 Okay, well, thank you for having me here and thank you for being here as well. My name is David Garces Urza-Inki. I'm a PhD candidate at Free University Amsterdam and I'll be presenting a joint work with colleagues from the University of Jerusalem on recent poverty dynamics and vulnerability patterns in Tanzania. So I'll maybe give some introductory remarks about how this work came together. So if we look at the public policy discussions and also research when we think about the evolution of poverty, what generally is discussed, we speak about the evolution of mainly poverty rates over a given period of time. What we have is a collection of snapshots of the income distribution and then we see how this rate or maybe the gap is evolving, but we tend to pay much less attention at the dynamics of particular households. What are the poverty trajectories in and out of poverty that different households follow? There are different reasons why this focus has been like this. One main issue is that data constraints as standard poverty dynamics analysis requires panel data and that's while the availability is growing quite a lot in developing countries, it's much less available than repeated cross-section service. And maybe also part of the reason for this focus on how poverty, like the static, the evolution of these snapshots of poverty is that at least until the pandemic, the policy discussion on the reduction of poverty was often dominated by a sort of optimism where poverty was going down, maybe not fast enough, maybe not everywhere, but in general it was about bringing this, like there was hope not to be this rate down, so it was mainly thinking about issues like the growth elasticity of poverty and how can we make this reduction go faster? Now the setback to poverty reduction efforts that has been delivered by the pandemic emphasizes that progress might be reversible and that it is important to think about a vulnerability to poverty households that in this particular moment where we look at the data are above the poverty line, but are prone to falling below it when we look at it again in a few years and in general about these dynamic issues. And while the pandemic has helped to bring these issues to the forefront, it's something not exclusive of it. Like you don't need a global shock for poverty dynamics and vulnerability to be important. Households face many shocks, in particular poor households and we need to be thinking about these issues generally. It's a structural feature of developing economies. So this work is part of an attempt, I mean, and this is important for public policy, but also for research to give insights on how are different households classified into different poverty dynamics and how that might help policy makers to think about how to address these different forms of poverty. So this study is an attempt to, is part of a collection of several countries studies over initially five countries to provide a systematic assessment of poverty dynamics in several countries that are, let's say, make an important contribution to global poverty accounts. And this is work that was funded under the data and evidence to end extreme poverty program coordinated by University of Copenhagen and Oxford Policy Management. And these five countries studies appeared in a recent special issue in the review of development studies. So something that all these studies share and that will be also apparent in the work I'm presenting today is that to overcome these data constraints they apply synthetic panel techniques on repeated cross-section service. And they also, due to the timing when they were conducted, we look always at pre-pandemic data but also try to think how these results can be interpreted in the, well in the face of a world that has suffered a pandemic that had some impacts on the income distribution of many different countries. So why is looking at Tanzania interesting? Well, Tanzania is among the countries accounting for, like if you rank counters by how many of the global extreme poor, according to the international line of the World Bank, the account for Tanzania is one of the countries that contributes most. It's depending on the years or how you of course you computed the price and so on, four or five percent is something reported. It's also known that it's a context of high vulnerability and a lot and transcend poverty and that will also appear in my work and when speculating about the interaction with COVID-19 and also for future work that will have actual data on that, it's a country that followed an unconventional response. As we will see there. These are aspects that make looking at Tanzania interesting apart from of course people that are interested per se in Tanzania. So what I will be doing, I will be looking at poverty dynamics and vulnerability in so-called normal times with, so speaking of 2011-12 to 2017-18 and applying synthetic panel methods to the most recent data from the large household survey that underlies the official poverty statistics and then look a bit like what are the levels of persistent poverty, what are the levels of transcend poverty, vulnerability profiles and then try to think how these relate to what we know by now of or what we knew by then on what has, which people have suffered more from the economic impact of COVID-19. So to give a little bit of background, Tanzania has experienced during the decade of the 2010s and until the pandemic sustained economic growth. We're speaking about 6% real GDP per capita growth in that period, but that did not translate into substantial poverty reduction. Now if we look at the last period between 11-12 and 17-18, we only see like a reduction by two points in the basic needs poverty rate with the national poverty line, which by the way is way below the international poverty line. So we have a situation where growth is not sufficiently translating into poverty reduction and we have well, relatively high levels of poverty. So on the data methodology again, and we'll be looking at the most recent household budget service and they fetch around 10,000 households and are the basis for national poverty statistics and at the time of writing, they were the most recent data on welfare or on consumption in the country. By now there might be, I think there's a new wave of the national panel survey, but even then there are some reasons why it's interesting to do this exercise with synthetic panic techniques on the HBS. First that the sample is much larger than what you can find in panels, which allows us to disaggregate better by different regions. I mean this last HBS is representative by region, so we'll be able to look at regions. And also, well, there are some discrepancies between both services and since this is the basis for national poverty statistics in terms to look at it. So to give a quick introduction how these synthetic panel methods work. So the standard situation, the standard board dynamics analysis would have panel data and then we observe a set of households at a given period of time and we observe, at least those that can be tracked, the same set of households in period two and there might be issues with waiting and household splitting and so on, but then it's a simple counting exercise. Know how many of the households that were initially in poverty remain in poverty in period two. Of course the complication with repeated cross sections is that we observe different samples at different periods. Here we have, at first period, we observe these people, the blue circle and in period two we observe these different households in the red circle. So the question is, well, I can know how these people are doing now and how these people are doing now, but I don't know how, for instance, these people would have been doing in period two. So what synthetic panel try to do is to combine the information system data set to make the best possible gates or to get the best possible estimate. So the core of the method is building regression models based on time invariant regressors. Regressors that do not change over time. This could be age of the household head, ethnicity, education of the household head, region of birth, and then basically we get two different models, one for each period of time, where we get the association of these time invariant characteristics with income. What we can do then is, okay, let's say, in this case, let's imagine we are predicting forward for households in period one. Let's try to guess what this household would have got in period two. And what we do is we take the characteristics from these households because they don't change over time. So six years later, they would have been the same characteristics and we use the returns because it's not causal of these characteristics that we know from the other household survey, from the household survey that was running period two. And this will give us a part of the mobility, the one that is predicted by the model. And then if we can try to make estimates of how large are these quantities of people that remain in poverty, people that transition in or out of poverty. The important assumption here, of course, is that the estimations we get, the estimates we get, the bitters we get for this population are valid for this one. So we require a certain degree of stability of the population and also the survey being run in a comparable fashion. And then we, of course, we need to think about these error parameters. Because a large part of income will remain unexplained and depending on how we model this, we will obtain one result or other. And well, I'll say a bit more about this, but in posting certain structure we can then estimate for each household make a guess of what would have been the probability, for instance, of having escaped poverty, of having fallen back into poverty, like being above the poverty line in period one and below in period two. And also, this for all the different states, also for being below poverty in this period and then aggregate up by subgroup or at the national level. And this will give us, like if our assumptions are correct, it will give us a decent estimate of what are these quantities, of households that stay in poverty, escape poverty, and so on. So for the details on how we model, how we deal around more in these residuals, we use the technique introduced by Dan and Nanjo in 2013, which has two elements. One is on the slide, which is we assume that the residuals are a log normally half a, by variate log normal distribution. This could be refined, but it's not very problematic. The question is, the key question here is to back out of what is the correlation of this residual, not the inter temporal correlation of this residual. How much of this unexplained component of income you will be carrying forward to the next period because that's going to be very important to know how persistent poverty is. And well, Dan and Nanjo's are just a way that this can be derived from the inter temporal correlation of income, basically how income is correlated over time. And what we do in this paper is since there is, we take advantage of this panel data from a neighboring period and use the inter temporal income correlation from there to from there derive the inter temporal correlation of these residuals. And with these elements at hand, we can, I mean, we have some, we have a distribution of assumptions for this probability of the, and we can compute the transitions. Our income model is going to include age, education and region of birth of the household head. And yes, so that's for the poverty transitions. For analyzing vulnerability, we also need a vulnerability line. And what we do in this study, instead of, let's say what's often done is choosing more or less arbitrarily a level directly. Let's say twice the poverty line or 1.5 to the poverty line. What we do here instead is we like, we set up a so-called vulnerability index, which is the conditional probability. So how, when I call a household vulnerable, what does it mean? What is the probability? No, what the probability that these vulnerable households fall into poverty. So we choose that level. And this case is gonna be 30%. So the vulnerable households are gonna be those that when we look at them, their conditional probability of falling into poverty is going to be 30%. And so this can be seen since we have data for two periods, we can see empirically, okay, at what level of the vulnerability line, we have that 30% of those households fall into poverty in the next period. You can imagine, and in this way, this index implicitly defines a vulnerability line. If you set the vulnerability line very high, then you're gonna have a relatively low probability of falling into poverty because it will include very rich households. As you make the vulnerability line lower, this probability of falling into poverty is gonna be higher. So at some point, you will hit the point where it corresponds to the index you are choosing. And in our case, it results in a vulnerability line that is around, well, it's 46.8% higher than the poverty line. Okay, coming to results. This was a bit too, to be a bit background of the technique. Coming to results, these are like the core results of the paper, which are the poverty dynamics at the national level. So if we try to understand this, this is the state in 2012, poor or non-poor, and the transition to the state in 2018, poor or non-poor. If we look as a share of the total population, we see that about 12.5% of the population remained in persistent poverty. Around 60% was out of poverty in both periods. And we have that transient poverty in total was around 30%. If we look at conditional probability, this is a lot of, this is very high transient poverty and a lot of mobility. If we look at conditional probabilities, we see that, well, the probability for poor people to escape poverty is quite high, but also for initially non-poor people to fall into poverty is around 20%. This is over the whole average, over the whole set of initially non-poor people. Okay, in the interest of time, let's look a bit of poverty and vulnerability profiles. What are the factors that are important for association with persistent poverty or with being vulnerable or being secure, which is the households that are above the vulnerability line? We see that the area of residence is very important. That, I mean, there is a lamb. Okay, this is presented in terms of ratios, ratios to the national average. So for instance, in Darn Salam, we would have that, like households are 60% more likely to be secured than in the average in the country. We also see that education is very important. Households above secondary education have almost non-existent probabilities of being persistently poor. Also much higher probabilities of being secure. Demographics, well, just shortly, like all their households and larger households have a bit more adverse situation, although that's not so marked as with education. If we look at the regional patterns, we can see that persistent poverty is very concentrated in the Northwest. And it's lowest, I mean, here, the lighter zones reveal a lower persistent poverty. So in Kilimanjaro and the area around Darn Salam, we're speaking about below 55%, while in Northwest we're speaking of levels close to 20, 25%, while in poverty, if we look at conditional probabilities to exit poverty, we see that they are highest, well, they are lowest in the region where persistent poverty is higher, but where they are higher, it's not like, let's say, the less the areas with the least persistent poverty, but in these areas in the Southwest. And this is something that we could not know without these dynamic analysis. So, then maybe to quickly think about the relation of these results in normal times to COVID, to COVID, Tanzania's COVID-19 response has been a bit unconventional, especially because after an early phase where some measures were taken for a relatively long period, the reaction of the country, let's say, was a bit different to the international standards and didn't take many measures, but from March 21, there has been a, with the change of president, there has been a gradual convergence to more standard response. So, this lack of preventive measures for macroeconomic performance, it seems that, I mean, it has led to a comparatively good, so there has been positive real GDP growth over 2020 and 2021, although lower than what was the regular rate and the economy is mainly recovering. So, if we look at what some reports are saying about the mainly affected sectors of groups, we're speaking about the tourist sector and there have been lots of formal jobs and also income losses for informal workers in urban areas. So, now if we look at how, and let's focus maybe here on the right figure, how these households that have been economically impacted by COVID, compared in terms of poverty dynamics to the national average, it turns out that this new vulnerable would be suffering from lower levels of persistent poverty and enjoying high levels of security, that's the yellow bar, then the average household. So, in a sense, we can't be speaking about new vulnerable. This means that we might need some specific new policies for addressing them, but also that we cannot forget the structurally vulnerable to poverty. And that's a bit the conclusion, one of the conclusions from this study, that I mean, while COVID-19 has been useful of bringing this, well, has been terrible, that has been useful in bringing this vulnerability considerations to the forefront, we need to think broader, when we think about vulnerability and this is going to affect also different households. Also, when we face different shocks like now food price inflation, yes, it's important to respond to current conditions, but there is also a structural vulnerability in the economy, which in the case of Tanzania is relatively high because at least 30% of people were in transit poverty between 2012 and 2018. And this again, when thinking more broadly about poverty reduction, if we look at the poverty assessment at a given point of time, it's important to remember that, I mean, yes, this line is meaningful and it's important to focus on the people that are below, but also the people that are right above or a bit above are going to be in risk of poverty, so policies should also be thinking about them. And again, that goes a bit back to the beginning, as long as we have that growth translated efficiently into poverty reduction, we're going to be suffering from these structural vulnerability issues. Okay.