Tax fraud is an important economic issue in every country. IRS (The USA Tax System) wants to predict the USA tax income in the next ten years based on people's income database in the past years. Clearly, every citizen prefers to pay lower income tax. If IRS wants to study PaidTAXAmount i =f (income i, age i, race i) e where i denotes person i in the income database, which types of model is more appropriate to predict the USA tax income in following years? Multinomial Probit because of violation of lIA property. High income people are similar to each other. Tobit Regression Truncated Linear regression Truncated Linear regression wherein incomei2 must be included to pin down non-linear trend, i.e., high income people want to pay less tax