Credit Card Efficiency: Italian Cost Analysis

The paper begins by establishing the context of retail payment instruments, noting that while cash remains dominant, payment cards (credit and debit) are experiencing rapid growth, especially in Europe. The adoption of electronic fund transfer systems has led to reduced production costs and processing times for non-cash transactions. However, a 'monetary illusion' regarding the true social cost of payment services is prevalent due to information asymmetry, cross-subsidization, and opaque pricing mechanisms. This sets the stage for the central investigation into efficiency issues within self-regulated retail payment systems, a topic of increasing interest for antitrust authorities. The introduction also highlights the historical business model of payment cards, where cardholders often benefit at the expense of merchants who absorb high merchant fees; some card-issuing costs are offset by 'interbank exchange fees' rather than direct consumer charges. The inherent value of payment instruments is tied to network acceptance, underscoring the significance of network effects and the cooperation needed for their success. This cooperation, however, can lead to cross-subsidies and a 'low-powered incentive environment,' potentially resulting in x-inefficiency, where firms fail to operate at minimum feasible costs. The 'chicken and egg problem' inherent in network goods and services is also mentioned, noting the challenges faced by new networks in achieving critical mass. The paper mentions several reports regarding competition and interchange fees in the UK and Australia, providing relevant context for the focus on the Italian credit card market. The authors acknowledge that the economic analysis is not considering welfare economics aspects and that cost inefficiency might be tolerated in the early stages of network expansion.

The paper then delves into the specifics of the Italian credit card market. In 2000, Italy had 590 million annual payment card transactions, representing 19% of total non-cash payments, a significant increase from 3% in 1990, but still lagging behind other industrialized nations in per capita transactions. CartaSi holds a dominant position in the Italian market, issued by Servizi Interbancari and linked to international open-loop networks like Visa and MasterCard. Deutsche Bank is identified as a major competitor. The discussion differentiates between open-loop systems (Visa, MasterCard), characterized by self-regulation and global networks, and closed-loop systems (American Express, Diners Club), with more contained operations. The concentration of the Italian market is noted, despite recent growth in competition. The paper emphasizes the significance of the 'interchange fee' within the payment card circuit. This fee, paid from the merchant's acquiring bank to the cardholder's issuing bank, helps share issuing costs. The difficulty of reducing merchant discount fees (which include interchange fees) over time is considered. The impact of explicit cross-border interchange fees for proprietary circuits or three-party schemes are also highlighted, especially within the context of institutions offering domestic cards linked to the same international brand across multiple countries. The mechanisms for calculating uniform interchange fees are criticized for lacking transparency and failing to account for technological progress and productivity gains. The Italian market's structure is further analyzed, showcasing the dominance of operators controlling both issuing and acquiring aspects, and how interchange fees serve more as a price reference rather than a cost-balancing mechanism. The existence of implicit interchange fees in closed-loop systems is noted. The paper further discusses the vertical integration of some companies, where they fulfill the roles of issuer, system, and acquirer simultaneously.

A key theme explored is the potential for x-inefficiency within the credit card industry structure. The analysis highlights how network agreements significantly shape the cost structure of credit card companies, noting that commission expenses, including domestic and cross-border interchange fees, payments to partner banks, and connection fees to international circuits, form a substantial portion of the total costs. The paper points out that the traditional view of justifying protective measures to help network growth might be outdated. The notion of x-inefficiency, where firms fail to operate at minimum cost, is discussed in the context of the incentive structures within the credit card industry. The study investigates the complex relationship between revenue generation, particularly from interchange fees, and efficient cost structures. Revenue streams, largely linked to transaction volume, might not directly correlate with cost-efficient practices. This discrepancy between revenue and efficient cost structures is suggested as a key driver of x-inefficiency, especially in concentrated and vertically integrated markets. The paper concludes this section by linking the network size managed by each company with the emergence of a 'low-powered incentive environment,' where the relationship between revenues and efficient production costs is compromised over time. This sets the stage for the empirical analysis that follows.

This section details the econometric approach used to analyze cost efficiency in the Italian credit card market. The core methodology is the stochastic frontier approach (SFA), a technique that allows for the estimation of a parametric cost frontier. The study utilizes the Battese-Coelli (1995) model, specifically adapted for panel data, to account for both firm-specific and time-specific effects on cost inefficiency. This model explicitly includes a one-sided error term (u) to represent x-inefficiency, distinguishing it from purely random variations in costs. The choice of a cost frontier approach, rather than a production frontier approach, is justified by the exogeneity of input prices and the demand-driven nature of credit card output, which is not storable. The use of a translog cost function is explained, noting that it allows for the examination of scale economies, technical change, and factor demands. The selection of this model is contrasted with alternative approaches such as the Fixed Effects, GLS, or FGLS models, which are noted but not selected for this analysis. The concept of an unbalanced panel data set is also acknowledged, meaning that not all firms are observed in every period. The paper highlights that while the stochastic frontier method offers statistical advantages, particularly Maximum Likelihood Estimation (MLE), there are limitations and the authors address this by performing robustness checks using deterministic methods.

The empirical analysis employs an unbalanced panel dataset covering four leading non-bank credit card issuers and acquirers in Italy (Servizi Interbancari SPA, American Express, Diners Club, Setefi) from 1990 to 2001, resulting in a total of 40 observations. The focus on non-bank intermediaries is justified by their homogeneity in the credit card business, their market share (over 70% of the national market), and the availability of more reliable cost data compared to bank operators. The data includes information on productive and cost structures from annual reports and the Bank of Italy's Supervisory Returns Database. The study uses a translog single-output total cost frontier, omitting fixed or quasi-fixed inputs. The choice of a cost function over a production function stems from the assumption that input prices are exogenous and output is demand-driven. The model components are detailed: a composite output (y), calculated as the sum of transactions processed as issuer and acquirer; input prices for labor (pL) and capital (pk); and a time dummy variable (t) to capture Hicks-neutral technological change. The inherent difficulties in measuring fixed capital in the context of credit card companies, which often rely on outsourced infrastructure, are acknowledged. The model specification includes standard symmetry and linear homogeneity conditions. The one-sided error term (u), representing x-inefficiency, is specified as a function of the log of issuing transactions, justified by the role of issuer's revenue from interchange fees and fixed annual fees, and by using this variable as a proxy for market power.

The paper presents the interpretation of the model's parameters, focusing on several aspects of the credit card industry's performance. The analysis of cost efficiency, using a simplified Battese-Coelli model, shows the significance of the relationship between the number of transactions handled and x-inefficiency. The positive and significant coefficient associated with issuing transactions indicates a positive correlation between transaction volume and inefficiency. The results suggest increasing returns to scale, but this potential benefit is offset by the rising x-inefficiency in larger networks. Additional results shed light on input demands and factor substitution. A positive estimate for capital's factor share confirms the capital-intensive nature of the sector, while the Morishima elasticity points to a high level of substitution between labor and capital, likely due to outsourcing and partnerships. This cost structure is further linked to network agreements, supporting the earlier emphasis on their impact on cost efficiency. Analysis of technological change, measured by a time dummy variable (βt), suggests a slightly upward, neutral shift in the cost frontier. The paper concludes this section by presenting predicted inefficiency scores (average 18.2%), calculated using FRONTIER Version 4.1 software, emphasizing that inefficiency increases with network size. Robustness checks are briefly discussed, highlighting the comparison of the SFA results with deterministic methods (a descriptive standard cost method and Data Envelopment Analysis) to validate the findings.

The results section begins by analyzing scale economies within the Italian credit card industry. The findings indicate a potential for significant increasing returns to scale, meaning that expanding the scale of operations leads to lower average costs. However, a crucial counterpoint is presented: the extent of the network (measured by the volume of transactions) is negatively correlated with cost efficiency. In other words, while larger networks could theoretically benefit from lower average costs due to scale economies, the observed data suggests that they instead tend to deviate further from the efficient cost frontier. This deviation highlights the presence of x-inefficiency, where firms operate at higher-than-optimal costs, even given the potential benefits of scale. The study suggests that the drive for increased transaction volume and resulting revenues may be overshadowing the pursuit of cost minimization, which is a key driver of the observed x-inefficiency.

The analysis then examines input demand and factor substitution within the Italian credit card industry. The estimated capital factor share (βk) confirms the capital-intensive nature of the sector, aligning with the observed high proportion of non-personnel costs. The Morishima elasticity is found to be greater than one, signifying a substantial degree of substitution between personnel (labor) and non-personnel (capital) costs. This suggests flexibility in the input mix, which could be attributed to outsourcing, partnership agreements, or other strategies to optimize cost structure. Despite this flexibility, the own-price input elasticities are negative and inelastic for both capital and labor, indicating that the demand for these inputs is relatively insensitive to price changes. This inelasticity is consistent with the cost structure being significantly constrained by network relationships and agreements, reinforcing the prior findings on the importance of network effects on cost dynamics.

The study investigates technological change by analyzing the impact of time (represented by the coefficient βt). A positive coefficient (0.009) indicates a slightly upward neutral technological shift in the cost frontier over the 1990-2001 period. This could reflect, according to the literature, potentially excessive increases in capital expenses, or improvements in the quality of service. This increase in costs despite technological advancements could be attributed to investment in automation, fraud prevention, increased payment speed, or better customer service. The authors note, however, that the lack of quality adjustments in output measures may influence the interpretation, and it's important to consider the potential of capital intensive quality improvements masking actual cost reductions. The positive shift in the cost frontier suggests that improvements in technology and service quality have increased costs in the Italian credit card industry, partially offsetting the potential benefits of increased scale and efficient input utilization.

Recognizing the limited research on the asymptotic properties of MLEs in stochastic frontier analysis with finite samples, the authors conduct robustness checks using deterministic methods. This addresses concerns about the reliability of the x-inefficiency scores obtained from the stochastic frontier model. The checks utilize a balanced panel dataset (1996-2001), comparing the stochastic frontier estimation with a descriptive 'standard cost' method and Data Envelopment Analysis (DEA). The standard cost method's results are consistent with the MLE estimation, showing high and increasing x-inefficiency levels with network size. The application of DEA, which addresses the challenges of handling panel data in a manner comparable to parametric models, is described. Two approaches within DEA were considered: a sequential frontier (calculating efficiency yearly based on cumulative observations) and an intertemporal frontier (merging data from all years). The DEA results, using the DEAP computer program, incorporate a composite output and input prices (labor and capital), with a capital input quantity index based on network access points for the 1996-2001 period. The paper mentions that the deterministic methods are more sensitive to outliers compared to the stochastic methods. Overall, the robustness checks enhance the confidence in the research findings by confirming the key results using different methodologies.

The concluding section summarizes the key findings of the study, emphasizing the significant aspects revealed about the Italian credit card market over the past decade. The research concludes that while the industry possesses the potential for increasing returns to scale, the growth of the network is negatively associated with cost efficiency due to x-inefficiency. Network agreements and interchange commission flows significantly impact the cost structure, playing a crucial role in final pricing within the concentrated and vertically integrated market. These factors contribute to a 'low-powered incentive environment' where revenues are not consistently tied to cost-efficient production, underlining the need for ex-ante regulatory measures. The study acknowledges that efficiency problems have historically been addressed ex-post by antitrust authorities, however, it emphasizes the importance of proactive measures to establish transparent and efficient intra-network pricing schemes. The authors position their study as one of the first empirical analyses of cost inefficiency in private retail payment circuits, particularly in the context of the Italian credit card market, suggesting the need for further research to explore the nuances and implications of these findings for policy and regulation.