 This paper is on trends in global inequality using a new integrated data set. It's about global inequality. Global inequality is when we look at inequality among all citizens in the world. So like we treat the world as a unit, as a unit of analysis. And we know something about the trends, mainly using the Gini index and mostly focusing on household service. There is different contributions, especially since the 2000s. For example, Annan Segal, Bourguignon and Morrison, later Bourguignon, Luckner and Milanovic, the Wall Bank, and then later also Milanovic. And basically in this literature, do you identify that there was an increasing inequality at some moment there is a period with the evidences mixed in the 80s and 90s, depending on the period you consider, the study, you can find different results. And then there is an agreement on a decline in inequality. Some papers have looked at some sensitivity analysis, looking for example at different measures of inequality with different emphasis on different parts of the distribution. For example, Luckner and Milanovic and Bourguignon, a more innovative approach was to look at to put the emphasis on the top of the distribution, looking at the evolution of the income share of top incomes, and also using different types of data, using basically survey data, introducing corrections based on either national accounts or using some statistical approaches. We have some people looking at the impact of correcting for the top incomes on the level and trends on inequality. And also this approach by the World Income Distribution held at the World Inequality Lab in Paris where they produce a new whole set of more complex data combining different sources. And of course this affects the trends and the levels, especially in some cases the trends as we will see later. And there is also a literature, a small piece of literature looking at absolute inequality instead of relative inequality that is the most dominant approach and basically identifying that inequality was increasing if you adopt this approach, like Ravallion or Nino, Zarasu and others. So in this case the main contribution of this paper is to use a new integrated standardized data set that you probably already know the width that we hold that you know wider. So this data set includes information at the percentile level since 1950 for each country and then it's aggregated to construct global inequality. It's based on household surveys so there is no correction in this case for top incomes even I will show you for analysis some of these corrections. It's based on the World Income Inequality Database that is there since 2000 and the idea is to provide a broad overview of trends in global inequality, more detailed and systematic that most of the literature I mentioned before and a comprehensive analysis looking at overall inequality and inequality between and within countries and with different approaches. So the data is, I'm being consistent, increasing the time coverage and geographical coverage and trying to make the data consistent across countries and over time. We work with the entire distribution and this gives us the possibility of looking at the sensitivity to different inequality approaches. We will look at absolute and relative or whether you put more emphasis at the bottom, middle or top of the distribution as I mentioned no correction at this moment for top incomes but they will do some exercise with some corrections to see the impact of that and also it's innovative in this paper that we quantify the contribution of each country to total inequality and to the changes over time to identify the main drivers in a more precise way. So about the data, everything started with Daning and Squire in 1996 so the width is gave continuation to this compilation of information of inequality. It has been used by some of the literature already in different formats and it's basically fed with information from the main income distribution providers like Povkalnet or now the platform in equality and poverty at the World Bank, Luxembourg Income Study, Eurostat Set, Lucky Clock and many, many others including research studies. So we have information for basically all countries and some other special territories with information with the genie and the shares for this at the side, quintile or bottom and top five percent, a mean median and more recently we added also other inequality indices like the entropy family, Atkinson, Palma, et cetera and has very rich information but in general it's a complex database because it has information, multiple options for each country and year so we made a selection of the best country series for each country given priority to those that are more internationally comparable and estimated for each country and year the percentile distribution from aggregated data using the short rocks and one algorithm that is quite known in this literature. Then we integrated these different series for countries and standardized them to make them comparable over time and across time. I'm not going into the details but basically taking advantage of the information they were lapping of different series in one country or doing some regression analysis starting the relationship for example the main issue here is to that's for some countries you have information on income from others you have information on consumption and how to translate that into the same measure that is per capita measure of welfare that is per capita income and then we studied empirical relationship between the distribution of income and consumption in those countries where you have both and then you extend that to other countries. So we end up with a balanced panel of countries from 1950 to 2020 and that is aimed at being each year updated with new information for the average per capita income in each country. We use GDP at this stage in PPPs. This GDP information comes mainly from the World Development Bank, World Development Indicators since 1990 and we reconstructed historical series backwards or for countries not included in WDI with the Madison Project or even with the Penn World Tables. I will also show and I can advance that this is a very relevant element that changing the source can affect the inequality trend in specific periods and it's not included in the database but in this study including also some projections for 2021 to 2027 projects on the evolution of GDP using IMF database. Then of course when you build a global distribution you need to make a lot of imputations or interpolations between survey years because if the country doesn't have a survey each year or extrapolation before the earliest survey or after the last survey and with a few imputation for countries with basically no information at all and you end up with a wall or regional percentile distribution and estimate just to give you an idea of the quality of this exercise of course if you look at the after 2000 we have information for within a window of plus minus plus five years for basically almost one hundred of the population especially after the 2000 this drop here is because in we don't have information for India more recently even that probably is going to be available soon and of course the more you go backwards the lower the quality because we relay on the surveys that are further away than five years but even in the worst scenario 1950 we have information for half of the wall with the survey falling between my plus minus five years no but of course the quality is better after 2000 obviously the data is open access is a tariff friend you can have information on all technical notes each all the necessary elements including state ago to reproduce all the process that we did okay let's go so what can we say with this data about the training in equality first I look at what is known as absolute inequality so it's not the dominant view on inequality but in terms of the academic research but some people claim that many people think in terms of global of absolute inequality when they think about inequality technically this means imposing the translation invariance principle basically the idea is that if there is growth you need that every person receives the same amount of dollars for inequality to be constant so it's quite demanding no so if you increase in the same proportion but not the same amount then inequality increases and the same applied for reduction so it's really quite demanding and what we see is in terms of you can see here the change over time between the big periods 50 and 80 18 to 2000 and 2000 2020 and you can see here that in every period basically the richer you are so the higher the percentile the larger is the increase in income so basically the story about absolute inequality is very short it's inequality increase all the time because you can see that for example for this period 80 to 2020 is what Martin Ravallion called the serpent graph but the serpent changes the shape but still always like a pro-rich growth in terms of absolute amounts and if you translate this into measures here you have the genie index you can see exactly that inequality globally inequality increase all the time except in specific periods of recessions but because we understand the recessions globally absolute the absolute approach basically says that the if the recession affects proportionally every income inequality decline so it's a bit weird to apply this criteria when you have a recessions no so basically increasing inequality both between countries and within countries and of course overall in fact there is Lawrence dominance if for people familiar with the Lawrence a criterion which means that any index that is Lawrence consistency would tell you the same story no as you can see here so for more details about this analysis by countries we have another paper in which we quantify that also for most countries in the world this is true inequality increased no and this is consistent with other papers like a Nino Zarathua or Rupen Tarp and Anand Segal all my uncle computations using data from the world inequality lab that you get basically the same story for the periods in which they apply okay let's go for the richer story that is when you look at in relative inequality then it's when you have a more nuanced story so this applies the scale invariance principle saying that what keeps inequality constant is when everybody sees their income increased by the same proportion not the same amount in dollars but the same percentage so if the percentage is higher for the richer people inequality increases if the percentage tends to be higher for poor people inequality tends to decline in this case these are the growth incidence curves and here you see how the growth pattern worked during these subperiods in particular for example in 1980 2000 you have this peculiar shape that is what Ronco Milano is called the elephant graph no and this was the elephant I found that matches this graph perfectly and you can see after 2000 the elephant the trunk of the elephant is going down and the first video was completely different okay trans how this translates into in terms of the evolution of trends in the income share held by different population groups you can see there was a large increase even that the scale doesn't allow to see how in because it was it was more than double with this data the increase in the bottom 40% after the 2000 but still is a very small amount considering that we have 40% of the population is receiving like 6% of the total income so is a the inequalities are still huge but there was a improvement especially after the 2000 you can see at the same time the top income 10% was relatively stable until there was a large increase around the period in which the Eastern Europe communist regimes collapsed and also there were some large increases in the top in the concentration of top income countries like the US etc there was a huge increase but since also around 2000 it was declining even that this is decelerating and getting softer so also for the middle 50% so this period between 80s and 90s is when you see that the top increased while the middle decrease and the bottom increase so there is if you look at the bottom 40% you will say probably that inequality declined if you look at the top 10% you will say that inequality increased no that is where the sensitivity to parts of the distribution is more relevant no however after the 2000s there is more agreement because even if you look at here unless I will say with with as I showed you you look at the very very top in general inequality also declined and the Palma index is just the ratio between top 10% and bottom 40% and it has been declining steadily since the end of the 70s and but also starting to be more stable sorry more yeah with decelerating at this decelerating rate here you can see also with the since 1920s had including the projections the IMF projections for the GDP and assuming constant inequality within countries to just see what could we can expect in the years to come and somehow even that we have an increase during the of or a higher concentration at the top and lower the bottom during the the pandemic the process seems to be resumed after that this is just what I mentioned before when you look at the very very top of the income distribution I don't I'm not doing this with my data because it's not designed to have an accurate estimate for these very precise estimates but it's when you find a different trend or like an increasing inequality or at the very base some stability in terms of inequality measures that put more emphasis on different parts of the distribution you can see the gene index started to decline basically since since the 90s and mainly during the 2000 but you find something similar with the mean log deviation that puts more emphasis at the bottom of the distribution so the fall of inequality started earlier sorry but if you look at the an index like the general entropy index that is related with the coefficient of variation then the declining inequality started later because it's it reflects this increasing inequality when the top 10 percent was going up the main difference here in my view is when you put more emphasis at the bottom of the distribution and you use an index like the general entropy measure for the parameter minus one because then the story is totally different you find that declining inequality in the initial years some stability and then was followed by an increasing inequality indicating that when if you care much a lot about the poorest people in the world then the story that inequalities declining is not true but it's increasing and how this match the system literature well you have here that it matches quite well estimates from other papers based on similar survey data like Lackner in Milanovic or Milanovic or Nino Tharatou and others also David's insurance that here the difference in levels is because this is a study on income and wealth and then they don't use PPPs so they use exchange rates to convert incomes into dollars and then of course the level of inequality is higher but the trend as you see is the same here the main difference is with the information from the world income distribution that the world inequality lab but you can see also that the trend after 2000 is similar maybe the decline is smaller and may the main divergence comes from between night meet 90s and 2005 but this is not due to the correction of top incomes as I will show but how we measure average income by country so to see this for example here you have in solid black to have the our estimates with a weed to ice is you know wider and with only one I is the world income inequality lab so do you have this discontinuous line that is the top 10% with their estimates that are higher especially after the 90s but you can see general that the trend is quite similar and here this this other curve in red discontinued is what I call the hybrid distribution that is the weed where I correct the top 1% getting in the information from the world inequality lab to see the impact of only changing the top 1% and nothing else and then you can see that the trend is quite parallel even that the gap increases at the end meaning that inequality the following inequality is smaller if you make these corrections for the top 1% okay and this impact that I estimate of correcting for the top 1% is in line with previous literature Anna and Segal they also used something same a similar approach with an earlier version of the top income database so very similar about two genie points initially that this increases during the early 2000s and then it's like four genie points this is between some estimates with that Milanovic and Lackner made initially or Milanovic later they basically impute the top so the part of the national income that is not reflected in household service to the top 10% of the population and they get these higher estimates or this is Jorda and Niño Tarazua that they may follow a completely statistical approach so they don't use any national accounts or tax data to make the correction and they find like a bigger impact okay then this this reduction in inequality you can see in general the trend in overall inequality is driven by the trend in between country inequality first increasing and then declining while the trend in within country inequality is the opposite it first decline and then increased especially during the 90s and then but keeps increasing or being more stagnant and so we can say that overall inequality decline after 2000 for example because inequality between countries decline even that inequality within countries increased this reflects also previous literature for example Burguignon for the period that we saw overlap if we do we estimate using the sharply the composition I'm not going into details now and see the relevance of the between country inequality to explain overall inequality you can see that with any index you find a similar curve in the sense that inequalities between countries became more important until mid 80s for example and then starting at that level or decline in many indices with different sensitivities to the income distribution at different parts you can say now that inequality within quantum within countries is more important than inequality between countries some robust analysis as I mentioned so the story after 2000 is if I use different measures for the average income in each country so the one I use in the week basically on WDI complemented with the other sources and I use the alternative the GNI instead of GDP or I use the Penwall tables or the Madison project basically the same story except the period between 90 meet 90s and early 2000s in which there is a large discrepancy between using WDI or Penwall tables estimate and why is that okay China no not a surprise because they how they perceive the the Chinese miracle during this period is different so in the Penwall tables China starts with a relative income compared to the average in the world at a higher level and then the growth afterwards is smaller than if you use the WDI so depending on the source basically this impressive growth in China that is behind the declining inequality between countries started later something force for India but in the case of internet is not that relevant because the trend in that period in the period that there are discrepancies not that important okay then I use an approach that I'm not going to the test with I estimate for each country or region the contribution of that aggregate of country of that country or a nation of countries to overall inequality to identify and quantify of course we already know that the decline in between countries what due to China but I want to quantify that in a more precise way inconsistent and and looking at the contribution to each component and we can see here that the contribution to inequality of East Asia and Pacific is shows this large decline that driven mainly by China and that is responsible for the decline in inequality between countries that we observe before there are contributions from other regions like Europe but much smaller and and basically was due more because Europe is less less important in terms of population the opposite to sub-saharan Africa no you can see it sub-saharan Africa is increasingly contributing to inequality but it's mostly due to a composition effect a higher level of population living in proportion of population of the world living in sub-saharan Africa and the other areas they are important but much less this is just the case of China where you can see the contribution of China to the between country inequality and to the within country inequality and how it contributed to more inequality between countries until the structural reforms and then China started growing very fast and then this huge contribution to the decline in equality but the opposite happens with the contribution to within country inequalities so in the period in which China declined in equality it has a large contribution to that but then the contribution goes in the opposite direction but this contribution is a smaller than the other one and then the net effect is contributing to reducing equality this is something similar to India it's a same similar story but very different levels and yes for the US or for other countries it has to finalize for example to as a way to quantify this contribution for example you can see that for the period 2020 there is a fall in genie in eight point two points and you can see this is a composition in inequality between countries in the world and inequality within countries and this is like a composition for you to change this population and of course here you can see that out of 11 points of the declining inequality between countries in the world 6.5 are explained by China 1.9 are explained by India so basically these two countries account for almost the whole effect and the same happens with increasing inequality within countries 1.8 points of the overall for the world 1.7 came from China 0.5 from India so India and China could also explain basically that increase it doesn't mean that in other countries didn't increase in equality but the net effect cancels out so some there were some increases and declines okay then that's it thank you very much