Emptying blocks: the Hazardous Incentive Scheme Behind Blockchain Fee Formation.

Abstract

Multiple cryptocurrencies suffer a bottleneck effect: blocks are limited in size and the protocols limit their expected arrival rates. On the other hand, this congestion is creates incentives to set transaction fees. We show that this incentive structure suffers from moral hazard, where miners have incentives induce congestion to increase fees. We present this result using two approaches: Auction Theory and Reinforcement Learning. While Game Theory studies strategic behaviour between rational players, Machine Learning is based on blind players finding optimal strategies by \textit{brute force} iteration of trials. The Auction Theory model presented in this paper is a multiunit discriminatory (or pay-as-bid) auction with single unit demand. We add to the standard model the element of supply reduction, characterize the symmetric equilibrium and present how to expand it as the number of players grows asymptotically. The ML part focuses on Q-learning, a well-known application of reinforcement learning algorithms. The main finding has significant policy implications: decentralization, one of the core strengths of proof-of-work protocols, doesn't necessarily apply to block-level incentives.