New Normal Weather Breaks a Traditional Clothing Retail Calendar

Background: Clothing businesses have complained of sluggish sales because of new normal weather, an increased variation of temperature and precipitation and the higher occurrence of extreme weather events. Traditionally, the business runs tied to calendar dates or retailing events, and the previous year's sales draw up a sales plan. This study questioned whether the sales planning method of the clothing business is valid and reliable for today. Results: Using weather observation data and Google Trends for the past 11 years, consumers' responses to weather changes were analyzed through the decision tree to learn about consumer insights. The month is the most significant predictor of seasonal clothing demand during a season, and consumers' responses to weather vary from month to month. Minimum temperature and maximum temperature were significant predictors in a particular month. Conclusions: Our results have important managerial implications. Rapid weather changes affect consumers’ demand. Clothing retail ers can apply the predictive model to quickly respond to unexpected weather changes, prepare products with rapidly increasing demand not to miss sales opportunities, and adjust quantities and prices for products with sharp declines in demand.


Background
Unseasonable warm winter, worst heatwaves, cold and wet springs, and cooler and longer summers become new normal weather. It is frequently cited in numerous publications to show the effects of unpredicted weather factors on clothing sales. The demand forecasting of seasonal clothing is an extreme challenge because the clothing industry runs its business tied to a calendar date or a retailing event and develops a sales plan based on last year's sales. The clothing industry is a highly fragmented global value chain, encompassing textiles producers, designers, manufactures, specialized factories, branded markers, and retailers [26]. Thus, the clothing product procurement requires long lead times and lesser flexibility cause overstocking and understocking problem.
Summer clothes are introduced in mid-April at the store, and hot weather appears to encourage sales. Most consumers purchase summer clothes from May to July. Even though the temperature is continuously high in August, consumers hesitate to purchase new summer clothes at full prices because the new fall clothes are introduced in August. Winter clothes are introduced in October, and the cold temperature encourages winter clothing sales. Before the end of the season, all stocks sell with deep price reductions called the end-of-season clearance sales to liquidate all items. The ideal weather conditions for clothing retailers seem to be one where the seasons exert themselves early, and thus new seasonal products sell quickly with full price.
There is a large body of evidence in clothing retailers to have challenges from unseasonal weather events. In 1995, estimates of the impact of the anomalous climate of heat sources in the UK on retail showed that the clothing and footwear retail market has the highest weather sensitivity [1]. A record-breaking El Nino event marked the winter of 1997-1998, and retailers in the northern states in the US suffered from a lack of winter clothing sales [12,25]. Gap Inc. reported poor sales in the first quarter of 2019 because of unseasonable cold and wet weather [39]. The leading US clothing retailers observed depressed sales for the winter clothing due to unseasonably warm temperatures into the high of about 15°C and some parts of the East Coast in the US [37]. While many professionals acknowledge the importance of weather as a sale forecasting factor [8,13,17,30,37], very few companies such as ZARA utilize weather information in the sales plan.
Since the retail clothing market is the most sensitive to weather change [1], it is questionable whether clothing retailers' traditional merchandising planning methods are valid and reliable today in new normal weather. In this study, we tested consumers' seasonal clothing demand using data from the past decade to develop the foundation for a predictive model that clothing retailers can predict a seasonal sales plan and respond to weather changes that can support supply chain activities. Therefore, this research proposed to answer the following questions: 1) when do consumers search for seasonal clothes? 2) how do weather factors affect consumers' seasonal clothing searches? 3) What are the essential predictors, and how the predictors interact with consumers' seasonal clothing searches?
Temperature leading to clothing sales up to a certain period of a season Researchers have long recognized that clothing sale depends on unique aspects of clothing, including trend, quality, and price, and is related to the interaction of several weather situations. While the objectives of studies are various, and the results of studies are inconsistent, numerous studies investigated the impact of temperature changes on clothing sales [1,2,3,4,6,7,22]. Warmer temperature increases sales of the clothing item (i.e., dress) in Spring and Summer and decreases clothing items (i.e., winter jacket) in fall and winter [6,7,22]. Specifically, when the temperature is 1°C warmer in Spring, there is a three percent increase in sales. In fall, the opposite applies; when temperate is up by 1°C, there is a three percent drop in sales [6]. The maximum temperature fluctuations are insignificant in clothing sales, but the sales increase when the temperature increases to some extent and then decreases [2]. Sunchine [1], Snow depth [2], and rain [22] were considered to be the significant factors in buying clothes. Consumers are physically and psychologically inconvenient to go to a clothing store to buy clothes during harsh weather such as snow or rain, resulting in decreased sales. An inch of snowfall reduces sales by 17%, but the snowfall effect is weak if the snowfall is historically common [32]. On a rainy day, store traffic is reduced by 7.4 % in street stores, increasing by 5.2 percent in a shopping mall [22].
The weather generally has a complex effect on daily sales in a brick-and-mortar store. In contrast, the degree and direction of weather effects depended on the store location and the sales promotion such as promotion theme [3]. Consumers' online searches and online shopping activities are a significant part of the market; consumers may shop through online stores when faced with unfavorable weather for clothing shopping to physical stores. Two studies showed contradictory findings. One study found that online clothing sales had a very positive relationship with rain [35], but another study did not provide evidence of bad weather online clothing sales [32].
Maybe retailers wished for supernormal weather that was warmer than average in spring and cooler than average in fall [28]. Early weather changes in the new season are significant because rapid weather changes increase new season clothing sales. However, towards the middle of the season, the weather, especially the temperature, does not spur sales [1,2,4,6]. Determining when to purchase when the weather is consistently hot or cold involves the possibility of comparison performance, and consumers hesitate to buy seasonal items at the end of the season because consumers feel bored with the clothes they are wearing now and want to change to the clothes of the new season to come.

Seasonal clothing search on Google
Clothing is the human's most immediate environment associated with ab individual's biologically protective functions. Weather changes are the most dominant exogenous factor and influence an individual's everyday clothes choices [20].
Consumers recognize the need for new seasonal clothes concerning physical and psychological comfort. The recognition of needs results in the creation of information search activities. The determination of an information search by a consumer before a purchase is either to enhance the quality of the purchase outcome or pleasure [9]. Online retailing has grown over the past decade.
According to the US Census Bureau News [41], online sales were 10.2 % of total sales in the first quarter of 2019, and clothing consumers use Google to search for ideas, find the best designs, and buy new clothes [10]. In this study, Google Trends (GT) was employed. GT is an extensive realtime dataset and indicates consumers' seasonal demand for any given product [34]. GT is not familiar with the clothing and textiles field, but it has gained popularity in various research areas such as finance, health care, hospitality, and tourism [15,29,31]. GT and complement survey data such as Consumer Index and Consumer Sentiment are positively associated [27]. There is a need for more study that provides conclusive evidence to confirm that GT can predict clothing purchase.
This study used GT, the volumes of search queries on a product, as a consumers' demand indicator because an online search for information is the strongest predictor of online purchase intention [33], and consumers may combine online and offline channels to search for information and to purchase seasonal clothes.

Rationale of the study
The studies mentioned above have provided a useful contribution to understanding the usefulness of weather in the clothing market. However, there are limitations in developing a prediction of clothing demand that responds to weather changes over a single season.
First, most of the studies mentioned above have found the relationship between weather factors and clothing sales in brick-and-mortar stores. Consumers feel that unpleasant weather conditions (i.e., rain, snow, snow depth) are a physical and psychological obstacle to traveling to a brick-andmortar store. It is reasonable to use Google Trend, consumers' online search, as an indicator of consumer demand in today's society dominated by online search and online shopping. Second, most previous studies have tested clothing sales in one season or year. However, this study tested each month during the winter months. Since extreme weather is getting more frequent and severe, and the lifecycle of clothing is getting shorter, not only the overall winter retail months need to test, but each month during the season needs to test. Third, under new normal weather, it is crucial to identify the predictable factors for consumers' seasonal clothes demand and reveal the relationship among the predictable factors during a season from a managerial perspective. Clothing sales increased when the temperature increased to some extent and then decreased [2]. Without considering nonlinear effects, the impact of extremely bad and good weather occurrences could be misestimated [3]. Therefore, the study used a decision tree, one of the most popular data science techniques, to understand consumer insight, the nonlinear effect of weather variables on clothing sales.

Methodology
Study area and study period There are three reasons to select one study area: First, individuals living in different cultures and environments show different psychological evaluations related to weather despite similar climate conditions [19]. Second, clothing reflects a wide range of individual psychological and physiological needs, so individual heterogeneity is relatively high.
Third, since this study is about seasonal clothing demand, a region of Goole Trends where longterm data are available, and areas with seasons were selected. New York City, which has a significant influence on global fashion trends and is a significant center of the clothing industry, was selected as the study area. The summer months in New York City are warm and humid, while the winter months in New York City are frigid and windy. According to the Climate Normals: 1981-2010 in NYC [42], the highest temperature in early October is over 12℃, and the lowest temperature is 15℃. The trend of temperature is dropped until late January and is increased in early February. The average annual temperature has increased about 1.1 over the last two decades, and the polar jet stream has brought the number of extreme precipitation events during the winter [14].
Data The coldest months in New York City are from January to February, and the cold winter lasts until early March. Clothes are usually displayed in stores from October to early November and sold during winter. Winter clothing is sold at all price points in October, and after Holidays, retailers generally lower prices to stimulate more sales. This study analyzed the data from October 1 to January 31, as winter clothing sales after January do not generate enough revenue for retailers, and retailers make an effort to try to introduce spring clothing products. Therefore, the data were obtained from October 1 to January 31 for the past 11 years. The data set was combined with two data sources. Consumers' seasonal clothing search data was from GT, and weather data was from NOAA. Table 1 shows the definitions of each variable in this study.  Maximum temperature (C°)* The highest temperature recorded during a specified period of time; The most common reference is to the daily maximum temperature.

MINT
Minimum temperature (C°)* The lowest temperature recorded during a specified period of time; The most common reference is to the daily minimum temperature.

PRCP
Precipitation(inch)* The process where water vapor condenses in the atmosphere to form water droplets that fall to the Earth as rain, sleet, snow, hail, etc.

SNOW
Snowfall (in)* The amount of snow that falls there during a particular period. With Pearson correlation tests, the relationship between WGT and weather factors were investigated. Then, to develop a predictive model's foundation, the critical predictors for consumers' seasonal clothes search were identified through the decision tree regression model. A decision tree is a popular data science technique for discovering the consumers' decision process [16,21,23]. It is to find and describe structural patterns in data as tree structures, explaining data, and making predictions using the data [40]. The primary purpose of using the decision tree is to achieve a more concise and perspicuous representation of the relationship between a target variable and predictor variables. This study used the classification and regression tree (C&RT) algorithm with WGT as a target variable and all other variables (i.e., MAXT, MINT, PRCP, SNOW, SNWD, & WIND) as predictors. C&RT is a recursive partitioning method used for both classification and regression. It is configured by dividing a subset of a data set using all predictors, creating two child nodes repeatedly, starting with the entire data set [11].

When do consumers search for seasonal clothes? Consumers' seasonal clothes searching pattern
during the four winter months is shown in Figure 1. The overall average of WGT is 32.87 (SD =

19.09). Consumers' searching was increased in late October, and the peak was in mid and late
November. Then, consumers' searching was declined in early January. A one-way betweensubjects ANOVA was conducted to compare the effect of the months of the winter retail season on consumers' seasonal clothes search. Welch's test showed F (3, 743.989)=30.351, p<.000, and the Games-Howell test indicated that the mean scores for monthly demand were significantly different (See Table 2   promotions from retailers are available throughout the years. One recent study reported that as more shopping moves online, the brick-and-mortar store events are less essential to the consumer [38]. Since clothing retailers rely on winter clothing sales to boost annual profit, selling without deep markdowns is a key driver of clothing business success.  Snow is an external stimulus to consumers, motivating consumers to prepare winter clothes.

What are the important predictors, and how the predictors interact with consumers'
seasonal clothing searches? A decision tree identified the consumers' seasonal clothing predictors and the relationships among the predictors. All variables were used to construct a decision tree regression model to predict WGT as a continuous variable. The data (n=1353) were split by using the sampling technique of 80/20, partitioning 80% into a training set (n=1090) and 20% into a testing set (n=263). The result of the C&RT identified important predictors and showed the relative importance of predictors (see Figure 2) factor. The importance of temperature changes in the early season was confirmed by previous studies [1,2,4,6], and the ineffectiveness of temperature changes in the late season [1,2,3].
However, the definition of early and late is vague and impracticable. Unlike previous studies, this study included the months of a retail season as a predictor and weather factors. The month of a season is a significant predictor of seasonal clothing demand, and temperatures next followed the significant factors. The results indicated that the month of the winter retail season, minimum temperature, and maximum temperature were significant predictors of consumers' searching for winter clothing.
In November and December, the minimum temperature was the critical predictor, but the maximum temperature was the critical predictor in October and January. If the minimum temperature is higher than 4.167°C in November and December, the consumers' searching for winter clothing activity was relatively weak. When the minimum temperature in November and December is lower than 4.167℃, consumers search for winter clothing in November is the strongest, and December is also relatively strong. In October and January, when the maximum temperature is higher than 18.611℃, consumers' winter clothing search activity is the weakest.

Discussion
Since we have experienced new normal weather, increased the variation of temperature and precipitation and the higher occurrence of extreme weather events [18,39], we initiated this study to question whether clothing retailers' traditional sales planning method is valid and reliable for today. From the findings of previous studies, this study built a predictive model in which a clothing retailer can control the seasonal sales plan in response to weather changes during a season.
Unlike previous research, this study adopted four approaches: First, Google Trends is used as a consumer demand indicator. Although GT is a relatively new method to predict consumers' clothing demand, it provides new data sets to complement survey data [27]. Since online shopping is part of everyday life and clothing retailers run multichannel stores, GT provides more holistic data, including brick-and-mortar stores, online stores, and omnichannel stores. Second, clothing retailers have suffered enormously from new normal weather over the past decade. This study used and analyzed weather observation data for the past 11 years to understand the historical consumers' response to weather changes. In carrying out place-specific research in NYC, we proposed a model for clothing retailers can develop a plan of action to deal with uncertainties. Third, consumer demand varies from month to month during a season, and the weather does not match the seasonality of temperature and the high incidence of extreme events. Thus, this study assessed the relationship between the whole winter retail month and monthly consumers' demand and weather factors to find the change in monthly consumers' demand during a season. Last, since the seasonal clothing demand associated with weather changes cannot be explained with a linear model [2,3], this study used a decision tree to reveal all possible outcomes of a decision, trace each path to a conclusion, and assess the importance of predictor variables.
The results of the analysis of the past 11 years of data are as follows. First, month, minimum temperature, and maximum temperature were the significant predictors of consumers' search for winter clothes. Consumers' demand pattern for each month was identified. The coldest month in NYC is January, based on the 1981-2010 Climate Normals, but consumers' demand was the highest in November. December, October, and January followed. Second, overall, demand for winter clothing increases when the temperature drops, but the demand of consumers every month varies depending on the weather. While the minimum temperature is a key driver that increases consumers' demand in November, the maximum temperature is a key driver that increases consumers' demand in October, December, and January. Third, consumers' demand increased due to extreme weather, such as snowfall in NYC in October.

Conclusions
Our results have important managerial implications. Clothing retailers will be able to procure seasonal clothing products based on a long-term weather forecast. For example, clothing retailers choose materials, fabrics, and designs that appropriate the weather, especially the temperature. With a short-term weather forecast, the clothing retailers will able to determine when to launch new products with price competitiveness in stores and reduce prices according to weather changes in the middle of the season. Rapid weather changes affect consumers' demand. Clothing retailers can apply the predictive model to quickly respond to unexpected weather changes, sufficiently prepare products with rapidly increasing demand not to miss sales opportunities, and adjust quantities and prices for products with sharp declines in demand. Thus, profit can be maximized, and unsold, excessive products can be minimized.
Even though we provide significant predictors and decision-making rules, consumer demand cannot predict with only weather factors. Consumers shop for new clothes for real needs and psychological enjoyment. Besides, clothing consumers' purchasing decision-making process depends on clothing characteristics such as style, quality, and price and is related to the interaction of various factors such as economy, society, and ecological environment. Hence, the results reduce the explanatory power of this study to describe considerable seasonal clothing demand. Since 11 years of data were used, it is insufficient to be used as a climate change model and inappropriate to apply to other places because this study is based on a place-specific case study. However, this study contributes to the building body of knowledge about weather changes in clothing sales by confirming the results of previous studies. It is also valuable to provide the importance of predictors and decision rules for new seasonal clothing demands in the new normal weather era.

Ethics approval and consent to participlate
Not applicable

Consent for publication
Not applicable

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding authors on request.

• Competing interests
The authors declare that they have no competing interests. The relative importance of predictors to Winter Jacket Google Trends Figure 3 The results of the trained tree diagram.