Research on Dynamic Pricing and Domestic Demand Expansion in Consumer Markets Under Reinforcement Learning
TL;DR
DE-CPPO is introduced and represents transaction, conversion, inventory, competitor-price and macroeconomic sequences as components of a Markov state and shows that it can represent observed retail-demand dynamics.
Abstract
Consumer-market prices, demand, inventory and competitive signals change simultaneously; therefore, revenue-only pricing can lead to unstable adjustments, delayed purchases and reduced access for price-sensitive consumers. This paper introduces DE-CPPO and represents transaction, conversion, inventory, competitor-price and macroeconomic sequences as components of a Markov state. Dilated temporal encoding and gated fusion support Proximal Policy Optimisation (PPO), while safety projection limits the price level, adjustment rate and segment disparity. Under controlled retail conditions, DE-CPPO increased revenue by 13.6%, raised the demand-expansion index to 1.154, improved coverage to 82.4% and reduced stockouts to 3.1%. Ablation and stress tests show that temporal encoding, the expansion-oriented reward and safety projection all contribute to performance. A public-data transfer test involving 811 products over 52 weeks yielded a normalised mean absolute error of 0.199 and a root mean square error of 0.265 for the temporal encoder, providing further evidence that it can represent observed retail-demand dynamics.