A section of this article (“Tactical issues in CLSCs”) is adapted from a portion of an article published by INFORMS. Reprinted by permission, Souza, G., Closed-Loop Supply chains with Remanufacturing, Tutorials in Operations Research, 2008. Copyright 2008, the Institute for Operations Research and the Management Sciences, 7240 Parkway Drive, Suite 300, Hanover, MD 21076 USA. The author would like to thank the Associate and Senior Editors for valuable suggestions on a previous draft of this article.
In this article, I present a review and tutorial of the literature on closed-loop supply chains, which are supply chains where, in addition to typical forward flows, there are reverse flows of used products (postconsumer use) back to manufacturers. Examples include supply chains with consumer returns, leasing options, and end-of-use returns with remanufacturing. I classify the literature in terms of strategic, tactical, and operational issues, but I focus on strategic issues (such as when should an original equipment manufacturer (OEM) remanufacture, response to take-back legislation, and network design, among others) and tactical issues (used product acquisition and disposition decisions). The article is written in the form of a tutorial, where for each topic I present a base model with underlying assumptions and results, comment on extensions, and conclude with my view on needed research areas.
In forward supply chains, the flow of material is unidirectional, from suppliers to manufacturers to distributors to retailers, and to consumers. In closed-loop supply chains (CLSCs) there are reverse flows of used products (postconsumer) back to manufacturers. As an example of CLSC, consider Cummins, the original equipment manufacturer (OEM) of diesel engines based in Columbus, Indiana, shown in Figure 1. Forward flows consist of new engines and/or engine parts (such as a water pump or a turbocharger), and reverse flows consist of used products, and remanufactured products. Remanufacturing is the process of restoring a used product, postconsumer use, to a common operating and aesthetic standard, which may involve upgrades to the original product's functionality. For a diesel engine or part, remanufacturing consists of six different steps: full disassembly, thorough cleaning of each part, making a disposition decision for each part (remanufacture it or recycle it for materials recovery), refurbishing parts to restore their functionality to that of a new part, reassembly, and testing.
Remanufactured engines or parts sell at a 35% discount relative to the corresponding new engine or module. Upon purchasing a Cummins product, customers receive a discount if they return their old product. Used products (also known as cores, or returns) are shipped from dealers to Cummins’ depot for used products. At the depot, customers are given credit for returning the used product, and products are then shipped to one of two plants: engine remanufacturing (plant A), or module remanufacturing (plant B). Remanufactured engines are shipped from plant A to the main distribution center, for distribution to the dealers. Remanufactured parts are shipped from plant B to either the distribution center, or to the engine remanufacturing plant A, depending on forecasts and current needs. Used parts not suited for remanufacturing are sold to recyclers.
Figure 1 illustrates a CLSC where the main source of cores are end-of-use returns, when the product has completed a full usage cycle but still has significant value left. This is in contrast with end-of-life returns, which have reached the end of their useful life, mostly due to obsolescence or substantial damage, and can only be recycled for materials recovery. Finally, there are consumer returns, which are products that have experienced little or no use by consumers—they are a result of liberal return policies by retailers and are mostly not defective (Ferguson, Guide, & Souza, 2006).
The focus of the article is thus on the management of CLSCs, which is only a subset of the much broader field of sustainable operations. I do not provide an exhaustive literature review of the CLSC literature—this would be too ambitious for a single paper, considering the scope—and when appropriate, I point the reader to exhaustive reviews in specific topics. Rather, I present a reasonably comprehensive review written in the form of a tutorial, where for a specific topic I present a base model formulation, assumptions, and key results, and then comment on extensions. The choice of papers included here is based on my view of the more influential research. I include only published papers, except in select topics, where I happen to know of significant unpublished research. Other general overviews of CLSC research are given by Guide and Van Wassenhove (2009), and Atasu, Guide, and Van Wassenhove (2008), albeit not in a tutorial format, and in Ferguson and Souza (2010), from a managerial perspective.
Table 1 provides an overview of decisions in CLSCs. Strategic decisions, such as network design, have a long-term impact (years) on the firm's operations. Tactical decisions, such as inventory policies, impact the firm's operations for weeks. Operational decisions, such as scheduling, are made daily, and have only short term impact. I focus on strategic and tactical decisions, as these seem to be the target of most of the recent research in the area. Strategic decisions also drive most of the profits and environmental impacts. The reader interested in operational issues is referred to Lambert and Gupta (2005), Pochampally, Nukala, and Gupta (2008), and Souza (2008) for reviews.
Table 1. Examples of strategic, tactical, and operational issues in CLSCs
Decisions and Issues
• Network design: location and size of collection centers, remanufacturing facilities, etc.
• Collection strategy: should customers return products to retailers or directly to OEMs?
• Should the OEM remanufacture?
• Leasing or selling?
• Trade-in and buy backs programs
• Supply chain coordination: contracts and incentives
• Response to take-back legislation
• Impact of recovery activities on new product design
• Acquisition of product returns—how many, when, and of which quality ?
• Returns disposition: remanufacturing, dismantling for spare parts, or recycling?
• Disassembly planning: sequence and depth of disassembly
• Scheduling, priority rules, lot sizing, and routing in the remanufacturing shop
STRATEGIC ISSUES IN CLSCs
Should an OEM Remanufacture?
Should an OEM offer a remanufactured version of its product? The basic modeling framework to answer this question uses vertical differentiation models. I describe a basic model formulation with the underlying assumptions; I then discuss its variations, extensions, and challenges. Consider a monopolist OEM who manufactures a (new) product and considers the introduction of a remanufactured version of the same product. Unit variable cost for the new product is c, and unit variable cost for the remanufactured product is . The addition of a remanufactured product has two implications: (i) a market expansion effect, because the remanufactured product (priced lower) reaches a segment of consumers who are not willing to pay for the new product; and (ii) a cannibalization effect, as some consumers who would have previously purchased the new product switch to the remanufactured product. Thus, pricing of the two products is critical. The price of the new product is and the price of the remanufactured product is , which are decision variables. One can equivalently work with quantities as decision variables, as shown below. This is a single period problem, which can be thought of as a period in an infinite horizon.
Assumption 1. Consumer willingness-to-pay (wtp) for the new product is uniformly distributed in the interval [0, 1]. The cumulative distribution function (cdf) of wtp is thus .
Each consumer buys at most one unit of the product. In this model, a consumer of type has a wtp of ϕ for a new product and her net utility from purchasing is . If the OEM does not offer remanufactured products, and if the market size is normalized to , then the quantity sold is , which is the demand curve if there are no remanufactured products. Thus, the assumption of uniform wtp implies linear demand curves. For many products an inverted S-shaped curve for demand is more realistic: as price increases from zero, demand decreases slowly up to a certain price level; demand then sharply decreases with subsequent price increases; at high price levels, subsequent price increases lead to a slow reduction in demand. A logit demand curve is an example of this type of behavior, and it can be obtained if one assumes a Gumbel wtp distribution (instead of uniform), that is . Linear demand curves can be seen as an approximation of the non-linear demand curve if prices are allowed to vary within a relatively small interval (the middle portion of the inverted S-shaped curve).
Assumption 2. A consumer of type ϕ (i.e., wtp for new product equal to ϕ) has a wtp for the remanufactured product equal to , where .
Remanufactured and new products are vertically differentiated—consumers prefer a new unit to a remanufactured one for the same price. If , consumers do not consider the remanufactured product as an alternative to the new product. If , consumers view the new and remanufactured units as perfect substitutes. Most products are in between these two extremes, so that . There is considerable evidence that consumers perceive remanufactured products to be of lower quality than new products. Hauser and Lund (2003) report 35–55% average discounts for a remanufactured product relative to new. Guide and Li (2010) empirically derive δ for power tools and Internet routers using online auctions, and they find . Subramanian and Subramanyam (2012) compare online prices of new and remanufactured products, and find that (a proxy for δ) ranges from 0.60 (for video game consoles) to 0.85 (for some consumer electronics).
A consumer of type ϕ thus has a net utility for the new product and for the remanufactured product. If , the consumer buys the new product; solving it for ϕ yields . If and , the consumer buys a remanufactured product; solving it for ϕ yields . These expressions for and are the demand curves; inverse demand curves are easily obtained as and . The monopolist's problem is thus:
It can be shown that the solution to Equation (1) falls into three separate regions. If , then , and thus the OEM does not offer the remanufactured product. If and , then the OEM offers both remanufactured and new products. If , then (but ); clearly this case is not realistic because in steady state the firm needs to sell new products to obtain used products for remanufacturing as one cannot remanufacture a product an infinite number of times (Geyer, Van Wassenhove, & Atasu, 2007). If the OEM only sells the new product, and a third-party remanufacturer offers the remanufactured product, then we have a game where the OEM maximizes the first term and the entrant maximizes the second term in Equation (1).
Vorasayan and Ryan (2006) use the basic demand framework above for a monopolist selling new and remanufactured products, however, they explicitly consider uncertainty in demand when setting the optimal prices in a capacitated setting through the use of a queuing network model. In Debo, Toktay, and Van Wassenhove (2005) the firm also optimizes the level of remanufacturability of a new product (which impacts the remanufacturing cost ), and they use an infinite-horizon model where sales of new products in a period constrain the amount of remanufacturing in future periods. Ferrer and Swaminathan (2006) also consider a multiperiod problem; they assume that the OEM's new and remanufactured products are perfect substitutes (), however, a third-party offers a remanufactured product with an inferior quality. This same competitive scenario is analyzed by Majumder and Groenevelt (2001) who do not model consumer behavior as above but equivalently assume linear demand curves. Debo, Toktay, and Van Wassenhove (2006) introduce a life cycle perspective into the problem, so that sales of the new product follow a Bass-type diffusion model, with remanufacturing being constrained by the amount of new products sold. They find that remanufacturing is more attractive for new products with a slower diffusion rate, due to more overlap between the new and remanufactured product life cycles, and thus more opportunity to sell remanufactured products.
Ferguson and Toktay (2006) consider the case where a monopolist OEM offers remanufactured and new products, as in Equation (1), in addition to the case where the remanufactured product is offered by a third-party remanufacturer. They use a two-period setting where remanufacturing in the second period is constrained by the number of cores available, which is the quantity of new products sold in the first period (i.e., ). They assume that the average cost of remanufacturing increases in the quantity of products remanufactured, that is, —this assumption is based on convex collection costs; moreover, higher remanufacturing quantities imply the need to use returns with lower quality. This assumption becomes standard later in the literature. They show how remanufacturing can be used as a strategic weapon by the OEM to deter the entry of third-party remanufacturers. In Atasu, Sarvary, and Van Wassenhove (2008), there is an additional green segment, which constitutes a fraction λ of the overall market, and that values remanufactured and new products equally, so that the monopolist OEM remanufactures if as long as . They also consider the case where the OEM offers both new and remanufactured products, but competes with a low cost producer of new products (subscript “c”). Consumers discount the value of the competitive product by α, and they analyze the case where , which is the worst case for remanufacturing. They find that if and (that is, the competitor has no relative cost advantage), then the OEM remanufactures as long as is sufficiently low. Focusing on consumer returns, Pince, Ferguson, and Toktay (2012) show that the monopolist OEM always remanufactures if, in addition to selling the product as a remanufactured product as in (1), remanufactured products can also be used to meet demand for warranty replacements, which is common practice. In their model, consumer returns are , , and demand for warranty replacements is , where ζ is a random variable.
Strategies for product take-back: leasing, trade-ins
The models discussed in the previous section do not explicitly model the manner by which the firm obtains its product returns for remanufacturing. They simply consider that remanufacturing is constrained by the amount of new products previously sold by an OEM. This section explores different mechanisms for obtaining product returns (or preventing third party remanufacturers from obtaining returns), such as trade-in programs, leasing, and relicensing fees. There is a stream of literature that determines the firm's acquisition strategy—how many product returns to acquire to support remanufacturing, given acquisition costs, demand for remanufactured products, and other costs. This stream of literature is reviewed later, under tactical decisions.
The CLSC literature published before 2000 typically made the assumption that product returns were exogenous and uncontrollable. The focus was on optimizing operations (e.g., inventory) given a probability distribution of product returns in a period. Guide and Van Wassenhove (2001) was one of the first papers to conceptually introduce the idea of product acquisition management where an OEM can control the timing, quantity, and quality of product returns through appropriate economic incentives, in order to increase the profitability of recovery activities.
Leasing is an attractive mechanism for obtaining used products for remanufacturing, as well as controlling the secondary market. The durable goods literature in economics addresses the profitability of leasing—for a review, see Waldman (2003)—but without considering the operational decisions for products off-lease. In the CLSC literature, I am aware of only two papers that study leasing as a mechanism for take-back. Agrawal, Ferguson, Toktay, and Thomas (2012) compare leasing relative to selling as strategies for the OEM, and they use Assumptions 1 and 2 to model consumer preferences between new and used products. With leasing, used products are remanufactured and resold by the OEM, whereas in the case of selling, consumers trade used products in the secondary market. They find that the profitability and environmental impact of leasing relative to selling depends greatly on the product durability δ and the product's environmental impact during the use stage of the life cycle. Robotis, Bhattacharya, and Van Wassenhove (2012) use optimal control in a Bass-type diffusion model to determine the optimal leasing price and duration when both production and servicing capacities are limited, in the presence of remanufacturing, where remanufactured and new products are perfect substitutes, which is the case of Xerox copiers. They find that if savings from remanufacturing are high, then the firm should offer longer leasing contracts.
Trade-in programs as a source of returns for remanufacturing have been explored by two papers: Ray, Boyaci, and Aras (2005), and Li, Fong, and Xu (2011). Ray et al. (2005) assume that traded-in products can be remanufactured (so they have a value), and that consumers also have a feel for the value of their used product. They derive optimal trade-in discounts under three different policies: a discount dependent on the used product's age, a discount independent of the product age, and no trade-in discount. Li et al. (2011) provide a methodology for forecasting trade-ins based on customer segmentation and signals (return merchandise authorizations, or RMAs).
Oraiopoulos, Ferguson, and Toktay (2012) explore the case where an OEM can directly affect the resale value of her product in the secondary market through a mandatory software relicensing fee charged to the buyer of the remanufactured product, where remanufacturing is conducted by a competing third-party firm. They find that it is suboptimal for the OEM to charge high relicensing fees, even though lower relicensing fees increase competition for the OEM's products by making the remanufactured product more attractive. This occurs because consumers are strategic and consider the lower resale value for a product with high relicensing fees when purchasing a new product.
Challenges to Assumptions 1 and 2
Recently, a stream of research has empirically addressed the market for remanufactured products, and some of the results run counter to Assumption 2, of linear demand curves for remanufactured products.
Ovchinnikov (2011) conducts a lab experiment to study the impact of prices of new and equivalent remanufactured laptops on the sales of remanufactured laptops, that is, the cannibalization effect. The cannibalization effect is measured by the fraction of customers who switch from buying new to buying an equivalent remanufactured product for a given percentage discount in the price of the remanufactured product relative to the new product's price. Assumption 2 indicates that this fraction should be monotonically increasing in the remanufactured product's price discount, but Ovchinnikov finds an inverse U-shaped curve, indicating that consumers may infer a low quality of the remanufactured products if it is priced too low. Thus, higher price discounts may decrease cannibalization, particularly for high-end segments.
Using a lab experiment, Agrawal, Atasu, and van Ittersum (2012) conclude that the presence of iPods remanufactured by the OEM (Apple) decreases the wtp for new iPods, because the remanufactured product induces consumers to infer a lower quality for the new product. This negative impact is weakened if the product is remanufactured by a third party; it is thus possible that third-party remanufacturers can actually benefit the OEM. This finding is in contrast with the research that originated from Assumptions 1 and 2, where the remanufactured product has no impact on the wtp for the new product, which concludes that third-party competition can only be detrimental to the OEM.
This recent stream of empirical research in the market for remanufactured products is a welcome addition in a field that has been heavily prescriptive and normative. One must be careful, however, about not reading too much into the results. Behavioral studies, in particular, do not simulate exact market conditions—there is a gap between purchasing intent and actual purchasing behavior. This is in contrast with studies that use actual sales data (e.g., Guide & Li, 2010; Subramanian & Subramanyam, 2012), which I believe to be ideal. In addition, studies are focused on a particular product category (such as iPods), and that does not mean the results can be extended to other products, such as medical equipment and automotive parts.
Product Take-Back Legislation
Take-back legislation is based on the concept of extended producer responsibility (EPR), which holds manufacturers physically and financially responsible for taking back used products at the end of their useful life, and disposing of them in an environmentally friendly manner. Take-back legislation exists for automobiles, appliances, packaging, and electronics, in different parts of the world. Because of the exponential increase in electronic waste (e-waste) generation, however, the bulk of the research has focused on take-back legislation for e-waste. The Waste Electrical and Electronic Equipment (WEEE) directive in the European Union is an early and important example of take-back legislation, as it regulates 11 different product categories as diverse as lamps, medical equipment, cell phones, and large household appliances. There are over 25 states in the U.S. with some form of take-back legislation for e-waste, and some are narrowly targeted to computers.
An advance recycling fee (ARF) is a fee collected from consumers or producers at the time of sale. The fees are used to subsidize recycling of obsolete products collected by municipalities. An example of such legislation is the state of California. In contrast, the WEEE directive, along with the majority of the other U.S. states’ legislation, stipulates collection targets (a minimum amount to be collected as a percentage of sales) and recycling targets (a minimum percentage of products to be recycled for materials or energy recovery, from the amount collected). In the WEEE directive, the overall collection rate for a given year is 65%, in weight, of the average amount sold in the previous two years, starting in 2016. The recycling rate for IT and telecommunications is 70%, and the total recovery rate (recycling plus energy) is 80%. For details on take-back legislation around the world, as well as a comprehensive literature review, the reader is referred to Atasu and Van Wassenhove (2012), and Atasu and Van Wassenhove (2010), respectively. I illustrate this stream of research with a base model, and commment on extensions and variations.
Atasu, Van Wassenhove, and Sarvary (2009) consider the design of take-back legislation that maximizes society welfare W. The policy maker's decision variables are the collection rate τ (defined as the fraction of products sold that are collected), the recycling rate ω (defined as the fraction of products collected that are recycled), and a subsidy σ given to each manufacturer per unit that is collected and recycled. There are n identical manufacturers competing Cournot-style on a market of size normalized to one. Each manufacturer's expected profit is , where is the equilibrium price, c is the unit production cost, is the take back cost per unit (convex increasing in ω and τ), and σ is the per unit subsidy. The equilibrium price is determined through , which yields the profit maximizing output for each manufacturer , and so is the total output. The total profit across all manufacturers is . Consumer surplus is . The environmental impact of such legislation is proportional to the number of products not recycled, or , where ε is the environmental impact per unit. Thus, the policy maker solves the following problem:
where is the advertising cost incurred by the policy maker in order to achieve a collection rate τ (collection is typically done by municipalities). They find that the policy maker's optimal collection rate is decreasing in collection and recycling costs φ, and increasing in the degree of competition n and the cost to the environment ε. This shows that the design of take-back legislation should take into consideration a product's environmental impact, as well as industry structure, in order to achieve better societal outcomes.
EPR laws typically require the brand manufacturer (OEM) to have full responsibility for reverse logistics and recycling costs, so other members of the supply chain (e.g., the OEM's suppliers) are not typically directly impacted. Jacobs and Subramanian (2012) study the impact of τ and ω on supply chain profits and environmental impacts when responsibility for product take-back is shared across supply chain members. They use a two-echelon model, with a single manufacturer and a single supplier, and a linear demand curve . They model the flows of virgin and recycled materials in the supply chain, for given τ and ω set by legislation, however the manufacturer or supplier may elect to recycle at a higher rate than the minimum mandated rate ω, if that is more economical than using virgin materials. Collection cost is convex increasing in the collected quantity, but recycling cost is concave in the recycled quantity, reflecting economies of scale; there is also a disposal cost for units collected but not recycled that is linear in the disposed quantity. They show that implementing regulation that requires some level of sharing for product take-back responsibility, such as a shared collection rate between supplier and manufacturer, may increase profits for all supply chain parties, resulting in increased society welfare W. That, however, may come at the expense of increased virgin material consumption.
Another stream of research has modeled a more granular reverse supply chain for recycling e-waste. Nagurney and Toyasaki (2005) propose a network-based modeling framework and algorithm for the numerical computation of equilibrium prices and product flows in a reverse supply chain with four tiers: e-waste sources, collectors, processors of e-waste (e.g., smelters), and demand nodes for the recycled materials. Hammond and Beullens (2007) extend the model by Nagurney and Toyasaki (2005) by incorporating the forward supply chain (for newly manufactured products), and hence their model includes the entire CLSC. Their model is in the context of the WEEE directive for e-waste. Limited numerical examples indicate that legislation that sets minimum recovery levels for product take-back (like the WEEE) can increase recycling and recovery activities, and thus can create economic growth. Toyasaki, Boyaci, and Verter (2011) present a model of a CLSC with two manufacturers, two recyclers, and a market for recycled material, in the context of take-back legislation. The manufacturers compete in the consumer market, and they assume linear demand curves with substitution effects so that demand for the product sold by manufacturer j is for . Recyclers compete with one another, and charge profit-maximizing recycling fees. There are economies of scale in recycling, so the recycling cost for recycler i is . They analyze two schemes: in a monopolistic scheme, a third-party NGO receives the used products collected by the manufacturers, contracts with the two recyclers, and allocates e-waste among them; it then charges recycling costs back to manufacturers according to their respective market share. This is similar to the e-waste take-back system used in Belgium or Sweden. In a competitive scheme, there is no third-party NGO, and manufacturers contract with recyclers directly for their own collected e-waste mandated by the legislation. They analytically show that the average recycling fee charged in the monopolistic scheme is always higher than in the competitive scheme, and if economies of scale in recycling are not too strong (θ not too high), then recyclers are also better off in a competitive scheme.
Another stream of literature analyzes the key question as to whether take-back legislation provides incentives for manufacturers to design greener products. Take-back legislation can be in the form of collective producer responsibility (CPR) or individual producer responsibility (IPR). In CPR there is a collective collection target (across all manufacturers), whereas in IPR each manufacturer is only responsible for its own waste. Although the WEEE directive was originally designed as IPR, it was implemented as CPR in practice, due to logistical difficulties associated with sorting e-waste at collection points. Plambeck and Wang (2009) show that IPR legislation indeed results in manufacturers designing products with higher recyclability. In addition, they show that an ARF decreases the frequency of new product introductions, and consequently decreases the amount of e-waste, benefiting the environment. Using different model setups, Atasu and Subramanian (2012) and Esenduran and Kemahlioglu-Ziya (2012) also demonstrate that IPR results in greener products (more easily recyclable) than CPR because manufacturers have incentives to free-ride on other manufacturers greener products under CPR. However, CPR may provide better operational cost efficiency, due to the impact of economies of scale in recycling costs.
The literature on take-back legislation is alive and well, although most of it is based on relatively simple economic models that overlook some of the details. I agree with Atasu and Van Wassenhove (2012), who indicate the need for more research into implementation of take-back legislation (the WEEE directive was implemented differently in the member states), and producer response. In other words, details matter. In addition, I believe that any modeling effort in this area should take a life cycle assessment (LCA) perspective, and report on environmental impacts of producers’ choices, not only in terms of virgin material consumption (a standard metric reported in the literature), but manufacturing, distribution, use with consumers, and end-of-life. Examples of papers using LCA include Atasu and Souza (2012), and Raz, Druehl, and Blass (2012); both papers focus on new product design, as we discuss next.
Impact of Product Recovery on New Product Design
There is a small but growing stream of literature that studies the impact of product recovery (remanufacturing and/or recycling) on the design of new products. This stream of literature uses predominantly stylized analytical models, and to keep tractability, several papers capture product design in a uni-dimensional variable s called product quality. Quality can be interpreted as a key performance measure for which consumers are willing to pay. As an example, for a printer it could be printing speed; for a digital camera it could be the quality of the picture as measured in megapixels. Assumption 1 is slightly modified such that a consumer of type ϕ has a net utility for a product of quality s with price level p equal to (Mussa & Rosen, 1978; Moorthy, 1984). Another common assumption is that the product's variable production cost is quadratic in quality . Thus, a monopolist not engaged in product recovery solves the problem , where is the quantity sold for a market with size normalized to one (product is sold to all consumers with positive net utility, ). This yields , and .
Atasu and Souza (2012) introduce product recovery, so that a fraction τ of total demand q is taken back at a collection cost , either as a result of take-back legislation or as a profitable activity, and used in the production of the new product as recycled material, or remanufactured components. The production cost for q units is . Here r is the per unit cost saving from recovery and reuse of components that are quality inducing, and ν is the unit processing cost (for used products) plus the cost of recycling components that are not quality inducing; if there is profitable recovery then . They show that, in general, the monopolist designs a product with higher quality than in the case of no recovery: . These results also hold when recovery is in the form of remanufacturing, and remanufactured and new products are imperfect substitutes (Assumption 2). Using a similar modeling framework, but a two-period model, Orsdemir, Kemahlioglu-Ziya, and Parlakturk (2011) study a situation where an OEM has a new product (offered in both first and second periods) that competes with a remanufactured product offered by a third-party remanufacturer (3PR), which is offered only in the second period. The remanufactured product's quality level (and hence its cost) is dependent on the quality level of the OEM's new product. In addition, the OEM—through its pricing and quality choices in the first period, which impact sales—can limit the number of cores that the 3PR can access for remanufacturing in the second period. They show that the OEM can deter the 3PR by serving high valuation segments with a higher quality product (which implies a higher price), because this strategy limits the amount of cores available for remanufacturing by the 3PR in the second period.
In Subramanian, Ferguson, and Toktay (2012), an OEM has an existing product line with two products: a high-end product with quality , and a low-end product with quality . Qualities are given; the decision variable here is binary: whether to implement component commonality in the product line or not. Commonality impacts the valuation of new products due to consumers’ perception of similarity between high and low end products: a consumer's perceived valuation for the high and low quality products with commonality are and respectively. Following Assumption 2, a consumer's perception of quality for a remanufactured product is without commonality, but with commonality. Commonality also impacts the marginal production costs for the high and low quality products as and , respectively, where m indicates the cost savings attributed to commonality, and is the overdesign penalty, as common components are over designed for the low-end product. Finally, commonality lowers unit remanufacturing cost to , . The authors compare the firm's profit with and without commonality for two cases: one where the firm remanufactures, and one where remanufacturing is conducted by a third-party. Among other findings, they conclude that a firm may not decide to introduce commonality in its product line if α is low and remanufacturing is done by a third party, because the third party firm reaps the commonality savings.
Yenipazarli and Vakharia (2012) consider a monopolist seeking to tap into a growing segment of green consumers. The firm has an existing product with some green features, but these are not advertised to consumers. Their model follows the quality choice setup of Moorthy (1984), but where s is now interpreted as the product's green quality. To tap the green segment, the firm has a choice of two strategies: (i) under the accentuate strategy, the firm advertises the existing product's green quality s1 at an advertising cost , or (ii) under the architect strategy, the firm redesigns the product to a green quality level s2 at a cost . The potential market for a product of green quality is , where is a strategy-specific parameter. They find that the choice between the accentuate and architect strategies depends critically on the ratio between marginal cost and demand elasticity parameters β1 and β2.
In Galbreth, Boyaci, and Verter (2012), a monopolist sells a product that undergoes steady innovation over time, such that in each period a fraction of the product must be redesigned (β is given). The product is sold in three versions: new, upgraded (used products remanufactured to the same functionality as new products, incorporating the redesign), and remanufactured (used products remanufactured to their original functionality). Building on Assumption 2, a consumer of type ϕ has net utility for a new product, for an upgraded product, and for the remanufactured product. Unit production costs are , , and for the new, upgraded, and remanufactured products, respectively. Here, c0 is the unit manufacturing cost if , c1 is the unit manufacturing cost if , K is an extra variable cost of making a product with entirely reusable components (), and d is unit cost of disposal. Similarly to Equation (1), the firm decides on the profit-maximizing quantities for each of the products in a single period. A key finding is that the optimal amount of product reuse (remanufacturing and/or upgrading) decreases as the rate of innovation β increases.
In Raz et al. (2012), the firm can invest effort in designing a product to improve environmental performance (decrease energy consumption) in both the manufacturing (e1) and use (e2) stages of the life cycle. There is a quadratic cost of effort. Improving environmental performance in manufacturing also decreases unit manufacturing cost (), whereas improving environmental impact in use increases product quality, which increases mean demand in the period (, where ε is a random variable). Using a newsvendor framework, the firm decides upon the production quantity q and efforts e1 and e2, where leftover units can be recycled with a positive salvage value. Due to the cost reduction impact of the effort e1, the firm produces more than if it does not invest in design for environment (), and as a result, there can be a higher environmental impact overall, despite the lower environmental impact per unit in manufacturing and use. Raz et al. (2012) does not focus, however, on designing for remanufacturability or recyclability. Other papers in this stream of research discussed in previous sections include Debo et al. (2005), who introduces a remanufacturability design variable that impacts the remufacturing cost , and Plambeck and Wang (2009) who investigate the impact of take-back legislation on the frequency of new product introductions.
I believe this stream of research overall has significant potential, due to its importance; based on my own experience with firms such as Volkswagen and Cummins, companies do design products with CLSC considerations, such as remanufacturability. Some design protocols place a high weight on recyclability, such as Cradle to Cradle (www.mbdc.com). More generally, though, firms design products with sustainability considerations, such as materials choice (toxicity, availability, recyclability, etc.), energy consumption, ease of disassembly, packaging, and so forth. Yet, research here—from a business, not engineering, perspective—has only started to scratch the surface. Most of the research, as seen above, uses a single abstract parameter or variable to indicate a product's recyclability or quality, and it is for the most part an extension of traditional marketing models of quality choice. Although it is easier said than done, I believe that higher impact research would need to either take a more empirical approach, or use more flexible and comprehensive choice models, such as discrete choice, or both. This is because designing a product involves trade-offs along multiple design dimensions, but that is not captured in existing research.
CLSC Network Design
In forward supply chains, facility location models provide the optimal configuration for the chain. A typical location model determines the optimal number, location, and size of warehouses that minimizes distribution and fixed costs, for a given network and size of customers. In a CLSC, the firm also locates consolidation centers for returns and recovery facilities (remanufacturing, recycling, or disposal). Consolidation centers aggregate returns from various sources, test them and route them to recovery facilities.
Exhaustive reviews of CLSC network design models are provided by Aras, Boyaci, and Verter (2010), and Akcali, Cetinkaya, and Uster (2009). I illustrate this stream of research with the general formulation presented by Fleischmann, Beullens, Bloemhof-Ruwaard, and Van Wassenhove (2001). It includes four separate levels: (i) plants, where new products are manufactured and/or recovery takes place; (ii) warehouses, for distribution of new and/or recovered products; (iii) consolidation centers; and (iv) customers. The flow is as follows: new and/or recovered products are shipped to customers via warehouses while returns are shipped to recovery facilities (or disposal) via consolidation centers. For example, Hewlett-Packard has a returns consolidation and testing center in Tennessee (Guide, Souza, Van Wassenhove, & Blackburn, 2006), to which all consumer returns across the U.S. are sent. After testing, returns are shipped to different remanufacturing facilities.
Plant locations, ; , where is the disposal option
Potential warehouse locations,
Fixed customer locations,
Potential consolidation center locations
Fraction of customer k's demand served from plant i through warehouse j
Fraction of customer k's returns returned to plant i through consolidation center l
Unsatisfied fraction of customer k's demand
Uncollected fraction of customer k's returns
Indicator variable for opening plant i ( and similarly defined)
Unit cost (transportation, production, handling) of serving k from i via j
Unit cost of returns (transportation, handling) from k to i via l
Unit disposal cost (including collection, transportation, handling) for k via l
Unit penalty cost for not serving customer k's demand
Unit penalty cost for not collecting customer k's returns
Fixed cost for opening plant i ( and similarly defined)
Demand for customer k
Returns from customer k
Minimal disposal fraction
The firm solves the following mixed-linear integer program (MILP) to optimally design its CLSC:
The objective (3) minimizes the CLSC's fixed and variable costs. Flow balancing constraints for demand and returns at each customer are given in Equations (4) and (5). Constraint (6) represents flow balancing constraints at each plant; the gap between outgoing and incoming products represents new production. Constraint (7) requires that disposal be a minimum percentage of all returns; this is a technical requirement because not all returns can be recovered. Constraints (8)–(10)) ensure that flows from a facility only happen if the facility is open.
This formulation is quite general and captures many different scenarios. For example, if the firm decides on two separate networks for reverse and forward chains, it solves two different problems: one where and the other where . Through the value of relative to and , plants can be dedicated solely to recovery or to new product manufacturing (e.g., Barros, Dekker, & Scholten, 1998). If there are no legislative constraints on collection quantities (Jayaraman, Guide, & Srivastava, 1999; Realff, Ammons, & Newton, 2004), then ; otherwise it can be made as large as desired (Spengler, Puechert, Penkuhn, & Rentz, 1997). Sahyouni, Savaskan, and Daskin (2007) propose an efficient Lagrangian relaxation-based algorithm to solve this problem. Future research in this area could address specific applications, for example, how take-back legislation impacts network design for a particular industry.
CLSC network design approaches other than the MILP optimizaton model have also been used. Guide et al. (2006) use queuing networks to demonstrate the value of speed in recovery—which is not captured in the MILP models above—on profitability for time-sensitive consumer returns such as consumer electronics. Their paper is based on data from two firms, Hewlett-Packard and Bosch. Wojanowski, Verter, and Boyaci (2007) use a continuous model for network design (where population is represented by a density) to study the effectiveness of a deposit-refund scheme in increasing a firm's collection rate. Increasing the collection rate is important when there are mandatory minimum collection targets as in take-back legislation. They show that a deposit-refund scheme increases the firm's collection rates for products with a high return value. Collection strategy is another example of a network design problem: should used products be collected by the OEM, the retailer, or by a third party? This coordination issue is analyzed next.
Incentives and Coordination in CLSCs
This stream of literature considers problems arising from the interaction of parties with non-aligned objectives in CLSCs. I illustrate the model that addresses “false failures” consumer returns by Ferguson et al. (2006). False failures are non-defective consumer returns, returned to the manufacturer due to liberal return policies (“no questions asked”) by powerful retailers. For example, about 80% of returned Hewlett-Packard printers in the U.S. are false failures. Ferguson et al. suggest that a considerable number of false failure returns can be avoided through increased retailer effort during the sales process, such as spending more time with the customer to suggest a product that best fits his or her needs, however, most of the benefits of reduction accrue to the manufacturer, who incurs most reverse logistics costs.
Specifically, consider a CLSC comprised of a manufacturer and a retailer under a single-period setting. If the retailer exerts a (sales) effort at a cost then the number of false failure returns is , a random variable with ; is the baseline effort level such that β is the baseline expected number of false failures in the period. Avoiding one false failure return (through some retailer effort ρ) implies a profit increase to the manufacturer , which is a combination of the return's processing costs (logistic cost, refurbishing if necessary, and remarketing cost). For the retailer, avoiding a false failure implies a profit increase of , which is a combination of the retailer's processing cost and the expected foregone margin (there is a probability that the consumer walks away after returning the product). If the retailer and manufacturer are vertically integrated (i.e., a centralized supply chain), the retailer's optimal effort level is found by solving:
The solution to this problem is . If the retailer and manufacturer are separate parties, then the retailer's optimal effort is given by solving Equation (13) under , which yields , so the retailer does not exert the system-optimal level of effort. To coordinate the supply chain, Ferguson et al. propose a target rebate contract, where the manufacturer pays the retailer a fixed amount u per return below a set target T. The retailer's profit under the contract is , where the last term is the additional payment by the manufacturer for false failures under the target T. The values for are chosen such that the retailer's optimal effort level , found through , results in the centralized solution, that is, . Ferguson et al. show that this contract is Pareto improving in a wide range of scenarios for normal and uniform distributions of .
The subject of consumer returns has been addressed by other papers in the literature (Ketzenberg & Zuidwijk, 2009; Su, 2009; Shulman, Coughlan, & Savaskan, 2009, 2011), although the focus of these papers is about the design of optimal returns policies for retailers: the combination of retail price and restocking fee that maximize retailer profitability. A restocking fee is a assessed on consumers for a product returned to the retailer; higher restocking fees reduce consumer returns but have a negative impact on sales. Restocking fees are imposed in order to decrease the operational costs associated with returns, as illustrated in the model by Ferguson et al. above.
Savaskan, Bhattacharya, and Van Wassenhove (2004) investigate the choice of a collection strategy in a CLSC where new and remanufactured products are perfect substitutes, demand is a linear function of price (as in Assumption 1), and remanufacturing saves costs for the OEM. The retailer can exert effort to collect used products and transfer them to the OEM for remanufacturing at a unit transfer price w; another alternative is for the manufacturer to collect used products herself; and finally another alternative is for the OEM to hire a third party for collection. They find that the retailer collection is preferred. Savaskan and Van Wassenhove (2006) extend these results to the case of competing retailers and find that retailer collection is still preferred for products where retailers compete on prices; manufacturer collection is preferred, however, for products where retailers have less impact on prices (e.g., toner cartridges). Karakayali, Emir-Farinas, and Akcali (2007) consider a CLSC comprised of a collector and a remanufacturer. The collector acquires used products, which have different quality grades (indexed by j), through a unit acquisition price (a decision variable) and sells them to the remanufacturer at a wholesale price . The remanufacturer remanufactures a single part and sells it in the market at a price (decision variable). Demand functions for number of used products acquired and number of remanufactured parts sold are linear functions of the respective prices and . They analyze the centralized case where the remanufacturer and collector are the same entity, and two decentralized cases, where the remanufacturer and the collector act as Stackelberg leaders in setting . They show that a two-part tariff (comprised of a unit wholesale price and a fixed payment) in the decentralized case can be used to coordinate the channel, that is, to achieve the same collection efficiency as the centralized case.
There is some opportunity for research in this stream, particularly when combined with take-back legislation. For example, how does take-back legislation shape incentives for players in a CLSC, such as municipalities, collectors, recyclers, OEMs, retailers, and suppliers? In addition, the literature on consumer returns is too prescriptive, and there is not enough empirical research demonstrating the impact of return policies on the behavior of consumers, manufacturers, and retailers.
TACTICAL ISSUES IN CLSCs
As seen in Table 1, most tactical issues in CLSCs revolve around product (returns) acquisition, and returns disposition, where disposition options include remanufacturing and salvaging (recycling or dismantling for spare parts). Remanufacturing planning has the peculiar characteristic of non-uniformity of inputs (returns), which has several implications: the firm needs to separate returns into different quality categories (grading); this leads to an uncertain mix of inputs in each period, as quantity, quality, and timing of returns are typically uncertain. More importantly, different quality grades may have different cost and processing requirements (with capacity implications), and graded returns can be salvaged or disposed of at potentially different salvage values depending on their quality. Some returns are unfit for remanufacturing, and need to be disposed of. I present here some of the more basic models in this large body of literature, and discuss some of the variations and extensions. For more comprehensive overviews, see Fleischmann, Galbreth, and Tagaras (2010), Inderfurth, Flapper, Lambert, Pappis, and Voutsinas (2004), and Guide (2000). I divide the discussion in three parts: single period acquisition and disposition models, multiperiod disposition models, and multiperiod capacitated planning models.
Single Period Acquisition and Disposition Models
This stream of research finds the optimal acquisition plan—how many returns to acquire—and/or the optimal disposition plan in a single period, when there is some source of uncertainty. Some papers assume only two categories of returns (good or bad); good units are remanufactured at the same cost whereas bad units are salvaged (Ferrer 2003; Zikopoulos & Tagaras, 2007). Other papers assume more granularity in the quality of incoming returns, which implies different remanufacturing costs, and focus on the optimal acquisition plan for product returns (Guide, Teunter, & Van Wassenhove, 2003; Bakal & Akcali, 2006; Galbreth & Blackburn, 2006, 2010). I illustrate this stream with the integrated acquisition and disposition model by Zikopoulos and Tagaras (2007), and comment on extensions.
Assume the firm has collected R returns, and the sorting procedure produces r remanufacturable returns (thus, returns are non-remanufacturable and should be disposed of at a unit cost ). Out of the r remanufacturable returns, the firm decides upon the remanufacturing quantity to meet stochastic demand D, which has a cumulative distribution function (cdf) . Unit sales revenue is , remanufacturing cost is , shortage cost is , and disposal cost for unsold refurbished items is . The optimal solution to the unconstrained version of this problem is given by the newsvendor model: . Here, the unit underage cost is (margin plus shortage cost plus opportunity cost of disposing an extra unit that could have been used to meet demand), whereas the unit overage cost is (difference in disposal costs plus remanufacturing cost). Given , the optimal solution to the constrained version of the problem is quite simple: if , then remanufacture and dispose ; if , then remanufacture all r units. This is the first part of the decision process. The firm should also decide upon the amount of returns to collect, R. They assume that returns originate from two collection sites; site i, , has stochastic yield (percentage of remanufacturable returns) ; these random variables are correlated with a joint distribution . The firm then determines the amount to collect from each site (such that , and ). They show that in some cases it is optimal to collect from a single site, and in others it is optimal to collect from both sites.
Zikopoulos and Tagaras (2008) consider a variation of this problem where there are inaccuracies in the sorting operation, so there is a probability α that a (good) remanufacturable item is thrown away, and a probability β that a non-remanufacturable unit is sorted as good, and it is further disassembled at a cost. They determine conditions under which setting up a sorting operation before disassembly is optimal. Ferrer (2003) considers an acquisition problem where demand D is deterministic and there is a single source for returns, but the firm also has the possibility of meeting demand by sourcing more expensive new parts from an outside supplier. He shows that the optimal solution is generally not mixed, that is, the firm either sources only new parts or only remanufactured parts if the supply of returns is large enough.
Another stream of research considers product acquisition for remanufacturing when returns have varying quality levels (that is, not only good and bad). Consider the basic framework of Guide et al. (2003). There are N quality categories for returns; category i has remanufacturing cost . To improve the quality of returns, the firm decides upon the acquisition price for quality category i and realizes a corresponding return rate , where is increasing in . Because the supply of returns, which is determined by , constrains the amount of products that can be remanufactured, the firm should set (optimally) the price of remanufactured products such that supply meets demand. The demand function (quantity sold) is , so the firm sets such that , or . Thus, the firm's optimization problem is
In Equation (14), the decision variables are the values of the acquisition prices for each quality category i. Note that remanufacturing yield is not considered in Equation (14). Bakal and Akcali (2006) consider a single acquisition price a and a single unit remanufacturing cost , but model the resulting quality of acquired returns such that yield rate (fraction of units fit for remanufacturing out of all units collected) is , where is a concave function (so that a higher acquisition price implies a higher yield), and γ is a random variable. Thus, the total amount of remanufacturable units is . The firm then decides on a and the remanufactured product's price , with the firm facing a linear demand curve . (Due to the random yield, the firm cannot set prices to exactly match supply and demand, as in Guide et al. (2003)). Motivated by toner cartridges, Galbreth and Blackburn (2006) consider an exogenous fixed acquisition price a, but assume that total remanufacturing cost (excluding acquisition cost) decreases in the total quantity of returns acquired R (the decision variable), such that the firm minimizes , where is the probability density function for remanufacturing cost (with cdf ), and , so that the firm meets demand D from the acquired R units. The trade-off is that the higher the total amount of returns collected R, the higher the acquisition cost, but the total remanufacturing cost is lower because there are more returns of better quality in the pool of acquired returns. Galbreth and Blackburn (2006) model the distribution of returns quality (i.e., cost) through a known probability distribution , which is assumed to be invariant with respect to the quantity of returns acquired. This assumption holds if the acquisition quantities are large enough such that the probability distribution associated with the sample closely matches the overall population of returns. Galbreth and Blackburn (2010) extend their 2006 model to the case where there is uncertainty in the condition of used products, that is, in the actual distribution of cost within the sample of products acquired. Their 2010 model is thus appropriate for cases where the acquisition quantities are smaller, or where the firm has more limited information into the distribution of quality of used products in the field.
Multiperiod/Continuous-Time Product Acquisition and Disposition Models
In multiperiod or continuous time acquisition and disposition models, there is more than one opportunity to acquire returns and/or make a disposition decision. Two broad streams exist (Fleischmann, Bloemhof-Ruwaard, Dekker, van der Laan, van Nunen, & Van Wassenhove, 1997): (i) repair systems, where there is a significant correlation between demand and return streams, in that the total number of parts in the system is constant (every return—defective part—triggers a demand for repaired part), and (ii) recovery (remanufacturing) systems, where, due to the long time lags between buying a new product and its end of use, return and demand streams can be assumed to be independent. Again, the focus here is on recovery systems; see Cho and Parlar (1991) for a tutorial and review on repair systems.
I present here the periodic-review model by Inderfurth, de Kok, and Flapper (2001) to illustrate some of the dynamics. There are n disposition options for stochastic returns at period t. Examples of disposition options include disassembling for spare parts, light remanufacturing, and comprehensive remanufacturing. Disposition option j faces stochastic demand at time t, has processing lead time , unit processing cost , and unit shortage cost . Unit holding cost for returns is h, and holding cost for option-j processed units is . The firm then decides on the disposal quantity and disposition quantities at time t. Denoting by α the one-period discount factor, the firm solves the following stochastic optimization problem:
In Equation (15), the expectation is taken with respect to ; and are the ending inventory of returns and processed units for option j at time t, respectively; they satisfy the balancing equations and . The optimal solution structure for Equation (15) is fairly complex, but Inderfurth et al. (2001) show how a heuristic policy with a simpler structure—a base stock policy for returns coupled with base stock policies for each of the disposition options—performs well under a linear allocation rule for returns to the different reuse options. (A base stock policy for an option j maintains the inventory position of processed units constant at , where the inventory position is defined as the inventory on-hand minus backorders plus in-processing, or in transit, inventory.)
Ferguson, Fleischmann, and Souza (2011) consider two disposition options: remanufacturing or dismantling for spare parts. Remanufacturing is more profitable, on a per unit basis, but has more uncertain demand than spare parts’. Using a formulation similar to Equation (15), but under profit maximization, they find the structure of the optimal policy for the special case where dismantling yields only a single part. When dismantling yields multiple parts, each with its own independent stochastic demand distribution, they are only able to provide the structure of the optimal policy for a single period model; the problem is much harder to solve in a multiperiod setting. Solving the disposition problem with multipart dismantling in a multiperiod setting would be a significant contribution to the literature.
Many models in the literature consider a hybrid system, where demand can be met by remanufacturing a returned item or by manufacturing (procuring) a new item from a supplier (Figure 2); these two options have different lead times and costs. Hybrid systems are particularly appropriate for managing inventory of spare parts, or in products such as Xerox copiers, where some parts can be obtained from remanufacturing used products. There are two broad types of models: continuous review and periodic review systems. van der Laan et al. (1999) derive optimal inventory policies for push systems (where remanufacturing is initiated as early as possible—as soon as there are enough returns for a batch), and for pull systems (where remanufacturing is initiated as late as possible, depending on forecasts, current inventory positions, etc.) in a continuous review setting with fixed remanufacturing costs. Toktay, Wein, and Zenios (2000) study a hybrid system with continuous review and no fixed ordering cost, and propose a base-stock policy for managing the manufacturing of new components; their biggest contribution is to provide a methodology for estimating the demand parameters, which depend on the products’ sojourn time with the customer and the probability that a customer returns a used product. Aras, Boyaci, and Verter (2004) also consider a continuous-time setting with no fixed remanufacturing costs, but they consider two types of stochastic (Poisson) returns: returns of higher quality have a lower remanufacturing lead time than returns of lower quality.
When parts are managed using a periodic review system, Inderfurth (1997) shows that the structure of the optimal policy for a hybrid system is not a base-stock policy if the two sourcing options—remanufacturing and manufacturing—have different and positive lead times. Thus, one needs heuristic approaches; some are provided by Kiesmueller and Minner (2003), Teunter, van der Laan, and Vlachos (2004), and Inderfurth and Kleber (2012). Kiesmueller and Minner (2003) propose a heuristic policy structure with two base stock levels for each period (one for manufacturing, and one for remanufacturing) and propose some basic newsvendor-based heuristics for computing the base stock levels. Zhou, Tao, and Chao (2011) consider that returns are of K different qualities, and show that if remanufacturing and manufacturing have identical lead times the optimal policy is comprised of different base stock levels, one for each quality grade (remanufacturing), and one for manufacturing. When lead times are different, they propose a heuristic based on base stock levels. Inderfurth and Kleber (2012) add a third option where in addition to the two options displayed in Figure 2, one can place a large final order (at a lower cost per unit) at the beginning of the planning horizon. They also propose newsvendor-based heuristics for computing the two base stock levels and show that they perform very well. Tao, Zhou, and Tang (2012) introduce random remanufacturing yield and propose three heuristics. Ferrer and Ketzenberg (2004) numerically extend the hybrid one-period model by Ferrer (2003) (discussed in the previous section) to a multiperiod setting. Ketzenberg, van der Laan, and Teunter (2006) provide an estimate for the value of information in a hybrid system, where the value of information is defined as the reduction in inventory costs (holding and shortage) as a result of reducing uncertainty in demand, returns, and remanufacturing yield.
In a series of papers, DeCroix and his co-authors study multiechelon inventory policies for hybrid systems under periodic review in multistage supply chains. DeCroix, Song, and Zipkin (2005) study a serial supply chain where product returns join new products directly in stock, without the need for remanufacturing. They show that an echelon base-stock policy is optimal. An echelon base stock policy maintains an echelon inventory position at stage j constant at , where the echelon inventory position at j is defined as inventory on-hand at j and at all downstream stages minus backorders at stage 1 plus on-transit inventory to j. DeCroix and Zipkin (2005) study complex assembly systems where there can be returns of end products; these returns are disassembled and a fixed subset of the recovered parts join existing inventory of those parts for future assembly. They propose some heuristic inventory policies and show that they perform well. DeCroix (2006) extends the single-stage hybrid system with remanufacturing of Inderfurth (1997) to a serial multiechelon supply chain. DeCroix shows that if product recovery occurs at the most upstream stage, then the optimal policy at that stage is similar to that of Inderfurth, with three parameters: remanufacture-up-to, order-up-to, and dispose-down-to levels, evaluated in terms of the echelon inventory position as defined above; the optimal policy for downstream stages is a simple echelon base-stock policy.
Periodic review or continuous-time product disposition approaches most often assume unlimited capacity, a stationary environment, and a single product case. The vast majority of these papers also assume that returns come in a single quality and thus, the unit remanufacturing cost is a constant. These assumptions can be relaxed with mathematical programming approaches, which we review next.
Multiperiod Nonstationary Disposition Models
In multiperiod production disposition, the firm matches supply of returns, remanufacturing capacity, and demand in a non-stationary environment. There are multiple quality grades for returns, and two disposition options in any period: remanufacturing or salvaging (a generic term that could mean dismantling for parts or recycling). The literature generally considers the expected value of demand and returns in each period and, similarly to regular (forward) production planning problems, suggests a rolling horizon approach as a practical way of dealing with uncertainty. Problems are formulated as a linear program where the decision variables are quantity of quality-i returns remanufactured in period t, salvaged in period t, inventory of quality-i returns at the end of period t, and inventory of remanufactured products in period t. The objective function minimizes production, holding, and backlogging cost, minus the salvage value, and constraints include inventory balancing constraints for returns and remanufactured products and production capacity constraints.
Ferguson, Guide, Koca, and Souza (2009) derive analytical properties for the optimal production plan, but the focus of their research is on finding the value of grading returns into a finite number of quality categories (as opposed to all returns going into a single pile), and the “optimal” number of quality categories. They find that most benefits are derived from using five quality categories (e.g., worst, bad, average, good, best) for categorizing returns. Guide, Pentico, and Jayaraman (2001) provide a framework and formulation for hierarchical remanufacturing planning when there are multiple products, although they do not conduct a comprehensive numerical study. Golany, Yang, and Yu (2001) do not consider multiple quality grades, but they assume that demand can also be met with new products (a hybrid system) and they consider a fixed cost to remanufacture in a period, which results in an integer program. Denizel, Ferguson, and Souza (2010) consider the case where the outcome of the grading process (that is, the number of returns obtained for each quality grade) is random in each period, and thus the linear program becomes a multistage stochastic program. They demonstrate the advantages of the stochastic programming formulation, as it detects infeasibilities that are not possible in the deterministic formulation. For example, returns with worse quality grades consume more capacity; if the firm realizes “bad” outcomes (many returns have bad quality) a few periods in a row then the problem may be infeasible if backlogs are not allowed or restricted, even though, in expectation, the firm has enough capacity.
In this article I present a literature review and tutorial on CLSCs, where I define the basic terminology, present basic modeling frameworks, and comment on main results and extensions. I divide discussion in terms of strategic and tactical decisions in CLSCs.
In terms of strategic decisions, research has focused on whether an OEM should remanufacture, trade-ins and leasing as a strategic source of returns, producer response to take-back legislation, CLSC network design, incentives and coordination among CLSC members, and to a lesser extent, on the impact of product recovery on new product design. Research in tactical decisions has focused on the used product acquisition strategy (when, how much, of which quality), and product disposition decisions (that is, what should a firm do with product returns: remanufacture, dismantle for parts, recycle, etc.). There is a significant stream of research in hybrid inventory management systems, where remanufactured and new products are perfect substitutes (that is, the firm has two sourcing options—remanufacturing and manufacturing); these systems are mostly found in the management of spare parts inventory.
There are many areas for future research. More empirical research is welcome (Vachon & Klassen, 2010), particularly on documenting acquisition and collection costs, remanufacturing costs, and the overall market for remanufactured products (including prices, warranties, and channels). The cost structure, in particular, drives a lot of the results, so it is imperative that research incorporates realistic cost structures (e.g., economies or diseconomies of scale, fixed vs. variable costs, and so forth). Research in consumer behavior and the market for remanufactured products is nascent, so a more comprehensive understanding of the market for remanufactured products is needed across different industries. Despite some attempts, I believe that the interface between new product design and recovery activities (remanufacturing and recycling) is an open area of research. In addition, the CLSC literature has focused heavily on remanufacturing, and there is relatively little research on recycling. Recycling has been incorporated in research dealing with take-back legislation, but the comprehensive design of products and respective CLSCs for recycling is a needed research area. From an environmental perspective, LCA studies have shown that remanufacturing is not always the environmentally preferred option, despite savings in material and energy during production (e.g., Quariguasi-Frota-Neto & Bloemhof, 2012). For example, old refrigerators should not be remanufactured but recycled, because the bulk of their environmental impact (in excess of 80%) is in the use stage of the life cycle, and thus, newer and more energy efficient refrigerators are preferred. When measuring environmental impact, it is thus necessary to take a life cycle assessment (LCA) perspective and report on the environmental impacts of producers’ choices, not only in terms of virgin material consumption (a standard metric reported in the literature), but also in manufacturing, distribution, use with consumers, and end-of-life.
From a methodological perspective, I believe there continues to be a need for practice-driven classical operations research and optimization models that can be used as decision support, and where numerical studies are based on values observed in practice. I also believe that there is a need for more empirical research, particularly econometric models that use actual (in the sense of not perceptual) data, as in Subramanian and Subramanyam (2012), and behavioral research with lab experiments.
Gilvan Gil C. Souza is an associate professor of operations management at the Kelley School of Business, Indiana University. Prior to Indiana, he was an associate professor at the Smith School of Business, University of Maryland, and he visited Georgia Tech's College of Management during his sabbatical semester in 2007. He also worked at Volkswagen of Brazil in new product development and product planning prior to entering academia. His research focuses in closed-loop supply chain management and sustainable operations. He is a senior editor for Production & Operations Management and an associate editor for Decision Sciences. He is the author of the book Sustainable Operations and Closed-Loop Supply Chains (Business Expert Press, 2012), and co-edited the book Closed Loop Supply Chains: New Developments to Improve the Sustainability of Business Practices (CRC Press, 2010). He won the Wickham Skinner EarlyCareer Research Accomplishments award in 2004 and the Wickham Skinner Best Unpublished Paper Award from 2008, both from the Production & Operations Management Society (POMS). He was the president of the College of Sustainable Operations of POMS from 2010 to 2012.