 It's a great pleasure and honor to be at this conference and have a chance to present. I look forward to learning a lot today and tomorrow. Before I even get to this slide, let me start by saying that the presentation I'm about to give is a contrast. I hope a useful contrast from the keynote we just heard. Most of the way through Marcelo Neri's wonderful presentation, I was thinking to myself, we don't really need my paper, the Brazilians have figured it out. And then at the very end, when he said, well, maybe we should worry less about inequality because growth is slowing down and the tax burden is so high. And I kind of thought, okay, well, that's what my paper is about, so that's useful. And thank you for that. We take a big picture look at these issues in the sense that we step back and look at what the cross-country evidence has to say. You know, in my own work, I find that when I work on the cross-country evidence, I wish I could look at the details more closely because it's, you know, ultimately the big picture when you look closely, the pieces are blurry. On the other hand, sometimes when you look at the details, you do want to see how it fits together or what the broad patterns are, and that's the spirit that we've been doing this work. So this line of work started many years ago with my colleagues at the IMF, Jonathan Ostry and then Jeremy Zettelmeyer, when we were looking not at inequality, but at the sources of sustained growth. And it actually came very initially out of the idea that we might, we were thinking in terms of crisis prediction, being the IMF, and we thought maybe if we could predict when growth spells were going to end, we might have a handle on when certain sorts of crises would come. So we wrote a paper that came out a couple of years ago on sources of sustained growth, and in that project one of the surprises was the salience of income inequality as a predictor of weakness of growth spells, that's to say the hazard that growth spells would end. And then Jonathan Ostry and I drilled down on that, or really went, ran with that particular result in a 2011 paper where we talked about the role of inequality in sustained, or of equality really in sustaining growth spells. So that's part of a literature that has, a big literature obviously, that has talked about the role of inequality, the role of inequality in growth, and I'll review it briefly a little later. But that has come on and since the 2012 crisis, and since of course the rise and the discussion about the rise and inequality, especially in advanced countries this topic has come to the fore. Now, we would tend to say that there is a tentative consensus that inequality indeed is associated with slower and more durable growth, at least over the long horizons in a sort of cross-country sense. When we finished the last paper, our 2011 paper, it was very hard to write the conclusion. This was supposed to be a policy, it was a staff discussion note, it's a policy oriented note, and we were trying to draw policy conclusions, and we drew some policy conclusions, but there was a key problem with drawing conclusions, which is kind of precisely what we really heard about at the end with respect to Brazil, which is that it's not clear really what is, what's driving the relationship between inequality and growth, and in particular there's a theory, the fiscal, there's a theory that one of the reasons inequality may lead to lower growth is precisely that countries try and do something about it. So the story would be that Brazil's high inequality at a sort of longer horizon has led to low growth exactly because it led to a consensus in favor of redistribution, which indeed has a lower growth. Now, by no means making that claim about Brazil, I'm just using that example to illustrate the argument. And that's in the literature, and that sort of makes it very hard to draw any conclusions about the policy implications, or at least some of the policy implications, about the fact that high inequality may cause lower growth. And another way to put that is there's very deeply embedded in the economics profession this view about the trade off between equality and efficiency. There's a book by Arthur Oaken from I think 1978 that was Brookings' best-selling book ever, at least until a few years ago, called The Big Trade Off. And Thomas Sargent in his, the Nobel Prize winner in his speech to the graduates, listed 12 things economists know, or eight things economists know, but it wasn't a very big number. And one of them was that there's a trade off between equality and growth, or some such trade off. So for all these reasons we wanted to analyze simultaneously the relationship between inequality, redistribution, and growth. So one relatively narrow objective was to see if controlling for redistribution we still see the same effects of inequality on growth. And the other was to try and get a handle on sort of the Oaken trade off, and does it exist in the data. So we look at the effects, when I say growth we mean two different things. We look at the average rate of mean per capita real growth over five year periods. And I'm going to usually call those panel growth or growth regressions. And we look at the duration of growth spells, which is how many, if we define growth spells in a certain way, these are sort of episodes of reasonably high growth, and you can sort of define when they come to an end, and you can ask about the risks that they will come to end and the determinants of those risks. We use a, it says recently compiled, the paper was in 2009, it's recent in some sense. A recently compiled cross country data set by Professor Sold that's based on the WID data, mainly, that distinguishes between net and market inequality and allows therefore for the direct calculation of redistribution, which we define simply as the difference between the market and net genie in a country. So it's not redistribution defined in terms of spending or tax rates or any effort at redistribution. It's just what seems to show up in the data in this sense. And I'll talk about that obviously, the limitations of that in particular. Seeing Nora in the audience is always scary when I present this kind of thing because I know what she's thinking, I think. But anyway, I'll talk about that. One of our main conclusions is that lower net inequality does seem to drive both faster and more durable growth for a given level of redistribution. So we recover our old results and now we have done it also in a growth panel as opposed to just duration. And we find redistribution in general, we find little evidence of a negative effect of redistribution on growth. In extreme cases, which I will define, of course, there is some evidence that there can be a direct negative effect of redistribution, but the combined direct and indirect effects of redistribution are on average pro-growth. And by indirect effect, I mean the effect of redistribution on inequality and thereby growth. So this graph may help, or it may not, but it may help to define terms a little more carefully. Of course, I'm using Gini coefficients, which most of you know, but vary between, it says, 0 and 100 or 0 and 1, where 1 is completely unequal and 0 is complete equality. Market equality is inequality of factor incomes before taxes and before transfers. And then we have redistribution, well net inequality is inequality of income after taxes and transfers, again measured as a Gini coefficient. And the difference between market, am I able to point in some way? I guess I can use this. The difference between market and net, of course, is redistribution. So we're going to look at several of these channels in this paper, but not all of them. In fact, there's many arrows that we could have drawn that we haven't, and we don't look too much at those. But one reason we think it's important to look simultaneously at redistribution and inequality is that there's a body of evidence and thought that suggests that redistribution is a function of inequality. That's this line A. It's the so-called Meltzer-Richard effect, which is that more unequal countries tend to redistribute more, at least that's the claim of some people in the literature. And then there's the potential direct effect of redistribution on growth. This is kind of the Arthur Oaken effect, the fact that high tax burdens can lower incentives to invest and that there may be labor supply effects of redistribution and so on. And then there's the direct effect of inequality on growth. There's various channels. And then the indirect or the total effect of redistribution would be the effect of redistribution on inequality and thereby on growth. Now, maybe the most interesting lines in this chart are the ones I haven't drawn. For example, redistribution could affect market inequality. Now, when I talk about the regressions, for some purposes that's not a major problem for us in terms of running our regressions, but for interpretation of some of the results, it's just going to be a maintained hypothesis, which I will talk about. Another is from growth to either inequality or redistribution. And we make some effort to isolate causality in the direction of these arrows, D&E, but it's always a challenge in these kind of exercises and it's a caution. So let me briefly review the evidence. I don't know if this is redundant or it's just going to be contradicted in the next presentation, but we'll find out. But there's a number of reasons why inequality could promote growth. Of course, it provides incentives. And we know, in fact, well, we know that at some point too little inequality can't be pro-growth. I mean, and so some of these channels must exist and it's just a question of this relative strength. But anyway, there's many old stories about that. There's a Caldor kind of story that rich people save more and invest more and so inequality could promote savings and investment. Barrow has an argument that this obviously depends on how rich the country is, but in relatively poor countries a higher level of inequality allows more people past the hump where they would have enough income to start a business, basically to generate some capital and get a business to the minimum required scale. And then as far as the evidence goes, obviously the literature is divided. There is some evidence that inequality promotes higher growth. We tend to see that the way we read it is that if you look at the variation within relatively short time periods within countries, you can get positive effects, negative effects, or no effects or nonlinear effects. It's a bit of a mess. So we tend to think that you do have to look at longer horizons and cross-country differences to get... Anyway, this support of evidence is from the short horizon, typically. So then there's reasons why inequality itself could promote growth. Obviously, it helps support, stay healthy, and accumulate human capital education and so on, because you might, of loosely speaking or broadly speaking, capital market imperfections that we know exist. It can support political and economic stability that promotes growth, perhaps through higher investment and so on. Danny Roderick has an old paper about the 80s crisis arguing that if you look at the terms of trade shocks and the global shocks that took so many countries onto a lost decade in the late 70s, early 80s, the punishment was a lot worse than the crime in terms of the size of the shock. It was small relative to the lost GDP and that if you try to understand who suffered so much and who didn't, it was sort of the unequal countries that had trouble adjusting to the shock and lost their way. In fact, Jeff Sachs, while I was student with Jeff Sachs, we wrote a paper on the debt crisis arguing that that was sort of a way of understanding which countries got into the debt crisis most in the 80s, following that same idea. And there's substantial empirical evidence supporting the idea that the quality can support growth. Now, we don't focus as much on this last and this next channel in the paper, and I won't talk about it so much here, but the evidence on whether market inequality does lead through political channels, does lead to more redistribution is mixed, I would say, and I'll come to that when we talk about the data. So then on redistribution, whether it will help or hurt growth, in a way there's relatively little on this topic, it seems to me, depending on how you phrase the question, but at the aggregate kind of level, there's not that much about it. Of course, there's a million papers on, there's a lot of papers on the effects of, say, taxes and sort of micro papers on labor supply effects of different specific policies and this huge literature on labor supply effects of things, but putting it together is harder. After all, redistribution, anyway, so it's, but there's, of course, the Oaken idea. There's some papers arguing that redistributive policies could help growth themselves, but those papers, in a way, combine the effects of redistribution itself and the effects of redistribution on equality and on growth. They don't necessarily distinguish that so much. I mean, the main, I think in the Beninbu paper, there's an idea of social insurance associated with redistribution. If you know that you can take chances and your failures will be alleviated but that may encourage innovation, which is sort of a direct channel from redistribution to innovation and growth. Anyway, we conclude that it's an empirical question and that it could go, that we shouldn't be so sure, as Sergeant is, about what the overall aggregate effects are. So the data. Now, most papers in the growth equality literature have kind of not paid that much attention to redistribution or really to the definition of equality in terms of is it market equality, is it net inequality that's being measured? And that's partly because the data sets sometimes often kind of contain, like the WID contains that information, but it doesn't make it, it's a hard problem to deal with. Sometimes people make kind of crude efforts to make an adjustment factor for the difference between market and net inequality and so on to make them comparable. But it's, you know, there's definition, in this crowd I guess I don't need to go through this, but there's lots of different ways you can measure inequality and the surveys in the cross country data sets vary in all these dimensions. This makes it hard to interpret growth regressions in, should we think in terms of market equality, market inequality, net inequality, and where is redistribution in these coefficients? It's also very hard to measure redistribution directly in general. Measures of redistribution, like the size of subsidies and transfers in the budget or the size of the government or the size of the tax take, those things are kind of correlated with our measure, pretty highly correlated with our measure, but the correlation coefficient sort of varies from 0.5 to 0.7. And most papers that, and so when you put something like the tax rate or something like that on the right-hand side, it's hard to know how to interpret that as a redistributive, as a measure of redistribution. Bronko Milanovic has a paper in 2000 looking at the question of whether unequal countries redistribute more and he actually measures redistribution directly kind of like we do, or I should say the other way around, we do it kind of like he did. He uses data for, you know, the data is hard to find so he does it just for the OECD and he finds relatively strong evidence in favor of this effect, this effect that unequal countries do tend to redistribute more. He once he measures redistribution correctly or more accurately. So we use this Solt standardized World Income Inequality Database that came out in 2009, he's been updating it. And briefly he takes, you know, the best, he takes the survey data that's out there and he uses an essentially regression-based method to try and create, you know, he categorizes every survey into one of, I think, 21 categories in terms of unit of account, you know, is it household, is it individual, is it net, is it gross, is it disposable income, is it consumption. And then he looks for the statistical relationships between these 21 categories and tries to kind of boil it down into one net and one market inequality measure for each country. Now in doing so, he does make use of, you know, there's some imputation and there's using these regressions to infer how to compare these different surveys. But the result is we get a large sample in principle comparable across countries in time of net and market inequality, which is what we need because for our techniques we need a big sample. For our regression techniques we need a big sample. So we see it as not perfect but the best for our purposes. We do quite a lot of sensitivity to try and see how, you know, different sub-samples affect the results and so on. And I'll show you some of that. I'll show you some of that. So just some stylized facts that we'll spend a lot of time on. The left panel is for the whole sample then OECD and non-OECD and OECD. So global median net inequality has been very stable but it's been declining. Market inequality has been rising in the OECD and declining in the rest. And the gap, of course, one important fact is that the gap between net and market is much higher. That's to say between, for example, this line much higher in industrial countries than in developing countries as we know that developing countries have less scope to redistribute. We heard from Brazil that it can be done. So anyway, that's just some of the patterns in the data. This is just a picture. This picture shows the inequality of market income on the horizontal axis and of net income on the vertical. So if a country doesn't redistribute the point will be along this line. And you can see here for the OECD that there's actually a relatively weak relationship between market inequality and net inequality. In other words, most of the difference across countries in market inequality on average is undone by the taxes and transfers of the fiscal system. Now that's true. This is a snapshot. We know that of the increase in the last 10 or 20 years a smaller than sort of typical fraction has been undone by redistribution. But anyway, that's... And if you look statistically at these relationships they hold up... This holds up in the OECD. It holds up in the non-OECD although it's weaker but it's statistically significant. I don't think I'm going to show that but that's just a regression of that picture. So we're just looking at the raw data. Here I'm just graphing the genie of net income on the horizontal axis with growth over the subsequent 10 years. Mean real per capita GDP growth. And you can see a fairly strong negative relationship. Now I don't know if those of you who've looked at, say, Barrow's book, you know, he shows a lot of pictures like this you can look at this two ways. Either it's a total cloud and, you know, there's just a bit of a slope or you can say, wow, that's clearly, you know, a strong relationship. But this is, as these things go, it's pretty good, it seems to me. You know, it's not as good as investment but some of the things we think about like good institutions or something don't necessarily do that great in these pictures either. And then for redistribution we see essentially no relationship or if anything a slightly positive relationship between redistribution in genie points and growth over the subsequent 10 years. Now this is for spells. I'll define spells later when I get to the results if I have time but we're looking at the net genie at the beginning of the spell on the horizontal axis and this is the length of the subsequent spell. And so they vary, you know, there's some that are 40, 50 years long and presumably get truncated at the end of the sample. Many of them do get truncated. Many have kept going. And you can see something of a negative relationship. For redistribution, here you do see a bit of a negative relationship but it seems to be driven by these high redistribution a small number of observations. So what do we do? Our basic, I said we have these two sets of results. One is kind of traditional growth, panel growth regressions where we put the five-year average mean per capita growth rate on the left. We put initial income on the right. So column one is our baseline. We put initial income on the right. We put net inequality contemporaneously and redistribution contemporaneously. And then we use system GMM which means essentially that we're using lagged variables as instruments. And I'll come back to that. And what is our basic result is that net inequality has a significant negative effect on growth and redistribution has sort of no effect. I mean, it looks, the sign is positive but it's highly insignificant. Now, what we do, we don't, we're not really interested in the definitive growth regression here which not that it exists but so we added progressively various variables just to see how these results hold up. So we added investment in population growth and then investment population growth in education to put in the basic factors and then we kind of threw in a bunch of other stuff. You know, debt, political institutions and so on. And for, for, for our purposes here the main point is that inequality really holds up strongly in this business and that redistribution is never significant. And in fact, the difference between those two is significant. We can estimate redistribution precisely enough that we can say that the coefficients are not equal. So that's our core, that's our, this is our core sort of test of the Okan hypothesis. We thought we would get a negative coefficient on redistribution because of, you know, redistribution being bad for growth a negative coefficient on inequality and we would see which one was more negative. That's kind of a test of the trade off. But in fact, we didn't find any negative coefficient really on redistribution, at least in these regressions. So there's no evidence of a trade off here. See how I'm doing on my list of conclusions. So I'm on to the fifth one. Yeah, so the results are just not consistent with the notion that there's this big trade off that's really salient in the data. I guess I'm done with this. Okay, so these are these results graphically. This is the effect of, the first column shows the effect of increasing the net genie from 37 which is like the U.S. to Morocco, which is like 40. We didn't pick these with any, except because they fit the numbers. And that reduces growth by 0.5 percentage points from 5 to 4.5 percent holding everything else constant. Five minutes, okay. And then the direct effect of redistribution is almost zero. And then this total effect combines the direct effect of redistribution with the resulting effect on inequality. Now, in terms of this issue I raised at the beginning about distinct, well, about causality, to estimate the direct effects, we don't really care what the relationship is between inequality and redistribution. Once we've got them both in the regression, we can estimate the direct effects. It's true that for this last column where we want to say what the effect is of redistribution, we are maintaining the hypothesis that market inequality is not affected by redistribution. Now, we know that can't be right. We can't be wrong about redistribution. I want to make one point, which is that redistribution in principle should, for example, lower the labor supply at the top and the bottom, because you tax the rich and give to the poor, you can reduce labor supply at both ends. What the effect is on the relative wages is not clear. It's kind of a second-order effect, so to speak. So it's not obvious anyway that redistribution as a general matter should have a big effect, but we have effects. I don't have time to talk too much about the GMM issue, how we really isolate causality and whether that works, but we do subject to a number of diagnoses and we find that our results, for example, are robust to weak instrumentation, you know, and I can maybe come back to that. We get the same results even if we assume that our instruments, our lagged instruments are not strongly related to the variables of distribution. So about the data. So there's various ways you can use this sold data. You can use every observation or you can apply some screens to try and get the highest quality data. Our baseline throws out a bunch of countries that sold things are relatively unreliable. It also only keeps a redistribution measurement if there's actually an actual series on some net inequality concept and some gross inequality concept for that country in question, so that you're not just using other country information to infer redistribution. Now we have various restricted samples, including one, for example, in which we keep only the five-year only an observation, which is a five-year average, if there's a net and a gross a net and a market measurement in that five-year period for that country. So we lose almost half the observations doing that and then we also put some other restrictions from in terms of throwing out questionable countries. And our results hold, essentially. Sometimes we lose... The result that inequality is significant and negative, that redistribution is not significant, the difference is usually significant, those hold even in these restricted samples. Now I don't have much time, I guess, to talk about spells. I have two minutes. Spells... I'm not going to say how we define growth spells, but we basically there's some interesting results on nonlinearities. As I hinted at the beginning, it's obvious that there must be a nonlinear relationship in all these variables. At some level, if inequality is zero, then increasing inequality is very likely to be pro-growth, and you think about China in 1980, for example, or the transition countries. In those previous regressions, we explored nonlinearities, looking if it mattered whether you already had higher or low inequality, and we found nothing in the data. It must be there, but it wasn't in our... Within the range we observed in the data, it was not there. In this growth sample, we did find that once you redistributed more than 75% of the sample, you started to see negative effects on the duration of growth. And so we divide into redistribution in the top 25% of the distribution of redistribution, that's to say the countries that redistribute the most, it can be adverse to the duration of growth. In this table, the number should be understood as like an odds ratio. So 1.06 here means that if you have a higher genie by one point, there's a 6% higher risk that the spell will end in the next period. So here, if you redistribute and you're already redistributing a lot, there is a 9.8% chance that the spell will... More, there's a higher chance that the spell will end. So we... We find... Yes, for redistribution, there is evidence of a nonlinear relationship. So we find that for countries below the 75th percentile, there's no negative effect. For countries above the 75th percentile, there may be one. And it's about equal to the magnitude of the effect of inequality. So even if you're above the 75th percentile, if you redistribute more, you know, that may not help growth. It doesn't seem like it hurts growth. And, of course, it may have its own merits on its own. And this is just a picture showing who's above the 75th percentile. You can see the UK and Australia are above. Canada's just below and so on. Now, the spell's results are more sensitive to the sample sizes. You're just asking more of the data. These spells, you have to measure the beginning. You need more data to get any kind of results with spells. So let me just conclude. My main conclusions already gave. Let me talk about caveats for one minute. Clearly, we don't want to overinterpret these results. They're just regressions. They're cross-country regressions. I think we haven't had a chance to talk about it. I think we made good effort to isolate causality, but we can't claim that it's iron-clad. It's very hard to do, that's for sure. And also, these are not controlled experiments. So we don't, you know, one of my colleagues at the IMF pointed out that countries may sort of be doing the amount of redistribution that they appropriately can do. We're not saying that Benin should go redistribute like Denmark. They don't have the institutions to redistribute like Denmark. So, you know, obviously it's dependent on the capacities of the country. It's dependent on the specificities of the case. And there's a number of caveats on those lines. We're just saying that this kind of if there were really a strong effect that redistribution, that Oaken was right and we should be afraid about redistribution, and that should be sort of the dominant thing in the data, we should see it, and we don't. So on average and across countries in time, government's efforts to redistribute do not seem to have led to bad growth outcomes, except possibly when they were extra. And I'll stop there. Thank you.