 Good afternoon everyone. I am thankful to the organizers and very excited to be here to present our work This is a very The data is very warm. So it's exciting to see the results a few years back we were interested in studying inclusive growth and We wanted to construct a consumption profile for countries within and across and look at people within and across countries and We attempted to do that initially with the existing data sets but we found that it was a very difficult effort both in terms of Methods were not clear and replicability was not there. So we embarked on an effort to create a database from scratch My co-authors here are Arjun Jayadev and Sanjay Reti At the outset before I start let me tell you that developing a global data set Which captures gives a consumption profile is a very difficult exercise as you know from yesterday's discussions also So we do not claim here that we know the best answers to everything But we are working on what has previously been done and trying to improve certain things and we are very open and flexible to your suggestions as well as ideas So let me start. What is GCIP? The global consumption and income project is basically consist of two global data sets a consumption data set and an income data set Which will give us a portrait of consumption and income within and across countries over time Main goal to start the project was we wanted to be open and transparent We want to provide both the code as well as data so that it's replicable as well as people can Do a change our methods use alternative assumptions and See how it looks The benchmark database which we currently have Gives us a consumption and a consumption profile and a separate income profile For our 133 countries spanning the last half century, which is 1960 to 2012 We have input tools for filling in missing data as well as creating portraits for aggregates of countries a user can create an aggregate For any country they want sorry these slides. I'm just moving the applications which we currently think about tracking the evolution of material living standards Over time as well as using the database we can calculate any poverty measure any inequality measure and inclusive growth measure Researchers might use it to both explain the consequences and the causes of inequality But we expect that once the data is public people will use it in many more ways So what is unique about GCIP and The several things a few are we provide annual portraits of consumption and income most current data sets I have to some benchmark years and Pick service we do a annual portrait and I'll get to why that is important in a minute We have a broader temporal and geographical reach going back say 50 years We provide separate consumption and income estimates Several data data sets currently Mix income and consumption distributions as well as mean incomes and use them interchangeably within that is wrong Because these are two different concepts, so we try to get that a separate consumption and an income estimate We provide tools for aggregation of user defined countries groups of countries So user may specify that they are interested in bricks or any group of three or four countries or A larger grouping and you can get a consumption Portrait out of it. And as I mentioned earlier, we'll provide full documentation of our methods Coming to construction of the data set we have four four steps The first step is we collect data on related consumption or income distributions The second is the standardized distributions. This is not complete standardization We but we convert the consumption distributions into equivalent income distributions and vice versa I'll come to the details of each of these steps in the third step we obtain an estimate of mean consumption from the service because we are interested both in the means as well as the Both in the levels as well as the distribution in the fourth state using these two We create a consumption and income profile So the first step is currently our data sources are UNE wide or wider WID pop cabinet and alias But we are open in principle to get data from other sources a Restriction which we put on our data is we restrict all our universe to per capita surveys per capita survey has the benefit that they're easy to understand They have a counterpart in GDP The the side effect the the negative point is they do not account for economies of scale But they are also the most commonly used Lastly is for country years we find after collecting the data for several country years. They would be multiple surveys So using the canberra group recommendations and our own measures we apply a lexicographical ordering To get a single survey for every year The ordering Basically, we prefer surveys which have mean information over surveys which do not have mean information for the consumption survey we prefer consumption consumption surveys when there is a choice Income surveys that are closer to arriving at the total net income concept are preferred But we do not here. I want to mention that we do not standardize The income concept across the board because we found that That not to be possible because of lack of data We prefer surveys from pop-cali net over alias and WID This is mainly because pop-cali net and alias have better means data than WID and some other other criteria So in the second step We convert income distributions to consumption for our GCD Database and vice versa for the other database So, how do we do this? WID has around 120 country years Which has both a consumption survey and income survey for the same country for the same year We use that to get a relationship at the quintile share level So for each quintile we do a regression and get a relationship of how the shares are related So you'll find that the relationships will vary by quintile For the bottom quintile you'll find that the consumption share Comes out to be 1.85 times the income share Whereas in the top quintile, it's the reverse. This consumption share is 0.86 times the income share This is a univariate regression We tried out other alternative Specifications, but we found that this worked the best But again, this is a work in progress and we are open to other alternatives here I adjusted our squares are pretty high and our confidence intervals are low. So we are a bit confident of these results Let me show an example of how this works This is Mexico 1989 which has an income survey. We want to convert it into a equivalent consumption survey the original income shares are these columns and We apply the regression coefficients mentioned earlier to get these shares, which is bottom moves from 3.93 to 4.66 But you'll find that because they are five different regressions. This doesn't add up to 100 So the choice which we made here was whatever shortfall is equi proportionally distributed across the five quintiles So now you have The bottom moves from 3.93 to 4.81 Whereas the top in share has moved from 56 Income share to a equivalent consumption share of 50.23 We do this across the across the data set where we have an income survey for the global consumption data set and the reverse for the income data set Coming to the third step now we determine the mean levels. We prefer we use survey means we do not use national account means so In the first step we Get the means data from the surveys very getting the related distributions from if the Mean is an income mean we convert into a consumption mean by using The consumption in nominal GDP ratio to get a equivalent consumption mean. This is the same method followed In earlier versions of crock and net Um For survey years without survey means we do interpolation or extrapolation by using growth rate data I will come to the exact interpolation method in a minute Lastly we convert all the consumption and income means to local 2005 LCU units which are finally converted into 2005 PPP Here let me mention that we realized that there are several problems with PPPs They are inherently flawed and the results might be quite sensitive to base years But this is the first version where we are trying to first use PPPs and At later point we will try to show the sensitivity to PPPs now the fourth step is can be We estimate the Lawrence curve from the distribution data We use the standard parametric regression methods to first prefer a GQ Lawrence curve, which is the generalized law quadratic form But if it fails to give a valid Lawrence curve, we fall back to beta Lawrence curves At points we have seen that both of them fail and we choose a piecewise linear method, which is our own method Once we have the Lawrence curves Lawrence curves are not usually estimated by the most existing data sets They assume an average Same average income for within deciles, which underestimates the within country inequalities In the second second stage we use the mean and the estimated Lawrence curves To get the mean consumption and income levels for each decile or we can go down to each Ventile level Now this is all for all for the survey years, but there are several gaps because 1960 to 2012 there are many non-survey years for that we estimate the consumption and income profile by using the national accounts growth rate data to interpolate interpolate is when a Non-survey year falls between two survey years We use the growth rate data and the time weighted average to get at the distribution and the mean levels But if the survey falls before the first Survey year or after the last survey year Then we extrapolate again using the national accounts growth rates So we have a database now. It's a global consumption database Which has 1340 country observations from 133 countries 45% of these surveys are consumption surveys all this data, which I'm showing is only for one data set The GCD not the GID But if you look at by decade coverage The coverage obviously in 60s and 70s is very poor around only 35 to 40 countries have So surveys in this period. So there's a lot of extrapolation happening here And we have to be cautious of results for 1960s and 70s There there are other things about 60s and 70s is we do not have data on planned economies and so That so that is another reason to be cautious of these results Overall, we find that Pavkanet forms around 26% of our surveys 62% of our surveys not preliminary results As I mentioned before I present the results There are several things to be cautious about one is the data coverage in 60s and 70s So we don't present much results in that period Second is the PPPs as I've mentioned and third is that we do not currently make any adjustments for the Non-coverage at the top level, but we plan to do that soon So let's look at how how these preliminary results look This is a kernel density plot of consumption log of consumption You'll find that in the 60s and 80s the 80s you'll have two peaks In the 2010 this has there's only a single peak almost a single peak This is basically the growth of China which has filled in the middle part These are the global generalized Lorentz curves generalized Lorentz curves are Lorentz curves scaled up by the means and These are welfare implications. So here you'll find that 2010 Clearly dominates 1990 and 2000 data So everyone in 2010 according to this is well off at least in the consumption sense The 2000 barely dominates 1990 Global inequality what has happened? It has declined from the 80s and 90s when it was in the 0.7 range to around 0.63 right now but if you look at this Excluding China the pattern looks different and And It's increased till 2000 and then a decline Excluding India doesn't doesn't make that much of a difference But again here these are relative measures of inequality and this might change Once you use different measures Global poverty also has declined dramatically, but again removing China doesn't show as dramatic a decline These are global growth incidence curves plotting cumulative growth rates You'll find the median as almost double over 1990 to 2010 period This is where the Chinese population is in the global median and the top which is the developed country population is stagnating But again, this was the relative measure the absolute measure looks different the gains In the lower percentiles are still lower than the top percentile So who is in the bottom quintile? So you'll find that in the 60s and 80s lots of Chinese when the bottom quintile around 60 60% of Chinese way But now less than 20% of Chinese are in the bottom quintile The space left behind by China is sort of occupied now by India and other countries Indians which were not in the bottom quintile now are around 40% in the bottom quintile What happens in the top decil in the top decil? Korea shows a dramatic change in the 680s It was nowhere in the top decil now around 40% of Koreans are in the top decil and There are several interesting results if you look at countries Inequality estimates I'll not go into details But this is to show that you can use the aggregation module to aggregate any sort of countries bricks or by region or by any other user defined countries This is interesting. These are Partial orderings we wanted to take this data as well as life expectancy and look at How partial ordering of countries would look this is for Asia Pacific So how this diagram operates is if there's a direct line connecting the two and the one at the higher level dominates the lower level so Japan here has a higher consumption profile Which means that it has higher consumption for each quintile as well as higher life expectancy all of all of that Vector is greater than in Korea if the line does not connect two countries then the domination. There's no relationship between those two Here you'll find interesting results that India does not dominate any other country So and as dominated by Nepal and several other countries in this whole This is just an illustration of an application of the of the data This is for OECD countries Yeah, you'll find us dummy does not dominate a lot of these countries as well as Germany Finland Island Israel it dominates Spain Korea and at lower levels This is for the world. I'm not again going details, but there's just an interesting plot to understand the various relationships So in short GCIP is a resource for understanding it's an evolution of material living standards within and across countries It it is a tool used it can be used to study poverty inequality inclusiveness and growth the implications of alternative assumptions and robustness of conclusions and We can study casual determinants of poverty inequality and inclusivity Finally to end. This is very much a work in progress. That's offers diverse possibilities We plan to be flexible in our approach as well as we are flexible and open to alternative methods and suggestions We seek to build and improve on this database in the months and years to come with involvement of interest specialists and world public and general You can follow us. This is just a webpage right now, but you can follow us here at gcip.com. Thank you