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.
=

# Beliefs Over Contracts

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=mC4ESywpIDc&t=19890s


## Talk (05:31:30 – 06:22:08)

[05:31:34] you very much for um the to the organizers for for inviting to present our paper. Um so this is one of those projects that has a pretty simple backstory. So I'll start with the backstory and I'll get into the paper.  
[05:31:46] Um four years ago I I chanced on having dinner with the CEO of a very large firm. It's one of the big techs in Africa very scaling. Um I've quite been curious really in understanding why firms do what they do. So at dinner I posed a question um you know why do you  
[05:32:01] pay these mobile money agents spread across entire country millions of them you know the same wage structure same compensation for the past 15 20 years.  
[05:32:10] He said you know what great question I keep making money so why do I want to change it? um why why try to fix it if it's not broken? uh for you know I even got more curious then I asked a second question um how do you know it's broken  
[05:32:22] if it's not tested um so eventually um you know he was stunned there was no answer so quickly what started as a conversation became a collaboration um within the space of months we were we all curios we wanted to learn and so that's the story I'm going to be telling  
[05:32:35] you okay that's how it started so I'll talk about beliefs and contracts and the motivation is pretty simple um when we think about firms more generally at least three things are true about firms, right? Um they tend to rely on contracts  
[05:32:49] typically to actually align the incentive of firms with workers and agents more broadly, right? So that we know is true about firms. The second thing that is so true is that uh you know having the incentives incorrectly can also be very costly to firms. Okay.  
[05:33:02] Perhaps what is more true is that when it comes to making choices and decisions about how to incentivize workers, more generally managers are going to become very central, right? executives and managers within the firm are very central in the way we think of, you know, actually making decisions about  
[05:33:16] the contracts. So the question though is that do managers really understand really you know the relative efficacy of different contracts in practice and if so how? Okay, that's what I'm going. So I'm going to do a basically this two-stage stuff. I'm going to combine  
[05:33:29] what I call this detailed elicitation of beliefs of workers in this firm. Okay, again going back to my the backstory along with a very large scale if you want call it an almost a nationwide experiment um with a vision of trying to actually show which contracts work and  
[05:33:44] what doesn't and to what extent does that allow us to learn more about the you know what managers actually know when we think of uh principles in the way we think of principal models the question is going to be very simple as I mentioned but I mean it's very fundamental but it's a very fundamental  
[05:33:57] question but you know super super simple when managers choose among different incentive contracts okay at any in time.  
[05:34:04] Do they correctly predict which contracts work better for firms? And if so, what are the consequences of that?  
[05:34:09] That's why I'm going with this paper. So, very super fundamental paper. Now, I'm going to basically tell you what the results are and then I'll get into it.  
[05:34:16] I'm not going to do so and I'll skip literature review and all of that if I have time. I I'll get to that. So, I'll show you four results. Four results.  
[05:34:22] First one, um, it's going to turn out that the best performing contract in our context is going to improve performance by over 20%. Okay? relative to the status quo contract. Okay, we're going to be curious about you know asking the  
[05:34:35] manager in this case the CEO despite similar participation across these contracts relative to the status quo. So which actually would imply that firms are leaving about 20% of money on the table. Okay, that's one way to think about this. The second thing is that  
[05:34:48] managers and by extension firms are going to be very very good at predicting what contracts are bad but less so about what contracts are good. Okay, which actually implies this systematic biases in the way managers make decisions when it comes to choice of contracts. Okay,  
[05:35:01] that's the second result. The third and fourth results are going to be, you know, thinking more about really what explains the differences in the relative performance of these different contracts. And I'm going to say that in this context, we're going to be have the ability to actually measure effort in a very precise what I'm  
[05:35:15] calling labor supply here in very high dimension. Bear with me and I'll get to that. By test, you're going to see that actually effort is going to basically change in the precise way that we like to see. you know what drives the performance differences. Okay. Um and in fact labor supply is also going to change in a very interesting way where  
[05:35:30] labor supply is going to be very persistent for some particular contracts and not other contracts. And when we get there we unpack what is going on. The fourth results and perhaps the last one is super descriptive but it's basically the idea that what drives actually the predictability of managers by extension  
[05:35:43] firms in this case. It turns out the complexity of the contract matters presumably this will be my prior getting actually into the project but perhaps more distinctively the hierarchy of the manager. the management actually matters a lot and in fact it's going to matter in a very precise way where senior level  
[05:35:58] executives are very good at predicting the IC constraint in the way we like to set our models of contracts but then junior level managers are going to be very very predictive or good at actually predicting the participation constraint so again it's really kind of speaks more broadly to how we actually want to design actually the incentives for the  
[05:36:12] executives themselves okay going forward so those are the four results I'm going to tell you I think by now you can make your own conclusions about the paper but here's one conclusion that I came up with it turns out managerial misperception and by extension you know principles within firms and having said  
[05:36:26] anything about heristics I'll use that word if you're not happy you know towards the end please tell me to some extent actually are a key source of you know inefficiency in the way we think of what you know contract design choices within firms okay so that's the paper that's where I'm going in the next you know I don't know how much time I have  
[05:36:40] done this is a good time to ask questions if you have one otherwise you know just wait let me tell you what I have and please um ask me the questions I'll actually walk you through the institutional setting and the data we have in fact we have more data per you know you perhaps more than we need but you know every data is going to be  
[05:36:55] useful okay I'll then talk about the experimental design the RCT itself and then I'll walk you through maybe some framework a model that allows us to think of how we design the contracts okay and there's a vision in doing that and then I'll walk you through the four results we just talked about two things  
[05:37:10] to pay particular attention to I'm going to be thinking of participation for the most part as the idea that you know workers in this case agents can exit or you know leave different contracts if they are not happy okay uh throughout the experiment and then performance is going to be coming in through along the  
[05:37:24] lines of revenues or profits and which by extension with profits in the way we set up the the paper and then lastly I'll talk about managerial predictability and then I'll conclude if I have time I'll tell you about the connection to the literature but there are classic literatures that we contribute to which I'll spare you if I  
[05:37:37] have time I'll talk about so that's my plan any questions please Yeah. So I would let me get into that.  
[05:37:55] [laughter] Okay. So so yeah. Okay. Let me let me actually tell you what they are. Okay.  
[05:38:00] So the setting let me start with the setting. I'll tell the status quo tell you what we propose. You know it's really more general but these are incentive contracts in way how do we you know typically pay workers. These are wage contracts if you want. Okay. So, so I'm going to be working with this  
[05:38:14] very large firm. Um, it turns out this was the very CEO that I was having dinner with. It's called MTN Mobile Money. Okay. Uh, it's by far one of the the largest, you know, multinationals that we have in the country. In fact, they do have footprints. Uh, for those  
[05:38:28] of you that are familiar with my work, I've been working with them for um, you know, the past. So, over time, I've developed these very deep collaborations with them to the extent that they allow for learning um, within the F. So very large multinational with a lot of footprints not only in Ghana but across Africa and to some extent some places in  
[05:38:43] the Middle East as well. Okay. As far as Ghana is concerned they have closed about 90% share of the market. Okay. Um so think of this as basically the monopoly basically the market for digital finance in this case mobile money in the country. As part of their  
[05:38:57] operational activities they have this dedicated set of retail agents that retail financial services on behalf of the firm. What we're going to call agents or workers by extension if you want. Okay. And in fact, the goal of these agents is basically think of this as human ATMs, right? Essentially taking  
[05:39:12] deposit and giving out withdrawals. Okay. All right. That's basically what they do. Now, as far as the organizational structure is concerned, I'll spare you the details. I want to just mention something on the right which is basically almost like a three-layered management structure where  
[05:39:25] you have folks at the HQ, some folks within the middle belt and you know some folks that are very actually you know on the ground you know interacting with these agents and workers across the entire country. Now something I want to highlight is that in my informal conversations at least with managers  
[05:39:39] especially at the HQ they tend to rely a lot when it comes to incentive design contracts they tend to rely on experience heristics less to do with experimentation okay uh and that's going to become something that you know I'll speak to down the line okay how is the  
[05:39:52] compensation environment um MTN again what going to be the platform in this case the firm is going to set basically the contract for these agents okay central one single contract these agents are going to take the contract as given and is going to exert the optimal  
[05:40:06] effort, okay, according to your contract. So, basically take it or leave it contract. Of course, agents are going to have discretion, okay, over how much effort to exert here. Think of effort in very precise labor supply, okay? How many hours I work in a day, how many days I work in a week, which days I  
[05:40:21] work, how early I show up to work, how late I leave, and we're going to observe all those high dimensional stuff. Okay?  
[05:40:25] Uh in the data So um the typical the typical mobile  
[05:40:40] money agent is going to basically uh be a shop owner. Uh it's going to be a corner store. In fact, this is one example of that. Um this is a classic one. This is a generic picture, but modern mobile money agents of course  
[05:40:54] bundle mobile money with other things. For example, in a a store they may be selling like a bag of rice or something else that actually are nonfinancial related. Okay. About 70% of their revenue is going to come from financial and about 40% is going to come from non-financial. For revenues,  
[05:41:09] profitability is about 50/50 across financial, non-financial. But of course, going back to whether the operator is the owner or just a delegated worker, about 70% of the time the owner is the operator and 30% of the time the owner is some delegated person who delegates  
[05:41:22] to. All right. So going back to the contracting environment um just to give you a sense of what the status quo look like in the status quo agents are rewarded on commission basis okay so a peace rate all right now to give you a  
[05:41:36] sense of what that looks like if you someone was to make a withdrawal okay take money from your account of say a th000 Ghana CDs the customer who actually conducts that transaction pays a 1% fee okay on that particular transaction roughly about 10 CDs right  
[05:41:51] out of that this is paid to the firm MTN they owns the platform Okay. Well, 40% of that 1% goes to the agent and the 60% is remained with the firm, the provider.  
[05:42:01] So, think of it as effectively what the agent is getting is about 0.4% of the total transaction value that gets transacted. Please, there was a hand up.  
[05:42:17] >> Absolutely. Yeah. Yeah. Yep. Yep. Yep.  
[05:42:25] Nope, it's not a labor supply. I'll walk you through a bunch of different effort dimensions. When I say labor supply, this is, you know, this is effort more generally. You can impact effort in different ways through liquidity, through labor supply, through even advertisement relationship and all of that. I'll get to those issues, but I  
[05:42:40] think most of the effects are going to be on labor supply, which is why I'm I'm thinking of labor supply here. But I please go ahead.  
[05:42:53] Yep. >> They are managers. I'm going to think of all of them as managers. Okay. Anyone who's an executive or a staff that actually interacts or may not interact formally either directly or indirectly with agents, I'm going to think of those guys as. So, everyone from the bottom all the way up is going to be what I'm  
[05:43:07] going to be calling managers for now. Okay? Bear with me now. Make that distinction clear. Okay? So, I think two things about this in setup, right? We have a centralized contract, right? that applies to every single worker if in this case agents across the entire country with decentralized effort  
[05:43:22] decisions that are made you know at the local level. Okay, that's one key thing.  
[05:43:26] The second thing is that we're going to be looking at the status quo contract which is very popular and widespread across you know this industry perhaps not only in mobile money or digital financial markets but in a bunch of other industries where peace rates are very popular. Okay, so that's what I want you to have in mind in terms of  
[05:43:40] what is distinct about a service. Okay, data data too much data and here's why actually this is actually good because you know we in the world where funding is limited and I think part of this relationships really has created an  
[05:43:53] opportunity where you can get remarkable data at a very low cost right part of the operational activities to track some of this data that we actually going to look at in a way that I don't have to go to the field to collect data so I think this in some sense is a blessing in  
[05:44:06] disguise okay we're going to do a baseline survey close about 6,000 agents across the country um across roughly 700 communities and I want you to think of this community as somewhere you know all over the country okay national you know just give a sense of where we are  
[05:44:21] where's the clicker okay so this is in our experiment these are the guys that we are going to be working with all over the country every door is basically a location for an agent okay that's what the baseline  
[05:44:33] service is going to be on um and I'll get to some some more details um later on in fact MTN actually has a slogan They call them everywhere you go. You can tell why that is, right? It tells you they're everywhere in the country because they have a footprint through the agents everywhere in the country.  
[05:44:47] Okay. So, we do a survey of these. We do a survey of these agents.  
[05:44:55] We ended too fast. Okay. Then we're going to do a version with managers.  
[05:44:59] Okay. In this case, managers will be everyone on that vertical chain. Okay.  
[05:45:03] Close about 500 of those um across all the regions. The country has 16 regions in the country. We're going to do our best to almost talk to everyone, okay?  
[05:45:11] In in the firm. Then we're going to combine this with this. This is the core of the data, the randomiz the RCT, the fluid experimental data. It's going to be an admin data that's going to give us every single transaction record that happens at every single agent point,  
[05:45:25] okay, in the country. Weekly payments, weekly compensations, investments that goes to the agent, you know, in the experiment and all of that, okay, across nine out of the 16 regions. So the experiment is of is quasi experimental.  
[05:45:36] the baselines you know sorry quasi national but you know the baseline service are national please >> yes so so okay two things two things one is that the  
[05:45:57] guy at the top makes the decision makes the call but they aggregate views from the lower guys okay they aggregate views and you know feedback from the guys who are in the failed which is what the second bullet is trying to get there.  
[05:46:07] That's number one. Number two is that um in fact when it comes to implementation of the scheme is the lower guy who does implementation. Okay. So so those are going to become very central. Okay.  
[05:46:18] So RC we're going to have this admin data. Okay. Uh that's going to allow us track actually in this case you know performance uh and you know dropouts and all of that. Um throughout the experiment we were at some point worried that I was Francis was going to come in and say hey you on a simple linear contract I'm going to switch you to a  
[05:46:33] tournament for example. And so to some extent agents might engage in some illicit activity for example imposing illegal markups which we know is very popular in this market and all of that right and fees. So we actually conducted a nationwide audit study of the agents that were in our sample right where we  
[05:46:48] randomly actually send consumers to go out there and also collect data to see what happens at the agent point with a lot of details on compliance and consumer protection. Well I was not satisfied enough. I still went ahead and did an endline survey. Okay. So I went back you know talked to agents and  
[05:47:01] managers at endline to get you know as some of the mechanisms. So this is the data that we're going to be working with. Each data has a distinct role to play. The design experiment is going to be market level across close to 425 communities in nine regions. They're  
[05:47:15] going to be five treatments. These treatments are going to be the the contracts in this case. And I'll bear with me just for two slides. You see the contracts. Um the way we do at least we make two innovations in the way we set up the paper. One is try to set up design contracts be expenditure  
[05:47:29] equivalent xante. Of course I cannot guarantee expenditure equivalent exposed. Okay. So that's one thing that we do better. We were very careful in the way we actually elicit the managerial beliefs which is finding a way to link you know the elicitation of the beliefs with the randomization of  
[05:47:43] the experiment itself. And I'll get to that. Oh actually that's so we stratify our treatment you know by three things baseline productivity of the agents um number of agents in a given market and some commercial zones that this firm has. So together instead of  
[05:47:58] compatibility when it comes to this elicitation of course contracts that that actually were ranked higher by managers actually were assigned the latter shelf of you know randomization.  
[05:48:07] So we finding a way actually to make you know to actually link randomization to kind of the elicitation in this precise way. Um the last thing is that the firm could only allow me three months [gasps] to run the experiment to mess up with the system. Okay at this skill. So we're  
[05:48:20] going to have 3 months of full experimental you know uh period where we're going to have weekly conversation that takes us to roughly to 12 different you know payments. So if you are someone who gets annual payments this is like almost a whole year. Okay so that's where we are going. In fact at some  
[05:48:34] point the f was like why don't you take this contract out it's not working. So why are you keeping it? I said you know just wait okay just wait just wait.  
[05:48:42] All right so that's that's where you're going. Um so a few things before I show you the contract. The next slide I promise you is the contracts. We're going to make weekly we made we made a a few trade-offs in terms of the way the experiment is set up. Okay. One is that we are going to make weekly payments to  
[05:48:56] these workers or agents for all transaction type whether withdrawals or deposits. In the status quo the agent receives an instantaneous payment the peace rate for transactions that are about withdrawals but they receive  
[05:49:09] monthly payments okay or commissions on transactions that are about deposits.  
[05:49:14] Okay. So we strike a good balance between you know how fast and how slow and we do just weekly for everything we manage to set up that system with the firm. So that's one change that we do key thing at any point in time agents could exit actually they could leave the  
[05:49:27] experiment and when they leave leaving means you revert back to this status quo we just talked about. Okay so that's going to be another way to think about it. In fact when you leave we can observe you right because you are still with the same firm which is going to become very key because then we can  
[05:49:41] actually track performance dropout everything really about the agent even condition on dropout okay that's going to become very important so the key observation as I mentioned is basically we measure almost everything whether you're in the experiment or outside the experiment so don't worry about  
[05:49:54] selection and all the IT stuff that typically you know okay here are the five contracts so the first one is the status code by the way we did a lot of piloting uh before we came down to this five contracts. Okay. Um we  
[05:50:08] started with about eight or nine you know it was operationally invisible to do some of them especially the nonlinear ones. Okay. And so we came down to these ones which we think actually very common and popular. So first one is a linear basically which we have talked about it  
[05:50:22] just you know get some commission W right on every output basically it's per output commission that you get. Okay. We also have a threshold contract which is one of the innovations. It's used a lot in marketing. The threshold is you receive this you know large payment over  
[05:50:36] and above your peace rate if you hit some predetermined threshold okay which is set by the firm. So the key point is how do we set this threshold t if you don't hit a threshold in that week you get nothing okay that's what this is you go home without any pay  
[05:50:50] the second innovation we have is a franchising people have talked about franchising today so here's some example of a franchising okay so and we think this is a future for the industry by the way at some point MTN do not want to handhold the agent right they want to say look pay me a platform fee and then  
[05:51:04] just keep all the commissions right and that's basically think of Amazon and all other platforms really how most of these is run. So we implemented a franchising.  
[05:51:13] What's a franchising? You get a boost in your pay but you pay some upfront fee which is basically a platform fee right in response to that. So the key thing is how do we set K and Z and we'll get to that. The fourth one is a tournament.  
[05:51:24] Again we are moving away from individual contracts to group based contracts. Okay a tournament. A tournament basically these are basically linear rank tournament. Every week agents in a given markets are all ranked. Okay. Of course the top guy gets a boost in their pay.  
[05:51:39] The total pay is the same okay relative to the the first contract if you want the simple linear except that the different ranks reallocates the money in some way. That's what we do with this.  
[05:51:48] I'll get to more details on this. The last one I've been curious about incentives. Economists have spent a lot of time talking about incentives. So we have a kind of incentives. So here's where we presume that MTN makes the agents actual workers. So actually pro  
[05:52:03] promise them some fixed pay regardless of the output or effort. Okay, that's what that's going to be. This is a particular one they wanted us to take out okay at some point but you know there was a very interesting story about that and maybe  
[05:52:18] if I have more time I'll share okay okay so these are the contracts so I say contracts this is where we are going so now let me walk you through how we make them expenditure equivalent okay that's the key part designing contracts very simple framework think of it as an  
[05:52:31] accounting exercise okay we have an agent okay who exists some effort ei generates some kind noisy output. Output here is those withdrawals and you know deposits. Okay. OI. Now um of course we have five contracts as we have just  
[05:52:46] seen. Okay. What contracts is basically mapping your you know your output to some kind of compensation structure.  
[05:52:52] Okay. To the agent. The utility is going to take at least two argument three arguments. You have basically you know of course you want more pay right that's the first part. Um you also get some intrinsic value right from working even if you don't have any pay. was the  
[05:53:06] second component and there's some cost of effort. Okay. Uh you could presume for now I'm just going to allow it to be very general. As I said it's going to be a very general framework. Okay. The typical IC and participation constraints all apply here. I'll not bore you with the details. As you can see this is a  
[05:53:21] word of commissions. So the firm ends some Z share of the output right that's what the firm gets and then you know whatever they get either they split it with the agent you know in some particular way. Our benchmark contract is going to be the simple linear right where we started which is the start the  
[05:53:35] simple linear contract. And let's call the output from the simple linear contract ODI. I will use this a lot. So please let's remind ourselves ODI. Okay.  
[05:53:44] O um you know subdi. So I'm going to move the key idea. What I'm going to walk you through is basically is that I'm going to try to start with a world where I make them on average equivalent right in expenditures to the firm. Okay. So then the only thing we might care a lot about is just  
[05:53:58] the fact that incentives are different or maybe the risks of the participation risks are different for different workers. So that I can get at the IC and the participation. That's where I'm going with all of this. Okay.  
[05:54:08] Okay. Simple linear. >> Yep.  
[05:54:20] >> That was one thing I worried about. So this was actually a motivation for why we wanted to start with expenditure equivalent. I don't want to lose money through this thing. So I said, well maybe I can make things equal to start with and then we see how the world is.  
[05:54:32] So one argument for exponential equivalent was to convince the firm to do this actually.  
[05:54:37] Of >> of course exposed we cannot guarantee that that I agree. Okay. All right.  
[05:54:45] Okay. But that was actually something that came out with a board meeting on this.  
[05:54:50] So we have you know simple and very simple you get some you know W share of the OI. Okay. And we know this it's about 0.4% 4% of of the output constant marginal incentives right is basically the industry standards are very popular used a lot in a lot of markets perhaps  
[05:55:04] because of the simplicity right of this and and a bunch of other reasons it's also going to do another job which will be outside option right because once you leave or experiment you default back to this particular thing okay now let me tell you how I make it  
[05:55:17] expenditure I take this as the benchmark and the normal let me introduce the nonlinear you know kind of the threshold for example and walk you through how I make it equivalent maybe I'll just do 10 or two and then I'll skip the rest because it's going to be the same exercise intuitively. So you get a WT  
[05:55:32] right the threshold some commission it's going to be a large payment capital B if you hit a particular threshold right in the particular week okay for that agent otherwise you go home hungry right you know to some or you get nothing okay maybe or you default back to the rice  
[05:55:45] that you sell okay so we're going to set a target there are so many ways of setting a target the most easily intuitive way to the frame was just to look at your previous say record 6 months or three months and set your target to be your past average okay so that's the  
[05:56:00] I mean in principle you could imagine setting a higher target or lower target and then vary the probability right to get an extreme right but right this is our target so how do we calibrate this we're going to set you know the the B such that the probability that you hit a target is equalized the previous simple  
[05:56:15] linear right the expenditure in the simple linear is what is W * the expectation ODI okay so all I'm doing on this part of the slide essentially is basically trying to estimate B right with some probability  
[05:56:29] that's you you're going to hit the threshold or not. Ensure that the equivalent place.  
[05:56:34] >> Yeah. >> Yep.  
[05:56:49] >> It's simple. Yeah. >> Yep. Yep.  
[05:56:55] >> Yep. Yep. >> Then you have these three other guys. Um is the way to understand like there's a side of stuff and that some stuff is good here some stuff bad people to know that or is that  
[05:57:08] our economic environment is one in which this is sort of in some deep sense right should be the best it should definitely not ever >> it's true maybe  
[05:57:23] >> can can I return to the philosophy of this once I show the results okay I have I have it's it's a very deep story perhaps it combines some of the insights you have along with something else that I think is very key when we think of contract theory more broadly and I'll  
[05:57:36] get to that u sorry sorry I didn't answer your question but I'll come to it yeah please have in the back of their mind that like I could be fired if I really do no work during this period I mean was there any chance of them being laid off if they  
[05:57:50] like revealed themselves as like people >> not as far as I can tell yeah this is a free market you can leave and exit anytime time and you know um really it's that's why it's performance contracts, right? It's all about you get to pay for whatever you do and but at some point if  
[05:58:05] you become inactive for a long time and the firm feels that you're a redundant, you know, at some point they're going to say, "Hey, we think we're going to take the license away from you to be an agent." But please [laughter] do this exercise. The whole point is to  
[05:58:24] estimate. So as you >> well I'm just like I said I'm doing maybe the 18th 19th century macro calibration for a minute.  
[05:58:31] >> Okay which is which is I have your history as a I'm just think >> yeah I know that I know I know all of that I you know I mean if to the extent that elasticity is infed in the previous historical data you know >> that's exactly that's all I'm doing  
[05:58:49] nothing nothing more than that. So when I say for example the target being expected ODI think of it as your previous all your previous transaction history which embeds you know what you know if you're in a simple linear whatever that is >> if you're very yeah well so so okay let  
[05:59:06] me get to the calibration exercise if it doesn't answer your question I'll get to that okay so immediately you can tell right that quickly um with some pre-pilot which we did with about 100 sample which is not a part of our experiment I can randomize the TE's and get at this particular you know  
[05:59:21] probability of hitting the threshold I use that to calibrate this okay so that's what we do the last the last bullet on the on the bottom okay so it turns out the B comes to roughly about 1.5 times your peace rate okay if you hit it you can do the same exercise with  
[05:59:34] franchising again I want to pick some K some platform fee such that I equalize expenditures okay that's again what I'm doing here remember the firm gets Z right share of of the ODI so all I'm doing I'm picking my K so I equalize expenditures that's again what I'm doing  
[05:59:48] here I go through the next one This is tournament maybe um you know one of the the extremes of of the philosophy. So tournaments of course may work under some conditions versus not. We were very careful with the way we set up tournaments. Okay. We have almost every  
[06:00:01] single data on all your performance records. Of course if you set up tournament groups where really really top performers are there with very weak performance is never going to work.  
[06:00:10] Right? So again to the extent that we promote a lot of homogeneity in terms of you know how similar or dissimilar okay agents might be based on their historical record. Okay. So tournament the total pay in a given market or tournament group is call it big P. Okay, it aggregates all the pay that goes in  
[06:00:25] the simple linear. All I'm doing is each week I rank the agents and all I do is I just really allow the ranks to determine how much pay you get. That's all I'm doing. So the best guy gets N times what the worst guy is paid. The next guy gets N minus one times you know the the best  
[06:00:38] guys paid. That's how Yeah. is everything is going to be for three months. Yeah. Yeah. Yeah. Yes.  
[06:00:45] Okay. The last one, incentives. Maybe I'll skip this because it's very trivial.  
[06:00:52] So, let me mention a few things and maybe I'll give you a graphical version of this. I don't know how much how well I'm doing on time, but essentially what we are doing here is having equal pay on average xante allow incentives to differ. Maybe the risk of participation  
[06:01:06] and all of that to be different. That's all. And then I'm going to basically track these things as what I care about in contracts. That's all. That's all I'm going to be doing. There's a graphical version. I'll skip it.  
[06:01:17] Please >> agents don't move. By definition, a market are special markets. So you're an  
[06:01:32] agent. If this is a village, you're agent within the village. So um Oh, if you leave Yeah. So in the way we set it up in you're going to generate more more dropout than it would have been in reality because in reality when  
[06:01:46] you leave the firm what happens you know you leave forever right uh so so think of this as maybe the upper you know an overestimation of what the actual participation rate look like in the way that we set it up because when you leave say I don't want to be part of this experiment you put me in threshold I've  
[06:02:00] tried 3 months or two months I'm not happy take me out I take you back to the default are we there no update They don't it's not like an exit in traditional sense that you know you quit a firm and you leave the firm.  
[06:02:14] You're still with the firm except that you're in a very different arrangement.  
[06:02:16] Okay. I think Oh, go ahead. Yep.  
[06:02:29] >> Yeah. Mhm. >> Yeah. Yeah.  
[06:02:39] remember yeah so that is that's a very great question except that I had told you that look at the last point even when you quit when you leave I observe you no there's nothing like new entrance in  
[06:02:52] our context >> so so let me tell you what I ask managers and And that's what I call beliefs. That's what end the title for the paper. I asked a lot of things about  
[06:03:07] managers. So let's get to the beliefs part. And if it doesn't address that, you tell me. Okay. This is the part that actually ends the title for the paper.  
[06:03:15] So I'm going to I call it beliefs over contracts. So all I'm doing is I'm going to present to managers these contracts in a way that we have talked about them and they basically elicit their beliefs over there. Okay. In some incentive comparable way. First elicitation pre-experiment. Everything is baseline.  
[06:03:28] Okay. Managers are going to be asked to rank these five contracts. Okay. Having done a great job, I can promise you I did a great job explaining to them they understand. They know what's going on.  
[06:03:38] And I I have a slide down there if you want. We can spend the whole time talking about it, but I don't think it's a better use of time. I promise you they understand. Um managers are going to rank these contracts on the basis of the ones they think actually would actually max actually maximize revenues to the  
[06:03:51] firm. And remember Xante, if you just believe in just Xante sense, they almost like profit maximization. Okay, because of the expenditure equivalence. This is what we find. Okay, the threshold is predicted to be the one that will generate the highest revenue to the  
[06:04:05] firm. Whereas the flat which sure all of us will do high schools probably you know people who have never seen scientists will predict you know the flat to be the worst and that's actually what you know we find a lot of confidence really they understand really we went through a lot of process we the  
[06:04:20] enumerator the field officers will go through the schemes one by one both in words visuals with graphs and then managers are asked repeat all of that to them you know they ask a lot of questions we go through really it's a it's almost like a whole like if you  
[06:04:33] want five hour type exercise you have to schedule meetings over and over to do this. Okay, so that's where we so this is what managers think. If they were to make a decision on the basis of this, you know, this is how they would have ranked things. Okay, please have this in mind. Second question, managers then  
[06:04:48] rank things on the basis of which one will generate the highest participation risk. Okay, the fact that people may drop out, you know, if they're not happy and all of that. Okay, going to this. It turns out of course managers think the simple linear will have the lowest okay  
[06:05:00] risk which is the status quo. But of course, the threshold will also have the highest risk. So the thing of course the threshold is the one that brings more money but it has a lot of risk. Okay, that's the way to think about this and then tournament bear in mind tournament franchising and and if you want you know  
[06:05:14] are not that different right that's how how they think of this okay second elicitation third question managers then are asked forget about the firm okay forget about MTN for a minute if you  
[06:05:27] were to make a choice okay about which contract you choose how would you which contract would you choose so that's why I'm going so again I'm not imposing any objective function okay in terms of what they're doing this is what they tell us it turns out the most their most  
[06:05:41] preferred contract in many many instances differs from what they believe will maximize profits to the firm in this case revenues to the firm. So the key question is what exactly are managers maximizing and I don't know I'm not going to be tell I think this is cool to show but about almost 40% of the  
[06:05:56] time there's departure from what they think will bring more money versus what they would have chosen. Okay I'll leave it there if you know you have some more thoughts please.  
[06:06:05] >> You said you thought the cost would be the same. Yep.  
[06:06:09] >> They believe that >> they do. Well, that's why I said I believe that I I they do. Okay. That's part of [laughter] part of all the exercise we do here.  
[06:06:20] >> Yeah. >> I I so we worked a lot with the firm itself. Um actually there all these full supervisors were part of the process.  
[06:06:30] Actually the firm actually backstory the firm actually had to set up a whole research team. Okay. a house at the HQ where we actually recruit actually have to hire like new master student in economics to become a staff hired at a firm for a year with five or six other  
[06:06:44] staff within the firm to actually get this to work. So really this was something that was serious uh with with all you know intentions you know on the deck. Okay. Okay. I guess these details are not necessary for for this talk but I can assure you that yeah I think to  
[06:06:56] the extent that managers believe >> okay no I mean and that's fine look there may be that. So let let's look at the results. Okay let's look at we can interpret them. Okay I'll spare you this. This is you know this world where you were to ask the manager you know  
[06:07:11] just describe for me what the optimal contract will look like. At some point I know when I submit this paper someone's going to ask me oh but you look at five contracts what the optimal contract right and so I may have an answer for you okay when you ask me that question which is we think we are within the space of what is optimal in the mind of  
[06:07:26] the manager okay that's what this is trying to get that maybe last question and then I see a hand up Christina then I'll do that so managers are then asked to state reasons for why they rank things the way they did okay this is going to become one of  
[06:07:39] the key things three things are obvious one is that things like risk which features a lot in our models is barely mentioned but incentives they think oh this one has ability to get people okay is the first one then the notion of simplicity  
[06:07:54] Aron this is the idea that they think of course I leave something simple okay to be able to do so it turns out there's a heristic some rule of thumb that they apply a lot which I'm going to argue is one of the sources for the misperceptions they may have okay regarding why managers may not predict  
[06:08:07] the right contract has been a key source of you know the inefficiency that we we elicit here so so Again have this in mind please and I'll tell you the three main effects so far is descriptive so don't hold me too much accountable to everything but go ahead yes the rest of the network  
[06:08:37] >> yeah I mean I don't think for now everything is think of it as markets are different they generate different revenues And you know and by the way this is like about 5% or less than 10% of what a manager's profile is like the manager a typical manager manages like  
[06:08:50] hundreds or so there only two guys that we picked for the experiment and so the notion of externalities which I think is what you have in mind to the extent that we can measure it I don't think it's it's the principal reason for why they did this >> Jim HQ should care about this but the  
[06:09:04] people asking yeah >> I have oh Christina just to be respectful Yeah, we do two things. One is that you know they are told the firm told them like you know this is a serious exercise. We are looking to change your  
[06:09:18] conversation structure. It was you know the firm made it clear right that their decisions are consequential. Okay, that's one.  
[06:09:30] Oh, so if you are someone who is a a lower level manager in the way that we classify it, you go around the agents, ensure that you know they their pin is not locked when someone shows up to do business, you know, operational things like that. Um, somebody may need  
[06:09:45] liquidity. Going back to your question about liquidity, maybe you can provide liquidity support in a way to rebalance the account because this is about, you know, deposit and withdrawals. It kind of goes on on and on. It may be some MTN sticker, some paraphernalia, whatever it  
[06:09:58] is, just you know all the typical things that a lot of these but let me move forward a little bit. I know I'm happy to talk more after this talk. Um treatment effects. Um I'm not going to do what everyone has done. I'll only show you graphs, nothing. I'm no regression. Okay? So that you know you  
[06:10:12] make the inferences. So that's my goal. If you want to see the regressions, I'll click the button. But everything hopefully the descriptives will be clear, okay? When I show you that. So I'm going to start with participation.  
[06:10:22] Whatever I call participation in my world. If you're not happy with the terminology, think of it as dropouts, okay? Or retention, how long they stayed in the contract. On average, tournament generated a high sorry um threshold generated the highest dropout. Okay,  
[06:10:36] Keith, that's the first highlight. The second highlight is that tournament dropout is not that different from the simple linear dropout. That's the second thing that I want you to take away from this slide. Second result, most dropouts are very quick and early. Okay, if you  
[06:10:50] want to drop out, you want to drop out very quickly. And that's what we see even you know for across all the contracts. Okay. So over time you see massive convergence in dropout rates during the experimental period. Second result I'm done with dropouts. I've  
[06:11:04] shown you the average effects and the dynamics. So I'm going to talk about revenues. Okay. Which one generates the highest performance? I'll focus on total performance. I can decompose this into one that goes to the firm and one that goes to the agent. I'll show you the  
[06:11:17] total. Overall your intuition is right that on average linear is not the best when it comes to raising more revenue to the firm. In fact is the second best tournament in this case is the best contract. Okay.  
[06:11:30] Okay. Despite it not being predicted as will be the one that will generate the highest. Okay. And I'll be going there.  
[06:11:34] Remember managers predicted threshold right to be the one that generate the highest uh performance. Okay. So that's the first result on average effects on performance dynamics. You see that there's a little bit of learning, right?  
[06:11:47] Tournament was not the the best one for the first week. Go ahead.  
[06:11:54] Uh think about it, but like you know a lot to say about like when >> previous  
[06:12:07] like what what is the >> what would theory say about like when tournaments would be better? contract and are we in that world or not in that world?  
[06:12:17] >> I think the hetrogenity I'll show you will speak to some of the what theory says. Yeah, hetrogenity. I'll get us some of that. Yeah, it speaks to some of that. Yeah, if a tournament works better if agents are very there's less volatility, right, in in things, you know, we know that we find evidence of  
[06:12:31] that. Um >> lot of like unobservedly specific.  
[06:12:36] >> Yep. Yep. That's right. Yep. Yep. >> Comic shocks. Yeah. So maybe I'll speak to someone that have heterogeneity slide and I think that's consistent what theory says okay for specific contracts but even all of that the average you know I like all these slicing the data  
[06:12:50] into different things but I think if as a firm right I'm looking to make a decision about where to make money you know maybe it's more of that both in theory but also in practice is important dynamics clearly you can see tournament took over the second week and then that was it okay became so again I don't know  
[06:13:05] how long this will go if you have to run this up to today but I can assure you I think tournament will still doing better because I'll show you some labor supply results. If you think this is coming from the effort stuff, I have a lot of data up to today on effort that I can actually show you and actually is very  
[06:13:18] consistent with this. Okay. Okay. So, I'm done with participation and all of that. Here is some hetrogenity stuff.  
[06:13:25] Okay. So it turns out if the owner of the store the agent is actually the operator and not the delegated person who is working on behalf then actually the these negative effects on on flat is actually  
[06:13:39] much worse. Actually all the flat wage results are coming from when the owner has a lot of flexibility to not to show up to work. You know if it's a worker the worker will show up but if it's the owner who's doubling as the operator then he decides not to show up. So again  
[06:13:52] it's very consistent with the idea that uh flexibility in terms of you know so the first bullet point actually owner operator run shops drives the effect on on the negative effect >> worse like but it's it's not worse  
[06:14:11] >> why like basically just like >> I think I think there's a lot of variance yep yep so so two things statistically franchise is not different from the simple linear Okay, if it's not look at the last bullet I didn't mention that if  
[06:14:25] I were to estimate treatment effects on there's a lot of variance actually a lot of the difference between the simple linear and and the franchising is a lot of hetrogenity that is driving if I estimate regression results now I don't see any differences by that so you're  
[06:14:39] right in fact my priority to this was franchising should be the thing right like I said is a f I think is a future for this market especially if the platform doesn't want to handhold agents at the downstream but It's I mean of course if you have risk  
[06:14:54] then franchising breaks down. If you have risk that's another another prediction. If you have risk then franchising bros. Yeah.  
[06:15:02] >> No >> contire.  
[06:15:14] >> Okay. But but I think in theory though if you have risk franchising thing breaks down very very easily. Okay. Um I'll spare you because I don't have all the time. Let me show you more results.  
[06:15:25] Effort. I'm going to look at effort in very precise way including things like liquidity and all of that that they invested in. But I'll focus on one. I'm going to construct this notion of labor supply as one measure of effort. Okay. I observe across all the contracts the number of days they work in a given  
[06:15:40] week, the number of hours they work in a given day. um whether they open their store early or they close their store late. And I'm going to use all these different dimensions to come up with some notion of some index. Okay, call it some labor supply index. Um and we're going to find evidence that in fact both the threshold and the tournament will  
[06:15:54] generate consistent increases in effort relative to the simple linear. That's all I'm going to show you. The evidence is there both in the survey data but also in the admin data. Okay. So the first one here is just the survey data and line survey data. You see that very  
[06:16:08] consistently. But let me show you the fun part. the admin data here is it this is actually coming from the platform like I said I could keep expanding this graph you know post experiment and we can learn more about you know these changes in labor supply of course there's a little bit of some pre-trend  
[06:16:21] before the intervention but what you see is that just showing the three different you know classic ones right this is simple linear the flat wage and then the tournament you see that of course in the course of experiment people you know the index actually went up and in fact it went up in a very precise even post  
[06:16:35] experiment it kept staying higher in fact all other contracts converged to the linear. Okay. After the the interventions, you know, the incentives were removed, which kind of opens this, if you want Pandora box on whether actually competition through tournaments could be kind of some micro foundations  
[06:16:50] for the way we think of, you know, persistence in labor supply, please.  
[06:16:56] Yeah. There but but remember the threshold.  
[06:17:02] Yeah. >> Yep. I agree. Yeah. Actually, we also measure the risk preferences of all the agents and all of that. We haven't used that a lot, but you know, um that's something that we can do more easily. Um  
[06:17:16] I'll skip we don't find much stuff on selection. The Christina story that came in selection to contracts or exit remember you know if you were an agent actually we also listed agents baseline whether you know which contracts they select into for example and then we randomize them. So I could test for  
[06:17:31] actually selection right if you are assigned to a contract that would have preferred to begin with and we don't see much happening there. Okay. Um okay the last part then I'll conclude managerial predictability I'll spend only one slide on that okay how predictive are managers that's where I'm  
[06:17:45] going what I'm going to do is just basically show you simple correlations okay of course putting all together I'll focus on performance you can do the same for dropouts or participation of course managers predicted threshold to be the  
[06:17:58] best we found tournament to be the best okay on almost all the different metrices they predicted flat bonus to be the worst or flat wage and that's indeed what we find okay in our experiment if you do a just simple correlation or run regression on one against the other you  
[06:18:12] get some coefficients in the within the magnitudes of 0.2 too. Okay, which is quite modest. Um, and I'm going to kind of if you want I think you know part of what is very key here is that this heristic the fact that managers are looking for something very simple  
[06:18:26] actually is part of the story for why again they may gravitate towards one particular contract. I see Andy is shaking his head. He doesn't like that argument but we can talk about that.  
[06:18:34] Okay. Now let me be more statistical. I'm going to create this what I call some notion of accuracy or inaccuracy called a prediction predictive error. So for each manager and contract pair, I'm going to define whether they rank right the rankings are similar, right? And then I'm going to run that as an outcome  
[06:18:48] against a bunch of things that I care about managers and contracts and all of that. That's what I'm doing on this. I only highlight two things. The first part is basically relative to the simple linear we see that things that are a little bit complex like threshold and what um the tournament seems to have the  
[06:19:02] highest you know predictive you know errors right from managers and we see that both regardless of whether you're looking at the IC constraint or the participation constraint. So which is kind of leading me to you know make this if you want uncomfortable argument that seems complexity of the contract  
[06:19:16] actually implies you know more inaccuracy. Okay. The second one which I'm more com comfortable with is basically doing a version where again I compare junior level managers to senior level executives or whatever however you want to call it. Okay. And we see very consistently on the on the IC part on  
[06:19:31] the performance part senior level managers are very very good right compared to the junior level guys. Okay.  
[06:19:36] Because the predictive errors are are lower. Okay. But then you see the opposite or the converse on the on the dropout or if you want the IC the IR part. Okay, which actually is very suggestive that I call this the ivory tower trade-off which is you may sit at  
[06:19:49] the HQ very good at doing the math the mathematics of figuring out you know how to incentivize agents because you have access to data you can look at markets from the sky and all of that. So very good at actually inferring what contra might do well with uh performance if you want revenues in this case but maybe  
[06:20:03] less to do with which agent to accept or drop out because you know the conditions locally are very different. Okay, so that's my interpretation. Let me conclude. That's all I have to tell you.  
[06:20:11] Remember I promised you four results. I've shown you four results. So let me conclude with two things. Contract design matter. Okay. It turns out at least in our context tournament like contracts seems to increase revenues are roughly 20%. Okay. Relative to the status quo. Managers how you know  
[06:20:25] miscalibrate you know the relative efficacy of these different contracts.  
[06:20:29] In particular, managers are very good at actually figuring out worse contracts or bad contracts but less of you know the optimal contract in a sense which will imply some kind of cognitive biases in the way we think of managerial decisions over contracts. Three implications and  
[06:20:43] maybe the philosophy can come in here. I have one minute. It turns out again managerial misperceptions may be a key source of inefficiency in the way we think of contracts more broadly. Okay.  
[06:20:52] Now and I think this is interesting because it challenges a very fundamental assumption that we make in a lot of models of contracts. you know it's not about physibility of a contract but it's about actually within a set of physical set of contracts which contract would I want to choose okay and I think we find  
[06:21:05] evidence that you know the principle may not be as sophisticated that we think principles are in our models more generally right and this is very kind of tangential to what the behavior I literature has focused on where firms know a lot than consumers and therefore they design exploitative contracts to exploit consumers because of you know  
[06:21:19] their biases well here the manager the firm themselves is not sophisticated so what do you do right we're in trouble and I think you know which obviously leads to the last point the fact that we want to think more carefully about you know kind of if you want cognitive biases when you think of principal agent  
[06:21:34] models and really how that allows us to think of performance more broadly I have a lot of literature I won't go through that Christina's paper is cited so you're right uh I'll stop here but thanks a lot Okay.