 So I'm going to take you east to South Africa. This is a slightly strange paper because the paper is based on a statistical observation and hence the title. And the statistical observation came about just through doing a standard. In fact, it was a background paper for the World Bank. We were just looking at standard labor market indicators and we thought, OK, let's look at real wage growth since apartheid ended over the percentiles of the distribution. So I'll say that again because that's quite important, that we looked at the average annual growth rate in real wages since 94, since the demise of apartheid across the percentiles of the wage distribution and the curve was U-shaped, which was very surprising to us. That suddenly we had a distribution and wage gains that seemed to suggest something similar to what is happening in the northern hemisphere, at least in Europe and the US, which is this hollowing out of the middle of the distribution. And so began the paper in a way, which was to try and figure out what was going on. Why did we observe this very peculiar, at least what we didn't expect, statistical outcome for the South African labor market? And essentially the paper takes you through firstly the explanations that we put forward, then to give you a little bit of the flavor in terms of the descriptive overview, and then we dip into quantile regression to try and tease out whether the results hold true in the sort of conditional environment. So as you may know, South Africa is probably, and certainly given the Latin American experience now, we can say we certainly, probably the most unequal country in the world I think. I guess the short little footnote would be for countries that are not experiencing war or famine or sort of structural crises. And the genie, depending on the data set, depending on the time period that you use, and so on, hovers around 0.65. There are some estimates that are north of 0.7 even using other data sets. Underneath that is of course a huge number of zero earners in the labor market. So an unemployment rate of about 25% using the Iowa definition is fairly consistent and constant across the years for South Africa. Genie decomposition work will suggest that labor market income. So that's really important for the paper because we look at wages knowing that the genie decomposition suggests that wage income is a crucial part, as we've seen in Latin America as well, going in the other direction, crucial part for the explanation in terms of overall household income inequality. We observe just based on the wage genie also consistent with the decomposition and increase in wage inequality. So that's all sort of not surprising, but then that second-last bullet is the one that just threw us, right? Which was when you looked at percentile-based wage growth rates strictly from 97 to 2015, it was U-shaped. And so we need to observe, and try and understand why we observe the sort of wage polarization if you like for South Africa. We settle on three explanations, and I'll go through each of them and we test them in the data. That's really the sort of the rhythm of the paper. The first is the, if you like, what was first the Tinbergen model, skills, bias, technical change, and it was then tested, as you all know, in the cats and Murphy work for the US, which is really a factor augmenting or a technology augmenting labor market story. So what you have is upgrading of technology and this race between technology and upgrading of skills leads to a high premium for highly skilled workers. In both the Tinbergen case in that period, but also for the cats and Murphy paper at the time, it explains, in a percentile world, the top end of the distribution. So it explains how you can see rising returns for tertiary skilled workers. So in the South African context, that makes sense because we have effectively a dual schooling system with a high-end, high-quality schooling system that produces, if you like, high-end, high-quality tertiary graduates on the one hand and then a low-quality schooling system on the other hand. So you then, you can see the scarce skills in a technology augmenting environment existing for South Africa. But that, this explanation doesn't give us enough for the middle of the distribution. So why this collapse in earnings in the middle of the distribution? So we then go to our second explanation. I'll come back to this time and time again with the actual data, which is, of course, as you know, the new literature from Orta and others goose and manning around task content, right? So the idea, and this slide doesn't do it very well, I really should have a table, which is to think of tasks in this new world as either routine or non-routine. Routine tasks are the ones that technology can replace, right? And we have examples there. Non-routine tasks are harder for technology to replace. Routine tasks that, and both of them exist either as manual or non-manual or manual and analytic, right? So just to give simple examples, a routine manual task is assembly line work, easily replaceable by robots, right? A routine task that's non-manual would be a computer programmer, also replaceable. But on the other hand, and this is the middle of the distribution story, you can get non-routine tasks such as a chef or a cook or domestic work, which given current technology possibilities are not easily replaceable. So a non-routine manual work, that's not easily replaceable. And then, of course, you've got the non-routine and litical work which is very difficult to replace. So most of the work for the US and Europe actually then goes to this middle of the distribution and says, okay, they're all these programming jobs, right? They're all these routine type tasks, assembly line work that sits in the middle of the wage distribution and those are the work, those are the types of occupations that are being replaced. And so you see this collapse in earnings. So that's, so the second driver is a really important one for us because we think it can explain this collapse in earnings in South Africa. The third is an unusual one, at least given the work in the North is that we think that the bottom end of the distribution, this U shape remember suggests that the 10th, the 20th percentile of the wage distribution, something's going on there. And we turn then to institutional factors. There was aggressive minimum wage setting across different sectors for South Africa since 2000. You also see, despite weakening trade union membership, so lowering of trade union membership, you do see the rise of the public sector union as a dynamic in the labor market. And other work that we've done shows this rise in the conditional wage premium for public sector workers. So there's something going on with respect to public sector unions that we think may explain the bottom end of the distribution as well. Just for later on, we don't have a minimum wage control. Our data runs from 97 onwards. And so we don't really, we aren't able to control for minimum wages. But we deal with that through the results we see. So there's our three explanations and we then effectively use post-apartate labor market series. It's a labor force survey data runs from 95 to 2015. And then what we do, although we're updating the paper to use the ONET task codes, if any of you are working in this area, you know that the ONET data is what you should be doing. At the time, this wasn't available to us and we use the old order measures, which is to differentiate. So if you go into the labor force survey, what you're starting to do is move away from the standard interpretations of occupations and going into tasks. And these are five standard tasks that precede the ONET work, right? From Orta, Jensen, Furpo has done it as well, and others. And you have effectively, as you can see there, ICT tasks. So I'll just quickly run through them because they're quite important for the results that we see. ICT type tasks are those such as typists, computer programmers, automated or routine tasks, machine operators, assemblers, face to face is quite important because they reflect a particular type of workforce, food vendors, teachers, where there's actual interaction, onsite, where you're required to actually be onsite, such as construction workers, site supervisors, and so on. And then the analytical jobs, which would be almost all of you in this room. So in many ways, what we do is we try and move away from occupation to tasks, and so each occupation is then coded into separate tasks. Okay. Keep that in mind because it comes back in terms of our regression results. Here's the statistical observation. So this is the average annual growth rate of real wages for that period, 97 to 2015. And there's a very, very strong and clear hollowing out in the middle of the distribution. You see the top end of the distribution seeing this significant increase in the average annual growth rates and the bottom end, right? You see certainly increases in wage rates. So just to go back, the obvious point is we're trying to understand what underlays the rising in inequality, right? So the genie goes up, the wage genie goes up, the total genie goes up, and we're worried that we haven't captured the nuances across the distribution. So taking each of our drivers, each of our explanations, and history has shown me, I've presented this once before in Bogota, and it doesn't work because the graphs are barely legible. What we do is we take local polynomials for each of the education levels that we have. So we've got no education, primary, incomplete secondary, complete secondary, and tertiary. And essentially the density functions here, the left panel is 97 and the right panel is 2015. I'll just read you through the results, right? So at the bottom end of the distribution, you've got the share of primary workers that are dominant in 1997. And that's the red line. So if you look at the red line in 97, there's a large number of workers at the bottom end of the distribution. That's to the left of the red vertical line, the red dotted line. If you look at the red dotted line, go to 2015 and you see this collapse. That's essentially the education system doing its work where primary schooling graduates are moving on to secondary schooling. So the dominant form by 2015 in the bottom end of the distribution are workers who have incomplete secondary education. The middle of the distribution, you see this rise, this bulge in workers with incomplete and complete secondary education. So the middle of the distribution is categorized really and we'll come back to that by incomplete high school leavers, early high school leavers and high school completers. The top end is the green line and that's the really powerful one in a lot of these graphs. The top end of the distribution to the 90th percentile or 75th percentile and onwards is the share of workers, tertiary workers in that high end segment and the green line, you see this massive spike in tertiary workers. We think that a lot of what's happening in this education story, which is consistent with the Timbergen type example, is this changing structure of the South African economy. In essence, what you've seen is the rise of the services economy. You've seen a decline in manufacturing or at least a flatlining or manufacturing output and in effect the driver of growth, the little growth that we've had has been financial and business services and high end services and you see that to a large extent in this diagram. Part of the structural change has also been increased automation in manufacturing and agriculture, which haven't generated the kind of jobs that you'd expect. So there's a structural change story that helps us in terms of the first driver. Let's take the second explanation, which is really targeting the middle of the distribution and we look at task content and you see at the bottom there, I've got all those ICT face to face and so on and again, left panel and right panel is 97 and 2015. The bottom of the distribution, what you see is automated and onsite jobs are prevalent in both 97 and 2015. The middle of the distribution, you see increased prominence of onsite, automated and face to face jobs and it reflects this rise of the services economy. I'm just keeping an eye on the time. But the top end of the distribution, again that green line is you see the spike of analytical type jobs and that's of course this growth of the financial and business services sector as well as the face to face type jobs. The third is what's going on with labor market institutions. Remember I don't have minimum wages so I take workers that are in the public sector and I take unionized workers and it's a really interesting comparison, much easier to read than the previous ones. We compare panel A and panel B and particularly if you look at the bottom percentiles, the 10th percentile and below, you see an increase in private and public union representation up to the 10th percentile and we think that's a lot to do with the crowding in of aggressive minimum wage policies. The middle of the distribution is stark because you see the hollowing out of workers in the middle of the distribution and we come back to this because we think if you look at the top end that a lot of these workers are part of a new labor elite for South Africa. Okay, so very quickly. Okay, right. We then, given all the descriptives, we then need some way to assess this in a conditional environment. Now your problem with the quantile regression is that the covariates sit at the mean. Furpo and others have given us the unconditional quantile using a resented influence function, right? Which is really important because together with the resented influence function, right? We are able to run our quantiles assuming that the covariates are unconditional. That's a key device for us when we need to explain using Oaxaca and Blinda, which you can't do in the standard quantile regression approach, Oaxaca and Blinda can work in the resented influence function world because we are trying to figure out what's the impact of both in terms of the Oaxaca Blinda, both endowments and coefficients, right? And the interaction between the two. So in simple terms, our question is, is it endowments? Is it the fact that people have become more educated that's explained inequality across the distribution, right, across the quantiles? Or is it the returns? If it's the returns, then we've got a story that's to do with technology. We've got a story that's to do with trade unions and so on. So very quickly our controls, the five education dummies, the task content variables are all in there, our institutional variables, and we have standard controls. So all of this goes into our model, which is the roof model as it does its work, and in the Oaxaca Blinda world, the key result for us is to see if we can replicate in the conditional world that U-shaped graph, right? And if we do get that U-shaped graph, the question is, is it endowments or is it returns? And this is what we get, which is actually a very stark result because the total effect is the solid U graph. So these are the conditional estimates across the quantiles, right? And you get a U-shaped outcome. So that's the first, if you like, affirmation of the descriptive statistics, right? Is that in the conditional estimates that we get, we get a U-shape. But the second U-shape graph, the dotted one, right, suggests that it's the coefficient effect. It's not endowments. People are not getting more educated, right? It's the coefficients. It's the returns, right, that in fact start explaining the U-shape of the distribution. We've got some simple statistics on the conditional effect. At the median, real wages in 2015 were 76% of what they were in 1997 for South Africa. So you see this collapse in earnings at the median. On the other hand, at the 90th percentile, real wages were 27% higher in 2015 relative to 1997. So effectively, we're comforted by the fact that we think that it's not about the endowments, but rather the coefficients that account almost entirely for the U-shape nature of this wage change. And that's quite important because we've driven the paper in that direction. We think that there's stuff going on in terms of skills, in terms of technology, in terms of union membership, and so on. So what I'm now going to do is show you more of these very illegible panels. So if you bear with me, I'm gonna now take out, in our conditional environment, I've got all those variables, and I'm gonna show you what they look like graphically. Five minutes, great. So, remember I had my education effects. These are now my education effects where the left bar is the compositional, so that's the endowments, the right is the returns, right? So, all we see if I just go to the wage effects for panel B, right? You see the spike in earnings for workers with primary schooling, right? And those reflect at the bottom end for us the impact of minimum wage policies. High school graduates, which is quite important, high school graduates, right? That's that bar there. That dotted line at the bottom. Those returns, sorry, those returns, so these returns here are negative, right? So high school graduates, right? The returns, and you can look at the actual results, the returns for high school graduates on average over the period across the quantiles were negative, right? So you see this hollowing out in the middle of the distribution in terms of the education returns. The solid line, you see the spike on the top right, which is these massive increases for those that have gained significantly. So now I'm just gonna concentrate on panel B, so the right side. This is now our second cluster of, the second drivers, which is all these tasks, right? And so I've given the first three tasks and you see there again the work of the minimum wage, right? So the top bar is on-site workers, security workers. There's a massive increase and aggressive minimum wage setting for security workers, for example, right? And so you see that growth at the bottom end, but there's a collapse in earnings in the rest of the distribution. There's a little bit of a U shape happening for the other tasks, but in essence, you see a pattern where top end workers by tasks have seen massive increases and those in the middle have seen a decline. If you look at analytical and automated jobs, you get the same result. Let me just, okay, almost there. We left the last two, right? Which was the automated tasks and analytic tasks and separated them out with public and, public employment and private union membership. So again, look at the right panel and what you see over there is this U shape in the distribution, right? So in essence, what you have is for workers in terms of automated tasks, if I just go back up, right? The returns to analytic and task content follow a distinct U shape across the distribution, right? There's a premium on analytical work due to the difficulty and there's a substitution with technology. The pattern in the automated task content is exactly the same. If you look at the public sector union membership, right? What you've got is returns to union membership that are negative, right? Up to a point and then they increase beyond the 40th percentile. And for us, it appears that what you've got is a trade union that's becoming increasingly representative of those workers at the top end of the distribution. And there's an insider market for public sector unions that show these massive returns. Okay, so what I very eloquently explained is much better explained in this table, right? So what's happening at the bottom of the distribution is zero primary schooling have seen steep increase return. So there's an education story and that's basically because of minimum wages. Those are these low end workers that are seeing increasing returns at the bottom end of the distribution. There's a positive return to automated jobs, but those are again minimum wage protected jobs, right? There's a little bit of a story about unionized and public sector workers that have increased, but they haven't seen an increase in wage growth rates. So you see nothing in terms of union effects for the bottom of the distribution. The middle of the distribution is where you see, I'll show you that high school graduates, right? You see a collapse in returns for high school graduates in the middle of the distribution. And you see that U shape for automated jobs that are seeing distinctly negative returns in the middle of the distribution. We think that alludes to the decline in manufacturing and mining where automated jobs are particularly prone to substitution. Public sector employment again in the middle of the distribution you see a hollowing out, right? And there's de-unionization of middle workers, right? So middle end workers used to be the rump of sort of your standard manufacturing, factory floor type worker. Because of the decline in manufacturing, you're seeing this hollowing out of workers in the middle of the distribution. Public sector unions are increasingly representing workers at the top end of the distribution. And so you see that collapse also in the middle of the distribution. What's happening at the top? So there's a little bit for everything, right? A little bit of the drivers in each of the parts of the distribution. The top end is almost the most powerful one is that you see these increases in both level and percentage change terms to tertiary education. So the question about South Africa is that you're seeing these massive increases in tertiary education across the percentiles of the distribution and overtime. And that's what you're seeing at the top end. It's also a reflection of this rise in importance of analytical jobs, right? And the service orientated economy. The final fact at the top end is we think deserves more explanation is on the one hand you've got minimum wage protection at the bottom end of the distribution. So that's the lip at the bottom. But on the top end, you've got a public sector union movement that's become incredibly strong, part of an alliance with the ruling government and have seen huge returns for their workers. And so the public sector union wage premium has grown dramatically. And that's what you're seeing in that top end as well of the distribution. So it's a little bit of a messy paper in that we've got all these balls in the air. And as you can guess, it's trying to still be refined. But in essence, what we're trying to show is that the U-shaped distribution that you see is a function of minimum wages, mainly at the bottom. In the middle, you've got this task content stuff that's doing its business. And at the top end, you've got public sector unions as well as the skills bias technical change that's driving it. We think that this is an interesting result to be good to know if it's replicated in other emerging markets because we haven't seen any other similar work, although it may be sitting in the great literature, but at least it begins a more detailed discussion about changing inequality dynamics for formal sector workers. Okay, thanks.