Deriving the Expected Value of the Tax Underreporting Rate
situations , these models must contain assumptions about macro-level phenomena that are as close to real world experience as possible ; including a tax authority ’ s audit rate , and the taxpayer population ’ s tax underreporting rate . These are critical design elements for ABM developers who seek to create a “ valid ” model .
Averill Law ( 2007 ) defines a “ valid ” ABM as one that provides an accurate representation of the real world system under study . In practice , validating a simulation model involves comparing the model ’ s output to data from the actual system ; if the two sets of data compare “ favorably ,” then the model is considered valid . Some authors have pointed out the lack of sufficient data on real world systems often makes it difficult to decide what constitutes a “ favorable ” comparison . The inability of ABM developers to conclusively demonstrate the validity of their models has made some traditionally-trained social scientists and non-academic practitioners reluctant to use ABMs ( Leombruni & Richiardi , 2005 ; Marks , 2007 ). For this reason , ABM and other model designers require tax compliance data that is as accurate as possible . Although enforcement data on elements such as the audit rate are available to researchers in designing tax compliance models , underreporting data are not .
The closest researchers have come to approximating the tax underreporting rate is when Mark Phillips ( 2011 ) manually counted the number of individual tax returns containing underreported income for all the returns sampled for the Internal Revenue Service ’ s “ tax gap ” study . Phillips found that approximately 32.5 percent of all individual returns contained underreported income .
While the work of Phillips remains impressive and is a significant contribution to the tax compliance literature , it has two limitations . First , it only counts underreported income , not net underreported tax . Measures of these two are not identical , which is further reason to attempt an approximation of the underreported tax rate . Yet , Phillips ’ s finding can help gauge the overall accuracy of the method derived in this paper . An intuitive assumption is that the rate of underreported income and underreported tax should not be unreasonably divergent . Stated another way , the two should be relatively close in measure .
Second , Phillips only estimates the rate of underreported income for individual tax returns . Similar rates for other reporting areas such as corporate , estate , and employment taxes remain unknown . Although “ tax gap ” studies estimate the magnitude of underreported tax in U . S . dollars , these studies do not report the relative frequency of returns containing net underreported tax . Given Phillips ’ s recent findings , the “ tax gap ” underreporting rate of about 18 percent cannot be used as a viable proxy for the net underreporting tax rate . The underreporting rate appears closer to 30 percent based on the work of Phillips .
Without an accurate expectation of the net underreporting measures across tax reporting categories , results of ABMs and similar research might not be as meaningful . The results certain will not be seen as “ valid ” per Law ’ s definition since the input might not represent the data of real world experience . A method that closely approximates the expected value of the under reporting rate given enforcement
4
The author ’ s previous use of this proxy by substituting the “ tax gap ” estimate for the underreporting rate was , therefore , in error ( Manhire , 2015 ).
6