Learning through Ratings under Endogeneous Product Quality
This paper analyzes how rating systems influence both consumer learning and seller incentives in markets with endogenous product quality, using structural estimation and counterfactual experiments to demonstrate that rating manipulation harms market outcomes while rating forgiveness improves revenue, quality, and welfare.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the vast, invisible marketplace of the internet, a strange problem persists: you cannot touch the product before you buy it. When you book a vacation rental or order a meal online, you are trusting a stranger's word and a few star ratings to tell you what the experience will be like. This gap between what a seller knows and what a buyer knows is called asymmetric information. Economists have long worried that this gap leads to two bad outcomes. First, the "lemons" problem, where bad products drive out good ones because buyers cannot tell the difference. Second, the "moral hazard" problem, where sellers might cut corners or stop working hard because they know buyers cannot see their daily efforts. To fix this, digital platforms built rating systems. These stars are supposed to do two things at once: they tell the next customer what to expect, and they push the current seller to keep their product in good shape. But it is not always clear if these two goals work together, or if they pull in opposite directions.
A new study by researchers at the University of California, Berkeley, and Nova School of Business and Economics in Portugal digs deep into this question using a unique window into the real world. They focused on the short-term rental market, the industry behind platforms like Airbnb, where hosts rent out homes for short stays. In this world, quality is not fixed; a house can get dirty, furniture can break, and a host can choose to fix it or ignore it. The researchers had access to a rare dataset from a management agency that handles hundreds of these rentals. Unlike most studies that only see the final star rating, this data showed the researchers exactly when a host paid for a deep clean, a roof repair, or new linens. They could see the money spent and the timing of these efforts, linking them directly to the reviews that followed. This allowed them to watch the invisible dance of effort and reputation in real time, rather than guessing at it.
The researchers built a detailed computer model to simulate how these markets work over a long period, treating the interaction between hosts and guests as a continuous game. They found that the system is alive with strategy. When a host receives a bad review, they do not just shrug it off; they are significantly more likely to spend money on maintenance in the following weeks to repair their reputation. This proves that the fear of a bad rating is a powerful motivator. On the other side, guests are constantly learning. When a property has a high average rating, more people book it. But the number of reviews matters, too. If a property has a high rating but very few reviews, guests are cautious. If it has a high rating and many reviews, guests feel confident. However, if a property has a low rating, adding more reviews actually hurts it, because the bad news is piling up and confirming the low quality.
The study also discovered that not all rentals are the same. The researchers split the listings into two groups: "value" properties, which are smaller, cheaper, and booked more often, and "premium" properties, which are larger, more expensive, and located in better areas. The premium hosts, who have more to lose, react more strongly to bad news and are more efficient at turning their maintenance money into better ratings. The value hosts, while also responsive, face a different set of pressures and costs. This difference matters when thinking about how to design the rating system itself.
To test how the system could be improved or broken, the researchers ran a series of counterfactual experiments. They asked: what happens if the platform allows fake reviews? Or what happens if the platform removes bad reviews that were caused by things outside the host's control, like a storm or a traffic jam? The results were stark and asymmetrical. When the researchers simulated a world where twenty percent of the reviews were fake five-star ratings, the market got worse. Even if the guests were smart enough to know that some reviews were fake, the hosts still reduced their maintenance efforts. Why? Because the fake reviews made it harder for the rating system to punish bad behavior. The hosts realized that they could get away with less effort, and the quality of the homes dropped. The booking rates and the actual quality of the stays declined, hurting everyone.
In contrast, when the researchers simulated a policy of "rating forgiveness," where the platform removed thirty percent of the negative noise—bad reviews that happened due to bad luck rather than bad service—the market improved. By removing these unfair penalties, the system became a clearer signal of true quality. Hosts, seeing that their hard work was less likely to be punished by a random bad review, invested more in maintenance. The quality of the homes went up, and guests booked more often. This effect was especially strong for the premium properties, where the hosts were already more sensitive to their reputation.
The core discovery of this work is that rating systems are not just passive scoreboards; they are active engines that drive behavior. Adding positive noise, like fake five-star reviews, is harmful because it dulls the incentive to work hard. Removing negative noise, however, is beneficial because it protects the good workers from bad luck, encouraging them to keep doing good work. The study suggests that for digital marketplaces to thrive, the design of the rating system must carefully balance the need for information with the need to motivate sellers. It is not enough to simply show the stars; the platform must ensure that those stars accurately reflect the effort behind them.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.