 Good morning, good afternoon, and good evening, everyone. So today we have Joanna or Boston Fed will hold this session. Joanna is all yours. Hey, good morning and good afternoon. So my name is Joanna Stavens, I'm an economist at the Boston Fed and I'll be the moderator for today's session. Our presenter today is Tony Achenert from the European Central Bank. And he'll be presenting the digital economy, privacy and CBDC. And Tony will have 25 minutes for his presentation and that will be followed by Rod Garats from the BIS as a discussant and Rod will have 10 minutes for the discussion and then we'll have 20 minutes for Q&A. So if you have questions, please put them in the Q&A and I'll give a three minute warning. So Tony, you wanna start? Thank you very much for the introduction. So we are very happy to have the opportunity to present our CBDC paper to this targeted audience. So this is a joint work with Peter Hoffman and Cyril Monet and both of them also here with us today and they're very happy to answer questions during the talk if some questions come up and they're also available to help me out with questions afterwards. So I should say that these, since some of the authors work at the policy institution, the usual disclaimer applies and in particular, we are not part of the digital year project that may introduce a CBDC in the Eurozone in three and a half years. So let me start by motivating the paper by observing that a digital economy requires a digital means of payment. So online sales have grown in the last decade, decade and a half quite a bit and at the same time, new payment solutions are merging Big Tech and FinTech are the buzzwords here and we've listed a few of the recent examples at the bottom of the slides. And as a result of these two trends, the use of cash is declining rapidly. And even in countries that have traditionally used cash a lot such as the Netherlands in green or Germany in red, we see that over the last decade, cash usage has gone down a lot. So if a digital means of payments are used, then the consequence is that a lot of data is generated very much in contrast to cash, which does not generate data. And indeed the business model of private firms such as Big Tech is to monetize data including payments data. And as a result, privacy concerns have grown and my discussion what has worked on this topic as well. So very much looking forward to his comments later. And these privacy concerns could take various forms. I mean, one could be more like political or oveian. For example, as we became aware from the Cambridge Analytica and Facebook scandal that kind of the use of data might affect voting outcomes. I mean, for example, there's the role played during both the Brexit referendum in the UK and the 2016 US presidential election. I mean, most of us are economists. So maybe the political applications are less obvious. So we often care about economic applications, but they're also their scope for price discrimination, preference manipulation and vent extraction. So indeed in a survey conducted in Europe about potential introduction of the digital euro, privacy was mentioned as the number one concern. So in this context of privacy and payments, what we try to understand in this paper, or kind of the research, the broad research questions we're trying to get is to kind of further our understandings about economic trade-offs involved and understand and devide implications for CBDC including the design of CBDC. So in this paper, we develop a parsimonious model of financial intermediation and payment choice in the digital economy in order to contribute to this ongoing debate. In the model, there are two rounds of lending and subsequent production and sales of goods. There are sellers who choose whether to distribute their goods online or offline. This is the venue choice of the seller. So the online distribution is more efficient because it allows for better matching with buyers. So the idea is just that you could meet more buyers or buyers that are a better match for you when you go online. For example, you distribute via the internet. If you do so, the seller requires a digital means of payment. So for example, bank deposits. This cannot be, say on Amazon, you cannot pay with cash. Offline distribution, however, in a prick and mortal store can be settled in cash. And the payment method, cash versus deposits affects the information received by the lender. And one aspect that we emphasize a lot in the paper and it's kind of key for our model is that the lender learns from a payment flows in deposit accounts, kind of consistent with a substantial empirical evidence and some recent theoretical papers that use this notion as well. Maybe the paper I like best is a paper by Lars Norton and Weber in the RSS in 2010 where they document that a bank that observes the payment flows generates a lot of additional information beyond the information they already have about the borrower like the credit score and all the standard information they have when making a lending decision. So the payment flows are very predictive. There's a lot of information in the payment flows. And when bank deposits are used, then this information is learned for free by the bank. Conversely, when cash is used, the bank does not directly observe the information and hence, if it wants to learn in a contracting design, in a separating contract, it must leave information rents to the seller to learn the same type of information. And the information about the seller is useful for two purposes. First, it's about rent extraction in the first round of financing and it's also about future lending decisions. So determining who should get funded in the second round. And as a result from the fact that cash preserves privacy in the sense that it generates additional information rents to the seller, the seller may choose to distribute goods offline and settle with cash, although this is the inefficient form of distribution. So in this benchmark setting, we then introduce CBDC. And we think about CBDC as digital cash. So like bank deposits, it's a digital means of payments and it can support online distribution of goods. And like cash, it preserves anonymity. So if the bank wants to learn the information, then it will have to give information rents to the seller. As a result, CBDC offers the best of both worlds to sellers. CBDC is both anonymous and digital. So just a question, I mean, it's a model obviously, but there are already various digital, private digital currencies out there, stable coins and so forth. So does what you say here also apply to those or is CBDC special and different from them? So we have an extension where the platform issues tokens. So we can talk about kind of, I mean, it's actually, I think on the next slide where we talk about the tension between tokens and CBDC. So here, so the wall of CBDC would be mostly as a means of payments. And I mean, with some private digital currencies, it's not obvious. I mean, especially in light of recent events that they can be a reliable means of payments. Of course, if there's a credible stable coin that might be different. So maybe we can see a bit what the results are on tokens issued by the platform and then maybe get back to it. And perhaps my courses will also wanna comment on the relation to that. So maybe let me ask a question differently. Maybe I shouldn't ask too many questions at this point, but imagine a well-regulated stable coin, something that's envisioned now in the MECA regulation where that's regulated like an error bank, but it's privately issued. And with that work just as well, does the center bank have to do this or is it just about making sure there's a digital but anonymous means of payment? Yeah, so I'm thinking a bit on my feet here, but I think up to this point, it's fine because the key thing is it needs to be digital to support online distribution. And it basically needs to make sure that the information doesn't flow to the lender, right? I mean, once you have this alternative option, we would then have to think about, say it will be private firm issuing that means of payment and maybe they have then incentives to sell that information to the bank. So we would have to think about those private incentives, right? And I mean, one thing that's been argued in the debate about CBDC is that the public authority does not have a profit motive. So it does not necessarily have an incentive to monetize data as much. So that could give it kind of a comparative advantage, actually an absolute advantage even. Okay. So when CBDC is introduced, banks still wanna learn about the seller and offers to separate in contracts but leaving more events to the seller and making online distribution more attractive. And as a result, there will be two efficiency gains of CBDC. Sellers choose the more efficient distribution channel more often, which is good for efficiency. And then also high types who deserve continuation finance are identified more often and therefore we finance more often. So the adverse election problem, I'll talk more about this in the model shortly, will be resolved more often. And for both of these reasons, efficiency goes up. And maybe this is also partly going at Harad Uliq's question. So we've talked about this public option, the CBDC, but then we should stop and ask maybe it's a scope for a similar, maybe even better, a private solution. So what we then do, I think you see what I see and I'd like you to see the top of the slides. I'm trying to get rid of this. Can you see the top of the slide? Yeah, the slide looks fine. Okay, good. So then it's just me, fine. And then we, thank you. And then we introduce digital platforms and these digital platforms can issue a payment token and they can make loans in the second stage. And these tokens foster competition in the market for continuation finance. So the platform learns from the token in as much as the bank learns from bank deposits. The idea again is this link between lending and learning from payment flows. And the bank can use a separating contract to learn. So we will have two lenders that are informed. So they compete with each other. And as a result, sellers are offered better loan terms. And sellers like that. So they strictly prefer tokens over CBDC and in fact tokens fully quad out CBDC. So this is what we call a CBDC adoption crisis. And even beyond this particular paper, I think it's important in this debate that CBDC is designed in a way that it's attractive and it's actually used. And my personal impression is that in some of the debates of policy makers around CBDC is that not enough attention is given to actual adoption. I mean, it's just because we design it doesn't mean it gets widely used. So I think much more emphasis needs to be put on adoption and the specific context of our model once tokens can be issued, CBDC will not be adopted. And this in our model is actually efficient. So the full crowding out of CBDC in that ZW will be efficient. One downside, however, is the downside, however, is that digital platforms may give rise to anti-competitive behavior, which is sometimes called a walled garden. So the concern is that if I'm on a digital platform and then I'm not just using it to buy and sell my goods, but I'm also using the token issued by the platform to pay, then I'm kind of locked in and I'm kind of, I cannot benefit from a future competition. So in one of the extensions, we allow for potential entry of a more efficient platform and how that competition is stifled by the incumbent platform. So, and then Zellers would not switch to the new platform, although that would be efficient. And in that setup, CBDC can be very useful. And here we rely on a very nice quote from this paper on the economics of privacy that states privacy is not the opposite of data sharing, rather it is to control over sharing. So privacy means you control when and with whom you share your data. And CBDC with data sharing will be privately optimal in our model because it induces a competition in the lending market. So when you're a high type seller, you wanna distribute that information to all potential lenders to make sure that you reap the benefit from continuation financing. And as a result, CBDC that is appropriately designed so it has this data sharing features will be adopted even when tokens are available. So if properly designed CBDC adoption is no problem and it will reach the first best allocation as given by the full information benchmark. So on top of distributing goods online and extending finance thing to high types in the second round, it will also mean that sellers switch to more efficient platform if it enters. And the policy implications for CBDC design is that an anonymous CBDC is already welfare improving but can lead to low adoption of once digital platforms issue tokens. And if a properly designed with these privacy features such as data sharing, then this can raise welfare and even reach first best. So what I now wanna do is, I mean, you see there's quite a few things going on but I wanna give you a sense of the model and show you the first two main results before concluding. So there's a bank that gives an initial loan to a seller. The seller chooses whether to distribute offline or online. It learns that it's type. It can be a high type with probability Q. If it goes offline, it randomly finds a buyer which again is high type with probability Q. And then we have this good match high type seller or H seller meets H buyer with probability Q squared generating this high surplus UH. Otherwise the surplus is lower, it's just UL. If you go online, however, you get a better match. For simplicity, we assume the match is perfect. So you always find a high buyer. So that means that this high surplus is reached more often with probability Q, which is bigger than Q squared. So that gives us immediately the efficiency gains from online sales. When you distribute your goods offline, you can use cash and cash means the bank does not learn automatically. When you choose online, however, and settle with bank deposits, the bank observes your payment flows and learns your type and the meeting in which you had with the buyer. And this information is useful for two reasons. It's about rent extraction for the first loan trying to get as much as possible out of the seller. And it's about making a decision about whether to extend the second loan and this second loan is subject to adverse selection because only high seller, high type sellers, H sellers will have positive production tomorrow. So I know that Watt is an H seller. So if I knew I faced Watt for sure, then I will always give him a second loan. But if I face a pool of sellers, then there are so many L sellers in that pool that the adverse election prevents me from making a loan because I cannot break even in that case. Also, sellers can scorn with a fraction of the loan or of sales, it's kind of a more hazard problem in reduced form that gives some minimal bargaining power of the seller vis-à-vis the lender. And we see in this setup, first best under full information is that goods are always produced, always distributed online and any H seller gets refinanced. These are the two dimensions of efficiency. In the interest of time, let me skip the timeline. And just go to the equilibrium right away. When this seller goes online, then the bank just learns the information about type and the meeting for free. However, when this seller stays offline, then the bank does not directly learn. So if it wants to learn, it must do so via screening by offering a menu of contracts. Full pooling of types is never optimal. So it basically chooses between partial pooling, whereby only HH sellers, so these are high type sellers having met a high type buyer, are separated and HL and LS sellers are pooled. Or there can be full separation whereby both HH sellers and HL sellers are separated. So the bank trades off the higher surplus from refinancing. So under full separation, all H sellers are identified and get a continuation loan, which generates a surplus, but there's a cost of separation. It needs to see some surplus to sellers. And this lemma characterizes when the bank offers a separating contract versus a partial pooling contract. Next, the seller chooses whether to distribute goods online or offline. And that's a trade-off between efficiency of distribution and the information rents earned. Offline distribution is inefficient, but the anonymity of cash generates additional information rents. So the specific sense in which cash preserves privacy in our paper, which goes back to other papers, I mean, all the way to work by Charlie Kahn and Krofus, for example, is that the sellers earn an information rent. Online distribution is more efficient. So you generate a higher surplus by selling your goods to buyers, but the bank immediately learns from the payment flows and extracts a lot of rents. So this proposition characterizes, depending on which contract the bank offers, whether the seller distributes online or offline. So with this benchmark in hand, you can now study the introduction of CBDC. So CBDC is digital cash. So it's anonymous but digital. So like deposits, it's digital and can be used for online sales, but like cash, it's anonymous. As a result, when CBDC is used, the bank will, for an online meeting, the bank will always use a separate 10 contract. And the intuition is that under online distribution, we have perfect matchings. We no longer have these HL sellers, this high type sellers that have met a low type bore. Three minutes left. Thank you. And as a result, the cost of separation is lower and the banks will choose more separation. In fact, they will always separate. And CBDC here allows the seller to capture the best of both worlds. They can distribute digitally but capture information events at the same time. So sellers always prefer CBDC over deposits. So bank deposits gets fully displaced. Nonetheless, in some circumstances, sellers still prefer cash over CBDC and stay offline. So there's some inefficiency left in the setup but the use of cash declines once CBDC is introduced. Let me just repeat the two efficiency gains of CBDC. Online distribution is used more often. So we get the better match more often. And the bank is more often informed about the type and can offer continuation financing to high type sellers more often. In the paper, we discuss free extensions. We allow for digital platforms that issue tokens as a means of payments and can make continuation loans. You also consider the potential non-competitive behavior of such platforms. And then we change the design of CBDC and allow for data sharing. So all of this is characterized in the paper and in introductory slides, I had given you kind of the main results on those extensions already. So let me conclude. In this paper, we develop a parsimonious model of financial intermediation with adverse election and payment and venue choice in the digital economy. Online distribution of goods is efficient but raises privacy concerns of sellers. As a result, sellers may prefer to stay offline and settle in cash. While this is privately optimal for the seller, it's socially inefficient for the economy. Allowing for CBDC, introducing a CBDC as digital cash, enables digital distributions without such privacy concerns. So CBDC is the best of both worlds. It's anonymous and digital raising efficiency. And in the paper, we discuss various extensions around digital platforms that issue tokens, potential anti-competitive behavior and a particular design of CBDC which allows for data sharing. Thank you very much. And I look forward to what's discussion. Thank you, Tony. So our discussant is Rod Garrett from the BIS. Rod, you have 10 minutes. Okay. Do you see my slides okay? Not yet. Yes, we do. Perfect. All right, well, thanks for the opportunity to discuss this paper and thanks to the authors for writing a paper that was clear enough that it was relatively easy to read and also thank you for your thoughtful references to my past work. Okay, so I'm gonna try to cover four things in this discussion. I'm gonna talk about the baseline model mechanics, summarize the results, then discuss some of the assumptions and then give three additional comments. And so I did the math and that's 10 minutes divided by four, so 2.5 minutes per item. So let's go. So baseline mechanics. So the banks, so first of all, the seller has to, as Tony mentioned, the seller has to trade off revealing their data to get a good match with the loss of these informational rents. So that's the seller's issue. And then what does the bank wanna do? The bank wants to learn two things. They wanna learn what type of seller they're dealing with. It can be a high type or a low type. And they wanna learn the sale price in the first round because if the sales are good in the first round they can charge a larger amount of repayment on the first wall. And so those are the two reasons why it wants to do this. Now when the seller accepts payment using deposits the bank learns everything. So it can price things accordingly and make the loan decision, the correct loan decision. But when the seller accepts only cash it has a complicated issue. So now the bank has to attempt to learn information. It needs to learn. It wants to learn if the bank is low or high type. As Tony mentioned, the assumptions are such that if completely uninformed the bank will never make a loan. And so it has to try and infer information or write contracts that are gonna reveal the information. So in particular it's gonna use these screening contracts. And so what power does it have when using these screening contracts? Well, the seller wants another loan. And so what the bank can do and a second loan is of more value to a high type seller than a low type seller. So what the bank can do is it can offer higher, as for a higher repayment on the first loan in exchange for coupling it with the permission to offer a second loan. And so through this it can offer contracts that will get the banks possibly to reveal their type and then it's willing to make this high loan. So whether or not banks are gonna do this or which contract emerges in equilibrium is gonna depend on various parameter values but this is the general idea. So the bank is uninformed, it's not going to make a loan when it's completely uninformed so it's gonna try and use these screening contracts. So it's a wonderful, wonderful model. Okay, so what are the key results? Well, as Tony already summarized in the baseline model the sellers are gonna choose online digital payments if the revenues they get from having better matching. Remember the key assumption in this is that if you use digital payments you get perfect matching with a buyer. If they exceed the loss of informational rents that is having an informed bank who will give you worse terms on repayment and on the future loan. If they introduce a CBDC this is greatly improving. Now keep in mind that the CBDC is kind of a magic bullet because the CBDC allows perfect matching without releasing data to the banks. So this is beneficial for the banks. They can have an efficient match they don't have to give up their data. So this is gonna improve the terms they can get on the second loan and it's gonna eliminate this partial pooling equilibrium. So we only have the separating equilibrium but this means cash is used less but when it is used it's more efficient. He didn't really go into detail but there are also these really interesting results related to digital platforms. And so digital platforms can create social efficiency. Again, the key idea here is that banks incur a cost if they reveal their data so they don't reveal their data. If they don't reveal, sorry sellers incur a cost if they were gonna reveal their data. If they don't reveal their data then we have inefficiency due to asymmetric information. So if you can make it costless for sellers to reveal their data you can get efficiency. And so basically the CDDC can do that and we can do that or the platform can do that and we can do that through a CDDC with data sharing. Okay, so now let's talk about the assumptions. So there's a number of assumptions in this paper and all papers of course have assumptions but I thought I would just highlight what assumptions that I think we need to think about. Number one, the sellers choose a distribution channel and a payment means before they know their type. So that is what sets up this screening framework but it's not entirely clear to me that that would be realistic in all cases. We could argue that but what I think is it becomes problematic from a modeling point of view is it makes it hard to think of a true multi-period model where the situation was repeated over and over and over again because then we'd have to sort of imagine the idea that the seller doesn't know their quality each time they do sales. Second, there's perfect matching with all digital payments. That's a key point of this paper because that's the carrot to revealing your data and so there isn't much explanation in the paper as to why we should think that online digital payments lead to a perfect match and by the way, what is the perfect match? In the model, there's high quality sellers and there's buyers that appreciate high quality goods and there are buyers that don't and so the question is does the seller get one of these buyers that appreciates high quality with digital payments they always do and so the question is how do we justify that? Next, the payment option has no impact on sales. So that's something that counters certainly some empirical work. So there's a lot of work that shows that whether or not a retailer offers cash or whether it offers credit or whether or not it offers cash and credit how that impacts their sales and so there's this bank of candidate estimate for example that an increase in 3% of revenue can come from offering multiple payment devices. So this is something to think about but what it also means more fundamentally is that the consumers don't care about privacy and so the consumers play a pretty small role in this model and maybe that's true so there's certainly some work out there that questions to how much consumers care about privacy but if we think consumers care about privacy then payment choice is gonna matter whether or not the privacy of the consumers is protected or not. So I mean remember this is a really interesting thing about this paper is that we're talking about privacy but it's the privacy of the seller not the privacy of the consumer and that also means that we don't have to think about things like Apple Pay for example different devices that might prevent the consumer data from being transmitted to the bank. Okay. Broadly of three minutes. Yep, yep. So there's another thing is that there's no big data collection in the model so we could actually imagine that there are wholesalers that sell to the retailers and the wholesalers produce data that could be used to infer what the sales are of the seller and actually the consumer data the consumers are purchasing from the seller. So I don't think this is a big point but there is this idea that when we think about big data analysis and collecting data from many sources we might have to think about this idea that there could be other ways for the bank to infer the quality of the seller and that actually spills into my last point which is that there's a sale of a single good and that rules out a signaling story and other interesting possibilities and so what I mean by that is that imagine that the firm so when I first read this paper I think I'm gonna go a minute over because I still have a couple more points but when I first read this paper I thought it was gonna be a signaling story not a screening story. I thought the idea would be that the seller would know their type, okay but that type is private information and that we'd be looking for essentially separating the equilibrium of a signaling game and you could imagine that would be true in a world where there were multiple where you could sell an uncertain quantity of goods and so you could think of a standard spent signaling story where the low quality merchant could work extra hard to generate higher sales so they could produce observable sales revenues that made them look like they were a high quality firm but then you could define a separating contract that wouldn't make it worth the firm to do that extra effort if they weren't in fact high quality. So two sort of points there one is that there's some interesting possibilities and I guess they're ruled out by having a single good and also this issue of whether or not if we go back to the first assumption you know, if we relax that assumption that I think it is interesting that we can actually get results that are pretty similar in the context of a signaling game. Okay, so now three comments to finish up. So first of all, as I think I already mentioned briefly it really is interesting to think about privacy from the merchant's perspective. So there's a paper by Richard Posner in the AER in 1981 that it's actually called The Economics of Privacy there's been other papers with that title since then but they actually make this really interesting or he actually makes this really interesting point that if we're gonna think about privacy as a consumer say the consumer's ability not to reveal so he was really talking about concealment and the idea of safe fraud but not revealing information about them that might be a bad signal for an employer then why don't we equally think about firms revealing not revealing flaws in their products. So he thinks about this idea of privacy in a symmetric way and so I do think it's interesting that they're addressing this idea of privacy from the point of view of the merchant. Also, I think it's really interesting to results related to data sharing on the CVDC even though Tony didn't emphasize these. I would make one quick point which is that the reference to acoustic at all there's a reference that makes the same point about privacy as being the power to as being the power to select the reveal information which comes from Eric Hughes and this result is I think so useful because it really feeds into this open banking agenda which is about consumers having the right to share data and the Euro system privacy objectives for CVDC so I think these results are very, very useful. It's probably worth emphasizing that one of the things that this paper doesn't do which a lot of the other papers that have been looking at privacy recently do is focus in on the externalities. So in the context of consumer privacy the externality has to do with the fact that that my decisions might tell you something about what Todd might decide to do for example but that would be the same thing with firms. So firms that look similar might have said we might be able to infer something about the sales of one firm about what the sales would be of a similar firm. So I think before we take these results fully we wanna sort of think about whether or not there might be a role for externalities. Again, just at a high level remember what the paper really does is it says is that if we sort of remove the frictions that prevents information sharing, we get efficiency. Well, that's of course what we expect in economics but when doesn't that happen it doesn't happen when there is externalities. And so that's why I do think investigating the role of externalities is so important. And then I'll conclude finally a very useful model, useful for thinking about a variety of new payment processing platforms that are also considering offing lending. So a great job and very relevant works and I'll stop there. Thank you. So while people put their questions in Q&A Tani, do you wanna respond? Yes, I can definitely get started and then hopefully Peter and Seville will join in. So there's a gives me the chance to pick one or two of the questions that at least I find easier to address. So what fantastic discussion. Thank you so much. There's several of these references. At least I don't know like use and also the AR 81 paper will definitely look that up. On the assumptions, I mean, your point one and five is a bit related, right? I mean, we purposefully modeled the type of the seller not to be known at the initial date. So we don't have to worry about kind of this signaling aspect. We kind of wanted to sidestep these issues. I'm not saying a lot important but just to kind of trying to simply convey kind of the message this way. And then of course externalities are important but also since you've already worked on those externalities. So we thought we wanted to focus on kind of the private incentives to preserve privacy, to make it complimentary. And then the third one I quickly thought is when we would have to work it out but intuitively at least I would expect that our results would generalize if online distribution has greater efficiency and more surplus and kind of going all the way to the limit that it's a perfect match. It's kind of mostly a normalization but I guess this has to be checked. So this is kind of mostly for tractability. And yeah, I'm hoping Peter or Seville may wanna say something to Giselle or the other comments. But excellent, excellent discussion. We'll definitely look at it carefully. Yeah, so let me chime in. So thanks a lot Rod for the discussion. So we actually have on the panel, I think we have like two of our previous discussions. So Harald and Charlie and both of them made the comment that the previous version of the paper was just really complicated. And so we really tried our best to sort of tune down the complications. So I am actually very happy to hear that you found it very easy to read. So that's great. But at the same time, it also explains why we have these very strong assumptions. It's in order to have an easy read and like for instance, the fact that you have perfect matching is definitely, I mean, it sort of gives a lot of power to online trading but at the same time you could relax that a little bit but that's going to be, that's going to complicate things as well. You had another, so the thing about the payment choice, that the payment choice might actually impact sales. We have an extension where we actually have buyers choosing the type of payments that they want to bring and we derive conditions so that they would bring any, they would be different between bringing any types of payment but it's definitely an interesting aspect to consider. And that's, I mean, your point about the assumptions are great because it gives like future papers to write. So that's awesome. And then no big data collection, of course we could have that, this would complicate things a bit again but I think it's related also to the better matching technology and this is definitely something that would be interesting to look at. Yeah, and the one good and the sellers choosing their channel of distribution. So essentially here, the problem is that suppose that you know your type before going online or offline. So the problem is that the choice of the platform of distribution platform is going to reveal some information and then it becomes tricky. I mean, the analysis is, you know, even more complex than what we have because it might be that as a high type you actually want to choose the same distribution platform as a low type in order to be able to pull together and sort of extract some rents out of the bank not knowing whether you're a high or a low type. So, you know, that's why we started actually with this assumption but then we realized that it was, you know, it had the layer of complication. So we chose the case where you don't know your type and then it's only going to be revealed with your sales. And the idea would be that it's a young firm and, you know, it's not, it doesn't yet know whether it's good. It's going to be very demanded by customers. And, you know, that's sort of the story that we have in mind. But thanks a lot for the great discussion. Okay, so we have a couple of raised hands. Mohammed, you wanna go next? I think Harold was the first, but okay. So I'll go first. Okay, so one, yeah, that's okay. So one question is a question or comment, I don't know. So it's about privacy of sellers. So in reality, what I can think about the sellers' privacy is mostly about tax evasion that they usually want, especially more kind of small businesses that they want to keep privacy. Of course, I'm talking about legal business, right? If a business is completely illegal, that's a different story. But if it's a legal business, so the main thing that comes to my mind in terms of privacy is just they want to do tax evasion and they want to kind of pay less tax, both sales tax and also kind of the whole business tax or income tax. Other than that, I think in reality, I don't know if really firms want to kind of hide anything. So if what I'm saying is true, perhaps one extension or one version of your model could consider tax evasion and basically tax evasion increases the size of the pie and that might have some implications for your results. But yeah, that's my comment or question. Thanks. You guys want to answer that or shall we move on? Okay, Harold? Yeah, it's also an overarching question and banks and credit card companies to considerably we are prevented from using information on the one hand. And I think the concern with CBDC may be exactly the other way around that people fear that once payment, once it's cash substituted for CBDC and the information goes through the central bank that now we have a government observing all the transactions and people, I think feel very nervous about governments being capable of observing all the transactions we do in daily lives. And I think it's a cool paper. I really like the paper and it's an important topic. So it's more of a constructive comment of going, writing more papers that investigate how much people are concerned about privacy arising when the central bank has access to all this information. And there are probably surveys of that type and what can be done about it. Okay, Katrin? Yeah, can you hear me? Yes. Yeah, okay. So I was, yeah, puzzled by the results you have with the platform and CBDC and that platform, if I understood correctly, is driving out CBDC except if CBDC is going to include something on data sharing. And I wasn't quite sure what you were thinking of because I fully agree with you that this adoption of CBDC is very important issue. And the more we know about features that drive adoption, the better. And from survey evidence, I would always think the more privacy, the better because that's what you get from survey results. But it seems that your model is giving a more nuanced result on that. So I would be happy to hear more about that. So what I could say on that is, so agents like to stay private in this model when they face a monopolist lender. So the moment there's a platform that can potentially compete with the bank, they actually like that the information is revealed. And because if that information is revealed and these two competing lenders, they can really compete with each other. So that's the main mechanism through which that works, right? So once you're facing monopolists, you love to hide your information because you can extract some information, rents in exchange for that information. Once you face competition on the other side, you're very happy that if a type gets revealed to everybody because that's the only way you can read perfect competition. So that's the mechanism that's at work here. But then the competition story behind that and not preferences for privacy, but rather enabling competition in one. So the preference about privacy or, so the preference about privacy is very much dependent because it's an economic preference, it's not something that falls from the sky. So that of course very much depends on the competitive environment. So yes, that's crucial, but it's also sort of to have a economically well-founded coherent story, right? That it's not something that falls from the sky, but that depends on the primitives of the model and the competitive environment is a key feature of them. Yeah, just to add it, so basically the amount of information is endogenous and information determines lending outcomes and depending on the competitive structure, these will be different, yeah? I can try to answer to Harald Wulig's overarching point a bit. So because I think it's basically related to trust, right? Whom do you trust? And if you're really worried about public authorities or the government, then you should not have a system where you give all that information to the government, right? So my personal interpretation is there's also a bit of a geographic heterogeneity. So it seems in Europe, there's a lot of trust in the government and in public institutions and maybe not so much trust in private enterprise, especially if the private enterprise is not based in Europe, but it's maybe based in some other country. And then maybe the European Commission has the story about sovereignty that is linked to that. While in North America, the fear might be a bit more about about giving all the information to the government. And I mean, there was this, it's also related to the events in Canada and the payment system and interventions in the payment system were used to deal with the truckers' strike, which maybe was maybe not the best way of supporting trust in the payment system and in public authorities because then maybe future adoption of CBDC is less likely. So I would, it's kind of beyond the model, but I think it's a lot about trust in institutions. And I think for the survey paper on the economics of CBDC, we had looked a bit on some of these surveys and kind of how people feel about that. I could just ask a quick question about the privacy versus security, because it seems to me that privacy is security of the data and information, but security is financial risk and financial loss and maybe I'm biased by the US and we've had lots of issues with Zell, for example, this recently and it's not necessarily the bad guys. Sometimes it's just irreversible transactions. I just wanted to get your thoughts on security versus privacy. I'm not quite sure what you mean by security. I mean, what I would understand is essentially, I mean, financing terrorism or something like that. Is that what you have in mind with security? Well, it just meant that there's a trade-off sometimes between privacy and security, right? So when you give out information, like with credit transactions, for example, you have much more financial security. So, yeah, so I haven't thought about this aspect in this dimension, but it's sort of related to what Harald was mentioning. I've been thinking a bit about the KYC ML regulation and the fact that if you issue a CBDC, then probably you want your central bank to satisfy KYC ML which means that the central bank will have to have access to the transaction data. And there's no... So either we're gonna have to design a CBDC that doesn't give access to the government about our data, but then we're gonna have to live with the fact that as with cash, CBDC might be used for those transactions that we might not agree with. But this might be efficient, so who knows? I mean, some other aspect, and it's kind of trying to take it a bit in a different direction, I mean, going back to the idea of adoption, I mean, if you think about security as the dimension that the CBDC system will be safe, there will be no cyber attacks on the CBDC infrastructure. That's probably a necessary condition for adoption. Thanks, Martin. Thank you, Joanna. Just kind of going back a little bit to the question that Harald asked and Tony's response like, well, like about the privacy vis-a-vis the government and that is very much depends on like the level of trust, the local level of trust in the government. When that level of trust is really high, people might be more willing to share data, but perhaps it is even more fundamental than that if we were to put it in an economic model because like basically trust in this context is like about information sharing. And then the question will, the person with whom you share that information actually abuse that data. So in that particular case, that will be the government. But when we think about designing or thinking about the privacy that you get in payments is actually more fundamental in the sense that do you actually need to trust or not? Because if you have the privacy, you don't need to trust. You don't have to worry if you share the data, when you share the data that will be abused because you just have the option not to share it. So there was just two cents on that. Like just, I think it's just a little bit more than just a geographical dispersion in trust that we need to take into consideration just the fact that you actually want to have a system where you would need trust. I mean, this brings some thought about this in the sense that it's true that if you share your data, you have, I mean, you might be willing to share some of your data with say a bank, whatever, but under the assumption that this bank is not going to resell the data to somebody else. And then you have to trust, you know. So trust is in payments, we know trust is of the essence. To me, this point sounds a bit like it's a question of actually, in some sense, where the data is kept, right? So because obviously we can think of designing ways of data being shared for it being in a given location. But so what do you say it sounds a bit like there's a role of thinking about where the data is kept, right? So can I kind of share the data? Can I start my own data with me privately and then I can share it with somebody? Or is it shared in a more centralized place and I give consent for it to be released from that centralized place or not, right? Because I think in the back of what you have in mind and also what Harald said, it's kind of like, there's some chance that the benevolent government or the benevolent center bank turns into a rogue actor with some probability and then something goes self. So like, I don't know, the seller gets expropriated or something like that. Well, there are data breaches, right? So you might give your data to a specific merchant or specific financial institution, but it doesn't guarantee it's not gonna go beyond that. Yeah, I think the general point is that there are just a whole host of contingencies that you can think of, right? You can assign high probabilities or low probabilities to those, fair enough. But the question is like, do you want to design a system where you actually need that trust that these contingencies do not happen? But it's not really a comment on your paper, right? It's kind of, as Harald said, it's really like, hard to capture an economic model and but perhaps something for the next level. Yeah. Yes. Okay, Russell, you wanna take over? Yeah, we did travel at East Coast and then even later in Europe. So thank you everyone for staying with us. So it was the last session of this year. So we will take a break for next January and February for the job market. So keep an eye on our website and our email announcement for any further webinar. Thank you, everyone. Happy holiday. Thank you. Thank you. Thanks for the discussion. Thank you. Thanks, everyone. Thank you. Happy holidays. Happy holidays. Rod, that was a great discussion. Yes. Thank you. That was really good. Thank you. Thank you, Rod.