 Thank you for very important and interesting presentation. My name is Irina Gerasimov. I am from Moscow Russian Academy of Sciences. I have two questions to the first presentation. The first one is about the social and economic factors of vulnerability. Could you say some more about what we mean, economic social factors, to fall in poverty and to change the statement of household? And the second one, if you have the task to estimate the dynamics of vulnerability in the big real, such as South Africa or South America and so on. But if each country has own level of poverty and the bottom lines of middle class, how to use your second approach? Thank you. Thank you very much. Then over there, the gentleman. My name is Niranjan. I work in the UN Regional Commission for Eastern Asia. Thank you for the excellent presentations. But my question is to Peter. First of all, the approach of defining vulnerability line and defining the middle class. This is something of my interest also, which you are doing in the Arab countries. So I liked the approach that you are proposing. But my question is, if you have a lower bound for middle class, which is the vulnerability line, what is the upper bound? Your former colleague specified $13 in PPP. But I didn't see that, so what is your response to that? My second question is specific to India. Since in my previous thesis, I was working on NSS data. You proposed this imputation method to calculate another condoms and expenditure for the households. But I think some years down the lane, Professor Diten had suggested, if we can use the grouped data, particularly because Indian households have very high fixed effects, which you have highlighted also, because of cultural expenses, maybe because of indebtedness, maybe because of seasonal migration. These are the factors which extremely affect Indian households' consumption. Now, if you impute, then I expect that the error variance may be high. But did you try with the grouping of households taking into account some common characteristics as you identified and calculating the mean expenditure for that and creating a panel to apply that method? Thank you. Thank you very much, and then we have a gentleman here. University of New South Wales. Another question for Peter. As you were introducing your talk, I was expecting what you would do would be to try to work out the probability of transition into poverty for people at the vulnerability line rather than the average for those above or those below it. Is there a reason why you didn't follow that strategy? I'm guessing there are practical reasons for implementation. Why didn't you do that? Thank you very much, and then we have a lady there a little bit further. First presenter. Let's mention for the Susanus data sources, you have calculated from 97 to 95 for the poverty. Actually, we have for panel data that from Susanus is 2002-2004 up to 2007 to panel data. My question is, are individuals in poverty data sources logical comparable among yours? That's my question. My assumption, you use the similar data sources from Susanus. If not, maybe my question will be disregarded. And for the second presenter for Davis, you calculate the log financial asset for per capita regression that stated the percentage of urban, the variable percentage of urban population, it was really small, 0.008. In this, in Aquilele, this is a very small coefficient. What is actually the purpose of this small coefficient in the analysis up in Aquilele? Thank you very much. Thank you very much. Let's stop here now for the time being and let's give our presenters a chance to make the first questions. Maybe would you, Jim, start and Peter can sort of put his comments together? Jim? Yeah, the idea about the percentage of urban population in financial assets regression is that it's an indicator of people, urban population has better access to financial institutions. So you would expect that with a more urbanized population, on average, access to financial institutions is easier and cheaper for that population. And so as an example of variables like that, that we tried in these different regressions, it turns out to be a significant variable, so we of course keep it in the regression. You're right, it's a pretty small coefficient, but you'd have to look into that a little bit further to find out, for example, what does a quantitatively, how large the impact is, it may be that it's quite small, but it is a theoretical reason for it being there and it's a significant variable, so we keep it. Thank you, Jim, and then Peter. Very cool, thanks very much for the questions. I'll try and answer, I'll try and move over. Thanks very much. I'll try and respond to at least some of these questions. Thanks very much. On the first, taking the last question first, the validation exercise that we reported on the Indonesian data based on the research that we had done in a separate paper was based on the Indonesian Family Life Survey, the IFLS rather than the Susanus, because we were interested in taking a true panel data set and then testing the method against the true panel data set. And in testing the method what we did is we treated the IFLS as though it was not a panel data set. So we took the panel subsample say from round one, we split that sample into two subsamples separately and then we implement the methodology across the two subsamples in such a way that we could treat them as though they're basically not panel data at all. We apply the methodology, we produce the transition probabilities and then we check against the true panel transition probabilities that are in the IFLS. So both the validation work using for Indonesia as well as the validation work for Vietnam was based using the actual panel data set but treating them as though they were not panel data, implementing the method as though they're not panel and then checking against the panel dimension of the work. So the Susanus data were not used in this exercise at all. I do have a separate study that's sort of ongoing at the moment with a colleague where we're going to try to implement the methodology using the Susanus data because it would then allow us to do some kind of panel type of analysis on a very much larger sample size in Indonesia if we could use the Susanus data for this purpose. But we want to probe and test that and validate that as well as we can before we do that in a large scale and that's kind of still ongoing at the moment. The question about calculating the probability at the vulnerability line, I think that's an interesting point. I can't say off the top of my head that we thought a whole lot about that. My quick reaction I think is that it will have posed some computational challenges, particularly sufficient sample sizes and so on at a particular point it's going to be much more difficult to do. Yeah, it would be like an additional level of fitting that would be going on but it's something that's worth thinking about and I'm grateful for the comment. The question of the upper bound on the vulnerability line, we sort of defined the middle class but we sort of let the middle classes range all the way up to the top income level. That's a very valid question and we did not really try to pursue that. We were really focusing on this vulnerability dimension and I think you'd have to use a very different line of reasoning, a different set of indicators as to try to find that line that delineates or separates the middle class from say the rich. I haven't really thought much about that. It's possibly that it wouldn't be based so much on a risk of falling into poverty criteria and it'd be some other criteria that we would want to use. I've called the middle class basically anyone that's above the vulnerability line but I acknowledge that that's probably not entirely satisfactory if you want to do a real focused analysis of the middle class and you want to also keep in mind the distinction between the middle class and those that could be considered as the rich in this society. So that's definitely not something that we've handled adequately in this particular paper. Fixed effects, I think it's very true that I think in India, as in many of these other data sets that we were looking at there is this very important fixed effect in the error term which means that our upper bound approach which assumes no correlation at all is clearly not satisfactory and that also motivates the work that we were trying to do where we tried to estimate this correlation coefficient, this parameter rho in the methodology and I didn't go into great detail here but what we do precisely is to work with cohorts and we use the cohort level information to estimate to produce an estimate of what this row parameter might be and we then plug that row parameter into the parametric version of the method to then generate the transition probabilities. So it's kind of a merging of these different strands. There's a long standing literature on producing pseudo panels based on cohort level analysis. We're trying to work with the unit record level analysis here but then we do draw on that cohort approach to estimate a particular parameter that we need for our purposes. So we're trying to integrate I guess to some extent these different strands of research. The first question on socio-economic patterns of vulnerability. I didn't go into that here but it is perfectly possible to calculate the vulnerable and identify the vulnerable population say or the chronically poor based on this methodology where you have these transition probabilities. You can focus on specific groups of the population and then you can look in the data at the characteristics of those populations and so you can look at the characteristics of the chronically poor and try to distinguish those from say the characteristics of those who are likely to be transitioning out of poverty or those who are more likely to be transitioning into poverty and to look at these interesting characteristics that might differ across these different groups. That's something that we've done and we've demonstrated that that's something that you certainly can do. You have to be careful in terms of how far you pursue it because of the cell sizes and the sample sizes and there is a whole error that's associated with prediction coming out of the imputation procedure that we're doing here and on top of that error comes the sampling error that comes from small cell sizes if you're looking at small numbers of observations. So you have to be careful and cautious in how far you take that kind of disaggregation for that kind of more disaggregated analysis but we certainly can, you know, show things about how the role of secondary education versus the primary school only, you know, how that is associated in a different way between those that are chronically poor and those who are more likely to be transitioning out of poverty over time. So there are opportunities to do that kind of work and I didn't manage to go into that in this presentation. I think that's my reply so far, thanks. Very good. Then the second round, I think we had a question over there and then a question here, so let's start with the lady at the back. Okay, my name is Prudence Maghejo from the Investor of the Witwatersrand in South Africa. Please can you put the mic a little bit closer? Okay, my name is Prudence Maghejo from the Investor of the Witwatersrand in South Africa and thank you for the very insightful presentation. My questions are directed to the first presenter, Peter. I'm not so sure about the role of time-varying characteristics in the predictions from round one to round two when we are in the synthetic panel. Can you please explain a little bit on that and also I'm wondering the role of unobservable characteristics. Is it going to be captured with the role parameter that you are talking about because I'm thinking maybe we may have similar characteristics which are time invariant but in terms of unobservables, how does your methodology deal with that? Thank you very much. Thank you very much and then we go here in the middle. Andrea Cornia, University of Florence. My question is addressed to Jim Davis. Is there any attempt to evaluate the Gini assets as well as the asset, the total wealth level and the wealth composition in relation to changes, large changes in interest rates? I think Tony, yesterday mentioned that quantitative easing interest rates are falling which means that everything has been equal, everybody will borrow and this is the intention and the value of housing goes up but the rents of bond holders goes down and so I think I'm puzzled by the fact that the values which are being presented, they are conditional on the level of interest rate and could one simulate changes in all these variables? Has it been done? I mean, in the US, if you take the value of assets before or after the housing bubble and after the housing crisis, it must be quite different. Thank you very much. We do have time. One, two other questions. Finn, in front. Okay, thank you very much. My name is Finn Taub. The first question I have for Peter, let me begin by first of all saying thank you very much for two brilliant presentations. They are fascinating. I mean, I'd like to convey my appreciation. The first sort of thought around Peter's presentation is just so where does that sort of leave those of us who spent a major part of our life doing panel work? I just kind of pondering about are we now becoming, if not superfluous, but we should then lower our level of effort? I was wondering whether you had a thought or a reflection around that. And Jim, you sort of, in your presentation, you sort of quietly mentioned would you attention or would you care? I was sort of wondering whether you could say or elaborate a little bit on that. I mean, because your presentation obviously comes across as incredibly convincing. It's very professional, but I was just sort of wondering, could you reflect on some of the big decisions or some of the sort of really tricky points where you would say that mistakes are often made or that that due attention is not being paid? Thank you. Thank you very much. Any other questions? Please. Right, thanks. I've been a little bit. I'm from Norway. Could I ask, I think, Jim, what you have of your final slide there, the wealth GDP ratio, I mean, the difference between low-income countries two and four in high-income countries is quite significant. But it's always intriguing to ask, are there any countries that doesn't fit the pattern, any outlayers, in particular low-income countries which are significantly higher than the ratio of two and high-income countries that are significantly lower than four? And for Peter, I was wondering, I mean, the vulnerability historically could also be related to people who have been in the category of four people who have then moved to the vulnerable group and have a risk of falling back into poverty. Is there any analysis where you would actually also look at the history of those in the vulnerable group, whether they have always been just above the poverty line or whether they have actually been middle-class people going down into the vulnerable group or poor people who have actually moved out of poverty temporarily? Thank you very much. I think that now we have to go to the answer so that we can go and have some coffee in due time. Would Peter start? Sure, I think I'll try and go backwards again in order. On this question of the kind of work that's been done based on this approach and looking at more closely at the vulnerable and where they've come from and where they've gone and so on, so far in terms of my own research, I've been very much focused on the sort of methodological aspect of the feasibility and not so much on describing these patterns of vulnerability and so on, but that is something that's work that's underway and I might point you to a study that's very soon to come out of the Latin America region at the World Bank which is a study of chronic poverty which uses these methods to sort of identify who the chronically poor are given that in many countries we don't have panel data and we can't actually estimate this directly. They use these synthetic panels to try to identify who the chronically poor are and who those are that have managed to escape poverty and so on and then there's a fairly far wide-ranging analysis of circumstances and characteristics and history and so on of these groups. So that kind of work that looks more and really tries to tell the story of what comes out of these types of data is, I think, obviously the much more interesting part of this whole agenda. It's one that I've not, just for reasons of time, not been focusing on so much myself so I don't have a whole lot to say at this point but there is work that's underway and I think it's a fairly rich agenda which I think in some sense takes me to Finn's question about the future of panel data. I would really want to emphasize very, very strongly here that I see this work as by no means suggesting that we should stop collecting real panel data. I acknowledge that collecting panel data is a difficult thing to do. I remember being quite influenced when I was doing my graduate studies by a paper that Orly Aschenfelter, Angus Deaton and Gary Solon wrote. I think it was an LSMS working paper where they were just asking this question of whether the LSMS study at the World Bank should be getting into the business of collecting panel data or not and they were just highlighting how difficult it is and costly and complicated it is to do and I think that's also the reason why we see relatively few countries that are collecting panel data on a very large scale. Having said that, you know, there are efforts in the Indonesia Family Life Survey in Mexico. There's work underway. There are a number of countries where there's panel data being collected. The World Bank itself and the LSMS team has a very active program of collecting panel data in sub-Saharan Africa now and I think the insights that can come from those type of data are enormous, particularly when they're done well and the kind of insights that we can get from these synthetic panels that I'm describing here is really very limited. I'm describing, you know, a few things that we can possibly get some insights out of using these synthetic panels but it doesn't even scratch the surface of what we can do if we have real panel data. So it's more of a, you know, we're trying to be opportunistic. We have cross-sectional data in many countries. Let's try and get as much out of those cross-sectional data that we can but it's not really a substitution for collecting panel data in settings where that's a realistic option. At the same time we should just recognize it is difficult to do good panel data collection. Finally on the first question about the time invariant characteristics, let me try to very quickly give you the intuition of what the role is of these time invariant characteristics and why they need to be time invariant. You think about, so you have two surveys. You have a survey in say time zero and another survey in time period one and you're interested in the data for time period one. So you have all the households in that data set and you know what their income level is or their consumption level is in that time period one. But what you would like to know but you don't because you don't have panel data, you would like to know what their income was at time period zero. So that's your situation. You have the data on their income in time period one and you would love to know what their associated income was in period zero but you don't know that because you just have cross-section data. Our proposal is to then estimate a model in time period zero that relates consumption in time period zero to household characteristics in time period zero. And that's, we can easily do that with the data from time period zero. Now if the variables that we use in that model are time invariant then we can take the parameter estimate on that model estimated in time period zero and plug that in to the data for time period one because the X variable, the characteristic is not changing over time and that will then give us a reasonable basis in predicting what the consumption level was for a household in time period zero for each of those households in our data set in time period one. I hope that's clear and I'm very happy to talk perhaps afterwards over coffee to explain it better if I haven't done so but it's really to try to find that bridge between period one and period zero and using these explanatory variables these time invariant characteristics to produce that bridge, I guess with that bridge. Thank you very much and then Jim your last comments. Yes, very interesting questions from Andrea Cornia. It would be fantastic to be able to do a simulation on a global scale what the impact of changes in interest rates is on the distribution of wealth. As we saw a country like Japan where so much of personal financial assets are in the form of deposits, there would be much less impact than there would be in the United States where the value of their equities and their defined contribution pension plans would be very, you know, really quite sensitive to the interest rate. I think that realistically speaking the work needs to be done at the national level in some countries that have really good data and also have, you know, a body of macroeconomists who are used to performing simulations with dynamic general equilibrium models where they have heterogeneous individuals so there's actually a distribution of income. So I know that, you know, for example at my home institution there are guys that, you know, do work of that type. The US would be the natural place to do this because of the, because it has such really excellent wealth data. I think it would be a fascinating thing to do because, you know, there are forms of assets that are very sensitive to interest rates. It depends very much on what expectations are about the future, you know, which interest rate is the relevant one, what do you do about the risk adjustment, etc. So it would be very interesting to do but it hasn't been done and so there should be some PhD theses looking at that I would say. FinTARP is a question of what I was signaling about with due care. I think the largest thing is to make careful use of survey data. And so that means comparing the survey aggregates with the aggregates from the household balance sheets or other sources where people can get them making adjustments if necessary. Thinking about things like personal assets that may not be included at all in the surveys. So for some reason, which I don't quite understand in the English-speaking countries there has been a strong move towards including employer-based pensions and trying to do a better job of including them. So in Canada in the last couple of wealth surveys we have estimates of how much each household's employer-based pensions are worth and it's a really big chunk of wealth. Australia does the same thing. The UK has tried to do this in their last survey with kind of uneven results. So that would be another indication of where due care is required in this respect. If in a country like the UK there's a new survey they're trying to do something important and new well, you're going to have to look at that carefully and more carefully than say in a country where there's an established practice and they've taken a long time and people are quite confident about the quality of that stuff. So I'd say that's the major area that I think is tricky. The pension thing is especially tricky and this would be a point which is independent of whether you look at survey evidence or balance sheet evidence and the national accounts the value of our pensions is taken to be the value of the assets that are sitting in the fund that is supporting the pension. So if you have a defined benefit pension plan and you're working for the government or for some employer which is very unlikely to go bust and the markets go down as they did in 0708 how much has your pension wealth changed? Maybe it hasn't changed at all because you're not bearing the risk the government or your employer is bearing the risk completely opposite for defined contribution people So those are some tricky things and then finally oh yes, are there any outliers in wealth income ratios? I wish I had a sheet in front of me it's very dynamic we were seeing that just in the last few years the wealth income ratios in France, UK, the United States have been changing quite a bit in some countries they've been trending up in other countries they've recently done these fluctuations so it would even matter whether we were looking at year 2000 or year 2010, 2013 and then think about developing countries well, if you have a nascent miracle economy that's starting from a low wealth income ratio and they have a very high saving rate you might think well the thing in the numerator is going to go up really fast well the denominator is going up fast as well so it's a bit ambiguous of what's going to happen there and you can I'm thinking more in terms of models than I am in terms of facts I have to admit but in a slow growing traditional underdeveloped country you can have a high wealth income ratio and Raymond Goldsmith actually called attention to this and he said the wealth and the capital are not being used efficiently in these cases of stagnation and there were more of those cases back when he was writing them in art today so you can have a high wealth income ratio in a Southeast Asian country that's not growing very fast and it may be that when growth improves the wealth income ratio actually comes down and the other thing is again if we go back to developed countries it depends how much of your economic activities being run through the public sector versus the private sector so a country like the US like to try and do everything through the private sector in Scandinavia, Canada, Europe generally hospitals and colleges and universities and so on all these things are in the public sector but it is possible to locate those in the private sector as we look at the transition economies many of them had a rejection of doing things through the public sector and so over time they may end up as outliers on the wealth income ratio although that's not the case at the moment thank you very much Jim this ends our session we are running 10 minutes late but as we started a bit late so I took the privilege of extending the session a little bit thank you once again for Peter and Jim for presentations that definitely provided food for thought we don't live on good ideas only so let's go and have some coffee and next session we'll start at 11 thank you