 Good afternoon everyone, and I would like to thank the organizers for giving me the opportunity to present my research. It's very much research in progress, so I would be very happy to receive a lot of comments and critical suggestions from you, especially from any South Africans in the room, and there might be some here. Well, actually I think my story kind of complements that of Haroon, because unlike Haroon, I will focus on a much shorter period, but I will look at the post-crisis, so post-2008 situation of South African labor markets, and also I will look at more, well, take more of a dynamic perspective, so I will look at individual transitions in and out of employment and unemployment. So the outline of my presentation is the following. I will first introduce the motivation of the study in the research question. I will then go into describing the two longitudinal data sets I used to analyze or at least attempt to analyze this research question. I will use the data to construct transition matrices and also decomposable mobility measures, so labor market mobility measures. I will say something about my empirical model and discuss the results, and then I will just close with a number of, well, summarizing my main findings and also looking at some avenues for the research. So as Haroon has already said, South Africa was heavily affected by the global crisis, because it was well integrated into the crisis, and so many studies have documented that due to the impact of external shocks, real sector shocks in private capital flows, trade, and for some countries also remittances, that growth got affected in developing and emerging market economies. So this also counts for South Africa. As Haroon said, South Africa entered recession for the first time since the fall of apartheid in 2008, so the fourth quarter of 2008, and a new than three negative quarters of growth. So and since the recovery has been very anemic and punctuated actually by renewed global economic slowdown. So my figure is well less fancy than that of Haroon, but basically showing them the same thing that there was a recession in growth, and that recovery has been has been quite an anemic ever since. Of course, this adverse macroeconomic trajectory has not been without consequences for South African households or South African individuals. The official figures indicate the net employment loss of about one million individuals over the most intense phase of the crisis, and then a slow recovery after that, and we also see a rise in unemployment rates over 2008-2012. As we all know, labor market stages is a critical determinant of household and individual well-being, and this is definitely no different for South Africa. Also, as Haroon explained, there was already high and very structural unemployment and segmented labor markets before the crisis. So most of the unemployment is structural due to some of the factors highlighted before, like skill-biased technology change. And so this unemployment has been described as South Africa's Achilles heel, so it's a logical question to ask how far can South Africa go in rising unemployment rates before something really drastic happens. Also from the literature, we know that economic recessions tend to have heterogeneous impacts on different kinds of workers. And so in my study, I tried to complement earlier crisis impact studies by using longitudinal data sets instead of repeated cross-sections. So my research question is fairly simple. Which individual household level and job-specific variables are associated with staying employed or not in South Africa during the height of the crisis and in its aftermath? So this is just to illustrate the importance of unemployment or of the unemployment rate in South Africa. We see that during the crisis, the narrow unemployment rose slightly, but there was a much greater increase in broad unemployment. And if you focus on broad unemployment, we see a number of heterogeneous, I mean, there are a number of group differences. So for example, in the second graph, I compare male with females and you see that whilst females have higher broad unemployment, the gap, well, diminished during the crisis. So males were apparently hit harder during the crisis. Of course, this kind of cross-sectional data only provides a very knitted out picture of changes in South African labour markets. And so to evaluate cross-changes, we need longitudinal data sets to identify which are exactly the people that move in and out of employment. And that is what I set out to do with two data sets. First, the NITs, the National Income Dynamic Study, which is South Africa's first nationally representative multi-purpose individual level panel data survey. It has had two waves so far. Well, there's a third wave that has become available just days ago, so I hadn't had the chance to look at it. My analysis is restricted to adults aged 20 to 55 because I don't want my transitions to be influenced by school leavers or pensioners. And there are six mutually exclusive labour market statuses. So there's regular wage employment, self-employment. There's casual and other employment. That other employment also includes subsistence agriculture. And then the searching unemployed, the discouraged unemployed, those people that are willing to work but not actively looking for work, I mean, not actively searching for a job. And then you have the not economically active, those other people outside the labour force. I must say before I show the results, of course, that there are a number of problems with NITs, as indicated by the people that constructed the data set themselves. So there's some misclassification between the different categories of the non-employed in wave two. So I will also let this inform my empirical model. And also the between waves attrition rates are overall quite acceptable, but they're much higher for better off whites. And these better off whites are exactly also the people that were less likely to participate in wave one. So this gives extreme weights and can make the estimates for this particular group less reliable. Then the QLFS. The QLFS is actually the official data source for unemployment rates of South Africa and has been there since the first quarter of 2008. This is not a panel of individuals, but it's actually a rotating panel of dwellings. And so the unit of analysis is the household. And normally the household identifiers are maintained over the different quarters, but not necessarily the individual identifiers. And so I use a matching technique developed by Renshet and Dinkelman to match individuals from one quarter to the next quarter using household ID, age, gender, race, education and marital status. And I achieve an average matching of 68 or almost 69%. I use inverse probability weighting techniques to correct for non-random matching on observables. So I use probits to estimate the probability of being matched to the next wave. And so I reweight my sample to kind of tease this out. I also restrict my analysis to the same age group as I do for NITs. And then here the labor market status are not 100% comparable. So in the QLFS I use formal sector employment in formal sector employment. And then three other categories are quite similar than those in NITs. Also this matched dataset has a number of problems. It's possible that of course there's non-random matching on unobservables for which I cannot control. For example households that migrated from one dwelling to another, they cannot be matched. And so if you assume that households or individuals that migrate are more likely to change their status than we're actually underestimating mobility in this sample. Also there's a possibility of false matches. I try to minimize this by checking consistent or doing consistency checks on educational status and marital status. But of course it's still possible that also false matches which would of course overestimate the labor market mobility. So what does this give? Looking at the data, I have here a transition matrix. You see that in fact there's quite some mobility in South African labor markets. You see that almost 24% of those people in regular wage employment in 2008 were no longer in regular wage employment by 2010-11. But there's also mobility in the other direction. So lots of people that were unemployed searching or discouraged, they actually found a job in 2010-11. So we can use this survey weighted transition matrix also to construct measures of mobility. So if you look at overall mobility, I find that actually just over half of all the individuals in my sample moved from one labor market state to another. And if I decompose that in upward mobility, meaning people transitioning from non-employment to employment and downward mobility, so people moving from employment to non-employment or within mobility, I get the following figures. So you see that here downward mobility was a bit higher than upward mobility. But the largest component was actually within non-employment mobility. But since there has been some misclassification, we need to take this with a grain of salt. So doing the same thing for QLFS, I'm not sure whether you can see the data from here. But the main point is, yeah, of course mobility is much lower in this transition matrix because the QLFS is a quarter to quarter data set. So of course quarterly mobility is much lower than mobility over a span of two years. And you also see that the states are far from stable and especially there's a lot of changing in states among the unemployed. If you look at it in an inter-temporal, I mean comparing years, you see that actually all states have become more absorbing. So this means that labour market mobility has actually decreased during the crisis. So both upward mobility as downward mobility and as already been noted in other studies, actually the rise in unemployment rates during the crisis is more due to reduced inflows into employment than actually outflows out of employment. But nevertheless in this paper I focus on downward mobility. So here I can calculate the decomposable mobility measures. So my empirical model is actually a very simple survey weighted binary probit model. So I estimate two kinds of probits, one for NITS, one for QLFS. And so my outcome variable equals one if an individual in regular wage employment in 2008 was again in regular wage employment by 2010-11 and zero if that individual was initially in regular wage employment but no longer so in 2010. So I leave out actually those people that did not have a job when the crisis hit. So meaning here looking at the NITS, I focus on the individuals in grey, so on the first row only. So of course this technique can be used to also look at the other transitions but just to focus here on the downward mobility. I take only those individuals. So the QLFS very similar but here we look at formal sector employment and transitions from one quarter to the next. And I pull quarter to quarter transitions over the years to see whether there has been any evolution in factors that influence this downward mobility. So my explanatory variables are both individual and household level demographic and locational variables. These are the standard variables also used in cross-sectional studies. And then I add a number of job specific variables, occupation types, industry types, union membership and contract types and durations. And I do all estimations separate for men and women because they have been shown, has been shown that there is a lot of different dynamics going on for men and women. And this also follows the literature. So looking at my estimates, these are the estimates using NITS. I just printed here the average marginal effects. You see that actually those people in the middle age group have a higher chance of keeping their job remaining in a regular wage employment over this two-year period. And you see that there are some buffering effects of education or of secondary or higher education but only significantly so for women. So remember that I focus on those people that were already in regular wage employment in 2008 so those are already the people that have typically have higher education. But so for men this education did not provide an extra buffering effect during the crisis or at least that's what I find using NITS. And you see also that the racial effects are not significant. Again of course these racial effects are highly significant if you just look at cross-sectional data, if you just look at who has a job in the first place. This is really very much determined by racial effects but we see no racial compounding effects during the crisis. This can also be due of course due to the unrepresentative of the better of whites in the sample. So that's an open question. Then adding some extra job specific variables. You see that semi-skilled and managerial or professional jobs provide higher job security for females but not for males. Looking at the industry dummies you see that especially construction and wholesale and retail trade were sectors where there was less job security during the crisis. So this is also found in the cross-sectional studies. The only thing that struck me is that there is no significant coefficient of manufacturing which was actually the sector that suffered most in cross-sectional terms of unemployment. But if we look at repeated cross-sections we also see that there has been quite a fast rebound in manufacturing employment. So it might be that those people have more transferable skills and over a two-year period we do not catch these temporary unemployment effects for manufacturing. Looking at union membership we see that union members are definitely have a higher chance of keeping their job. Also written contracts and permanent contracts lead to higher job security so that's kind of logic. Then looking at the QLFS this enables us to compare these effects over time. Now I find actually that it's basically the younger employees that lose out during the crisis and this is consistent over the number of all the years. And especially sorry it's especially high for female workers. Looking at the education effects we do find now a buffering effect of higher education for men and also for women. But we see that has declined generally declined over time. I do not have really a good explanation for that so if somebody would have a suggestion that would be very helpful. And we also see some racial effects especially for white males so this is in contrast to the niche findings. Then I just added one more estimation with the industry effects and we see again that construction and retail trade are those sectors that provide less job security. And also here transport is okay I will move to my conclusions. So my main findings are that first of all there's considerable mobility in South African labour markets if you look at it from an individual dynamic perspective. And this corresponds well with the findings of other periods. I see that NITS and QLFS both suggest that likelihood of continued employment differs significantly between different groups of workers. So lower for younger workers, workers with less than secondary education and males employed in construction and trade. And chances were higher for trade union members and those with written or permanent contracts. Looking at the evolutions we do find that there is some time variation in the economic significance of some of the variables. But it's very difficult to actually connect them empirically or theoretically to the broader evolution of the South African economy. And so these are some avenues for further research which I'll just leave the slide on so you can read it. Thank you very much.