 So this is a very big topic. I think we are going to be quite selective in taking on particular aspects of this. You'll notice that this conference actually was about the mapping of the future of development economics. It's not about economic development, it's about development economics, and indeed wider is a university that teaches and researches in development economics. So let me start by making a few observations about what progress there has been in development economics. Then I want to look at what do we tend to do in microeconomics today relative to what we have been doing for the last 30 years. I'm going to extract some shortcomings of what we do in relation to what has happened to development economics. And then I want to present some results which derive from the way we do some of that microeconomics, especially using field experiments. And then I want to reflect about some of the issues that are going to be on our agenda for the years to come. So one has to be really impressed. Oh, I don't have a screen here. We have to be really impressed by the fact that what happened to development economics in the last 30 years has been a stunning story. It used to be a field that was a maverick in the econ departments. It was considered to be not sufficiently rigorous, not sufficiently empirical. It was considered to be too interdisciplinary as opposed to really focused on what the economic theory had to contribute. And then what we see is that the field has been mainstreamed. Most major econ departments in major universities have by now a field in development economics. The importance of the field and the research coming out of it is illustrated by the fact that publications used to be very much into field journals. And now we see that the top journals are very receptive to papers in the field of development economics. And indeed, there is a very large number of such papers which are published in the top journals today, which is unique and quite new. And there has been a huge growth in the fields, in the number of students, in the number of positions, in the dynamics of the job market. And this job market is not only in academia. It's also in the institutions. It's also in the private sector. It's also in the philanthropic sector. And there's quite a lot of positive energy. If you go to the NEUDC meetings, if you go to some of the specialized development meetings, you cannot be but impressed by the fact that there are so many ideas. It's kind of bubbling with enthusiasm and ideas. And that's really the fantastic achievement that development economics has had over the last 30 years. What do we do in this microeconomics of development that characterizes what is sort of now looking forward as opposed to what we have been doing over the last 25 or 30 years? I think there are two big revolutions in a sense. The first one is how we do things. And so here what we have is sort of no longer big ideas, but the big ideas are not in terms of what should be done. The big idea is how to do things. New methodologies, new approaches, new ways of thinking about development, new ways of conceptualizing the development problem, and new ways of using data. So clearly new approaches, much more pragmatic based on diagnostics, recognizing heterogeneity, customization, experimentation, sort of coming up with solutions that are being constructed in the case of particular context, giving us in a sense a lot of degrees of freedom as to how we are able to address the development problem in a particular area. So the big revolution is how do we use that information? A lot of rigor as opposed to kind of looking for policy recommendations, which are going to be based on spurious correlations, on partial correlations. What we want is causality, what we want is identification. There are going to be different ways of achieving it, right? And there is no gospel as to how this can be uniquely achieved, fortunately. But clearly the filter and I think the reason why we have kind of been upgraded in the development profession and upgraded in the hierarchy of journals is precisely because rigor has become a commonly shared criterion as to how do we do development economics today. And this is important because it links to kind of results-based management, which is what the agencies and the governments want to have today. Namely that if we do policy, we want to make sure that there is causality between the instruments that we propose for policymaking and the expected outcomes of those policy interventions. We have discovered the possibility of experimentation the last 10 or 15 years. A very important revolution in the way in which we kind of manage information. We go to the field, we partner with local institutions, we organize experiments, we collect new data, we do baseline survey, we do follow-up survey. So we are eventually in the field for extended period of time in relationship with kind of partnering with local institutions who are development agents. So it has taken sort of academia into the field. It has sort of democratized government. It has made us direct agents of development instead of development being done by the big agencies in a kind of top-down fashion. So kind of bottom-up interventions where a PhD dissertation can be a mechanism to explore new ideas and those new ideas can have very significant policy relevance if the ideas which are being explored are worthwhile. And then the last dimension of this methodological advances is the access to big data. The fact that we use now data that are kind of satellite-based, night-by-lights or it could be kind of micro data of different types it can be data, administrative data that we collect in enterprises much more access now to census data on a kind of sampling basis, etc. And the fact that we work with those massive data overlaying and combining those data gives us the possibility of moving into much more natural experiments. So we kind of combine data access to the regular revolution if you like looking for ways by which we can find exogeneity which is going to give us ways by which we can achieve causality and derive policy implications not only from the RCTs from the randomized control trials which is one way of doing it but also from those natural experiments which tend to have the advantage that they tend to be much broader in terms of scale. Hence they have second round, third round effect. They are things that have been sustained over a longer period of time. They achieve eventually scaling up. And so the policy implications of what has been done in terms of analyzing them via natural experiments has much more value than what we likely will ever be able to do with randomized control trials that tend to be local, tend to be confined, are difficult to sustain over time. And as we will see, has the TRN or what is the external validity that those particular experiments are going to have. And then in terms of the new emphasis I think the easy way out because it's so multidimensional is to think of the SDG versus the MDG. The MDG last 30 years. The SDG next 20 years or so. What else kind of some of the highlights, especially in the dimension of more microeconomics that we see appearing on the agenda for research. Certainly the emphasis on equality equity but especially on inclusion. Inclusion in terms of labor market, inclusion in terms of process, inclusion in terms of job satisfaction, inclusion in terms of kind of progressing, developing oneself in the context of a particular job. Think of the counterfactual which is youth rebellion which is social instability. And so here is really an agenda item which is very big which is not the same as just inequality is closer to equity but certainly there is more to it which is that what you do in terms of income generation, whatever is going to take you above the poverty line also has to be satisfactory for the standpoint of a life program. Focus on vulnerability, a big missing element in the Millennium Development Goals. We know that vulnerability to uninsured shocks is a source of new poor. That vulnerability to uninsured shocks is also a very high cost in terms of economic growth. That a lot of irreversibilities are due to exposure to uninsured shocks. So that links to social programs in a very important way. But what in a sense we need to do is get a better understanding that there is a lot of instability. That instability is very costly but that instability can be managed. As a consequence, it is a very big research item that is coming in the years to come. The role of behavior, Joe Stiglitz talked about it quite extensively. If you look at the Banerjee DuFlo, it starts in a sense by saying behavior and experiments. Those are the two pillars of what J-PAL and this approach to research is doing. But in a sense, if you think of the Millennium Development Goals and the ones that succeeded and failed, it's quite likely the case that the ones that failed more than succeeded are the ones that require more behavioral changes than the ones that could be handled mainly via money transfers or via kind of hard instruments. So important to understand rationality and irrationality, important to understand how people are going to react to what is going to be proposed. One world development, I think that's quite obvious, the sustainable development goals arriving from the Rio agenda tell us that it's a shared responsibility. And that's kind of new if I look at myself in the trajectory of 30, 40 years of research and development. That's not the way we used to think some years back, namely that there are lots of zero sum or at least we need to collaborate and we need to coordinate better in the sense that we are jointly facing two issues which are going to have very strong externalities, very strong spillovers across boundaries. Environment, resources, trade, but also migration and kind of emerging issues which are going to be very high on the agenda. And then finally the focus on state and governance and governance which made its way into the sustainable development goals. Stiglitz very much insisted yesterday that there cannot be success without a strong developmental state. We know that most of the poor by the year 2030 are going to be located in failed state countries or in post-conflict areas. We are going to have to learn as to how to deliver aid in hard places where we cannot count on the institutions as easily as you can in the context of China or in Brazil or in Mexico. You have to either find surrogates to the institutions in terms of access and delivery but also very much on the agenda, building the state. In other words, in the longer run, those aid programs will have to be delivered via the state apparatus and hence creating a development state. Development state is certainly high on the agenda. And yet this is one we know, one of the elements about which we know least as to how to proceed with this when we start looking into it, every case historically has been quite different. One from the other is hard to extract kind of simple lessons and guidelines as to how to proceed towards the construction of a developmental state. So all of this is fine but what we see in a sense is that there is a cost to this mainstreaming. The cost to the mainstreaming is that the way we do, especially the microeconomics of development in this sort of elevated university context, if you like, has some potential cost of which we should be aware of in order to kind of think as to how to counteract those tendencies. The first one is that I think the micro link has been very severely lost. It used to be that when development was more a maverick field, we were all both microeconomists, microeconomists, but also trade economists and also quite interdisciplinary. Our friends in anthropology and sociology didn't quite know if we were with them or if we were in the econ department. It was kind of all one thing and it has the great advantage that we were able to kind of capture the multi-dimensionality of what is the reality of the development problem. Things have become much more partitioned. Things have been much more driven by the quality of the analysis and in a sense the kind of, especially the micro-micro link has been lost in part because the micro has not progressed as rapidly in terms of empirical rigor and satisfaction for the micro people. And then the second is kind of a loss of the policy commitment. The pioneers of development were all the Hirschmans and the others, Rodenstein Roden and the Liebenstein were always strongly policy committed and engaged into policy advice and policy making. The fact that now advanced research in development becomes more of an economic exercise. The prioritization of topics is different. It's much more the urge to public in top journals as opposed to the policy implications that they are going to have. If you think of the refereeing of journals, quite often they are more concerned with the rigor of the analysis than they are concerned with the content and the value of the policy implication. The field experiments, and there has been interesting debate on this and I don't want to go into the randomized controversy that Angus Dieter and others and Martin Ravellian and Mark Rosenswaig have important contributed to this. But it's quite clear that what we can do with randomized control trials is a subset of the policy questions that can be raised. Those things tend to be small and tend to be short run. They are focused on particular policy instruments which are mainly transfers, nudges of different types and subsidies. There has been a bit of a development populism in the sense that a lot of interest as to how do we transform poor people and true entrepreneurs, when in fact the way out of poverty is much more likely to be via wage employment in medium and large firms with good jobs where you have kind of wage progression as opposed to trying to transform people into entrepreneurs. The rate of failure in entrepreneurship is extremely high. The cost of failure is incredibly high because there is no risk sharing mechanism. And then you incur that and then you find yourself into bankruptcy without even having started a successful business. And then there is clearly in the context of those experiments, the loss of the second round effect, the loss of general recovery effects, those things are hard to sustain over time. So let me mention some of the results that we get via those approaches of field experiments. Let me take them into agriculture. I mentioned just a few issues. The first one in a sense is the importance of self-selection. Quite often what we want to do is to avoid placement bias and selection biases. In fact, in terms of what we do, quite often we are going to be better off with the self-selection of the right people who can, for example, make a better use of the cash transfers, achieve higher multipliers. I think the work that Kerry is going to be describing, achieve higher multipliers out of cash transfers. If you want to have a demonstrator for your new seeds, what you want to have is a farmer who is going to be able to use those seeds productively. So in fact, quite often what we want to do is to start our experiments with a self-targeting, a self-selection procedure, a bidding process, a rolling transaction that allows you to reveal types and hence to choose with whom those experiments are going to be run. We have to be very careful as to what it is going to mean subsequently in terms of external validity. But it is one way by which we can secure more success in the experiments that we make. Second, we can use those randomized control trials to estimate a full demand function. One of the big puzzles, I think, in our field is the kind of vanishing demand for what we consider to be useful product in health, in fertilizers and seeds and vaccination, et cetera. High demand with full subsidy, but very highly price-elastic demand, and the demand collapses way before it reaches market price. So what we need to know is should we subsidize? Is it a one-time thing? Should it be recurrent? Or is there kind of dynamic learning, especially if the outcomes are themselves stochastic, to only learn when, say, for example, a bad shock happens as to what is the value of an insurance product? But then what you learn is also something that you de-learn, because if there's no shock during a certain time, you're going to learn what you have learned in the context of a previous shock. So it could very well be that those optimal subsidies are in fact permanent subsidies that need to be constantly adjusted to past events, to how much learning and de-learning there has been, and hence we kind of use those randomized control trials to estimate a full demand function and to, as a consequence, adjust the subsidies that should be consequently offered in order to sustain a desired level of demand. Puzzle of low uptake of index-based insurance. I think that games in the field have been quite revealing here that basis risk is a very strong deterrent to the demand for index insurance, that in fact index insurance should be elevated from the individual level to the institutional level, and then to count on internal redistributive mechanisms whereby premiums that are being paid to a group, to a cooperative, to a state, to a governor who is insuring his programs via an index insurance, how those things are being redistributed internally. Risk-reducing technology, flood-resistant rise, for example, what we find, which is quite interesting, is that sure, they are protecting yields in bad years, but they're also creating incentives to invest better and more in the normal years. And it could very well be that over time, you have expected gains which are as much derived from what you do in the good years, knowing that your bad years are going to be protected than the avoided losses in the bad years themselves. And then finally, the importance of learning in the context of heterogeneity, that what we find is the extension services have typically not worked because they are addressing an heterogeneous clientele without adjusting what is being said. And that we need is much more customization or recommendation to allow for heterogeneity that people in social networks tend to learn more not from the big people, but from people who are more like them. Why so? Because they know that heterogeneity is important. They will only use the observable part of heterogeneity to notice what other people looking like them are actually doing. And it could very well be that instead of, if there is a lot of heterogeneity, instead of making recommendations, it's better to help people make their own assessment. So they will be able to adjust what they do to their own conditions. Those conditions are going to be changing anyway. So for example, an example is leaf color charts. You help people tailor the fertilizer application and the pesticide applications they are going to make to the way the plant looks like and you give them a way to assess in a simple fashion what the plant seems to be needing. So instead of telling them what to do, you give them a mechanism whereby they are going to assess what they should be doing. So let me finish with just a few recommendations as to how to do better in the future than what we have done in the past. I think that there's a very big issue that the rigor of revolution is creating a certain degree of myopia in terms of what we do. We kind of fail to elevate our sites. We fail to address the big issues. And that's why I look at an institution that's being so important and this kind of periodic chance that we have of coming back here. It allows us to kind of look beyond the demands of rigor to look at the big issues and to think, well, how do we go about using what we can do in order to fit our questions to what are the big challenges and the big gaps in research. I think there's a serious need for professional guidance to achieve more relevance. First you rigor, yes, but with a plan. That is, we do a succession of experiments but those experiments should be expected like converged towards something which is bigger than what each of those experiments is doing. And for that, we also need to have some kind of a vision as to what matters in the end and how we can break down the successive steps that we're going to take research-wise in order to converge towards this bigger issue. We need to inform external validity. In general, we are very concerned with internal validity and respect. If you go to seminars on campuses now it's all about internal validity and endless discussions as to whether the criteria of rigor have been sufficiently satisfied. But we fail to talk about where was the external validity. We should, in each of those experiments, make significant effort to characterize the external validity so that we can accumulate what we could call mega-domains. Place by place, where do those results apply and what are the places as a consequence where policy recommendations are going to apply out of experiments. We obviously need to identify channels but for this we need to plan the experiments to identify the channels. We need to measure general equilibrium effects. The first results, the immediate results of most of the experiments are quite fallacious in terms of subsequent implications. The prices are still high. There's little competition. The soil has not been exhausted. So second round, third round, general equilibrium effects will be quite different from what has been measured here. So here to avoid the fallacy of aggregation we have to be able to either sustain those experiments over a longer period of time or then go into empirical modeling where we model those phenomena using the experiment to identify the parameters that are going to be used to construct those models and then to engage into policy simulation. Suddenly we need to use more natural experiments back to the big data which allow us to take a broader perspective, a longer perspective and to incorporate into results the general equilibrium effects that we see. And then finally policy relevance. I said that in a sense the mainstreaming of development economics takes it away to some extent of immediate policy concerns. But we have to be careful of sort of bring those experiments, bring this research back into contact, intercourse with government demand, with institutions so that there is more likelihood that the results of our research are going to be policy relevant. Thank you very much.