Decentralized Finance (DeFi) has been skyrocketed since 2020 (Figure 1). Existing literatures either give a comprehensive review of the industry from various perspectives, propose frameworks or formulate, define DeFi, or dive deep into money laundering, price prediction, token distribution, etc. However, seldom quantitative research adopts statistical methods to analyze the relationship between internal protocol design and external market performance. This research aims to expand the current research area and fill the gap.

According to Figure 2, Ethereum is still dominating the DeFi industry with over 70% of market share. In addition, Figure 3 shows that lending market is the biggest sector in DeFi. Therefore, this study focuses on Ethereum lending protocols, and 6 protocols are selected based on market cap and data availability, namely Compound, Aave V1, Aave V2, Maker, DeFiner and Cream Finance.

As multi-linear regression model is used for analyzing the linear relationship between dependent and independent variables, it is adopted to derive a model for each selected protocol.

Ordinary least square (OLS) is used to estimate the parameters.

Market performance of each protocol is represented by 24hr trading volume, in short trading volume.

Independent variables are selected also based on whether they fluctuate daily, fixed parameters and design are not considered. While data of cryptocurrency fluctuates, which means it is likely to update every 7-12 seconds in the case of Ethereum, all data used in this research is as Table 1 disclosed, calculated on a daily basis to avoid overfitting. According to the white papers of each protocol, only Compound has internal protocol design regarding the exchange rate, and only Aave provides two kinds of rates, which is stable borrow rate and variable borrow rate, so stable borrow rate, variable borrow rate and exchange rate are available only for Aave and Compound respectively.

APIs are the main data sources in this research, including those from DeFi Pulse, CoinGecko, Aave, Compound and Cream Finance. Independent variables selected are displayed in Table 1. Data are collected via Python, mainly requests module, and process through json, pandas and others. Overall, the collection process is as follow:

This analysis traces back the historical data as early as possible, considering data availability and completeness from the APIs provided by popular cryptocurrency price tracking websites and the official web APIs from the protocol itself. In this research, APIs from DeFi Pulse, CoinGecko, Aave, and Compound are adopted and merged while DeFi Pulse and CoinGecko are two of the most popular, prestigious cryptocurrency price trackers. As seldom platforms provide complete historical data and most of them just launched in recent years, some protocols are only open for limited time data.

The results are shown in a simply way as Table 2:

For Compound, price and cash shows positive, significant relationship with trading volume and exchange rate shows inverse, significant relationship with trading volume. Borrow rate and price shows positive, significant relationship and lend rate shows significant inverse relationship with trading volume in DeFiner’s model. For Aave V1, price and cash have huge positive relationship on trading volume, while for Aave V2, variable borrow rate shows significantl positive relationship and price with strong negative relationship with trading volume. For Maker, borrow rate shows inverse, significant relationship and price with strongly positive relationship with trading volume; Cream Finance only has price as a statistically significant and positive relationship with trading volume.

Looking at Table 2, the relationship of internal protocol design and external market performance isn’t stronger than external factors. $R^2$ and adjusted $R^2$ are in general not high enough to explain most of the changes in the trading volume of each protocol: the maximum is 0.561 for Aave V1 and minimum 0.066 for Cream Finance, indicating that at least about half of the changes in trading volume are possibly explained not by protocol’s internal design but by external factors like external market environment. Comparing $R^2$ of Aave V1 and Aave V2, it is obviously found that the $R^2$ and adjusted $R^2$ of Aave V2 model are way lower than V1’s.

Looking at coefficients and p-values of independent variables of each protocol, we can see that for these six examined protocols, namely Compound, Aave V1, Aave V2, Maker, DeFiner and Cream Finance, from the examination period, price always shows high statistically significance in each protocol and has positive relationship with trading volume in all protocols except Aave V2. Cash in Compound and two Aave versions also show a relatively outstanding connection with trading volume. Borrow rates show positive relationship in DeFiner and Cream Finance, and negative in Compound and Maker; stable borrow rate shows inverse relationship in Aave V2 and positive in V1. On the other hand, borrow rates and lend rates, in general, don’t show obvious trends or inferences to trading volume, which is an interesting finding as, according to their white papers, most of the protocols consider market supply and demand when designing the formulas of borrow rates and lend rates. Another finding is the exchange rate of Compound shows outstanding statistical significance within the relationship of trading volume.

This research acts as the first study to discuss about the relationship between internal protocol design and external market performance. It can be referenced by investors when conducting analysis of DeFi protocols and assist them when making investment decisions when selecting targets and protocols to invest in. From the data collection perspective, as all data used in this research comes from public, free and trustworthy API resources, more people can be inspired by this analysis and take advantage of it. As demand of using APIs increases, the maintenance and documentation of these APIs are expected to catch up. For further research, like mentioned previously, protocols on other blockchains can be analyzed; comparison between different blockchains or adding more protocols other than only select popular protocols can be made in further research. From the model perspective, as this analysis only applies multi-linear regression model to each protocol, other models may be analyzed in the future. From data perspective, different timeframes and granularity should be taken into consideration when conducting further research; methodology of grouping tokens under a protocol can also be adjusted.

Author: Chiao-An Tsai(Joanne)