Page 99 - Contributed Paper Session (CPS) - Volume 4

CPS2134 Yutaka Kuroki et al.

A multi-factor modelling for retail demand
forecasting: An empirical analysis of restaurant
visitors prediction
1
2
Yutaka Kuroki ; Takayuki Shiohama
1 Graduate School of Engineering, Tokyo University of Science
2 Department of Information and Computer Technology, Tokyo University of Science

Abstract
Analyzing cross-sectional and time-series retail sales data is important for
multi store retail managements, especially in service related and retail
businesses. This paper presents a use of factor model for numbers of
customers forecasting in retail business and tests the validity of the proposed
model. The factors are constructed by means of fundamental factors which are
common tools for analyzing asset pricing models in financial market analysis.
Data analysis using Japanese restaurants data are illustrated and showed that
the effectiveness of the multi-factor modeling with high forecasting
performances.

Keywords
Marketing; factor model; panel data econometrics; structural time series
analysis.

1. Introduction
There has been an enormous growth in needs for big-data analytics in
marketing science. Point-of-Sales (POS) data can be helpful to provide
accurate demand forecast of a retail shop and be used to analyze consumer
buying behavior. Big data analysis makes one-to-one marketing possible,
which improves management effectiveness and accurate decision making in
their supply chain. Forecasting demand in multi store sales is especially
important for effective managements such as franchise chains. Since demand
not for a single store, but for whole stores is dominated by calendar effects,
which can be considered as an undiversifiable risk called a “systematic risk”.
On the other hand, demand forecasting for a single store is not enough
explained by such a common effect. We need to model and manage an
“idiosyncratic risk” which arise in single store retail businesses by using
appropriate statistical approaches.
In this study, we propose the factor models for number of customers of
restaurants in Japan. We use seasonal and calendar effects as dominant
factors, other factors are also proposed using the similar idea of analyzing the
Capital Asset Pricing Model (CAPM) in financial econometrics. These factors
include market, size, and volatility factors, which are considered as anomalies
of the dynamics of the cross-sectional restaurant visitors time series. The
88 | I S I W S C 2 0 1 9