 Thank you very much. So I am a post-doctoral researcher at the University of Croningen. This paper is part of a larger project on structural change and productivity growth funded by DIFFIT. And basically, this would be the outline of the presentation. I would run very quickly through the motivation to do this paper, then some theoretical predictions. Although I'm not going to test these predictions. These are just two frames, some of the results. I'm going to describe the data and methods I use, then show you some results, and then finally get to the discussion and conclusions. So the motivation I had for this paper was that I've seen a lot of increasing interest in special economic zones or export processing zones, industrial clusters in sub-Saharan African countries. Especially promoted by the World Bank through their country reports. However, when I went back to the literature in search for papers analyzing the actual distribution of employment over space in Africa, I couldn't find much. That is basically so maybe then spread around country reports, but nothing really systematic that tells us how employment is actually distributed within countries in sub-Saharan Africa. So the aim of this paper is very simple, but maybe can contribute to filling this gap is to analyze the special distribution of wage and non-wage employment in a set of sub-Saharan African countries before and after the reforms. Why non-wage and wage employment? This is because non-wage employment has become increasingly more important, and I have reasons, theoretical reasons to believe that the patterns of distribution of wage and non-wage employment are different. So I want to also contribute by analyzing these two types of employment separately. And then I'm going to try to contrast some of these stylized facts with the predictions that I have from particularly in new economic geography models to then inform policy about the feasibility of this industrial cluster idea for sub-Saharan African countries. So now for the theoretical predictions, I am going to be talking mostly about new economic geography models because this is the framework that is used in papers analyzing the possibility of industrial clusters, particularly in 2009 war development report reshaping economic geographies is heavily based on new economic geography predictions. So as you may know in a nutshell, new economic geography predicts that large firms producing on their increasing returns to scale will agglomerate to locate near consumers and from these larger agglomerations, they will source smaller markets at some transportation cost T. That's a very basic outline of the theory, but it's enough for the purpose of the argument that I'm making here. And if you look at sub-Saharan African countries, you have very high transportation costs within countries and a very small scale of production as evidenced by a very large share of non-wage employment or family business or small scale units, particularly informal units. So when you take these two things together, you say, okay, what is the relevance then of new economic geography theories for sub-Saharan Africa if these are our initial conditions. So what I try to do is to trace back some models that may apply better for the case of sub-Saharan African countries. So first, just to be clear for new economic geography for the conditions you have in countries that I just described with high transportation costs and small scale of production, the prediction is just that firms with disperse over space. So you won't have any agglomeration steaming from increasing returns to scale. In particular for resources countries, then the prediction is that all the distribution of employment over space is predicted by the uneven distribution of endowments. So be it labor if it's immobile or natural resources. So that is the source of the disparities in the distribution and employment and not agglomeration economies. So for particularly for the case of resource rich African countries, Golding and Company have developed a model where they explain why Africa has urbanized without structural change. And this is related to the previous point. So they say that these resource rich countries have agglomerations that are based on consume consumption of non-tradable. So you basically have the rents from resource activities and people spend those rents in urban areas. This generates a type of city that they call consumer city that is very different from the city that will be generated by the normal channels of new economic geography. Those are productive cities. So you can think of Chinese cities along the coast as productive cities that are brought about by agglomeration economies from the production side and you can think contrast those with African cities where you have large agglomeration, but basically based on the consumption of non-tradable. So the predictions on the benefits of agglomeration are very different from for these two types. So in particular, you would have less benefits for benefits of agglomeration from consumer cities. With a colleague of mine, I also developed a new economic geography model where we try to integrate an informal sector. So the idea is to try to understand what happens when you have a context like the one in sub-Saharan African countries where 80% or more of employment is on small-scale activities. And basically in our model, we say that people can substitute formal and formal manufacturing varieties. And if this substitution is high, the government might improve connectivity, build roads, but that even then they won't be a development of a formal manufacturing sector. Why? Because people still prefer to consume informal goods locally. So that's just to point out how the predictions of energy might change when you introduce some of the particulars for countries like sub-Saharan African countries. And finally, Berens and Polovala developed this model also based in a new economic geography model. When you have a set of skilled workers that can either be productive working in a normal factory or they can become part of an unproductive urban elite that benefits from rent seeking. So then you have urban agglomerations as well with wage workers, but these wage workers are more unproductive than workers that work at factories. In that case also you would have lower benefits from agglomeration. So we have basically also the prediction of agglomerations, but we don't certain benefits steaming from them. Not so clear, as is the case of the standard case of energy. So now with that on the back of your head, I'm going to try to describe the data and methods I use for the paper that are very simple, but this is more a comparative exercise, so I'm not trying to add anything to the existing methodology just to use the available data and some standard methods to inform some of these questions. So I use the population census samples that are provided by PUMS for these countries, so Tanzania, 1988, 2002, Guinea, 1993, 1996, Senegal, 88, 2002, Malawi, 87 and 98, and Mali, 87 and 98. Why this sample? Because these of all the countries that are listed in IPUMS are the ones that have at least two year data points for the census and have some of the variables that I need for the analysis. So I basically chose all of the ones that were available. I had to do some adjustments to this data to make it comparable. So I use the unit of analysis is basically the second level of disaggregation, be it the province department of region. I'm not claiming that this absolutely these areas are comparable because you can think of regions in Malawi that are ten times larger than some regions in Tanzania. But they are relatively comparable. So within the country the division is fairly the same across. And for changing boundaries, I basically use the oldest boundary. So I match, whenever there is a changing boundary, I match the the latest data points to the previous one. So I have a uniform set. Then I classified workers using this variable class of workers that is available from IPMUS and it's supposed to be comparable across time and across countries. And I classified them as wage. If this variable takes the value of work at someone else, works for someone else as wage salary worker and non-wage worker if they are self-employed with or without employees or if they are unpaid workers such as family workers or apprentices. The underlying assumption for this is that basically non-wage employment corresponds to small scale activities that fall mostly into the production of non-tradable goods and services. So this would be the opposite case of what new economic geography considers to be increasing returns to scale. Wage employment corresponds then is a mix of employment between public and private institutions, but I assume that these are fairly larger in scale regardless. So that's as much as I can do with the available data. I also break down employment by these four industries. They are initially available for 12 industries, but I break it down into agriculture and mining, secondary sector, market services and non-market services. Unfortunately, this industry variable is not available for these countries in years. So sometimes I cannot make the over time comparison, but I still use all the information I can. These are some general characteristics of the studied countries. So basically I have a mix here of coastal, non-coastal, resource rich, resource poor, but not large enough to draw any conclusions for these larger categories. So just as a reference, by far the largest is obviously Tanzania. So and then you have the others are fairly similar in size and while you can see there some of the other characteristics, so that's just for your reference. So basically to measure the concentration of employment, I use two types of measures, the A spatial measures of concentration and spatial measures of concentration. So I combine these two to give a bigger picture, a more complete picture of what is happening with the distribution of employment in sub-Saharan African countries. So first the coefficient of variation is a very simple measure. It's just the standard deviation over the mean. So for the two types of employment, if this measure is zero, that means that there is a uniform distribution. If it increases over time, it means that employment is becoming more concentrated. Why is this indicator a spatial? Because it doesn't take into account the proximity or of the units. So basically every unit is taken independently and not the fact that they might be approximate. But I still take it because this is comparable over time in across countries. So that is one of the aims I have. And then I measure the degree of concentration using fairly two standard measures, two fairly standard measures of inequality, the tail index, and they have the square coefficient of variation derived from the general entropy measures. And here I won't go through the results, but a larger value indicates more concentration. For the spatial part of the measurement, I have a very standard index that is the global moran index. This index expresses the overall degree of similarity between spatially close regions. So what is called spatial autocorrelation? With respect to an numerical variable that in my case it can be wage or non-wage employment. So the spatial interaction is measured through an inverse instant matrix. And then I measure the distance in a very standard way, which is the bilateral distance measured as the crowd flies. And this global moran can take basically a positive value, in which case it is indicating that approximate regions exhibit similar values. So that's that would be indicating clustering. If it's negative, it indicates that approximate regions exhibit dissimilar values of employment. And if it is statistically significant, it means that the distribution of employment does not follow any particular pattern or space. So the distribution can be taken as to be as random. And an extension of this global moran is the local indicator of spatial association, which is basically going to tell us where this clustering happens. So it identifies where clusters happen, the center of the cluster. A cluster would be then a region that has high employment and is surrounded by regions that also have high employment. You can also have also something called cold spot, which is a region with low employment surrounded by regions of low employment. So now for the interesting part of the presentation, the results. First I'm going to go through the coefficient of variation. Up there you can see the coefficient of variation on the same scale for the five countries here. So you can compare the absolute values across. And on this axis I have the share of wage employment in total employment. So the first thing you notice is that for most of these countries the share of wage employment decreased sharply. But this is a slice fact that is already well known in the literature. Except for Senegal, when you see it staying at a level of around 15 percent. So very sharp decreasing countries like Guinea. Mali never really had much of wage employment anyway. But you see that even though this share is decreasing, employment is becoming more concentrated. So you see it here, the change over time. Wage employment becoming less representative and more concentrated. This is the finding that I have. Then non-wage employment, as you can compare here, the red with the blue. Non-wage employment is far less concentrated than wage employment. So that's one of the main results here, that the two types of employment have very different patterns of concentration. And you see it clearly there. For the case of Guinea it's very extreme, the difference in concentration levels of these two types of employment. And you see that there is not really much of a change over time. So we're talking about 10 years difference at least. But it seems like nothing really happened with the concentration of wage employment as this is expected. Because non-wage employment, if you remember, is related to small scale activities that disperse in space. So there is no reason why this type of employment would become concentrated. So this makes a lot of sense in that sense. So you have different cases like in Tanzania, it actually became more dispersed over time. You have some other countries where it remained fairly stable. And in Senegal and Guinea in Greece, but it was initially at very low levels. So for the industries, I'm not going to show the table with the results, but I'm going to just run through the main results. So these indexes of LG E1 and G2 are very close to the lower bound. Meaning that absolutely, industries in Africa are very low levels of concentration to start with. The highest level of concentration comparatively between the four sectors that I have is on market services, as this is expected. So for instance, retail services, retail trade, sorry. And they are lowest in agricultural and mining. This is also to be expected because of the nature of agricultural activities. And this remained true before and after structural reform. So another finding is that basically not much happened between before and after reforms with the special distribution of employment in the countries that I analyzed. No wage employment is absolutely less concentrated than wage employment in these three sectors, so in all sectors except non-market. And for the two countries where I can analyze patterns over time, I find that basically there is not much of a change because of two reasons. Why? Because there is no movement of employment towards sectors that display larger concentrations. So not much change, for instance, from agriculture to market services in that time period. And there is an increasing share of dispersed non-wage employment. So those are the sources of the lack of change. Now for the special part, here the ones that are shaded are not significant at the 95% level, and here then I wouldn't take them into account because already I have a very small number of observations. So it's fair to say that these values are not to be interpreted. But for the ones that I can interpret, then I see that there is no special autocorrelation in wage employment for all countries except for Senegal, which is also, you can see here, this is wage. Levels of special autocorrelation are positive, meaning that there is some global pattern of autocorrelation, but it's not very strong. And it's only significant for the country that if you remember, I had some significant share of wage employment in both periods. So it's not surprising that I don't find anything for wage for the other countries because it represents such a small share of employment anyway. And then you observe here that the patterns for non-wage employment are also relatively small. Now, where these closers are located? You have here for wage employment, I do not find any significant local indicator of special association. For the case of non-wage, I find some here in the area around Lake Victoria. But if you can see here, comparing over time, there is a high inertia in this. So the distribution remains fairly the same across. And then for the only case where I find a clear wage employment cluster is for Senegal, around the Dakar area. So this is the only country of the five that I analyzed, where you can see some evidence of clustering for wage employment. And the interesting part is that there is also clustering of non-wage employment but happening not in the same areas as the other one. So for the conclusions very quickly, so I'll just go through all of them. So basically there is not much happening in wage employment, but it is the type of employment that as you saw has different special patterns and is the one that is relevant for industrial clusters. So all these ideas of creating clusters since at odd with this evidence. It seems that most of the distribution in space in sub-Saharan African countries is explained by the distribution of resources. Like it was very clear in the case of Tanzania. And new urban settlements are dispersed in the interior area. So the migration is within the same region. So there is not much of movement to the coast or anything like that. Like we are observing in China. And then basically the relationship between structural change in space and structural change in industries, you can see clearly the connection because there has not been change. By industries you don't also observe any change in space. So as long as you don't change the distribution in wage and non-wage employment and the distribution across industries, this clustering in coastal areas is not going to happen. So my prediction would be that at least for the case of Tanzania, this idea of export processing zones and clusters in the area of Dar es Salaam does not make much sense. Thank you and sorry.