De-Crypto-Ing Signals in Initial Coin Offerings

Home , Huobi, Initial coin offering, Telegram Open Network

De-crypto-ing Signals in Initial Coin Oﬀerings: Evidence of Rational Token Retention

Tetiana Davydiuk, Deeksha Gupta, and Samuel Rosen∗

Original Draft: October 2018 This Draft: July 2019

Abstract

Using the market for initial coin offerings (ICOs) as a laboratory, we provide evidence that entrepreneurs use retention to signal their quality to investors in the presence of asymmetric information. ICO investors face a high degree of uncertainty because of the unregulated and opaque nature of the ICO market. Using a detailed dataset on 7,450 ICOs, we show that ICOs that retain a larger fraction of their tokens are more successful in their funding efforts and are more likely to develop a working product or platform. Specifically, we estimate that a 1 p.p. increase in token retention leads to a 0.1-0.3 p.p. increase in fundraising success. Moreover, we find that signaling incentives are stronger in periods of elevated market noise.

Keywords: asymmetric information, signaling, entrepreneurial ﬁnancing, ICOs

JEL Classiﬁcations: D82, G32, L26

∗Tetiana Davydiuk ([email protected]) and Deeksha Gupta ([email protected]) are at the Carnegie Mellon University. Samuel Rosen ([email protected]) is at the Fox School of Business at Temple University. We would like to thank James Albertus, Gurdip Bakshi, Hugo Benedetti, Dean Corbae, Selman Erol, Marvin Goodfriend, Todd Gormley, Hanna Halaburda, Zhiguo He, Burton Holliﬁeld, Stephen Karolyi, Theresa Kuchler, Bryan Routledge, Harald Uhlig, Ariel Zetlin-Jones, attendees at the 2019 Toronto Fintech Conference, the 2019 NBER meeting on Blockchain, Distributed Ledgers and Financial Contracting, and seminar participants at Temple University, the Oﬃce of Comptroller of Currency, and Carnegie Mellon University for their helpful comments and discussions.

Electronic copy available at: https://ssrn.com/abstract=3286835 1 Introduction

A classic concept in economics is that a privately informed seller may want to signal information about quality to less informed buyers. Although the theoretical literature has long proposed costly retention as a method for signaling quality, studying this question empirically poses a number of challenges. First, the presence of regulators, intermediaries, and legal protections reduce the problem of asymmetric information in many markets. Second, in-person interactions between buyers and sellers can also affect the information asymmetry problem through the conveyance of soft information. Third, a researcher must have sufficient data to control for information that could affect a buyer’s decision besides retention. In this paper, we address these empirical challenges by studying the role of retention in an ideal laboratory that has not yet been considered: the market for initial coin offerings (ICOs).

ICOs can be thought of as a hybrid form of crowd-funding. They have emerged in the last few years as a novel form of financing for young firms with $33.4 billion raised in aggregate through 2018. In simple terms, an ICO is the sale of digital assets, “tokens”, by an entrepreneur or company to fund the creation of an online platform or ecosystem that uses blockchain technology.1 The token being sold can be used on the platform that the issuer is creating. Typically, the token is the sole currency that can be used to purchase products and services on the platform. Thus, the value of the token is derived not from profit-sharing as most tokens do not have any cash-flow rights. Rather, the token gets its value because once the platform is completed, the token can be exchanged for the product that is being sold on the platform, making it similar to crowd-funding. Unlike crowd-funding, however, investors can sell their tokens in liquid secondary markets (exchanges).

To study how retention aﬀects signaling, we ideally want a market that is liquid while being prone to a high-degree of information asymmetry. The ICO market has grown rapidly, but remains largely unregulated and is inherently opaque. The lack of regulation is due in part to the fact that ICOs take place globally over the internet, which makes potential enforcement diﬃcult. The market is opaque because there are neither intermediaries nor

1We will refer to the digital assets sold in an ICO as “tokens” and reserve the word “coins” only to refer to cryptocurrencies. We realize this may seem confusing given that the word “coin” is in the term ICO, but it is growing convention to refer to these digital assets in this way. In fact, some parties in the crypto industry are pushing to replace the term ICO with either “token sale” or “token generation event” (Token Alliance, 2018).

Electronic copy available at: https://ssrn.com/abstract=3286835 face-to-face interactions between ICO investors and entrepreneurs. This combination of features is unique relative to other forms of ﬁnancing (e.g., initial public oﬀerings or venture capital). As a result, ICO investors face a uniquely high degree of information asymmetry with prospective issuers.

The contracts underlying ICOs are unique in that their terms are enforced through software instead of the legal system. This software, which is publicly available and veriﬁable, is known as a “smart contract.” This feature of ICOs creates a clear distinction between promises that are enforceable and those that are not. Essentially, something is contractible through a smart contract if and only if it can be written in computer code.2 For example, an issuer can clearly state the total number of tokens to be created and the number of tokens available for purchase by investors. However, any piece of information (e.g. resumes of entrepreneurs) or future promised action (e.g. legal enforcement) that cannot be translated into computer code could not be included a smart contract, thereby being unveriﬁable from an investor point of view. As such, ICO issuers would face no ex-post repercussions of misrepresenting such information.

There is no comprehensive public database available on ICOs so we compile our own rich dataset of 7,450 ICOs. We do so through a combination of web scraping, PDF scraping, collection by hand, and manual review. Our primary sources are websites that provide information about ICOs to potential investors, but we also rely on documents provided by issuers and manual validation. For each ICO, we have numerous ex ante characteristics as well as several ex post outcomes of interest. We use the fraction of tokens available for sale to potential investors as our measure of retention. It is important for our analysis that this retention measure can be coded into a smart contract so that it is a credible ex ante commitment.

Before describing our empirical analysis, we formalize the intuition using a stylized model of the ICO market. The model is similar to standard models of retention such as Leland and Pyle (1977) and Demarzo and Duﬃe (1999). Unlike equity value which is based on proﬁt- sharing, the value of the token derives from the expected value of the services that it can be traded for if the platform is completed successfully. In the model, penniless entrepreneurs raise funds to develop their platform through an ICO from a large measure of deep-pocketed

2A smart contract can be thought of as containing “if-this-then-that” statements. It is important that the “then-that” is something that can be executable through a computer and not something that requires any real-world actions (for example, legal action).

Electronic copy available at: https://ssrn.com/abstract=3286835 investors. Entrepreneurs have private information about their type. Greater token retention by insiders is costly as it forces ICO entrepreneurs to hold a large share of their wealth in a single, risky asset. However, it is less costly for ICO issuers with high-quality projects to retain a share of tokens because they are more likely to successfully build their platform making their tokens more valuable. In the model equilibrium, high-type entrepreneurs choose to retain a larger fraction of tokens than low-type entrepreneurs to signal their quality and separate from low-type entrepreneurs. This outcome causes a positive relationship between the fraction of tokens retained by entrepreneurs and the quality of the ICO. When noise in the market increases and the quality of information investors get through a public signal about ICOs decreases, this relationship becomes stronger and high-type ﬁrms retain a larger share of tokens on average.

We perform numerous empirical tests to assess our theoretical predictions. First, we examine whether ICOs that retain a greater fraction of tokens are more likely to fund raise successfully. When a ﬁrm announces its ICO, it usually states a “hard cap,” which is the maximum amount of funds it wants to raise and is willing to accept. For each ICO in our dataset, we divide the funds they raised by this hard cap to get a continuous measure of ICO fundraising success. In line with retention as a means of signaling quality, we show that a 1 percentage point increase in the fraction of tokens retained by entrepreneurs is associated with an increase in fundraising success by 0.1-0.3 percentage points.

Second, we test if the relationship between token retention and fundraising success is stronger when the market is noisier. We measure noise in the market at the time an ICO is happening by the number of other ICOs that are happening at the same time, adjusted for the amount of information provided by token issues in ICO prospectuses (commonly referred to as “whitepapers”). If only a few ICOs are happening in a month and/or whitepapers have less content, investors can undertake more due diligence on each ICO - they have more time to study the various ICOs, examine their prototype, read analyst reports etc. On the other hand, at a time when many ICOs are taking place it is harder for investors to undertake such detailed due diligence. We therefore believe that the number of ICOs happening at the same time as the ICO of interest, adjusted for the amount of information provided in the whitepapers, is a good measure of how noisy the market is. We ﬁnd that in accordance with our model predictions, the relationship between retention and fundraising success is stronger at times when ICO markets are noisier.

Electronic copy available at: https://ssrn.com/abstract=3286835 Third, we show that greater retention is related to better ex-post performance. We measure ex-post performance in a variety of ways. In line with Amsden and Schweizer (2018), we measure ICO success by future tradibility - if the token begins to trade on a reputable exchange following the ICO. The provision of ex-post liquidity to buyers is an important source of value and a key distinguishing feature of ICOs from venture and angel capital, the other prominent forms of early stage financing. However, only relatively high- quality ICOs list their token on an exchange. Exchanges engage in due diligence prior to accepting a token for listing to protect their reputation both with customers and regulators. Furthermore, the fees ICO have to pay to list on an exchange are non-trivial and can range from $1 to $3 million.3 Therefore exchange listing is an ex-post measure of quality. We find that ICOs that retain more tokens are more likely to be listed on exchanges, further confirming our hypothesis that greater retention signals higher quality. We additionally measure ICO success by whether the company still has a working website, has a live platform and has an application available to download on the Apple store. In line with our hypothesis, we find that ICOs that retain more tokens have better ex-post performance using all our measures.

The technology of smart contracts ensures that our results are not confounded by hard information that is typically not possible to control for. In other markets, there are reputation costs and legal consequences for misrepresentation of certain hard information (e.g. entrepreneur’s education or work experience) allowing investors to rationally condition their decisions on it. Accounting and controlling for all such information in an empirical design is not feasible. In ICO markets, we have a limited set of enforceable promises that can be coded into a smart contract. Since these promises are public, we control for relevant hard information.

Furthermore, traditional ﬁnancing contracts which rely on legal enforcement can be quite complicated with multiple clauses and terms. Clauses such as anti-dilution protections make inference between the share retained by the entrepreneur and the price investors are willing to pay for shares tenuous. The current contracts in ICO markets are relatively simple, making the interpretation of our results straightforward.

Moreover, the lack of face-to-face interactions between investors and ICO entrepreneurs greatly limits soft-information exchange that would otherwise be a concern in other

3https://www.bloomberg.com/news/articles/2018-04-03/ crypto-exchanges-charge-millions-to-listtokens-autonomous-says.

Electronic copy available at: https://ssrn.com/abstract=3286835 markets. For example, in-person meetings convey information to potential investors about an entrepreneur’s presentation skills, perceived intelligence, demeanor, manner of speaking etc. If these variables are related to skill, then an investor would ask an investor with good soft skills for a smaller share of the ﬁrm given a set fund raising goal. This would create a correlation between retention and project quality, that would not be due to a signaling channel. ICO fund raising takes place online with little to no interaction between the team developing the platform and purchasers of the token. We therefore have less confounding soft-information exchange that we need to worry about.

Our OLS results are unlikely to be confounded by most soft information exchange that occurs in personal interactions. However, we may still be concerned about soft information that is conveyed online which is unverifiable but costly to produce. To assuage this concern, and further uncover the causal effect of signaling on an ICO’s fundraising success, we implement an instrumental variables (IV) strategy. We use the Financial Development (FD) Index of the country where the ICO team is located as an instrument for token retention. Token retention is relatively more costly for entrepreneurs who have limited access to financial markets and cannot easily diversify risk away. We argue that an entrepreneur’s location and its corresponding level of financial development is independent of the project quality. As all token sales take place online and are accessible to investors worldwide, the entrepreneur’s location has no effect on an ICO’s fundraising success beyond its effect through the cost of signaling. In the first stage regression, we find that entrepreneurs with limited risk-sharing opportunities, as captured by the lower FD Index, sell a larger share of their project to outside investors. Using this exogenous variation in cost of signaling across countries, we further document that a 1 percentage point increase in token retention results in a 1.1 - 2.0 percentage point increase in fundraising success.

Our paper contributes to the empirical literature on signaling through retention. Ours is the ﬁrst to focus on the ICO market, which is unique in its lack of regulation and intermediation and the use of software to enforce contracts. The combination of a high degree of information asymmetry and high-quality data makes the ICO market is an ideal laboratory. Other studies look at the markets for initial public oﬀerings (IPOs), mortgages, and commercial real estate.4 The empirical evidence is mixed in the IPO setting where

4Adelino et al. (2019) provide evidence of a signaling mechanism through trade delays by originators in the residential mortgage market. Contrary to expectations, Agarwal et al. (2012) do not find systematic differences between subprime mortgages sold in the secondary market and those retained on banks’ balance sheets. Purnanandam and Begley (2016) find that higher levels of equity tranches in private label residential

Electronic copy available at: https://ssrn.com/abstract=3286835 retention is defined as a fraction of ownership, which is similar to our measure token retention in ICOs. Downes and Heinkel (1982) and Clarkson et al. (1991) find positive support for the share of equity retained by insiders in an IPO as a signal while Krinsky and Rotenberg (1989), Ritter (1984), and Schultz and Zaman (2001) do not. In the ICO setting, we not only find positive evidence for signaling through retention but we also show how the strength of this signal depends on market environment. This latter aspect of our results is similar to a finding from Purnanandam and Begley (2016) that the signal from higher level of equity tranche in RMBS deals becomes less strong after the passage of antipredatory lending laws.

Our paper also contributes to the growing literature that specifically analyzes ICOs. One strand of this literature investigates what drives ICO success.5 Broadly speaking, these papers rely on the same data sources as we do but they focus on a different goal: uncovering the set of key factors associated with ICO success. A consistent message in their results is that disclosure and transparency are important. Ahlers et al. (2015) reach a similar conclusion in the equity crowdfunding market. Another strand of this literature studies post-ICO returns in the spirit of the IPO literature.6 In contrast to the papers that analyze the determinants of ICO success, we use the ICO market to empirically study retention as a means of signaling in an environment with high information asymmetry. Our contribution is to show that greater token retention is, in fact, associated with ICO fundraising success and higher quality issuers. Moreover, we find that this relationship is stronger during periods in which markets are noisier. We provide evidence that given the lack of regulation in the ICO market, entrepreneurs and investors may be using tools to signal their quality and pick high quality investments respectively. mortgage-backed security deals are associated with lower delinquency rates and higher prices. Downing et al. (2008) find that mortgages sold to special purpose vehicles tended to be of lower quality. Fuchs et al. (2016) provide evidence that higher-quality firms wait longer to have their IPO. Garmaise and Moskowitz (2003) find no evidence that informed sellers of commercial real estate signal their information through retention. 5See Adhami et al. (2018); Amsden and Schweizer (2018); Boreiko and Sahdev (2018); Bourveau et al. (2018); Deng et al. (2018); Fisch (2019); Jong et al. (2018); Howell et al. (2018); Lyandres et al. (2018). Similar to us, these papers use the achievement of stated funding goal, amount of funds raised, or listing on an exchange as measures of success. Howell et al. (2018) uniquely focus on ex post trading measures (liquidity and volume). 6Benedetti and Kostovetsky (2018) observe an average first-day return of 179%. They find that cryptocurrency prices and social media activity are significant determinants of ICO underpricing. While they believe their results could indicate an irrational bubble, they conclude that the observed ICO underpricing is also consistent with high compensation for risk. Momtaz (2018a) and Momtaz (2018b) explore both short-term and medium-term returns after an ICO. He argues that his results are consistent with a “market liquidity hypothesis” that proposes ICO issuers want to underprice to generate ex post market liquidity for the token. Lyandres et al. (2018) conclude that post-ICO empirical patterns resemble those observed in the IPO market.

Electronic copy available at: https://ssrn.com/abstract=3286835 In addition to empirical studies, researchers have begun to analyze the properties of ICOs in a theoretical setting. In general, these papers highlight channels through which the novel aspects of an ICO can improve or worsen outcomes relative to traditional financing. Catalini and Gans (2018) highlight that the ICO mechanism can elicit the value that consumers place on the underlying project. Both Li and Mann (2018) and Sockin and Xiong (2018) find that an ICO can solve the coordination failure problem in creating a new platform. In a dynamic setting, Cong et al. (2018) highlight endogenous user adoption in ICOs. Cong and He (2018) point out the nonlinear relationships between equilibrium values for token price, platform productivity, and network size. Sanchez (2017) proposes an optimal ICO mechanism to improve upon the current set of models. Chod and Lyandres (2018) analyze the decision to use venture capital or do an ICO in a model with agency conflicts between issuers and investors. Bakos and Halaburda (2019) study how token tradability and broader crowdsourcing of due diligence affect the decision to use an ICO. As part of our analysis, we develop a simple model to formally evaluate the intuition behind our key empirical findings. We regard our contribution as primarily empirical as our theoretical model follows closely the analysis already done by seminal theories in the literature. Through the model we are able to formalize the intuition of retention as a form of signaling in the ICO market.

The rest of this paper is arranged as follows. Section 2 gives an overview of blockchain technology and the ICO market. Section 3 describes the model and empirical predictions. Section 4 describes our data and provides summary statistics about the ICO market. Section 5 describes our results. Section 6 concludes.

2 Background: Blockchain, Cryptocurrencies, and Initial Coin Oﬀerings

In this section, we provide key institutional details, a brief history, and an overview of the regulatory treatment for ICOs. First, however, we describe the basics of the crypto space because ICOs would not exist without blockchain technology and cryptocurrencies.

A blockchain is a digital, decentralized, distributed ledger for a specified digital asset. It is a file that contains the sequential list of transactions for that digital asset. Blockchain technology refers to the combination of this file with a network of computers (nodes) and

Electronic copy available at: https://ssrn.com/abstract=3286835 software that dictates how the blockchain is monitored and updated. Using cryptography and an incentive scheme to reward participation on the network7, this technology enables the direct and secure transfer of the underlying digital asset without a central authority to verify the transfer of ownership.8 A cryptocurrency is a digital asset that is intended to be used as a medium of exchange. Although they may diﬀer in technical details, all cryptocurrencies exist and are traded on a blockchain. In 2008, Bitcoin was the ﬁrst cryptocurrency ever created (Nakamoto, 2008).

As discussed in the introduction, the tokens that are sold in an ICO have a functional purpose in the product or platform that the issuer is creating. This “utility” aspect is the key diﬀerential between cryptocurrencies and tokens. It is also an important characteristic in the current regulatory debate (discussed more below). To provide a concrete example, let us consider the STORJ token (https://storj.io/). The underlying product is a blockchain-based decentralized cloud storage solution that allows users to store their data in the otherwise unused storage space owned by other people. One can either rent space on the devices of others using the STORJ token or lease out their own space in exchange for STORJ tokens. In this example, it is easy to see how the token is an integral part of the blockchain-based platform. This example also highlights the early-stage nature of platforms that choose to ICO. The Storj project issued STORJ tokens in May 2017. As of October 2018, the company’s website promises that the Storj network is launching in early 2019.

ICOs are a relatively new phenomenon. In 2013, a software developer named J.R. Willett launched a social media fundraising campaign that promised 100 newly created MasterCoins for every Bitcoin received. He used the approximately $500,000 of funds raised to both develop the software and build an exchange to trade the newly created tokens. The success of MasterCoin encouraged other similar campaigns between 2013-2015. Campaigns during this time period did not actually use the term “ICO” but instead used terms such as “crowd sale”, “donations”, or “presale” (Boreiko and Sahdev, 2018).

The ICO market did not begin to resemble its existing form until 2016 after the development of the Ethereum blockchain. The Ethereum blockchain, which is denominated in the cryptocurrency Ether, was itself funded through a crowdfunding-style ICO in 2014. Ethereum was the “world’s ﬁrst programmable blockchain” that allowed users to design

7Mining refers to a process in which participants use computers to solve mathematical puzzle and compile recent transactions are compiled into blocks. 8For a more detailed description, see, e.g., Narayanan et al. (2016).

Electronic copy available at: https://ssrn.com/abstract=3286835 “smart contracts.” As a result, users could easily create proprietary tokens that existed on a platform on top of the Ethereum blockchain. Before this advancement, an ICO would have to build and maintain their own blockchain. In September 2015, Augur was the ﬁrst ICO to issue a token that ran on the Ethereum blockchain.

Since 2016, the stylized process for an ICO is as follows:

1. Conception. A team of entrepreneurs conceives a blockchain-based project that requires funds and users

2. Announcement. The team announces project details and the planned sale period on cryptocurrency forums/websites

3. Payment. In the ICO period, prospective buyers submit orders by sending cryptocurrency payments to the issuer

4. Distribution. At the sale’s conclusion, the token contract automatically sends the purchased tokens to successful buyers

Despite their relatively short history, ICOs have evolved to a relatively standardized procedure in many ways to reduce investor uncertainty (Massey et al., 2017). First off, issuers typically produce a “white paper” beforehand that describes project in detail and provides key aspects of the sale such as token supply and pricing. This document serves like a prospectus in an initial public offering. Second, most ICOs choose to sell a certain number of tokens on a first-come, first-served basis at a fixed price. This pricing mechanism means that there is a cap on the amount of money to raise. Third, most ICOs set a time limit for the sale to occur.

In the United States, ICOs eﬀectively remain unregulated and their future regulatory treatment is still uncertain. Since 2014, cryptocurrencies have been formally designated as commodities and, accordingly, fall under the regulatory jurisdiction of the Commodity Futures Trading Commission (CFTC).9 The key issue for ICOs is whether the token being sold is a “security.” If so, they will be regulated under existing securities laws with oversight from the Securities and Exchange Commission. As part of their 2017 investigation into

9Testimony of CFTC Chairman Timothy Massad before the U.S. Senate Committee on Agriculture, Nutrition and Forestry (Dec. 10, 2014), http://www.cftc.gov/PressRoom/SpeechesTestimony/ opamassad-6.

Electronic copy available at: https://ssrn.com/abstract=3286835 the post-ICO hack of The DAO, the SEC concluded that the tokens issued were, in fact, securities.10 This ruling put forth the notion that future ICOs could similarly be categorized as securities. Based on the limited precedent so far, it is still not clear how ICOs will be treated in the US moving forward.11

The regulatory response has diﬀered widely across the rest of the world. On one hand, China and South Korea made ICOs illegal in 2017. On the other hand, several countries including Singapore and Switzerland have adopted a much friendlier approach. Singapore puts new ﬁnancial innovative ventures in a “regulatory sandbox” from which they can emerge with a government license.12 Switzerland has been described as the world capital of cryptocurrency, with the city of Zug as the heart of “Crypto Valley.”13

Industry participants have responded to the events of the past few years with their own forms of self-regulation. The now common practice of issuing a white paper is an early example. More recently, many ICOs are consulting with law ﬁrms to avoid potential legal issues from regulators. For example, ICOs now voluntarily comply with Know Your Customer (KYC) and Anti-Money Laundering (AML) regulations by gathering personal info from prospective buyers. These types of positive changes reduce uncertainty about future regulatory actions.

3 Theory and Testable Implications

In this section, we develop a simple model to illustrate the relationship between investor supply and how successfully a company fund raises in an ICO. The model is similar to Leland and Pyle (1977) and Demarzo and Duffie (1999) with the key difference being that, unlike equity, tokens do not entitle the holders to profits from the platform in the form if

10https://www.sec.gov/litigation/investreport/34-81207.pdf 11The SEC has only brought enforcement actions against cases involving fraud despite statements from regulators that imply almost all tokens are securities. In a December 11, 2017 “Statement on Cryptocurrencies and Initial Coin Offerings”, SEC Chairman John Clayton wrote “By and large, the structures of initial coin offerings that I have seen promoted involve the offer and sale of securities and directly implicate the securities registration requirements and other investor protection provisions of our federal securities laws.” (https://www.sec.gov/news/public-statement/statement-clayton-2017-12-11). See Token Alliance (2018) for more detailed description of the US and international regulatory environment. 12See Monetary Authority of Singapore, “A Guide to Digital Token Offerings” (November 14, 2017). 13https://www.wsj.com/articles/ switzerland-wants-to-be-the-world-capital-of-cryptocurrency-1524942058

Electronic copy available at: https://ssrn.com/abstract=3286835 dividends. Instead, token value is determined by the success of the platform being developed as they can be exchanged for the services being sold on the platform, or sold to someone on an exchange who wishes to purchase a service on the platform. Through the model, we formalize the key empirical relationships we later test. In the model equilibrium, high-quality entrepreneurs retain a higher fraction of tokens than low-quality entrepreneurs to signal their type to investors. As the market becomes noisier, the quality of ex-ante information available to investors deteroriates. In such an environment, a larger number of high-type ﬁrms, facing uncertainty from investors about their type, need to retain tokens to signal their quality causing the relationship between retention and fundraising to become stronger.

3.1 The Model

The model has three dates, t = 0, 1, 2 and two types of agents - penniless, entrepreneurs and deep-pocketed, risk-neutral investors. Entrepreneurs can undertake a project at t = 1 to build a platform using blockchain technology. There are two types of entrepreneurs - a proportion q are high-type entrepreneurs (H) and a proportion 1 − q are low-type entrepreneurs (L). A project undertaken by a high-type entrepreneur succeeds with probability πH . A low-type entrepreneur succeeds with probability πL < πH . If the project is successful it delivers a payoﬀ of R. Otherwise, the project returns 0.

At t = 1, the entrepreneur can undertake an ICO to raise money for investment. Entrepreneurs can monetize their projects by either selling tokens today or holding on to the tokens and getting the payoﬀ R. Entrepreneurs choose the fraction of tokens they wish to retain, s, and sell the remaining share of tokens to investors. Both entrepreneurs and investors can invest in an outside option that gives them a gross risk-free return of Rf . For simplicity, we normalize Rf to 1.

Entrepreneurs maximize the following utility function:

u(WT , s) = E[WT ] − γC(s),

where WT is their period-2 wealth, γ > 0 can be interpreted as a coeﬃcient of risk aversion and C(s) is their cost of holding tokens. The cost function C has the following standard properties: C 0 (s) > 0, C 00 (s) > 0, C(0) = 0. We think of the holding costs C (·) as a reduced form way of capturing the cost to entrepreneurs of holding a large portion of their wealth in

Electronic copy available at: https://ssrn.com/abstract=3286835 a single, risky asset and not being able to diversify (see e.g. Leland and Pyle (1977)). This cost implies that entrepreneurs are eﬀectively risk-averse.

At t = 0, before entrepreneurs issue tokens, there is a public signal which perfectly reveals the entrepreneur’s type with probability α. With probability 1−α the signal is uninformative. The informativeness of the signal is the key parameter we vary when we think about markets being noisier. We can interpret an increase in α as a decrease in noise in the market. For example, when a small number of ICOs are taking place, investors have more time to read up on an ICO, have time to study their whitepaper, business plan, prototype, talk to other investors etc. In the model, such a time is proxied by a high α. When many ICOs are happening simultaneously it is harder for investors to undertake such detailed due diligence - we interpret this as a low α.

t = 0 t = 1 t = 2

With probability α Entrepreneur chooses Payoﬀs realize:

entrepreneur i’s type token share si to retain With probability πi is revealed i ∈ {H,L} project is successful

Figure 1: Model Timeline

Entrepreneur’s Problem

At t = 1, there are 4 types of entrepreneurs - a high-type whose type has been revealed, a low-type whose type has been revealed, a high-type whose type has not been revealed and a low-type whose type has not been revealed. Let i = {H,L} represent a high-type versus a low-type entrepreneur and let j = {R,U} represent whether an entrepreneur’s type is revealed or unknown. At t = 1, an entrepreneur solves the following problem,

j max (1 − s)P (si ) + sπiR − γC(s). s∈[0,1]

Equilibrium Deﬁnition: A perfect Bayesian equilibrium in pure strategies of the model is given by,

Electronic copy available at: https://ssrn.com/abstract=3286835 R∗ R∗ 1. The share of tokens, sH and sL , retained by the high-type and low-type entrepreneur respectively whose types have been revealed by the public signal.

U∗ U∗ 2. The share of tokens, sH and sL , retained by the high-type and low-type entrepreneur respectively whose types have not been revealed by the public signal.

3. Investor beliefs, µ(θ|s), that an entrepreneur is of type θ = {H,L}, if he chooses to retain share s of tokens.

4. Prices investors pay per share of tokens sold to them s.t. their participation constraint is satisﬁed.

P (s) = µ(H|s)πH R + (1 − µ(H|s))πLR.

3.2 First Best

In the absence of information asymmetries, when participating in an ICO investors know if they are buying tokens from a high- or low-type entrepreneur. Investors will therefore set 14 Pi = πiR.

The entrepreneur’s problem therefore becomes,

max (1 − s)πiR + sπiR − γC(s). s∈[0,1]

We can rewrite this as

max πiR − γC(s). s∈[0,1]

This is always maximized when s = 0. In the ﬁrst best with no asymmetric information, entrepreneurs therefore always prefer to sell all of their tokens to investors. Since entrepreneurs are risk-averse while investors are eﬀectively risk-neutral, entrepreneurs prefer to sell all the risk associated with the tokens to investors and monetize all tokens at t = 1. Formally, we can establish the following proposition,

14Since there is a large mass of deep, pocketed investors, competition between them sets token prices s.t. investors earn a return of 1, equal to their outside option.

Electronic copy available at: https://ssrn.com/abstract=3286835 Proposition 1 [First Best] In the ﬁrst-best, with no asymmetric information, both high- FB FB and low- type entrepreneurs choose to sell all their tokens to investors, sH = sL = 0.

Since in the ﬁrst best, both types of entrepreneurs want to sell all their tokens to investors, we should not see diﬀerential retention between high- and low-type entrepreneurs. Therefore, in the absence of asymmetric information about entrepreneur quality, we should see no covariance between the amount of tokens retained by an entrepreneur and the quality of an ICO.

We deﬁne ρv ≡ Cov(pR, s) as the model-implied covariance between the value of the ICO-ing ﬁrm to an investor and retention.

Corollary 1 [Retention and Quality in First Best] In the absence of asymmetric information about entrepreneur quality, there is zero covariance between the value of an ICO and the

fraction of tokens retained, i.e., ρv = 0.

3.3 Equilibrium with Asymmetric Information

In the equilibrium with asymmetric information, with probability α an entrepreneur’s type will be revealed. In this case, the entrepreneur’s and investors’ problems are the same as in the full information case. Both high- and low-type entrepreneurs whose type is revealed will R∗ R∗ select sH = sH = 0 and retain no tokens. Investors will buy all the tokens at a price of πiR where i ∈ {H,L}.

If an entrepreneur’s type is not revealed, then there exists a separating equilibrium in U U which the high type chooses an sH > 0 to signal his type while the low type choose sL = 0. The following investor beliefs can support such an equilibrium,

1 if s ≥ sU  H µ(H|s) =  U 0 if s < sH

Since this is a fully separating equilibrium, the prices investors are willing to pay per fraction of tokens sold to them conditional on retention revealing the entrepreneur-type are the same as those in the full information case. In a signaling equilibrium in which the high-

Electronic copy available at: https://ssrn.com/abstract=3286835 type separates from the low-type, s much be high enough that the low-type doesn’t want to copy the high-type,

U U U (1 − sH )πH R + sH πLR − γC(sH ) ≤ πLR. | {z } | {z } | {z } Expected t=2 Wealth Cost of Retention Expected t=2 Wealth

Additionally, the high-type must not want to deviate to selling a larger fraction of tokens to investors,

U U U (1 − sH )πH R + sH πH R − γC(sH ) > πLR. | {z } | {z } | {z } Expected t=2 Wealth Cost of Retention Expected t=2 Wealth

Under the Intuitive Criterion, we can show that the least-cost separating equilibrium is the unique equilibrium of the model. In this equilibrium, the high type chooses to retain a U∗ strictly positive amount of tokens , sH > 0, of tokens. This fraction is set such that the low-type entrepreneur’s incentive compatibility constraint is just satisﬁed,

U∗ U∗ U∗ (1 − sH )πH R + sH πLR − γC(sH ) = πLR | {z } | {z } |{z} Expected t=2 Wealth Cost of Retention Expected t=2 Wealth Proposition 2 [Equilibrium with Asymmetric Information] Under the Intuitive Criterion, an equilibrium in which a high-type entrepreneur who has not had his type revealed at t = 0, U∗ signals his quality by retaining a fraction, sH > 0, of tokens and a low-type entrepreneur U∗ sells all tokens, sL = 0, exists and is unique.

In the presence of asymmetric information, high-type entrepreneurs retain a positive share of tokens to signal their type to investors. This will therefore cause there to be a positive relationship between token retention and fundraising success in the ICO market. A number of factors will aﬀect the fraction of tokens the high-type chooses to retain.

Corollary 2 [Equilibrium Retention] The fraction of tokens a high-type entreprenur retains, U∗ sH , increases when πH increases, R increases, πL decreases and γ decreases.

When πH increases, R increases and πL decreases, the gain to the low-type from pretending to be the high-type increases and the high-type needs to retain a greater fraction of tokens

Electronic copy available at: https://ssrn.com/abstract=3286835 U∗ to signal quality. Therefore sH increases. When γ decreases, the cost of retention decreases. U∗ This makes it cheaper for the low-type to copy the high-type and and therefore sH increases.

Corollary 3 [Retention and Quality in Equilibrium with Asymmetric Information] In the presence of asymmetric information about entrepreneur quality, there is a positive covariance between the value of an ICO and the fraction of tokens retained, i.e., ρv > 0.

Retention in Noisier Markets: In the model equilibrium, high-type entrepreneurs who have their type revealed by the public signal sell all their tokens to investors. High-type entrepreneurs who do not have type revealed retain a fraction of tokens to signal their type. We deﬁne Ret∗ as the total fraction of tokens retained in equilibrium. This is given by,

∗ U∗ Ret = q(1 − α)sH q(1 − α) is the measure of high-type investors who do not have their type revealed and they U∗ each retain a fraction sH of tokens. A decrease in α can be interpreted as an increase in noise in the market. If only a few ICOs are happening in a month, investors have more to study a company’s whitepaper, business plan, examine their prototype, read analyst reports, talk to other investors and so on. This can be thought of as a time of high α when a large fraction of ﬁrms have their type revealed to investors. On the other hand, at a time when many ICOs are taking place it is harder for investors to undertake such detailed due diligence. We interpret this as a low α. As α decreases, a greater fraction of high-type entrepreneurs retain tokens.

Then we can establish the following proposition,

Proposition 3 [Retention and Market Noise] In the unique signaling equilibrium, ICO- value covaries positively with the fraction of tokens retained, ρv > 0. As α decreases and the quality of public information deteriorates, the magnitude of this covariance increases, i.e.,

∂ρ v < 0. ∂α

Electronic copy available at: https://ssrn.com/abstract=3286835 4 Data

We collect data from several of ICO tracking websites that provide information to potential ICO investors and the general public. These websites include ICO Data (https:// www.icodata.io/), Token Data (https://www.tokendata.io/), CoinMarketCap (https: //coinmarketcap.com/), Cryptoslate (https://cryptoslate.com), ICO Bench (https: //icobench.com), ICO Drops (https://icodrops.com), ICO Rating Agency (https: //icorating.com), and ICO Checks (https://icocheck.io). To merge ICO characteristics across diﬀerent websites, we employ three identiﬁers: ticker symbol, token name, and website URL. The combined database includes 7,450 distinct ICOs between January 1, 2016 through December 31, 2018, including only completed ICOs.15 We augment this data by hand- collecting data from ICO whitepapers, websites, and Apple’s App Store. To the best of our knowledge, this is among the most comprehensive ICO database to be used for academic research.16

We collect data on both ex-ante characteristics of ICOs that are known to investors at the time of an ICO (ICO Start Date, ICO End Date, ICO Sale Price, ICO Hard Cap, ICO Soft Cap, ICO Token Retention, etc) and the performance of the company following the ICO (ICO Funds Raised, Market Cap, Price, Volume etc.). The full list of collected variables along with their deﬁnitions is provided in Table 1.

Table 2 reports summary statistics for all ICOs that have been competed by December 31, 2018. When announcing a project, token issuers typically produce a whitepaper that provides project details and key terms of an ICO. Not surprisingly, we find that half of ICOs in our sample indeed has such whitepaper. A growing number of ICOs now require their contributors to pass Know Your Customer (KYC) verification before purchasing tokens. This is the procedure of identity verification in which each buyer has to provide his credentials (e.g. passport or driver’s license number) in order to participate in an ICO. KYC ensures transparency of transactions which helps to guarantee the safety of investors’ assets and prevent scammers from participating in ICOs. On the other hand, its contradicts one of the

15We also collect data for 183 planned 481 ongoing ICOs. However, they are excluded from our sample because we focus on performance measures of ICOs. 16At the largest end, Lyandres et al. (2018) gather data for 4,379 completed or ongoing ICOs betweeen 2013 and 2018, Deng et al. (2018) use a sample of 3,573 completed ICOs between August 2015 and July 2018, and Benedetti and Kostovetsky (2018) use a sample of 2,390 ICOs that were completed on or before April 30, 2018. The remaining empirical papers referenced in our literature discussion use samples that include anywhere from 64 to 659 ICOs.

Electronic copy available at: https://ssrn.com/abstract=3286835 main principles of the crypto world – anonymity – and as such can be viewed as a barrier to enter an ICO. Given its recent adoption into the ICO market, only 27% of ICOs in our sample implement KYC. An even smaller fraction of token issuers (21%) request investors to pre-register on a “whitelist” before their token sale launch. Only “whitelisted” customers are able to purchase tokens during sale. Similar to KYC, whitelists create barriers to participate in an ICO.

One key measure of ICO success in fundraising is the amount of capital that token issuers collect for future product development. Table 2 and Figure 2 demonstrate that the amount of funds raised by ICOs is highly skewed. While the mean and median value of funds raised is $16.6 million and $5.1 million, respectively, the largest ICO in the history EOS, developer of infrastructure for decentralized apps, has raised $4,234.3 million after year-long crowdsale. Other large ICOs include Telegram Open Network, Petro, and Dragon. A related measure of ICO success is whether the issuer raises any funds at all. One third of all initial coin oﬀerings in our sample have raised a strictly positive amount of funding as captured by Successful Dummy. This is an indicator that equals 1 if ICO Funds Raised > 0 and zero otherwise.

The ICO market has grown rapidly since its inception. The number of ICOs increased from 52 in 2016 to 1,527 in 2017 and then 3,870 in 2018. To date, the ICO market has raised over $33.4 billion in total. The average ICO in our dataset raised $16.6 million and the largest ICO in history, EOS, raised over $4.2 billion (see Figure 3). For reference, the 2017 global ﬁgures for IPOs and VC investments are $189 billion and $155 billion, respectively (EY, 2018; KPMG, 2018). Nonetheless, this rapid growth in ICOs as a form of ﬁnancing for companies make them an interesting market to study in and of themselves.

In the recent period, the average size as measured by the amount of capital raised has nearly doubled indicating a higher interest of investors in the ICO market (see Panel (b) of Figure 3). We also observe a steep increase in the aggregate amount of capital that has been attracted by token issuers at the end of 2017. Panel (a) of Figure 3 shows that the ICO market has reached its monthly peak of $6,816 million in June 2018. Most of the funds raised in this month is attributed to EOS that completed its ICO on June 26, 2018 and raised $4,234.3 million. One unique feature of the EOS ICO is that it occurred over multiple rounds, raising $150 million in June 2017, $180 million in July 2017, $100 million in November 2017, $420 million in December 2017, $750 million in January 2018, and $500 million in February

Electronic copy available at: https://ssrn.com/abstract=3286835 2018, which is not reﬂected in the time-series of total funds raised17. Another peak coincides with the completion date of Telegram Open Network raising $1,700 million. Excluding the 10 largest ICOs as measured with the amount of capital raised demonstrates that the ICO market has stabilized at around $2,000 million throughout 2018.

In our analysis, we measure issuers’ signal of project quality with the percentage of the tokens that are retained by the ICO team during the fund raising campaign, which we denote as Token Retention. As discussed earlier, this measures the fraction of all tokens that are not sold to investors at the time of the ICOs. It is important to note that token retention is declared before the ICO begins and it is often written into the token smart contract. Figure 4 demonstrates that there is a lot of heterogeneity among token issuers when choosing what percentage of tokens to sell to outside investors. Token Retention ranges from 0% to 100%. An average ICO retains 44% of all issued tokens (see Table 2).

5 Empirical Findings

In this section, we test our main hypothesis that token issuers signal the quality of their project to potential investors by retaining a higher fraction of tokens during the ICO. This proposed signaling mechanism is consistent with the model in Section 3. It is also consistent with rational investors that respond to issuers’ signal in making their investment decisions.

Studying retention in the ICO market allows us to overcome challenges in identifying an effect of retention that are significant in other empirical settings. First, the presence of regulators, intermediaries, and legal protections reduce the problem of asymmetric information in many markets. ICO markets are liquid, opaque, and largely unregulated. The absence of legal repercussions combined with the unverifiability in ICO market makes the market prone to a high degree of asymmetric information and token retention is one of the few verifiable actions a company can take to convey information to investors. This gives us a unique setting to study signaling via retention, which is very close to theoretical settings.

Second, in most markets there is a lot of soft-information exchange through in-person interactions between buyers and sellers. For example, consider the interaction between a

17Importantly, since we conduct all empirical tests at the ICO level, not breaking down the funds raised by rounds should not aﬀect our ﬁndings.

Electronic copy available at: https://ssrn.com/abstract=3286835 venture capitalist and a potential entrepreneur. The venture capitalist may be inﬂuenced by a variety of “soft” variables such as a person’s presentation skills, perceived intelligence, demeanor, manner of speaking etc. which are impossible to control for in an empirical design. If these variables are related to skill, then an entrepreneur may value the business of an entrepreneur who has such soft skills highly and ask for a smaller share of the ﬁrm for a given amount of funds the entrepreneur is trying to raise. This would create a correlation between retention and project quality, that would not be due to a signaling channel. ICO fund raising takes place online with little to no interaction between the team developing the platform and purchasers of the token. We therefore have less confounding soft-information exchange that we need to worry about.

Third, it is hard to control for all of the hard information that could affect a buyer’s decision besides retention. These include, but are not limited to, information on the entrepreneur’s schooling, resume, previous experiences with start-ups, etc. In most markets, there are reputation costs and legal consequences for misrepresentation of such information allowing investors to rationally condition their decisions on such hard information. Accounting and controlling for all such information empirically is not feasible. In ICO markets, such information is unverifiable, not possible to commit to, and misrepresentation carries no legal repercussions. The only promises that can be committed to are ones that are codeable into smart contracts. Smart contracts therefore give us a limited set of observable promises that a seller can commit to, greatly reducing the number of omitted variables we need to worry about. Since these promises and smart contracts are public, we can control for all necessary variables. Moreover, traditional financing contracts which rely on legal enforcability can be quite complicated with multiple clauses and terms. Kaplan and Strömberg (2003) find that VC contracts often have anti-dilution provisions such as ratchet protections. Such clauses make inference between the share retained by the entrepreneur and the price investors are willing to pay for shares tenuous. The current contracts in ICO markets are relatively simple, making the interpretation of our results straightforward.

Electronic copy available at: https://ssrn.com/abstract=3286835 5.1 Signaling with Token Retention

As a ﬁrst step, we test whether ICOs that retain a larger fraction of tokens tend to raise a larger amount of funds using the OLS. The speciﬁcation is as follows, for ICO j:

ln (ICO F unds Raisedj) = α + β · ICO T oken Retentionj + γXj + j, (1) where the dependent variable is the natural logarithm of the total funds raised during the ICO.18 The coefficient estimate on ICO Token Retention captures the relationship between signal quality and ICO success. We control for a number of other observed ICO characteristics Xj that might affect the ICO success. In particular, we control for whether an ICO requires its investors to pass KYC verification and register on its whitelist, whether the issuer posts its code to an online GitHub repository, and whether an ICO has conducted a pre-sale. Additionally, we control for the GDP per capita and Corruption Perception (CP) Index of a country where an ICO team is located. The CP Index, developed by an international non-governmental organization Transparency International, is a score on a scale from 0 (highly corrupt) to 100 (very clean) computed based on surveys of experts and business people.19

In Table 3, we report the estimates of the above specified regression. In Columns (1)-(5), we include only ICOs that have been completed (i.e., have the end date of ICO) before December 31, 2018. We find that a 1 percentage point increase in the token retention by the issuer, or equivalently a 1 percentage point decrease in the investor supply, leads to an approximately 1.1 − 1.6% increase in the total raised funds. This result is consistent with the signaling mechanism described in our model. Importantly, this relationship is both economically and statistically significant.

As mentioned before, a number of ICOs achieve their fundraising goal before the ICO end date. As such, we also run the regressions including the ICOs that are ongoing at

18The total funds raised are denominated in US dollars. Some websites report raised funds in alternative currencies, e.g. Ethereum or Bitcoin. We convert these amounts to USD using the exchange rate prevailing on the last day of the ICO. This may introduce bias in our measure of the funds raised as most successful ICOs manage to achieve their funding goal way before the announced end date of ICO. This bias might be also attenuated due to the fact that cryptocurrency prices are highly volatile. For robustness, we run all the regressions restricting our attention only to ICOs that report the amount of raised funds in USD. All results continue to hold. 19See the Transparency International website for more details https://www.transparency.org/ research/cpi/overview.

Electronic copy available at: https://ssrn.com/abstract=3286835 least for a month as of December 31, 2018 but haven’t been completed yet. The results are reported in Columns (6) -(10) of Table 3. Including these active ICOs allows us to increase our sample size by 30 additional observations, and it doesn’t aﬀect, neither qualitatively nor quantitatively, the relationship between the signal quality and the ICO fundraising success.

We also find that ICOs with KYC or Whitelist raise 0.6 − 0.8% more funds on average. These figures would be 1.2 − 1.5% if an ICO implements both measures. Both KYC and Whitelist can be interpreted as barriers for investors to participate in ICOs. When token issuers choose to require their contributors to pass identify verification and/or register for a token sale in advance, they limit their potential pool of investors and as such reduce their chances to achieve funding goals. At the same time, by creating these additional constraints entrepreneurs attract more credible investors and as a result are more successful in their fundraising campaign. Additionally, we see that ICOs that post their code to the GitHub repository are more successful in their fundraising efforts, as they in general advanced more with the development of their platform. We also find that ICO teams located in countries with higher CP Index, i.e. lower level of corruption, are more successful.

Token issuers very often set their maximum fundraising goal in the details of their initial announcement. This ﬁgure is known as a Hard Cap. Our initial results could be driven by few large ICOs (as measured by the funds raised) that have a high token retention but, at the same time, fail to achieve their fundraising goal. For token sales that feature a hard cap, a better success metric would be the amount of funds raised in ICO as a fraction of the ICO hard cap, which we call as Fundraising Success. This metric captures how close token issuers are toward achieving their fundraising goal regardless of the ICO size. In Table 4, we report the estimates of the following regression:

ICO F undraising Successj = α + β · ICO T oken Retentionj + γXj + j, (2) where the dependent variable is the total funds raised as a fraction of the hard cap. We again find a strong positive relationship between the token retention and ICO fundraising success, implying that indeed investors respond to a signal of token issuers and invest more in projects of higher quality. In line with our hypothesis, when issuers retain a larger fraction of tokens, investors learn that they are of high quality, and consequently these ICOs fund raise more successfully. Specifically, we find that a 1 percentage points increase in the token retention results in a 0.1-0.3 percentage points increase in the fundraising success. Again,

Electronic copy available at: https://ssrn.com/abstract=3286835 this estimate is both economically and statistically signiﬁcant.

A set of ICOs choose not to report the amount of funds raised at all, if they failed to achieve their fundraising goal. This creates a concern that our coefficient estimates may be subject to a selection bias. Our regressions might include only ICOs that have been relatively more successful and as such chose to report the amount they raised. Figure 5 shows that around a quarter of ICOs in our sample have achieved their financing goal. At the same time, our sample includes around 30% of ICOs that raised close to $0. We further address this concern by setting ICO Fundraising Success to zero if it is not reported. The results are consistent with our previous findings as shown in Columns (6)-(10) of Table 4. In these specifications, we only focus on completed ICOs as active ICOs might still report the funds raised upon their completion. There are around 1,100 ICOs that do report the amounts of tokens retained in ICO but have missing funds raised.

As discussed previously, there is very little soft information exchange possible in ICO markets due to their opaque nature and the unveriﬁability of any information not codeable in the smart contract. However, there could be soft information exchange through some information conveyed online that is costly to produce. For example, having a more-technical, better designed and well-written whitepaper may convey information about the skills of the ICO team. As such, investors may value the tokens of teams with better whitepapers more. For a ﬁxed fundraising goal, this would create a positive correlation between token retention and fundraising. In such a case, we would observe a positive correlation between ICO’s fundraising success and token retention that is not due to signaling through retention.

To address this challenge, we estimate regression (2) using an instrumental variables (IV) strategy, where use the Financial Development (FD) Index of the country where the ICO team is located as an instrument for token retention. The FD index was developed by the International Monetary fund, and summarizes how developed financial institutions and financial markets of a given country are in terms of their size, liquidity, access, and efficiency. More details on the methodology of how the index was constructed are provided in Appendix B. By holding on to tokens, entrepreneurs add an illiquid asset to their portfolio, thereby increasing its risk exposure. This is particularly costly if such risk cannot be diversified away through the financial markets. To the extent that financial development promotes access to financial markets, ICO teams located in countries with limited risk-sharing opportunities, as captured by the lower FD Index, have a relatively higher cost of retention. Consequently, such

Electronic copy available at: https://ssrn.com/abstract=3286835 teams tend to sell a larger share of their project to mitigate their risk exposure. Plausibly, the level of ﬁnancial development of a country where ICO team is located is exogenous to fundraising success, except its eﬀect through the cost of signaling, as all token sales happen online attracting investors worldwide.

For our IV to hold, investors should not respond to the reported team location country. If investors accounted for the diﬀerential cost of retention across countries when determining how they value a token, the relationship between token retention and quality would change with the country of location, invalidating the IV. We think the assumption that investors do not respond to team location country is reasonable as it is unable to be written into a smart contract. Hence it is unveriﬁable and arguably costless to misreport (no legal repercussions). Any equilibrium in which investors value tokens from one country more than another is not sustainable as all ICOs have an incentive to falsely report their country as the country whose tokens investors value most highly.

Importantly, there is a substantial heterogeneity in the location of ICO teams, as shown in Figure 6. Though token issuers are mostly concentrated in United States, Singapore, United Kingdom, Russian Federation, Estonia, and Switzerland, we still observe ICO activity across wide range of countries. The full list of locations is provided in Appendix C, Table C.1. Figure 7 further demonstrates that there are no systematic shifts in team locations over time.

To extract exogenous variation in token retention, we estimate the following ﬁrst stage of the IV regression:

FS FS ICO T oken Retentionj = α + β · F inancial Development Indexj + δXj + εj. (3)

This step is important to demonstrate the relevance of our instrument. As shown in Table 5, we find a positive and significant estimate of βFS, indicating that indeed ICO teams located in more financially developed countries have a lower cost of retention and as such tend to hold on to more tokens. We also include ICO specific characteristics as controls, as well as time fixed effects. Across all first-stage regressions, the Cragg-Donald F-statistic ranges between 11 and 23, further confirming that the IV is not weak.

Electronic copy available at: https://ssrn.com/abstract=3286835 In the second-stage regression, given by

IV IV ICO F undraising Successj = α + β · ICO T oken\ Retentionj + γXj + j, (4) we exploit the variation in token retention coming from entrepreneurs’ access to financial markets and estimate the effect of token retention on ICO fundraising success. The associated identifying assumption would be that the ICO team location and the corresponding level of financial development is independent of the project quality and as such has no effect on the team’s fundraising success, beyond its effect through the cost of signaling.

One possible concern is that ICO investors are subject to the home bias, i.e. ICO investors from the United States are more inclined to allocate their funds to ICOs registered in the United States rather than in Russian Federation. Conditional on the presence of home bias, one could argue that ﬁnancially developed countries are also the countries with wealthier investors and, consequently, teams located in countries with high FD index would more easily achieve their fundraising goal. Though the presence of home bias in the ICO markets is questionable given the global nature of ICOs, to further strengthen the validity of our exclusion restriction we directly control for the level of the country’s GDP per capita. Column (5) of Table 5 conﬁrms that indeed GDP level has no explanatory power for token retention beyond FD index.

Another potential concern would be that less ﬁnancially developed countries are also the countries with higher level of corruption and lower level of trust. Presumably, ICOs located in these countries would have harder time to attract investors and meet their fundraising goals. At the same, one could argue that investors in these location face a higher degree of asymmetric information, calling for a stronger signal. Admittedly, this would mean that entrepreneurs located in countries with lower FD index retain on average more tokens, which is not the case in the data. To further address such concerns, we include the CP Index as a control in our regressions. Similarly to GDP, corruption index has no explanatory power for token retention beyond FD index (see Table 5).

Table 6 presents the results of the second stage of the IV regressions. The effect of token retention on ICO fundraising success is positive and economically large. Specifically, a 1 percentage points increase in the token retention results in a 1.1-2.0 percentage points increase in the fundraising success. The IV estimate is about 10 times larger than the OLS estimate. To understand this result, note that the IV captures a specific local treatment

Electronic copy available at: https://ssrn.com/abstract=3286835 effect (LATE). In particular, it captures the signaling effects stemming from the variation in token retention across countries with different level of financial development rather than the variation in token retention across all ICOs regardless of their team location.

5.2 Signal Quality during Periods of Increased Market Noise

In this section, we test whether the relationship between token retention and the quality of the ICO grows stronger during noisier markets as predicted by the theory. We proxy for noise in the market by increased number of ICOs, as well as increased amount of information provided by token issuers in the whitepapers. The intuition for this measure is that investors have less time to perform detailed due diligence per ICO when many ICOs are taking place and/or whitepapers become lengthier. They therefore rely more strongly on signals such as token retention. We propose using the number of ongoing ICOs at a given time scaled by the average number of words in the whitepapers at that time as a measure of the noise in the market. Speciﬁcally, for each ICO j, we count the number of ICOs that are active within the 15 days range from the launch date of ICO j and multiply it by the average number of words in the whitepapers of ICOs active at that time.

We start by analyzing whether signaling in the ICO market has a larger eﬀect on fundraising success during periods of increased noise when we expect there to be an increased degree of asymmetric information in the market. We test whether the signaling mechanism is stronger during periods when there are a large number of active ICOs and/or amount of information provided in the whitepapers, using the following speciﬁcation:

ICO F undraising Successj =α + β1 · ICO T oken Retentionj + β2 · Scaled # of ICOsj (5)

+ β3 · ICO T oken Retentionj × Scaled # of ICOsj + γXj + j,

where all variables, besides dummies, are normalized to capture the average effect of token retention on the total funds raised as a fraction of hard cap. We are interested in the coefficient estimate on the interaction term that captures the differential affect of token retention on ICO success when the market noise is above as compared to below its mean level. Regressions are estimated both using the OLS and IV, where we instrument for T oken Retention × Scaled # of ICOs with F inancial Development Index × Scaled # of ICOs. Admittedly, for this analysis we restrict our sample to ICOs with the beginning date between September 2017 and November 2018 inclusive. This allows us

Electronic copy available at: https://ssrn.com/abstract=3286835 to focus on the months when there is a suﬃcient number of ICOs launching in diﬀerent locations. This particular restriction ensures that there is at least 20 distinct countries in a given month (see Figure C.2).

Table 7 demonstrates that there is a positive economically meaningful relationship between token retention and ICO fundraising success during periods of increased market noise. In this specification, we normalize both the dependent and independent variables. We do so in order to more easily interpret the estimated coefficients, which now reflect the effect of a given variable for the average ICO. When the scaled number of ICOs is at its mean level, a 1 standard deviation increase in token retention leads to approximately 0.9-1.0 standard deviation increase in the fundraising success. In line with our model predictions, we also find that signaling project’s quality is more important during high ICO activity when market noise is high as it becomes harder for investors to separate out worthy investments. In particular, for each standard deviation increase in the scaled number of ICOs, a 1 standard deviation increase in token retention leads to an additional 0.5-0.6 standard deviation increase in fundraising success. Additionally, we consider alternative measures of the noise in the market. Specifically, we alter the window when we count the number of ICOs happening around the launch date of ICO j, as well as apply alternative scaling – by the the average and median number of pages in the whitepapers. The results are very similar. For brevity, we do not report those results here.

5.3 Ex-Post Measures of ICO Performance

Reaching the fundraising goal during the ICO does not necessary result in successful development of the product. As such the amount of funds raised is only a proxy for the project’s quality and it important to explore alternative measures of ICO performance. In this section, we consider whether ex-post measures of ICO performance.

First, we measure performance by whether a token lists on an exchange after an ICO. Crypto exchanges engage in due diligence prior to accepting a token for listing to protect their reputation both with customers and regulators. Allowing fraud or scam tokens on a trading platform erodes customer trust. Regulatory compliance is important for staying in business. In March 2018, the SEC issued a statement noting that any platform “providing a mechanism for trading assets that meet the deﬁnition of a ‘security’ ... must register with the

Electronic copy available at: https://ssrn.com/abstract=3286835 SEC as a national securities exchange.”20 Given the regulatory uncertainty about whether tokens are “securities” (see section 2), many exchanges that feature tokens have chosen to register with the SEC (e.g., Coinbase). Part of being a “national securities exchange” is complying with anti-money laundering (“AML”) regulations and associated requirements, which means understanding the parties with whom they interact (Token Alliance, 2018).

Recent literature points out that ICOs that result in successful product development are more likely to have their tokens traded on a reputable exchange. In this sense, an indicator of whether ICO’s tokens are listed can serve as an alternative measure of the project’s quality. Within the industry, it is well known that exchanges screen tokens for quality. For example, the Coinist, an ICO advisor and data provider, writes that “in general, a strong team with a better product that is upfront with information should have a relatively simple time getting listed.”21

We further test our signaling hypothesis using the following speciﬁcation:

ICO Listed Dummyj =α + β1 · ICO T oken Retention j (6)

+ β2 · ICO Residual F\ undraising Successj + γXj + j,

whether the dependent variable Listed Dummy is an indicator that equals to 1 if ICO tokens are traded on an exchange. The dependent variable ICO Residual F\ undraising Success is the residual term from the following regression:

˜ ICO F undraising Successj =α ˜ + β · ICO T oken Retentionj +γX ˜ j + εj. (7)

It aims to capture the portion of the Fundraising Success that can not be explained with the token retention and other ICO characteristics. The analysis is performed using our IV strategy but for comparison we also report the estimates from OLS regressions.

In line with our previous findings, we find that ICO investors indeed rely on token retention as a signal of the project’s quality. As reported in Table 8, a 1 percentage points increase in token retention leads to a 1.3-2.1% increase in the probability of tokens being listed on the exchange. We also find that ICOs that have achieved their hard cap are much more likely to have their tokens listed on an exchange. Specifically, a 1 percentage points increase in

20https://www.sec.gov/news/public-statement/ enforcement-tm-statement-potentially-unlawful-online-platforms-trading 21https://www.coinist.io/how-to-get-your-digital-token-listed-on-an-exchange/

Electronic copy available at: https://ssrn.com/abstract=3286835 the fundraising success results in a 0.5-0.8% higher listing probability. This is beyond the signaling channel though token retention.

Additionally, we use three alternative measures of ex-post ICO success: (1) whether the issuer continues to have an active and working website, (2) whether the issuer has produced a product that is live to users, and (3) whether the issuer has an application for download in Apple’s App Store. In line with our hypothesis, we ﬁnd that a 1 percentage points increase in token retention leads to a 0.20-0.28% increase in the probability of the company still having a working website. We additionally ﬁnd that a 1 percentage points increase in token retention leads to a 0.15-0.17% increase in the probability of the company having a live product and between a 0.12-0.14% probability of having an application available on the Apple store. All results are reported in Table 9. Hence, high-quality ICOs that have been successful in signaling their quality through token retention will more likely have their tokens traded and potentially develop their product.

6 Conclusions

In this paper, we provide evidence of signaling in ICO markets. We first establish that higher token retention is associated with fundraising success and ex-post measures of quality. We then implement an instrumental variables strategy to provide further causal evidence for the effect of token retention. When the market for ICOs is noisier, we expect the degree of asymmetric information to be higher. In line with this conjecture, we find that the correlation between the fraction of tokens that investors retain and ICO quality increases. In other words, the strength of the retention signal becomes stronger. All of these results confirm the hypotheses formalized in our stylized model of the ICO market.

Our paper contributes to two distinct areas of the literature. First, we contribute to the literature that empirically analyzes signaling through retention in various settings. Ours is the first to focus on the ICO market, which is unique in its lack of regulation and intermediation. In this setting with a relatively high degree of asymmetric information, we not only find positive evidence for signaling through retention but we also show how the strength of this signal depends on market environment. Second, our paper contributes to the general study of the ICO market, which has grown rapidly and is emerging as an important source of early-stage financing. In particular, we show how investors and entrepreneurs are

Electronic copy available at: https://ssrn.com/abstract=3286835 rationally responding to asymmetric information frictions inherent in the ICO market. We also provide causal evidence for the role of token retention as a signal of project quality.

Electronic copy available at: https://ssrn.com/abstract=3286835 References

Adelino, M., K. Gerardi, and B. Hartman-Glaser (2019). Are lemons sold ﬁrst? dynamic signaling in the mortgage market. Journal of Financial Economics 132 (1), 1 – 25.

Adhami, S., G. Giudici, and S. Martinazzi (2018). Why do businesses go crypto? an empirical analysis of initial coin oﬀerings. Journal of Economics and Business 100, 64 – 75.

Agarwal, S., Y. Chang, and A. Yavas (2012). Adverse selection in mortgage securitization. Journal of Financial Economics 105 (3), 640 – 660.

Ahlers, G. K., D. Cumming, C. G¨unther, and D. Schweizer (2015). Signaling in equity crowdfunding. Entrepreneurship Theory and Practice 39 (4), 955–980.

Amsden, R. and D. Schweizer (2018). Are blockchain crowdsales the new “gold rush”? success determinants of initial coin oﬀerings. Working paper.

Bakos, Y. and H. Halaburda (2019). Funding new ventures with digital tokens: Due diligence and token tradability. Working paper.

Benedetti, H. and L. Kostovetsky (2018). Digital tulips? returns to investors in initial coin oﬀerings. Working paper.

Boreiko, D. and N. Sahdev (2018). To ico or not to ico – empirical analysis of initial coin oﬀerings and token sales. Working paper.

Bourveau, T., E. T. De George, A. Ellahie, and D. Macciocchi (2018). Initial coin oﬀerings: Early evidence on the role of disclosure in the unregulated crypto market. Working paper.

Catalini, C. and J. S. Gans (2018). Initial coin oﬀerings and the value of crypto tokens. Working paper.

Chod, J. and E. Lyandres (2018). A theory of icos: Diversiﬁcation, agency, and information asymmetry. Working paper.

Clarkson, P. M., A. Dontoh, G. Richardson, and S. E. Sefcik (1991). Retained ownership and the valuation of initial public oﬀerings: Canadian evidence. Contemporary Accounting Research 8 (1), 115–131.

Electronic copy available at: https://ssrn.com/abstract=3286835 Cong, L., Y. Li, and N. Wang (2018, September). Tokenomics: Dynamic adoption and valuation. Working Paper 18-46, National Bureau of Economic Research.

Cong, L. W. and Z. He (2018, March). Blockchain disruption and smart contracts. Working Paper 24399, National Bureau of Economic Research.

Demarzo, P. and D. Duﬃe (1999). A liquidity-based model of security design. Econometrica 67 (1), 65–99.

Deng, X., Y.-T. Lee, and Z. Zhong (2018). Decrypting coin winners: Disclosure quality, governance mechanism, and team networks. Working paper.

Downes, D. H. and R. Heinkel (1982). Signaling and the valuation of unseasoned new issues. The Journal of Finance 37 (1), 1–10.

Downing, C., D. Jaﬀee, and N. Wallace (2008, 12). Is the Market for Mortgage-Backed Securities a Market for Lemons? The Review of Financial Studies 22 (7), 2457–2494.

EY (2018). Global ipo trends: Q4 2017 a busy 2017 with more mega-ipos on the horizon. Technical report, EY.

Fisch, C. (2019). Initial coin oﬀerings (icos) to ﬁnance new ventures. Journal of Business Venturing 34 (1), 1–22.

Fuchs, W., B. Green, and D. Papanikolaou (2016). Adverse selection, slow-moving capital, and misallocation. Journal of Financial Economics 120 (2), 286 – 308.

Garmaise, M. J. and T. J. Moskowitz (2003, 08). Confronting Information Asymmetries: Evidence from Real Estate Markets. The Review of Financial Studies 17 (2), 405–437.

Howell, S. T., M. Niessner, and D. Yermack (2018, September). Initial coin oﬀerings: Financing growth with cryptocurrency token sales. Working Paper 564/2018, European Corporate Governance Institute.

Jong, A. d., P. Roosenboom, and T. v. d. Kolk (2018). What determines success in initial coin oﬀerings? Working paper.

Kaplan, S. N. and P. Str¨omberg (2003, 04). Financial Contracting Theory Meets the Real World: An Empirical Analysis of Venture Capital Contracts. The Review of Economic Studies 70 (2), 281–315.

Electronic copy available at: https://ssrn.com/abstract=3286835 KPMG (2018, January). Venture pulse q4 2017. Technical report, KPMG Enterprise.

Krinsky, I. and W. Rotenberg (1989). Signalling and the valuation of unseasoned new issues revisited. Journal of Financial and Quantitative Analysis 24 (2), 257–266.

Leland, H. E. and D. H. Pyle (1977). Informational asymmetries, ﬁnancial structure, and ﬁnancial intermediation. The Journal of Finance 32 (2), 371–387.

Li, J. and W. Mann (2018). Initial coin oﬀering and platform building. Working paper.

Lyandres, E., B. Palazzo, and D. Rabetti (2018). Are tokens securities? an anatomy of initial coin oﬀerings. Working paper.

Massey, R., D. Dalal, and A. Dakshinamoorthy (2017). Initial coin oﬀering: A new paradigm. Technical report, Deloitte.

Momtaz, P. P. (2018a). Initial coin oﬀerings. Working paper.

Momtaz, P. P. (2018b). Putting numbers on the coins: The pricing and performance of initial coin oﬀerings. Working paper.

Nakamoto, S. (2008). Bitcoin: A peer-to-peer electronic cash system.

Narayanan, A., J. Bonneau, E. Felten, A. Miller, and S. Goldfeder (2016). Bitcoin and Cryptocurrency Technologies: A Comprehensive Introduction. Princeton University Press.

Purnanandam, A. and T. A. Begley (2016, 07). Design of Financial Securities: Empirical Evidence from Private-Label RMBS Deals. The Review of Financial Studies 30 (1), 120–161.

Ritter, J. R. (1984). Signaling and the valuation of unseasoned new issues: A comment. The Journal of Finance 39 (4), 1231–1237.

Sanchez, D. C. (2017). An optimal ico mechanism. Working paper.

Schultz, P. and M. Zaman (2001). Do the individuals closest to internet ﬁrms believe they are overvalued. Journal of Financial Economics 59 (3), 347 – 381.

Sockin, M. and W. Xiong (2018). A model of cryptocurrencies. Working paper.

Svirydzenka, K. (2016, January). Introducing a New Broad-based Index of Financial Development. IMF Working Papers 16/5, International Monetary Fund.

Electronic copy available at: https://ssrn.com/abstract=3286835 Token Alliance (2018, July). Understanding digital tokens: Market overviews and proposed guidelines for policymakers and practitioners. White paper, Chamber of Digital Commerce.

Electronic copy available at: https://ssrn.com/abstract=3286835 Appendix

A Proofs

Proof of Proposition 1. For a fraction (1 − si) of tokens sold by an entrepreneur of type i ∈ {L, H}, investors will pay πiR(1 − si). An entrepreneur of type i choose s ∈ [0, 1] to solve,

max (1 − s)πiR + sπiR − γC(s). (A.1) s∈[0,1]

We can rewrite this as

max πiR − γC(s). s∈[0,1]

This is always maximized when s = 0. With no asymmetric information, high and low- FB FB type entrepreneurs will choose sH = sL = 1.

Proof of Corollary 1. In the equilibrium with full information, the covariance between the value of an ICO and retention is given by,

ρv = Cov(πiR, s) FB = E[πiR · s] − E[πiR]E[s ]

= qπH R · 0 + (1 − q)πLR · 0 − (qπH R + (1 − q)πLR) · 0 = 0.

Electronic copy available at: https://ssrn.com/abstract=3286835 Proof of Proposition 2. In the equilibrium with asymmetric information, if an entrepreneur’s type is revealed, his maximization problem is the same as in the full information case given by (A.1). In this case, both high- and low-type entrepreneurs whose type is revealed will R∗ R∗ select sH = sL = 0. Investors will buy all the tokens at a price of πiR where i ∈ {H,L}.

We now show that a separating equilibrium in which a high-type entrepreneur who doe U∗ not have his type revealed chooses sH > 0 exists and is unique. Consider an equilibrium in U∗ which sH > 0 is given by the implicit solution to the following equation,

U∗ U∗ (1 − sH )R(πH − πL) = γC(sH ) (A.2)

U∗ U∗ The RHS is increasing in sH . The LHS is decreasing in sH . Therefore there can be U∗ only one sH for which the two sides are equal. Such an equilibrium always exists as for U sH ∈ [0, 1] the LHS takes values between 0 and R(πH − πL) while the right hand side takes U∗ values between 0 and inﬁnity. Therefore they must intersect for some sH ∈ [0, 1].

The following investor beliefs can support such an equilibrium,

1 if s ≥ sU∗  H µ(H|s) =  ∗ 0 if s < sH

First, we show that given these beliefs, a low-type entrepreneur whose type has not been U∗ U revealed will choose sL = 0. The low-type entrepreneur chooses sL ∈ [0, 1] to solve,

1 1 max sπH R s≥sU∗ + sπLR s

U∗ If a low-type entrepreneur chooses an s ∈ [0, sH ), it is a dominant strategy to choose U∗ s = 0. Similarly, if a low-type entrepreneur chooses an s ∈ [sH , 1], it is a dominant strategy U∗ U∗ to choose sL = sH . It follows that a low type-entrepreneur will choose sL = 0 iﬀ,

U∗ U∗ U∗ (1 − sH )πH R + sH πLR − γC(sH ) < πLR

U∗ This just holds for sH from (A.2).

Electronic copy available at: https://ssrn.com/abstract=3286835 U A high-type entrepreneur whose type has not been revealed chooses sH ∈ [0, 1] to solve,

maxs ∈ [0, 1] sπH R 1 U∗ + sπLR 1 U∗ + (1 − s)πH R − C(s) s≥sH s

Using the same reasoning as for the low-type entrepreneur, if a high-type entrepreneur U∗ chooses an s ∈ [0, sH ], it is a dominant strategy to choose s = 0. Similarly, if a high-type U∗ U∗ entrepreneur chooses an s ∈ [sH , 1], it is a dominant strategy to choose s = sH . Then a U∗ high type-entrepreneur will choose s = sH iﬀ,

U∗ U∗ U∗ (1 − sH )πH R + sH πH R − γC(sH ) ≥ πLR

U∗ Since πH > πL, this is always satisﬁed for sH from (A.2).

Therefore a high-type entrepreneur whose type has not been revealed will choose to sell U∗ sH fraction of tokens to investors. This is the least-cost separating equilibrium since the low-type’s incentive compatibility constraint just binds. Under the Intuitive Criterion, this is the unique equilibrium.

Proof of Corollary 2. We apply implicit function differentiation on A.2 to determine how U∗ U∗ sH varies with different model parameters. For example, differentiation sH w.r.t. R

U∗ U∗ ∂s 0 ∂s − H R(π − π ) + (1 − sU∗)(π − π ) = γC H ∂R H L H H L ∂R

Rearranging,

U∗ ∂s 0 H γC + R(π − π ) = (1 − sU∗)(π − π ) ∂R H L H H L

U∗ U∗ ∂sH (1 − sH )(πH − πL) = 0 ∂R γC + R(πH − πL)

U∗ This is positive. We can similarly determine how sH varies with the other parameters of

Electronic copy available at: https://ssrn.com/abstract=3286835 U∗ the model. sH is increasing in R and πH and is decreasing in πL and γ.

Proof of Corollary 3. The covariance between ICO value and retention is given by,

ρv = Cov(pR, s) = E[pR · s] − E[pR]E[s] U∗ U∗ = (1 − α)qπH RsH − (qπH R + (1 − q)πLR)(1 − α)qsH U∗ = (1 − α)qsH R(πH − πL)(1 − q) (A.3)

Since πH > πL, q, α < 1, ρv is always positive.

Proof of Proposition 3. Using (A.3), it is straightforward to show that

∂ρ v < 0. ∂α

Electronic copy available at: https://ssrn.com/abstract=3286835 B Methodology Behind Financial Development Index

Data are from the International Monetary Fund (IMF) website.22 The Financial Development (FD) Index is computed based on the depth, access, and efficiency of the financial institutions and financial markets. It is computed from six sub-indices that are created from financial market, survey, and accounting data at the national level. Most of these data are from well-known cross-country databases maintained by either the IMF or the Bank for International Settlements (BIS). The annual data series begin in 1980 and cover over 180 countries. The motivation for creating the index and a brief description of its construction are provided below. See Svirydzenka (2016) for further details.

The motivation for the creation of the FD Index was to provide a single indicator that better captures the level of financial development in a country. Historically, empiricists had relied upon simple measures to approximate financial development such as the ratio to GDP of private credit or stock market capitalization. These simple measures, however, cannot capture the multidimensional process of financial development and the multifaceted nature of financial systems.

The procedure for computing the FD Index follows the OECD Handbook on Constructing Composite Indicators. Broadly, there are three steps. First, all of the indicators used in the sub-indices are normalized. Second, the normalized indicators are aggregated into sub-indices that each represent a particular aspect of ﬁnancial development. Speciﬁcally, a sub-index is computed as the normalization of a weighted linear average of the underlying series, where the weights are obtained from principal component analysis. Third, the sub-indices are aggregated into higher-level indices following the same procedure.

There are eight sub-indices and one final index. The final index is the aggregation of two broad sub-indices. The two broad sub-indices are the Financial Insitutions Index and the Financial Markets Index. For each broad sub-index, there are three sub-indices representing depth, access, and efficiency within the given area. The specific indicators used to construct each sub-index are shown in the following Figure B.1.

22http://data.imf.org/?sk=F8032E80-B36C-43B1-AC26-493C5B1CD33B

Electronic copy available at: https://ssrn.com/abstract=3286835 Figure B.1: Indicators Used to Construct the Financial Development Index

Source: Svirydzenka (2016).

Electronic copy available at: https://ssrn.com/abstract=3286835 C Tables and Figures

Electronic copy available at: https://ssrn.com/abstract=3286835 Table C.1: ICO Counts by ICO Team Location

Rank Country Count Rank Country Count 1 UNITED STATES 228 41 SERBIA 5 2 SINGAPORE 142 42 MEXICO 5 3 RUSSIAN FEDERATION 113 43 COSTA RICA 5 4 UNITED KINGDOM 110 44 LUXEMBOURG 5 5 ESTONIA 80 45 IRELAND 4 6 SWITZERLAND 74 46 ST. KITTS AND NEVIS 4 7 HONG KONG 43 47 ARGENTINA 4 8 AUSTRALIA 31 48 CAMBODIA 3 9 CANADA 26 49 CROATIA 3 10 GERMANY 26 50 NORWAY 3 11 CHINA 26 51 SWEDEN 3 12 SLOVENIA 23 52 INDONESIA 3 13 NETHERLANDS 21 53 NEW ZEALAND 3 14 MALTA 16 54 HUNGARY 3 15 LITHUANIA 16 55 MAURITIUS 3 16 INDIA 16 56 KAZAKHSTAN 2 17 FRANCE 15 57 GREECE 2 18 UNITED ARAB EMIRATES 14 58 MARSHALL ISLANDS 2 19 BELIZE 14 59 NIGERIA 2 20 SEYCHELLES 14 60 BARBADOS 2 21 UKRAINE 14 61 BRAZIL 2 22 BULGARIA 12 62 MALAYSIA 2 23 ISRAEL 12 63 TURKEY 2 24 POLAND 11 64 FINLAND 2 25 CZECH REPUBLIC 11 65 COLOMBIA 1 26 CYPRUS 11 66 ARMENIA 1 27 SOUTH AFRICA 10 67 VIETNAM 1 28 REPUBLIC OF KOREA 10 68 ST. VINCENT AND THE GRENADINES 1 29 JAPAN 10 69 GUINEA-BISSAU 1 30 PANAMA 9 70 SLOVAK REPUBLIC 1 31 AUSTRIA 8 71 PERU 1 32 LATVIA 8 72 KYRGYZ REPUBLIC 1 33 ROMANIA 7 73 BAHAMAS 1 34 ITALY 6 74 PAKISTAN 1 35 GEORGIA 6 75 SAMOA 1 36 SPAIN 6 76 ECUADOR 1 37 PHILIPPINES 6 77 DOMINICAN REPUBLIC 1 38 BELARUS 6 78 DENMARK 1 39 BELGIUM 5 79 EGYPT 1 40 THAILAND 5

The Table reports the list of countries where ICO teams are located. Only ICOs that have been completed before December 2018 are included in the corresponding counts.

Electronic copy available at: https://ssrn.com/abstract=3286835 h iuedpcstenme fdsic onre nwihIOtasaelctdb h ae nyICOs included. Only are date. 2018 the 31, by December located on are teams and ICO before which completed in been countries have distinct that of number the depicts Figure The # of Countries 10 20 30 40 50 2017m7 iueC2 ubro itntIOTa Locations Team ICO Distinct of Number C.2: Figure Electronic copy available at: https:/ /ssrn.com/abstract=3286835 2018m1 43 Date 2018m7 2019m1 References