Volume 10, Issue 2, 2016

Influence of Factors on Clothing Sales and Its Future Trend:

Regression Analysis and Time Series Forecast of Clothing Sales

Avoni Gharde
Ph.D. Program in Textile Technology Management
College of Textiles
North Carolina State University
Raleigh, North Carolina
27695-8301
USA

ABSTRACT

Forecasting is crucial for many industries. Sales forecasting for retail is critical to understand
the volatile product demand and supply chain lifecycle. This paper is a study conducted on
clothing store sales with data of previous 20 years. Variables such as GDP, GDP per capita,
population, marriage rates and their effect on clothing store sales were studied. Regression
analysis, ARIMA model for time series analysis with JMP software were used to forecast sales for
the next 20 and 50 years.

Keywords: Sales forecasting, clothing stores, Regression analysis, ARIMA

Introduction between the fashion industry and the

The apparel supply chain has one consumer and between the various segments
purpose – to provide an appealing and in the supply side chain. In the narrowest
desirable product to satisfy consumer needs, sense, forecasting attempts to project past
wants, or aspirations. When successful, the trends into the future.
connection results in a sale. Because this Cost reduction, customer
context is the purpose of the process, every requirements, globalization and technology,
forecast begins with the customer – by have impacted the textiles, apparel and
observing the client's adjustments to the fashion industry and have contributed to the
marketplace and in the unexpected ways the varying enablers of seasonal looks in the
customer adjusts the market to his or her market.
lifestyle and preferences. Forecasting, in the
apparel and textiles industry, is a process Sales Forecasting
that spans shifts in color and styles, changes
Sales forecasting refers to predicting
in lifestyles and buying patterns, and
future demand (or sales), assuming that the
different ways of doing business (Brannon,
factors which affected demand in the past
2000). What appeals to be near random
and are affecting the present will still have
activity is, in fact, a process of negotiation
an influence in the future. It is an important

Article Designation: Graduate Student Article 1 JTATM

Volume 10, Issue 2, 2016
task but is tough to accomplish (Liu, Ren, textiles from creation to distribution is
Choi, Hui, & Ng, 2013). In the fashion shown in figure 1.
retailing industry, which is defined as the
retailing business of fashion products
including apparel, shoes, and fashion beauty
products, forecasting itself can be treated as
a "service" which represents the set of
analytical tools that facilitate the companies
to make the best decisions for predicting the
future. Undoubtedly, a good forecasting
service system assists in avoiding
understocking or overstocking in retail Figure 1. Production Planning of
inventory planning. It further relates to Autumn-Winter Textile Items
critical operations of the whole supply chain The purchasing managers need to know
management, production planning, pricing almost one year in advance the quantities of
(C.-H.Chiu, T.-M.Choi, & Li, 2009; Chiu & raw materials to order which will correspond
Choi, 2010), and achieving high customer to the forecast total volume of each textile
service level (Frank, Garg, Raheja, & product range. Production managers also
Sztandera, 2003). require item quantities to manufacture
Over the past decade, some research particularly early in the case of imported
studies have been reported in the literature. items from distant countries. Moreover, it is
However, each forecasting method has its necessary to fit in the forecasts after the
limits and drawbacks. For example, the launch of the textile items according to the
traditional statistical methods depend highly restock achieved by local manufacturers.
on the time series data's features, and this Therefore the textile industry requires both
will affect the forecasting accuracy a lot. mean (one season) term and short term (one
Artificial intelligence (AI) methods can to six weeks) forecasting models (Vroman,
perform better regarding accuracy than the Happiette, & Vasseur, 2001).
traditional statistical forecasting models, but In the literature, Green & Harrison
they usually require a much longer time and (1973) apply a Bayesian approach to explore
a larger requirement on computational forecasting for a mail order company which
power. Thus, many researchers propose to sells ladies dresses. After that, Thomassey,
combine multiple methods together to form Happiette & Castelain (2003), used item
a new “hybrid method” to achieve an classification to examine the accuracy of
efficient and effective forecasting task (Liu, sales forecasting for new items. They found
Ren, Choi, Hui, & Ng, 2013). that a larger number of item families and
appropriate classification criteria are
Statistical Models for Sales Forecasting required in the particular prediction
Traditionally, fashion sales procedure to achieve an improved
forecasting is accomplished by the statistical forecasting precision. They concluded that
methods. A lot of statistical methods have product family and aggregated forecasting
been used for sales forecasting, which are more accurate than the individual item’s
includes linear regression, moving average, forecasting. It is accepted that there is no
weighted average, exponential smoothing, forecasting technique appropriate to an all
exponential smoothing with a trend, double sales time series (Ferrer, Hoyo, & Arroyo,
exponential smoothing, Bayesian analysis, 1997).
and so forth. Statistical time series analysis Witt & Witt (1994), have reported that
tools such as ARIMA and SARIMA are also different forecasting techniques might
widely employed in sales forecasting. The perform differently in handling stable vs.
production planning of autumn-winter unstable data. It is well accepted that no
forecasting technology is appropriate to all

Article Designation: Graduate Student Article 2 JTATM

Volume 10, Issue 2, 2016
data patterns. But no research has been
investigated to compare the effects of
various methods on different sales data
patterns from apparel retailing, therefore,
leaving room for further exploration.

Time-Critical Decision Modeling and

Analysis
The ability to model, perform decision
modeling and analysis is an important
feature of many real-world applications.
These range from medical treatments to
military systems. Existing formalisms and
steps of inference have not been capable of Figure 2. Potential Choices of Forecasting
real-time applications where trade-offs Methods
between decision quality and computational
tractability are essential. In practice, a Data Collection
practical approach to time-critical dynamic
US Department of Commerce’s
decision modelling should provide explicit
database provided with the data of estimated
support for the modelling of temporal
annual sales of US Retail and Food services
processes and for dealing with time-critical
with various categories including clothing
situations (Arsham, n.d.). Almost all
and accessories stores around the US.
managerial decisions depend on forecasts. A
Population, marriage rates, divorce rates as
decision becomes operational at a stage in
well as GDP and GDP per capita were
the future, so ideally, it should be based on
gathered from US Census Bureau and CDC
forecasts of future conditions.
(Centers for Disease Control and
Two approaches to forecasting are
Prevention). The variables chosen to study
the estimate of future value, based on an
their effect on sales were GDP, GDP per
analysis of factors that influence future
capita, population and marriage rates of
values, i.e., the explanatory method, or else
USA.
the prediction builds on an inferred study of
past data behavior over time, i.e., the
Influence of Factors on Expenditures
extrapolation method. Both approaches can
lead to a creation of accurate and It is quite obvious to the fact that as
meaningful forecasts, however, even for a the household income shall increase, costs
modest degree of the desired accuracy, the shall go up. About clothing, consumer
former method is often difficult to behavior is highly unpredictable and
implement and validate. variables such as income, season, political
climate, environmental factors have no
effect on the shopping pattern. No matter
what, going shopping is a major source of
relaxation as well as a household chore
(Badar, 2008).
According to the study conducted
by Badar, decision making is a critical
component of consumer behavior that
significantly affects the choices made. In his
study among the sampled respondents, more
than 57 percent were married, around 41
percent were singles while less than one

Article Designation: Graduate Student Article 3 JTATM

Volume 10, Issue 2, 2016
percent were divorced and widows. He Regression Analysis
found that family size contributed
Regression analysis is a statistical
significantly in shopping pattern of the
technique to analyze quantitative data to
consumers; larger the family size, higher
estimate model parameters and make
will be the shopping expenditure. In another
forecasts. The X-axis is the horizontal line,
study by Smith (2004), there were many
and the vertical line is referred to as the Y-
single households in the US than married
axis. Regression analysis looks for a
households with children. Smith found that
relationship between the X variable
shopping as a family was three times more
(sometimes called the “independent” or
likely in single mother households than
“explanatory” variable) and the Y variable
married ones. (Martin & Ramsaran-Fowdar,
(the “dependent” variable) (Guerard Jr.,
18 - 20 November, 2013).
2013). Forecasts using the regression line,
assume that the relationship which existed in
Rate the past between the two variables will
continue to live in the future.
10 Consumption makes up for a
8 majority of real Gross Domestic Product,
6 known as GDP (Guerard Jr., 2013). GDP per
Marriage Rate
4 Capita compares GDP on a purchasing
Divorce Rate power parity basis divided by population as
2
of 1 July for the same year. From the figure
0
below, we see that the real GDP has
1994 1999 2004 2009 2014
increased substantially with a slight dip
during the recession of 2009.
Figure 3. Rates of marriage and divorce
(USA)
In the case of married couples, an Nominal GDP (in billions $)
interesting observation was made. An $20,000
increase in the wife’s income relative to her
husband’s income was associated with $10,000
greater expenditure on food, childcare, and
children's and women's clothing, and $0
reduced expenditure on transportation 1990 1994 1998 2002 2006 2010 2014
(Phipps and Burton, 1998). Lundberg,
Startz, and Stillman (2003) noted that wives Figure 4. Incremental GDP (in billions $)
preferred to save more than their husbands of United States
and that relative control over household
decisions are affected by control over
market income (ZIOL-GUEST, DELEIRE,
GDP per capita (in US$)
& KALIL, 2006). 60,000.00
From these research studies, a 40,000.00
prominent hypothesis which formed was the
influence of marital status on the consumer 20,000.00
shopping behavior – in this case, whether or 0.00
not it is directing affecting the clothing store 1994 1997 2000 2003 2006 2009 2012
sales. To start with, a regression analysis
was done on identifying factors which Figure 5. Incremental GDP per Capita (in
significantly affected clothing store sales in $) of United States
the period of 1994-2014.

Article Designation: Graduate Student Article 4 JTATM

Volume 10, Issue 2, 2016
 developed for clothing store sales by other
Clothing stores (in millions variables like marriage rate and GDP per
$) Capita. Equation generated was as follows:
200000 Clothing stores = 220509.66-
100000 8421.855*Marriage Rate
0
1994 1997 2000 2003 2006 2009 2012

Figure 6. Clothing Store Sales

This corresponds to the trend shown by
GDP per Capita and clothing sales during
the specified period.
Using JMP software, after analyzing
the data, the best-fit line and a regression
model were developed with the clothing
sales by GDP of USA plot; for the period
1994-2014. As we can see by the plot, the
increase in GDP through the years has
guided the simultaneous incremental
clothing sales during 1994-2014. Figure 8. Best fit of Clothing Sales by
Marriage Rate (USA)
R-square as well as R-square adjacent
values were produced as 0.91195 and 0.9073
respectively. There is a very high negative
correlation amongst the two with inverse
proportionality. The probability value was
found to be <0.001 for marriage rates, which
proves a large amount of interaction. The
equation generated for GDP per Capita by
clothing store sales was:
Clothing stores =
1930.1785+1.8162133*GDP per capita

Figure 7. Best fit of Clothing Sales by

GDP (USA)
An estimated regression model equation
generated by the software is as follows:

Clothing stores (in millions $) = 17530.288

+ 4.8744465*GDP in billions ($)

The predictor variable here being GDP (in

billions $) and the response variable is
clothing sales (in millions $) and R-square
as well as R-square adjacent values as
0.9918 and 0.9913 respectively. Figure 9. Best Fit of Clothing Sales by
Regression models were also GDP per Capita

Article Designation: Graduate Student Article 5 JTATM

Volume 10, Issue 2, 2016
R-square as well as R-square adjacent values marital status, consumers with ages 27 to 29
were produced as 0.98958 and 0.9890 years, earn more, spend more, are more
respectively for figure 9. There is a high likely to have a bachelor's degree. Also, they
interaction between GDP per capita and are more likely to own a home than their
clothing sales, so as the income increases, early twenties counterparts. Singles in their
expenditure on clothing stores shall 20 to 24 years earn less in income than
proportionally rise. early-twenties married couples but spend
For figure 10, R-square, as well as R- about the same amount per capita. However,
square adjacent values, were produced as late-twenties singles earn per-capita incomes
0.9641 and 0.9622 respectively. The similar to those of late-twenties married
estimated regression equation generated by couples but spend significantly more per
the software was: capita. Income appreciably affects GDP of
Clothing stores = - 153624.8 the nation, and this is now in accordance
+0.0007939*Population with the analysis done below with
independent variable as marriage rate.

Figure 10. Best fit of Clothing Sales by

Population
Figure 11. Best fit of GDP by Marriage
About the hypothesis presented earlier of Rate
marital status affecting clothing store sales,
as seen in the previous analysis, there is a The equation generated was:
high negative correlation. The next step was
to figure out the impact on income and GDP GDP in billions ($) = 42003.908 -
since sales and GDP were proportionally 3825.873*Marriage Rate
related to one another. At a fundamental
level, every marriage creates a new R-square as well as R-square adjacent values
household, an independent economic unit were produced as 0.9423 and 0.9392
that generates income, spends, saves, and respectively. Also, the probability value was
invests. The vast majority of these new <0.001 which shows a high significance.
homes produces babies and transforms what Using the Graph Builder tool on JMP
are mostly teens, into responsible adults, software, a visualizing correlation between
contributing to the necessary next generation data becomes easier. The figure given below
of human capital to the economy. shows us clothing store sales in relation to
According to the U.S. Bureau of population and GDP per Capita of USA
Labor Statistics, Consumer Expenditure (period of 1994-1014).
Survey (2007-2008), irrespective of their

Article Designation: Graduate Student Article 6 JTATM

Volume 10, Issue 2, 2016
 of the series. These indicate how and to
what degree each point in the series is
correlated with earlier values in the series.
Using this platform, data was studied,
and a growing trend pattern was discovered
in annual clothing store sales from 1994-
2014 while a decreasing trend was seen in
the marriage rate vs. GDP (in billions $).

Figure 12. Correlation between Clothing

Store Sales by Population and GDP per
Capita
As population and GDP per capita were
Figure 14. Trend of Clothing Store Sales
increased, there was a significant rise in the
in the US (1994-2014)
clothing sales as reported, which is, again, in
correlation to the analysis done above.

Figure 15. Trend of Marriage Rate by

GDP in the US (1994-2014)
Figure 13. Marriage Rate vs. Clothing ARIMA
Store Sales Theoretically, ARIMA models are
the most general class of models for
As per the graphs, there has been a forecasting a time series which can be made
decrease in marriage rates during the period "stationary" by differencing, perhaps in
1994-2014. The trend seems to be the same conjunction with nonlinear transformations
with an increase in clothing store sales such as logging or deflating. A random
despite lower rates of marriage. variable, like a time series, is stationary if its
statistical properties are all constant over
Time Series Forecast Model on JMP time. A stationary series has no trend as its
The Time Series platform allows you variations around its mean have a constant
to explore, analyze, and forecast univariate amplitude, and it wiggles in a consistent
time series. A time series is a set y1, y2, ... , fashion, i.e., its short-term random time
yN of observations taken over a series of patterns always look the same in a statistical
equally-spaced time periods. The analysis sense. The latter condition means that
begins with a plot of the points in the time its autocorrelations (correlations with its
series (SAS Institute Inc., n.d.). Also, the prior deviations from the mean) remain
platform displays graphs of the constant over time remains constant over
autocorrelations and partial autocorrelations time (Duke University, n.d.). A random

Article Designation: Graduate Student Article 7 JTATM

Volume 10, Issue 2, 2016
variable of this type can be seen as a
combination of signal and noise, and the
signal could be a pattern of fast or slow
mean reversion, sinusoidal oscillation, or a
rapid alternation in sign. A seasonal
component can be implemented, if
needed. An ARIMA model is used as a
"filter" that tries to separate the signal from
the noise, and the signal is then extrapolated
into the future to obtain forecasts.
To choose an appropriate model,
first, a determination of the order of
differencing is required to stationarize the
series. Normally, the correct amount of
differencing is the lowest order of
differencing that yields a time series, which
fluctuates around a well-defined mean value Figure 17. AR (1) Model for Marriage
and whose autocorrelation function (ACF) Rate
plot decays fairly rapidly to zero (either
from above or below). Second is to
determine the orders p and q of the The plots of ACF and PACF can be
integrated time series of order d by studying interpreted as follows:
ACF and partial ACF i.e. PACF. a) If the ACF abruptly stops at some point,
after q spikes, this can be an MA, i.e.,
Moving Average (q) type model.
b) If the PACF abruptly stops after p spikes
too, here AR (p) model would work
best.
c) Had neither been present but a decline
towards zero was evident in some
fashion, the appropriate model would
have been ARMA (p,q).
JMP software helps to work with this
model as well as SARIMA. With the given
input data, a forecast was generated as
follows:

Figure 16. AR (1) Model for Clothing

Store Sales

Figure 18. Predicted Sales for Next 20

Years (2015-2034) with AR (1) Model

Article Designation: Graduate Student Article 8 JTATM

Volume 10, Issue 2, 2016
 0.8227 and 0.81339 respectively while for
marriage rates vs. GDP (figure 20), R-square
as well as R-square adjacent values were
0.7554 and 0.7425 respectively. Finally, we
see a predicted rise in clothing store sales
(figure 21) as GDP rise and R-square as well
as R-square adjacent values were found as
0.84211 and 0.8338 respectively. Once the
plots are made, predicted values give us
accurate annual sales data with a confidence
Figure 19. Predicted Change in Marriage interval range of values.
Rates for Next 20 Years (2015-2034) with All the probability values obtained
AR (1) Model were less than 0.001 which proved that they
are significant; as there is a rise in marriage
rate with increase in GDP, there should be a
rise in the clothing store sales as well. This
proves the hypothesis of marriage rate being
a factor influencing the clothing store sales!

Conclusion and Future Scope

In this paper, a short literature review
regarding the tools and models that are used
for forecasting is provided. Real-time data
from US Department of Commerce (retail-
clothing stores sales) as well as
Figure 20. Predicted Change in marriage demographics and economic survey data
Rate as GDP Increases with AR (1) Model from US Census Bureau was studied to
analyze and predict future sales for a span of
20 years and 50 years.
From the regression analysis done on
the data, there is high significance and
interaction present between GDP, GDP per
Capita, population to clothing store sales
and also marriage rates by GDP in the USA
in the specified period of 1994-2014. Also,
in the trend observed (of clothing store
sales), there seems to be an increase in these
sales while there is a decrease in marriage
Figure 21. Predicted Rise in Clothing rates.
Store Sales as GDP Increases with AR (1) With time series analysis, there is a
Model growth in expenditure as the household
income per capita or GDP of USA rises per
AR (1) models (depicted above) were year; or so we expect this. However, the
plotted with probability values and 0.05 forecasts show a dip in the overall sale and a
(95% confidence intervals). In figure 18, R- large confidence interval is plotted with
square, as well as R-square adjacent values, variance. We also found there will be a rise
were found as 0.83867 and 0.83018 in marriage rates as GDP increase which in
respectively for clothing store sales. For turn should lead to a rise in clothing store
marriage rates (figure 19), R-square, as well sale. Just forecasting based on one factor
as R-square adjacent values produced, were isn’t yielding accurate results.

Article Designation: Graduate Student Article 9 JTATM

Volume 10, Issue 2, 2016
 The AR (1) model produced a large Frank, C., A. G., Raheja, A., & Sztandera,
confidence interval between predicted L. (2003). Forecasting women’s
values. Newer models such as pure AI and apparel sales using mathematical
hybrid models could be tried to predict more modeling. International Journal of
refined forecasting. Gathering many Clothing Science and Technology, 15
historical data creates noise, therefore (2), 107–125.
optimizing the data, although a hassle, can Green, M., & Harrison, P. J. (1973). Fashion
lead to improved results. forecasting for a mail order company
using a Bayesian approach.
References Operational Research Quarterly, 24
(2), 193–205.
Arsham, H. (n.d.). Time-Critical Decision
Guerard Jr., J. (2013). Regression Analysis
Making. Retrieved from
and Forecasting Models. In J. Guerard
http://home.ubalt.edu/ntsbarsh/stat-
Jr., Introduction to Financial
data/forecast.htm#rgintroduction
Forecasting in Investment Analysis
Badar, H. (2008). An Exploration of Female
(pp. 19-45). Springer.
Shopping Behaviour: A Case Study of
Liu, N., Ren, S., Choi, T.-M., Hui, C.-L., &
City Faisalabad (Pakistan). Pakistan
Ng, S.-F. (2013). Sales Forecasting for
Journal of Life and Social Sciences, 6
Fashion Retailing Service Industry: A
(2), 75-79.
Review. Mathematical Problems in
Brannon, E. (2000). Color Forecasting In
Engineering.
Fashion Forecasting. New York:
Lundberg, Shelly, Richard Startz, and
Fairchild Publication Inc.
Steven Stillman. (2003). The
C.-H.Chiu, T.-M.Choi, & Li, D. (2009).
Retirement Consumption Puzzle: A
Pricewall orwar: the pricing strategies
Marital Bargaining Approach. Journal
for retailers. IEEE Transactions on
of Public Economics, 87 (5–6): 1199–
Systems,Man, and Cybernetics Part A,
1218.
39 (2), 331–343.
Martin, K., & Ramsaran-Fowdar, R. R. (18 -
Chiu, C.-H., & Choi, T.-M. (2010). Optimal
20 November, 2013). An
pricing and stocking decisions for
Examination of the Consumer
newsvendor problem with value-at-
Buying Behaviour of Single-Parent
risk consideration. IEEE Transactions
Households. Proceedings of 23rd
on Systems,Man, and Cybernetics
International Business Research
Part A, 40 (5), 1116–1119.
Conference. Melbourne, Australia.
Choi, T. M., Hui, C. L., & Yu, Y. (2003).
Phipps, Shelley A. and Peter S. Burton.
Intelligent Fashion Forecasting
1998. What’s Mine Is Yours? The
Systems: Models and Applications.
Influence of Male and Female
New York, NY, USA: Springer.
Incomes on Patterns of Household
Duke University. (n.d.). Introduction to
Expenditure. Economics, 65 (260),
ARIMA models. Retrieved from
599–613.
Forecasting:
SAS Institute Inc. (n.d.). Time Series
http://people.duke.edu/~rnau/411arim.
Analysis. Retrieved from JMP SAS:
htm
http://www.jmp.com/support/help/T
Ferrer, A. G., Hoyo, J. D., & Arroyo, A. M.
ime_Series_Analysis.shtml
(1997). Univariate forecasting
Smith, J. (2004). A single-minded
comparisons: The case of the Spanish
marketplace. Marketing
automobile industry. Journal of
Management, Vol. 13, Issue 4, p. 52.
Forecasting, 16, 1-17.

Article Designation: Graduate Student Article 10 JTATM

Volume 10, Issue 2, 2016
Somer, E., & Ruvio, A. ( 2014). The Going Vroman, P., Happiette, M., & Vasseur, C.
Gets Tough, So Let’s Go Shopping: (2001). A Hybrid Neural Model for
On Materialism, Coping, and Mean-Term Sales Forecasting of
Consumer Behaviors Under Textile Items. Studies in Informatics
Traumatic Stress. Journal of Loss and Control, 10 (2), 149-167.
and Trauma, 19, 426–441. Witt, C. A., & Witt, S. F. (1994).
Thomassey, S., Happiette, M., & Castelain, Forecasting International Tourist
J. M. (2003). Mean-term textile Flows. Annals of Tourism Research,
sales forecasting using families and 21, 612-628.
items classification. Studies in ZIOL-GUEST, K. M., DELEIRE, T., &
Informatics and Control, 12 (1). 41– KALIL, A. (2006). The Allocation
52. of Food Expenditure in Married-
and Single-Parent Families. The
Journal of Consumer Affairs, 40 (2),
347-371.

Article Designation: Graduate Student Article 11 JTATM

Volume 10, Issue 2, 2016