Job market paper — reading script (from practice)

Job market paper — reading script (from practice)


My job market paper is about e-commerce return policy. Now suppose you are a consumer who shops on Taobao. At the checkout, you’ll see a green label next to the product name. This label means the product is covered by return shipping insurance. So whenever you return a product with this return shipping insurance, part of your return shipping fee is reimbursed. This is the kind of policy I’m interested in. A return policy changes how costly a return is, and therefore a consumer’s willingness to pay. My job market paper evaluates existing return shipping insurance, then proposes a new policy that outperforms the existing one, and shows how the best return policy differs across products.

A return policy should encourage purchases and returns that generate positive social surplus. A return should occur if the seller’s value of the product exceeds the consumer’s value plus the shipping fee and the processing fee. A purchase should occur if the expected value generated when consumer keep the product exceeds the expected return cost. If a product is sold to a consumer who is most likely to return it, then this transaction shouldn’t happen in the first place. Our main result is that there is a trade-off between purchase and return efficiency.

Consumers tend to over-return under full refund because they do not internalize the cost that falls on the seller. The return product may have a stain on it, which takes time and effort to clean and resell it. Tightening the return policy solves this excessive return, but it also makes consumers more likely to keep the product, which strengthens the seller’s incentive to raise price. We call this force surplus extraction. And this force to raise the price can exclude valuable transactions when the profit margin is high. On the other hand, a more lenient return policy performs well by expanding valuable transactions when the profit margin is high, but at the cost of more over-return. This suggests a return policy should depend on the profit margin.

Actually, which return policy the platform should use depends on factors other than profit margin. It also depends on how much scrap value the seller can recover from the return product and the degree of match value uncertainty. So if the scrap value is high, the return doesn’t generate much loss. Therefore the seller doesn’t need to worry about the returned product, and the priority should be given to expanding purchases. In this case, a lenient return policy will do the job.

Now we’ll move to match value uncertainty. If the match value uncertainty is high, this means the consumer either super likes it or hates it. This creates a large option value, which makes the transactions valuable. Moreover, high match value uncertainty also makes the return rate less sensitive to the leniency of the return policy. This is because when the match value uncertainty is large, whether the consumer returns or keeps the product is mainly determined by the realized match value rather than how lenient the return policy is. So a generous return policy doesn’t affect consumers’ return rate a lot, and the distortion on the return margin is small.

So to conclude, whether a more generous return policy should be adopted depends on three things: profit margin, how much scrap value the seller can recover, and the degree of match value uncertainty.

So specifically, we apply this general intuition in the context of return shipping insurance. There are two types of return shipping insurance, depending on who pays the premium. If the seller pays the premium and bundles the insurance with the product, then we call it retailer return shipping insurance, RI. Under RI, the seller provides automatic coverage for all products sold in the store. So the consumer chooses between buying both the product and the insurance or not buying. On the other hand, if the seller doesn’t offer RI, the consumer can self-decide whether to add on the insurance at checkout, which we call consumer return shipping insurance, CI. In this case, the consumer chooses between three options: buy the product with insurance, buy the product without insurance, and not buy.

Using the data from a medium-sized Taobao store selling women’s clothes, we estimate a model of purchase, return, and insurance decisions. Within the same product, return rates tend to be higher under RI than under CI, and prices tend to be lower under RI, which is consistent with the surplus extraction story.

RI is a more generous return policy than CI because all purchases are insured. This is in contrast to CI, where only 3% of the consumer purchase the insurance. RI can expand valuable purchases but at the cost of more overreturn. Even without RSI, consumers return more than the socially optimal level. Reimbursing them the return cost just strengthens this over-return tendency. RI is better than CI in high margin, high scrap value, and high match value uncertainty categories such as glasses. In that category, RI greatly outperforms CI in terms of purchase efficiency by enabling many efficient purchases that do not occur under CI, while it causes slightly more over-return than CI. So in total, a great outperformance in terms of purchase but slightly worse performance in terms of return means RI is better than CI in this category when the three primitives are high. On the other hand, CI is better than RI in low margin, low scrap value, and low match value uncertainty categories such as sweater. In that category, CI greatly outperforms RI in terms of return efficiency by screening out some valuable transactions, while it is slightly worse in terms of purchase efficiency. So in total, a great outperformance in terms of return but slightly worse performance in terms of purchase means CI is better than RI in this category when the three primitives are lower.

The starkest contrast is in category tops, like T-shirts, where both the scrap value and match value uncertainty are very low. In this case, return is very costly and reimbursement strongly increases the return rate such that RI performs worse than CI in both purchase and return efficiency. In this category, RI leads to more than a 50% loss in welfare relative to first-best welfare by enabling many inefficient purchases and returns.

RSI is imperfect itself because it creates both moral hazard and adverse selection. It causes consumers to over-return because return is subsidized. It also attracts those low-value consumers who are most likely to return into buying the insurance, leading to expansion at the low end.

We then consider an alternative policy which we call a return penalty discount, RPD. For that policy, a consumer chooses between buying the product at the full price with a full refund or purchasing it at a discount rate, but later on needs to pay a penalty when returning the product. RPD is better in that it partially solves the ex-post over-return. Under the discount branch, the consumer returns without receiving the full refund, so the over-return tendency is mitigated. There is also no expansion at the low end, since the consumers who choose the discount option are those high-value consumers. Moreover, it allows the seller to screen out inefficient purchases using the product-level information. Now imagine a scenario where the return probability is low but return is extremely costly to handle. In this case, the insurance company, which sets the premium mainly based on the return probability, will set it too low, such that many consumers buy the insurance and return, creating a large handling cost for the seller.

We first consider a full RPD where the buyer receives no refund when they return the product. In this case, the return policy can be so strict that the intention to correct over-return overshoots. Consumers keep the product even if returning it is socially efficient. Even so, full RPD outperforms RSI in return efficiency. However, because the return rate under RPD is so low, the seller is incentivized to set a higher price to extract surplus, which would exclude many efficient transactions. In categories other than dresses, the exclusion is not as significant, such that a better performance in return efficiency still makes full RPD outperform or nearly match the existing RSI scheme. In dresses, full RPD performs worse than RSI by excluding too many efficient purchases. This is because the seller sets a very high price with a large discount such that all consumers choose the discount branch. For low-value consumers who return frequently, they simply do not buy.

Partial RPD with a 10 to 15 percent restocking fee as a penalty largely preserves the return discipline of full RPD, while enabling more efficient purchases. We show that it outperforms existing RSI in all categories other than dresses. And for dresses, it nearly matches — it only underperforms the best-performing RSI by 5 percent. So the takeaway is that if the seller has information about those three primitives — margin, scrap value recovery, and the degree of match value uncertainty — then the platform should set a lenient policy when those three primitives are high, while tightening it when those three primitives are lower. When the platform cannot make the return policy contingent on that information, partial RPD is a practical alternative that better balances purchase and return efficiency.