Estimating the Causal Effects of Pricing Interventions in Two-Sided Marketplaces Under Interference and Spillovers

Abstract

Two-sided marketplaces such as ride-hailing, food delivery, and home-sharing platforms connect buyers (riders, diners, guests) with sellers (drivers, couriers, hosts). A price change aimed at one side of the market often affects both sides. For example, raising rider prices can dampen rider demand and simultaneously influence driver earnings and availability. These cross-side effects are commonly called spillovers. This paper estimates effects that can be interpreted causally from naturally occurring price variations (such as surge pricing events or promotional discounts). Three sources of information are combined: (1) a short online survey of roughly three hundred participants (both buyers and sellers) asking which price changes they observed and how they reacted; (2) public information on dynamic pricing events (surges and discounts) in the marketplace; and (3) time-based proxies (peak hours, weekends, holidays) that correlate with shifts in prices or pay. Using these signals, the study constructs an indicator for an “own” price change and a measure of the share of neighboring areas that also experienced price changes at the same time. We then estimate an ordinary least squares (OLS) regression for each outcome of interest, regressing the outcome on the own price change indicator and the neighbor price change measure, controlling for routine factors and including fixed effects for areas and time periods. We carry out this analysis for five buyer-side outcomes (number of orders, wait time, cancellation rate, switching to a competing app, and spending per order) and five seller-side outcomes (earnings per hour, utilization of working time, number of jobs completed, acceptance rate of job offers, and short-run retention on the platform). According to the regression results, a 10 % increase in prices on the buyer side is associated with approximately a 6 % drop in the number of orders, about a 1 percentage point increase in the order cancellation rate, around a 5 percentage point increase in the probability that buyers switch to an alternative app, and a slight decrease in the average spend per order. Price increases in neighboring areas contribute additional negative impacts: for example, if all adjacent areas raise prices, local orders decline by roughly another 2 %, and switching to other apps rises by about 2 percentage points. On the seller side, a 10 % increase in pay (through surge pricing or bonuses) corresponds to roughly a 4 % increase in earnings per hour, about a 3 percentage point increase in offer acceptance rate, and around a 1.5 percentage point increase in the probability that a seller remains active the following week. At the same time, it is observed that when pay spikes, the utilization rate (the share of a seller’s time spent actively serving customers) tends to dip slightly, and total completed jobs do not rise much, suggesting that an influx of additional sellers may outpace the growth in demand. Notably, when buyer prices rise, the average wait time for buyers tends to decrease modestly; this counter-intuitive outcome likely occurs because higher prices suppress demand, allowing available drivers to be matched faster. Likewise, when seller pay surges, some sellers quickly come online or shift to the high-paying area, which can increase competition among sellers and result in a small decline in utilization. These findings show the importance of evaluating both sides of the market and accounting for local network spillovers when analyzing pricing changes. This work provides a simple framework for measuring direct effects and spillovers using limited data, which can support more informed pricing experiments and clearer communication of their expected results.