 Hello, everyone. Thanks for having my paper on the program. I'm going to talk about cryptocurrency returns today. So basically the blockchain technology has shortened the gap between having ideas and developing financial instrument based on those ideas and because of this we have thousands of cryptocurrencies in the market. There are some of these cryptocurrencies are competing for more efficient payment system while the others provide specific services based on blockchain technology. And these cryptocurrencies have a very active trading ecosystem with more than 200 cryptocurrency exchanges all around the world, where people have been investing actively in this market. And because of such an active trading market and because of unique features of the market like the extent of the speculation that's going on, people in finance and economics academia have been interested in pricing of these assets. But so far we have a limited knowledge of what's actually driving price of these assets and these cryptocurrencies. So, for example, what drives cryptocurrency prices, what determines the return of structure of these assets, what's the source of the underlying value and importantly how do investors think about the underlying value. In this paper, I'm going to study the drivers of cryptocurrency returns and hopefully shed lights on these questions. And when we talk about prices, basically we have two broad frameworks for pricing assets in the finance literature. One is trying to explain returns based on characteristics such as size, market ratio, past returns, industry, and we have a huge literature on that framework. And basically in the context of cryptocurrencies, people have tried to explain prices using using some characteristic like this and, for example, we have Leo and Sivinski, who explain cryptocurrency prices based on characteristics such as size, momentum, volatility. We also have the second framework that explains prices based on investor demand. And the idea here is that assets who have exposure to the same investor clientele should show prices should move together or basically show a co-movement in this context. So, in this paper, I'm going to look at characteristics of cryptocurrencies and basically find that the characteristics matter for prices, but the main focus of my paper is on number two. And the idea is that I hypothesized because of unique features of cryptocurrencies, the demand actually matters a lot for pricing of these cryptocurrencies. So basically demand for holding cryptocurrencies can be perceived as a sign of user adoption, which can affect the underlying value of these cryptocurrencies due to the network effect. There are a lot of academic papers on these, Kami and Bang, and Sakin and Jung, and some other papers, but also people in crypto community think about these assets in the same way. And I'm going to show you that I basically use comments from Reddit pages of cryptocurrencies to quantify their reliance on this network effect. All right, so basically I studied the structure and drivers of cryptocurrency returns within the framework of examining crypto investor demand. And how do I proxy for investor clientele having a similar investor clientele, I use these 200 cryptocurrency exchanges, and based on their trading location, I create a connectivity measure where two cryptocurrencies trade on exactly similar set of exchanges. I label them as connected cryptocurrencies, so they have a high connectivity, as opposed to cryptocurrencies that trade on entirely different set of exchanges who have zero connectivity. So I quantify the level of overlap in trading location of cryptocurrencies, and that's my proxy for exposure to similar investor clientele and see to what extent that measure can explain co-movements of these assets. So what do I find? I first find that characteristics such as size, trading volume, age, consensus mechanism, token industries, explains part of the co-movement of cryptocurrencies. But the highest variation is explained by exposure to similar investor basis, which I proxy by cryptocurrencies trading exchanges. I find that cryptocurrencies with the one standard deviation more overlapping overlapping exposure exhibit point 22 standard deviations higher correlation. So the magnitude is large is larger than what all other characteristics combined can explain. And to just give you a sense of what the number means, if you move two cryptocurrencies, two cryptocurrencies that trade on entirely different set of exchanges is if you move them on the same set of exchanges, their correlation goes up by almost 40% of the mean. So it's a very large number. And I find that the effect increases in time horizon. Like if you look at daily returns, the effect is weaker than when you look at weekly returns or bi-weekly returns. So the longer the horizon that you measure return, the stronger the relationship. I also find that this effect is not explained by unobservable characteristics. So it's not the case that cryptocurrencies that trade on the same exchange. They have similar characteristics and they co-move because of similar characteristics. That's not the case. I would honestly move some cryptocurrencies on the same exchanges, which I'm going to use a cause a natural experiment to examine that effect. If you move them exigenously their correlation goes up with each other. And I find that basically what I'm finding reflects common policies in crypto investor demand. I'm looking at the trading order flows because for any of these assets, like for Bitcoin, Bitcoin trades on more than 200 exchanges, you have the order of the same asset at the same time from 200 different sources. You can decompose the order flow or investor demand into exchange specific currency specific and a market wide order flow. The exchange of specific component drives most of the order flows of cryptocurrencies, even if you control for currency specifically. And I also find that the network externalities of user adoption basically explains a significant part of this effect. The exposure to common demand shocks translates into 36 to 51% additional co-movements for cryptocurrencies that heavily rely on the network. And I'm going to show you how I measure this network effect using social media. I'm going to skip the literature. I'm going to take up time. So I use three types of data, data on trading and prices from more than 200 exchanges. I use technological features of these cryptocurrencies from various sources and then I use Reddit social news platform, I use 25 million currency specific comments from Reddit to capture the network. So my data starts in January 2017. I applied a lot of filters to capture basically the larger and more liquid cryptocurrencies, like the market cap should be above $1 million and certain trading volume. So my sample starts with 23 cryptocurrencies and I have 17 exchanges at the beginning of the sample and the average listing per currencies on 4.1 exchanges. And my sample ends with 485 cryptocurrencies, 130 exchanges at the end after all the filters and the average number of listing for each cryptocurrency is on 9.6 exchanges. This is the geographical distribution of exchanges. And there are other differences between exchanges beyond geographical restrictions. So first, there are these variation in geographical rotation, sometimes it comes with some restrictions, like beat hump is a South Korean exchange that's only open to South Korean people. And third, ZEF is very popular among Japanese investors because it only accepts wire transfers in Japanese yet. So there are differences in nationality of people who actively trade on different exchanges. There are also differences, identity verification, limits on deposits withdrawals, transaction fees. So basically different exchanges can attract different investor clientele based on nationality, the size of the trade, how active they are, how educated they are or how, how risk taking they are. So we have different investor clientele on different exchanges and we have friction between exchanges that the money cannot easily flow between different exchanges. And importantly, we have a large variation in share of cryptocurrencies on different exchanges, for example, repo traded heavily during my sample period in those Korean exchanges that are only open to South Koreans, whereas Bitcoin and Ethereum traded more heavily on other exchanges. So based on similarity in the exchange exchange of different cryptocurrencies, I created connectivity measure. So I hypothesize that those cryptocurrencies that also trade on that South Korean exchange, they are more connected to repo, and their price should move much closer to repo than Bitcoin and Ethereum. Based on this connectivity measure, I can classify different cryptocurrencies into different clusters. So each color you see shows a cluster of very close cryptocurrencies. For example, in this context based on my measure, Litecoin and Ethereum Classic are more similar in terms of trading locations than Ethereum Classic and Ethereum or Litecoin and Bitcoin. Even though these are Litecoin and Bitcoin are technologically basically much more similar, they have the same history of blockchain. Litecoin is more similar to Ethereum Classic in terms of trading. repo, for example, has its own isle here. Ethereum and Bitcoin, Ethereum and Bitcoin Cash are basically more similar than Bitcoin Cash and Bitcoin. So based on these measures, I'm going to estimate if higher connectivity translates into more similar price movements as well. So I calculate correlation each month, I calculate correlation of cryptocurrencies based on market adjusted daily returns. And I see connectivity in the previous month. So the lag connectivity can explain future correlations. And I also look at similarity in other characteristics. And the first set of results that they see is that connectivity, various timely explains co-movement. So one is standard deviation higher connectivity translates into points 24 standard deviation higher co-movement. And I find that size also has an effect volume number of trades age being a coin or being token in a univariate regression that also has an effect after a control for everything, the impact of connectivity is by far the largest and we have other measures as well that they kept with the economic magnitude in terms of the marginal effect on the RS square or to what extent they can explain the design a spread of crisis and the connectivity measure is by far stronger than all these other characteristics combined. So if I consider a control for lag correlation connectivity measure still has a very strong effect. That's that holds if I look at the subset of coins subset of tokens, if I look at a more limited set of cryptocurrencies with much larger set of characteristics to control for. And the effect is almost monotonically increasing in connectivity. If I sort every month if I sort the co-movement based on the size of connectivity, and I look at return correlation this is a standard so it doesn't mean negative correlation. And those cryptocurrencies that are highly connected in terms of trading location, they show a very large co-movement in prices and those that are not connected that much they don't move together. These holes, specifically in many different settings like whether I use prices in Bitcoin prices in dollar different settings, but important to eat also holds when I sort cryptocurrencies between exchanges and within denominator crisis. So because different currencies on different exchanges they trade in different denominators in different fiat currencies. Because exchanges might have some noise at the exchange level that noise in reporting prices or converting the fiat currencies into US dollar or Bitcoin, the noise can generate these co-movements. But here, I'm looking at within that South Korean exchange between Bitcoin and between all cryptocurrencies that are denominated in South Korean one. I, some of them are more connected because they are connected on other exchanges, and some of them are less connected. I see that those that are more connected within that Korean exchange using Korean prices, they basically show significantly larger co-movement so the effect is not coming from price difference between exchanges or any notes. And when I create a portfolio of connected portfolio returns based on how connected other cryptocurrencies are to you and how large those cryptocurrencies are. I find that the effect increases in time horizons. And what it means is that like if you look at this four week return of 82 basis point. It means if repo generates 1% higher returns than Bitcoin in a given month. Those cryptocurrencies that are connected to repo, they generate 82 basis point higher return than those connected to Bitcoin. And it's a very large effect when these cryptocurrencies move other cryptocurrencies that are connected to them, they also move. And the effect also holds on a lead black basis, which I'm going to skip here. Let me just briefly talk about this experiment so I use the shutdown of Chinese exchanges as an exogenous shock to the location of cryptocurrencies. In 2017, Chinese government shot down all these cryptocurrency exchanges. And because of that we have a change in location of cryptocurrency. Just imagine two cryptocurrencies, half of their trading volume happened in Chinese exchanges, the other half and two other exchanges. Before the shutdown, these two cryptocurrencies were partially connected, but after the shutdown they're completely disconnected. They were completely exposed to the same investor clientele. They had exposure to Chinese clientele before, but they have their exposed to completely different investor clientele afterwards. So when I look at the impact on correlation, I see that actually correlation between those assets significantly drops. And we also have some assets that their connectivity actually increases after this check and you see that their correlation actually goes up. And the magnitude is very comparable to the baseline results. So it suggests that exogenous variation in trading location of cryptocurrencies, because it exposes them to different investor clientels, it also changes their level of co-movement. Alright, so the next thing I want to talk about is that is this network update. So the idea is that demand for cryptocurrencies may matter specifically because demand can reflect investor and user adoption of this cryptocurrency, which can affect the underlying value of these cryptocurrencies. We have an academy literature and that's a growing literature. And people in crypto community also consider adoption and community building and a network effect, a key source of cryptocurrencies values. And we can see that in the comments that I'm going to show you from Reddit pages of these cryptocurrencies. And many investors, many people in crypto community perceive buying pressures on cryptocurrency exchanges as a sign of user adoption. So they see that people are buying on Coinbase, which is a cryptocurrency exchange, and they believe that's a sign of adoption. Because of these, these demand pressures that we see on exchanges can have an amplified effect on cryptocurrencies because they may be interpreted as a sign of adoption. So if we can somehow quantify the reliance of different cryptocurrencies on adoption, if there are some cryptocurrencies that rely more on adoption and some of them rely less, we should see the effect of demand on prices, much more significantly for those who rely more on this adoption than those who don't rely as much if this channel has an effect. So I'm going to examine that. I'm going to try to divide the cross section of cryptocurrencies into those that rely more on this network effect and user adoption and those that rely less. So I use 25 million currency specific comments on Reddit. Different cryptocurrencies have different pages on Reddit so you can see that this is Bitcoin subreddit, this is Ethereum subreddit, so these are comments related to Ethereum. These are related to Bitcoin. And I read and label 10,000 of these comments as a training sample, as whether they're talking about, they're basically relating the value of cryptocurrency to the network effect, user adoption, community building. And then I feed a random forest model to extract important features that distinguish these comments. And then I feed the rest of 25 million comments into the model to label them. And I quantify the percentage of comments each month that talk about these concepts for each cryptocurrency. So these are some example of example comments. This one says you should look at community support and number of developers working on projects for a certain platform. There is no other project with network effect, even closely failure. People on Ethereum community, they actually talked a lot about the network effect community building developers users. This one says how many users can Coinbase on board everyday Coinbase is a cryptocurrency exchange. The more people that own one light point, the faster the value grows. So looks at the demand on these exchanges and says the higher demand for light point, the faster the value grows. So many other examples. So I feed this random forest model and I find the important features that distinguishes those comments that talk about the network effect and user adoption. And these are some features of the measure that they want to show you. If I look at this measure over time for Bitcoin, Ethereum and recall, we see that in Ethereum community in Ethereum subreddit, a huge percentage of comments actually talk about these issues, much larger than Bitcoin and recall. And, for example, the percentage of Ethereum comments that include the terms network effect user adoption community building and user demand are six 7.7 13 and almost 17 times that of repo, respectively. And that's consistent with common sense about Ethereum, the source of value, but Ethereum, which is built on this community of users and developers trying to add to the ecosystem, as opposed to ripple that's basically governed centrally and it signs contract directly with major banks. And it has a very different ecosystem. And if we sort different major cryptocurrencies based on this network effect we see Ethereum, Ethereum Classic, Cardano, Tezos, basically these platform tokens, which they allow for having decentralized apps and smart contracts on these platforms, they score very high on this network measure and cryptocurrencies such as Binance, Coin, Ripple, Litecoin, they actually score relatively low. So I use this measure to basically divide the cross section of cryptocurrencies into high net for reliance cryptocurrencies and low net for reliance cryptocurrencies. I mean, I see if this measure interacts with demand pressures that I see. Yes. Yes, it's Catherine. I think you should conclude now so that we have a couple of minutes for questions. They can be done in one minute. So, when I, when I look at the interaction of this measure, I find that those cryptocurrencies that rely heavily on this network measure, like if you have two Ethereums, Ethereum one and Ethereum two, when they have when they're exposed to similar ways, their co-movement actually goes up, but by 36 to 51% more than low net for cryptocurrencies. So here, for example, the Y axis shows the return co-movements, the difference between high net for cryptocurrencies and low net for cryptocurrencies. If cryptocurrencies are not connected, there is no difference in their co-movement. But as they become more and more connected, those high net for cryptocurrencies show a significantly larger co-movement, reflecting this demand effect, the significant impact of this demand effect. So in conclusion, I find that cryptocurrencies return structure is mainly explained by exposure to similar investor clientele, praxis by their trading location, the effect increases in time horizon, and leads to a strong prosperity activity. So it's not explained by an observable characteristic, it praxis for exchange specific, a strong exchange specific component in crypto investors demand, and it's largely amplified by this network effect that I just explained. And I conclude by saying that understanding the demand side of the market is probably a key to understanding cryptocurrency prices. Thank you very much.