3.3.1. Excluding Capital Gains

Among the different sources of income (wages, dividends, interests, business incomes, etc.) the capital gains declared are realized, and only a fraction is reported on the tax returns. Moreover, capital gains feature an extreme volatility in the short-term, and keeping their aberrant values in the sample skews the curve of the overall trend. Unequivocal comparisons therefore require the exclusion of all capital gains from the upper income time-series. Piketty and Saez (2004) note (p. 33):

‘From 1913 to 1933, 100% of capital gains were included in net income (there was no capital gains exclusion); from 1934 to 1937, 70% of capital gains were included in net income (i.e. 30% of capital gains were excluded); from 1938 to 1941, 60% of capital gains were included in net income (i.e. 40% of capital gains were excluded); from 1942 to 1978, 50% of capital gains were included in net income (1942-1943) or in AGI (1944-1978) (i.e. 50% of capital gains were excluded); from 1979 to 1986, 40% of capital gains were included in AGI (i.e. 60% of capital gains were excluded); from 1987 on, 100% of capital gains were included in AGI (there was again no capital gains exclusion).’

After withdrawing the share of declared capital gains from total income, one problem occurs. The subtraction of capital gains in each fractile series leads to a significant reduction, if not the elimination, of the highest incomes. This is the case for these tax units filing returns with capital gains as a main source of income. In other words, the removal of capital gains generates a new series whose ranking is inconsistent with the initial income-class intervals as displayed in the IRS publications. The rank reversal is an issue that needs to be corrected for. According to Piketty and Saez, the top fractile AI99.99-100 needs to be increased by 40 percent, the fractile AI99.9-99.99 by 2 percent, AI99-99.5 and AI99.5-99.9 by 1 percent to preserve the initial income ranking. AI90-95 and AI95-99 are left unaffected. These corrections are here applied to the fractile series uniformly across states. It seems reasonable to apply the treatment of capital gains uniformly across states because splitting capital gains into 51 different areas also divides the biasing effect by the same number.