Do Cryptocurrencies Have Fundamental Values? Evidence from Machine Learning

The rise of FinTech is one of the most critical developments in finance over the past decade. One important FinTech development is initial coin offerings (ICOs), whereby investors can purchase blockchain-based cryptocurrencies directly from entrepreneurs. To those new to the crypto world, ICOs provide a mechanism to raise external funding through the issuance of digital coins. Such coins can be used to purchase products or services provided by the ICO project, or can be traded on the cryptocurrency market if the coin is listed.

While ICOs provide a new way of fundraising, there are extensive debates among practitioners and researchers about how to understand cryptocurrencies. On the one hand, there are growing concerns about whether speculation fuels the development of the market. For example, Satis—a security token advisory firm—claims that over 80 percent of ICOs in 2017 were scams.

There is also evidence of price manipulation in Bitcoin and other cryptocurrencies. On the other hand, many cryptocurrencies, such as Bitcoin and Ethereum, are highly valued on the market. The blockchain technology behind them is often referred to as the new internet, and some investors believe that it will bring revolutionary changes to every aspect of our lives. One focus of these debates is whether cryptocurrencies have fundamental value and to what extent the fundamentals affect the valuation of cryptocurrencies.

Investors participate in an ICO for two primary reasons. First, they may prefer the underlying products and the convenience coming from the security of the blockchain. One example is utility tokens, which give their holders the right to access products or services. Second, participants may purchase tokens for speculative motives. Studies show that investor sentiments can drive cryptocurrency prices. If cryptocurrency prices are purely driven by speculation, fundamentals should not matter when investors choose which ICOs to participate in. In our recent paper, we provide an empirical study on measuring the fundamentals of cryptocurrencies and their role in determining the performance of cryptocurrencies.

Identifying cryptocurrency fundamentals from disclosure documents

Cryptocurrencies are in a special asset class. Unlike traditional financial assets like stocks, cryptocurrencies do not distribute dividends, and there is no traditional accounting information for them. Cryptocurrencies are also different from fiat currencies in the sense that their value is not backed by any government. Many investors invest in cryptocurrencies because they trust the blockchain technology embodied in these digital coins. They believe that the blockchain technology is an important innovation and that at least some coins are assets that represent a stake in the future of this technology. Because of these distinct features of cryptocurrencies, we focus on the technology aspect when measuring the fundamentals of cryptocurrencies. Several recent theoretical papers in the cryptocurrency literature echo this viewpoint and emphasize the importance of technology in determining the viability and valuation of coins.

Measuring the fundamentals of an asset is always challenging in the finance literature, and it is particularly true for cryptocurrencies because of the limited information available. To overcome the challenge, we utilize ICO white papers to measure their technology components. In particular, we use several machine learning methods to construct measures of technology importance, or technology indexes, from a comprehensive database of ICO white papers.

First, we construct a technology index using supervised machine learning methods. We mimic the way investors evaluate white papers and manually assign scores to 200 white papers, where coins with higher scores are deemed to employ more sophisticated technologies. We train the model based on these 200 white papers and extrapolate scores to the remaining white papers. The supervised machine learning method we employ is a top-down approach that closely imitates the way investors assess ICOs.

Next, we use two unsupervised machine learning methods, word embedding and Latent Dirichlet Allocation (LDA), to construct the second and third technology indexes. The unsupervised machine learning methods are bottom-up approaches to study the textual elements of white papers. The advantage of the unsupervised methods is that they require little human input.

Lastly, we construct a composite index as the fourth technology index. We carefully validate these four indexes with measures of code quality from GitHub webpages of ICOs. The result suggests that the technology indexes offer a good proxy for the technological fundamentals of ICOs.

The figure below illustrates one of the methods we adopt—LDA. We classify all words in white papers into 20 topics. The size of each circle represents the proportion of words assigned to each topic, and the distance between the circles reflects the relationship between the topics. The technical topics (blockchain, algorithm and information), shown in red, are proxies of cryptocurrency fundamentals.

The effect of cryptocurrency fundamentals on pricing

Capital raising

To understand the role of fundamentals in cryptocurrency pricing, we start by studying the relationship between the technology indexes and ICO successes. We first examine whether the technology indexes are related to ICO fundraising. The ability to raise funding is one of the most important features in a successful ICO. If ICO performance is driven by speculation, investors would not care about the technology associated with the ICOs and the technology indexes would not predict ICO success. Instead, we find that ICOs with high technology indexes are more likely to raise capital and more likely to subsequently be traded in the secondary market. The economic magnitude of the effect is significant. For instance, a one standard deviation increase in the composite technology index is associated with a 10.4 percent increase in the listed probability, which is a 40.1 percent increase of the average of listed probability. The results suggest that the underlying technology of the ICO is important to investors.

Long-run performance

Next, we investigate whether the underlying technology is associated with subsequent performance. To test this conjecture, we examine the relationship between the technology indexes and the long-run performance of ICOs. We examine long-run performance rather than short-run performance because the process to fully incorporate technology-related information may take months due to the complexity of blockchain technology. We measure long-run performance using cumulative post-ICO returns, abnormal returns, and liquidity measures. We find that the ICOs with higher technology indexes tend to have better performance in the long run compared to other ICOs. A one standard deviation increase in the composite index is associated with a 23.9 percent increase in cumulative returns at the 300-day horizon.

We present additional evidence in support of the delayed reaction mechanism and attempt to rule out potential alternative explanations. An implication of the delayed reaction mechanism is that investors should be able to quickly incorporate the fundamental information if the white papers are written clearly. Consistent with the implication, we show that among the white papers with better readability, the long-horizon predictive power of the technology indexes is weaker. We also find that there is no return reversal phenomenon, suggesting that the return predictability results are unlikely to be driven by investor overreactions.


We also investigate whether our indexes help understand ICO failure, measured by a delisting. We find that the ICOs with higher technology indexes are less likely to be subsequently delisted. The economic magnitude of the effect is also large. For instance, a one standard deviation increase in the composite technology index leads to a 2.52 percent decrease in delisting probability.


Overall, these results suggest that the underlying technology is an important determinant of cryptocurrency prices and support the argument that investors do take the technological components in ICO white papers into consideration. However, it takes time for investors to differentiate the fundamentally sound ICOs from others. The delayed reaction from investors may be caused by investor inattention and the complex nature of the technologies, both of which necessitate more time to process related information.


Yukun Liu is an Assistant Professor of Finance in the Simon Graduate School of Business at the University of Rochester.

Jinfei Sheng (Corresponding author) is an Assistant Professor of Finance in the  Merage School of Business at the University of California, Irvine.

Wanyi Wang is a PhD student studying Finance at University of California, Irvine.

This post is adapted from their paper, “Do Cryptocurrencies Have Fundamental Values?,” available on SSRN.

Leave a Reply

Leave a Reply

Your email address will not be published. Required fields are marked *