The impact of a lenient return policy on
return rate
2
The impact of a lenient return policy on
return rate
1rd edition, June, 2016
LONNEKE EIJDEMS University of Groningen Faculty of Economics and Business
Marketing department Master Thesis Westerbinnensingel 42A 9718 BV Groningen +31 6 53 98 89 31 l.r.m.eijdems@student.rug.nl s2092174
3
Management summary
Especially with the rise of e-commerce, defining the proper return policy is of vital
importance to minimalize product returns for a retailer. There has been a debate about the impact of return policies on the return rate, but results are inconclusive. Wood (2001) favors a lenient return policy, as her research yields evidence that it will increase net sales. She states that a lenient return policy increases purchase proclivity, without a similar increase in product returns. The purchase proclivity has been investigated a lot in the academic field (Wood, 2001; Ketzenberg & Zuidwijk, 2009; Wang, 2009; Janakiraman et al., 2015) and is not covered in this paper, since all papers prove that a lenient return policy increases purchase proclivity. Opposite to Wood (2001), Davis and Gerstner (1998) found that a lenient return policy does increase the number of returns. These contradictory results fostered new research into product returns and the impact of a return policy on the return rate. Janakiraman et al. (2015) conducted a meta-analysis of recent studies that took product return in combination with return policies into account. Based on their analysis, four dimensions are tested, which are the time that a customer can return the product, the monetary costs with regard to the return for the customer, the scope of the products that can be returned and the amount of effort that a customer needs to put into the process of returning a product. Together with these four dimensions, a fifth and sixth dimension is defined in this study. This fifth dimensions incorporates the way in which the return policy is stated on the retailer’s website and the sixth dimension controls for the size of the company. The size of the company is also used as a moderator. These dimensions are analyzed based on secondary data of a large wholesaler in the Netherlands, Quantore, which operates in the the office supply industry. The data captures the return rate of 20 retailers for 9 quarters. The retailers operate in the Netherlands and have Quantore as their main wholesaler, although some might have a different wholesaler for some specific products. The data is based on all products that are ordered at Quantore and returned to Quantore. Since there is data over time, a random effects panel analysis is done to examine the effect of the different dimensions on the retailer’s return rate. The return rate ranges from 1.16% to 23.67%, with an average of 6.07%.
4 policy changed over the last 3 years. The size of the company is predicated upon the number of orders placed per quarter.
To estimate the model, a random effect model is tested, based on panel data with 171 observations. Two models are estimated, where the second model includes a lagged dependent variable. The study shows that the lagged dependent variable is an important predictor for the future return rate. Further results prove that monetary costs and the timeframe in which a product can be returned affect the return rate significantly. When only the return costs are charged, the return rate is 49.6% lower compared to extra fixed return costs and 107.3% lower compared to extra return costs that are based on a percentage of the order. The timeframe only differs between 7 and 30 days, where 30 days increases the return rate with 71.1%. The amount of effort also affects the return rate, where a higher effort leads to a higher return rate. No significant results are found for the number of restrictions placed on products that can be returned, the communication of the return policy on the retailer’s website or the size of the company.
Model 2 outperforms model 1 in explaining the difference in return rate (33.84% explained vs. 18.56% explained). This indicates that model 2 is preferred over model 1.
All results taken into consideration, a lenient return policy lowers the return rate and is therefore preferred. For managers, this has several implications, namely that if the retailer wants to keep their returns as low as possible, it should adopt a more lenient return policy, instead of a strict return policy. A tool is created to allow retailers to easily compare the difference in expected return rate when they change a dimension of their return policy.
5
Preface
First of all, I would like to thank prof. Dr. T. H. A. Bijmolt for his advice and feedback in the process of writing this master thesis. By meeting face-to-face meetings with my peer
students and the emails in between these meetings, I was able to finish this master thesis. I also would like to thank my second supervisor, A. Minnema, for grading my thesis.
Furthermore, I would like to thank Quantore for providing the data that was needed to perform this study. Special thanks go to Matthijs Donkers and Mike Vaneker for the
cooperation. Lastly, I would like to thank my family and friends, who supported me during the entire process.
6
Table of Contents
Management summary ... 3 Preface ... 5 1. Introduction ... 8 1.1 Background information ... 8 1.2 Research questions ... 81.3 Introduction to the company ... 9
1.4 Relevance of the study ... 10
1.5 Structure of the thesis ... 10
2. Theoretical framework ... 11
2.1 Conflicting study results ... 11
2.2 Product evaluations ... 11
2.3 Dissonance after purchase ... 12
2.4 Return policy dimensions ... 13
2.4.1 Time ... 13
2.4.2 Monetary costs ... 14
2.4.3 Restrictions ... 15
2.4.4 Effort ... 15
2.4.5 Communication ... 16
2.4.6 Size of the company ... 17
2.5 Conceptual framework ... 18
3. Research design ... 19
3.1 Data collection and research method ... 19
3.2 Plan of analysis ... 20
3.3 Model specification ... 24
4. Results ... 26
4.1 Descriptive statistics ... 26
7 4.3 Estimation... 28 4.4 Validation ... 30 4.4.1 Statistical validity ... 30 4.4.2 Predictive validity ... 31 5. Conclusion ... 33 5.1 Timeframe ... 33 5.2 Monetary costs ... 33 5.3 Restrictions ... 34 5.4 Effort ... 34 5.5 Communication ... 34 5.6 Size ... 34
5.7 Lagged return rate ... 34
5.8 Overall conclusion ... 35 6. Recommendations ... 35 6.1 Managerial recommendations ... 35 6.2 Research recommendations ... 36 References... 39 Online references ... 44 Appendix ... 46
Appendix A: Descriptive statistics ... 46
Appendix B: Dummy coding ... 47
8
1. Introduction
1.1 Background information
With the rise of e-commerce, retailers face new challenges. One of these challenges is the increased number of product returns (Guide et al., 2006). Since product returns cost the firm lots of money - estimates have shown that product returns cost retailers on average $100 billion per year, which equals a 3.8% decrease in profits (Blanchard, 2007; Stock et al., 2002) - decreasing them has become a point of interest in marketing research. A partial explanation for the high product return rate for online purchases is the lack of physical contact with the product in the pre-purchase stage. This consequence of online buying results in customers who are more likely to be dissatisfied when receiving the product. Therefore, the likelihood of a product return is much higher for online shopping (Ofek et al. 2011). The absence of “sense and feel” of the product makes an online purchase more risky, which reduces the probability that customers will eventually buy the product (Wood, 2001; Bahn & Boyd, 2014). In order to reduce the risk of making a bad decision, flexibility is needed to reduce uncertainty and motivate customers to buy online.
One way to reduce uncertainty for customers is to adopt a lenient return policy. A lenient return policy reduces the burden for customers to return a product (Van den Poel & Leunis, 1999), as it allows to mimic the flexibility of offline buying (Ketzenberg & Zuidwijk, 2009). Stiner (2004) conducted a survey about return policies, indicating that there are many different return policies across retailers, even in similar product categories. Research has already proven that a lenient return policy increases purchase proclivity (Wood, 2001; Ketzenberg & Zuidwijk, 2009; Wang, 2009; Janakiraman et al, 2015). A survey from Market Wire (2007) concluded that 90% of the customers value a lenient policy and that online purchases would increase with 70% when the activity of returning the product would be more convenient. Thus, many studies provide evidence that a lenient return policy fosters purchase behavior. In order to make more purchases profitable for a firm, the product return rate should remain constant or decrease.
1.2 Research questions
9 reason for these inconclusive outcomes has been addressed by Janakiraman et al. (2015). Their research claims that previous research only investigated a few factors in return policies. Their meta-analysis suggests that returns do increase, but not as much as the increase in purchase proclivity. Since this increases net sales, a lenient return policy is preferred. To test their outcomes in an empirical setting, the following research question is defined:
What is the effect of a lenient or strict return policy on product returns?
To answer the main research question, several sub-questions are investigated:
1. What is the effect of the amount of costs involved in a return on the number of product returns?
2. What is the effect of the timeframe in which the products needs to be returned on the number of product returns?
3. What is the effect of the number of restrictions on returnable products on the number of product returns?
4. What is the effect of the effort a customer needs to put in the process of returning on the number of product returns?
5. What is the effect of the way a retailer communicates the return policies on the website on the number of product returns?
6. To what extent can the return rate in the previous quarter explain the return rate in the next quarter?
1.3 Introduction to the company
10
1.4 Relevance of the study
The practical advantage of the study has to do with the drawbacks of product returns. Returns do not only increase extra costs for retailers, but also create more cash-flow uncertainty (Heiman et al., 2001). Returns also increase the complexity for supply chain management and waste of products (Hess et al.,1996). Therefore, a firm can amplify net sales by minimizing product returns. Moreover, lower returns reduce handling costs. The factors that play an important role in determining the return decisions of customers are revealed.
The relevance in the research field is based on the meta-analysis of Janakiraman et al. (2015). Their analysis has not been tested in an empirical setting, where return policies from different retailers in the same industry and country are investigated. This is done in this study and makes their study practically more relevant for managerial purposes.
1.5 Structure of the thesis
The thesis is structured as follows: The next chapter outlines the theoretical framework to provide theoretical support for the hypotheses and identify important factors for our model based on literature from previous research. Then, in chapter three, the data collection and methodology are elaborated on. The discussed methodology is performed in the chapter four, where the results of the analysis are presented. Based on the previous chapter, chapter five covers the conclusions and managerial implications. The implications are divided
11
2. Theoretical framework
2.1 Conflicting study results
Return policies vary greatly across retailers, sometimes even within a firm (Preddy, 1998; Anderson et al., 2009; Suwelack & Krafft, 2012). Since research abounds that return policies do have an effect on returns (Davis et al., 1998; Petersen & Anderson, 2013; Janakiraman et al., 2015), exploring the factors that increase returns, help to define the best return policy for a retailer. The question that follows is whether to adopt a strict return policy or allow customers to return products easily with a lenient return policy.
Many retailers are reluctant to adopt a lenient return policy, as they expect an increase in returns instead of a decrease. People might argue that the easier it is to return a product, the more likely people will return. This anxiety for higher returns is not without any empirical support, since this argument draws on research conducted by Davis and Gerstner (1998). They found that products that can easily be returned are also more likely to be returned. Nevertheless, more recent research suggests that a lenient return policy does not necessarily need to increase returns, in contrast to what many retailers think (Janakiraman et al., 2015). The factors that support this point of view are based on the perceptions that consumers have about the retailer and its product, both in the pre-purchase stage, as in the post-purchase stage.
2.2 Product evaluations
The decision to return a product is an evaluation of both the product and the retailer. Brown and Dacin (1997) studied product evaluations in relation to the image of a company.
Corporate ability is “those associations related to the company's expertise in producing and delivering its outputs” (Brown & Dacin, 1997, P. 68). Their study is relevant for product returns, since a product is returned if the customer evaluates the product negatively (Wood, 2001). To improve product evaluations, customers need to believe that a company is able to meet their expectations. If the company fails to meet the demands of the customer, the product is likely to be returned. Lenient return policies enhance the perceived quality of the retailer and the product (Heiman et al., 2001; Bonifield et al., 2010). By offering a lenient return policy, the evaluation can be altered during the entire decision process.
12 policy suggests that the quality of the products meets the expectations. Customers will buy more, which leads to an increase in the purchase proclivity. Research suggests that a purchase increase does not need to increase the average return percentage. This leads to higher net sales (Petersen & Anderson, 2013). Wood (2001) researched the perceived quality of products when a lenient return policy was used. She discovered that a lenient return policy increases pre-purchase quality, but also quality in the post-purchase stage. As perceived quality reduces returns (Bonifield et al, 2010), a lenient policy leads to lower returns compared to a strict policy. The premise that follows is that a lenient policy increases perceived quality and improves the reputation of a retailer. When perceived quality and image are high, it also positively influences the valuation of the product in the post-purchase stage. This positive valuation in turn affects the level of returns negatively, which is
favorable for retailers (Walsh et al., 2016). From this line of reasoning, a lower return rate is expected for a lenient return policy.
An analogy with low-priced products can be made here. Shiv, Carmon & Ariely (2005) found that low priced products adverts a poor quality, even if the quality is the same as similar products. This effect is due to perceptions of the customers. The same mechanism might be in place for returns. A lenient return policy affects the quality of products in a positive way, which in turn reduces the likelihood of a return (Eisenbeiss et al., 2014).
2.3 Dissonance after purchase
Another ground that increases product returns, is when customers experience a dissonance after buying a product. It is therefore meaningful to prevent customers from this feeling. Powers and Jack (2013) found a direct link between a liberal return policy and the decrease in product dissonance. The underlying construct is the image of a retailer. A positive image decreases the dissonance and increases the preference for the bought item (Sweeney et al., 2000; Keaveney et al., 2007). Therefore, products are less likely to be rejected when
receiving it, lowering the likelihood of a return (Kim & Wansink, 2012; Lee, 2015).
Even when satisfaction cannot be reached for the product that has been bought, a return policy can still alter product perceptions. Since online sales do not allow for a physical product inspection, dissatisfaction is more likely to occur (Mukhopadhyay & Setaputra, 2007). Fruchter and Gerstner (1999) found that when retailers allow for refunds, it can work as insurance for dissatisfaction. These insurances, which are granted by a lenient return policy, motivate customers to buy the product and increase the perceived quality. All things considered, perceived quality can be altered by a lenient return policy and
13 The higher satisfaction decreases the returns. Therefore, theoretical evidence prefers a lenient product policy above a strict policy.
2.4 Return policy dimensions
Some previous studies have investigated return policy based on two monetary extremes, no return or a full return (Lee, 2001; Sarvary & Padmanabhan, 2001). However, Muhopadhyay & Setoputro (2004) already noted that these two extremes do not give a full representation. In order to analyze the effect of return policies, multiple factors play a role in classifying a return policy and these can vary per dimension. Even per dimensions, scales might differ from a full refund, to a partial refund, to no refund. Several studies have investigated these factors, whereas the number of factors varies between one and five. Heiman et al. (2001) developed a framework with four dimensions, whereas Janakiraman et al. (2015)
constructed a framework that categorizes return policies on five dimensions. Davis and Gerstner (1998) did this as well, based on two categories, named “no hassle” and “high hassle”. The first indicates a lenient policy and the latter a strict return policy. Again, these were constructed based on five dimensions. On the other hand, there are also many studies that only consider one factor (Bower & Maxham, 2012; Lantz & Hjort, 2013).
In this paper, product returns are analyzed based on six dimensions. These are listed below and explained in the following sections.
1. Time that a customer has to decide whether to return the product. 2. Monetary costs for a customer.
3. Effort involved in the process of returning a product.
4. Restrictions that are placed on returns, so the scope of the product that can be returned.
5. The clearness with which the return policy is communicated on the retailer’s website. 6. The size of the company.
2.4.1 Time
Time is a component that has been studied quite often for return polices (Ketzenberg & Zuidwijk, 2009; Wood, 2011; Janakiraman et al., 2015). It is also a highly valued dimension for a return policy, the second after a restocking fee (Posselt et al. 2008), in this study listed under monetary refund.
14 a Google application (Albanesius, 2012) to a return policy with no specific time to be
returned (Heiman et al., 2001).
For a retailer, the time to allow for returns can be used as a quality signal, since a longer period indicates a higher product quality level (Posselt et al., 2008). A higher perceived quality in turn results in lower product return rates. This is confirmed by Mosterd et al. (2005), who investigated the period with the highest return rate, which is between two to three weeks after purchase. This suggests that a return policy below 7 days or above 21 days decreases returns.
Other research suggests that the longer the time frame, the higher the perceived
endowment of the product will be, leading to lower returns (Janakiraman & Ordóñez, 2012). This is supported by the idea that a longer deadline reduces cognitive effort, which leads to a lower product return rate. Their research argues that reducing the timeframe will only increase product returns. Another reason to increase the return period is that consumers need to experience the product. After receiving it, they need a period in which they can perceive the benefits that the product has to offer. When retailers limit the time to return the product, the customer cannot fully understand all the benefits of the product (Davis et al., 1998). This is also in line with research done by Payne et al (1993). They found that a longer time period allowed for more cognitive processing. Customers who have more time to return a product are more capable of processing all the benefits.
In line with previous literature, a larger timeframe will decrease the return rate and a shorter time frame increases the return rate.
H1: The larger the time frame to return a product is, the lower the product return rate will be.
2.4.2 Monetary costs
As already discussed, the perceived risk when buying online is much higher than in
traditional brick-and-mortar stores. In order to encourage customers to buy online, refund methods can play a major role (Nasiry & Popescu, 2012). A money-back guarantee shows to be the most important risk reliever and increases the likelihood of buying online (Van der Poel & Leunis, 1999). Now it can be said that a refund increases the likelihood to buy, the questions arises whether it reduces returns as well.
15 that a customer can get for a product return. When these costs are high, a return has
financial implications for the customer. The other way around, when the costs are handled by the retailer, it reduces the risk but might also increase the likelihood that a product is returned. Therefore, retailers must decide how much cost a customer must bear for the return of a product. Davis and Gerstner (1998) and Yan (2009) argue for a partial refund, to tackle opportunistic behavior of customers. Even though product returns might increase with a full refund (e.g. no return costs) policy, research suggests that it increases the perceived trust in the retailer and also increases purchase intention. Despite these positive long-term effects, a negative relationship is expected between the level of monetary costs and the percentage of product returns.
H2: The higher the monetary costs for the customer to return a product are, the lower product return rate will be.
2.4.3 Restrictions
The restrictions include the items that are not allowed to be returned by the customer. The reason to include this, is that some retailers allow all products to be returned, while others limit their return policy to specified items. Some product categories show higher returns than others (Peterson & Kumar, 2013). In order to tackle these high returns, a retailer might limit the range of products that can be returned to lower the overall return rate.
On the other hand, an earlier discussion concluded that a higher perceived quality is linked to lower return rates. A limited range of products that can be returned might lower the perceived quality.
Literature states that no clear effect of the direction on return rates can be found for the number of restrictions. In order to test the effect, two opposing hypotheses are tested (Armstrong et al. 2001). As there is not enough prior knowledge, opposing hypotheses give way to test for both directions. The following hypotheses are tested in this study:
H3a: The higher the restrictions on returns are, the higher the return rate will be. H3b: The higher the restrictions on returns are, the lower the return rate will be.
2.4.4 Effort
16 customer needs to take to return an ordered product . For example, the ways in which customers need to notify the retailer plays a role in the amount of effort to return a product. Bonifield et al. (2010) found that 40% of the customers are reluctant to buy online, since the effort in returning a product is too high. This indicates that effort plays a major role in the decision to buy a product online or not. Griffis et al. (2012) also included the complexity of buying in their analysis of returns. The easier it is for customers to buy a product, the more likely they will make a purchase. Customers want to spend as little effort as possible in their shopping process (Clarke & Belk, 1979).
Owens and Hausknecht (1999) did a similar study with effort as a predictive variable, but for compliant handling purposes. When customers could easily make a complaint, they were more likely to do so. Based on their research, customers will be more eager to engage in a certain behavior, when it is easier to do so. The same might be true for returns. When it is easy to return a product, people are more likely to send it back. So making it easier for customers to return a product can increase the amount of purchases, but also the amount of returns.
H4: The more effort it takes to return a product, the lower the return rate will be.
2.4.5 Communication
Since many of the reasons explained in the previous sections have to do with considerations in the pre-purchase stage, customers need to be informed before buying a product. This information can be found on the website. The ease in which the information can be found and the quality of information can have an impact on the perceived quality of the product and the retailer (Park & Stoel, 2002). When a customer needs to invest a lot of time in searching for the right information, he or she might be distracted or annoyed. This might lead to less satisfied customers. As previously noted by the theory, a higher perceived quality might affect the number of returns. When the level of communication increases the
perceived quality, it will have a negative effect on the return rate of the retailer. Moreover, a return policy that is hard to find increases the effort that a customer needs to put into a purchase. Cardozo (2016) has done a recent study that reveals that more effort is related to lower satisfaction. As discussed before, a lower satisfaction increases the return rate. Based on the above literature, the following hypothesis is constructed.
17
2.4.6 Size of the company
Direct effect
The size of the company might have an impact on the return rate and is therefore included as a control variable. The size of the company is not included directly in the study of Janakiraman et al. (2015), but they do include variables that are company-specific. Size is used in this study as the company-specific variable, as all retailers operate in the same industry. The effect is of a large company is based on the assortment and perceived quality. Larger companies typically offer a broader assortment and serve more customers. In line with these findings, perceived quality is higher for large firms than for small firms (Berger et al., 2007). From previous discussed literature, perceived quality is an indicator for a lower return rate. Thus, a large firm will have lower product return rates than a small firm.
H6a: The larger the company, the lower the return rate will be.
Interaction effect
Companies need to adjust their return policy according to what they seem best. The best return policy is likely to be different for small companies than for large companies. To test this assumption, a moderation effect is included in the model. The impact of restrictions might be perceived as higher when the company is already small. Small companies typically adopt a smaller assortment. A smaller assortment decreases the flexibility that customers perceive. Since online purchases are linked to high uncertainty, a large assortment might work as an insurance and decrease uncertainty (Lehmann, 1991). Therefore, the impact of the number of restrictions on the return rate might be higher for small companies.
18
2.5 Conceptual framework
Based on the analysis of previous research, the conceptual framework for this study is displayed in figure 1. A lagged dependent variable will also be analyzed, which will be discussed in chapter three.
19
3. Research design
3.1 Data collection and research method
For this study, the data is gathered from different sources to collect enough data to perform the study.
First, secondary data is used which is collected from the database of Quantore. Quantore is a wholesaler for office supplies in the Netherlands and Belgium. Their clients include Dutch and Belgian retailers, but for this research, only Dutch companies are researched. Some retailers operate solely online; other retailers also have brick-and-mortar stores. However, all the retailers have a main focus on online sales. Most retailers operate in the western part of the Netherlands, although Quantore is located in the south-east. When considering the size of the orders made at Quantore, there are six main retailers who place more than 3000 orders per quarter. Eight retailers range between 1500 and 3000 orders and there are five small retailers, who place less than 1500 orders per quarter at Quantore.
Quantore provides the retailers with all kinds of products in the office supply category. Their product range includes mainly stationery products, but also cartridges for printers, computer accessories and other products relevant for the office supply industry. When a product is returned, the retailer declares the return at Quantore and the product is returned to
Quantore’s warehouse. Quantore charges the retailers for handling these returns with a fixed amount. Due to the new handling system that started in the last quarter of 2013, Quantore has information about the sales and returns to and from the retailers’ clients. For competitive reasons, the names of the retailers cannot be named in this study.
The data has been gathered from every quarter in 2014 and 2015 and also from the first quarter in 2016 and the last quarter in 2013 from twenty retailers. As Quantore started a new handling system for returns in the last period of 2013, these observations are not a fair representation and contain a lot of outliers and missing data. Therefore, this quarter is deleted from the analysis and nine quarters remain.
20 different cases, the retailers are asked to fill in a survey to prevent misinterpretation and missing data.
After the qualitative analysis on the websites, the retailers received an online survey about the return policy. The purpose of this extra questionnaire was to verify the return policy as stated on the website and to uncover extra requirements. Moreover, questions were asked about the difference in policy between B2B and B2C customers and whether the return policy in general has been changed over the last three years. Since most customers of the analyzed retailers are focusing on B2B customers, the B2B return policy will be used in the analysis. It is not possible to define exactly for every year whether the returns were B2B or B2C returns. Since 85.6 percent of the returns were based on B2B returns in 2016 (Appendix A, Table 8: B2B vs B2C in 2016-1), the return policy that is used for businesses will be taken as the leading return policy.
For all other quarters, data of purchases and returns is available. When dividing the number of returns by the number of purchases for each retailer per quarter, the return percentage is calculated. A bias might exist as the number of sales in a previous quarter might affect the return rate in the next quarter. This is not taken into account in this study, but no large sales fluctuations exist in the sample.
The data is based on all returns that the wholesaler receives from their retailers. Retailers include a reason for return, which can either be damaged, defect, too many, not enough or a wrong order. In order to analyze the impact of the different factors in a return policy, it is important to exclude the returns that are due to failures of the retailer. Since the policies for returning products that are defect or damaged caused by the retailer or production are similar across all retailers, these returns are taken out of the dataset. Classification of these returns is based on the accepted offers that have the reason “defect” or “damaged”. One of the companies is taken out of the dataset, as the company operates under the same name of another retailer that is investigated. Two of the nine observations are missing for this retailer and interpretation is therefore not reliable. For further analysis, the other 19 companies are investigated. As data is available for nine periods, a total sample of 171 observations is researched.
3.2 Plan of analysis
21 qualitative study that was done for every return policy on the website of the retailer. The category levels are in line with recent research of Janakiraman et al. (2015).
Table 1: Dummy coding scheme
Description 0 1 2
Time Timeframe that a product
can be returned
7 days 14 days 30 days
Money Monetary costs for the
customer to return a product Return costs Fixed amount (€9.95 - €10.00) Percentage (7.5%)
Restrictions Scope of the products that
can be returned
Low High
Effort Amount of effort that a
customer needs to put into the process of returning the product
Low Moderate High
Communication Way in which the return
policy is stated on the website
Unclear Clear
Size The size of the company,
based on the orders at Quantore < 3000 orders per period > 3000 orders per period
Time will be categorized according to a path-worth scheme. It is not likely that the difference in effect will be the same between 7 and 14 days as the effect between 14 and 30 days. Moreover, most retailers adopt policies based on a timeframe of 7, 14 or 30 days. For example, time periods of 17 days do not occur in the dataset. When retailers place a return period of 10 working days as standard, it is labelled as 14 days. Bahn and Boyd (2014) also analyzed the time that a product could be returned. They wield the same scheme. They allowed for a time frame between 1 and 30 days. However, since 1 day is not used by any of the retailers in the dataset, the minimum is set on 7 days.
22 return costs. The retailer takes all the handing costs and the customer only needs to pay for the delivery costs. In the other two options, the retailers ask for an additional fee to cover the return costs. This can either be done by a fixed amount, which equals €9,95 to €10,00 or by a fixed percentage, which equals 7.5-10%. It depends on the order amount what option is most economic for the customer. Since most customers are firms that typically order for a large amount at once, the fixed amount is taken as more lenient. When a percentage from the order is charged, it is treated as most strict. This is in line with the study of Bonifield et al. (2010), where the strict return policy is also treated as a percentage of the order. The restrictions are categorized according to low or high restrictions. Low restrictions are retailers who only limit the returns to products where personal characteristics from the client are printed on. High restrictions do not only limit their returns to these products, but also to specific product categories and items that are on sale or in the outlet store. A note should be made here, since there are some product categories that are never applicable for return or not every retailer offers this set of products. This includes cartridges for earplugs, printers, toner cassettes and software packages (Appendix B: Table 10: Coded restrictions). Hence, these products are ignored when defining the level of restrictions.
The effort that a client needs to put into a return is based on the same scale as restrictions, but with an extra level. Low effort indicates that the client only needs to fill in a form or has the opportunity to let someone collect the product. A moderate effort lacks this pick-up service, but has the same way of notifying the company that he or she wants to return the product. A high effort indicates that the client needs to send a separate email or call the client service of the company before the companies allows for a return. This requires additional steps and takes therefore more effort. No pick-up service is present in this the high effort case (Appendix C: Table 11: Coded effort).
Then, the communication of the return policy is coded in two levels, hard and easy to find (Appendix C: Table 12: Coded communication). Several factors are included, whether there is a direct link on the homepage, the way the return policy is displayed and whether all relevant information about time, money, effort and restrictions are covered. The way in which the return policy is displayed can differ between a clear overview of the return policy or when the customer needs to find the information in terms and conditions. As it takes more effort to find the return policy in the terms and conditions, displaying the return policy in terms and conditions is evaluated as "unclear".
23 For the research, different programs will be used. The return percentages are calculated in Microsoft Excel 2010 and exported to Stata and SPSS. Since SPSS 23 has no option for panel data, these analyses will be done in StataSE 14. Some descriptive statistics and tests are done in SPSS 23.
The dependent variable in this study is the return rate. A return can be treated as a binary variable, since customers decide to return or not. The return rate percentage is a number between 0 and 1, as it shows the probability of a return in that period. Here, a 0 indicates no returns at all and 1 would indicate that every offer is sent back. The return percentage rate will have a value between 0 and 1. Due to the binary characteristics of the dependent variable, no normal distribution can be found for the error term. Moreover, many outcomes are close to 0 and therefore, a transformation is done. The dependent variable is
transformed by the following function to create a log-odds ratio (Tekens & Koerts, 1972).
After the transformation, a multivariate regression can be done based on panel data, also known as longitudinal data (Leeflang et al. 2015). The panel data allows for the same cross-sectional data that is used over a specific time period (Wooldridge, 2010).
Since the tested coefficients are log-odds ratios, it is hard to interpret those. A logit transformation, done by taking the exponent of the coefficient, allows for a more intuitive interpretation. The coefficient can now be interpreted as the odds ratio.
In order to properly interpret the results, an anti-logit transformation is done after
estimation. The anti-logit transformation after estimation allows for direct interpretation of the results, which increases the practical relevance. The anti-logit transformation is done by the following equation (Tekens & Koerts, 1972).
24
3.3 Model specification
For the estimation, two models are estimated. The difference between the models is that the first model is without a lagged dependent variable, where the second model is with a lagged dependent variable.
The reason to run both models is that the captures the combined effect on y of all variables that are unobserved and do not change over time. It is also independent from all the explanatory variables that explain y. Here is where a statistical problem arises. If affects , it cannot be independent of (Allison, 2015). Since the independence of
is a requirement for a random effect model, a lagged variable will violate this assumption. Despite the statistical problems, lagged dependent variables are of major importance for managers (West, 1996, Leeflang et al. 2015). Returns in the past can be an important predictor of returns in the future. For practical purposes, a second model will be generated, including the lagged variable. Model 1 will therefore be a nested model of model 2.
Since a random effect model is used, the model will look like the one below
Where:
β0 = Intercept
T1= A dummy variable for the time that a product can be returned, 1 = 14 days a 0 = otherwise
T2 = A dummy variable for the time a product can be returned, 1 = 30 days, a 0 = otherwise M1 = A dummy variable for the monetary costs for the customer, 1= fixed amount, 0 = otherwise
M2 = A dummy variable for the monetary costs for the customer, 1 = percentage based, 0 = otherwise
R = A dummy variable for the number of restrictions placed on the products that can be returned, 1 = high, 0 = low.
E1 = A dummy variable for the amount of effort a customer needs to put in the process of returning a product, 1 = medium, 0 = otherwise
E2 = A dummy variable for the amount of effort a customer needs to put in the process of returning a product, 1 = high, 0 = otherwise
25 S = A dummy variable for the size of the company, 1 = larger than 3000 orders per quarter, 0 = less than 3000 orders per quarter
RS = Interaction effect of the size of the company on the importance of the restrictions U = Random intercept
ε = Error term
i = Company, 1, 2 …. 19.
t = period 2014-1, 2014-2 …. 2016-1 = Rho correction for autocorrelation
However, when a lagged variable is included, the model will look like this:
Where:
Yi(t-1) = Return rate in the previous quarter (t-1)
A Hausman test will be used to decide whether it is allowed to use a random effects model. It tests the null hypothesis that the unique errors are not correlated with its regressors (Greene, 2008). A random effects model is preferred in times of efficiency. A Breusch-Pagan Lagrange test is done to verify the Hausman test.
The two models will be compared based on the Akaike's entropy-based Information Criterion (AIC) and the Bayesian information criterion (BIC) (Burnham & Anderson, 2004). The BIC has a stricter penalty for an additional parameter, which is the case is model 2. This can only be done by computing the log Likelihood of the model. Therefore, next to a GLS regression, a maximum likelihood regression will be performed, in order to compute the log likelihood. Based on these two criteria, the best model fit will be selected. Akaike’s information criterion (1998) is defined by
Schwarz’s Bayesian information criterion (1978) is defined by
26 The model comparison is done based on the log likelihood of the model. Moreover, the Mean Absolute Percentage Error (MAPE) is calculated to assess the predictive validity. The formula of the MAPE is shown below (Leeflang et al, 2014).
4. Results
4.1 Descriptive statistics
From the 171 observations, there is an average return rate of 6.07%, where the highest return rate is 23.67%. The minimal return rate is 1.16%. In Figure 2: Return percentage over time, the return percentage of five retailers is shown, to provide an indication of the return behavior over time. Although there are some fluctuations per retailer and per quarter, the return rate behaves quite stable over time.
In Figure 4: Return rate per company, which can be found in appendix A, the return
percentage of all retailers is plotted. Here can be seen that the return rates are quite stable, with some exceptions. For some retailers, this is due to the limited number of returns per quarter, making it more vulnerable to large fluctuations. No correction is made for these large fluctuations.
Figure 2: Return percentage over time
0,00% 2,00% 4,00% 6,00% 8,00% 10,00% 12,00% 14,00% 16,00% 1 2 3 4 5 6 7 8 9 R e tu rn p e rc e n tage
Return percentage over time
27 In Table 2: Frequency table , the distribution is shown for the return dimensions of the retailers. Most retailers give customers a limited time to return the products and only four retailers allow for a return after one month. The level of monetary return is almost equally divided, whereas the amount of effort is not. Most retailers offer a low effort return policy. However, restrictions on the products are very high, were only two retailers allow for almost all products to be returned.
Table 2: Frequency table of the independent variables
Time Money Effort Restriction Communication Size
0 8 6 9 2 11 9
1 9 8 4 0 8 11
2 4 5 6 17 N.A. N.A.
4.2 Pre-estimation
No multicollinearity is detected for the independent variables. The lagged dependent variable shows the highest correlation with the return percentage. This hints that the lagged
dependent variable has a lot of explanatory power.
The variables are treated as random effect variables, to account for the differences in return rate per retailer over time. For that reason, a fixed effect model cannot be used (Hedges & Vevea, 1998). A fixed effect model includes the variance in the fixed effect dummy or error term. The purpose of the model, to define whether the variance is due to the return policy, cannot be tested with a fixed effects model. To provide statistical evidence for the use of a random effect model, a Hausman test is performed (Hausman, 1978). The Hausman test compares the standard error of a random effect model with the standard error of a fixed effect model. Ideally, the standard error of the random effect model is smaller. The Hausman test is done in Stata, by estimating both models and storing the error term. The results of the Hausman test show a chi-square of .4568, which is not significant within an alpha of .1. Therefore, the null hypothesis is rejected and allows for a random effects model. The Breusch-Pagan Lagrange multiplier (LM) (Breusch & Pagan, 1980) is also a statistic to test between the fixed or random effects model. In line with the Hausman test, the LM also favors a random-effects model (p<.05). Therefore, a random effect model is most
28 To test for serial correlation, also known as autocorrelation, a Wooldridge test is done, which is based on an F-test (Drukker, 2003). The Wooldridge test is specified to test
autocorrelation for data over time. For both models, autocorrelation is present (Model 1: F (1, 18) = 21.384, p<.05. Model 2: F (1, 18) = 49.109, p <.05).
To correct for autocorrelation, the data needs to be transformed based on the rho, as shown in Table 3: Autocorrelation. After revising the model by adding the rho, the autocorrelation was not only removed, but the results of the GLS also improved.
F-test P-value Rho for AR
Model 1: Without lag_DV 21.384 0.002 0.414
Model 2: With lag_DV 49.109 0.000 0.662
Table 3: Autocorrelation
4.3 Estimation
The first model is estimated based on an intercept, eight dummy variables, a random intercept and an error term. To interpret the beta’s directly, a transformation needs to be done. This is done in the third column in *= significant for
**=significant for
Table 4: Estimation model 1, by taking the exponent of β. By subtracting the exponent of β with 1, the percentage change from the reference level can be interpreted. All the
coefficients are odds ratio’s and need to be interpreted accordingly.
From the table, the intercept (β=-3.858, p<.05) and the dummies for monetary cost (Dum1: β=.442, p<.1. Dum2: β=.715, p<.05) are significant. When applying a larger threshold for the significance, ( , the second dummy for time is also significant (β=.530, p=.062). The random intercept has a value of .262 and the error term has a value of .440 (Appendix C: Table 13: Random intercept and error term). The other variables tested in the model are not significant.
When looking at the time that a product can be returned (time), it is clear that the first dummy is highly insignificant. Though, no significant difference is found between a timeframe of 7 days and 14 days.
The second dummy for effort also shows a very high p-value (β=.031, p=.925), indicating that this variable is not significant at all.
The same holds for the communication variable. The way in which the return policy is
29 a p-value of .905 (β=-.041), the variable is highly insignificant. Hence, it is taken out of the model.
Model 1: No Lagged_DV
Independent variable β Exp(β) Exp(β)-1 Std. Err. P-value
Timeframe (Dummy 1) 0.067 1.069 0.069 0.271 0.806
Timeframe (Dummy 2) 0.537 1.711 0.711 0.306 0.080**
Monetary cost (Dummy 1) 0.442 1.556 0.556 0.230 0.055**
Monetary cost (Dummy 2) 0.715 2.044 1.044 0.283 0.011*
Effort (Dummy 1) 0.561 1.753 0.753 0.376 0.135 Effort (Dummy 2) 0.006 1.006 0.006 0.276 0.983 Restriction 0.457 1.579 0.579 0.522 0.382 Communication -0.034 0.9660 -0.034 0.250 0.891 Size 0.013 1.013 0.013 0.251 0.958 Restriction#Size -0.308 0.735 -0.265 0.522 0.555 Intercept -3.858 0.021 -0.979 0.605 0.000* *= significant for **=significant for
Table 4: Estimation model 1
The second model is estimated with a lagged dependent variable, as can be seen in Table 5: Estimation model 2. Due to the lagged dependent variable, the model loses 1 period of data for quarter 2014-1. The random intercept has a value of .278 and the error term of model 2 is .291 (Appendix C: Table 13: Random intercept and error term).
The constant in the model is significant (β=-3.795, p<.05,). The same is true for the lagged variable (β=6.340, p<.05,). This is a highly significant variable with a high coefficient,
indicating that the previous return percentage has a large influence on the return percentage of the current quarter. In model two, the monetary costs variables (Dum1: β=.403, p<.05 and Dum2: β=.729 p<.1) are both significant. Therefore, the costs for the return itself are significant predictors for the variance in return rates.
30
Model 2: with lagged DV
Independent variables β exp(β) Exp(β)-1 Std. Err. P>z
Return rate in t-1 6.340 567.051 566.051 0.600 0.000*
Timeframe (Dummy 1) 0.034 1.0347 0.035 0.270 0.899
Timeframe (Dummy 2) 0.480 1.617 0.617 0.307 0.118
Monetary cost (Dummy 1) 0.403 1.496 0.496 0.237 0.089**
Monetary cost (Dummy 2) 0.729 2.073 1.073 0.288 0.011*
Effort (Dummy 1) 0.651 1.917 0.917 0.387 0.093** Effort (Dummy 2) 0.055 1.056 0.056 0.270 0.839 Restriction 0.486 1.626 0.626 0.540 0.368 Communication -0.127 0.880 -0.120 0.258 0.622 Size 0.013 1.013 0.013 0.199 0.948 Restriction#Size -0.378 0.685 -0.315 0.507 0.456 Intercept -3.795 0.022 -0.978 0.623 0.000* *= significant for **=significant for
Table 5: Estimation model 2
Not all variables show sufficient explanatory power for the return rate. These variables show a low significance level in both models. However, literature shows that it does have an impact. Therefore, no variables are taken out of the model. Besides that, the outcomes contain relevant findings. When no significant effect is found for a dimension, it indicates that is does not enlarge the return rate. Therefore, a lenient return policy does not
necessarily increase return rates. A return policy with very strict conditions might decrease the sales, but does not decrease the return rate. The other significant results even show that dimensions of a return policy can lower the return rate. In the next session, some relevant conclusions and managerial implications from these results will be elaborated on.
4.4 Validation
4.4.1 Statistical validity
31 certifies the level of explained variance in the dependent variables by the explanatory
variables. In the first model, when all the variables are included except the lagged dependent variable, 18.56% of the variance in the return percentage is explained by the model.
However, when the lagged variable is included in model 2, 33.84% is explained, which is 15.28% more than without the lagged dependent variable. This endorses the large impact of a lagged return percentage on the current return percentage in period t.
For both model 1 and 2, an F-test is done to check the significance of the overall model. For model 1, the Wald chi statistic is 11.30 and the F-test computes a p-value of .1851. Within an alpha of .1, the model is in both ways not significant.
Model 2 has a Wald chi statistic of 143.35 and the F-test shows a p-value of .000, indicating that the model is highly significant.
LL (Null model) LL AIC BIC R2
Model 1 -121.8537 -117.7981 255.5962 287.0129 18.56
Model 2 -121.8537 -90.83431 203.6686 238.2269 33.84
Table 6: Validation results
From table 6, both models outperform the null model. This indicates that the independent variables do explain more than a model with only the constant. Based on the AIC and BIC, model 2 is better than model 1. The difference is smaller for the BIC, as the BIC is stricter on adding an extra parameter. However, the lagged variable explains enough to cope with this extra penalty, which was already expected from the estimation. Although model 2 loses 9 variables due to the lagged variable, the information criteria and R-square outweigh the disadvantage of losing one period of data. Moreover, the variables are more significant in model 2 than in model 1. Based on both outcomes, the extra parameter in model 2 adds sufficient explanatory power and increases the model fit. Therefore, model 2 is preferred over model 1.
4.4.2 Predictive validity
32 -4,5 -4 -3,5 -3 -2,5 -2 -1,5 -1 -0,5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
Actual return rate Predicted return rate
(2014-2 until 2015-3). The last 2 quarters are used here as a holdout sample for validation (2015-4 and 2016-1). Therefore, T* has 114 observations and T-T* has 38 observations. The MAPE shows a percentage of 15.91%. This indicates that 15.91% of the actual value is the absolute error in this model.
In Figure 3: Actual vs predicted return rate, two lines are plotted. The predicted return rate is based on the model 2. From the graph, it is visible that the predicted return rate follows a similar pattern as the actual return rate. Deviations exist, as the MAPE pointed out, but the model is able to predict the returns quite well.
33
5. Conclusion
Based on the results generated from the analysis, several conclusions are drawn. The outcomes of hypotheses that were tested are shown in Table 7: Hypothesis overview. For interpretation of the coefficients, the second model is used as it outperforms the first model.
Model 1 Model 2
H1: Time Partly accepted Rejected
H2: Money Accepted Accepted
H3a: Restrictions (+) Rejected Rejected
H3b: Restrictions (–) Rejected Rejected
H4: Effort Rejected Partly accepted
H5: Communication Rejected Rejected
H6a: Size Rejected Rejected
H6b: Restriction*Size Rejected Rejected
Table 7: Hypothesis overview
5.1 Timeframe
For the timeframe to return a product, a positive relationship is found between a 7 day and a 30 day return policy in model 1. This indicates that a timeframe of 30 days is not favorable for a retailer, as it increases the return rate by 71.1%. However, no significant results can be found for a 7 and 14 day timeframe. Based on this result, a 14 day return policy increases the perceived quality (Posselt et al. 2008) and does not increase the results significantly compared to a 7 day return policy. Hence, the optimal value is to allow for a return period of 14 days.
5.2 Monetary costs
34 tactic has the opposite effect. A more expensive return policy is linked to a higher return rate.
5.3 Restrictions
No significant effect can be found for the restrictions a retailer poses on the returnable products. Although no significant results can confirm one of the opposing hypotheses, the insignificant coefficient shows that higher restrictions increase the return rate with 62.6%, which is in line with hypothesis H3a. As said, no evidence can be found for the significance of the hypothesis for the number of restrictions.
5.4 Effort
A negative relation was expected between the number of returns and the amount of effort needed to return. This relation is only visible for the difference between a low and high effort rate. The more effort a customer must put into the process, the lower the return rate will be (Clarke & Belk, 1979; Owens & Hausknecht, 1999; Griffis et al., 2012). A medium effort level increases the return rate by 91.7%, which might indicate that the difference between low and medium effort is too small and does therefore not explain the difference in return rate.
5.5 Communication
Hypothesis 5 is rejected, implicating that there is no significant effect for the way the return policy is communicated on the return rate. Still, since other variables show an effect, it is wise for a retailer to be transparent about the return policy. The positive effects of a lenient return policy only apply when the customer is aware of the return policy. When a company adopts a strict return policy, purchase proclivity might increase as the customers are less likely to be aware of the strict return policy.
5.6 Size
The size of the company has no effect on the differences in return rate. This indicates that the size of the company, related to the assortment, do not affect the return rate. Moreover, there is also no significant interaction effect between restrictions and size. A possible explanation might be that customers do not perceive the size of a retailer when it operates online. The perceived quality does not differ when there is no perceived difference in size. Therefore, the interaction effect does not hold either.
5.7 Lagged return rate
35 Other factors, both internal and external might play a role for explaining return rates. These were captured in the return period one time before. Possible explanations are the area in which the retailer is based, the type of customers that a retailer has or the experience that customers have with returning products. However, it might also be due to the lack of quality or the poor quality of information on the website. These variables are not included in this research, but are likely predictors of return rates. Since these characteristics are probable to be the same in this current quarter and in the previous quarter, it explains why the lagged return rate plays such a large role in predicting a return policy.
5.8 Overall conclusion
The overall conclusion is that a lenient return policy is linked to a lower return rate. A lenient policy might help to decrease the return rate, when two retailers come from the same starting point. However, independently from any return policy, fluctuation of return rates is due to other factors. These factors might be external or internal. Nevertheless, a stricter return rate does not decrease the return rate. The opposite is true, as a strict return policy seems to increase the return rate. Therefore, this study argues that a lenient return policy is the best return policy.
6. Recommendations
6.1 Managerial recommendations
The results of this study provide further guidance for managers who need to define the return policy for their online sales. When deciding which return policy to adopt, a focus on perceived quality of the customer is an important point to bear in mind. A stricter return policy is not necessarily linked to a lower return rate. The results show that it is wise to adopt a lenient return policy, to increase the perceived quality and by those means reduce the return rate. Trying to cover the handling costs by reckoning them up shows the opposite effect, based on the set of retailers that were analyzed. To set the optimal time frame, 14 days is the best option. It does not increase the return rate by giving too much freedom, but provides the customer with enough time to make a deliberate choice whether to keep the product or not.
The restriction on certain products should only be placed on products that cannot be sold otherwise. Think of personalized products or products that cannot be returned for sanitary reasons (e.g. earplugs). All other products should be available for return. No
36 it has on the return rate.
When retailers want to decrease the return rate, they should take a look at the previous return rates and uncover what factors influence these rates. A lenient return policy can help, but other factors might also be important.
On page 38, a table is shown that can be used by retailers to check the change in return rate when adopting a different return policy. The coefficients are included in the table and
retailers can choose one option per dimension. As the dimensions are mutually exclusive, the table will give an error (see the red highlighted area) when two options for one dimension are chosen. The table provides the retailer with a prediction of the return rate, based on the return policy dimensions, the size of the company and the return rate in the previous period. By choosing for another dimension, the return rate is adjusted accordingly.
6.2 Research recommendations
This study has been executed based on product returns in the Dutch market only. Previous research already pointed out that there are differences in return frequency in different countries. For example, in North America, product returns tend to be much higher than in the European market (Guide et al., 2006).
Moreover, Bahn and Boyd (2014) found that no return policy fits all products. Across product categories and even within assortments, return policies might vary. Although this research is only based on the office supplies wholesalers, even conclusions within this specific product category might differ across certain products. For further research, data from different industries must be used to expand the results to other fields.
A major drawback from the sample is that there was no specified difference between B2B and B2C companies. Further research should investigate whether it would have a large difference on the decision for a customer to return a product. No research is done so far and since the majority in this sample is B2B, the focus is on non-private customers.
39
References
Akaike, H. (1998). Information theory and an extension of the maximum likelihood principle. In Selected Papers of Hirotugu Akaike (pp. 199-213). Springer New York.
Albanesius, C., (2012). Google Sued Over Bogus Android Apps, Return Policy. PC Magazine, March 2012, 1
Altug, M. S., & Aydinliyim, T. (2016). Counteracting Strategic Purchase Deferrals: The Impact of Online Retailers, Return Policy Decisions. Manufacturing & Service Operations
Management, (20160125).
Anderson, E. T., Hansen, K., & Simester, D. (2009). The option value of returns: Theory and empirical evidence. Marketing Science, 28(3), 405-423.
Armstrong, J. S., Brodie, R. J., & Parsons, A. G. (2001). Hypotheses in marketing science: Literature review and publication audit. Marketing Letters, 12(2), 171-187.
Baayen, R. H. (2008). Analyzing linguistic data: A practical introduction to statistics using R.
Cambridge University Press.
Bahn, K. D., & Boyd, E. (2014). Information and its impact on consumers׳ reactions to restrictive return policies. Journal of Retailing and Consumer Services, 21(4), 415-423. Berger, J., Draganska, M., & Simonson, I. (2007). The influence of product variety on brand perception and choice. Marketing Science, 26(4), 460-472.
Blanchard, David (2007), "Supply Chains Also Work in Reverse,”. Industry Week.
Bonifield, C., Cole, C., & Schultz, R. L. (2010). Product returns on the Internet: A case of mixed signals? Journal of Business Research, 63(9), 1058-1065.
40 Breusch, T., & Pagan, A. (1980). The Lagrange Multiplier Test and its Applications to Model Specification in Econometrics. The Review of Economic Studies,47(1), 239-253.
Burnham, K. P., & Anderson, D. R. (2004). Multimodel inference understanding AIC and BIC in model selection. Sociological methods & research, 33(2), 261-304.
Davis, S., Hagerty, M., & Gerstner, E. (1998). Return policies and the optimal level of “hassle”. Journal of Economics and Business, 50(5), 445-460.
Clarke, K., & Belk, R. W. (1979). The effects of product involvement and task definition on anticipated consumer effort. Advances in consumer research, 6(1).
Drukker, D. M. (2003). Testing for serial correlation in linear panel-data models. Stata Journal, 3(2), 168-177.
Eisenbeiss, M., Cornelißen, M., Backhaus, K., & Hoyer, W. D. (2014). Nonlinear and
asymmetric returns on customer satisfaction: do they vary across situations and consumers?
Journal of the Academy of Marketing Science, 42(3), 242-263.
Fruchter, G. E., & Gerstner, E. (1999). Selling with “satisfaction guaranteed”. Journal of Service Research, 1(4), 313-323.
Greene, W. H. (2008). The econometric approach to efficiency analysis. The measurement of productive efficiency and productivity growth, 92-250.
Guide Jr, V. D. R., Souza, G. C., Van Wassenhove, L. N., & Blackburn, J. D. (2006). Time value of commercial product returns. Management Science, 52(8), 1200-1214.
Hausman, J. A. (1978). Specification tests in econometrics. Econometrica: Journal of the Econometric Society, 1251-1271.
Hedges, L. V., & Vevea, J. L. (1998). Fixed-and random-effects models in meta-analysis.
41 Heiman, A., McWilliams, B., & Zilberman, D. (2001). Demonstrations and money-back
guarantees: Market mechanisms to reduce uncertainty. Journal of Business Research, 54(1), 71-84.
Hess, J. D., Chu, W., & Gerstner, E. (1996). Controlling product returns in direct marketing.
Marketing Letters, 7(4), 307-317.
Huppertz, J. W. (2007). Firms' complaint handling policies and consumer complaint voicing.
Journal of consumer marketing, 24(7), 428-437.
Janakiraman, N., & Ordóñez, L. (2012). Effect of effort and deadlines on consumer product returns. Journal of Consumer Psychology, 22(2), 260-271.
Janakiraman, N., Syrdal, H. A., & Freling, R. (2015). The Effect of Return Policy Leniency on Consumer Purchase and Return Decisions: A Meta-analytic Review. Journal of Retailing. Keaveney, S. M., Huber, F., & Herrmann, A. (2007). A model of buyer regret: Selected prepurchase and postpurchase antecedents with consequences for the brand and the channel. Journal of Business Research, 60(12), 1207-1215.
Ketzenberg, M. E., & Zuidwijk, R. A. (2009). Optimal pricing, ordering, and return policies for consumer goods. Production and Operations Management, 18(3), 344-360.
Kim, J., & Wansink, B. (2012). How retailers’ recommendation and return policies alter product evaluations. Journal of Retailing, 88(4), 528-541.
Kumar, V., Dalla Pozza, I., & Ganesh, J. (2013). Revisiting the satisfaction–loyalty
relationship: Empirical generalizations and directions for future research. Journal of Retailing,
89(3), 246-262.
Lantz, B. and Hjort, K. (2013), “Real E-customer Behavioural Responses to Free Delivery and Free Returns,” Electronic Commerce Research, 13 (2), 183–98.
42 Lee, D. H. (2015). An alternative explanation of consumer product returns from the
postpurchase dissonance and ecological marketing perspectives. Psychology & Marketing,
32(1), 49-64.
Leeflang, P., Wieringa, J. E., Bijmolt, T. H., & Pauwels, K. H. (2014). Modeling Markets: Analyzing Marketing Phenomena and Improving Marketing Decision Making. Springer.
Lehmann, D. R. (1991). Modeling choice among assortments. Journal of retailing, 67(3), 274-299.
Mostard, J., R. de Koster, R. Teunter. (2005). The distribution-free newsboy problem with resalable returns. International journal of Production Economics. 97(3) 329–342.
Mukhopadhyay, S.K., Setaputra, R., (2007). A dynamic model for optimal design quality and return policies. European journal of operation research. 180 (3), 1144–1154.
Nasiry, J., & Popescu, I. (2012). Advance selling when consumers regret. Management Science, 58(6), 1160-1177.
Ofek, Elie, Zsolt Katona, and Miklos Sarvary (2011), ""Bricks and Clicks": The Impact of Product Returns on the Strategies of Multichannel Retailers," Marketing Science, 30 (1), 42-60
Owens, D.L. and Hausknecht, D.R. (1999), “The effect of simplifying the complaint process: a field experiment with the Better Business Bureau”, Journal of Consumer Satisfaction, Dissatisfaction, and Complaining Behavior, Vol. 12, pp. 35-43.
Hye Park, J., & Stoel, L. (2002). Apparel shopping on the Internet: Information availability on US apparel merchant Web sites. Journal of Fashion Marketing and Management: An
International Journal, 6(2), 158-176.
Pasternack, B. A. (2008). Optimal pricing and return policies for perishable commodities.
43 Payne, J. W., Bettman, J. R., & Johnson, E. J. (1993). The adaptive decision maker.
Cambridge University Press.
Pei, Z., Paswan, A., & Yan, R. (2014). E-tailer׳ s return policy, consumer׳ s perception of return policy fairness and purchase intention. Journal of Retailing and Consumer Services,
21(3), 249-257.
Petersen, J.A., and Kumar, V. (2010) "Can Product Returns Make You Money?" MIT Sloan Management review 51.3 :85-95.
Petersen, J. A and Anderson, E.T. (2013), “Leveraging Product Returns to Maximize Customer Equity,” in Handbook on Customer Equity, V. Kumar and Denish Shah, eds. UK: Edward Elgar Publishing.
Posselt, T, Gerstner, E, and Radic, D. (2008) "Rating E-Tailers' Money-Back Guarantees."
Journal of Service Research 10.3: 207-219. Print.
Powers, T L., and Eric P. J. (2013) "The Influence of Cognitive Dissonance on Retail Product Returns." Psychology & Marketing 30.8 (2013): 724-735.
Rogers, D.S., Tibben-Lemke, R.S., (1999). Going Backwards: Reverse Logistics Trends and Practices. Reverse Logistics Executive Council Press, Pittsburgh, PA
Sarvary, M. and Padmanabhan, V. (2001), “The informational role of manufacturer returns policies: how they can help in learning the demand”, Marketing Letters , Vol. 124 No. 4, pp. 341-50.
Shiv, B, Carmon, Z and Ariely, D. (2005): "Placebo Effects of Marketing Actions: Consumers May Get What They Pay For." Journal of Marketing Research 42.4 383-393.
Suwelack, T and Krafft, M (2012), “Effects of Money-Back and Low-Price Guarantees on Consumer Behavior,”. Quantitative Marketing and Marketing Management. 531–61.
