No Elephant in the Room? Noisy Convergence and the Global Growth Incidence Curve

In recent decades, worldwide incomes have grown enormously. But how has this growth been distributed? According to data from 1988-2008, the percentiles in the middle and at the top of the world income distribution have grown the fastest, while percentiles at the bottom and around 80 (the "global upper middle class") have stagnated (Lakner and Milanovic, 2013). This striking pattern was dubbed the “Elephant curve†and, in media coverage, commonly interpreted as a redistribution of incomes from the global upper middle class towards the top earners, plus fast-growing middle classes in emerging economies (especially China). However, I show that there is a simpler, statistical, explanation for the Elephant pattern: random (“noisy†) growth affecting individuals whose initial incomes follow a bimodal (two-peaked) distribution. Famously, the world income distribution was indeed bimodal in the 1970s-90s. I simulate a process of income growth with these two features, plus a third - a modest amount of convergence (expected income growth is decreasing in income). The simulated growth incidence curve almost exactly replicates the observed data. Thus, this exercise suggests that the Elephant curve may be a statistical artifact reflecting a historically unique situation, rather than revealing anything about the forces of globalization. Finally, I review further evidence that also supports the artifact hypothesis, and I discuss additional implications of noisy growth for the very top ("one percenters") and very bottom (eradication of poverty) of the world income distribution.