The Microvelocity of Money in Ethereum - Francesco De Collibus Carlo

Transcript

you thank you so much M for the warm introduction thank you all for being here needless to say it's an honor being able to present one of the main contribution I thought my thesis um uh was about which is about the micro velocity of money in ethereum which is of course a topic uh one of the main reason for studying the crypto beside the many fascinating option offered by defi ftech smart contract and all the kind of amazing possibilities crypto offering is as well having an economic system which is completely transparent for us when we can really see all the detail of the

transaction happening so beside the pure algorithmic cryptography achievement we has as well as an economic perspective that could be very much insightful as well for other more traditional branches of economy like digital payment digital currency and so forth so velocity uh my main research question during this article whose conclusion I'm now going to present you was about what is the role velocity play in account based crypto such as but not only ethereum yes now I can expect a question what is velocity how do we Define velocity velocity or uh it's a it's a

concept that can be defined in many way according to in according if we are in physical branches if we are in economical branches if we a fleet like uh the girls competing here but uh in economics if you're according to the Fisher law it's one of the fundamental macroeconomic variable bounding together the monetary mass m Price p the economic output q and the velocity of money B this is one of the fundament of the foundation contribution from iring Fisher and we see that the price for the economic output are supposed to balance with the monetary mass and the velocity of money of course we know how much complex

is economic system here we are at the University we have cafeteria we have to pay fees for tuition for enroll outside there are shops how we going to take uh keep track on a real economy of all this value of all this Vel we can't that's why we usually take aggregate velocity proxy measure to understand how fast the economy is going our uh if our velocity is increasing because because it could have consequences of prices if we and of course on the monetary Mass the Central Banking have to issue or have not to issue but if we take all it just from a data science perspective and if we consider the crypto ecosystem

independently from the what makes a crypto system we now uh we can have a different granularity on this topic instead of just analyzing the aggregate measure we can go to the individual contribution of a single agent into the overarching economic system like we can ask how much the velocity of thisle is concurring to the overall velocity watch that what does it mean velocity in such an aggregated context somehow uh the concept of micro velocity is a matter of scale when we think about temperature it's easy to measure a temperature if you have fever we have to measure our

temperature but Atomic level we know all the particles have different temperature that somehow aage on aaz and on a micro scale but if you look at the micro scale we can see hey what's really happening inside this economic system so we can somehow disaggregate a look at individual contribution of every agent so uh since this is a novel approach of micro velocity I try to introduce exactly how we computed it and what are the results that took that we could measure imagine uh I got Alice received five ethereum at block 10 and this uh

receive another 10 ethereum at block 20 and she's giving 25 uh 15 ethereum to Bob at block 30 what is our overall contribution to the velocity of this system uh measured by heras a block 10 we see that the ethereum she received a block 10 are gone in Block 30 so it means every unit of money as this have to be divided by 20 so they have an individual contribution of 0.25 but the ethereum she receives a block 20 as a higher relative velocity contribution because it's given in a shorter time frame of 10 and if we sum

up the contribution we got we got an average contribution of a velocity of 1.25 at Pro block that's of course a simplification nobody will really giveing 20 block yes you could give 20 Block in ethereum such amount but let's say if we have this discretized quantity of money that's how we can contribute that's how we can compute it I leave some time from the question at the end of the talk so that's that's even mean we can measure the absolute contribution to the velocity we have proactor and we see how concentrated how centralized is the velocity contribution

of every individual agent we see that the first five agents are somehow taking more than 60% in a single block of every velocity the phenomenon of concentration is one of the most fascinating Topic in crypto because we could see what happened to economic system where no regulations outside the algorithmic one take place and I have to say the level of concentration we see from any possible point of view it's quite impressive but we have other findings as well I like to share with you we can measure what is the mean olding time of every unit of money and we see that H is going to hold we see how to the right are the amount of feather is quite an in

liquid Supply we see that it's uh mostly M the near if not more even more near logarithmic scale we see how much people are tend to uh hold and not using it really real economic system it be is much more than a uh investment than really as a currency in like Franks or traditional corren fat currency uh I have to precis the the data set we had to manage to do this study was quite huge because was not only a transfer was the payment of fees was the smart contract transaction rejection and so forth so was a bit complicated sound easy to get um amount of data you need

to do this kind of computation but it was quite challenging for me so basically the data set end on 2029 2021 uh but I mean could be extended and still it's a 6 time frame of research 20 from 2015 up to 2021 another finding that if uh it's very interesting if we divide M if we divide here the what H is is sending Byer balance we got the pure amount of velocity of every single unit of money like like the atom of M of economic matter how fast is it moving

and what we find is quite interesting and this value of course is highly concentrated when we don't take balance in account but it's even highly uneven if we divide by the wealth and we find out that the distribution equates to a power law this is quite fascinating and we ask ourself that if we don't take in account balance how could it be that we have a power law even there we try to investigate if some sort of bro Brownian motion inside the matter this kind of by following snet if this kind of distribution could arise we try different mixing method because the one we tried is uh Lio yeah it's uh last 10 first out no uh

yeah first Fe is fifo but we try a random mixing with the lifo and as well an equal mixing of two and the results were consistent velocity doesn't change stay power law distributed and if we compare to the traditional way we have to measure Fisher velocity and that's what make our approach fascinating I try to explain this slide a bit if we take the traditional way so we measure the transaction by the monetary Supply at the time and we have our approach where we know every individual balance and we take care of every individual transaction we get that our way to measure velocity that you see here in blue is much more accurate rate

detecting Spike here you see we plotted it against the the price of ethereum in dollar you see the spike and you see how much more nuanced is our analyzed V the very good thing we see of course is this value is right sensored because I don't know if I going to spend my unit of money uh one block after the cut off data point of my data set so of course it's right sensored the the time frame but we see that the the error of our um our approach as a prediction value and we are trying to understand if in a real economic system we could use the micr

velocity approach to predict inflation in real economic system the micro velocity approach is something um uh Carlo cayola CL have been working on and we are uh further improving this method I share the source code to plot it to extract this value at the end of the talk so we see with our approach we are we could have we could be in the in in possibility to have a a prediction value for the future a predictive value for the future another very interesting finding in my opinion is how much does this correlate with the token velocity with the overall token ecosystem one of the main focus of

my doctoral fees was um I don't know the diversity of the token ecosystem I mean I was fascinated by the possibility of investigating 1 millions of token which right now we are yeah 2024 are my May 2024 are 100 Milli 1,100,000 different token think to it think of the multitude think of the many economical experiments you can have with with such diversity and you can analyze it on an aggregate way which fascinated me so I tried to uh see how the average velocity of token ecosystem compare with

the one of ethereum problem is of course we are talking about very different scale somehow and the box plot is quite uneven on if you see but still you can see that when the token specification was issued in 27 RC 20c 721 you can see how somehow it correlates with the overall um with the overall velocity of average velocity of this multitude of token we analyze thousand of them here another results we have is since we are we are or we are supposed to we were in the past we still are a network science department we try to analyze how uh does it compare with the network

topology of the transaction networ network network network science is a branch of complex system that study how U the relationship between uh complex system and agent in this case transaction transaction network if I send money to somebody I have a relationship with him is send him to another one you can have multiple way to model it it could be direct and indirected multi-layer in our case we just plotted a weekly graph of transaction we try to extract the topology of the network and we find out a core periphery structure of this network like you have a core which where there is a high number of transaction

between the core is strictly connected plus you have the branches which are quite peripheral to the uh to the network and what we found is that velocity is extremely concentrated at the core of the network when we take weekly snapshots it means most of the action happen inside this core and the periphery contribute um not very much to that uh this is true even if we take the pure velocity of man of money so if we divide by our balance or if we take or we don't take this in account it's even more unequal if we don't don't divide by the balance at

that moment of course one of the biggest advantage of etherum is not not just telling a this add is doing something but let's see what that address is is it a smart contract does it is it known to belong to an exchange to belong to any other entity and so we we use the labels from matter scan to understand how much the who are this top contributor not not many of them are smart contract the vast majority are uh EA funny think if we look if we don't if we don't divide by monetary monetary

balance we find that exchange contribute like 56 56% are of uh in weekly top 10 or if we take weekly snapshot and we see how much often they appearing it of course doesn't sum up because they could have multiple label so in this case uh binance is both in exchange and binance uh so that's why but you you see how much it's concentrated into exchange which play the line which play The Lion Share in contributing to the overall velocity of the system uh while with normal other others not really

contributing by much what I find fascinating about look look at the scale here we have more than 60% if we divide by the balance what we obtain is uh mining is the um other SS usually associated with mining are uh doing big part of the of moving big part of the Velocity but for the if you see the contribute 3% 2% is nothing comparable to this one to the other B um and we I like I like the the label Criminal by the way which tell a lot and uh uh so uh the conclusion of this article

are going to be published they been accepted we just have a problem with my mail address so I can't really give the license at the very end to on a European Journal of data science the source code for computing velocity the way we did on another kind of crypto asset is available here it has been extended to token or to any other um currency you might wish for uh and what what are what is our main consequence we see that velocity emerge as a property of holder as an aggregate we don't see a thing like velocity of money per seale it's always velocity of the olders which sum up from the micro

scale to a mesos scale and to a micro scale and we really see it rising and we wish for the future to understand why it's power L distributed and how this could explain the phenomenon concentration and aggregation the Matthew effect we see many economical system this said I want to thank the coauthor um Dr clayola from City College London and Professor CL which you probably already know thank you so [Applause] much perfect thank you Franchesco are there any questions from the room yep one second um hello hi so one thing that I was

wondering is um how about these uh me Bots and in general Bots that clearly are just uh equalizing um liquidity and trades on chain are you categorizing them separately and if so how much do they actually make up for all of the transactions on the ethereum network that's a very good question and I have to admit I haven't encountered them in my computation up to 2021 because they weren't labeled there are many possible reason which uh could make a very good follow-up research question on this article but I I haven't encountered them specifically in but it's a very good point and maybe a follow-up question on

this I think what would be interesting is also to see maybe you find a way how you're able to label humans uh because right changes everything is all programmatic and and then to see how humans actually um yeah play a role in in this velocity uh discussion uh another approach with uh we tried in a different paper was using e name and to see which one where Ser name on eum naming system uh but of course even this is not per se guaranteed that it's an end user uh a very useful approach to such huge

amount of data we're talking terabyte would be having a more llm machine learning approach to categorize them because sometime you're just overwhelmed by the the sheer number of information like I mentioned is 1 million token how we categorize them even if you read the the white paper sometime you just can make sense of them or to categorize them properly so labeling it's a quite a rough first measure agree with you I think really AI will help us in uh doing this first step part seeing what are the real agents yes one more question yeah and how is the velocity in ethereum compared to the Velocity in real life some of the same is it different did you

compare it uh yeah Carlo is doing some experiment with in Japan about this so personally I didn't compare it with another real economic system but I think this is a research question myel for are following right now so I can't tell you more what I would like to okay that's it thank you Franchesco we have uh five minutes uh break until the next talk which will be a live coding talk [Music]