Absolute Improvement and Relative Decline: Platform Ranking, Distribution Compression, and Moving Benchmarks

Online platforms commonly construct relative-performance rankings based on sellers' recent signals such as platform traffic and reputation. Owing to search frictions and related factors, sellers with higher rankings are more likely to obtain traffic and transaction opportunities. In practice, however, the number of sellers is large and rankings evolve continuously, so an individual seller does not compete against any specific rival, but instead chooses its strategy according to its position within the population. Against this backdrop, a key question arises: can a platform's relative-performance ranking continuously raise the competitive benchmark faced by sellers and drive platform competition toward an ecology characterized by "absolute performance improvement but relative position decline"? To address this question, this paper develops a mean field game model to characterize a dynamic competitive process in which the platform forms a composite ranking based on traffic and reputation signals, sellers improve their states through promotion and service investment, and consumer demand depends on relative rank positions. The results show that the intensity of ranking competition depends not only on the value of top display positions, but also on the shape of the population distribution: the more concentrated the distribution of the composite index, the denser the local quantile intervals, the greater the marginal return to rank improvement, the stronger the sellers' incentives to invest, and the higher the competitive benchmark. Furthermore, even when sellers' absolute performance continues to improve, their relative ranking may still decline as long as the density of the relevant local quantile interval rises more rapidly, thereby giving rise to a dynamic paradox of "absolute improvement with relative decline." The paper also shows that the transaction expansion and market concentration induced by top-position incentives do not necessarily arise from the same mechanism: the former mainly results from the platform's resource tilt toward top display positions, whereas the latter is driven more by distribution compression induced by rule parameters such as recommendation precision, refresh frequency, and multidimensional weights.