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.
=

# Liquidity Constraints and Capital Allocation in Agriculture: Theory and Evidence from Uganda

Authors:   
Discussant: None  
Video: https://www.youtube.com/watch?v=PHYYOyrSItw&t=21279s


## Talk (05:54:39 – 06:48:40)

[05:54:39] Thank you. Max.  
[05:56:14] Yeah. Well, Good job.  
[05:56:56] Everyone get  
[05:57:34] There's the You're  
[05:58:07] >> terrible. So we have one last session.  
[05:58:23] >> All yours. >> Okay. Well, excellent. Thanks so much uh for including the paper on the program.  
[05:58:30] Um this is joint work with Jonathan Dquit who's at Queen Mary in London, Benedict Alva who works for the World Bank is and is here and Stephanoi who uh is working in the private sector now. So  
[05:58:44] what's the kind of background u for our interest here? So the the broad background is that oops this why is this not Okay. Uh so the broad background is um  
[05:59:04] this observation that uh Africa hasn't really seen much of a green revolution uh yet. So if you look at different measures of output agricultural yields across the world, you see that they have  
[05:59:19] increased in much of the developing world um often by multiffold. um while they have more or less stagnated in Africa and together with that comes the observation that many of these other  
[05:59:32] developing regions in the world have started to adopt modern inputs such as fertilizers and seeds and uh again that hasn't happened much uh in Africa and so of course people have been thinking a lot about this there are kind of entire  
[05:59:46] research agendas and programs uh dedicated to understanding this um uh both in terms of how the green revolution unfolded in the rest of the developing world and the kind of constraints that farmers face especially  
[06:00:00] in Africa um to adopt modern uh inputs right and there's been a lot of work on trying to understand the role of credit constraints insurance constraints behavioral constraints information constraints and so on and so forth and  
[06:00:14] so what we are trying to do is trying to understand in that setting and in the face of all these constraints how is the market for these inputs actually functioning. Okay. And so what we'll do  
[06:00:28] in particular is uh three things. So first thing is to ask how well are fertilizer markets functioning uh in the status quo. And so what we'll find is that there's a group of people who  
[06:00:41] actually has high returns to fertilizer that could be induced to adopt these uh inputs by a relatively modest uh subsidy. um we're trying to calculate the optimal subsidy and find that to be  
[06:00:56] uh 30%. Now, you know, I don't want to uh necessarily uh stick my hat out for that exact number. You'll see that these things are estimated with a fair amount of uncertainty. Um but but that's our  
[06:01:09] our best guess. Now if you think about alternative subsidies in particular a full subsidy so handing out fertilizer for free um uh we find that that would  
[06:01:22] not generate uh average profits that would uh uh that would justify such a program for sure the average profit impacts of of that are well below the market price. Now I think this um kind of connects to of course the large  
[06:01:37] literature sort of some of the closest connection is the long um standing debate about whether we should subsidize these uh kind of inputs. Um there is also a fair amount of work trying to estimate what are the average impacts of  
[06:01:51] these um of these inputs. Um with you know I would say mixed findings. Um then I think there is the kind of important work by Tavitsuri highlighting that  
[06:02:04] really we should be thinking about these inputs having very heterogeneous returns. Um and that's going to be key for what we are doing. Um and there's a another paper by Mai Mammud who does uh who's interested in a very similar  
[06:02:18] question I think and um uses a different methodology also um finds a different uh conclusions um but it's a it's a great uh paper that's closely connected to  
[06:02:31] what we're doing here. Um so that's the first part. Then the second part is the question whether we can actually address uh these distortions uh that generate this motive for the subsidy at the source. And so what we'll be doing is  
[06:02:45] we'll go after one particular distortion liquidity constraints. We'll be relaxing those by handing out cash transfers. Um and we'll find that these cash transfers actually increase adoption. they  
[06:02:57] increase um uh profits and importantly that uh we find that after uh cash transfers are handed out um the optimal subsidy now drops to zero right and so  
[06:03:11] we find that uh you know that uh also connects to a large literature some of the closest connections are with the work of uh uh Colin in uh in Ghana and  
[06:03:23] Dean and Corus in in Mali I think methodologically it's very closely connected to the work of Swan and uh Josh you'll hear more hear more about related work um right after this and  
[06:03:37] then uh you know in the kind of broader scheme of things we feel this is related to these old ideas by Lip and Lancaster on the theory of of the second best. So their you know idea is that uh in the  
[06:03:50] presence of distortions in some markets um really all first best conditions might no longer be um optimal. But you know of course when you remove these distortions then the optimality of this  
[06:04:04] first best first order uh conditions uh is being restored. And then in the last part what we are trying to do and that's uh probably a bit more uh speculative is to ask well what would a policy maker do  
[06:04:18] here? uh if a policy maker can either try to correct the distortions with a subsidy or can try to address the distortions at the source by u you know whichever way um what would the policy  
[06:04:32] maker want to do and you know the conventional wisdom of course in economics is that you should address the distortion at the source but the problem is that many of these distortions exist for good reasons and and addressing them is actually uh costly and so um we'll be  
[06:04:47] thinking a little bit about uh that trade-off and and and what the optimal solution might be. And it turns out that it will the answer will depend on the government's um opportunity cost of funds. Particularly we find that governments which have a very high  
[06:05:01] opportunity cost of funds might find it optimal not to dis uh address distortions at the source but live with them and try to correct them with a with a subsidy.  
[06:05:12] Okay. Um so I'll talk about a very simple theory theory uh truly very very simple um uh uh that uh you know I think you can think about ways of making it  
[06:05:25] more complex uh it's a little harder to think of ways of making it more simple um this [clears throat] but you'll see it is is exactly motivating our uh experimental  
[06:05:39] design and highlights number of population moments that our experimental design is trying to um allow us to estimate. We'll talk about that estimation. Then uh I'll talk about  
[06:05:53] the setting uh and and related issues and then talk about the results and in the very end hopefully uh get to talk about this question of what the government would want to choose amongst these different options.  
[06:06:06] Okay. So here's the simple theory. Think of uh uh an investment good that yields a heterogeneous return theta I. Okay. So that's the increase in profits that your experiments experience by adopting this investment good. And so the question is  
[06:06:21] you know how do we ensure that that investment uh good ends up in the right hands. Okay. So it will be useful to think of that to to think about that question to consider the space of theta i and wi where wi is the individual  
[06:06:36] willingness uh to pay. Now without distortions um of course we would think that the willingness to pay off anybody uh any of the users corresponds to the return theta i. So these are all on the  
[06:06:48] 45 degree line. Now that gives us these dots are just some examples really there's a distribution of of of willingness to pay. Now if you're charging a price um the price does what it's you know the market does what it's  
[06:07:02] supposed to do. It's going to select the people with the with a high willingness to pay and those just happen to correspond to the people with a high return. Right? Now notice one thing in this graph. Um the way this graph works  
[06:07:17] um the the the market really always does uh selection by willing to pay. So that's a selection in horizontal direction. Um the first best would be a selection in vertical uh uh in the  
[06:07:30] vertical dimension. Okay. So here the price is the market price. C is we think of the market price corresponding to the social opportunity cost of producing this good. So there are no distortions on the supply side of this input. And so  
[06:07:44] the first best selection would be one in the vertical direction. And and here those of course coincide. Okay. So what are distortions doing here? Um you can think of uh many distortions. uh but all  
[06:07:58] of these distortions we feel can be thought of as inducing a shift in these dots primarily actually a shift in the horizontal direction right so in the presence of liquidity constraints somebody who has a high return might  
[06:08:11] have a depressed willingness to pay for example the same would be true for um you know insurance constraints um information constraints might work either way behavioral constraints uh likewise but these dots are are shifting  
[06:08:26] around. Now, in this world, if you're charging a price, um you're selecting out a couple of people uh who have a willingness to pay just below the price.  
[06:08:36] And that's a mix of people. That's a mix of some people who truly have a really high return and some people um who have a low return. Right? So now uh you can think about introducing a subsidy. uh  
[06:08:49] what would uh now when you're thinking about introducing a subsidy uh it's important to think about what is the return of the people that are induced to buy um but because the market will only select on willingness to pay you know you cannot distinguish between these  
[06:09:04] dots of course and so you the only thing that's really relevant is what's the average return of these uh these people okay and so that's what we are plotting here so this is the average return uh conditional um on the willingness to pay  
[06:09:18] And so you can uh think about what is the surplus that would be generated by different prices. So at any uh price p prime. So we denote with p as the market price and p prime any you know other  
[06:09:32] potential price. Um you know the surplus is just the integral over um all the willingness to pay levels from that p prime up to infinity and at each willingness to pay uh level it's the uh  
[06:09:45] average profit that's generated minus the social opportunity cost of producing this and then weighted by how many people are at that weddings to pay.  
[06:09:52] Yeah. Andrew. effect say you have other capital input other than say fertilizer and therefore you know your willingness to pay to capture that ability to return right in  
[06:10:06] that case this is a obviously I'm making the problem model more complicated just dismiss that but right but but it's because you've excluded that possibility that you want to focus on the horizontal >> yes so you could actually perfectly think about shifts here that are not  
[06:10:20] just in the horizontal dimension and the same argument would go through uh so I think that is is a generalization that would be kind of relatively easy and straightforward here uh in fact we are kind of implicitly making that in a  
[06:10:32] second I'll show you um okay so now uh the other question you can ask is um uh you know what how how do we maximize surplus here well uh we maximize uh  
[06:10:46] what's the price that maximizes surplus right and so in this graph um uh you can see that that's this particular price P S uh S star uh at which further a further um subsidy so a lower price  
[06:11:01] would induce people who actually have average return below the opportunity cost of funds. So we wouldn't want to do that. Now the surplus that's being generated here by choosing this uh subsidy level would be that um uh that  
[06:11:14] dash area up there. Okay. So what do we learn from that? Well, what's important for us is that once you know the average return conditional on the willingness to pay and once you know the distribution  
[06:11:29] of willings to pay those two objects highlighted uh down there um you can figure out what the optimal uh subsidy um will be and in particular you can also uh figure out what would be the surplus that would be generated if you  
[06:11:43] actually reduce the price all the way down to zero. So that's the uh the you know the answering the question what would happen if we hand this out for free. Okay.  
[06:11:53] Now how does this change if we relax some of these constraints? So in this paper you know we are particularly going in particular going after one constraint liquidity constraints and we'll index um  
[06:12:06] the the these different worlds by L L0 being liquidity constraints have not been relaxed L1 being liquidity constraints have been relaxed okay and so the one is the orange world the other  
[06:12:20] one will be the blue world um okay so what's happening if we relax uh uh liquidity constraints Um generally you would think that some of these dots would shift to the right.  
[06:12:31] Um and so the distribution of willingness to pay also shifts to the right. Importantly for us the uh conditional returns function. So the average return of the the average of  
[06:12:45] theta conditional on w will also change. That is this blue line here. And you know that will happen no matter whether only the W's change or also the thetas change. Right? And so the important  
[06:12:56] thing is that now the um uh in in this new scenario the optimal subsidy might look very uh different and that's a consequence of the surplus now uh being different at a different piece and will  
[06:13:10] be indexed by yeah sorry >> I'm asking asking this because you're talking about optimalities of subsidy and other policies. So why you know why are we thinking about a liquidity constraint rather than a credit constraint? So why is why is credit not  
[06:13:24] the optimal response rather than just a just a transfer?  
[06:13:29] >> Uh yeah so the answer is is is probably a little sad. Um uh so the answer is that we felt that that's something we can experimentally address. So we will actually hand out cash to people. We  
[06:13:42] will not offer credit. And the reason why we didn't do that is because we were worried that we are not very good at running a bank. Uh so we were worried that they would basically we would offer the credit we couldn't enforce repayment and uh then we don't really know what we  
[06:13:57] did there. Is that credit or is that cash anyway?  
[06:13:59] >> I see that makes sense. So so basically from your paper it could be a liquidity constraint but it could be that it's a credit constraint that could have been addressed.  
[06:14:07] >> Absolutely. And and in the end as well, I think the only thing I will be talking about is policies that relax liquidity constraints by handing out cash cash.  
[06:14:17] That of course is a by now a real uh uh option in the policy mix of a policy maker. Um uh but uh you could of course think of alternative and maybe better ways of  
[06:14:31] relaxing liquidity constraints um and and they might have different benefits and different costs. Yeah.  
[06:14:38] Okay. Um so what we'll be doing is uh we'll try to create this blue world and then in this blue world um see uh first of all how adoption changes. So that's just uh you know how many uh what's the  
[06:14:52] kind of the density of individuals above the market price in the blue world relative to the orange world. Uh how do um what's the direct profit impact of giving out the cash? uh so it might have  
[06:15:05] a direct uh impact on on on farm profits and how does the optimal um subsidy change. Now to answer all of these questions again the thing that we really need is this conditional return function in the blue world and the distribution  
[06:15:19] of willings to pay in the blue world uh along with the same objects in the uh more distorted orange world. Okay. And so that's what we are trying to um get at with our experimental design. Yeah.  
[06:15:42] Yeah. Yeah. I mean the blue line could be really uh uh uh in many way anything uh um yeah could change in in in really quite complicated ways. So so we we don't have a prediction on that yet.  
[06:16:02] Exactly. That would be a leading reason for that. Yeah. Exactly.  
[06:16:08] >> Yes. Exactly. Um so we will not be able to talk about negative. We only consider positive subsidies. Uh you you'll see why we we can't think of negative  
[06:16:20] subsidies. Um okay. So how do we get at these uh um population moments? So what we do is we run um uh you know a standard backard growth marshall uh  
[06:16:33] mechanism implemented through a multiple price list where individuals are asked for each potential price whether they would want to purchase the a bundle of fertilizer in our case at that price yes or no. We'll figure out what's the  
[06:16:46] maximum at which they would um purchase that. That's the how we elicit their willingness to pay. Of course, you know, this is an incentive compatible way to elicit their willingness to pay. Um and  
[06:16:58] then uh we draw um you know, a random price. In fact, uh in our case that has been predetermined. So we implement this through this scratch card design where we tell the go with the scratch card and  
[06:17:12] tell people look the price has already been determined. Imagine it's zero.  
[06:17:15] Imagine it's 20,000. Imagine it's 40,000. What would your choice uh be? we we we feel that sort of reduces a little bit the complexity of of of the procedure in there uh from their perspective. Um and uh so then what is  
[06:17:30] the price distribution from which we are drawing uh these prices? Well, it's this heavily biodal uh uh price distribution.  
[06:17:38] Um the reason is the following. Well, you know, individuals who draw a high price will not get the fertilizer.  
[06:17:43] Individuals who draw a low price will get it. In our case, most of the people who draw a low price actually um get it for free. And the reason why we chose this biodal distribution is because it  
[06:17:56] ensures that roughly 50% of the participants get the fertilizer and 50 don't at each uh willingness to pay level even when the willingness to pay kind of moves to the extremes. um and uh  
[06:18:09] that increases our power to then estimate the profit impacts of getting the fertilizer conditional on the willingness uh to pay. Right? So to taken together what this allows is first the distribution of willings to pay and  
[06:18:23] then at each willing to pay level roughly half of the individuals get the fertilizer the other half doesn't get the fertilizer and that's how we estimate the returns conditional on the willings to pay in this orange one. How do we create? I should say this is you  
[06:18:37] know this is a version of what um Shasang Pad Miguel Snow call a selective trial in the 2012 AER paper. Um there are other people who have used this. So there's a paper that has been around for  
[06:18:50] a long time uh by Barry Fisher and guitars um selling a health product that does something you know at least has many elements of this and the work by Sana and Joshu is is also uh using  
[06:19:05] something like this though I think you know the way you use the data is is maybe a bit different from what we're doing. Um so how do we create then the blue world? Well, we uh proceed this whole thing with a lottery where uh  
[06:19:19] individuals either um get get a low consolidation prize of 5,000 shillings or win 200,000 chilling which corresponds to the market price of the fertilizer bundle that we are offering  
[06:19:32] through the BDM. Okay, what do we do with that data? We'll um uh basically run a regression with a bunch of fixed effects for each one of these uh willingness to pay levels as  
[06:19:46] well as all of those uh interacted with getting the fertilizer for free. Uh yes or no. And then we include village fixed effects as well as um the baseline farm  
[06:19:58] size as controls throughout. um we'll actually exclude uh individuals who drew a price that is lower than their willingness to pay but bigger than zero.  
[06:20:09] That's a you know random subsample conditional unwillingness to pay is innocuous to to drop those but that gives a very clear interpretation of the findings as giving fertilizer um for  
[06:20:21] free. um if we you know had much more power uh you know much bigger sample size it might actually be very interesting to to to to keep them and look at the impact of of paying different prices.  
[06:20:35] Okay. So those are these individual um uh uh level the individual returns conditional on the willingness to pay.  
[06:20:44] We'll also uh aggregate them for uh a couple of reasons. Um so first of all if we look at the average effect of um fertilizer amongst people who are induced at different prices uh we'll you  
[06:20:58] know take the weighted sum of all these beta hats and uh similarly we can look at the direct impact of the lottery on um farm outcomes by um taking the difference of the alpha hats and the  
[06:21:13] blue world and the and and the orange world um suitably weighted. Right? So basically we're exploiting that some people get the lottery um uh and then some people don't get the fertilizer at  
[06:21:25] random and so that allows us to um estimate the pure impact of the lottery on farm profits.  
[06:21:32] Okay. The setting uh where do we do this? We do this in in eastern um uh Uganda in around Teruro in three districts. Um this is of course very  
[06:21:46] close to um the area where the work by Duflow Kramer and Robinson was done on the returns to fertilizer. So they work with the same type of fertilizers just on the other side of the border  
[06:22:00] basically in in in in Kenya. Um we are selling uh to farmers a fertilizer uh a bundle of two fertilizer and planting fertilizer top dressing fertilizer um  
[06:22:14] that is meant to be suitable for one acre and the market price of which is 200,000 till chilling. We did this in the first agricultural season which is the main agricultural season in this  
[06:22:26] area in of 2017 and uh 2018 because that's a while ago and um please don't ask but uh but I was uh I was happy DIP to  
[06:22:40] see that your data is even older but uh yeah anyway so so what we do there is in the first season we um uh sample LED five 51 villages in the  
[06:22:53] second season. 102 um villages stratified by population density conducted a census of the households selected eligible household households by finding households that engaged in commercial farming, cultivated maze,  
[06:23:08] self-reported cultivating 2 to six acres of land and had mobile money account.  
[06:23:13] None of this was really very restrictive. And then we selected um 48 farmers in the first season and 816 farmers uh in the second season.  
[06:23:24] Okay. Within the season what we did is first the census then a baseline survey then the willingness to pay is elicitation shortly after which um the the fertilizer was actually delivered  
[06:23:36] five days later. Um then during the season we did a phone survey to elicit uh labor supply responses of the farmers. Um and then uh at towards the end of the season we tried to measure  
[06:23:50] the yield responses in two different ways. First we did a crop uh survey before the harvest and then we did a follow-up survey after the harvest and  
[06:24:00] we are um yeah I can maybe talk later uh about how we are trying to combine that data to get as precise a measure of of of yields as as we can. Uh you know it's  
[06:24:14] still going to be uh uh noisy. We also uh went to the markets and um collected prices on agricultural outputs to value those um yields and then we conducted another phone survey in the following  
[06:24:28] season um to look at the long run um responses. Yeah.  
[06:24:38] >> Are they to adjust to the plot size or do they is it multiple plots or like how do you get those things to map up correctly?  
[06:24:46] Yeah. So, so most uh of these farmers cultivate at least one acre of of land.  
[06:24:52] Um what farmers do some farmers do is that they store fertilizer which they didn't use in the first season. Um there's a kind of yeah surprisingly  
[06:25:05] small resale market. Uh so some people also sell some of it but that's a very small uh proportion. And when but we measured that and and when we look at  
[06:25:15] profits we um uh you know uh you know any any kind of revenue that they generate from reselling some of the fertilizer will be counted towards their  
[06:25:26] farm revenue and profit. >> Yeah. I mean we told them that this is suitable for one uh acre and Yeah. And so and we well and we gave them  
[06:25:39] instructions how to uh how to use it right so which quantity to use per hole basically. Um and uh you know our sense is that it's not going to be uh enough  
[06:25:53] for the typical total maze area that these farmers have. and uh and and and farmers will have used it for as long as it lasted for the holes that last you know for long as it lasted and then uh  
[06:26:06] not beyond that but uh but you know I we could look more into that.  
[06:26:14] >> We are interested in total in actually total revenue total farm revenue. Yeah.  
[06:26:21] Okay. So um so let me show you the the the results. Um so first of all in terms of what the average what is actually the distribution of willingness to pay that  
[06:26:35] we are eliciting. And so the um the the distribution of the willingness to pay in the status quo world looks like this. And in particular what we find is that you know of course  
[06:26:49] most people have a very uh low willingness to pay. On average the willingness to pay is roughly 56,000 Uandian shillings. Um there's a small fraction of individuals who has a willingness to pay equal to the market  
[06:27:03] uh price which was the maximum they could state in our elicitation procedure. It's 4.2%.  
[06:27:10] um and uh that corresponds pretty closely to what we actually find in the baseline survey in terms of um fertilizer used. So 4.2% had used the DAP fertilizer in the previous season.  
[06:27:23] 1.5% had used KHN in the previous season. Um so that gives us some level of conffort that this uh willings to pay elicitation procedure um had had worked  
[06:27:34] reasonably well. Um yeah that's no. Um so how does this change uh in the blue world once liquid once liquidity constraints are relaxed? Well there's a  
[06:27:47] noticeable uh shift of that distribution to the right. In particular uh also the number of uh the fraction of individuals who have a willingness to pay equal to the market price u increases from you  
[06:28:01] know 4.2 two to 6 uh uh 6%. Right? Okay. Now, the average willingness to pay goes up to 72,000 UN chillings in this case.  
[06:28:12] Okay. So, that's on the willingness to pay. Now, let me start with kind of the average um profit impacts of the fertilizer. Yeah. Sorry. Yeah.  
[06:28:23] to interpret this as like relaxation liquidity constraints to buy or relaxation liquidity constraints that you would use to buy and I ask in part because it's not an enormous difference but looking at  
[06:28:37] transfer trials directly they don't find very big impact adoption so should we think about labs >> yeah um so we try to avoid that we try  
[06:28:51] to make clear that they can make the money any way they want. Um our findings in terms of um the kind of profit impacts actually are pretty similar to other the other kind of  
[06:29:05] studies the the the the work of Udrian courses in Ghana and Bean and and courses in in in Mali. Um now of course you could also in those cases ask whether there was some labeling u uh uh  
[06:29:18] effect and uh so it's a experimental demand effect or whatever you call it.  
[06:29:23] Uh and uh but at least uh you know it's it's not off from that. Yeah. I mean but but is there some of that? I I can't confidently say that that that's not the case.  
[06:29:40] assets and localiz. >> Yeah. Yeah, sure. We could do that. It's  
[06:29:50] a good idea. Um Okay. So, uh so what do we find in terms of the average um in uh impact of getting fertilizer on uh revenues? So, first of all, revenues go  
[06:30:05] up by roughly 140,000 in China. Okay. So uh that is of course a substantial and highly significant increase in revenues.  
[06:30:14] Uh so fertilizer works that is kind of maybe unsurprising. Uh it's also substantially below 200,000 organ right it's not uh uh u exceeding the the price  
[06:30:28] of um fertilizer in the market. So um surely it doesn't make sense to hand this out for free. Now this is uh especially uh uh the case if you think of the profit impact. So this is just a  
[06:30:42] revenue impact um together with that revenue impact comes an increase in the cost. So there's an increase in non- labor uh input costs a very very small one but basically zero here but there are some changes in behavior but  
[06:30:57] especially there's a change in uh the labor cost. So individuals who get the fertilizer hire more labor. Um they also provide more uh household labor on the farm. We we value that household labor  
[06:31:11] at 60% of the uh of the uh market wage. Uh but the findings here are not uh you know really particularly uh relying on any particular value that you choose  
[06:31:25] there. So the total costs on the farm actually go up by roughly 75,000 Ugandan shilling. Uh meaning that profits only increase by 65,000.  
[06:31:34] Yeah. >> Yeah, that's a that's a great question.  
[06:31:54] Um so actually we're doing sort of some work along those lines right now. Um we also see in this experiment that uh there's a uptake of fertilizer usage in the following season. Uh so which is consistent with this uh with this idea  
[06:32:08] that maybe some of the individuals here experience high returns and real noticed that they have high returns and started adopting fertilizer.  
[06:32:17] >> Yeah. >> If I could follow up on that. Um yeah, I guess what I was going to ask about was to what extent might these estimates on impacts be understated for a couple of reasons I was thinking about the one one was what you just mentioned increased  
[06:32:32] adoption in future seasons right which it seems like you're not taking into account here >> um and then secondly um a different issue uh potential positive spillovers under the control group which may lead again these to be underestimates. Yeah,  
[06:32:45] I think the third thing is actually that the uh fertilizer might have longer lasting effects on the soil. Uh so that that that also adds to that. So so um  
[06:32:58] yeah that I I agree with that. So all of those would change the calculation.  
[06:33:02] Yeah, we we should think about that. Um okay, so um you know find roughly kind of uh similar impacts in the in the in in the blue world.  
[06:33:14] Sorry, slightly smaller profit impacts. In fact, um you can think of that through the through the lens of the model. But um in the interest of time, let's uh let's maybe move on. The one thing I want to highlight is that the average profit impact here is 65,000  
[06:33:29] Ugandan shillings. The average willingness to pay in the orange world was 56,000 Uganda chillings. Right? So that's not wildly off. Uh so this would certainly not lead us to reject the idea that they have a rough idea of what the  
[06:33:42] profit impacts um are at least uh on average that they're right. Okay. So how does this uh comp compare to um to to the work by Duflow Kramer and Robinson?  
[06:33:56] They find slightly um higher uh uh impacts. Um now uh they don't take into account the uh the the labor cost when they look at the the the profit in um  
[06:34:09] packs. There also a bunch of other uh reasons uh why our estimates might uh be different. First of all, they were conducted in uh different seasons. They are uh pricing things uh differently  
[06:34:21] than than we do too. And so uh there are a number of reasons to think why their their estimates are slightly higher than well are higher than ours. Um okay so now uh let me uh look at the impact of  
[06:34:36] the lottery. So just giving people the lottery also increases um revenue and profits. So uh farm revenue goes up by roughly 80,000 in response to getting a  
[06:34:47] lottery of 200,000. Um and uh here the increase in cost is much more modest uh relative to that um revenue increase. So that you uh see a profit in uh impact of  
[06:35:02] of of 50,000. Okay. So giving people just the lottery itself increases uh uh farm profits. Now that again is uh um you know roughly in the ballpark of what  
[06:35:13] others uh uh found as I uh said before it's in fact pretty much between those uh studies in terms of the the direct impact of on farm profits of giving  
[06:35:25] farmers cash. Okay. So now let's uh get to the um the the returns conditional on the willing to pay. So these are estimated with you know substantial noise as I said but the one thing to  
[06:35:39] notice is that in the in the orange world um so where liquidity constraints um have not been relaxed there uh are very substantial returns amongst the farmers who have a willingness to pay of  
[06:35:53] 140 160 180 uh000 chilling. so close to the market price, right? And so that's really what's driving this idea that it might be a good idea to subsidize um a fertilizer.  
[06:36:08] And so in particular if you think of different uh price levels so if the induced price is uh you know anything from 0 to 120 um you can ask well what's the average um increase in uh profits  
[06:36:21] amongst people who are induced and that is the to buy and that's the orange cur curve on top. um how many people are induced to buy that's given in the graph below and you can think about what's the  
[06:36:33] optimal um subsidy that maximizes surplus and that in our case uh happens to be uh a subsidy that gets the price down to 140 um uh thousand shillings. So  
[06:36:46] that's a that's a 30% um subsidy. Now in the blue world you can see that no subsidy here has any uh positive impact on surplus. The uh now if you introduce  
[06:36:58] sub such a subsidy actually um uh we would find that uh the adoption rate goes up from 4.2% 2% to I think 11.1% so  
[06:37:09] roughly 7% um uh increase and uh and and we don't find this um in the unconstrained scenario. Okay. So in the kind of remaining 10 minutes let me  
[06:37:23] quickly talk about um this idea of uh this question of what the policy maker might want to do uh when choosing between these different options. answer >> just one question about the market price.  
[06:37:36] So if the global price of the three input chemicals falls, do we expect to see much pass through? Is it likely that the market price?  
[06:37:47] I'm trying to figure out like how how contingent is this pattern of results on price.  
[06:37:54] >> Yeah. Yeah. I think you you that's a great question. I think you can ask the same question on the output price, right? So I think um relating to the the earlier presentation um uh I think it's um  
[06:38:06] uh it's not implausible that uh input uh you know input prices are higher in those places relative to say an American uh what an American farmer faces output prices might be lower and that makes it  
[06:38:19] less attractive to to adopt these uh technologies. And if we would find solutions to both of those problems, you know, we would likely see an increased adoption of these uh these inputs. I  
[06:38:31] think that's probably right. Yeah. Um okay. Um so let me think about the optimal policy.  
[06:38:41] Now coming back to uh Mushri's uh uh uh earlier question here you know we will think of of a very uh simple scenario where the government chooses between those two options either choose a  
[06:38:54] subsidy of any level or hand out uh cash transfers because those are the only two options we can uh say something about here and and and uh uh that it's not of course true that those are the only  
[06:39:08] options that a policy maker effectively has um now uh what we'll think about is a world where the policy maker either introduces the different levels of subsidy levels or um it gives a cash  
[06:39:21] transfer of the size that we were uh giving out to a fraction d of all um households. Okay. And so then we are trying to figure out what is the wolfer  
[06:39:33] maximizing um combination of the D the share of people who get a a cash transfer and the subsidy. And you can see how you know giving changing the D basically uh creates a mixture of the  
[06:39:47] blue and the orange world and as a consequence the optimal subsidy um will change as well. Okay. And so what do we find? So what we find is that at a high opportunity cost of funds the government will not want to do either of those  
[06:40:00] things and you know just leave adoption at the current 4.2%. If adopt if opportunity costs are falling somewhat the uh the first thing the government might want to do is to have a very small  
[06:40:14] subsidy 10% subsidy right that has a very high return it increase for every dollar that the government spends $2.18 of uh of of surplus increases adoption  
[06:40:27] from 4.2 two to 6 uh 6%. And uh it's not a particularly uh uh costly in terms of total budget costly program that the government runs um but it realizes these  
[06:40:41] high returns that uh people with willingness to pay of 180,000 had. Okay.  
[06:40:47] Now if the opportunity cost of funds falls further then the government would want to have a bigger subsidy and also induce the people who had a willingness to pay of uh 160 and 140,000 to uh adopt  
[06:41:00] fertilizer and so that would increase the adoption to 11.1%.  
[06:41:05] And now the return uh for every dollar spent for the government is 2.01 in terms of surplus that's being generated.  
[06:41:15] Now only when the uh opportunity cost of funds falls below uh you know 1.2 roughly the government would actually want to switch and instead uh roll out a  
[06:41:28] program where uh cash is handed out to individuals to all everybody in fact and the moment you do that of course you don't want to have the subsidy any uh any longer. You can see that that's a much much bigger um program. um it  
[06:41:43] creates um more total uh surplus but per dollar spent it only creates this um additional surplus of 1 uh25%.  
[06:41:53] So, you know, I don't mean I don't want to over interpret this, but I think this might be one uh uh explanation for why a lot of governments in uh developing  
[06:42:05] countries adopt these uh kind of uh uh subsidy schemes um o over um uh o over other programs. They are of course big schemes uh relative to uh their GDP  
[06:42:20] often but um um you know in in relative to these other programs here it's it's much much smaller still Rachel you had a question so  
[06:42:40] so all but those things plus the increase in in farm profits, all of that. Yeah. [snorts] Okay. Uh and so  
[06:42:51] now, uh of course, if you choose to go for this 30% subsidy, um that um leads to a relatively high adoption rate of 11.1%.  
[06:43:05] And um it's true that some of that adoption might be individuals who actually don't have uh very high returns, right? So it is truly uh a second best policy. It doesn't achieve  
[06:43:20] necessarily in the first place at all, but you know, it might be the very best thing that the government can do in the face of uh of those distortions. Yeah.  
[06:43:29] So it >> depends.  
[06:43:33] Sometimes a thought about replacing these subs with kind of schemes where basically every household, every farmer gets a fixed amount of like fertilizer or something which is obviously going to be worse given the selection you've  
[06:43:47] documented but that's kind of a you know would be another nice >> okay yeah great to them yeah [snorts] >> so so um it seems like you're you know basically you're you're what everything  
[06:44:01] you're doing here is about tying your hands to simply thinking about there being only these two policy levers, the price and the cash transfer. Uh, and you know, in that context, you're basically sounds like you're basically agnostic  
[06:44:15] about what the underlying distortions or market failures are that are that are operating. I mean, you listed a number and I think you're sort of trying to not pin yourself down to and you're not attempting to try to sort of reveal what the underlying market failures are. But  
[06:44:27] I guess the um uh it feels like in the bigger picture having some insight into what the underlying market failure is uh could you know point to potentially you know more attractive set of policy  
[06:44:41] levers uh like you know mush mentioned credit there could be it also could be that there's a imperfect information going on say on the returns to fertilizer um uh so I guess what what are your thoughts about  
[06:44:55] >> I feel like it it feels a little premature to me to say, okay, here's the menu of policies that policy m should should think about when we haven't really revealed what the underlying market failures are and whether there are other policies out there that might, you know, be more um desirable if  
[06:45:09] they're directly affecting the the relevant market failures.  
[06:45:12] >> I I I couldn't agree more. I uh I I I was trying to caution this sufficiently in the beginning. I uh but let me just highlight that again like this is what you find when you restrict yourself to those two policies. If I were a policy  
[06:45:27] maker, I would definitely ask those questions about what else could be done and and what uh uh yeah and but I think the general point here is that uh that addressing those distortions uh often  
[06:45:41] will be costly too, right? and and it's uh uh and so uh it it creates this uh trade-off between addressing the distortion at the source or dealing with the distortion uh by uh you know  
[06:45:54] changing the market price basically and I think that's always a trade-off you would need to uh you would need to consider >> credit so I mean I really like this idea that sort of combination of credit  
[06:46:08] constraints and energy easier to Exactly the right people. That's really interesting kind of develop.  
[06:46:16] >> Yeah, I'm not sure I would sign that statement, but okay.  
[06:46:19] >> No, exactly. It's not exactly the right. It's just the best thing you can do, right? It's the subsidy. What it what the subsidy does is it it it does potentially induce people to buy who truly have a low return. But it's the  
[06:46:32] best thing you can do if in this graph you can only select >> on the unwillingness to pay. And that's what you restrict yourself to with a market mechanism.  
[06:46:47] >> Your unlimited time my statement depends whether the thought >> so we are actually trying no not here we I can tell you we're trying to do this right now. We we're we're teaching farmers how to run experiments on their  
[06:47:00] own farms. uh and so uh from those we can get individual level uh return estimates right and so we can look at heterogeneity and individual level returns um but uh yeah it it will take a  
[06:47:14] time to uh to to to come back with those findings and we haven't done that here >> I guess in a different country  
[06:47:29] suggesting that credits were less importance than missing insurance.  
[06:47:34] So, uh this different by all men would that be something that also consider >> I mean first of all I I also wouldn't  
[06:47:46] sign that statement. I so uh I I I my reading of that paper is that uh credit is pretty important too uh um as is our  
[06:47:58] insurance constraints. Um but uh uh you know I I also agree that in this of course we are going after one of these distortions and you know I think what we find is that once you go after that distortion um you don't want to have a  
[06:48:13] subsidy again um any longer you could potenti you know if you went to after some other distortion who knows what happens and we could have of course done that we didn't we didn't do that  
[06:48:25] in conclusion in conclusion I uh in conclusion In conclusion, we're sorry. I hope we delivered on uh on what I promised. Thank >> you. Thank you all for staying for the