Reaping benefits from management research: Lessons from the forecasting principles project, J. Scott...

2
Table 1 Performance statistics of the ‘‘efficient’’ and ‘‘inefficient’’ ANN tests Hidden nodes Type Phase RMS error MAPE error MdAPE error 11 Inefficient Training 0.028 60.07% 7.88% 11 Efficient Training 0.032 19.58% 3.66% 11 Inefficient Test 0.11 27.97% 31.58% 11 Efficient Test 0.08 21.63% 12.76% Reaping benefits from management research: Lessons from the forecasting principles project, J. Scott Armstrong and Ruth A. Pagell, 2003, Interfaces 33 (6), 89–111. I’m hard pressed to think of any other endeavor in academe resembling the ‘‘Forecasting Principles Project’’ described in Armstrong and Pagell (here- after, A and P). This project was specifically designed to evaluate all the useful knowledge in forecasting, and to distill it in the form of princi- ples. No small task! The project, begun in 1997, resulted in the publication of Armstrong’s (2001) Principles of Forecasting, and the creation of a web site, www.forecastingprinciples.com. Does A and P’s paper contain interesting, surpris- ing, and relevant findings? By my estimation, seve- ral. First, I was disturbed to find that only 3% of published articles on forecasting contained useful findings—variously defined as that which can help people improve their forecasting, and direct evidence concerning the principles found in Principles of Forecasting. Second, I was stunned to discover that invited (‘‘special treatment’’) papers were 20 times more likely than regularly submitted manuscripts to pro- duce useful results. Third, I was amazed that a total of 139 principles of forecasting had been identified in the summary chapter of Armstrong (2001). This is an astonishing figure in view of Armstrong and Schultz’s (1993) failure to locate any similar principles in nine market- ing textbooks. Nevertheless, this testifies to what might be accomplished within the academic commu- nity when there is deliberate, focused attention direct- ed toward the discovery of principles. Fourth, I was disappointed to read that forecasting textbooks do not contain many principles about fore- casting. This seems strange, especially since 139 principles have been outlined by Armstrong (2001). errors for our experiments in the Table 1 below. The reader may insert these values in Table 7 of our article, if they wish. For test data, all the error (fit) metrics are very consistent and favor the use of ‘‘efficient’’ data for training the ANN. For researchers, who may be interested in expanding the findings of our study, we have identified a few possible directions. We used the BCC-DEA model for computing the efficiencies of the units. Perhaps, a comparison between CCR-DEA based screening and BCC-DEA based screening and its impact on perfor- mance may be a good idea. We believe that CCR-DEA based screening may be a good approach, when linear structural representation of the forecasting model, in addition to monotonicity, needs to be preserved. An- other valuable contribution could be to compare how the method works with multiple-inputs and multiple- outputs. Our conjecture is that it will work well with multiple-inputs and multiple-outputs, but our conjec- ture is open to future tests. Parag C. Pendharkar School of Business Administration, Pennsylvania State University at Harrisburg, 777 West Harrisburg Pike, Middletown, PA 17057, United States E-mail address: [email protected]. Tel.: +1 717 948 6028; fax: +1 717 948 6456. James A. Rodger Department of MIS and Decision Sciences, Indiana University of Pennsylvania, 1011 South Drive, Indiana, PA 15705, United States E-mail address: [email protected]. Tel.: +1 724 357 5944. doi:10.1016/j.ijforecast.2004.09.003 Research on forecasting 740

Transcript of Reaping benefits from management research: Lessons from the forecasting principles project, J. Scott...

Page 1: Reaping benefits from management research: Lessons from the forecasting principles project, J. Scott Armstrong and Ruth A. Pagell, 2003, Interfaces 33 (6) 89–111

Table 1

Performance statistics of the ‘‘efficient’’ and ‘‘inefficient’’ ANN tests

Hidden nodes Type Phase RMS error MAPE error MdAPE error

11 Inefficient Training 0.028 60.07% 7.88%

11 Efficient Training 0.032 19.58% 3.66%

11 Inefficient Test 0.11 27.97% 31.58%

11 Efficient Test 0.08 21.63% 12.76%

Reaping benefits from management research: Lessons

from the forecasting principles project, J. Scott

Armstrong and Ruth A. Pagell, 2003, Interfaces 33

(6), 89–111.

I’m hard pressed to think of any other endeavor

in academe resembling the ‘‘Forecasting Principles

Project’’ described in Armstrong and Pagell (here-

after, A and P). This project was specifically

designed to evaluate all the useful knowledge in

forecasting, and to distill it in the form of princi-

ples. No small task! The project, begun in 1997,

resulted in the publication of Armstrong’s (2001)

Principles of Forecasting, and the creation of a web

site, www.forecastingprinciples.com.

Does A and P’s paper contain interesting, surpris-

ing, and relevant findings? By my estimation, seve-

ral. First, I was disturbed to find that only 3% of

published articles on forecasting contained useful

findings—variously defined as that which can help

people improve their forecasting, and direct evidence

concerning the principles found in Principles of

Forecasting.

Second, I was stunned to discover that invited

(‘‘special treatment’’) papers were 20 times more

likely than regularly submitted manuscripts to pro-

duce useful results.

Third, I was amazed that a total of 139 principles

of forecasting had been identified in the summary

chapter of Armstrong (2001). This is an astonishing

figure in view of Armstrong and Schultz’s (1993)

failure to locate any similar principles in nine market-

ing textbooks. Nevertheless, this testifies to what

might be accomplished within the academic commu-

nity when there is deliberate, focused attention direct-

ed toward the discovery of principles.

Fourth, I was disappointed to read that forecasting

textbooks do not contain many principles about fore-

casting. This seems strange, especially since 139

principles have been outlined by Armstrong (2001).

errors for our experiments in the Table 1 below. The

reader may insert these values in Table 7 of our article,

if they wish. For test data, all the error (fit) metrics are

very consistent and favor the use of ‘‘efficient’’ data for

training the ANN.

For researchers, whomay be interested in expanding

the findings of our study, we have identified a few

possible directions. We used the BCC-DEA model for

computing the efficiencies of the units. Perhaps, a

comparison between CCR-DEA based screening and

BCC-DEA based screening and its impact on perfor-

mance may be a good idea. We believe that CCR-DEA

based screening may be a good approach, when linear

structural representation of the forecasting model, in

addition to monotonicity, needs to be preserved. An-

other valuable contribution could be to compare how

the method works with multiple-inputs and multiple-

outputs. Our conjecture is that it will work well with

multiple-inputs and multiple-outputs, but our conjec-

ture is open to future tests.

Parag C. Pendharkar

School of Business Administration,

Pennsylvania State University at Harrisburg,

777 West Harrisburg Pike, Middletown, PA 17057,

United States

E-mail address: [email protected].

Tel.: +1 717 948 6028; fax: +1 717 948 6456.

James A. Rodger

Department of MIS and Decision Sciences,

Indiana University of Pennsylvania,

1011 South Drive, Indiana, PA 15705,

United States

E-mail address: [email protected].

Tel.: +1 724 357 5944.

doi:10.1016/j.ijforecast.2004.09.003

Research on forecasting740

Page 2: Reaping benefits from management research: Lessons from the forecasting principles project, J. Scott Armstrong and Ruth A. Pagell, 2003, Interfaces 33 (6) 89–111

Perhaps the future will see their incorporation into

textbooks. I would also like to know how many of

these 139 principles are of the ‘‘grounded’’ (defined by

A and P as having support beyond expert opinion)

variety.

Fifth, I was surprised to learn that only about 20%

of the 139 principles are currently used in software

packages. No doubt more of them will be included in

future packages.

A and P note that much academic work is of little

value to practitioners. Armstrong (2001) goes some

way toward correcting this. Moreover, by suggesting

what researchers, journal editors, textbook writers,

software developers, web site designers, and practi-

tioners can do to facilitate the discovery of forecasting

principles, A and P perform a valuable service. With

collective and concerted effort, we can begin codify-

ing a discipline’s knowledge. This is A and P’s

message. Congratulations are in order.

References

Armstrong, J. S. (Ed.). (1995). Principles of forecasting. Boston,

MA: Kluwer Academic Publishing.

Armstrong, J. S., & Schultz, R. L. (1996). Principles involving

marketing policies: An empirical assessment. Marketing Letters

4 (3), 253–265.

Raymond Hubbard*

College of Business and Public Administration,

Drake University,

Des Moines, IA 50311, USA

*E-mail address: [email protected]

Tel.: +1-515-271-2344.

doi:10.1016/j.ijforecast.2004.04.002

Research on forecasting 741