 Thank you very much. OK, so good afternoon. So this is joint work with Kyle Emerick, who was a former student and is now starting as assistant professor at Tufts, Allah, and Mansour Da, who is at the International Rise Research Institute in New Delhi. So let me start by saying we all know that risk is important in agriculture. And risk has some very high costs on the farmers. This is something that we have been concerned for a long time. But here I want to look at something in more detail. Then there is, in a way, two mechanisms through which risk of extreme weather events, so we're going to concentrate on extreme weather events, affect farmers. One type of cost are the direct costs, who damages to crops, homes, livestock. So this is the immediate cost after the storm of everything you have lost. The second type of mechanism that negatively affects the farmers is all what they have to do, the self-insurance action that they have to take, conservative decision they have to make to reduce their exposure to risk. And this often at the expense of lower profitability. So choice of crops, which are less sensitive to weather shock, crops of variety, level of input use, not undertaking expensive input use when you may lose them, and some of management techniques. So our question here is to focus on the second mechanism. And the way we're going to do that is to infer the negative cost of risk. We're going to do something on the positive side. We're going to see what sort of the positive consequences of alleviating the risk. And so this is how we're going to infer negatively what would be normally, what was the cost of risk. So the question that we are asking is can a new technology that make agricultural output less susceptible to weather extreme carding investment that farmers did not undertake before because they were too risky? Now if this is the case, and it's going to be the case, I will show you, it means that the technologies that sort of reduce risk for the farmer for the bad years also are going to boost agricultural output in the normal years. And I think this is sort of a new insight into the way to address risk, which is quite interesting. The approach that we take in this paper is pretty straightforward. It's a two-year field experiment that we conducted in 128 villages in Odisha, in India. We allocate the first year, we allocated five kilogram of flood-tolerant rice seeds, which is called swanah sub one, and come back to it in a minute, to a random subset of farmers, very standard randomized control trial, five farmers randomly selected in each of 64 villages, half of the villages randomly selected as well. The first year that was in 2011, the summer of 2011, there was very large floods that occurred in many villages during the first year. This is really bad for the farmers, terrible for them. But in a way, good luck for us. Because then, right the first year, we could verify the agronomic properties of that seeds that we had given to the farmers, and verify in the farmers' field and cultivated by them with no help and no support from any one of us, and as opposed to being an experimental station done by Erie before. Second year, 2012, this was a normal year with normal rainfall, so good luck for the farmers this time, finally, but good luck for us as well. Because then, we are going to see all the changes in behavior that they undertook because they knew of the seed, and we are going to see large changes in investment, but also we're going to see, I'm going to show you that, changes in savings and credit use, and we're going to see a large aggregate effect on yield and on production. Remember, immediately the second year, this is extremely fast. Now, because there was no flood, we know that all the results that we find, the aggregate effect on production is solely due to the behavioral change. So we can really separate how much can be due to the agronomic property of the seed. This is the first year. How much is due to the farmer's behavior respond the second year, and we can compare the order of magnitude. So let me show you very briefly the result of the first year. So swanna sub one, this is the new seed, the flood tolerant rice variety. It's similar to swanna. It's actually exactly the swanna, which is a common variety used throughout eastern India, in which the scientists have introduced just one specific gene that make it flood tolerant. And it's very, it's particularly flood tolerant for the period which is five to 14 days of flood. You see this on the graph on the horizontal axis, you have the number of days of flood. And in our experiment, we had filled with no days, no flooding at all, a little bit higher. And then we had fields in any one of these area, but you see more, a lot of field at 10 to 12 days of flood. And what's represented on the vertical axis is the difference between percentage term, between the yield in the different field, according to the number of days they were flooded. Two things very important to see on this graph. First, at the lower end, when there is no zero to five days of flooding, there is no difference. So there is no trade-off. That swanna sub one is no worse in good year than the swanna. And then we see the great value is really between five and 14 days. Beyond 15 days, we had actually had almost no experiment, but we know that it's not doing too well anywhere. It turns out, so what this technology does, and I'll come back on that, is that it's going to cut on the bad years to raise the yield. So cut the losses on the bad years with no negative effect on the good years. So that means that it both increases the expected value, the expected return of whatever you do, the expected yield, and reduce the variance of yield. So we're going to have to be careful to distinguish those two. It also turns out, but that's just because this is the way India is, it also turns out that this is poor poor. Why is it poor poor? Well, who do you think is cultivating the most marginal land and the most which are more prone to flood? Well, the lower cast, for a number of reasons, historical marginalization. And therefore, it turns out that in fact, the lower cast are going to benefit even more because from this protection from flood. How does sub one works? Just a little bit of minimum of agronomy. Now, most of the rice seeds, when flooding comes, they elongate to try to keep their head above the water and keep that. And then after when the water recede, they just fall and have lost all their energy, which has been used to grow up. What sub one does is just hold his breast, stay underneath, stay put, wait until the water recede, and then has all this energy and start to grow up. And that's why it's called actually the scuba, scuba rice in India. And you see here the fields after the flooding, still a bit underflowing. The theoretical framework we have in mind is very standard. It's a standard multi-period household model where production is risky. There's very simple in the model, flood or no flood. And the flood happened after the input choice. That's very important. It comes late in the season after you have decided. We have planted and put the fertilizer, or most of the fertilizer. The return in the model, the return to inputs, is lower on the flood conditions. So this is the. And the modern technology increases the return to inputs on the flood, but has no difference on the normal conditions to respond to what we saw the first year. So it's a protection on the downward risk with no penalty on the good circumstances. The prediction of the model, there's two sort of prediction. One is sort of the productivity effect, the fact that technology increases the expected return to inputs, then you should increase your inputs and your investment in any way. Even if it was not a risk issue. And then there's the insurance effect, which is the fact that the technology reduces losses in bad states. And therefore, it reduced the risk. And it allows for you to reduce the risk mitigation behavior and the precautionary savings. And we should see that increasing with risk aversion. In term of econometric specifications, so we're going to have a, because it's one of my control trial, a very simple specification, where it's going to be either at the farm or the plot level, depends on the question we're asking. Essentially, why on the left hand side is some of the outcomes that we are interested in and we'll look at that in a minute. As a function directly, a linear function, of whether the treatment variable is whether the farmer had received, so it's randomized control trial again, so that's orthogonal to anything else, whether the farmer had received those seeds the first year and then a block fixed effect and some control variable as well. We had some farmers who disadopted for different reasons. Their seed bed was flooded or they ate their seeds or they mixed them with the others, different reasons. And some we figured out to get those seeds from some of their friends. But of course, this is endogenous, so we are going to everywhere use the intention to treat effect. I mean, what we call the treatment is having received seeds, whatever you did with it. And if you haven't received it and you managed to have it, you're still counted in the control group. This is pretty standard. The outcome of the interest that we will be looking at, so the why that I had in the previous slides, are a number of inputs and management practices, rice area, fertilizer use, especially we're going to contrast the fertilizer of the earliest fertilizer that you put at the beginning of the season as opposed to the one that you put later in the season, use of traditional varieties as a way to protect against risk and planting method than yield, which I mentioned before, credit and precautionary savings. So let me go through a number of tables here. You can just listen to me or look at those numbers, whichever. So the treatment variable is the one which is written original mini-kit recipient, which you have there. And here what we see is that all these are three different ways to show that flood tolerance increases the area cultivated. We see the number of plots increasing by almost 0.7 on a mean value of 3.6, so this is a large increase. The worst area of 0.10 hectares on an average value of one and or in log of 9% increase. So a 9, 10% increase in the log areas coming through increasing in number of plots. We keep that here. I don't have the detail here, but those plots that were brought into cultivation were the more flood prone and marginal plots. Now look at the impact on the use of fertilizer. So here there's three types of fertilizer. I don't know very much what they are, but maybe some of you do. One which is called DAP is put very early on just after the transplanting. The one that MOP is put later on in the field in the middle of the season and the latest one, the urea is put just at very late. And in a sense, those who are more subject to be more at risk of being lost are the earliest one because you have to put them before do you know whether you have flooding or not as you only put the urea if your field has not be destroyed. And we see here, again, the parameter which is on the original many case perceptions, a very strong increase in fertilizer, the early fertilizer by 18 kilogram on the mean value of 80 kilograms. So this is really a large increase. A little bit of some increase in the one in MOP which is put in the middle of the period and no change in urea which is the latest that you've put in. So this is some indication of an increase in fertilizer but very specifically the fertilizer which is used early. Then there is some traditional varieties which are not the modern variety which have much lower yield, whatever the condition but have one particular advantage is that there is less risk of complete failure especially beyond 10, 12 days of flooding. And then the third input that we want to look at is the different way to seeds the field. There's a very labor intensive method which is a transplanting which we actually, if you go a tiny bit outside of Hanoi you're going to see this extensively nowadays which is the transplanting one plant at the time. And there is a low, much less labor intensive way which is simply to broadcasting those sending the seeds but has much lower yield output after that. Okay, so let's look at those three inputs that we look at. So again, the original mini-kit recipients those are the one that receive five kilogram of the sub one seeds the first year and we see that while they use less one that's sort of what we would expect because we just gave them a replacement seed which was no worse but they also use less of the traditional variety by 4% of them on an average of 28% of the farmers do use some amount of traditional variety, number of observations. So this is the other plot level, 4% of the plot when 20% of the plot. And then in term of broadcast again a very large decrease in the use of broadcasting therefore using transplanting instead and a decrease you see the last column of the number of plot which is not which are not cultivated as well which is what I mentioned the first. Overall when you aggregate all these sort of more intensive input use we see a shift on the average productivity of the yield, therefore average yield in kilogram per hectare of 283 kilograms or about 10%. Okay, let's keep this number in mind. I think we want to, if you compare that we're going to compare that to what we had. It's comparable to what we were obtaining just from the agronomic result on the first, the year where there was flooding. So this year there was no flooding. All this average productivity increase is solely due to the input or to the behavior of the farmers, 10% increase in yield. Two more results here, one on credit so we see a large increase in credit. Now we don't know whether this is higher demand from credit which could be but it's also be higher supply that actually the bank actually will give credit to those who have those seeds because it's less risky. What interesting here in this when you contrast column two and three is the credit which is given early in the season which increases by 5.4% on a mean value which is pretty small which is only 14% of the farmer received credit. Finally something which is very interesting is we see less savings of seeds, of less saving of grains as we call them, less saving of grains for the next year. Now they had a higher production so you would expect that they could sell more. When inside you would, they had a higher production so in many way you would expect that they could save more of that if they wanted to, if they were short or you could say that they would expect to have more the next year but the fact that we see that there is less saving of grains at any level of production suggests that it is really due, it is really a response to the risk of not having enough for the next year. Now so let me summarize all these results. So the conclusion here is that we see the results suggest that there is the risk strongly influence the decision on the inputs. So this is on the second year. An input 10% increase in rice area, 11% in fertilizer use, especially at the beginning of the season, 15% less reliance on the traditional variety, 33% less use of broadcast planting and this aggregate to an approximate 10% increase in yield, 5% less savings of output for future consumption and 36% increase in graduate use. This is just one year after the introduction of the seed. Now the main explanation that we've given is that the reduction is due to the reduction in risk but of course we could have other explanation which are possible and the one that I mentioned in the beginning, maybe it's solely due to the aggregate expected return that we have and because the fact that we have that it's protected and downward risk without penalty in good years mean that it increased the expected return and reduced the risk at the same time. Now both of these effects could be responsible for much of the result that I have. So to distinguish them we have to look in more detail and I think I mentioned them as I presented the result. One of the indicators that is due to risk rather than expected return is the increase in fertilizer is concentrated on the early fertilizer and not the late fertilizer. Now those are complementary, those fertilizer, so you would not increase one without the other if it was solely due to the expected return. We saw also the land brought into production is the low quality land and the land which is more prone to flood and we see the result, I didn't show them to you, the result are generally larger for most risk-averse farmers and the share of production which is saved is lower at any level of production even among those we have relatively low level of production. So it has that. There's two other potential explanation, the wealth effect, could it be due to the fact that we gave to some people some seeds, some new seeds at the beginning? No, I won't, and so they are wealthier and therefore they can afford more input. We look into a fair amount of detail in this, to this but it's unlikely, first we give them very little on a five kilogram of seeds and then even if we condition on the harvest that they had the first year because they had much better harvest, we have no change of result. Could it be to the output prices? You would expect that this miracle seed would be much higher price. It turns out that it's not really much higher price, only 4.6% higher price and there's very few farmers, well, not very few, there's only 40% of the farmer who sold any output at all and the results are focused on only considering the farmers that did not sell anything so it cannot be coming from the price. So let me conclude, and this is my last slide. So we see a very large reduction. No, we see that when we have last risk reduction from these weather, flood resistant and technology and we see that it helps cope with shocks in bad years, this is the reality and it helped reduce risk management and self-insurance in normal years. In a sense, we knew that before, right? I think we've learned that models tells us that, we've learned that in all our textbooks that risk is really bad, not only because you need to have instruments of risk coping afterwards to manage it, but then it pushed you to do risk management. We knew that and we probably did not need in our city to do it. What really we learned, at least for us, at that is the order of magnitude and I think we completely, at least us, in our team, we completely underestimated the order of magnitude. So let's look at those two numbers together. The average gain in normal years from risk management was 283 kilograms in the scope of 2,000 and some farmers. And that's about equal to the average avoided loss in bad years from shock coping, which is 250 kilograms. And this was a really bad years when we don't have time series, but that was a really bad flood years. So the order of magnitude of what we gain in the good years is as large as the size of what you, the avoided loss in bad years. Now, how many good years, how many bad years we have? Well, we don't have a very good time series on that, but actually we have a bit more than that. I think we have had two of these bad years in the last seven years in this area. So we definitely have more normal years than we have bad years. That's mean that the behavioral spillover effects exceed the gains from the shock coping, the agronomic intended purpose effects from the scientists in expected value. Of course, it's not the same years. So of course, for Erie or counterparts who were really developing that seeds for those bad years, this is very good news. I think it's very good news for all of us to see how reducing risk has very strong positive impact, even for the good years and overall for the production of rice. Thank you.