Скачать 117.91 Kb.
|A SPATIAL DECISION SUPPORT SYSTEM|
FOR BANK LOCATION: A CASE STUDY.
David J. Willer
National Center for Geographic Information and Analysis
Department of Geography
State University of New York at Buffalo
Buffalo, New York 14260
Technical Report 90-9
I would first like to thank my parents for providing me with the support and opportunity for this educational experience. Thanks also to my wife Teri and her parents for their friendship and support which has been greatly needed and appreciated. I thank Dr. Paul Densham for guidance and the opportunity to conduct research under his advisement, also to Dr. Barbara Buttenfield, Greg Theisen, Lew Jean King, and SUNYAB students who helped enrich my university experience. Partial support for this research was provided by Manufacturers Traders and Trust Company and the National Center for Geographic Information and Analysis (NCGIA). This research is a contribution to Initiative 6 of the NCGIA.
This paper presents a computer-based system that is designed to help solve semi-structured location problems that exist within the banking industry. Bank-branch location problems are complex, often requiring large data sets and sophisticated modeling techniques in their solution, providing a good testing ground for such a system. The first chapter of the paper discusses new developments in locational analysis and in bank-branch location, and an introduction to the spatial decision support system (SDSS) literature. Chapter two is a discussion of an SDSS designed and implemented to help solve bank location problems. This chapter also discusses a new area for the use of Geographic Information Systems. Chapter three examines the design and construction of a prototype SDSS which will make use of currently available software and hardware technologies. Chapter four focuses on a case study which uses the aforementioned SDSS to address a complex, bank-branch location problem, and to develop new solution strategies. This research consisted of a two-phase research project. The solution techniques and results for each phase will be examined. The final chapter consists of a summary of the research conducted designing, building, and implementing a prototype SDSS.
CHAPTER I. INTRODUCTION
There is a growing need for computer-based systems that are able to store, manipulate, retrieve, and, most importantly, analyze the large quantities of geo-referenced data that are available today. These data are a valuable resource for use in solving many problems that have not been easily addressed in the past because of their lack of structure.
The objective of this paper is to present a spatial decision support system (SDSS) that helps to solve a subset of semi-structured locational problems that exist within the banking industry. The bank-branch location problem is complex, often requiring large data sets and sophisticated modeling techniques in its solution. Such problems provide a good testing ground for the system I have implemented. In addition, this system is designed to provide a base for solving other types of semi-structured locational problems. A SDSS includes computer-based software that manages geo-coded data, and human operators (or analysts) who interact with the databases, analysis modules, and decision-makers. SDSS are unique in that they require decision-makers actively to participate at various levels of the analyses.
This introductory chapter will examine new developments in locational analysis, bank-branch location methods, and in location-allocation modeling, followed by a discussion of the SDSS literature.
A. Location Analysis
"People are distributed unevenly in earth space and they must obtain many kinds of goods and services from facilities located at widely separated places. They have an obvious interest in the location of these facilities being 'most accessible' to them." (Rushton, 1979, p. 31)
One of the early location models is known as the gravitational or gravity model. These models have been developed in a wide range of disciplines including economic geography, sociology, psychology, and marketing. In some cases a theory of behavior was used to supply the rationale for the gravity model, in some cases the model had no theoretical basis and was strictly empirical, and in other cases the model resulted from a direct transfer of physical science principles into the behavioral sciences, with neither theory nor empirical reasons (Lundsten, 1978, p.21). An extensive review of the gravitational research literature is given by Lundsten (1978) and Tocalis (1979) and interested readers may refer to those works for a detailed discussion.
The gravity model is often used to locate supermarket chains and gas station franchises (Chelst, Schultz, and Sanghvi, 1988) because the impact of competition may be incorporated in the context of facility location. A great deal of store location research has been directly impacted by the gravitational models developed by Reilly (1931) and Huff (1964). The primary weakness of gravity modeling, however, is that it cannot accommodate treatment of a network of facilities as a whole.
A second class of models is based on Applebaum's (1968) classic analog technique. This technique addresses trade area definition and market penetration. The analog model is used most effectively when a company is trying to locate a single facility or a few widely dispersed facilities over a large metropolitan area. Similar to the gravity model, the principal weakness of the analog model is that it evaluates single facilities while ignoring the effects of an entire network of sites (Chelst, Schultz, and Sanghvi, 1988).
Ingene (1984) and Mahajan, Sharma, and Srinivas (1985) have developed a third technique which draws on econometric methods of market potential estimations using multiple regression to estimate demand. Mahajan et al. (1985) present a "Portfolio Approach" which was developed for evaluating existing and identifying attractive potential locations for financial institutions. This approach is used for analyzing multiple-store locations.
B. Bank-Branch Location Analysis
One of the most widely known procedures used for evaluating and selecting a possible bank branch office location is the technique suggested by the American Bankers Association (ABA) in their book "A Guide to Selecting Bank Locations". A flow chart of the procedure is displayed in Figure 1. A checklist method is presented, leaving to the discretion of the locational analyst the choice of potential facility locations or candidates, the definition of the primary service area, the estimation of the proportion of the population who bank near home or work, estimates of realizable market share and growth, and final adjustments to accounts and balances (Lundsten, 1978). The ABA method suggests that analysts evaluate factors that are critical to the success of a branch such as socioeconomic and demographic characteristics of consumers, the level of competition, and consumer expenditure patterns. In addition, there are site-specific factors such as traffic patterns, parking availability, route access, and visibility, that are considered by decision-makers.
A number of different models have been employed for locating bank branches (Clawson, 1974; Littlefield, Burney, and White, 1973; Olsen and Lord, 1979; and Soenen, 1974). Clawson (1974) implements a stepwise linear regression in an attempt to more effectively screen new branch locations, set realistic performance standards for different areas, and pinpoint remedial actions for poorly performing branches. The analog technique, however, was the most widely used primary location model in the banking industry during the late nineteen-sixties (Fenwick and Savage, 1979).
C. Location-allocation Models
Recently, some analysts in the banking community have realized the potential of location-allocation models for solving the bank-branch location problem. Location-allocation models simultaneously optimize facility locations and the allocation of demand to those locations. Solution techniques for location-allocation models have existed for nearly twenty years, and have been formulated for problems represented in continuous and discrete space. The latter are more widely used, in part because available solution techniques are tractable for large problems. Formulations in discrete space are based on a network which is made up of a series of node-link relationships. Links can be characterized as transportation routes and nodes can represent route intersections, disaggregate demand points such as individual homes, aggregate demand points such as census tract centroids, and existing and potential facility locations.
There are two distinct solution techniques for location-allocation models: programming and heuristic methods. The former can employ linear, integer or mixed-integer programming methods and always yield the optimal solution. These methods often are computationally intensive and for large problems become intractable, as well as prohibitively expensive to solve (Armstrong and Densham, 1990). Heuristic techniques employ spatial search algorithms that produce the desired optimal solution with regularity. They do not guarantee the optimal solution as the programming methods do, although the Teitz and Bart Vertex Substitution Heuristic (Teitz and Bart, 1968) has been
Figure 1. Flow chart of the procedure suggested in the A.B.A. Green Book.
shown to produce the optimum solution with a high degree of regularity (Rosing, Hillsman, and Rosing-Vogelaar, 1979).
Heuristics are used when suitable programming methods may not exist, or may be prohibitively expensive, or when a set of alternative solutions using varied objective functions is desired. Used in an appropriate framework, the versatility of heuristic techniques enables decision-makers to examine the original problem statement and objectives through sets of alternative solutions, and permit and encourage any restatement or clarification of the problem.
Hillsman (1984) demonstrated that many location-allocation models have similar structures and that, by editing the objective function coefficients, each can be derived from the well known p-median model. Collectively, these variants are known as the Unified Linear Model (ULM). A heuristic, such as Teitz and Bart's (1968), that can solve the p-median problem can also solve many of the models in the ULM.
D. Spatial Decision Support Systems
There has been a separate but parallel development of systems to support decision-making in the management science and geographic literatures. Decision support systems (DSS), from management science, have evolved out of Management Information Systems (MIS) which have existed since the 1960s. Spatial decision support systems, from geography, are evolving from Geographic Information Systems (GIS) which have existed since the early 1960s. Most GIS provide capabilities for map overlay but do not support the analytical and statistical modeling required by many decision-makers (Armstrong and Densham, 1990).
According to Geoffrion (1983) there are six characteristics that distinguish a DSS:
1. 'They are used to tackle ill or semi-structured problems - these occur when the problem, the decision-makers objectives, or both cannot be fully and coherently specified.
2. They are designed to be easy to use, the often very sophisticated computer technology is accessed through a user-friendly front end.
3. They are designed to enable the user to make full use of all the data and models that are available, so interfacing routines and database management systems are important elements.
4. The user develops a solution procedure using the models as decision aids to generate a series of alternatives.
5. They are designed for flexibility of use and ease of adaptation to the evolving needs of the user.
6. They are developed interactively and recursively to provide a multiple-pass approach which contrasts with the more traditional series approach - involving clearly defined phases through which the system progresses.'
Spatial decision support systems represent the role of spatial computer systems in the decision making process. For the purposes of this research, the term SDSS will include more than the interactive spatial systems used by managers. It will also incorporate understanding of the decision making process involved in solving a specific bank location problem and fulfilling the characteristics set forth by Geoffrion.
Conceptually, a SDSS can be thought of as providing an integrated set of flexible capabilities; the implementation of such a system can be achieved using a set of linked software modules (Armstrong, Densham, and Rushton 1986; Densham and Armstrong, 1987; Armstrong and Densham, 1990). There are several components that form the core of a SDSS, these can be seen in Figure 2. Typically, a SDSS contains a spatial information system (GIS) integrated with a modeling system. Specifically, the system includes a geo-referenced database, analytical tools, and display and reporting capabilities (Armstrong and Densham, 1990).
Within a SDSS there are two interface components, a user interface and a system interface. The user interface is often the most important component in a system's development and perceived success. It provides users with access to the databases, modelbase (analytical routines), and graphical and report generation. System designers may provide different levels of access for different levels of users. Bank management, for example, may require access to all databases and analytical routines, while clerical personnel have restricted access. The user interface also enables users to generate output (maps, charts, etc.) directly from the database management system (DBMS) or from the results of analyses which may be stored in the DBMS (see Figure 2).
The system interface expedites the transfer of data between the DBMS and the modelbase and contains routines invoked automatically during SDSS execution. A modelbase, or a core of analytical routines, is integrated into the SDSS through the user and system interfaces.
Figure 2. Interaction of SDSS components.
CHAPTER II. SDSS AND BANK LOCATION PROBLEMS
There has been little research dedicated to large scale regional analysis of banking systems using aggregate data sources (Brundage, 1989). Similarly, there has been a lack of development of comprehensive bank-branch location modeling techniques which could exploit these data resources. Developments in information systems technology, specifically SDSS, are described below. These developments will enable bankers to address many of the problems that exist in bank-branch location and to develop more comprehensive solution strategies that use available data resources.
A. Data Issues
A substantial amount of preprocessing is required to format survey data. These data are often specifically formatted and stored on transportable media, such as ASCII flat files on floppy disks. Bank managers and analysts often use these data as input for statistical software packages that require a particular file format. A great deal of time and money is often spent reformatting such data. Reformatted data can be used: as input for analysis modules; to build a computerized database; and for producing presentation graphics such as maps, charts, and tables.
SDSS can reduce the burden of reformatting by the implementation of a system interface. This interface helps to expedite the transfer of data between the database management system and the available analytical routines in the modelbase (see Figure 2). Widespread use of telecommunications in businesses will enable SDSS designers to include sophisticated capabilities in future systems.
B. Bank Location
Three fundamental questions must be answered for bank-branch location problems:
1. How many bank branches should there be?
2. Where should these branches be located?
3. What services should be provided at each branch?
Bank management must address several preliminary tasks in order to answer these questions, such as (Chelst, Schultz, and Sanghvi, 1988):
1. defining and measuring branch performance;
2. evaluating the existing banking network;
3. estimating current and future potential for the targeted market areas;
4. characterizing different types of branches with respect to competition within an area;
5. monitoring short-term corrective measures and developing long-term strategic plans; and
6. creating methods for generating alternative solutions and for evaluating these alternatives.
The prototype SDSS developed in this research will help management focus on a number of these tasks, the most important being the generation and evaluation of alternative solutions to a specific bank-branch location problem. The system will also provide management with a tool that will allow the other tasks (outlined above) to be addressed.
Several scenarios may confront bank management in a bank-branch location problem (Chelst, Schultz, and Sanghvi, 1988). The first concerns an oversized, unprofitable branch network. Decision-makers may consider closing the poorly performing branches or, possibly, consolidating a number of these branches into a single, more profitable office. The second scenario is to consider expanding a profitable banking network to increase customer coverage. Bank management, relying on market potential projections, may forecast that a presence in a currently unserved market area will be profitable in the future.
The restructuring of an existing bank network in areas that are either under or overrepresented is a third scenario. The effect of overrepresentation is that one office may be cannibalizing another branch's market. As in the first scenario, bank management may consider consolidation of these 'competing' branches. A fourth scenario is to enter a new market. A successful bank may achieve a new market presence by securing an existing bank-branch network or by establishing a few strategically located branches. The final scheme combines elements of the first four scenarios. Successful banks are capable of purchasing a failed bank, which will often require measures to remedy poorly located, or cannibalizing bank branches.
The complex nature of these scenarios and potential solution strategies provide a good testing ground for a prototype SDSS. Such systems are likely to be a valuable tool in the development and application of modeling techniques and implementation strategies that focus upon specific bank location problems.
C. A SDSS for Bank-Branch Location
The SDSS developed in this research is capable of storing and manipulating large quantities of geo-referenced data applicable to the bank-branch location problem, as well as analyzing and displaying the data and subsequent results in a real-time, interactive environment. This prototype SDSS can be described as a system which supports decision-makers, not replaces them.
Bank management must develop a set of strategies that specifically focus on bank-branch location problems. These strategies should contain a preliminary problem statement and set of objectives. An SDSS enables decision-makers to participate in any or all phases of the analyses. Interaction between analysts and decision-makers within the SDSS often leads to more structured and well defined problems.
The analysis module implemented in this prototype SDSS, a location-allocation model, is applied to the location problem using the decision-makers' preliminary set of objectives. These objectives effectively determine the initial set of alternative solutions and frequently provide decision-makers with a new perspective on or understanding of the problem. These insights generate discussions between analysts and decision-makers which alter the objectives defined for subsequent solutions. The application of a SDSS may be described as a process of problem-structuring and objective building, alternative solution generation and evaluation, and strategic implementation (Figure 3).
SDSS encourage the interaction and participation of analysts and decision-makers at all stages of analyses. Refinement of the problem statement through decision-maker
Figure 3. Problem refinement process in a SDSS environment.
participation is made possible by the speed with which alternative solutions can be generated, displayed and evaluated. Analysts and decision-makers continue to iterate between refining objectives and generating alternative solutions until they are confident that they have a well-defined problem and objectives and are able to select a solution. This process helps decision-makers to understand more clearly location problems and to improve their subsequent decisions.
CHAPTER III. SDSS COMPONENTS AND CONSTRUCTION
This chapter presents SDSS design considerations for integrating a GIS and a specific location analysis model. There are five major elements in the prototype SDSS developed in this research:
A. a GIS engine;
B. analysis modules (modelbase);
C. a system interface;
D. a user interface; and
E. outputs (maps, charts, etc.).
The first step in designing the SDSS was to choose the GIS that would form the core of the system with which the analysis modules and interfaces would interact. A GIS called TransCAD (1988) was chosen for its geographic data management and cartographic display capabilities, as well as its interface for external analysis routines.
The next step was to choose an analysis module that would help solve the bank-branch location problem. A location-allocation package developed by Densham (1990) was chosen. Densham's implementation possessed desirable characteristics, such as:
1. capable of solving large location problems (up to a 3000 node network); and
2. speed and efficiency.
The third step in the construction of this prototype system was to implement the location-allocation software into the geographic information system by developing a system interface.
Finally, the prototype SDSS was tested on a bank-branch location problem. A local bank's proprietary data and the expertise of its decision-makers were used to test the efficiency and effectiveness of the SDSS.
A. GIS Engine
A geographic information system is capable of collecting, storing, retrieving, and creating output from geographically referenced data. Currently available GIS provide these capabilities but lack sophisticated analytical tools or models (Densham and Goodchild, 1989). The prototype SDSS designed in this research was driven by TransCAD (1988). This GIS is designed to model transportation networks using node-link relationships. TransCAD contains a set of transportation procedures and necessary data management capabilities to support them. The location-allocation package requires network-based data in the form of nodes and links files (Goodchild and Noronha, 1983). TransCAD can support these data structures using these data management capabilities.
TransCAD is a DOS-based GIS that was installed on a Hewlett-Packard Vectra i486 personal computer, running at 25 Mhz, with 10 megabytes of RAM, a 150 megabyte hard disk drive and a color VGA monitor. TransCAD can be installed and running in less than an hour.
A menu-driven system, TransCAD provides five fundamental functions:
1. storage and retrieval of geographically based data;
2. digital mapping;
3. transportation and operations research models;
4. statistical procedures; and
5. presentation graphics.
Additional, and particularly important, capabilities include:
1. the ease of analysis module integration; and
2. the construction and manipulation of databases to support analytical models.
TransCAD weaknesses are:
1. a lack of sophisticated cartographic display capabilities; and
2. binary file structures for the import of data, which are difficult to create and use.
A description of this GIS can be found in the TransCAD users manual (1988).
B. Analysis Module
As stated above, location-allocation models optimally locate a given number of facilities while simultaneously allocate demand to these facilities. Densham's (1990) package requires a topological network represented by two data files: the nodes file generally contains the locational and demand data associated with each node on the network; the links file represents the networks topology.
The location-allocation model used in this research was designed using widely known operations research techniques, implemented using geographic principles. These models have been applied to a number of location problems, such as the location of fire stations and road maintenance garages. Densham's (1990) implementation uses the Teitz and Bart (1968) vertex substitution heuristic. This implementation was chosen over exact programming techniques because it can solve large problems on a micro-computer in times which are an order of magnitude faster than other micro-computer codes (Densham, 1990). The major drawbacks to this implementation are that it lacks data file management and cartographic display capabilities. These shortcomings are directly addressed in Chapter IV of this paper.
The data required to run a location-allocation analysis are structured in relatively simple ASCII files. TransCAD contains a database query module, using the well-known Structured Query Language (SQL), that is capable of producing these ASCII files quickly. The specific file structures are presented in the appendix of this work. Densham's (1990) software produces a series of ASCII solution files that need to be converted, using the system interface, into a specific binary format. These binary solution files are then imported into TransCAD and stored in a database, where they can be manipulated and displayed.
The speed of this modeling technique allows the analysts and decision-makers to generate sets of alternative solutions, evaluate them, refine the input parameters and problem objectives, and proceed with another round of analyses in single problem solving sessions.
C. System Interface
Compatibility between separately developed software packages (TransCAD, 1988 and Densham's Location-allocation software, 1990) was a problem during the development of this prototype SDSS. This system was designed to provide a steady data flow between GIS and analytical modules. A system interface contains software components designed to be transparent to users in order to maintain smooth system performance.
The prototype system interface is automatically executed when any data is transferred between TransCAD and the location-allocation package. Although the modules are specific to this system, their design can be a model for future interfacing routines.
D. User Interface
The user interface is the link between users and software and is critical to a systems success. Due to the modular nature of a SDSS, this prototype system contains separate user interface techniques. TransCAD's (1988) user interface contains a series of mouse-driven, pull-down menus. Although much of the interface is fixed, TransCAD provides SDSS builders with the ability to guide users through analytical processing. In this prototype system, I present users with a brief software description and warning concerning Densham's (1990) analytical module. These messages were implemented in order to provide users with the option of returning to TransCAD immediately to fulfill analytical input requirements before entering the location-allocation module.
Densham's (1990) software is a set of DOS-based, Microsoft Pascal modules. These modules are driven through a scroll-bar menu (see Appendix) and were altered for implementation into this prototype SDSS. The general user interface contains a series of prompts to which users reply with input and output file names and model parameters.
Upon completion of a location-allocation modeling iteration, users reenter TransCAD, and are prompted for database locations to store analytical solutions or to forego any database input in times where results were undesirable or incorrect.
The fifth major element in this prototype SDSS is the output formed from the results of Densham's (1990) location-allocation software. TransCAD (1988) provides users with three forms of graphic output, including:
2. charts; and
During execution of a location-allocation analysis, Densham's (1990) software produces results with respect to solution facility locations (nodes), and allocation of proximal demand to those locations (see Chapter 1). TransCAD enables users to display these results on an interactive map display, bar or pie charts, or spreadsheets. Each of these displays may be sent to a graphics printer for hardcopy output.
F. Merging Existing Technologies
The prototype SDSS employs existing technologies to expedite the development of an interactive problem solving environment for bank-branch location. These technologies include: a GIS (TransCAD); a location-allocation model (Densham, 1990); a personal computer (Hewlett-Packard, i486) capable of running the software; and text and color graphics printers.
Using SDSS, decision-makers may alter the way they solve their problems. Decision support technology encourages decision-makers to become part of the analytical process, providing analysts with expert knowledge about location problems. This involvement helps to educate decision-makers about the characteristics of their problems, and its underlying relationships, and to identify solution strategies that have, heretofore, been improperly formulated.
Merging existing technologies provides several benefits. First, TransCAD will have additional analytical tools - location-allocation models. These tools will open new application areas for many GIS where such algorithmic solutions to location problems are required. Second, the data management and display capabilities in TransCAD will help increase the number of analysts that may apply location-allocation models to semi-structured location problems.
|Web based multimedia support in computer networks course: a case study||Cropping system: a case study in brazil|
|[Abel 1992] Abel, D. J., Yap, S. K., Ackland, R. G., Cameron, M. A., Smith, D. F. et Walker, F. G., Environmental decision support system project: An||Keywords Distributed decision, constraint propagation, negotiation, decision support. Introduction|
|Supply, Installation, Testing, Integrating and commissioning of Boundary layer wind profiler at igi airport –as part of strengthening of Instrumentations for suitable completion of the project of Aviation Weather Decision Support System (awdss)||14 General observations 14 The “London Lens” 20 Case study 1: Innovative London ‘high streets’ 21 Case study 2: Tesco in London 25 Unresolved questions 27|
|Chapter 9 e-business Decision Support Learning Objectives||Decision Support and Expert Systems (mis 538)|
|Enabling Decision Support and Costing of Product Designs by using Visual Metaphors||Representing and using legal knowledge in integrated decision support systems: DataLex WorkStations|