Learning from interactions: tax evasion in an evolutionary perspective

The work studies the effect of interactions between fiscal authorities and taxpayers on income tax evasion decision, both in a theoretical and empirical perspective. Particularly, from a theoretical point of view, an evolutionary game (Friedman, 1998, Sandholm, 2010) is used to investigate the role of repeated interactions on the taxpayers’ decision to be honest, cheating or ghost, formalizing both the individual and population perspective. It will therefore be assumed a large population, divided into two sub-populations, taxpayers and fiscal agencies, were individuals are characterized by heterogeneous intrinsic characteristics (the propensity to declare their income to fiscal authorities varies from one individual to another) but by homogeneous decision making processes. For the purpose of studying the role of learning from interacting with other individuals, the replicator dynamic equation (Taylor and Yonker, 1978) will be introduced and used, and because all individuals behave the same way (they share the same decision making and learning process), the replicator equation represents both the individual and population perspective. Within this setting, we will present two models, the first one (in Chapter 2), which includes in the formalization different features on fiscal agencies’ side, such as effort in inspecting activities and the cost of inspections. The second one (presented in Chapter 3) captures instead fiscal agencies’ opportunity costs, as well as taxpayers’ costs related to the circumstance of being inspected. We will therefore investigate the effects of interactions between taxpayers and fiscal agencies in light of two different behaviours of fiscal authorities. In the model presented in Chapter 2, fiscal authorities’ actions are not related to costs associated with the circumstance of missing the right target in inspections, whereas in the second one, in Chapter 3, fiscal authorities are subject to an opportunity cost associated with the circumstance of missing the right target in inspections, when non inspecting the cheating and ghost taxpayers. Results show that in the long run equilibrium, the proportion of characters chosen by the members of each subpopulation does not change over time. Particularly - in equilibrium – the proportion of honest individuals is positive as long as the costs of inspections are less than the amount of taxes and fines recovered from audits, and the effort put by fiscal agencies is higher than the ratio between the costs and the amount recovered from inspections. In case the cost is equal to the amount recovered with inspections, the subpopulation of taxpayers is formed only by cheating individuals, which include also cases of ghost individuals. Moreover, higher tax rates are associated with a distribution of characters oriented to a higher proportion of cheating individuals (which in turn, can hide a certain proportion or their entire income), but also a higher proportion of inspecting tax agencies with respect to non inspecting ones. Higher penalty rates are instead associated with a higher proportion of honest individuals with respect to cheating taxpayers, and with a lower proportion of inspecting fiscal agencies. As lower costs of controls induce a higher proportion of honest with respect to cheating, it seems the effect of penalty and the perception of penalty leads to an adjustment in taxpayers’ behaviour towards honesty. Then, Chapter 4 presents an experiment conducted among students at the Faculty of Economics of the University “La Sapienza” in Rome, which aims at verifying whether learning affects individuals’ propensity towards honesty, as well as some of the results of the models presented in Chapters 2 and 3, such as the effect of tax rate and penalty rate on individuals’ choice of being honest or cheating. To estimate the effects of learning on individuals’ propensity to be honest, we use both a fixed effects and random effects models, controlling and non controlling for scenarios, as well as a modified version of the Directional Learning model, as in Selten and Stocker (1986). Results show that the effects of tax and penalty rate on individuals’ propensity towards honesty are statistically significant and share the same direction as shown in the models presented in Chapters 2 and 3 of this work. Moreover, learning affects significantly individuals’ propensity to be honest, and the results appear to be consistent with directional learning.