RESELLER ASSORTMENT DECISION CRITERIA, Jerker Nilsson
Contents
1. 640K 256K 640K 192K Coprocessor support Yes Yes Yes Forecasting step identification Model diagnosis Plotting series Raw series Yes Yes Yes Yes Smoothed series Yes Yes Yes Yes Box plots No Yes No Yes Autocorrelations Yes Yes Yes Yes Cross correlations No Yes Yes No Preprocessing Lead and lags Yes Yes Yes Yes Dummy variables No Yes 5 Yes Differencing Simple Yes Yes Yes Yes Seasonal Yes Yes Yes Yes Log transform Yes Yes Yes Yes Component decomposition Yes Yes Yes Yes Estimation Univariate Curve fitting Yes Yes Yes Linear Exponential smoothing Simple Yes Yes Yes Yes 2 parameter Holt Yes Yes Yes Yes 3 parameter Winter Yes Yes Yes Yes Box Jenkins Yes No Yes Yes Multivariate Spectral analysis No No Bivariate No Regression OLS Yes Yes Yes Yes WLS 7 No Yes Yes Two stage No No Yes No State space models Yes No No No Vector autoregression Yes No No No Dummy variables are created in SPSS by using the COMPUTE command The FP package does offer Harrison s 1967 harmonic smoothing FM accommodates hetereoscedasticity by using the ARCH model Engle 1982 A no in the table means only that the package does not include a specific command to implement the step For example with a package that lacks a specific com mand to plot a smoothed series one usually can create the smoothed series save it as a variable and then plot the new variable against time Likewise with a package that has OLS multiple regression one c2. first be transformed to stationarity before aggre gation If they are not a forecaster may be tempted to apply the automatically determined optimal procedure for the total to a component for which it is patently in appropriate The forecasting tools offered by SFII are at the low end of the sophistication scale This level may be con sistent with its positioning which seems to be for series 131 having relatively few cases This package has the po tential of requiring the least mental and physical labor from a forecaster It is particularly suited for rapidly and efficiently analyzing large numbers of relatively short less than 150 cases series The package is less attractive for forecasters who want full understanding and control of a sophisticated analysis of a possibly complex series Forecast Plus Separate versions of the package are available for IBM PC XT AT machines that do and do not have math co processors It is not copy protected Data are stored in comma delimited ASCII files though DIF files can be read and written Either single or multiple variable files can be created Multiple variable files can contain up to 30 variables Year month day or numbered labels must be spec ified for each variable Missing values are estimated by linear interpolation All of the functions and arithmetic logical operators of BASIC are available by using an SPSS like command structure i e SELECT RE CODE COMPUTE e
3. litera ture Chapter 2 is continued in Appendix B where 34 studies containing assortment decision criteria lists are reviewed in detail Chapters 1 and 2 plus the contin uation of Chapter 2 in Appendix B and the Bibliography fill 87 of the book s 176 pages The theoretical discus sion and literature review alone are worth the purchase price to any academician or graduate student researching reseller assortment decision criteria Chapter 3 Empirical Bases is a description of the data used in the study The data were collected from a very large vertically integrated organization of Swedish con sumer cooperatives in 1975 The cooperative organiza tion included more than 2300 retail outlets of several dif ferent types with a total assortment of more than 5000 items The latter part of Chapter 3 includes the data col lection instruments and procedures as well as the vari able specifications Chapter 4 is the Statistical Analysis chapter Multiple logit regression analysis was used to analyze the data because of the binary nature of depen dent variables This chapter includes a brief discussion of the logit model selection of the explanatory vari ables and goodness of fit measures A more detailed matrix algebra description of binary choice models is contained in Appendix A Chapter 5 Findings de scribes the results of the data analysis in terms of both the seven decision categories and the 14 decision criteria studied Conclusio
4. than it is for most multiple package programs A characteristic of this package is that numerous small files are likely to proliferate First all series initially are entered as univariate series stored in ASCII files a max imum of 2000 observations can be included in a single series Associated with each of these series is a KEY file that names the series and describes its calendar char acteristics such as number of periods starting date and JOURNAL OF MARKETING RESEARCH FEBRUARY 1989 periods per year Second each analysis must be pre ceded by a step in which the variables to be analyzed are selected Leads and lags of variables are specified at this stage as are created variables a constant and some user specified transformations This specification is saved in a file that is accessed subsequently by the selected analytic program Academic forecasters are likely to want to try different transformations lags variables and so on and probably would create a file for each analysis Finally the user has the option of creating an audit trail of all the steps in a forecasting session This trail is stored in a file and can be used to reconstruct all the steps that were performed to create a forecast It is easy to see that an active forecaster especially an academician could create a plethora of small files on the storage disk The program requires 512K of memory More mem ory is preferable because the package relies on ex
5. NEW BOOKS IN REVIEW RESELLER ASSORTMENT DECISION CRITERIA Jerker Nilsson and Viggo Host Aarhus Denmark Aarhus University Press 1987 176 pages This softcover book contains an extremely detailed study of the relative importance of criteria used in reseller as sortment decisions As the authors state in the Preface The literature on reseller assortment decision criteria consists mainly of descriptions of what criteria have been used in single reseller companies on certain occa sions What is lacking are explanations of how the re sellers criteria are related to the various situational fac tors Their study seeks to remedy these shortcomings by focusing on situational factors that can affect the use of certain classes of decision criteria The chapters are logically laid out with a preceding outline or introduction to each and an alphanumeric paragraph numbering sys tem A Summary chapter is provided at the beginning for readers not willing to work their way through the following five chapters The study is written in a journal article style that probably will not appeal to practition ers Chapter 1 Problem is a detailed dissertation quality theoretical introduction to assortment decisions The per spective is primarily that of the reseller but all channel levels are discussed Chapter 2 Literature Survey con tinues the theoretical discussion with an extremely thor ough review of the reseller assortment decision
6. an achieve two stage least squares by performing the steps separately Overall SMII and FP are probably the best choices for the nonspecialist practitioner who wants to make short term forecasts in an operating environment For the aca demician who is familiar with SPSS especially at insti tutions where the package is used for instruction the Trends package and manual could be used in a fore casting class For the academician or practitioner who wants to do relatively sophisticated multivariate analysis multiple dependent variables FM is the choice Stan dard datasets e g Lydia Pinkham were analyzed by using the packages The results agreed with published results and the programs were sufficiently user friendly that a user would be indifferent between doing the anal ysis on a PC or a mainframe computer However none of the manuals can stand alone as introduction to time series analysis nor do they provide particularly well or ganized references to guide a nonspecialist user SmartForecasts Il On an IBM PC XT AT computer this package re quires 256K of memory DOS 2 0 or higher and two 5 25 inch drives or one drive and a hard disk Seven hundred and twenty kilobytes of hard disk space is needed to store the programs For the IBM System 2 two 3 5 drives or one drive and a hard disk are required The program is not copy protected and an installation batch file is provided No explicit reference is made to copro cess
7. ark CA Sage Publications Inc 1988 133 Kelly Janice R and Joseph E McGrath On Time and Method Newbury Park CA Sage Publications Inc 1988 Kimmel Allan J Ethics and Values in Applied Social Research Newbury Park CA Sage Publications Inc 1988 Krueger Richard A Focus Groups A Practical Guide for Applied Research Newbury Park CA Sage Publications Inc 1988 Maynes E Scott ed The Frontier of Research in the Consumer Interest Columbia MO American Council on Consumer Interests 1988 McCracken Grant Culture and Consumption Bloomington IN Indiana University Press 1988 Wittink Dick R The Application of Regression Analysis Needham Heights MA Allyn and Bacon Inc 1988 Software 1 2 3 Version 2 01 Lotus Development Corporation Cambridge MA 1986 MARKUPS Version 1 1 McGraw Hill Book Company New York 1988 Reprint No JMR261110 J of Marketing i URNAL OF MARKETING Published in January April July and October this journal contains articles furnishing information on marketing discov eries techniques trends and new ideas that contribute substantive findings and demonstrate workable solutions to today s marketing problems Journal of Marketing is available for individual as well as corporate library school business etc use Rates subject to change AMA Members Nonmembers Corporate Journal of Marketing 28 56 90 Copies may be purchased on an individual basi
8. at Statistical Model Iden tification JEEE Transactions in Automatic Control AC 19 6 716 23 1976 Canonical Correlation Analysis and Infor mation Criterion in System Identification Advances and Case Studies Mehra and Lainiotis eds New York Aca demic Press Inc Box G E P and G M Jenkens 1976 Time Series Anal ysis Forecasting and Control San Francisco Holden Day Engle R F 1982 Autoregressive Conditional Heterosce dasticity with Estimates of the Variance of United Kingdom Inflation Econometrica 50 987 1007 NEW BOOKS IN REVIEW Harrison P J 1967 Exponential Smoothing and Short Term Sales Forecasting Management Science 13 11 821 42 Litterman R B 1981 RATS User s Manual Version 4 1 Minneapolis VAR Econometrics Rosenberg B 1973 The Analysis of a Cross Section of Time Series by Stochastically Convergent Parameter Regression Annals of Economic and Social Measurement 2 399 428 Velleman P F and D C Hoagling 1981 Applications Basics and Computing of Exploratory Data Analysis Bos ton Duxbury Press BOOKS AND SOFTWARE RECEIVED Books Devinney Timothy M ed Issues in Pricing Theory and Research Lexington MA Lexington Books 1988 Ehrenberg A S C Repeat Buying 2nd ed New York Oxford University Press 1988 Grady Kathleen E and Barbara Strudler Wallston Research in Health Care Settings Newbury P
9. d multiple regression However the last does not provide procedures for handling auto correlation and heteroscedasticity such as Cochrane Or cutt or WLS SPSS PC Trends Trends for SPSS PC is a five disk add on for the SPSS PC statistical series It can be used as a stand alone package but that would entail learning the SPSS command language structure in order to use the Trend 132 package The user may also want to consider a version of Microsoft CHART as an add on to get high quality graphics The set will require 10 megabytes of hard disk space a coprocessor and 640K of RAM For this over head the forecaster gets a comprehensive univariate forecasting package that gives considerable control over the application of sophisticated estimation procedures For example the ARIMA command uses a maximum likelihood procedure to model a dependent series with or without the inclusion of independent regressors The procedure can accommodate missing data in the series Trend includes a missing data estimation command that estimates missing values by using linear trend the mean or median of surrounding values the mean or median of the series or linear trend at the point The CREATE command will create series that are a function of existing series including smoothed series based on the Fast Four ier Transform moving averages and the T4253H func tion Velleman and Hoagling 1981 The NPPLOT com mand displays normal probability plo
10. enkins 1976 identification estimation and diagnosis Identification is applied to the historical series with the objective of tentatively determining an appropriate model Estimation refers to the task of es timating model parameters Diagnosis is applied to the series of residuals with the aim of determining the ap propriateness of the tentative model The process is it erated as necessary The capabilities of the packages in these respective areas are reported in Table 1 In this table the term di agnosis differs somewhat from the its usage by Box and Jenkins It refers to a set of tools such as time plots or autocovariance function plots that can be used to determine the sorts of models that might be appropriate for a series Preprocessing refers to transformations that might be applied to a series to make it amenable to a particular estimation technique For example prepro cessing might be done to a nonstationary series to make it stationary Trends seasonality and varying autocor relations result in nonstationary series Many useful tools and techniques such as the autocorrelation and cross correlation functions simple exponential smoothing and ARIMA models assume that the series is stationary i e the mean and variance of the series are independent of time 130 JOURNAL OF MARKETING RESEARCH FEBRUARY 1989 Table 1 OVERVIEW OF FORECASTING PACKAGES FM SMII Trends FP Requirements RAM 512
11. ns are presented at the end of this chapter This study is very well done but it does not seem to be aimed at any particular market other than research 129 academicians and graduate students interested in reseller assortment decisions The chief value of this book will be as a literature review and as an excellent example of how to organize and conduct dissertation level research on a very large complex database The study represents an application of one particular method to one very fo cused marketing management decision area The empir ical results are probably not generalizable to U S mar kets given the social and economic system from which the data were collected This lack of generalizability and the authors dissertation style of writing will probably preclude the book from appealing to practitioner mar kets GEORGE C HOZIER JR University of New Mexico SMARTFORECASTS II Smart Software Inc Bel mont MA 02178 FORECAST PLUS Walonick Associates 6500 Nicollet Avenue South Minneapolis MN 55423 SPSS PC TRENDS SPSS Inc 444 N Michigan Avenue Chicago IL 60611 FORECAST MASTER Scientific Systems 35 Cam bridge Park Drive Cambridge MA 02140 FOUR TIME SERIES ANALYSIS PACKAGES Four packages for analyzing time series data on PCs are reviewed here SmartForecasts I SMID Forecast Plus FP SPSS PC Trends Trends and Forecast Master FM The discussion follows the steps suggested by Box and J
12. or support but it does run on a system with a co processor No mention is made of laser printer support in the manual or in the installation procedure This omis sion is somewhat surprising as the package has a strong graphics orientation NEW BOOKS IN REVIEW A single SFI datafile can accommodate up to 60 vari ables with up to 150 observations per variable Data en try can be from the keyboard using a spreadsheet for mat or data can be read from and written to files stored in SFI Lotus DIF or ASCII formats In the spread sheet format titles column variable labels and row case labels can be added to the datafile The spread sheet structure applies only to data entry however new variables are not created with cell formulas but by spec ifying an operation or a formula on a command line Useful operation commands include dummy variable creation primarily for seasons differencing percent age changes cumulative totals and leads lags of vari ables SFII s characterization of time series is limited to year quarter monthly etc structures it does not pro vide for nonuniform trading days traditional and special holidays and the like The screens are well designed and make excellent use of a color monitor The program expects a minimal re sponse to a command line and then it prompts for ad ditional information as required For example the com mand for regression analysis requires only that the dependent va
13. riable be specified The program then prompts for the independent variables their seasonality and so forth Context sensitive help screens are pro vided This package has two distinguishing characteristics At one extreme it provides the ability to forecast by simply eyeballing the data Thus the forecaster can view a plot of the historical data and the results of two naive forecasting rules a linear function of all data and no change from last These can be replaced with other naive rules such as constant percentage change In ad dition the forecaster can move the position of a forecast on the screen and a new set of forecasts will be calcu lated At the other extreme the forecaster can have the pro gram conduct a tournament among a set of procedures and determine the procedure that does the best job of fitting a set of historical data Though very convenient and relatively speedy the risks of capitalizing on chance with such data crunching can hardly be ignored This problem is not emphasized in the manual The program has the capability for aggregating up to 60 series and then automatically determining the best forecasting procedure for the total and its components This pro cedure would be handy for example for forecasting the sales of shoes of different sizes and widths or deposits of different kinds in bank branches However little guidance is given as to whether the component series should
14. s as well as on an annual subscription basis For subscription information contact The American Marketing Association e Subscriptions Department 250 S Wacker Dr Suite 200 Chicago IL 60606 5819 or call 312 648 0536 Copyright of Journal of Marketing Research JMR is the property of American Marketing Association and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder s express written permission However users may print download or email articles for individual use
15. tc The package offers relatively sophisticated trading day adjustment capability Nine U S A holidays are part of the default adjustment table Up to 24 holidays can be entered The user can specify whether the holiday occurs always on a fixed day such as July 4th or on a fixed day of the week such as Labor Day furthermore one can specify whether the adjacent day is also a nontrading day e g if a holiday occurs on a Saturday the previous Friday may be a nontrading day In addition to the diagnostic plots listed in Table 1 seasonal box plots and a spread versus level are avail able The latter has options for evaluating six variance stabilizing transformations on the fly FP has the least satisfactory screen graphics of the four packages pri marily because many of the graphs do not fit within a single screen and the user must scroll to see the entire graph In fact the default output for FP is the printer the only trouble encountered in installing the program was figuring out that the default option had to be changed to get any screen output at all If the printer has contin uous sheet paper large graphs covering several pages can be printed For large series the diagnostic detail these reveal is worth the time it takes for them to print FP provides an extensive set of smoothing algorithms including an adaptive filtering algorithm not listed in Ta ble 1 ARIMA models implemented in the standard Box Jenkins framework an
16. ternal editors to edit datafiles Though the DOS EDLIN pro gram will fit in 512K the manual says that memory problems have been encountered in using some editors or attempting to run some programs under the DOS shell With 640K I had no problem using WORD as the editor or running 1 2 3 under the DOS shell the shell allows one to exit the forecasting program run another program under DOS and then reenter the forecasting program at the point of exit Being able to exit to a spreadsheet program is handy because the data transformation ca pabilities of Forecast Master are limited to exponents logs and arithmetic operations They do not include trigonometric functions dummy variable creation log ical operators select if type operators and the like It was reasonably convenient to use Lotus 1 2 3 to create complex transformations The state space forecasting can accommodate up to 10 endogenous and exogenous variables to forecast multiple dependent variables simultaneously using a Kalman fil terlike recursion algorithm Akaike 1974 1976 The Bayesian vector autoregression is the Litterman 1981 VAR model which appears in the RATS package The FM package also includes a varying parameter regres sion program applicable to regression situations in which the coefficients are time varying either deterministically or stochastically Rosenberg 1973 JAMES B WILEY Temple University REFERENCES Akaike H 1974 A New Look
17. ts of the residual Most of the estimation procedures include subcommands for performing log differencing and other transforma tions on the fly Trend s manual has the most professional appearance and it is the only one that could stand double duty as a supplemental text in a course Each of the estimation tools in the package is discussed in the context of a prob lem For example data on the ozone layer are analyzed to demonstrate the handling of missing data and the use of WLS to analyze data in which observations are mea sured with equal accuracy Forecast Master Forecast Master is the only one of the four packages that has a substantial multivariate capability Multivari ate procedures are state space forecasting and Bayesian vector autoregression The package can work with or without a math coprocessor but it is desirable to have one Installation on a hard disk system consists of cre ating a subdirectory copying all of the disks to the sub directory running an INTSTALL BAT batch file and answering the questions that appear on the screen Com mands are written to existing config sys and auto exec bat files If problems arise in running other pack ages examine these files to make sure the inserted commands are not a problem Once installed the pro gram is invoked by running the MASTER BAT batch program The program can be run on a two drive or one drive hard disk system However a hard disk is even more desirable
Download Pdf Manuals
Related Search