liquidity, similar to the zero-intelligence agent by Gode & Sunder (1993) without any real constraints besides their own budget. The other category is called chartist, which used different strategies involving genetic algorithms to optimize its revenue. They could confirm that the daily trading returns of absolute prices are indeed autocorrelated and follow a negative exponetialy shaped cumulative distribution function with a fat tail. In their next work, Cocco & Marchesi (2016) modeled the Bitcoin blockchain in-cluding the mining system. It added the new agent type miner. This category could acquire mining hardware and generate new Bitcoins. Its goal is also to maximize its revenue. The authors were able to reproduce stylized facts, such as the unit root and volatility clustering. This thesis is based on this model. In their newest publication, Cocco et al. (2019) solely focus on the mining aspect of Bitcoin and ignores the trading aspect. It compares the profitability of mining Bitcoin to that of mining gold and comes to the finding that mining Bitcoin is superior compared to mining gold. The authors state that if everything included, the costs of using Bitcoin might actually be smaller than those of a traditional currency.
Zhou et al. (2017) use a simple agent-based model to try to capture the effect of the amount of agents. Their findings, while limited, say that the amount of agents directly affects the Bitcoin price and the deal rate of transactions. A high number of agents decreases the price fluctuation and the overall deal rate is higher. Finally, Lee et al.
(2018) first use inversed reinforced machine learning to deduct trading rules from the real Bitcoin transactions. In a second step they applied these rules to an agent-based model in order to forecast the price and performance of Bitcoin. They were able to establish an 80% directional accuracy of their price prediction within the first six days of forecasting.
2.3 Summary
This chapter gave an introduction to the technology and design of Bitcoin, including its blockchain and parameters such as the block size and the Bitcoin generation. It also discussed current developments of Bitcoin and the resulting problems, such as conges-tion, high electricity consumption and security concerns. The second part summarized the economic literature of cryptocurrencies and Bitcoin. Several papers come to the conclusion, that the current protocol is not efficient from different point-of-views.
Dif-‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
ferent approaches have been theorized and proposed, such as totally new protocols and different parameters for the blockchain. Lastly, some literature has been introduced on agent-based modeling of a cryptocurrency. The model used in this thesis will be explained in the next chapter.
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
3 Methodology
In order to get more insight into the design of a cryptocurrency and its calibration, an agent-based model is created. The goal of the model is to find a better way to design a cryptocurrency that allows for a high security and high economic efficiency simul-taneously in a Pareto optimal fashion. Such an agent-based model approach has been shown to be successful in the economic literature on policy design. Chen & Chie (2008) for example used an agent-based model for the design of a lottery market, while Marks (2006) introduces market design in agent-based modeling in a more general fashion.
In the first step, a model was created that tries to simulate the real cryptocurrency Bit-coin. This chapter will describe the model and its agents in detail.3 Section 3.1 will begin describing the model by giving an overview of its core features and agents. Sec-tion 3.2 will explain the different types of agents used. Section 3.3 will describe the model with its market and transaction system. Section3.4describes the initialization of the model. After the model has been created, it will get calibrated and optimized which is described in Section3.5. Lastly, in Section3.6, the chapter will be summarized.
3.1 Model Overview
The model used is based on an existing one by Cocco & Marchesi (2016), which was then adapted and extended. The core features of the original model include a Bitcoin market, where agents can sell and buy Bitcoin in a realistic fashion. Furthermore, agents are able to create (i.e., mine) new Bitcoins. The full list of the features of the original model is:
• Three kinds of agents, including chartists, random traders/users and miners;
• A realistic order book to imitate a real cryptocurrency market;
• Agents join over time and decide to engage in the trade and/or to mine Bitcoins;
• A power law wealth distribution for existing and later added agents;
• Miners are all in mining pools, meaning that they have a steady stream of income.
This is because since 2010 miners collaborate in groups to share their hashing
3The full code can be found under https://www.comses.net/codebases/3d2dfc87 -78e0-47ff-ad8f-7c8f5f13fcdd/releases/1.0.0/
‧
國立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
power to decrease their investment risk by avoiding unnecessary mining (Cocco
& Marchesi, 2016). The miner agents also can decide at given time points to invest in or divest their hardware.
This list of features was then extended by a transaction system, similar to the one used in Bitcoin as of today. It includes a transaction capacity limit, that allows users to submit transactions and set a fee for their transactions as well as letting miners take fees in general into their investment consideration. This transaction system allows us to analyze the development of the transaction fees as well as a better representation of the mining network hashing power. The model simulates each day between February 23rd in 2014 and April 3rd 2018 in a single time step, totaling 1500 simulation steps, where one step is equal to a whole day. This particular time frame was choosen as it was used by the creators of the underlying model from Cocco et al. (2017). It also captures the time before and during the time of Bitcoins highest congestion around the end of 2017 (Bitcoinfees.info, 2019).