가계패널조사 저축과 순자산 변화의 불일치에 관하여

I compare savings and net asset changes of households using four Korean household panels--the Korean Labor and Income Panel Study, the Korea Welfare Panel Study, the National Survey of Tax and Benefit, and the Survey of Household Finances and Living Conditions. Accounting principles require savings (= income - expenditures) to be identical to net asset changes (= asset changes - debt changes); this identity does not hold in actual data. There are two reasons for the failure of the identity: First, income, expenditures, assets, and debt items are not consistently defined due to inconsistent treatment of unrealized capital gains and other issues. Second, respondents of household surveys provide incorrect answers. My analysis shows that the gaps between savings and net asset changes (which I call cash flow errors) are too large to be explained by inconsistent definition of survey items, and that household and survey characteristics likely to be correlated with survey response accuracy have strong explanatory power for cash flow errors. I first examine summary statistics of cash flow errors calculated for each household-year of the four Korean household panels. The mean of absolute cash flow errors exceeds 60 million KRW (about 43,000 USD at the exchange rate of 1 USD = 1,400 KRW), and the mean of absolute cash flow errors divided by expenditures exceeded 2.0. Then I examine how household and survey characteristics are associated with absolute cash flow errors. I find that absolute cash flow errors are especially large when household members change and when households move. Income and education levels of heads of households reduce absolute cash flow errors; assets and ages of heads of households increase absolute cash flow errors. In the National Survey of Tax and Benefit, respondents are asked to submit copies of tax return files; when respondents comply with this request, absolute cash flow errors are smaller. In the Survey of Household Finances and Living Conditions, starting from the survey year of 2017, some income, expenditures, and liabilities items are filled by administrative data; this use of administrative data, however, does not reduce absolute cash flow errors. Cash flow errors imply inaccuracies of household savings and net assets variables. Concerns for the inaccuracies of household net assets have been expressed by many empirical researchers. To explore the possibility that excluding observations with large cash flow errors may improve empirical analysis utilizing net assets of household panels, I estimate the wealth effects from the household consumption function after applying alternative sample selection criteria. I find that wealth effect estimates vary across household panels when observations are not selected. However, when observations with large absolute cash flow errors are excluded, wealth effect estimates from the 4 household panels become similar to one another.