Selection and industrial applications of panel data based demand forecasting models
| DOI | https://doi.org/10.1108/IMDS-08-2015-0345 |
| Pages | 1131-1159 |
| Date | 11 July 2016 |
| Published date | 11 July 2016 |
| Author | Shuyun Ren,Tsan-Ming Choi |
| Subject Matter | Information & knowledge management,Information systems,Data management systems |
Selection and industrial
applications of panel data based
demand forecasting models
Shuyun Ren
The Hong Kong Polytechnic University, Kowloon, Hong Kong, and
Tsan-Ming Choi
Business Division, Institute of Textiles and Clothing,
The Hong Kong Polytechnic University, Kowloon, Hong Kong
Abstract
Purpose –Panel data-based demand forecasting models have been widely adopted in various industrial
settings over the past few decades. Despite being a highly versatile and intuitive method, in the literature,
there is a lack of comprehensive review examining the strengths, the weaknesses, and the industrial
applications of panel data-based demand forecasting models. The purpose of this paper is to fill this gap by
reviewingand exploringthe featuresof various main streampanel data-baseddemand forecasting models.
A novel process, in the form of a flowchart, which helps practitioners to select the right panel data models
for real world industrial applications, is developed. Future research directions are proposed and discussed.
Design/methodology/approach –It is a review paper. A systematically searched and carefully
selected number of panel data-based forecasting models are examined analytically. Their features are
also explored and revealed.
Findings –This paper is the first one which reviews the analytical panel data models specifically for
demand forecasting applications. A novel model selection process is developed to assist decision
makers to select the right panel data models for their specific demand forecasting tasks. The strengths,
weaknesses, and industrial applications of different panel data-based demand forecasting models are
found. Future research agenda is proposed.
Research limitations/implications –This review covers most commonly used and important
panel data-based models for demand forecasting. However, some hybrid models, which combine the
panel data-based models with other models, are not covered.
Practical implications –The reviewed panel data-based demand forecasting models are applicable
in the real world. The proposed model selection flowchart is implementable in practice and it helps
practitioners to select the right panel data-based models for the respective industrial applications.
Originality/value –This paper is the first one which reviews the analytical panel data models
specifically for demand forecasting applications. It is original.
Keywords Data systems, Demand forecasting, Model selection, Panel data forecasting,
Technical review, Use of information
Paper type Literature review
Nomenclature
ithe cross-section dimension, i¼1, 2, …,N
tthe time-series dimension, i¼1, 2, …,T
NNindividuals
TTtime periods
yit the demand of individual iat time
period t
bparameter matrix (1 K)
Xit the itth observation on Kexogenous
variables
gcoefficient
uit an error term usually be modeled as
random variable with a zero mean and
a fixed variance
Utthe disturbance vector form of uit
Industrial Management & Data
Systems
Vol. 116 No. 6, 2016
pp. 1131-1159
©Emerald Group Publis hing Limited
0263-5577
DOI 10.1108/IMDS-08-2015-0345
Received 20 August 2015
Revised 24 December 2015
Accepted 1 February 2016
The current issue and full text archive of this journal is available on Emerald Insight at:
www.emeraldinsight.com/0263-5577.htm
This paper is a part of the first author’s (Shuyun Ren) PhD dissertation.
1131
Selection and
industrial
applications
aconstant
aiunobservable individual-specific effect
ltunobservable time effect
WaNNspatial weight matrix whose
diagonal element are zero
uthe vector of individual effects
ftthe remainder disturbances which are
independent of u
dspatial autoregressive coefficient
rserial autoregressive coefficient
1. Introduction
Forecasting is an integral part of industrial operations and production management.
Demand forecasts are important for understanding market situation and the
competition, production planning, including promotions, pricing, advertising, and
distribution (Frees and Miller, 2004). However, forecasting the future demand is a truly
challenging task. Various methods including statistical methods, intelligent methods,
and hybrid methods have been developed to conduct forecasting. In recent years,
with the emphasis on “big data”and the data driven knowledge-based operation
management, panel data-based forecasting models have been widely adopted in
various industrial settings. Panel data, also called time-series and cross-section data or
pool data (Hsiao, 2003), follows a given sample of individuals over time. It involves two
dimensions: a cross-sectional dimension N[1], and a time-series dimension T, and thus it
provides two-dimensional observations on each individual in the sample. The panel
data method is timely in the big data era, although the collection of panel data is more
costly than the one-dimensional ones[2]. Panel data models have some advantages over
the time-series econometric models. They usually give a larger number of data points
and incorporate much richer information from both time-series and cross-sectional
dimensional data. Panel data models consider variables observed over time and across
different units, and can identify effects that simply are not detectable through the
purely cross-section or time-serial analysis of data. Hence, panel data methods improve
the efficiency of econometric estimates (Hsiao, 2003). In a recent study, Ren et al. (2015)
suggest that panel data-based forecasting models outperform both time-series methods
and artificial intelligent methods in fashion sales forecasting. Panel data approach
also reduces the problem of multi-collinearity and provides a higher degree of freedom
in the model estimation (Song and Li, 2008). Therefore, it is especially suitable for
the forecasting problem when: the time series for all variables are shorter; and
cross-sectional information on these variables is available.
Over the past decades, panel data models and forecasting analysis have been used in
many research areas. Baltagi (2008b) give a pioneering survey of forecasting with panel
data and find that panel data estimators perform well in forecast performance, though the
performance of various panel data estimators and their corresponding forecasting
performance may vary from one empirical example to another. Different from Baltagi
(2008a), the current paper aims at providing guidance for panel data forecasting
procedures and further discussing on the strengths, the weaknesses, the application
performance, etc., of the panel data-based demand forecasting models. This paper
contributes to the literature and advances knowledge in three ways: first, to the best of our
knowledge, this paper is the first one which reviews the analytical panel data models
specifically for demand forecasting applications. Second, we provide a novel model
selection flowchart to let decision makers choose the right panel data method for their
specific demand forecasting tasks with respect to the proper data tests. Third, we
reveal the strengths, the weaknesses, and the industrial applications of different panel
data-based demand forecasting models and discuss future research directions.
1132
IMDS
116,6
To continue reading
Request your trial