Abstract
Turning to the implementation of FCM on an international banking dataset, Section 3.1 summarises the main empirical questions and their context. Section 3.2 discusses the measurement difficulties that must be confronted in banking research, especially in a multinational setting, while Section 3.3 describes the dataset constructed for this study. Finally, Section 3.4 presents the parametric specification, the main econometric issues, the chosen estimation techniques, and the dual characterisation relations that emerge from the chosen parametric specification.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
In the UK in 1996, for instance, commercial banks alone numbered 348,000 employees and their operating expenses amounted to £23.6 billion. [The data sources used for these and other facts presented in this Section are Carrington et al. (1997) and the FDIC]
For a general review of the bank cost literature, see Gilbert (1984) or Kolari and Zardkoohi (1987).
The levelling of banking employment has been compared to the peaking in employment for other industries transformed by automation in the 20th century: for farming ∼1910s, for steel and motor vehicles ∼1950s.
In the United States, for instance, where banking employment peaked at ∼1.5 million employees as early as 1986, technology spending increased from ∼$14 billion in 1989 to ∼$20 billion in 1996.
This may be due to the focus of most previous studies on the US banking industry, which is the one a priori most likely to be operating close to optimal capacity levels.
Consider for instance Benston et al. (1982), Berger et al. (1987), Gilligan and Smirlock (1984) and Gilligan et al. (1984).
In some cases, recent mergers have also been justified by an expectation of realising scope economies (e.g., for the Morgan Stanley/Dean Witter merger) or gaining market power in a context of generally declining margins (especially for in-market transactions such as the merger between New York-based Chase Manhattan Bank and Chemical Bank).
There is no British, French, German or Japanese equivalent to the Functional Cost Analysis (FCA) program run in the United States by the Federal Reserve System.
For example, if a bank clears in a given year 500,000 checks and examines 40,000 loan applications, the idea here would be to consider that examining a loan application is equivalent in work load terms to (for example) clearing 50 checks and use this ”product equivalency ratio” to define a one-dimensional output aggregate, normalised in units of checks cleared, of 2,500,000 (500,000 plus 40,000 times 50). Conceptually, this approach could be extended to cover the number of deposit accounts monitored, the number of electronic payments, etc…
The ”product equivalency ratios” are difficult to estimate. Also, while a few banks do provide some transaction processing statistics, the vast majority does not.
Output price is of course defined as the value of output divided by the latent output aggregate, so that differences in output prices could be due to differences in product price levels and/or to differences in product mixes.
The estimation of the shadow cost function in the two-step econometric approach discussed in Section 2.3.3. would not be impacted either since would not change. Notice however that replacing output with its value, or more generally measuring output with error, would generate estimation problems for both the traditional shadow cost approaches discussed in Section 2.2. and the neo-classical cost approach.
It is critical here to use the logarithms of prices and output, and not their levels.
If the production function is specified to be Cobb-Douglas as in (2.13), the same line of reasoning can be followed for inputs, and FCM is essentially robust to the replacement of inputs with their values.
Across the entire sample, customer & short-term funding and customer loans respectively represent 75% and 60% of the balance sheet.
”Total assets”, frequently used in national studies as a proxy for output, proves to be an inconvenient metric in a multi-national context because of the considerable crosscountry differences in the level of inter-bank business and because of the difficulty of consistently removing this business from reported balance sheets. Also, to account for different degrees of risk taking, risk-adjusted definitions of output were tested (e.g.: total revenues minus loan-loss provisions), but did not yield significantly different econometric results; most accounting provisions indeed occur ex-post and are therefore not timed to coincide with the relevant revenue inflows: this issue is unlikely to be resolved until banks upgrade their risk management infrastructures and build RAROC-type reporting systems.
I wish to thank Jean-Paul Ndong, from the Mitchell Madison Group, for his very kind assistance.
Throughout this study, three classes of bank sizes are considered: ”small” banks with less than 4,000 employees, ”medium-sized” banks with 4,000-20,000 employees and ”large” banks with more than 20,000 employees. Size classes were defined in terms of number of employees and not output levels because of possible output measurement error.
Remark however in the first table presented in the Appendix that most of the 2,778 observations are in fact on American and Japanese banks, on smaller banks and in the period 1992-1995.
Notice from the technology price computation that, given data limitations, it was not possible to obtain different technology price estimates for different countries (i.e. beyond PPP-induced differences).
1996 was an exceptional year for Japanese banks which, benefiting from the decline in interest rates initiated in 1995, earned large returns on their bond portfolios, on trading activities and on balance sheet lending.
It is particularly important for the group/time effects on θit to be specified flexibly because, when output is replaced by its value, these group/time effects are also those assumed for output prices.
Subscripts are dropped throughout this Section.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 1999 Springer Science+Business Media New York
About this chapter
Cite this chapter
Sarkis, Z. (1999). Empirical Modelling. In: General Cost Structure Analysis. Springer, Boston, MA. https://doi.org/10.1007/978-1-4615-4599-6_3
Download citation
DOI: https://doi.org/10.1007/978-1-4615-4599-6_3
Publisher Name: Springer, Boston, MA
Print ISBN: 978-1-4613-7078-9
Online ISBN: 978-1-4615-4599-6
eBook Packages: Springer Book Archive