Product Pricing using Adaptive Real-Time Regime-Based Acceptance Probability Estimates

In a complex supply chain, in which several traders compete in a component procurement market as well as in a sales market where assembled products are sold through reverse auctions, product pricing is a vital, yet non-trivial task. In this thesis, a product pricing approach using adaptive real-time regime-based probability of acceptance estimations is proposed. Based on economic regime estimations, price distributions are approximated, which are adapted using relevant available information on prices and characteristics of customer requests for quotes. Artificial neural networks are trained to act as adapter and estimate parameters for the double-bounded log-logistic function assumed to be underlying the prices. This adaptation differs per market condition and is corrected using an error factor, which is updated on-line. Given the parametric approximation of the price distribution, the probability of acceptance is estimated using a closed-form mathematical expression. This expression can then be used to determine the price yielding a desired quota. The approach is implemented in the MinneTAC trading agent and tested against a price-following product pricing approach in the TAC SCM game. The new product pricing approach yields a significant performance improvement; more orders are obtained against higher prices. Profits are more than doubled.