 Thanks, I want to thank the organizers for inviting me and thank Alon and Karen for covering so much ground before I start so that I can focus on a couple of narrower issues. I want to start with just a picture, which is not going to be a surprise to anybody here, but which shows non-agricultural employment labor force of the active labor force at the country level where the vertical axis is the share who are own account workers and the horizontal axis is the share who are employers and the square boxes are sub-Saharan Africa and the diamonds are Europe. Basically the point is half of the labor force in sub-Saharan Africa is an own account worker, very large numbers of self-employed, at least in the cross-section and we can ask whether this is true in longitudinally as well, but at least in the cross-section development is a process of moving from the upper left towards the lower right, that is firm sizes grow, the large firms become larger as incomes rise, at least in the cross-section. So what do we know about moving downward and to the right on the graph? I would argue not very much, at least yet anyway. And what I want to talk about today is some of the small progress that we've made and why progress is so difficult using the techniques and methods that have been popular in the last ten years, particularly field experiments and what's out there and what's going on now and what I see as a very new and exciting area. Let me step back and say I think if I look back thirty years and ask about development research, my view is that firms have been tremendously understudied and have been sort of marginalized in the research agenda. This reflects I think Santiago's point this morning that productivity matters even for things if we think about inequality and so forth, we have to think about what's going on with firms, what's going on with the private sector. So we can ask a series of questions, where do large employers come from? Are large firms born? Can they be built? Are exports necessary to produce large firms, to generate large firms and so forth? I want to focus on one and this is far from an exhaustive list. I want to focus on one, do the micro ever become medium or even small, what do we know about that? Well, we can think about just a standard AKL production function. Firmers take capital and labor, they combine it with technology, ability, other kinds of other ways that they become more productive and we ask what are the constraints to growth? Well, it's likely all three of these. It's likely labor, capital and skills or technology. I'm going to focus first on capital constraints given that I have limited time and also for a point that I'll make later, I'll say something about the A part, the technology or ability part of the production function. So this is a place, Alon said that the micro has disconnected from the macro. I think in firms research, this is a place where the macro has sort of driven in the last three or four or five years especially a renewed interest in firms and especially in large firms. So at the macro level, there's evidence of serious misallocation of resources across firms and across sectors and across individuals. Banerjee and the Flo's handbook chapter pointed this out, Cenk Deshe and Pete Cleanel pick up on it looking at firms in India and India, China and comparing them with the U.S. in a 2008 QJAE paper. That paper's probably done more to generate research and development on firms than anything that's been done in the last 10 or 15 years. We've known for a long time or we've thought for a long time that there are constraints on access to credit, constraints on capital, particularly among the poor, but until recently I'd argue that convincing evidence on this has been lacking. So what does the more recent evidence say? So I want to say I think there's something of a puzzle from several field experiments and I've listed self-servingly three that I've been involved in here in Sri Lanka, Ghana and Mexico, but there are others as well. We have evidence of quite high marginal returns to capital in small enterprises. So these are all projects which were field experiments in which capital was given to micro-enterprise owners, existing micro-enterprise owners, and we look at what happens to profitability sales, other measures of outcomes for these firms. The Sri Lanka work was the first, this is all work that I've done with David McKenzie and others, the Sri Lanka work with Suresh Damell as well. The Sri Lanka work was the earliest, we've done a six-year follow-up, we find continued higher returns among the people that got a small grant six years earlier, we're just about to go to the field with a ten-year follow-up on these same firms. But we find returns of around 6% per month much higher than microfinance costs, or microfinance interest rates. Similarly in Ghana, similarly in Mexico, we find marginal returns which are far in excess of standard interest rates. And yet, we now also have several field experiments on microfinance itself. There was a special issue of the AEJ applied in January 2015 that reported on six of these, D-Mexico, Bosnia, Ethiopia, Mongolia, maybe they're only five, I think they're worth, maybe either I've missed one or they're only five. But in any case, reported on several of these. And I think the takeaway from that is that the effect of microfinance on businesses and on firms is fairly modest, it's underwhelming. And so there's a puzzle. If firms have really high returns to capital, why is microfinance not having a larger effect? Why don't we see a larger effect? So puzzle is a good result. And I think it's an example of how the research gets pushed forward when we have evidence that people find credible and that they're able to move back and forth on. The puzzle stimulates thinking about why the differences are there. And I think there are a couple of interesting experiments, one that's underway now, but one that's by Erica Field and others that was published in the AER last year, where they said, look, maybe what's going on here is that the need to start repaying immediately on these microfinance loans is a constraint to risk taking. I can't make an investment that has any kind of long-term return because long-term, even two months or three months, return because I have to start paying right away. So they offered a subset of the borrowers a two-month grace period. They show up for the loan, unannounced, they're offered a two-month grace period. They don't have to start making payments for a period of two months. And what do they find? The good news is it works. Borrowers who have the grace period invest very differently. They invest more in the business. They take riskier investments, and those riskier investments have much higher returns. They estimate the returns of around 9% a month from the investments that are made by the people who are given the grace period. But the bad news is that default rates also go up from about 2% to about 9%. And when they sort of put the number to it and look at the numbers and develop a sort of very simple structural model to ask, does it make sense for microfinance institutions to offer this kind of a contract, they say no. So obviously if they offered it, they'd have to raise interest rates. If they raise interest rates, they're adverse selection, moral hazard problems, and so forth. And they sort of take that through a model and they say, by the time they get to where the equilibrium interest rate would have to be given the selection process as we go, no one would be left to take loans at that interest rate. But I think this does at least say maybe there's something, and this is far from the final word, but I think it's a way of saying that... This is an area where I see there's been some progress with the methodologies that have been quite common in recent years with the experimental techniques. It's an area where the experimental techniques have taken us forward. But capital is the easy one, because if I believe that the only constraint to the enterprise is capital, there's only one form of capital that matters. I just give them cash. Before I go and ask them what do they want to buy, and I buy it for them. So if I believe that's the only constraint, then there's just one thing to design the intervention route. We still talk about external validity, we can do this in different places with different samples, that's still going to matter. But it's a potentially solvable problem. The dimensionality is not too big because there's only one relevant intervention. If I think about the A in AKL production functions, with respect to micro-enterprises, typically we think of this as training for the entrepreneur. David and I have a review of 16 experiments, this is from a few years ago. So at the time, 16 randomized experiments on micro-enterprise training. And the punchline that we take from that is we really haven't learned anything at all from these experiments. And what's the problem? The problem is that there is no single A, it's not like A, you can't give the same training to every populate. Some of these 16 experiments are on university graduates, some are on people who are illiterate, some are women, some are in specific sectors. The obviously some are rural, some are urban. There's so many different contexts. What's relevant as training varies in so many different dimensions that I guess I wonder if we're ever going to make really substantial progress on figuring out what works for training through sort of standardized randomized experiments. I think this is a place where we've reached the limits. Now that's not to say that there aren't some good experiments that we can't learn something and that we can't learn something about specific policies and specific contexts. But are we going to learn something general? It's going to be a whole lot harder because the dimensionality is so much greater when it comes to these. And micro-enterprises are the easy one as well. It's not clear that moving from the upper left to the lower right is about getting small businesses to grow. There is some evidence. John Sutton has done some very interesting work on sort of looking at the histories of large enterprises in several African countries that suggest that most of them started not as micro-enterprises in manufacturing, say, but as traders in some other sector. And then when they started the actual firm that's large, it was large from the beginning. There are a lot fewer large firms, so it's a lot harder to do a large randomized experiment with large firms. They're more challenging to work with. Suresh and Trilanka would occasionally get a call from one of our micro-enterprise panel people who said, who would say, nobody came to visit me last month. Aren't you going to come? I'm doing some work now with large firms in Bangladesh and let me tell you none of them ever call us and say anything like that. In fact, usually what they call and say is you were supposed to come today, don't. It's much more difficult. It's much more challenging to work with large firms in this kind of way. But I think there are other opportunities out there and there's a lot of work that I think is very exciting that's going on now. There's now, I run a very large grant, organized a very large grants program for DFID on private enterprise development in low-income countries. We funded a lot of projects. The International Growth Centers funded a lot of projects. There's now an increasing amount of funding going into this area. And I think there are opportunities in a couplers. There are opportunities in administrative data. Alon mentioned this as well. A lot of firms have very good administrative records. IT costs have fallen a lot faster than analysis costs. There's an opportunity in a sense for us to say you have all these data. You have some people who are able to analyze them, but those people are even busier than we are. If you share the data with us, we'll give you some reports back and we'll help you understand what the data is saying. And we found that some firms at least are willing to open their administrative data under those kinds of terms. Government data also exists in a lot of countries. Unfortunately, at this point, access still tends to be pretty personalized. The program I run is funded by DFID and they want us to fund only things in the 35, now it's 31 low-income countries. We fund things in Brazil because Brazil has very good matched employer-employee data, something that exists in some of the low-income countries but is currently unavailable. So I think there's an opportunity there if we can figure out how to get secure access to those data. But I want to say I think we're beginning to see progress on large firms from a variety of methods. Some of it is experimental. I'm doing some work in Bangladesh and other places in garments that's experimental, but a lot of it is not. I've mentioned John Sutton's enterprise maps. Nick Bloom and co-authors have done a lot of work. Nick and John VanRenen have done work mapping management practices across countries. There's Nick and some others have an experiment on management practices in India. And then I think we're beginning to see a return. When I was in graduate school, industry studies was sort of a pejorative word. It was something that you didn't want to be doing. I think we're beginning to see a return to industry studies. And again, this goes back to something Alan said where I think it's people diving into an industry really trying to understand the institutional structure of the industry, the microanalytics of the industry. And several examples, David Atkin and several others have projects on soccer balls in Pakistan and rug exporting in India. Dan Keniston is doing some work on the brick sector in India, Raka Makavello and Amit Maurya doing work on coffee and so forth, and I'm working with several others on garments and places. And I think we'll begin to see some interesting work coming out of these projects. They're not quick projects. Dan Keniston's been doing survey work for three or four years. I don't think he has a paper yet because the survey work is all set up to kind of thinking about how do I design experiments? How do I think about getting variation? How do I think about using the data? So I think in common with other topics in economics generally, we've had an evolution or at least with development, but we've had an evolution in research methodology shown from theory intensive work and big kind of thinking to more observational data driven work and experimental work. And I think now we're sort of pushing back towards theory. I think the best work that I see is work that's combining experimental work and theory. And so let me just sort of give a couple of thoughts and concluding thoughts on the bigger picture. So when I was in graduate school, I remember somebody one time mentioning Ronald McKinnon, and that person was sort of ridiculed because the idea that capital markets could explain under development was just seen as something that was just crazy. There are too many other things going on. That couldn't be part of the, that couldn't be part of it. I think we've moved a long way from that. But for the first half of the last 30 years, at least the formal domestic private sector has hardly been on the development agenda. There's been a lot of work on informality since from the 70s. There's been a lot of work on informal firms. There's been work on trade policy and trade. But if we think about the domestic formal private sector, there's been much less. So I expect when we have the 50th anniversary of UNU Wider that we'll see the private sector playing a much bigger role, I think the work that's going on there now finally is catching up with the rest of development. And it's using a mixture of the methods that have been prominent in the last 10 years, experiments and so forth, and new administrative data from various sources. So I'll leave it at that. Thank you.