Monte-Carlo and recency-weighted learning methods for conjectural variations in dynamic power markets

Conjectural variations based bidding strategies have been proved to be a more appropriate model to analyze bidding profile of players in an electricity market than other game theoretic models. The equilibrium quantities and market clearing prices result from Nash-Cournot equilibrium are far from real markets data. However CV has been criticized for having no definite meaning in static form. In this paper we proposed the dynamic form of quantity setting conjectural variations. Dynamic optimization and conjectures learning which set the collection of dynamic nonlinear state equations are obtained. Monte-Carlo and recency-weighted learning methods are introduced and their effects on equilibrium and MCP are investigated. The results of simulations verify that learning would lead to greater social welfare and more realistic price.