Session 1: Infrastructure of Digital Economy
3:23 pm – 3:46 pm
Harald Stieber (European Commission) – Modelling the digital economy as a network of contracts
So thank you very much for this opportunity Sheri. And I was actually listening very carefully to Niki, I know this area very well, I’ve worked in this area for several years. And I had a project actually that was very much linked to identifiers. 2017 I concluded the first blockchain project in the commission, where we showed that we can actually report in a unique manner and automatically have a unique presentation in a unique manner only once on the same type on the same financial contract using basically a smart contract implemented on the blockchain and we demonstrate this also with live demonstrator, how this can be done.
So, following up on this, we had published a paper on how basically the data infrastructure needs to be done and I would actually mention to Sheri’s point of over engineering that in a way our conclusion was that to get this done really well using a whole host of techniques from Document engineering, Blockchain various data management techniques and so on. Actually, you need a lot of engineering, you actually need to engineer the entire business process of reporting using all kinds of techniques and to get this done and when we saw that this is feasible, we wondered actually, if we cannot represent all kinds of contracts in this manner, not only financial contracts that differ in a very different very regulated environment and where you link to such things as the LEI, but you also linked to well defined recipients of the required data and so on, can we not do something similar with just any contract, the private contract to trade anything, you know, the most general form of contract and then actually, if we can do this, can we not model actually the entire economy as such a network of contracts and this is of course, not a new idea. In economic theory, this has been argued for a long time. But the practical feasibility had not been, had not been demonstrated.
Okay. So, let me start by this basically stylized fact when you look at sort of say basic performance characteristics of large companies, and you look at sort of, say old companies, non digital companies, if you want and some large digital companies, then you would notice that the digital companies have comparable market cap figures, but their number of employees is lower by an order of magnitude. Yeah. Not by 5% 10%? No, by an order of magnitude. And how can this be explained? Is this just overvaluation on financial markets? Or is there something more substantial behind it? So it relates to the fact that GDP, or the economy as we used to measure it over the last, let’s say, roughly speaking, since the big depression and classical macroeconomics has, has been introduced by Keynes, and then the macro economic models, in the 30s and 50s, we have started to model the economy with a number of implicit assumptions that have not been questioned for a long while, because actually, the models worked well for policy purposes, and to explain what is going on.
But clearly, there are several reasons why this is maybe not a good way to look at the economy anymore. And we have we found that Coyle’s work on GDP is very nice. And a very good introduction, if you are interested in this topic, we can highly recommend it, we have tried to probably go a bit further. And at the same time, I like this reference of David Hume, who was so much ahead of his time and had already this view of the strong interrelationship of human action that value of human action very much depends on who you’re interacting with. And in the same action, or the same asset can have very different values, depending on in which networks they are integrated, and who you’re interacting with.
So, and this is, I mean, this is quite a while ago, a long before we came up with modern macroeconomics. So we try to see what happens if we take a dedicated network modelling approach to this digital part of the economy and, and really model agents as somebody who has a certain probability to enter into contract with Nate with another agent he or she meets randomly, right, so there is a process that allocates you with with a potential trading partner, and then with some probability you enter into contract with a trading partner. And then there is a second round, which of course, Sheri smiling is creating the power law distributions. Because this agent has another co worker probability attached to him that you also enter into contracts with agents that are already connected to this agent that you’re meeting. And then we also had the ambition and we could not completely deliver on this one this turned out to be very, very hard to model bankruptcy so sort of say the case where an agent goes out of business and is not no longer available to contract with us. This is a very hard one.
My friends at Oxford say there are some PhDs that try to provide solutions but I have not seen them yet. But I take their word for it. But still already the first part, the first two elements are very challenging because as you will see going forward they create very interesting distributional properties of such economies.
So quick overview of what this means model design results. So, basically, we wanted to see that in the output of such a model you can see sort of say the value of the benefit that is generated for each agent engaging in these contracts, in these trades, we want to of course be able to model the cost of contracting, we want to be able to replicate patterns that we observe in the real world in the digital parts of the economy such as Uber or Airbnb etc. And of course, it would be nice if any such model would be abstract enough to have some general value that is, that does not only apply to specific case you’re looking at as you saw already on the previous slide, I mean, you have basically this probabilities of engaging into a contract with your matches, you have contract cost explicitly model and of course, you have a measure of success. So, if agents contract and increase their welfare, they are more likely to contract even more and what we could show in the in the study and with this enterprise is that, we were able to replicate power law distributions that we found in actual data on digital economies, we also saw that, we can then basically once we can replicate such data, it is a first step enabling us going forward to simulate changes to such systems and of course, the ultimate goal being that we can test different policy regimes or different you know, regulatory regimes on such economies and to see how it would change the not only the cost of contracting, but the distribution of properties, you know, like speed of contracting, how the welfare is playing out, inequality and so, on.
So, here you see, I put together some, pictures to illustrate also for those that are maybe not familiar with network model. So, in the upper in the north eastern part, you simply have an illustration of what it means, in a network view of contracting to have a different likelihood to enter into contract, when you meet an agent that is part of the population. So, any dot here in this distribution is an agent that is a potential trading partner, but when the probability to enter into a contract with a random match is only 0.005 You see that, that the network is not forming really an interesting structure, when you slightly increase that probability of entering into a contract you start to structure emerging, but what is really typical of such systems is that there is a point where the probability of entering into contract for each individual match is still rather low, but suddenly the network that you create becomes very, very dense and quite complex at least for visual inspection, it is already too complex to draw any conclusions from what you see on the right hand side, upper right hand side, you see, what is meant with power distribution.
So, you see here that, in the left panel, you see that, in more classical supply chains, you will see distribution like that, where of course, some there are fewer bigger players and there are many more smaller players, but suppliers and then customers so, to say are quite similar in the sense of these distributions, whereas, in the digital economy, in the case we have seen here in the case of Airbnb, you see that the supply side and demand side really qualitatively different in terms of the distribution. So, both half of course, fewer larger players and many more smaller players, but the supply side has significantly larger players than the demand side and you’ll see that these distributions really do not lie very close to each other and are qualitatively different and we wanted to understand this another aspect that is something you will meet often in the literature and that is what’s important for us to point out in this project is that very often in theoretical work, that is, by the way also true for financial network, for Financial System Research.
Very often you have symmetry assumptions concerning the distribution properties of populations of systems that you’re looking at. And this is even true for network based analysis. And we saw it very very important here to point out that we probably have to work with and accept the not so nice properties and working off highly non symmetric, highly asymmetrical distributions of such systems. And if we want to learn about real world digital economy, we need to sort of say come composed with such highly asymmetric degree distributions. So, I think for the interest of time, I stopped on this slide, just a question on the data. And that is related to the previous talk as well. I mean, in principle, there’s a lot of data around public authorities own a lot of data, there is more and more open data, there are even some more and more data vendors and data that on the digital economy as well that that can be found on the internet, we could actually have see a future development that even private firms are starting to share and share more of contractual data, if only they are provided with an environment that enables, you know, sharing of such data without sharing strategic intelligence or strategic issues that are relevant for the firm, that would should not be shared.
So, we are aware of course methods, how this can be done. And what was surprising when I met halfway around halfway through the project, roughly, I presented this work to a roomful of contract management practitioners. And we did a short poll. And it was quite surprising to us that actually the readiness to share data beyond your usual business partners or within your own firm was quite high if people were given the opportunity to share data in a privacy preserving privacy enhancing environment, this was quite an interesting etc. And still on the data. So what is also here, important to know and this is also relates a bit to the first talk is that we use din this in this project, basically two different data sets. One was Airbnb data on Airbnb contracts, which is in the public domain. And another data set, there was a third one as well. But I mentioned the other data set was on a contractual network between Italian SMEs that have written contract with each other to share, in sort of, say, in exchange of innovation, and to be able to share data with each other without getting into a position where it would be a problem from a competition point of view.
Yes, so is it a directed network that you’re constructing or it will be undirected, just as long as two parties are entering into a contract? You say that they are connected? The direction doesn’t matter?
Yes, yes, no direction does not matter. So I want to get to the basics of the takeaways. So this is second to last. So just to mention that we used, it’s not only the RPM V data, which is in the public domain, also the data that we got, actually from these private companies, and the data we modelled from the business registry. So we had the third data set that was from business registers. I mean, Niki had mentioned this kind of data, so data that you would link with something like the LEI, for example, you know, it’s actually not I mean, I can confirm what you had on the last slide. I mean, this data is not easy to work with, actually, our experience in the project was that you need a whole host of data modelling techniques so this is a real data science challenge.
So this data is highly, highly, highly heterogeneous. And actually, in the project we had, you see just the number of software that we used to model this kind of data. And we had really a massive data science Challenge to make this data operational and be able to ask meaningful economic questions to this data. So I really confirm the finding of the project. So finally here, this is the important Slide, because the takeaways that we had, are now the project is closed, we can share the, you can actually, if you want to read the report, you can basically go on Google and Google, the EU bookstore, so EU capital letters for European Union and bookstore, when you google this, you get the website of the European publications office, the EU bookstore, and there you Google modelling the EU economy as a, as an ecosystem of contracts, and then you find the two volumes of this study, and you can read it at your ease. And basically, the three takeaways from this project were that we looked into so one, one member of this project was legal tech company, who is helping companies to implement smart contracts as a contract management tool. And they did this in a way that is blockchain agnostic.
So actually, you can use any blockchain that can run smart contracts underneath it, sitting on top of it, and while most of the time it still uses Etherium you could use in principle of any blockchain. And we saw that that actually pros objects have a lot of potential once they are embedded in a sufficiently rich data structure. So, what is that? I mean, contracts are typically written in natural language, right. And in there in plain English and English law applies, many contracts are written in English and English law applies. And then, basically, how can prose objects become useful, you know, in terms of managing your data and through basically populated the digital economy. And there were very interesting propositions already how to work with pros objects to keep actually the full information that is in those contracts. And, and, as if you read the report, you’ll see that we have put quite some work on how to make this prose objects, part of a structured data model, where they interact with other kinds of data, but remain sort of, say, in plain English, and can really interact with other data by using this prose object approach. And this is really, I think, a very, very interesting and encouraging finding, that the contracts can really be this connecting elements that can connect into a network of very, very different parties that have otherwise nothing in common, you know, they are in different industries, they have different levels of sophistication, etc, etc. And they’re just connected by this contract. And this contract can be made itself a digital agent that is connecting to different kinds of datasets. And this can be done in a way that you can really get into a complete sort of, say representation and even replication of, economic relationship.
So the second takeaway was that there seems to be a sweet spot based on the numerical simulation model that was developed in the project, where you have two things going together, a winner takes all logic on the supplier side, where you have this, extreme power distribution, so very, very large agents, connecting to many, many other agents, but going together with very low entry cost for small suppliers and, and consumers at the lower end of the distribution. And that is, of course, very interesting, because, this is a bit of a curveball, and a challenge to an institution like the Commission where we have competition policy and Consumer Policy and so on.
And I think this is something where we need to follow up and to see how we can strike a good balance there. So we would not like to, you know, to regulate the way a large agent in the digital economy and sort of say at the same time regulate the way these increased opportunities for very small agents in the digital economy. So very interesting point too for future work. And the third one, and that will, I think should be a lot of interest to have to Sheri. The third one was that maybe the digital economy is not the end of inflation. Yeah, but we certainly have strong indications that it is the end of certain assumptions that are underlying assumptions or necessary assumption for the pure play price control competition.
And now, I’m always pointing out that this point of the talk that one of my favourite economists Schumpeter, who is from my home country, Austria, more than 100, almost exactly 112 113 years ago pointed out that the micro economic assumption of duality that you can look at cost minimization and profit maximisation interchangeably, is a completely unrealistic assumption. And much, much later, Joe Stiglitz got the Nobel Prize, for the impact of asymmetric information in markets, but Schumpeter died before the Nobel Prize was invented, as well. He already pointed out in 1911, that typically the entrepreneur only knows its own costs, and actually is not in the position of the relevant information to actually engage in meaningful profit maximisation and the digital market structure is actually also not compatible with the assumption of atomistic agents. Yeah. So, another implicit assumption of standard macroeconomic theory is that, that agents can be modelled as atomistic elements in a market and they are not connected by anything else, but the price mechanism. So the price is basically watched by everybody. And the price is coordinating actions of atomistic agents, what we see in digital economies is very, very different. There are many very small agents, but they are not atomistic. And they are not disconnected from each other, quite to the contrary, that there are highly connected, the digital economy is highly structured. And you can quite to the contrary, discuss very different actually, dimensions of this connectedness and of the relative importance of agents. And price is only one information, the price stays relevant.
Prices do not lose all the coordination capability. But they become one parameter amongst several. And when we did the forensic research into actually the process of contracting on a platform like Airbnb, we discovered that these relationships that connect agents with one another are strongly managed, they are very intensely managed. And there is a lot of detailed work actually going into the process. So what actually digital companies are doing, they are not so much competing on price. they’re competing on quality, and they’re competing on process. And process is linked to the cost that users of platforms are facing, but the link is highly non trivial.
So the link is, so to say like, the cost is one parameter that the user is observing, but it links to many, many, many different parts of a structured process. So you would not have a simple functional relationship between the two. Because the process is so structured and multi process, that it would not be easy to say, I changed this part of the process and it would trigger this particular change in the cost, actually would not be able to do that. But this is really something that came out of the forensic analysis of the contracting onboarding process and all the different steps that actually define the contractual relationship between the host and the host key in a case like Airbnb and this applies to other digital economy agents as well. Okay, I’ll stop here.