Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=

# Relational Frictions Along the Supply Chain: Evidence from Senegalese Traders

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=mC4ESywpIDc&t=24607s


## Talk (06:50:07 – 07:40:58)

[06:50:07] >> Okay. >> Okay. Hey, thank you very much for uh for being here and for having our paper on the program. This paper is called relational frictions along the supply chain, evidence from Seneagalles traders. It's joint work with Davy Wex who is also here in the audience. So  
[06:50:21] this is a paper about the extent to which information frictions make it hard for small firms in lower income countries to buy goods from foreign markets. Okay, imagine you are a small wholesaler in Sagal and you want to start selling highquality Europeanmade jeans. It's really hard. First of all,  
[06:50:35] you got to find the supplier, most of whom are in Europe. And you need some way of seeing what they sell. Now, even if you do that, you find this supplier, you order 200 pairs of some variety of jeans. What happens? Typically, you pay and then you wait. And how do you know that what arrives is the quality you  
[06:50:49] thought you were paying for? And you can think of that as an adverse selection problem where suppliers have types, but you don't know which one is which. Or it's a moral hazard problem where suppliers are all the same, but they have an incentive problem.  
[06:51:01] Now this these issues are very general and indeed in Grafe's famous studies of medieval trade he even referred to this as the fundamental problem of exchange.  
[06:51:08] It's particularly severe however in lower income settings for a number of reasons. For example, large informal sectors often make search very complicated and contracts are particularly hard to enforce.  
[06:51:19] Economists have thought about this for a long time with theoretical work dating back to at least the 70s and 80s. But trying to understand how big these forces are empirically and also how they interact with each other has been a lot harder for a number of reasons. But one  
[06:51:32] is that it's hard to find variation that is both exogenous and that kind of closely maps onto these theoretical objects.  
[06:51:39] Okay. Now against that backdrop, there's been a huge recent growth in the past 5 to 10 years of something called social commerce, which is basically firms using social media as a way of directly buying and selling goods. And so the majority  
[06:51:53] of urban Africa today has a smartphone and uses social media regularly.  
[06:51:57] And the key is this is true very much also for firms. And you might have seen news headlines talking about things like the Facebook economy or even seen headlines about things like large formal B2B platforms have actually struggled to compete with firms directly transacting  
[06:52:11] over social media. Why is this relevant? Because one of the reasons that firms might be doing this is precisely because it helps them alleviate some of those frictions. So the supplier can now at a very low cost send high quality photos and videos of goods that they're  
[06:52:25] selling. And on the trust side, it's much easier and very low cost for firms to share information with each other for about suppliers or for example or to coordinate behavior, for example, to discipline a supplier that cheats.  
[06:52:38] So this is potentially helpful to these firms that try to alleviate some of the frictions. It's also going to be helpful to us as experimenters because some of these technologies are very easy to sort of tweak certain features to design an experiment that can give us interesting variation and learn something more  
[06:52:51] general. That's what we're going to try and do in this paper. So this is a paper about information frictions and social commerce in B2B trade. Paper has four parts and my presentation will go over each of those parts one at a time. The first part is we bring new descriptive  
[06:53:04] evidence from just under 2,000 garment retailers and wholesalers in Sagal that try to document this fact that they really use social media a lot to learn about and interact with suppliers both domestically and abroad. Now this is  
[06:53:19] actually a very salient fact when you spend time in these markets and you see that this this really is happening a lot but we think that these this sort of has not really been picked up so much in the literature. One fact I'll give you now is that the median firm in our study is  
[06:53:32] in four what I will call supplier WhatsApp groups. I'll show you pictures of those later, but that's basically a group that a supplier maintains where they advertise goods that they sell.  
[06:53:42] We're going to focus on WhatsApp in the study because it's convenient for the experiment, but this is also happening through Facebook and Tik Tok and Instagram. So this is not specific to WhatsApp.  
[06:53:52] Okay. The next part of the paper, we develop a dynamic model of relational contracting between a firm and a supplier. And we use this for three things. One, we use it to help us think about in an experiment, how would you design variation to try to separate  
[06:54:05] these different frictions? Two, what do we think social commerce is doing sort of theoretically? And three, how do these frictions interact with each other? One thing I'll tell you now on that third point is that because we're going to have very treatments that  
[06:54:19] alleviate both adverse selection and moral hazard, it was not obvious to us exanti how those things interact with each other. And what we'll show in the theory is that in general those interventions are complimentary. And I'll explain to you the intuition for why that is.  
[06:54:33] The core of the paper and of the talk is we run a field experiment where we try to use social media in ways related to how firms use it in real life to try to generate variation that targets those different frictions in a major  
[06:54:46] international import market. To target the search piece, what we're going to do is connect treated firms in Seneagal with three new suppliers in Turkey by basically adding them to these types of WhatsApp groups that the suppliers already use and maintain. So we are  
[06:55:01] directly creating matches. So that part is a matching intervention.  
[06:55:06] Then among the ones that we've connected, we're going to cross randomize treatments that target adverse selection and moral hazard. I'll show you exactly what that means later, of course, but basically adverse selection, we're going to give them information about the types of those suppliers they've been matched with. And for moral hazard, we're going to give them help  
[06:55:20] them. We're going to give them information about the incentives of those suppliers that we've matched.  
[06:55:24] We'll see how all of this affects those firms through three independent data sources. One, we will do a mystery shopping exercise where we'll directly go to every firm in the study and try to buy highquality foreignade goods and we'll see what they sell us. So, do  
[06:55:38] these interventions actually help them sell higher quality goods to real customers?  
[06:55:43] Then we will use data from the largest mobile payments provider in Sagal to actually track some of these relationships dynamically. And finally, we'll do a follow-up survey.  
[06:55:53] A preview of what we find is that all groups, so all of the different treatments have roughly similar shortrun effects. But really what we see is that they only really develop meaningful relationships with these suppliers that we have matched them with. When we alleviate those trust frictions,  
[06:56:08] we see a bit more action in the adverse selection group than the moral hazard group. But really what you'll see is that the bulk of the effect is coming from the group where we cross randomize both, which is consistent with a theoretical prediction that in general those things should be complimented.  
[06:56:22] Finally, in the last part of the paper, we will use the experimental treatment effects as moments. We will map them directly onto the theoretical treatment effects in the model and use those to estimate the parameters of the model.  
[06:56:34] What we we don't what we will do is basically say well in our experiment we connected them to firms to specific suppliers and then alleviated those frictions. What we will do in the structural part is we will say what happens if we were to alleviate those frictions throughout the entire marketplace. And what we find is that  
[06:56:49] interventions doing that would generally have a very high return. So these frictions are quantitatively large.  
[06:56:56] Okay, in sort of two sentences, the main takeaways are one, both adverse selection and moral hazard are quantitatively important frictions in trade and that they complement each other.  
[06:57:07] And relatedly, if you think of the search part as a matching intervention, we connected them to new suppliers. We only really see that having an effect when we also alleviated the trust friction at the same time.  
[06:57:18] The second part is that social commerce is really changing how firms go about mitigating some of these frictions in lower middle inome countries.  
[06:57:26] Okay, good. I'll be quick on the contribution and then get straight into some of the details. So there's a huge literature on information frictions in firmtofirm trade of course. Um, and what we think we're adding here is that we are directly designing an experiment  
[06:57:40] that helps us generate variation that is both exogenous and that is aimed to separately target those different frictions as well as highlighting how small firms are using new technologies to try to alleviate literature on technology and trade.  
[06:57:54] We're also documenting this new phenomenon. So this the literature has often focused on platforms like Alibaba.  
[06:58:00] Instead, we are saying that actually a lot of firms in the world we live in today are avoiding these platforms or at least the platforms have not found a way of penetrating some of these markets and are instead using social media directly.  
[06:58:11] Finally, there's a huge literature that's looked at community- based enforcement across a number of settings from trade networks in GRE's famous studies to micro credit where joint liability was a big deal initially and to some extent still is today to informal lending and and so on. And what  
[06:58:26] we're highlighting is that these new technologies may facilitate a modern manifestation of some of these same ideas.  
[06:58:33] Okay. So, as I said, I will tell you first about the context, then the model, then the experiment, and then the structural part.  
[06:58:43] Good. So, we're going to be studying um Turkishmade garments in Sagal and in NAR in particular. Uh Turkey is the second largest source of ready toear garments in this market. China is the biggest. So  
[06:58:57] the reason we chose Turkey is because um in this market 30 Turkish goods have a reputation of being higher quality. It is actually fairly normal in this setting to go into a store and say I would like to buy a t-shirt that was made in Turkey because that is a signal  
[06:59:11] of indicating that I am willing to pay for I am willing to pay a quality premium. And the implication of this is that if you are a firm and you are buying from a supplier and you are buying Turkishmade goods, you are paying a quality premium and you are worried that you're actually not going to get  
[06:59:23] it. So there is this that is the concern here.  
[06:59:29] Okay, there's two main actors in our study. In fact, the real main actors are what I will call firms. These are retailers and wholesalers in uh DARP.  
[06:59:39] They are in many respects typical of what you would have imagined in this setting. They are small, majority owner operated. Um, at baseline, a fifth of them have a supplier in Turkey directly who they will access by either flying to Turkey periodically or by using some of  
[06:59:53] the technologies that I'm going to tell you about over the course of this talk.  
[06:59:56] The others will buy locally from local wholesale markets as they'll go down to the wholesale market and buy from there.  
[07:00:02] Of course, many firms will do both, some combination of these things.  
[07:00:08] Um the other group of people in this study are what I will refer to as suppliers. Now to be clear all randomization is at the firm level. So they are really the main actors here.  
[07:00:17] The suppliers are really a way of delivering treatment. But the suppliers are 30 suppliers that we recruited from an area of Istanbul that is known for exports to West Africa. So this is a huge garment export hub and there's significant infrastructure for exporting  
[07:00:32] already there. Here are two photos I took on the same street. one that specializes in exporting to Seneagal, another that specializes in exporting to Mali. So the infrastructure is very much already there. This is really an enormous wholesale market.  
[07:00:46] Um now depending on your perspective, you may find this obvious or very weird, but um all all of the inter the suppliers who are going to be in the study, these are not manufacturers. They are all intermediaries of Seneagalles nationality whose job is to live in  
[07:01:00] Estanbul and specialize on being the go-between between the manufacturer and the wholesaler and the Seneagalles wholesaler or retailer. This is how these markets often operate and um indeed there's a long literature documenting the importance of these  
[07:01:14] types of intermediaries um in a number of settings. This is going to help us eliminate issue well one this is just the natural cadence of how a lot of the markets work. It's also going to help us eliminate issues of worrying about language and payment technology so we  
[07:01:27] can really focus on some of these other frictions of interest.  
[07:01:34] >> Do you know how many there are? >> Hundreds. Hundreds. We we did a there's a whole procedure where we recruited them. There's a lot. So we have nowhere near the full the full set of the market here.  
[07:01:47] Okay. Let me tell you about social commerce.  
[07:01:52] So we just we we sort of discovered talking to firms that they use social media a lot in ways that really look a lot like some of these frictions which is partly how this um project started. So the first one is something called supplier  
[07:02:07] WhatsApp groups. Now to be clear we did not invent these. These things exist in the world.  
[07:02:12] What are these? This is a WhatsApp group where a supplier who may be domestic but may also be abroad creates a WhatsApp group and adds 50 to 100 of their regular customers. To be clear, these are not discussion groups. In fact, most of the time they are restricted so that  
[07:02:26] only the supplier is even allowed to post anything. So this is not a chat group. They not people are not discussing these goods. This is a one-way communication for the supplier to advertise their goods. If you want to ask about prices, you can obviously send them a private message and say how much  
[07:02:39] is this and how much is that and so on. At baseline of our 2,000 garment wholesalers and retailers, 86% of firms are in at least one of them and the median firm is in four. Okay, so these  
[07:02:53] things are are quite ubiquitous. So firms are seeing a lot of information a lot of the time directly to their phones about different varieties, prices and things like that.  
[07:03:06] Okay. >> Yeah. want to think that this sort of social commentary it looks a lot more like a website say than a Facebook group or something like that right I mean  
[07:03:21] you've been emphasizing that >> no you're right so this is one of the reasons it's convenient for our study so this is this really shuts down a lot of the ability to discuss but the parts where this these similar things happen on Facebook where it's a more open-ended group and then that kind of mingles some  
[07:03:35] of the other pits together so this is useful for our experiment because it really just has the search part but Facebook and Tik Tok really have a sort of back and forth and interaction that has much more of the social bit. The other part point I want to say about the emphasizes the social is you can directly message the supplier and talk  
[07:03:49] to them. Whereas on a platform it's often much more anonymous. You don't know who the person is. You don't know anything about them. And here it's much easier to immediately call up the person and talk to them.  
[07:04:05] uh they will have multiple I think the mean is something in the order of three or four a number of suppliers that firms have it's something like that yeah there's there'll often be specialized by product so I'll have this this type of t-shirt will come from this supplier and  
[07:04:18] this one will come from another supplier >> why there is no Turkish Alibaba like would seem a much more efficient way of doing exactly the same thing >> it's a good question so we uh asked in our survey well first of all so there is no real platform that has penetrated to  
[07:04:32] this for B2B at least that has penetrated in this market to do this kind of thing. There's a number of reasons as to why that might be. We asked surveyed firms have you I won't talk about it much today but we surveyed them have you used formal B2B platforms  
[07:04:44] and something like 80% say no they've all heard of them but one of the reason and we asked them why and the two reason main reasons they give are one they're very complicated uh you have to have a requirements of requirement have a bank account or  
[07:04:57] formal address and things like that the other reason they give is that they don't trust the platform which is that you have to believe that the third party who is operating the platform is going to do whatever it is that they say that they're going to do and one of the reasons we think social commerce may be having a big impact is precisely because  
[07:05:12] you don't have to believe WhatsApp you could directly you are talking to the person on the other side you don't really care whe WhatsApp is not really any meaningful intermediary I don't know if that's necessarily the reason but that's what happens when we ask >> what does it take for a supplier to ask  
[07:05:32] >> uh it takes well it it's it's costless So usually these so you Yeah. Yeah.  
[07:05:43] Yeah. So these are often these are don't take the form of like random links online that you can click. They are typically suppliers are quite cy about who they add. And one of the reasons they do that is because you can see they've actually posted a price up here and they're actually quite um they view when when you talk to them they view  
[07:05:58] this as quite sort of they quote this is a commercial secret and that they only willing to give these types of prices to their regular customers. So often the entry will be by referral through a buyer.  
[07:06:12] Say that again. Uh usually they will you don't have to buy from normally what the other thing they do is they will often periodically  
[07:06:26] prune the group. So they will every couple of months look through the group and say this person hasn't bought anything in the past 3 months I'm going to remove them. So they they take this seriously.  
[07:06:35] >> How do I think about the Seneagalles intermediaries and why they are not solving some of these information? I mean, are they just playing like a logistical role here or you know, because I might have like I might know  
[07:06:48] this guy's family in the car and yeah, >> if he sends me bad goods, I'll go, you know what I mean?  
[07:06:54] >> So, yeah. No, no, no. Totally. So, that that can happen. I would say it was rare in our setting. There are enough the diaspora is large enough that the average person does not know the average person. I think a large portion is is logistical and a large portion is is also an understanding of the culture of  
[07:07:08] how commerce might work and also what types of goods are going to be of interest to this market. Um we generally found it was fairly rare that someone in our study told us that they knew either directly or through some secondary channel who the supplier was. It happened occasionally but not often.  
[07:07:23] These are fairly big markets. There's hundreds of these suppliers in in Turkey.  
[07:07:30] At least it's because >> it's a good question. So we surveyed the  
[07:07:46] suppliers what portion of your exports go through different channels and there's basically three. One is it goes on a boat. That's the cheapest but it takes the longest. The second one is it goes by air air freight. That's the most common. It's faster but is more expensive. And the most expensive is or then there's the fastest which is  
[07:08:01] someone takes it on a plane and avoids the import taxes. About 60% in our setting go through formal air freight.  
[07:08:08] So I don't think so. But if you if you do want to do it then you can do it that way. But most of it actually doesn't go that way.  
[07:08:15] Okay. Good. So that's the first type of WhatsApp group. Okay. Then we also observed a second type of WhatsApp group which sort of maps onto something closer to the trust friction. What we're going to call firm discussion groups. Now these are a much more nebulous concept because they're basically just WhatsApp groups that firms are in to sort of  
[07:08:29] share information. Now these can the boundaries of this thing what really is a group about information is a little hard to define. What I will say is 25% of the firms are in at least one where the explicit purpose of this group is to  
[07:08:42] share information between businesses. Many others are in groups with social networks that are not explicitly for this purpose but still involve sharing business information.  
[07:08:53] So these things exist at a baseline. We did not invent them, but they are con.  
[07:08:56] We're going to try to mirror them in the experiment to try to generate interesting variation.  
[07:09:01] There are no further questions. I'll get on to the theory and then into the experiment. Yeah.  
[07:09:12] >> Uh you may well know a lot of these are based on social networks. So you of I suspect you will the ones we'll create in the experiment they are not going to be. But the ones in real life, these are based these are things that are usually based around networks. And so yes is the answer.  
[07:09:27] Okay, good. Now I'm going to tell you about the theory, but I'm going to give you as the bare minimum that I possibly can for the purpose of this talk and leave the rest to the paper. Uh the goal here with the theory is really to highlight one how to sort of motivate the gener the v variation we will use in  
[07:09:41] the experiment to separate the different frictions. two to sort of be a bit more precise about what we think social commerce is doing and three to talk about the interactions between the frictions.  
[07:09:51] In one sentence, it's a game of relational contracting between a firm and a supplier with sequential search for suppliers. There's three friction.  
[07:09:58] Oh, well, there is a firm who buys some inputs and then resells them for some downward sloping residual demand curve.  
[07:10:04] This is simply a wholesaler or retailer who buys some inputs and then resells them.  
[07:10:09] They can always buy those inputs, excuse me, from a local wholesaler with no frictions. They go to the wholesale market, they buy the good, done. Or they can engage in international trade. To do that, they pay a search cost S. They get a match with a supplier. This will come from some match. They'll get a match  
[07:10:24] specific productivity distribution, but that's not important for today. Now, the reason for doing international trade could be to access newer varieties.  
[07:10:32] Garments is a setting where fashions change regularly. It could be to access higher quality varieties or to access the same varieties but at a lower price.  
[07:10:41] In the paper, the model is very simple and is CES and so these things are going to end up being isomeorphic at least in the way that we put put them in the model. We will actually in the experiment have some way of trying to separate these different mechanisms.  
[07:10:53] Now, assuming they've met with a supplier and they've decided they want to buy something, they're going to play a repeated game with these two frictions, there's a moral hazard problem, which is that what happens is the buyer it pays upfront. In our setting, that happens for 99% of the first time transactions and most 90% of  
[07:11:08] repeat transactions. That is a fact of the market. The buyer pays up front and then the supplier has received their money but the supplier can choose to basically sherk which means they can avoid some share sigh of the cost fixed  
[07:11:20] marginal cost C times quantity Q. If they do that then they get lucky with probability lambda and the good arrives as expected. But if they don't then the good does not arrive. Okay, very standard. What does this mean? It means  
[07:11:34] that the supplier has received the money and they're only going to bother exerting effort if the future value of the relationship justifies it. That is, they have to pay this extra s time cq to guarantee high quality.  
[07:11:45] And if they do that, then they'll get the future value of the relationship. If they don't and they get unlucky, which happens with probability 1 minus lambda, the buyer never buys from them again and it's done.  
[07:11:57] >> Description just like you have these supplier WhatsApp groups. Is there a buyer group with that group? But as a buyer, I can set up a bunch of suppliers >> for a buyer to have you mean five suppliers and one buyer in a group. You mean?  
[07:12:10] >> Yeah. And then the buyer just says, I need of this type of bid.  
[07:12:16] >> It's a good question. I mean, the answer is empirically we don't see them. I suspect the reason is that suppliers would view that as very complicated if they had many many there's the number of buyers per supplier is very big. So, I think if I've never asked that question, but my guess is they would respond and  
[07:12:30] say, "I I can't monitor a hundred different people asking me, do you have this jeans or that jeans?" But the buyers will send private messages to the suppliers.  
[07:12:39] Okay? So, what is this equation telling us? What I want you to bear in mind is that they're only going to do the effort if the future value of the relationship justifies it. And what does this do? It puts a ceiling on this quantity Q. Q cannot be higher than the relationship justifies. If I pay the supplier too  
[07:12:53] much money relative to the value of the relationship, he'll run away.  
[07:12:57] So it's going to constrain how much I can order to the extent that it binds.  
[07:13:01] The other problem is going to be adverse selection. So shums share mu of suppliers or bad types who just always sh if you like. They are unskilled and they are unable to exert effort or to take the good action. The firm is going to update its beliefs as a basian. It's  
[07:13:15] going to order some things see whether it was high quality. If it was high quality then it will increase its belief that this is a good type supplier and so on.  
[07:13:24] All very standard. Okay. What does social commerce do? What we think it's doing up here for the search piece is it's making this kind of remote contracting feasible. So it's not directly helping you pay the search cost s at least not the way that we're thinking about it. It might also be doing that. You could maybe more easily  
[07:13:39] find them. But what we think is going on is really this kind of remote relational contracting is made is really feasible because you can you can see the goods not as good as in person but you can see them at least um through these WhatsApp groups or through Facebook or through  
[07:13:52] Instagram. For the moral hazard piece, we think this is really kind of a shift towards joint punishment. So if the supplier cheats me, I can simply advertise on Facebook or wherever else and we have pictures in the paper of this exactly this kind of thing happening that this I  
[07:14:07] ordered from this person and they cheated me. And what that means is this dynamic incentive compatibility constraint is now this N on the right hand side. I can go and tell n other buyers and they will all stop buying from this person.  
[07:14:20] Okay. And the last one, what we think it's doing here is that it's facilitating a form of social learning.  
[07:14:24] That is buyers to the extent that they share information with each other, which they may do because they are they share social ties.  
[07:14:32] That means that we may I don't have to update all on my own. I can use the signals of other firms to increase the speed of my learning.  
[07:14:39] That's how we think about it. Okay. Now, in the paper, we characterize the optimal dynamic self-inforcing contract and we derive some of the predictions that it has. I'm not going to get into that today. um but that is there in the paper if you want to see it. The one thing I do want to highlight  
[07:14:53] today before I get onto the experiment is simply to understand what these trust frictions do. And what we show in the paper is that the optimal contract has this nice wedge equation where on the left you have marginal revenue. Okay.  
[07:15:04] And on the right you have marginal cost this C fixed marginal cost of the supplier and in between you have the two wedges that correspond exactly to the frictions.  
[07:15:12] >> Yeah. Or are they learning about >> Yeah. No, you're you're right. So I'm assuming that mu is fixed. There is a they all mu is fixed and everyone knows  
[07:15:26] what it is. They have correct beliefs about the share of suppliers. They don't know for a given person in Turkey. Is this a good type or a bad type. And the way we think about social learning is you and I have both found this person in Turkey. We both make an order. That's two signals rather than one. So it's  
[07:15:40] learn social learning about one particular supplier.  
[07:15:45] >> Yeah. You and I know each other and we communicate either through WhatsApp or you post on Facebook and I see it.  
[07:15:56] This seems like a good technology for sharing this information. But why aren't you getting to an optimal amount of doing social commerce anyway? So why have you got any traction in your  
[07:16:11] experiment? Is it because it's a new technology and it hasn't got to the optimal spread yet? Is that the idea?  
[07:16:18] >> Uh so there are two ways of thinking of answering this question. One is we don't have a particularly strong view that what we are doing is efficient in the sense of it's not clear that it it is not optimal. What we are trying to do is trying to understand what this technology does but then also learn  
[07:16:32] something in general about the frictions. So we don't have a deep view that this it hasn't reached its level and we need to push it further. We think we're learning something about what it's actually doing. Now that said, you may be right and it may be that actually firms are still learning and technology adoption is still nent and is growing  
[07:16:47] which it is in this setting and therefore we are helping to or not helping but we are seeing what happens as it accelerates. I'm not sure we have a strong view on which one or the other it is, but both are way both are sort of valid ways that we at least think about it.  
[07:17:12] >> Yes, I think that the well that yes is the answer. Um, empirically we don't we don't measure exactly how many. What I will say is there are certainly many more than an average. There are hundreds of them in Turkey and they are  
[07:17:25] constantly entering. Um and part of the reason why we think these frictions may be severe is the ent commerce may be doing is not what we get to today but it actually makes the cost of entry very low which may end up worsening some of the friction. That's not what we're getting at in this paper  
[07:17:39] but uh that is another separate thing that we think is super interesting.  
[07:17:45] Nathan technologies technology >> that doesn't seem right to me just because less sophisticated players like you and I in a market right they're now influencers and you know we have lots of  
[07:17:58] developments of you know new innovations to solve exactly this type of problem and then in journalist experiments you see that traders >> are you know behave in more rational way  
[07:18:12] >> yes I I agree my preferred view is that we We don't have a strong stance on the optimal market structure whether this is too much or too little but this is a thing that is happening empirically and we are trying to understand what it does by perturbing it. That is my preferred view of this. Yeah.  
[07:18:35] >> Yes. >> Yeah. So how do you think about that?  
[07:18:43] >> So a few things. So the um bas the contract at baseline is that the buyer pays always upfront. That shuts down a lot of the concern of the other side of the moral hazard problem or the adverse selection. The concern the suppliers are going to ship the good and then the  
[07:18:58] buyer will try to hold them up. That's largely shut down in our setting. The reason suppliers are still concerned is they're worried that for example the person may miss may not be very good at dealing with garments and then may badmouth them or def or or you know take their name down or something. I'm not  
[07:19:13] sure that we have a great deal to say about it as our experiment is entirely about the frictions on the supplier side. We took the view that because the baseline contract is payment upfront, the frictions that we want to get at are the ones on the supplier side. Um but I I think it's a super interesting  
[07:19:26] question in general and we don't >> Yeah.  
[07:19:42] surely when people were not allowing anybody >> they do. Yeah.  
[07:19:57] >> Yep. It does. All I can really say at this point is it's beyond the scope of what we're going to do. But I I um there was a world a long time ago when we considered doing this study with having randomizations on both sides and we quickly decided not to do that because  
[07:20:10] it was in beyond our our capabilities. But um I I do think it's it's an area that is super interesting and and in many modern real world settings most goods are actually paid for with trade credit where the buyer pays later and then you have precisely the opposite problems because that doesn't happen in  
[07:20:24] our setting. We chose to put it to the side.  
[07:20:36] certain parts of contract contract. >> Uh it is it's a little bit it it it strictly speaking it's lump sum. I make an order, I pay for it and I wait. But because this is a relational contract, there is an understanding of future business. So it's there's no the  
[07:20:51] contract doesn't say you must pay me in the future if this order goes well. But there is an understanding that it will.  
[07:20:56] >> It's done. It's not like you know x >> no >> empirically that doesn't happen.  
[07:21:15] >> Yes. >> Uh it's a good question. So that could happen. Um, one of the things that I that's precisely what I think social learning is or so this social why social is important. It's leveraging social  
[07:21:29] ties among buyers. If you and I are friends, I do not have an incentive to try to send you off to a bad supplier so that you lose money. That's that's why we think the social part is really important.  
[07:21:43] Okay, good. Now, just to get out to finish this up. So, what's this adverse selection wedge doing? It's you order less than you would like to because the supplier might be a bad type which happens with probability mu and there's a t subscript because you're  
[07:21:58] updating it as your beliefs improve. Moral hazard you're reducing q because you want to make sure the supplier doesn't cheat. If you order too much the supplier is too tempted to sherk and that's proportional to this side parameter but also to row where row is the lrangege multiplier as to how  
[07:22:12] binding actually is that constraint. Okay. Now the one thing that we was not obvious to us was let's suppose you have an intervention that alleviates these things which we're going to do and you do both should they be supplements substitutes or compliments and the  
[07:22:26] answer is they are compliments in general and the intuition is this let's say that I I tell you this is a good supplier it's a good type that means you want to order more from them so your desired queue goes up but then you're hitting this quantity ceiling this moral hazard ceiling really heavily you want  
[07:22:40] to order a lot but you can't and then an intervention that improves moral hazard is really value so in general doing both should be more than the sum of the parts.  
[07:22:50] Good. Now, why did I belabor this theory? Because it highlights these frictions that we're now going to try to get at in the experiment. So, for search, you would like a treatment that facilitates matching or lowers this cost s. For adverse selection, some share of suppliers are bad types. You would like  
[07:23:05] a treatment that gives information or facilitates learning about those types.  
[07:23:08] And finally, for moral hazard, you have to give them incentives. So one would like a treatment that either strengthens those incentives or at least strengthens firms perceptions of those incentives.  
[07:23:17] Our goal is going to be to try to generate those treatments in a way that leverages social commerce in a plausibly organic manner.  
[07:23:25] Let me tell you how we do that. Good. So we take these 2,000 garment wholesalers and retailers and we put them into five arms.  
[07:23:33] Okay? There's pure control with whom we do nothing.  
[07:23:36] Then 80% of them we're going to give them the search treatment which means we're going to match them to some of these suppliers in Turkey. I'll tell you exactly how many and who we match them to in a moment. This one we'll just do search only. That's it. We connect them.  
[07:23:49] End of story. Then we'll do something about we'll give them some give this group some extra information about those suppliers to do with adverse selection.  
[07:23:55] We'll give this group some extra information about moral hazard. And the last group will do both. Okay.  
[07:24:01] All randomization is at the buyer level. the actual suppliers and their incentives are fixed across the study.  
[07:24:07] So there is no sense in which different suppliers are matched to different groups differentially. Everything is at the buyer level.  
[07:24:14] What we're going to do is divide these 30 suppliers into 10 buckets of three.  
[07:24:18] Each supplier manages one supplier WhatsApp group. That's exactly the thing I showed you a few slides ago for the purpose of the study.  
[07:24:25] And importantly, suppliers are only told that they would be connected to a large pool of new customers. We didn't tell them there's going to be all these different treatments with different information given to the buyers. We simply said, "All we're going to do is put you in touch with a bunch of new interested customers."  
[07:24:38] Good. So, what is the search treatment? Those supplier WhatsApp groups I told you about, we had these suppliers divided into 10 buckets of three. They all run a WhatsApp group. We simply put the treated firm into three of those supplier WhatsApp groups. So, they directly get a match. They can see what  
[07:24:53] the supplier is selling. I remind you the buyers cannot post in these groups.  
[07:24:57] This is purely a way for them to see the goods.  
[07:25:00] We make clear that we are matching, not endorsing. This match is not meant to be a we have valid verified the supplier in any way. We are simply an intermediary who is putting you in touch.  
[07:25:09] That's it. So we think of that in the model is we pay this search cost s three times on their behalf.  
[07:25:17] >> Good. Is there is there an endorsement component like the supplier let this person in that is not for your >> um so one thing is I will say is that that we made new groups for the purpose of the study and I'm going to tell you  
[07:25:31] why in a few slides so one thing is that these these do not go into the direct the reg the suppliers regular groups they're already a little separate and the firms can see that because the groups have the name of the study on them we stress repeatedly that we are not endorsing them and I will say empirically we had usually the opposite  
[07:25:46] problem of firms saying I'm very skeptical of this person. Are you going to ensure my order? And we would say no, we are not going to ensure your order.  
[07:25:52] So we usually have to more the reverse problem in the endorsement. And even if you are worried about endorsement, we're going to do these extra things over and above it.  
[07:26:01] Good. So the adverse selection treatment, we're going to try to replicate those firm discussions where firms talk to each other. So what we'll do is we'll add a treated firm here to a fourth group. They're already in three with these new suppliers, but then a fourth one. The fourth one has no  
[07:26:14] suppliers. It only has other buyers. And we structure it so that it has 20 to 30 other buyers, all of whom were in the same bucket and therefore matched to the exact same three suppliers. And we tell them, the purpose of this group is for you to share information about the suppliers that you have all been matched  
[07:26:29] with. Yes. Although uh which which pre-existing supplier groups >> said suppliers in your study already have  
[07:26:46] existing groups known. >> Yes. Are those firms in your set buyers that are >> probably but I suspect that the market is so big that I would be surprised if that has a if that that that there are  
[07:27:00] many who are who are in with the same supplier or who meet someone who is matched with a supplier with whom they knew already would be my guess I I don't know the answer but but I suspect that would be minor there are very many suppliers and at baseline remember 80%  
[07:27:13] of firms do not have a supplier abroad so it probably does happen but I don't think it happens a lot >> question. If I'm in a supplier group as a buyer, can I see the list of everybody else and just invite them to a separate  
[07:27:27] but yes that uh some so the ones in real life often you cannot actually sometimes you can there is an option in WhatsApp that you can disable the feature of buyers of people to be able to see other  
[07:27:40] participants some suppliers use that the ones in the study actually we chose not to do that um and I've actually never heard of that happening but uh I suspect it's just a very costly thing to do and the person doesn't know who you are anyway. Like if I add everyone in the  
[07:27:54] WhatsApp group and you may know three or four people in that WhatsApp but everyone else you they don't know you.  
[07:27:58] They don't necessarily I don't know apart from potentially a large household cost but I've not heard of it happening in period.  
[07:28:12] >> Say that again. Sorry. >> Multiple shoots. There is no relationship between a supplier and them and the treatment groups. It is so they will a given supplier will have an even number of buyers in each of these  
[07:28:26] different boxes. Yes.  
[07:28:34] I don't think so. And again, I want to emphasize suppliers don't know that these buyers are in these different groups. They're just there's a bunch of buyers who they've been matched with and that's it.  
[07:28:45] Okay. Now, the other thing we do is to get this group going, we seed it with a positive review about one of the suppliers they've been matched with.  
[07:28:51] This is a real review based on a real order. We commissioned a team of firms to make mystery orders from every single supplier in the study several months before the study started. So, we have some baseline information about which ones supplied higher quality goods than others. So, we have one of these firms  
[07:29:05] who is not in the study call up and post in the group describing a real order that went well. We structure this so that it's always possible for them to give a positive recommendations. This is a positive information shock.  
[07:29:33] Uh so 80% are buying locally, 20% have a supplier in Turkey. I'm not sure I understand the So it is solving the search part is solving a marketing problem. The supplier is marketing. I'm not sure how does this is this for do you mean this or the other or the search  
[07:29:47] treatment? Oh  
[07:29:59] yes. >> Yeah. So it's a very good question. So  
[07:30:12] to both of them um we have questions in our baseline survey about um well first so first of all I will say we did the main way that we advertise this study is we will we will connect you to suppliers abroad and in general we had a great deal of interest  
[07:30:26] in doing that. So I'm not so concerned that we went to them and told them you need to do this. It was we found it quite easy to recruit firms who were interested in finding a supplier abroad but said we can't because we don't know how to find them and we don't know how to trust them which sort of motivated the study and we have baseline we have  
[07:30:41] facts from our baseline survey in the paper that say exactly the firms described these frictions as real um so I I they they described the frictions and they sort of were very interested in  
[07:30:54] being a part of that's how we marketed the study okay good Um the and so the goal of this treatment is to provide information about the types but in a way that doesn't particularly affect the incentives of the supplier because the supplier doesn't know that this group is around. All they're doing is sharing  
[07:31:09] information. Let me tell you about the moral hazard and then get into the results as fast as I can. For moral hazard, what we're going to do is we are basically going to give going to give all of the suppliers high-powered incentives, but we're going to randomize whether we tell that to the firms. So we're going to tell all the suppliers  
[07:31:23] that we're going to ask the firms to rate you and if we hear bad feedback, we will punish you. Punishment means we will remove you from the study which means that supplier WhatsApp group that you created for the study. Well, we are admins on it. So we can take them off it and replace them with a supplier from  
[07:31:37] the weight list. The idea is this is meant to say we are giving you incentives. Now the treatment is to the treated firms. We tell them hey by the way we have given these guys incentives.  
[07:31:47] Now whether you want to buy from them that's up to you but we have provided them with incentives.  
[07:31:52] So we are not directly shifting the suppliers incentives. We're telling the buyer that we've given them high powered incentives.  
[07:32:04] >> It's it depends on your perspective. So we usually one of these groups has about this much which is quite large.  
[07:32:11] So they stand to lose quite a large number of customers if something like this does happen.  
[07:32:21] So we will enforce. So the idea is we are acting as the as the coordination mechanism is we will literally remove the supplier from the group. So what happens in the real world will often be will the supplier may not know it's even happened or it will be a post on it will be a we literally have in the paper  
[07:32:35] pictures of a supplier's face with a big X through it that gets spread across social media.  
[07:32:42] >> Yeah. Yeah, the I don't know. It's a good question. It's something that I I I wish we had collected more data on and I  
[07:32:56] don't know the answer. Um certain it's it's I don't know the answer is is is the point. I I I I wish we had collected more data on exactly that point. Certainly there are these things happen and suppliers what I can tell you  
[07:33:09] is suppliers are very worried about it. Okay, good. So overall we have these five groups pure control search only and then the cross randomized trust treatments. Um we're going to do two sets of comparisons in general with this  
[07:33:22] design. One is we can simply say how big are these frictions so I can compare pure control to search only to simply this group together version of the trust frictions. You don't have to believe me that we have perfectly separated the adverse selection and moral hazard.  
[07:33:35] However, if I I still it is our goal to try to sort of target the different variations and so we can compare within this group to understand what which of these things are more important than the others.  
[07:33:45] That's what we're going to try to do. So, we're going to look at three sets of outcomes. I'm going to summarize the first one in the interests of time. We did a mystery shopping exercise where we hired mystery shoppers to go to all of the 2,000 firms in the study within a couple of weeks of entering the study  
[07:33:59] and try to buy highquality foreignade goods. How how does that work? Each good they try to buy a particular good. They have we had a bank of about 30 to 40 products. Each product is defined by five horizontal criteria. We picked them so that they are nothing to do with  
[07:34:13] quality. This is a red polo. It is red. It is one color etc. No quality here.  
[07:34:19] And we defined when we ask do they have a good that is like this. So has it increased their ability to source a particular variety?  
[07:34:27] We then buy the good and we have these two tailor come and measure the quality on a detailed 50point scorecard.  
[07:34:33] We bought a lot of clothes for the purpose of this study and we still have quite a lot of them in a warehouse in Dar and if anyone is interested we can uh they may have to pay some tariffs but  
[07:34:47] we can um we can find a way of getting them to you. Okay, I'll summarize what we get out of this which is what we find is that the treatments kind of jointly increase um the probability that they have a good that is like the good that  
[07:35:00] we were looking for by about a quarter. And that's similar across all the groups. And it also increases the likelihood that the good is high quality by about a third. We find no effect on the price. So that suggests there are real gains here in terms of increasing  
[07:35:13] the set of varieties they can sell and increasing the quality of those varieties.  
[07:35:18] Now the question is do these kind of gains translate into meaningful medium and long run effects? So that's what we then try to look at. So we have we look at their relationships using two sources. We have data from the largest mobile money provider. The fact that the suppliers in Turkey are Seneagalles nationals means that a lot of them use  
[07:35:33] this mobile money provider to accept payments, which means we can actually track some of these payments dynamically.  
[07:35:39] We also do a follow-up survey after 3 months where we directly survey the firm. So, the nice thing about this data is that it's real and dynamic. The problem is not 100% of payments go through this channel. So, we can't claim that it's everything. This one does capture everything, but it is  
[07:35:52] self-reported in a survey. Okay? Unless we look at profit and sales and we measured that in a survey. Let me show you the mobile money data. So what you have here on the y- axis is the cumulative order value in dollars. In  
[07:36:05] each of the four groups, I've emitted pure control, sorry, the order value from the study suppliers in the mobile money data. This is money all money that flows from firm in study to supplier in study over time cumulative as the study goes on. There's no pure control because we  
[07:36:20] didn't match them to any suppliers. And what you see is they buy some stuff and it's sort of similar across the groups. If I put arrow bars, we would not be able to separate them. Okay. But I've put this mystery shopping ends because one of the sort of secondary goals of the mystery shopping was to  
[07:36:34] give them a nudge to if they wanted to experiment with the supplier. Here is a customer who wants to buy a high quality for and make good.  
[07:36:42] We don't force them to do that in any way, but if they want to, they have an opportunity. So what happens once that's over? We stop. We're just we're out. The study is finished. Do they continue these relationships? In the search only group, the answer is maybe, but nowhere  
[07:36:55] near like what it was before. I want to point this is 30 months after the study started. So now we're getting into sort of the long run. In contrast, in the trust groups, it looks more like this. So here you this one here is the middle one is adverse selection, then it's moral hazard, and then the one with  
[07:37:10] both. This big spike is January from a few years ago, which is the biggest period of retail sales in this market.  
[07:37:17] Okay. So what we really see is the group where we did both are kind of have developed some meaningful long-term relationships.  
[07:37:23] Now what I'll do is pull these together to any trust and I will test is this increase after we exit bigger than this increase after we is the trust doing anything over and above simply connecting them. And at least at the 10%  
[07:37:35] level the answer is yes. Let me now show you the same thing from a completely independent data source. So here we do a survey after 3 months and we ask them do you have a regular supplier in Turkey where a regular supplier is defined as you've made two  
[07:37:49] orders and you have the intent to continue this relationship.  
[07:37:53] The control mean is 17%. When we pull the treatments together it increases by about four percentage points. Fine. But when you've seen the previous graph you might wonder what happens when I disagregate. And the answer is it looks like this. So search  
[07:38:06] only maybe but we don't see anything. adverse selection. There's some effect, but again, the group effect is concentrated in the one where we did both. And I want to emphasize this is a completely independent data source from the previous one. If you're worried about multiple hypothesis testing, here's when we adjust for the fact that  
[07:38:20] we've tested four of them. This last one remains significant.  
[07:38:32] Yes. What we find is that we ask them, how many suppliers do you have in total?  
[07:38:36] I don't know about varieties but we say how many suppliers do you have in total and then we find a null on that and how many suppliers in Sagal decreases of a similar magnitude to these increases. So it looks like to the extent they have formed new relationships they have substituted away from buying from a  
[07:38:50] local wholesaler and towards importing directly.  
[07:38:53] We don't have enough granular detail about the varieties to know how they entered new products. I would infer probably yes based on the mystery shopping but what we can say is they've substituted from a local supplier that's coming next. Yeah. Um, and just  
[07:39:07] to check, we test that these are all equal with the idea being is this formal way of testing are the search bits doing any more than are the trust bits doing any more than simply search only? And the answer is again yes.  
[07:39:21] >> Yeah, let's do that. So, uh, the last thing we look at is profits and sales.  
[07:39:25] We have some more outcomes in the paper, but the last thing I'll show you here is profits and sales. And we see a very similar picture. So again, we find that there's a large actually a huge increase in profits um coming again from this last group where we did both of the trust spreadsheets together. And again, we can reject that they're all equal,  
[07:39:39] which is telling us that those trust things are different than simply connecting. We find the similar pattern in sales. So overall, my takeaway here is that um connecting them alone did not have a long-term impact, but solving the trust frictions did. In the last 20  
[07:39:54] seconds, what I'm going to do is summarize this very quickly, which is what we do is we take those experimental treatment effects. We take theoretical treatment effects from the model, map them onto each other, gives us parameters. With those parameters, we can say what happens if we were to alleviate the frictions in the whole  
[07:40:08] market rather than just a few suppliers and we find the effects of that are very large. So the gains of alleviating these frictions, which you can do with many policies are potentially quite large. With that, let me wrap up and say I have two takeaways. One, trust frictions  
[07:40:22] contribute to the efficacy of matching interventions. We made matches, but we only find they developed into profitable relationships when we solved the trust the trust frictions. And social commerce is changing how firms go about mitigating these barriers in real life.  
[07:40:35] Uh, and um, it makes it makes it easier for small firms to do remote relational contracting. Thank you very much for all of your comments and questions.  
[07:40:49] >> Great. So, thanks very much. We will uh start again tomorrow morning uh coffee 8:30 and we'll start at 9.