Algorithm produces cleaner data for robust pricing
CAMBRIDGE, Mass., March 18, 2020 – Data is an important part of decision-making for pricing, but what happens when the data can’t be trusted? If the data is generated strategically by buyers, it could lead to bad pricing decisions by sellers. In a recent study, MIT Sloan School of Management Prof. Negin Golrezaei created a model which limits price manipulation by buyers, allowing sellers to collect reliable data to set optimal prices.
“In many marketplaces, buyers and sellers have access to data about products being sold over time. This can help sellers differentiate products and set possibly personalized prices, but it can also influence the willingness-to-pay of buyers. Buyers could be motivated to act strategically and trick the seller into lowering prices,” says Golrezaei. “A key question is whether there is a way to make the data more robust and limit buyers’ ability to manipulate prices.”
In a recent study, Golrezaei and her colleagues analyzed online advertising markets to see if they could find a way to mitigate this issue. They studied a model in which a seller runs repeated second-price auctions. Buyers are aware of the learning policy used by the seller and act strategically to submit lower bids. The untruthful behavior of buyers makes it difficult for the seller to learn the optimal reserve prices, which can prevent the seller from making effective data-driven decisions.
The researchers found that the key to limiting this price manipulation is using the right algorithm. Golrezaei explains, “We identified an algorithm that uses a binary signal to determine whether the seller wins in the prior auction. This limits manipulation by the buyer, who would need to change the binary signal to influence price. In other words, if they don’t manipulate the price, they can win the auction. But if they manipulate the price, then they won’t win. It becomes costly for the buyer to manipulate prices.”
Golrezaei adds, “This is good news for sellers because we showed that there is a solution to working with data that is prone to manipulation. And if it works in the advertising market, it should also be helpful in many other marketplaces too.”
She is the coauthor is “Dynamic incentive-aware learning: Robust pricing in contextual auctions,” with Adel Javanmard of the University of Southern California and Vahab Mirrokni of Google Research.
The MIT Sloan School of Management is where smart, independent leaders come together to solve problems, create new organizations, and improve the world. Learn more at mitsloan.mit.edu.
