- Email: [email protected]
A computerized cash concentration system
034EG.4 Th~ [nt Jl of M~mtS,:l. Vol. 8. No .1. pp. 45q to 44~
0305-Oa83,80/0701-0~5q$02.(]OiO
C Pergamon Press Lid 1980. Printed in Great Britain
A Computerized Cash Concentration System M ANVARI Concordia University, Canada
N MOHAN Standard Oil Company, Ohio, USA
(Received April 1979: in reL'ised form October 1979) The Standard Oil Company of Ohio generates revenues at a large number of geographically dispersed retail outlets. These funds are initially deposited with local banks and are subsequently transferred to the Company's major bank accounts in Cleveland. This paper reports on a computerized system developed to introduce efficiency in this transfer pcocess. The core of this new system is a modified inventory model which is used to issue transfer orders. The model captures the essential features of the process and at the same time is simple enough to be used in toni|ruction with a large commercial data processing system that contains the nece~ary data. The new system is a decision support system which allows management to intervene in the procedure for determining parameters of the decision model.
INTRODUCTION THE Standard Oil Company of Ohio (SOHIO) operates directly or indirectly, a large number of revenue generating outlets in the East and the Midwest. The managers of these outlets deposit their receipts in the company's local bank accounts from where they are transferred to central bank accounts in Cleveland. In 1976 a preliminary study of the company's cash management practices revealed that as a result of an inefficient method of cash concentration large quantities of idle funds were on deposit with the local banks. Subsequently, company management autfiorized a project to design and implement a new system for cash concentration. This paper deals with a description of this new system.
revenues and is therefore subject to the usual fluctuations in sales levels. The Company transfers funds from the local bank accounts by issuing special checks called Depository Transfer Checks (DTC) and depositing them with the major banks. The resulting concentration system is depicted in Fig. 1. It should be apparent that if the company were to use, instead of DTC, a different instrument for transferring funds, e.g. wire and/or a more efficient means of transmitting deposit information, e.g. telephone, then the characteristics of the resulting concentration system would be different. Available cash concentration systems have been discussed widely in the literature [2, 4].
D E V E L O P M E N T OF NEW DECISION MODEL
DESCRIPTION OF THE CASH CONCENTRATION SYSTEM
Notification of daily deposits received from Thousands of company agents deposit their the stations form part of the daily input to a daily receipts in over 500 local bank accounts multi-purpose computer system (CASHCOL) and forward notifications, through US mail, to and are available. the Head Office in Cleveland. The amount of The decision problem faced by management cash deposited by each agent reflects his daily is essentially an inventory type problem: when 459
460
Anvari, Mohan--A Computerized Cash Concentration System
Moil
Cash
I ad o.ice I
Prepare DTC
Local Bank t
[ Cleveland
l
tollect -
I
si,
Sank Clearing i-~ House
---I Sohio's Major Ranks
FIG.
to write a check against a local bank and for how much. This decision has to be made based on the information available at the time, which as a result of check clearing and mail delays is not accurate (see Fig. 2). The 'current' decision model programmed as part of CASHCOL works as follows: IfBB < D D F If B B > D D F If Z < $1,000
Z = 0 Z = BB -
DDF
Z = 0
where = bank balance as shown by company records (Book Balance) D D F = Draw Down Factor supplied by management Z = check size BB
The DDFs are supplied for each local bank based on past experience and in view of the large number of banks involved are rarely revised. The 'current' model did provide a solution to the decision problem, but was not leading to satisfactory results. A new decision model was thus to be found which, it was agreed, should satisfy two criteria: (1) it should be simple enough to operationalize and (2) it should be more 'efficient' than the current model. The first constraint was necessary, given that the large number of banks involved made the use of the computer mandatory, and thereafter the decision model had to be amenable to efficient
1. computerization. The second criterion, although seemingly self-evident, requires explanation. One way to compare the efficiency of two models is, to compare the total costs of the two models. In the case at hand, the total cost is the sum of (I) the opportunity cost of idle funds, (2) cost of checks issued by the decision center, and (3) costs associated with 'red balances'. An alternative way of comparing two decision models is to see which one can better achieve a certain 'balance goal'. Since, traditionally, 'payment' for banking services can take the form of leaving a certain minimum--or average--balance in the account, a decision model that can achieve a predetermined balance level while minimizing the number of orders issued can be considered efficient. (For a detailed discussion of the rationale for this criterion, the various forms that it assumes and the impact of changes in banking industry on its applicability see [I].) Two new decision models were then developed reflecting the two ways of viewing efficiency described above.
Deposit Data Mailed I T, Moil Log
Deposit Info Received / DTC Issued Tz
"Ignorance Period" FIG. 2.
DTC Arrives
I T3
461
Omega. Vol. & No. 4
The first 2 months of data for each bank were used to estimate the parameters of each model which were then used to issue withdrawals based on the remaining 10 months of deposit data. It was contended that historically used DDFs were not appropriate and if the current model were to be used as a benchmark for comparison, 'good' factors should be found and used in the simulation. Whereas for the inventory model the estimation procedure was quite straight forward through the use of an optimization routine, determination of the parameters of the 'goal attaining' (GA) and the +current' models were essentially a trial and error procedure. The simulation model provided three main measures of performance for each model: (1) Average daily balance (2) number of 'bounces' and (3) total cost. Extensive Monte Carlo experiments were performed which clearly indicated that the differences observed between performance measures of various models were stable and significant [1]. Tables 2, 3 and 4 summarize the results for the simulation run in which the 'actual' deposits were used. Table 2 summarizes the cost related Eli.) measures of performance. The inventory model PILOT STUDY led to lower costs and higher number of To compare the performance of the three bounces when compared with the current models a simulation program was developed. A model for all the six banks in the sample. data base was prepared which contained 12 Given that the actual dollar penalty of a months deposit histories for a number of repre- bounce, if any, is very small, it can be seen that sentative banks to be used in the study. The the inventory model is superior to the current inadequacy of the existing information systems model by as much as 269/0. Except for Bank 5 left the monthly bank statements as the only the (GA) model also comtSares favorably with source of accurate deposit information and the the current model. The difference between the sheer amount of work excluded the possibility overall performance of the (GA) and inventory of including all the local banks in the study. models, although discernible, is small (6.2~). Table 3 presents the balance oriented The banks in the sample represented a variety of possible situations in terms of amount of measures of performance. Note that the total daily receipts, number of depositing agents and 'required balance', i.e. the compensating location of the bank (See Table 1). balance determined by management for cover-
The first model was a theoretical inventory model designed to minimize the total cost of the system. It was found that with some reasonable assumptions, the ordering policy for the system is an (s, S) policy [3]. Furthermore, it is possible to compute a number of (s, S) pairs corresponding to different probabilities of 'bounced' checks. This is a desirable feature because, depending on the nature of the relationship between a particular bank and the Company, it may be imperative to constrain the number of 'bounces'. A second decision model was designed in congruence with the criterion of maintaining a certain balance at the local banks. It is an ad hoc myopic model which works as follows. To write today's check, first compute the expected bank balance on the day when today's check arrives at the local bank by forecasting the deposits during the 'ignorance period'. Next, compute the cummulative deviations to date from the 'balance goal'. Prepare today's checks by comparing these figures while making sure that small checks are avoided. (For a rigorous mathematical formulation of this model see
TABLE I. SUMMARY OF SAMPLE BANK CHARACTERISTICS
O.m.~. 8 4--F
Bank
Average daily deposit ($)
Standard deviation of deposits ($)
Number of agents depositing
Average mail lag (days)
Average check clearance lag (days)
1 2 3 4 5 6
526 3074 1077 5666 38,768 16,578
672 3246 1031 3246 35,768 15,748
1 I 5 11 42 53
1 3 2 1 1 1
2 4 2 2 2 2
462
AnvarL Mohan---,4 Computerized Cash Concentration System TABLE 2. COST RELATED MEASURESOF PERFORMANCE
Current model Bank Bank Bank Bank Bank Bank Bank
1 2 3 4 5 6
Total
(GA) Model Annual cost ($)
Inventory model
Annual cost ($)
N u m b e r of bounces
N u m b e r of bounces
Annual cost ($)
N u m b e r of bounces
462.43 254.02 749.76 1,243.96 6,399.60 3,653.09
----1 --
300.02 165.60 462.24 1,027.19 6,551.74 2,381.23
2 3 2 1 2 --
278.14 108.64 499.20 835.06 5,897.52 2.577.35
1 3 2 1 2 --
12,762.86
1
10,905.52
I0
10,268.84
9
Note: Cost figures are quoted to the nearest cent for comparative purposes only.
ing costs of banking services, is greatly exceeded for all three models. This is quite reasonable if one notes that the total required balance is less than half of the average daily deposit and thus in view of the time lags of the concentration system it can hardly be achieved regardless of the particular decision model used. Relatively however, the inventory model consistently out performs the current model although it leads to more bounces. The same is true for the (GA) model except for Bank 5.
Again, the average daily balances for the inventory and the (GA) models are comparable, with a slightly lower overall figure for the former. It is interesting to note that the inventory model which did not attempt to achieve any particular balance goal led to virtually the same average daily balance as the (GA) model. Table 4 summarizes the percentage differences in average balance and costs for the three models for each of the six banks. It should be emphasized that the performance measures of
TABLE 3. BALANCEORIENTED MEASURESOF PERFORMANCE
Current Model
Bank Bank Bank Bank Bank Bank Bank
1 2 3 4 5 6
Total
(GA} Model
Inventory Model
Average daily
Number of
Average daily
Number
balance ($)
Number of bounces
balance ($)
bounces
balance ($)
bounces
1233 751 1671 3760 8921 14,178
7383 4068 12,008 19,931 106,010 60,903
----1 --
4797 2420 6869 16,320 108,544 39,411
2 3 2 1 2 --
4409 1483 7487 13,046 97,993 42,733
1 3 2 1 2 --
30,514
210,303
1
178,361
I0
167,106
9
Average daily
Required t
Average daily
deposit ($)
balance ($)
1077 526 3074 5666 38,768 16,578 65,689
of
t Supplied by operating management.
TABLE 4. COMPARISONOF PERFORMANCEMEASURES
Bank Bank I Bank 2 Bank 3 Ban k 4 Bank 5 Bank 6 All banks
(GA) Model
Inventory model
Inventory model
Percentage improvement
Percentage improvement over current model Average Total balance cost
Percentage improvement
over current model Average Total balance cost 35.0 40.5 42.8 18.1 -2.4 35.3 17.9
31.3 34.6 38.3 17.4 -2.3 34.8 17.0
40.3 67.7 37.6 34.5 7.6 29.8 25.8
24.1 57.2 33.4 32.8 7.8 29.4 24.3
over (GA) model Total balance cost
Average 8.0 40.6 -8.2 20.0 9.7 -7.8 6.7
7.4 34.4 -7.4 18.7 10.0 -7.6 6.2
Omega. Vol. & No. 4
the current models are based on a 'good" set of DDFs and that the 'actual' performance for these banks during the period was considerably worse. As is evident from Table 4 the differences between the (GA) and inventory models were minimal. Nonetheless, the inventory model was selected over the (GA) model for implementation essentially based on operational factors. IMPLEMENTATION The (s, S) model was to be implemented with the view that minimum changes were to be introduced in CASHCOL. Consequently, the only modifications made were: (1) the system was ctianged to accept two parameters, s-High and s-Low instead of the DDF, and (2) the check writing routine was modified to use the new decision rule. All other changes, including the procedure for determining the two parameters were relegated to a :new system which was called the Transfer Bank System (TABS). The overall relationship between these two computer systems is depicted in Fig. 3. TABS was designed to prepare a series of s-High, s-Low) parameters instead of just one. Each set corresponds to a different expected level of bounced DTCs and is reported to management along with expected performance measures. Operating management upon reviewing the report and taking into account the particular bank-company relationship
--'~
463
decide on the best trade-off by selecting the appropriate set of parameters. This set of values is then passed on to CASHCOL and forms the basis for issuing DTCs until the next revision. TABS consists of a handful of programs written in FORTRAN.The total execution time of the system is about 2 hr each month of which only about 40 rain is due to the optimization routine. The timing of execution of TABS, both as far as the day of the month and the time of the day are concerned is, of course, quite flexible and hence it is run during slack periods in the company owned computer center. Prior to going on line, the historical deposit files for the past 12 months for all the banks were prepared and detailed cost and lag files were created. Subsequently, one full year of operation was simulated month by month and the behavior of the system was closely monitored. It was found that if TABS were in operation during the year in question, the balances at the local banks would have been reduced by about $2,000,000 with fewer checks bouncing. Finally the new system went on line in January of 1978. Operating management exercised considerable caution during the initial period by selecting conservative bounce levels. Nevertheless, the balances were reduced by the expected amount, and further the system's predictions about balances and bounces that could result from a particular set of parameters were very accurate.
CASFICOL
nonths
posit I
Transfer Bank System ( TABS ) FIG. 3.
464
Anvari, Mohan--A Computerized Cash Concentration System
CONCLUSIONS In terms of costs, this project was completed for about $50,000. The benefits of the new system are many fold. The most tangible and immediate is, of course, the substantial reduction in idle balances. Also, the new system has eliminated the need for the cumbersome and time consuming process of manual monitoring and revision of the parameters of the decision model. The flexibility of the new system will allow incorporation of new technical development in funds transfer. For example, should at some time in the future it becomes desirable to use faster means of data transmission instead of US mail, the system can easily handle the resulting change. In effect, the similarity of the
procedures for employing various transfer systems ascribe more generality of this project than may have been implied here. REFERENCES 1. ANVARZM (1977)The Problem of Collecting Cash Sale Revenues from Geographically Dispersed Locations. Unpublished doctoral dissertation, Case Western Reserve University. Cleveland. Ohio, USA. 2. LORDANJF (1973) The Banking Side of Cash Management, Financial Publishing Company, Boston. Massachusetts, USA. 3. SCARFH (1960) The optimality of (s. S) policies in the dynamic inventory problem. In Mathematical Models in Social Sciences (Eds ARROW K, K^RLIN S & SUPPEs P). Stanford University Press, USA. 4. S~R8¥ FW (1968) Use your hidden cash resources. Harv. Busin. Rev. March/April, 71-80. Aaoness f o r CORRESPONDENCE:M Anvari, Esq, Department of Quantitative Methods, Concordia University, Montreal Quebec, Canada.
0305-Oa83,80/0701-0~5q$02.(]OiO
C Pergamon Press Lid 1980. Printed in Great Britain
A Computerized Cash Concentration System M ANVARI Concordia University, Canada
N MOHAN Standard Oil Company, Ohio, USA
(Received April 1979: in reL'ised form October 1979) The Standard Oil Company of Ohio generates revenues at a large number of geographically dispersed retail outlets. These funds are initially deposited with local banks and are subsequently transferred to the Company's major bank accounts in Cleveland. This paper reports on a computerized system developed to introduce efficiency in this transfer pcocess. The core of this new system is a modified inventory model which is used to issue transfer orders. The model captures the essential features of the process and at the same time is simple enough to be used in toni|ruction with a large commercial data processing system that contains the nece~ary data. The new system is a decision support system which allows management to intervene in the procedure for determining parameters of the decision model.
INTRODUCTION THE Standard Oil Company of Ohio (SOHIO) operates directly or indirectly, a large number of revenue generating outlets in the East and the Midwest. The managers of these outlets deposit their receipts in the company's local bank accounts from where they are transferred to central bank accounts in Cleveland. In 1976 a preliminary study of the company's cash management practices revealed that as a result of an inefficient method of cash concentration large quantities of idle funds were on deposit with the local banks. Subsequently, company management autfiorized a project to design and implement a new system for cash concentration. This paper deals with a description of this new system.
revenues and is therefore subject to the usual fluctuations in sales levels. The Company transfers funds from the local bank accounts by issuing special checks called Depository Transfer Checks (DTC) and depositing them with the major banks. The resulting concentration system is depicted in Fig. 1. It should be apparent that if the company were to use, instead of DTC, a different instrument for transferring funds, e.g. wire and/or a more efficient means of transmitting deposit information, e.g. telephone, then the characteristics of the resulting concentration system would be different. Available cash concentration systems have been discussed widely in the literature [2, 4].
D E V E L O P M E N T OF NEW DECISION MODEL
DESCRIPTION OF THE CASH CONCENTRATION SYSTEM
Notification of daily deposits received from Thousands of company agents deposit their the stations form part of the daily input to a daily receipts in over 500 local bank accounts multi-purpose computer system (CASHCOL) and forward notifications, through US mail, to and are available. the Head Office in Cleveland. The amount of The decision problem faced by management cash deposited by each agent reflects his daily is essentially an inventory type problem: when 459
460
Anvari, Mohan--A Computerized Cash Concentration System
Moil
Cash
I ad o.ice I
Prepare DTC
Local Bank t
[ Cleveland
l
tollect -
I
si,
Sank Clearing i-~ House
---I Sohio's Major Ranks
FIG.
to write a check against a local bank and for how much. This decision has to be made based on the information available at the time, which as a result of check clearing and mail delays is not accurate (see Fig. 2). The 'current' decision model programmed as part of CASHCOL works as follows: IfBB < D D F If B B > D D F If Z < $1,000
Z = 0 Z = BB -
DDF
Z = 0
where = bank balance as shown by company records (Book Balance) D D F = Draw Down Factor supplied by management Z = check size BB
The DDFs are supplied for each local bank based on past experience and in view of the large number of banks involved are rarely revised. The 'current' model did provide a solution to the decision problem, but was not leading to satisfactory results. A new decision model was thus to be found which, it was agreed, should satisfy two criteria: (1) it should be simple enough to operationalize and (2) it should be more 'efficient' than the current model. The first constraint was necessary, given that the large number of banks involved made the use of the computer mandatory, and thereafter the decision model had to be amenable to efficient
1. computerization. The second criterion, although seemingly self-evident, requires explanation. One way to compare the efficiency of two models is, to compare the total costs of the two models. In the case at hand, the total cost is the sum of (I) the opportunity cost of idle funds, (2) cost of checks issued by the decision center, and (3) costs associated with 'red balances'. An alternative way of comparing two decision models is to see which one can better achieve a certain 'balance goal'. Since, traditionally, 'payment' for banking services can take the form of leaving a certain minimum--or average--balance in the account, a decision model that can achieve a predetermined balance level while minimizing the number of orders issued can be considered efficient. (For a detailed discussion of the rationale for this criterion, the various forms that it assumes and the impact of changes in banking industry on its applicability see [I].) Two new decision models were then developed reflecting the two ways of viewing efficiency described above.
Deposit Data Mailed I T, Moil Log
Deposit Info Received / DTC Issued Tz
"Ignorance Period" FIG. 2.
DTC Arrives
I T3
461
Omega. Vol. & No. 4
The first 2 months of data for each bank were used to estimate the parameters of each model which were then used to issue withdrawals based on the remaining 10 months of deposit data. It was contended that historically used DDFs were not appropriate and if the current model were to be used as a benchmark for comparison, 'good' factors should be found and used in the simulation. Whereas for the inventory model the estimation procedure was quite straight forward through the use of an optimization routine, determination of the parameters of the 'goal attaining' (GA) and the +current' models were essentially a trial and error procedure. The simulation model provided three main measures of performance for each model: (1) Average daily balance (2) number of 'bounces' and (3) total cost. Extensive Monte Carlo experiments were performed which clearly indicated that the differences observed between performance measures of various models were stable and significant [1]. Tables 2, 3 and 4 summarize the results for the simulation run in which the 'actual' deposits were used. Table 2 summarizes the cost related Eli.) measures of performance. The inventory model PILOT STUDY led to lower costs and higher number of To compare the performance of the three bounces when compared with the current models a simulation program was developed. A model for all the six banks in the sample. data base was prepared which contained 12 Given that the actual dollar penalty of a months deposit histories for a number of repre- bounce, if any, is very small, it can be seen that sentative banks to be used in the study. The the inventory model is superior to the current inadequacy of the existing information systems model by as much as 269/0. Except for Bank 5 left the monthly bank statements as the only the (GA) model also comtSares favorably with source of accurate deposit information and the the current model. The difference between the sheer amount of work excluded the possibility overall performance of the (GA) and inventory of including all the local banks in the study. models, although discernible, is small (6.2~). Table 3 presents the balance oriented The banks in the sample represented a variety of possible situations in terms of amount of measures of performance. Note that the total daily receipts, number of depositing agents and 'required balance', i.e. the compensating location of the bank (See Table 1). balance determined by management for cover-
The first model was a theoretical inventory model designed to minimize the total cost of the system. It was found that with some reasonable assumptions, the ordering policy for the system is an (s, S) policy [3]. Furthermore, it is possible to compute a number of (s, S) pairs corresponding to different probabilities of 'bounced' checks. This is a desirable feature because, depending on the nature of the relationship between a particular bank and the Company, it may be imperative to constrain the number of 'bounces'. A second decision model was designed in congruence with the criterion of maintaining a certain balance at the local banks. It is an ad hoc myopic model which works as follows. To write today's check, first compute the expected bank balance on the day when today's check arrives at the local bank by forecasting the deposits during the 'ignorance period'. Next, compute the cummulative deviations to date from the 'balance goal'. Prepare today's checks by comparing these figures while making sure that small checks are avoided. (For a rigorous mathematical formulation of this model see
TABLE I. SUMMARY OF SAMPLE BANK CHARACTERISTICS
O.m.~. 8 4--F
Bank
Average daily deposit ($)
Standard deviation of deposits ($)
Number of agents depositing
Average mail lag (days)
Average check clearance lag (days)
1 2 3 4 5 6
526 3074 1077 5666 38,768 16,578
672 3246 1031 3246 35,768 15,748
1 I 5 11 42 53
1 3 2 1 1 1
2 4 2 2 2 2
462
AnvarL Mohan---,4 Computerized Cash Concentration System TABLE 2. COST RELATED MEASURESOF PERFORMANCE
Current model Bank Bank Bank Bank Bank Bank Bank
1 2 3 4 5 6
Total
(GA) Model Annual cost ($)
Inventory model
Annual cost ($)
N u m b e r of bounces
N u m b e r of bounces
Annual cost ($)
N u m b e r of bounces
462.43 254.02 749.76 1,243.96 6,399.60 3,653.09
----1 --
300.02 165.60 462.24 1,027.19 6,551.74 2,381.23
2 3 2 1 2 --
278.14 108.64 499.20 835.06 5,897.52 2.577.35
1 3 2 1 2 --
12,762.86
1
10,905.52
I0
10,268.84
9
Note: Cost figures are quoted to the nearest cent for comparative purposes only.
ing costs of banking services, is greatly exceeded for all three models. This is quite reasonable if one notes that the total required balance is less than half of the average daily deposit and thus in view of the time lags of the concentration system it can hardly be achieved regardless of the particular decision model used. Relatively however, the inventory model consistently out performs the current model although it leads to more bounces. The same is true for the (GA) model except for Bank 5.
Again, the average daily balances for the inventory and the (GA) models are comparable, with a slightly lower overall figure for the former. It is interesting to note that the inventory model which did not attempt to achieve any particular balance goal led to virtually the same average daily balance as the (GA) model. Table 4 summarizes the percentage differences in average balance and costs for the three models for each of the six banks. It should be emphasized that the performance measures of
TABLE 3. BALANCEORIENTED MEASURESOF PERFORMANCE
Current Model
Bank Bank Bank Bank Bank Bank Bank
1 2 3 4 5 6
Total
(GA} Model
Inventory Model
Average daily
Number of
Average daily
Number
balance ($)
Number of bounces
balance ($)
bounces
balance ($)
bounces
1233 751 1671 3760 8921 14,178
7383 4068 12,008 19,931 106,010 60,903
----1 --
4797 2420 6869 16,320 108,544 39,411
2 3 2 1 2 --
4409 1483 7487 13,046 97,993 42,733
1 3 2 1 2 --
30,514
210,303
1
178,361
I0
167,106
9
Average daily
Required t
Average daily
deposit ($)
balance ($)
1077 526 3074 5666 38,768 16,578 65,689
of
t Supplied by operating management.
TABLE 4. COMPARISONOF PERFORMANCEMEASURES
Bank Bank I Bank 2 Bank 3 Ban k 4 Bank 5 Bank 6 All banks
(GA) Model
Inventory model
Inventory model
Percentage improvement
Percentage improvement over current model Average Total balance cost
Percentage improvement
over current model Average Total balance cost 35.0 40.5 42.8 18.1 -2.4 35.3 17.9
31.3 34.6 38.3 17.4 -2.3 34.8 17.0
40.3 67.7 37.6 34.5 7.6 29.8 25.8
24.1 57.2 33.4 32.8 7.8 29.4 24.3
over (GA) model Total balance cost
Average 8.0 40.6 -8.2 20.0 9.7 -7.8 6.7
7.4 34.4 -7.4 18.7 10.0 -7.6 6.2
Omega. Vol. & No. 4
the current models are based on a 'good" set of DDFs and that the 'actual' performance for these banks during the period was considerably worse. As is evident from Table 4 the differences between the (GA) and inventory models were minimal. Nonetheless, the inventory model was selected over the (GA) model for implementation essentially based on operational factors. IMPLEMENTATION The (s, S) model was to be implemented with the view that minimum changes were to be introduced in CASHCOL. Consequently, the only modifications made were: (1) the system was ctianged to accept two parameters, s-High and s-Low instead of the DDF, and (2) the check writing routine was modified to use the new decision rule. All other changes, including the procedure for determining the two parameters were relegated to a :new system which was called the Transfer Bank System (TABS). The overall relationship between these two computer systems is depicted in Fig. 3. TABS was designed to prepare a series of s-High, s-Low) parameters instead of just one. Each set corresponds to a different expected level of bounced DTCs and is reported to management along with expected performance measures. Operating management upon reviewing the report and taking into account the particular bank-company relationship
--'~
463
decide on the best trade-off by selecting the appropriate set of parameters. This set of values is then passed on to CASHCOL and forms the basis for issuing DTCs until the next revision. TABS consists of a handful of programs written in FORTRAN.The total execution time of the system is about 2 hr each month of which only about 40 rain is due to the optimization routine. The timing of execution of TABS, both as far as the day of the month and the time of the day are concerned is, of course, quite flexible and hence it is run during slack periods in the company owned computer center. Prior to going on line, the historical deposit files for the past 12 months for all the banks were prepared and detailed cost and lag files were created. Subsequently, one full year of operation was simulated month by month and the behavior of the system was closely monitored. It was found that if TABS were in operation during the year in question, the balances at the local banks would have been reduced by about $2,000,000 with fewer checks bouncing. Finally the new system went on line in January of 1978. Operating management exercised considerable caution during the initial period by selecting conservative bounce levels. Nevertheless, the balances were reduced by the expected amount, and further the system's predictions about balances and bounces that could result from a particular set of parameters were very accurate.
CASFICOL
nonths
posit I
Transfer Bank System ( TABS ) FIG. 3.
464
Anvari, Mohan--A Computerized Cash Concentration System
CONCLUSIONS In terms of costs, this project was completed for about $50,000. The benefits of the new system are many fold. The most tangible and immediate is, of course, the substantial reduction in idle balances. Also, the new system has eliminated the need for the cumbersome and time consuming process of manual monitoring and revision of the parameters of the decision model. The flexibility of the new system will allow incorporation of new technical development in funds transfer. For example, should at some time in the future it becomes desirable to use faster means of data transmission instead of US mail, the system can easily handle the resulting change. In effect, the similarity of the
procedures for employing various transfer systems ascribe more generality of this project than may have been implied here. REFERENCES 1. ANVARZM (1977)The Problem of Collecting Cash Sale Revenues from Geographically Dispersed Locations. Unpublished doctoral dissertation, Case Western Reserve University. Cleveland. Ohio, USA. 2. LORDANJF (1973) The Banking Side of Cash Management, Financial Publishing Company, Boston. Massachusetts, USA. 3. SCARFH (1960) The optimality of (s. S) policies in the dynamic inventory problem. In Mathematical Models in Social Sciences (Eds ARROW K, K^RLIN S & SUPPEs P). Stanford University Press, USA. 4. S~R8¥ FW (1968) Use your hidden cash resources. Harv. Busin. Rev. March/April, 71-80. Aaoness f o r CORRESPONDENCE:M Anvari, Esq, Department of Quantitative Methods, Concordia University, Montreal Quebec, Canada.