Research of the Shadow Income Estimates Robustness on Microdata

The study attempts to test the stability of hidden income estimates to the analytical definition of self-employment. The theoretical model of household consumer behavior was implemented. The main assumptions were checked: the category of expenses is reported correctly; the propensity to hide income only exists for households with a certain percentage of income from self-employment. Clothing costs were used as a category of expenses, in contrast to the classical function of food consumption. The model was evaluated on RLMS data for the period 2009-2018. The use of a log-linear model with fixed effects made it possible to avoid defiling. The first task was to check the stability of the hidden income estimations to the border, relative to which households were qualified as self-employed or employed. The results were obtained for 3 periods. The estimates obtained differ between groups within the same period. We conclude that additional research on the method of selecting the correct border should be done. Based on the results two hypotheses were tested. The main hypothesis of the study about the sensitivity of hidden income estimates to the analytical definition of self-employment was confirmed. However, the assumption that the largest step into the shadows was made by Entrepreneurs after 2014 was not confirmed (1.8%). The Combined group showed a higher concealment rate (3.6%). This may be due to a decrease in the other income part (which is up to 50% of their wealth) during the crisis period. That motivates households to hide income to maintain it at the same level. At the next stage, we divide households into 9 groups by the share of food expenses and the share of the total number of children and pensioners in the family. On the obtained subgroups, we estimate a similar model in two ways: by introducing a random effect on the constant, and by introducing an additional random effect on the dummy-self-employment coefficient. Estimates were obtained for the period 2009-2018, with 20% as the self-employment limit. Note that our results are sensitive to the method, which also depends on the original specification. The results are contradictory: the hypothesis of a low percentage of concealment in poor, socially obligated groups is rejected based on the results of a model with fixed effects. At the same time, the estimation of income concealment based on the mixed model confirms the hypothesis. Based on the overall results of both models, we tend to trust the results of the evaluation of the mixed one more. However, both models show similar trends and dependencies.