information. We can characterize the conditional mean of the cost minimization process (β'X) by the exogenous random transactions affecting the reserve balance and some exogenous and endogenous factors affecting the banks. In this context, the opportunity and penalty costs depend upon the value and volume of the uncertain transactions affecting the balance. For instance, a bank performing payments of high value might end up with a higher outstanding balance than a bank performing payments of low value. In addition, some endogenous factors depict the characteristics of the firm. For example, a bank with a network of nation-wide branches might experience more logistical problems aggregating information than a regional bank. The lack of aggregation might be reflected in higher costs. Exogenous factors can also reflect the financial and economic environment. For instance, if the prevailing rates in the market in a given day are higher than the rates on a subsequent day, banks will end up with different costs. The above descriptions suggest different levels of factors affecting the OPC of each bank: Transaction Factors (TF), Firm Factors (FF) and Economic Factors (EF). Transaction Factors

10

include the magnitude of incoming and outgoing operations that affect the bank's reserve account. We can split them into daylight and overnight operations. Daylight operations affect OPC because they are subject to operational errors, omissions and delays. Overnight operations are generated by third parties and create random amounts unknown to the bank. Firm Factors reflect intrinsic characteristics of a bank and therefore have an indirect impact on the costs. For instance, the size of a bank is an indication of its operational complexity. A large number of branches can be an indication of economies of scale or the level of geographical dispersion. Thus, firm-specific characteristics will either facilitate or complicate the processing of information used in the cost minimization process. Economic Factors have a direct impact on the costs. Higher interest rates will generate higher opportunity and penalty costs, ceteris paribus. Similarly, changes in the monetary base will generate changes in the balances of banks affecting cost. Hence, the conditional mean of the OPC process (β 'X) depends on a time variant vectors of variables {TF, FF, EF} ∈ X and a time invariant vector of parameters β ∈ θ 2 . The details of the elements included in the set of factors {TF, FF, EF} are described in the next section of this paper. To summarize, we have proposed a general model to evaluate whether the adoption of a network technology can improve firm performance, i.e., help banks reduce the opportunity and penalty costs generated by excess balances in their reserve accounts. We hypothesized that conditional on improvements in the information set derived from the electronic network, firms can improve their cost minimization process. We modeled the changes in the information set as an initial stand-alone network effect and a subsequent network externalities effect generated when more users adopt the network. Finally, to characterize the cost minimization process, we identified three levels of factors (transaction, firm, and economic factors) affecting costs. We now elaborate upon this general model by incorporating elements relevant to our empirical context. 4. Empirical Context, Measures and Data In this section we begin with a general description of the Mexican financial system, followed by a description of inter-bank operations. Then, we describe the implementation process of the electronic inter-bank payment network, and finally we depict the characteristics of the data set.

11

4.1. The Mexican Financial System In 1990, Mexico’s financial system included a Central Bank and 19 commercial banks. The Central Bank of Mexico (Banco de Mexico) offers banking services to commercial banks similar to those offered by commercial banks to their corporate customers. The main vehicle for settling inter-bank operations is the reserve account that banks have at the Cent ral Bank. Commercial banks pay and receive funds in the reserve account from 9 AM to 5 PM. The balance of the account changes as different departments of the bank perform operations that require electronic (or paper-based) payments. However, just before 5 PM, the bank has to decide how much money to leave in the account to support the overnight withdrawals generated by the settlement of the clearinghouse. If the end-of-day balance is positive, there is an opportunity cost. If the end-ofday balance is negative, a penalty is charged on the account. 4 During the first six months of our study, the aggregated outstanding balance of the 19 commercial banks was equivalent to $676 million dollars. This balance generated a combined holding and penalty cost equivalent to $87.7 million dollars. Consequently, an improvement in the management of the reserve account has the potential to generate significant benefits. Figure 3 provides an illustration of the intra-day behavior of the reserve account of a bank. Figure 3 Illustration of intra-day behavior of the balance of the reserve account of a bank Balance

Daylight Operations

9 AM

Overnight Operations

5 PM

12 PM

time

4.2. Inter-bank Operations The operation of inter-bank payments is relatively uncomplicated. For example, if Bank A buys a bond from Bank B, Bank A generates a payment from its reserve account to the reserve account of Bank B. The settlement of the payment is performed by the debit of the reserve account of Bank A and the credit of the reserve account of Bank B. Prior to adoption of the

12

electronic inter-bank payment network, fund transfers were performed by banks completing forms (similar to checks) that were later sent by courier to the Central Bank. When the forms reached the Central Bank, they were keyed into the computer system that controlled the reserve accounts. The system processed the operations and automatically credited and debited the reserve account of each bank in the Central Bank’s general ledger. At the end of the day, the computer system printed a balance statement showing the operations performed during the day. This statement was sent to each financial institution by courier the following day. Despite the existence of a computer system in the Central Bank, inter-bank operations were paper-based, because commercial banks had no direct access to the Central Bank’s computer system. Inter-bank operations affecting the reserve account can be divided into daylight and overnight operations. Daylight operations include fund transfers (used to settle foreign exchange and money market operations), federal tax related operations, and cash deposits and withdrawals. 5 Overnight operations were performed after all banks had closed and included the settlement of the check clearing houses. 6 Once daylight and overnight operations have been included, the final balance on the reserve account generates the respective opportunity or penalty cost. Although the optimal strategy for a bank is to leave a zero balance at the end of the day, possible errors, delays or omissions in the processing of daylight operations and uncertainty about the value of overnight operations make this goal infeasible. 4.3. Electronic Payments In 1990 the Mexican Central Bank started the implementation of an electronic inter-bank payment system to provide an on- line connection to every institution bearing an account at the central bank. The system enables financial institutions to switch from paper-based to electronic mode for the following operations: •

Fund transfers from reserve accounts (pesos checking account) to the accounts of other

banks, •

Fund transfers from U.S. dollar accounts to the accounts of other banks,

•

Government-bond transfers to the accounts of other banks, and

•

REPOs and direct purchasing and selling of government bonds to the central bank. 7

13

In addition, the electronic inter-bank payment system provides on- line queries to display the operations affecting the reserve account during the day. For instance, a screen could show outgoing payments keyed into the system by different divisions of the bank as well as incoming payments keyed into the system by other institutions. Hence, the system improves the level of aggregation by providing an automatic list of all operations affecting the bank's account (standalone effect). Similarly, the electronic inter-bank payment network improves the timeliness by providing on- line information of the incoming operations generated by the other banks (network externalities effect). In our analysis, we focus on examining the network technology impact on the reserve accounts of the banks in the network. We concentrate on the reserve account because it is the most significant in terms of value and number of operations. Moreover, the reserve account is the only account that is subject to the uncertainty of overnight operations and therefore makes the proposed evaluation more meaningful. 4.4. Implementation Process The implementation of the electronic inter-bank payment system took place from November 1990 to April 1991. We can distinguish three phases that characterize the process: the first, when the banks were only operating with paper-based payments; the second, when the banks were gradually switching to electronic payments; and the third, when all banks were operating using the electronic inter-bank payment network. A graphical representation of the system implementation process can be seen in Figure 4. Figure 4 Implementation Process 6 months

6 months

6 months

Bank 1 Bank 2 Bank 3 Bank 4 Bank 5 Bank 6

Bank 19 paper - based

Implementation

electronic - based

14

This implementation process has several characteristics of a quasi-experimental design (i.e., an experiment in which the researcher cannot control subject assignment to the stimuli): •

All banks adopted the same information system. This is important because, in general,

firms adopt different technologies making it difficult or impossible to conduct technology adoption evaluations across firms. •

For every bank, we have the equivalent of a within-subject interrupted time series design

in which we measure performance both before and after the implementation of the system. •

During the six- month implementation phase, the banks adopted the system sequentially.

About once a week, a new bank was added to the network while the remaining banks used paperbased operations. This strategy provides a between-subjects design for a given time t during which some banks have adopted the system while controlling for other banks that have no access to the electronic inter-bank payment network. In addition to the controls provided by our quasi-experimental design, we can control for other relevant aspects of the environment. First, the reserve account was the only account that banks used to settle inter-bank operations. Hence, there was no alternative account that we could not monitor. Second, the implementation of the system changed neither the rules for inter-bank transactions nor the deposit reserve requirements. For instance, trading practices and operational rules remained the same. Third, paper-based and electronic inter-bank payments were restricted to banks. Third parties such as companies and individuals did not have access to either paperbased or electronic inter-bank payments. Fourth, we have access to a wide set of variables to control for changes in the environment. For instance, we can characterize the magnitude of daily operations affecting the reserve account as well as exogenous factors such as inflation and interest rates. The disadvantage of this and every quasi-experiment is that the adoption sequence (i.e., assignment to the experiment) is not random. The lack of randomness can make it difficult to protect against some systematic bias. For instance, less efficient banks could adopt the system before more efficient banks. However, in our study, banks freely decided when they would adopt the system. Thus, despite the absence of random assignment, every subject had an equal chance to be assigned during the implementation process. Further, as we describe in our subsequent econometric analysis, we control for the possible impact of adoption sequence on performance.

15

4.5. Data The data we used in this study were collected from the electronic archives of the Central Bank of Mexico and from reports in the Bank´s supervisory office (Comision Nacional Bancaria). 8 The data include all daily reserve account operations from May 1, 1990 to October 31, 1991. We examine 6 months of daily operations before the adoption of the system, 6 months of daily operations during imp lementation, and 6 months of daily operations after all banks were operating electronically. Observations include both longitudinal (i.e., time-varying) and crosssectional (i.e., time-invariant) data. After eliminating holidays and weekends, there are 377 working days. This provides us with 377 time series observations for a cross section of 19 banks to yield 7,163 data points. In the following sections, we explain the details of each variable in the data set. 4.5.1. Opportunity and Penalty Costs (OPC). The observed OPC value is based on the ex-post observation of the end-of-day balance of the reserve account. The end-of-day balance is computed using the initial balance (Y) and the net of daylight (DL) and overnight (ON) operations. The opportunity and penalty costs incurred in the management of the reserve account are the result of multiplying the end-of-day balance by the given market rate of deposit for overnight funds (RD) if the balance is positive; or by the given penalty rate (RP) in the event of an overdraft. Thus, the accounting relationship of a bank’s OPC is: OPC = RD (Y + DL + ON )1( Y + DL +ON )≥0 − RP (Y + DL + ON )1(Y + DL+ ON ) 0.10). Finally, we explored if commercial banks could have developed internal information systems to improve the cost minimization process implying the existence of an embodied technological change. To test for this effect, we included a variable that equals an exponential time trend (TREND) in the first six months of the data set and zero when the system had not been implemented. The estimated model is given by: Ln(OPCit ) = α i+γ4 TRENDit +γ1 SAit +γ2 NEit +β'Xit +ε it

(12)

The coefficient of the time trend was not significant (p>0.10). Thus, we find no evidence of embodied technological change. 5.2 Results In this section, we examine the results from our final model (in equation 10) shown in Table 4 in the column labeled “Full Model” of the FGLS Fixed Effects specification. We start by examining the results for the stand-alone and network externalities effects of the network technology adoption. We then analyze the results for the variables that characterize the cost minimization process. The coefficient of the variable (SA) representing the stand-alone adoption of technology is significant (p