Competitive Demand Learning: a Data-Driven Pricing Algorithm
Dynamic pricing is used to maximize the revenue of a firm over a finite-period planning horizon, given that the firm may not know the underlying demand curve "a priori". In emerging markets, in particular, firms constantly adjust pricing strategies to collect adequate demand information, which is a process known as price experimentation. To date, few papers have investigated the pricing decision process in a competitive environment with unknown demand curves, conditions that make analysis more complex. Asynchronous price updating can render the demand information gathered by price experimentation less informative or inaccurate, as it is nearly impossible for firms to remain informed about the latest prices set by competitors. Hence, firms may set prices using available, yet out-of-date, price information of competitors.
In this paper, we design an algorithm to facilitate synchronized dynamic pricing, in which competitive firms estimate their demand functions based on observations and adjust their pricing strategies in a prescribed manner. The process is called "learning and earning" elsewhere in the literature. The goal is for the pricing decisions, determined by estimated demand functions, to converge to underlying equilibrium decisions.
The main question that we answer is whether such a mechanism of periodically synchronized price updates is optimal for all firms. Furthermore, we ask whether prices converge to a stable state and how much regret firms incur by employing such a data-driven pricing algorithm.