Blockchains employ internal and external incentive structures to influence participant behaviour, maintain network security, and ensure stable throughput. Internal incentives, like block rewards and transaction fees, are embedded within the blockchain design, while external incentives arise from market forces and competition. Both incentive structures are crucial for shaping blockchain behaviour and network efficiency.

The primary motivation of this thesis is to examine how misaligned incentive structures can negatively affect stability in Proof-of-Work blockchains, focusing on stable block and transaction throughput. The thesis aims to provide novel insights into the negative impact of unstable throughput on individual agents and the network as a whole. Additionally, it evaluates potential attack vectors resulting from misconstructed incentive structures, past exploits, and proposes fairer and more robust mechanisms to align incentives, ensuring higher throughput stability and network security.

The contributions of this thesis include the development of an open-source simulation framework called PoolSim. It enables the analysis of miner behaviour under different mining pool reward distribution schemes, including the profitability evaluation of queue-based manipulation strategies and pool-hopping between such pools. The thesis introduces the uncle traps attack, specific to Ethereum queue-based mining pools, which adversely affects block throughput and presents a fix to the uncle block reward distribution mechanism.

Furthermore, this thesis examines the impact of difficulty adjustment algorithms on block throughput. It identifies instability in block solve times due to cyclicality observed in Bitcoin Cash and analyses how miners' behaviour contributes to this phenomenon. A novel difficulty algorithm based on a negative exponential filter is derived, eliminating oscillations and ensuring more stable block solve times.

Lastly, the thesis addresses transaction throughput improvement by presenting a gas price prediction model for Ethereum. It combines deep-learning-based price forecasting with an urgency-based algorithm, optimising the trade-off between timely inclusion and transaction cost. Empirical analysis and real-world evaluation demonstrate significant cost savings with minimal delays compared to existing mechanisms.