Source code for taxcalc.behavior

Tax-Calculator elasticity-based behavioral-response Behavior class.
# pycodestyle
# pylint --disable=locally-disabled

from __future__ import print_function
import copy
import numpy as np
from taxcalc.policy import Policy
from taxcalc.parameters import ParametersBase

[docs]class Behavior(ParametersBase): """ Behavior is a subclass of the abstract ParametersBase class, and therefore, inherits its methods (none of which are shown here). Constructor for elasticity-based behavioral-response class. Parameters ---------- start_year: integer first calendar year for behavioral-response elasticities. num_years: integer number of calendar years for which to specify elasticity values beginning with start_year. Raises ------ ValueError: if start_year is less than Policy.JSON_START_YEAR if num_years is less than one. Returns ------- class instance: Behavior """ JSON_START_YEAR = Policy.JSON_START_YEAR DEFAULTS_FILENAME = 'behavior.json' DEFAULT_NUM_YEARS = Policy.DEFAULT_NUM_YEARS def __init__(self, start_year=JSON_START_YEAR, num_years=DEFAULT_NUM_YEARS): super(Behavior, self).__init__() self._vals = self._params_dict_from_json_file() if start_year < Policy.JSON_START_YEAR: raise ValueError('start_year < Policy.JSON_START_YEAR') if num_years < 1: raise ValueError('num_years < 1') self.initialize(start_year, num_years) self.parameter_errors = ''
[docs] def update_behavior(self, revision): """ Implement multi-year behavior revision leaving current_year unchanged. Parameters ---------- revision: dictionary of one or more YEAR:MODS pairs see Notes to Parameters _update method for info on MODS structure Raises ------ ValueError: if revision is not a dictionary. if each YEAR in revision is not an integer. if minimum YEAR in the YEAR:MODS pairs is less than start_year. if minimum YEAR in the YEAR:MODS pairs is less than current_year. if maximum YEAR in the YEAR:MODS pairs is greater than end_year. if _validate_assump_parameter_names_types generates errors. if _validate_assump_parameter_values generates errors. Returns ------- nothing: void Notes ----- Given a revision dictionary, typical usage of the Policy class is as follows:: behavior = Behavior() behavior.update_behavior(revision) In the above statements, the Behavior() call instantiates a behavior object (behavior) containing behavior baseline parameters, and the update_behavior(revision) call applies the (possibly multi-year) revision specified in revision and then sets the current_year to the value of current_year when update_behavior was called with parameters set for that pre-call year. An example of a multi-year, multi-parameter revision is as follows:: revision = { 2016: { '_BE_inc': [-0.3] }, 2017: { '_BE_sub': [0.2] } } Notice that each of the four YEAR:MODS pairs is specified as required by the private _update method, whose documentation provides several MODS dictionary examples. IMPORTANT NOTICE: when specifying a revision dictionary always group all revision provisions for a specified year into one YEAR:MODS pair. If you make the mistake of specifying two or more YEAR:MODS pairs with the same YEAR value, all but the last one will be overwritten, and therefore, not part of the revision. This is because Python expects unique (not multiple) dictionary keys. There is no way to catch this error, so be careful to specify revision dictionaries correctly. """ # check that all revisions dictionary keys are integers if not isinstance(revision, dict): raise ValueError('ERROR: revision is not a dictionary') if not revision: return # no revision to implement revision_years = sorted(list(revision.keys())) for year in revision_years: if not isinstance(year, int): msg = 'ERROR: {} KEY {}' details = 'KEY in revision is not an integer calendar year' raise ValueError(msg.format(year, details)) # check range of remaining revision_years first_revision_year = min(revision_years) if first_revision_year < self.start_year: msg = 'ERROR: {} YEAR revision provision in YEAR < start_year={}' raise ValueError(msg.format(first_revision_year, self.start_year)) if first_revision_year < self.current_year: msg = 'ERROR: {} YEAR revision provision in YEAR < current_year={}' raise ValueError( msg.format(first_revision_year, self.current_year) ) last_revision_year = max(revision_years) if last_revision_year > self.end_year: msg = 'ERROR: {} YEAR revision provision in YEAR > end_year={}' raise ValueError(msg.format(last_revision_year, self.end_year)) # validate revision parameter names and types self._validate_assump_parameter_names_types(revision) if self.parameter_errors: raise ValueError(self.parameter_errors) # implement the revision year by year precall_current_year = self.current_year revision_parameters = set() for year in revision_years: self.set_year(year) revision_parameters.update(revision[year].keys()) self._update({year: revision[year]}) self.set_year(precall_current_year) # validate revision parameter values self._validate_assump_parameter_values(revision_parameters) if self.parameter_errors: raise ValueError('\n' + self.parameter_errors)
[docs] def has_response(self): """ Returns true if any behavioral-response elasticity is non-zero for the current_year; returns false if all elasticities are zero. """ # pylint: disable=no-member all_zero = (self.BE_sub == 0.0 and self.BE_inc == 0.0 and self.BE_cg == 0.0) return not all_zero
[docs] def has_any_response(self): """ Returns true if any behavioral-response elasticity is non-zero in any year; returns false if all elasticities are zero in all years. """ for elast in self._vals: values = getattr(self, elast) for year in np.ndindex(values.shape): val = values[year] if val != 0.0: return True return False
[docs] @staticmethod def response(calc1, calc2, trace=False): """ Implements TaxBrain "Partial Equilibrium Simulation" dynamic analysis. Modify calc2 records to account for behavioral responses that arise from the policy reform that involves moving from calc1 policy to calc2 policy. Neither calc1 nor calc2 need to have had calc_all() executed before calling the Behavior.response(calc1, calc2) method. Returns new Calculator object --- a deepcopy of calc2 --- that incorporates behavioral responses to the reform. Note: the use here of a dollar-change income elasticity (rather than a proportional-change elasticity) is consistent with Feldstein and Feenberg, "The Taxation of Two Earner Families", NBER Working Paper No. 5155 (June 1995). A proportional-change elasticity was used by Gruber and Saez, "The elasticity of taxable income: evidence and implications", Journal of Public Economics 84:1-32 (2002) [see equation 2 on page 10]. Note: the nature of the capital-gains elasticity used here is similar to that used in Joint Committee on Taxation, "New Evidence on the Tax Elasticity of Capital Gains: A Joint Working Paper of the Staff of the Joint Committee on Taxation and the Congressional Budget Office", (JCX-56-12), June 2012. In particular, the elasticity use here is equivalent to the term inside the square brackets on the right-hand side of equation (4) on page 11 --- not the epsilon variable on the left-hand side of equation (4), which is equal to the elasticity used here times the weighted average marginal tax rate on long-term capital gains. So, the JCT-CBO estimate of -0.792 for the epsilon elasticity (see JCT-CBO, Table 5) translates into a much larger absolute value for the _BE_cg semi-elasticity used by Tax-Calculator. To calculate the elasticity from a semi-elasticity, we multiply by MTRs from TC and weight by shares of taxable gains. To avoid those with zero MTRs, we restrict this to the top 40% of tax units by AGI. Using this method, a semi-elasticity of -3.45 corresponds to a tax rate elasticity of -0.792. """ # pylint: disable=too-many-statements,too-many-locals,too-many-branches # nested function used only in response def trace_output(varname, variable, histbins, pweight, dweight): """ Print trace output for specified variable. """ print('*** TRACE for variable {}'.format(varname)) hist = np.histogram(variable, bins=histbins) print('*** Histogram:') print(hist[0]) print(hist[1]) if pweight.sum() != 0: out = '*** Person-weighted mean= {:.2f}' mean = (variable * pweight).sum() / pweight.sum() print(out.format(mean)) if dweight.sum() != 0: out = '*** Dollar-weighted mean= {:.2f}' mean = (variable * dweight).sum() / dweight.sum() print(out.format(mean)) # begin main logic of response assert calc1.array_len == calc2.array_len assert calc1.current_year == calc2.current_year mtr_cap = 0.99 # calculate sum of substitution and income effects if calc2.behavior('BE_sub') == 0.0 and calc2.behavior('BE_inc') == 0.0: zero_sub_and_inc = True else: zero_sub_and_inc = False # calculate marginal combined tax rates on taxpayer wages+salary # (e00200p is taxpayer's wages+salary) wage_mtr1, wage_mtr2 = Behavior._mtr12(calc1, calc2, mtr_of='e00200p', tax_type='combined') # calculate magnitude of substitution effect if calc2.behavior('BE_sub') == 0.0: sub = np.zeros(calc1.array_len) else: # proportional change in marginal net-of-tax rates on earnings mtr1 = np.where(wage_mtr1 > mtr_cap, mtr_cap, wage_mtr1) mtr2 = np.where(wage_mtr2 > mtr_cap, mtr_cap, wage_mtr2) pch = ((1. - mtr2) / (1. - mtr1)) - 1. # Note: c04800 is filing unit's taxable income sub = (calc2.behavior('BE_sub') * pch * calc1.array('c04800')) if trace: trace_output('wmtr1', wage_mtr1, [-9e99, 0.00, 0.25, 0.50, 0.60, 0.70, 0.80, 0.90, 0.999999, 1.1, 1.2, 1.3, 9e99], calc1.array('s006'), np.zeros(calc1.array_len)) print('high wage_mtr1:', wage_mtr1[wage_mtr1 > 0.999999]) print('wage_mtr2 them:', wage_mtr2[wage_mtr1 > 0.999999]) trace_output('pch', pch, [-9e99, -1.00, -0.50, -0.20, -0.10, -0.00001, 0.00001, 0.10, 0.20, 0.50, 1.00, 9e99], calc1.array('s006'), calc1.array('c04800')) trace_output('sub', sub, [-9e99, -1e3, -0.1, 0.1, 1e3, 1e4, 1e5, 1e6, 9e99], calc1.array('s006'), np.zeros(calc1.array_len)) # calculate magnitude of income effect if calc2.behavior('BE_inc') == 0.0: inc = np.zeros(calc1.array_len) else: # dollar change in after-tax income # Note: combined is f.unit's income+payroll tax liability dch = calc1.array('combined') - calc2.array('combined') inc = calc2.behavior('BE_inc') * dch # calculate sum of substitution and income effects si_chg = sub + inc # calculate long-term capital-gains effect if calc2.behavior('BE_cg') == 0.0: ltcg_chg = np.zeros(calc1.array_len) else: # calculate marginal tax rates on long-term capital gains # p23250 is filing units' long-term capital gains ltcg_mtr1, ltcg_mtr2 = Behavior._mtr12(calc1, calc2, mtr_of='p23250', tax_type='iitax') rch = ltcg_mtr2 - ltcg_mtr1 exp_term = np.exp(calc2.behavior('BE_cg') * rch) new_ltcg = calc1.array('p23250') * exp_term ltcg_chg = new_ltcg - calc1.array('p23250') # Add behavioral-response changes to income sources calc2_behv = copy.deepcopy(calc2) if not zero_sub_and_inc: calc2_behv = Behavior._update_ordinary_income(si_chg, calc2_behv) calc2_behv = Behavior._update_cap_gain_income(ltcg_chg, calc2_behv) # Recalculate post-reform taxes incorporating behavioral responses calc2_behv.calc_all() calc2_behv.records_include_behavioral_responses() return calc2_behv
# ----- begin private methods of Behavior class ----- @staticmethod def _update_ordinary_income(taxinc_change, calc): """ Implement total taxable income change induced by behavioral response. """ # compute AGI minus itemized deductions, agi_m_ided agi = calc.array('c00100') ided = np.where(calc.array('c04470') < calc.array('standard'), 0., calc.array('c04470')) agi_m_ided = agi - ided # assume behv response only for filing units with positive agi_m_ided pos = np.array(agi_m_ided > 0., dtype=bool) delta_income = np.where(pos, taxinc_change, 0.) # allocate delta_income into three parts winc = calc.array('e00200') delta_winc = np.zeros_like(agi) delta_winc[pos] = delta_income[pos] * winc[pos] / agi_m_ided[pos] oinc = agi - winc delta_oinc = np.zeros_like(agi) delta_oinc[pos] = delta_income[pos] * oinc[pos] / agi_m_ided[pos] delta_ided = np.zeros_like(agi) delta_ided[pos] = delta_income[pos] * ided[pos] / agi_m_ided[pos] # confirm that the three parts are consistent with delta_income assert np.allclose(delta_income, delta_winc + delta_oinc - delta_ided) # add the three parts to different records variables embedded in calc calc.incarray('e00200', delta_winc) calc.incarray('e00200p', delta_winc) calc.incarray('e00300', delta_oinc) calc.incarray('e19200', delta_ided) return calc @staticmethod def _update_cap_gain_income(cap_gain_change, calc): """ Implement capital gain change induced by behavioral responses. """ calc.incarray('p23250', cap_gain_change) return calc @staticmethod def _mtr12(calc1, calc2, mtr_of='e00200p', tax_type='combined'): """ Computes marginal tax rates for Calculator objects calc1 and calc2 for specified mtr_of income type and specified tax_type. """ _, iitax1, combined1 =, wrt_full_compensation=True) _, iitax2, combined2 =, wrt_full_compensation=True) if tax_type == 'combined': return (combined1, combined2) elif tax_type == 'iitax': return (iitax1, iitax2) else: raise ValueError('tax_type must be "combined" or "iitax"')