T AXPAYERS ’ R ESPONSES TO P ENALTIES : E VIDENCE FROM A T AXPAYER E XPERIMENT IN N EW Z EALAND Norman Gemmell Chair in Public Finance, Victoria University of Wellington, New Zealand and International Fellow, Tax Administration Research Centre, University of Exeter
Outline 1. Background & relevant literature 2. Research questions & hypotheses 3. Modelling responses to actual/perceived penalties 4. The experiment: Design & Results 5. Compliance conclusions
Background Slemrod (JEP, 2007, p.38): “There has been no compelling empirical evidence addressing how noncompliance is affected by the penalty for detected evasion, as distinct from the probability that a given act of noncompliance will be subject to punishment .” Questions: Does ‘Allingham -Sandmo v s social norms’ debate extend to penalties? or: Is evidence on responses to penalties consistent with the AS model?
Background - Related Literature Slemrod et al. (1995) – US taxpayers early and late filing of IRS tax returns appears inconsistent with simple utility-max. Why? - Decision to complete tax return ‘today’ based on stochastic opportunity cost of filing. Lubell & Scholz (JAPS, 2001) – introduction of penalties leads to less cooperation (compliance?) among experiment participants: “ suggesting that the increased deterrence motivation did not compensate for the changes higher penalties bring about in how people frame their decisions ” (quoted in Slemrod, 2007, p.39). Hallsworth et al. (2014) – Distinguish liquidity-constrained and unconstrained taxpayers. Latter make rational time choice of when to pay a £1 of tax based on e.g. penalties, social norms, salience. – Based on utility-max framework but no penalty-specific evidence.
Questions General question: Do taxpayer responses to penalties and interest on unpaid tax conform to ‘crime & punishment’ model? Specific research questions: “How important is knowledge of the tax penalty regime for taxpayer compliance?” - does awareness of the existence of penalties matter? - does the size of penalties make a difference? - do compliance enforcement interventions affect actual payments or just intentions to pay? Examine via penalties for late payment of New Zealand’s Goods & Service Tax (GST - New Zealand’s VAT )
NZ Penalty Regime NZ Penalty Regime • Late filing and late payment penalties at similar rates across taxes (PAYE, GST, FBT etc) • Use-of-money interest (UOMI): 8.4% p.a.: r = 0.084 GST Late Payment Penalties: 5% initial one- off ‘fixed’ penalty: f = 0.05 1. 2. 1% per month ‘incremental’ penalty: f = 0.127 (annual) • Incremental penalty ceases ( f = 0 ) when in instalment arrangement Effective penalty = (1 + f + f)(1 + r) In Instalments: Effective penalty = (1 + f )(1 + r) Effective marginal penalty = (1 + f)(1 + r) + f r In Instalments: Effective marginal penalty = (1 + r) + f r • Annual effective penalty rate: 27.6% in 2015 [max. 34.4% in 2007] or 13.8% when in instalments. (Marginal: 22.6% V s 8.8%)
Modelling Taxpayers’ Choices - 1 • Model tax payment (not declared tax ) compliance: taxpayer, j , and tax authority agree on outstanding tax liability • Two period model with individual decision: pay tax liability now versus delay till next period Penalty and interest regime: f , f , and r • Each taxpayer has subjective discount rate, r j • • Prefer payment option that maximises NPV of expected after-tax income over 2 periods (pre-tax incomes given) = minimise NPV of tax liability. where = (1 + r j )] [Consistent with C-D utility fn. 3 rd payment option - instalment arrangement. Sets f = 0 if fraction, • a j , of tax liability paid now, (1 – a j ) paid in period 2.
Modelling Taxpayers’ Choices - 2 • Delaying payment to period 2 involves: Probability of debt written-off or otherwise reduced o perceived probability p j 1, that debt is fully repaid Marginal cost of non-compliance , c j > 0 , of avoiding payment o till period 2. [can include non- pecuniary and ‘social norm’ costs e.g. social norms against delayed payment increase c j .] ’ = ( p j + c j ) Let p j • • NPV of tax liability (per $ of initial GST debt) for: Pay now (P) Delay (N) Instalments (I)
Modelling Taxpayers’ Choices - 3 Solve for values of p j ’ , r j , and a j , where taxpayer is indifferent (where N = P, N = I, P = I) ∗ and 𝜌′ 𝑘 ∗ : Indifferent between all 3 options at 𝜍 𝑘 ∗ = 1+𝜚 𝜌′ ∗ = 0.892 𝜌′ 𝑘 1+𝜚+𝑔 ∗ = 1 + 𝜚 𝜍 ∗ = 0.138 𝜍 𝑘 1 + 𝑠 − 1 ≥ 0 Note : values independent of a j
Loci of equal NPVs Probability, p 1 N = P P preferred 0.95 0.9 "(1+ f ) " p * /"( 1+ f +f) " = N preferred 0.85 0.8 N = P locus 0.75 r * 0 0.1 0.2 0.3 0.4 discount rate, r = (1+ f )(1+r) - 1
Loci of equal NPVs Probability, p N = P N = I 1 I 0.95 0.9 N "(1+ f ) " p * /"( 1+ f +f) " = 0.85 P N 0.8 N = I locus N = P locus 0.75 r * 0 0.1 0.2 0.3 0.4 discount rate, r = (1+ f )(1+r) - 1
Loci of equal NPVs Probability, p P = I N = P N = I 1 I I 0.95 N 0.9 "(1+ f ) " p * /"( 1+ f +f) C * " = Expect fully informed indebted 0.85 taxpayers here before instalment offer P N = I locus N N = P locus 0.8 I = P locus [ N = I locus (alpha = 0)] 0.75 r * 0 0.1 0.2 0.3 0.4 discount rate, r = (1+ f )(1+r) - 1 ⧪
Modelling Penalty Perceptions - 1 Question: How is analysis affected if taxpayers misperceive penalties? Hypothesise penalty perceptions for 3 experimental groups as: A : no penalties: f = f = 0 ; & no instalment reduction (reduced ‘time penalties’) B : unspecified penalty ( 0 f B f ) without instalments; f B = 0 with instalments C : specific f > 0 , f > 0 without instalments; f > 0 , f = 0 with instalments p ′ ∗ r * Groups: r A : 1 1 r B : 1 + 𝜚 B 1 + 𝜚 (1 + 𝑠) 1 + 𝜚 1 C : 1 + 𝜚 + 𝑔
Modelling Penalty Perceptions - 2 Instalments r* r
Modelling Penalty Perceptions - 2 Instalments r* r
4. The GST Experiment
Experiment Questions 1. Does being in penalty information/awareness group A, B or C make a difference to payment decision? ’ &/or r j affect What characteristics correlate with p j 2. propensity to enter instalments or ‘pay now’? 3. How well do actual repayment outcomes (6-7 months later) align with experiment responses?
The Sample Taxpayers : ~ 4,400 with GST debts for ‘60 -90 days ’. • • Randomly selected 3 x 333 taxpayers (groups A, B, C) for phone contact with 3 alternative ‘scripts’; in August 2014 . • Script : different penalty information/reminder given before offer of instalment arrangement with future penalties turned off. • Remaining 3,400 taxpayers: ‘business -as- usual’ (BAU) ‘comparator’ group. • BAU : mixture of IR-initiated, debtor-initiated, & no contact. • A,B,C groups : exclude some debtors, e.g. if debt > $1m.
The Script IR initiate phone call. First seek payment in full. If no success, then: A: “Would you like to enter into an instalment arrangement ”. No mention of penalties B: “Did you know that you are being charged penalties on your debt to IR? If you enter into an instalment arrangement, we’ll stop your penalties. ” C: “ Did you know that you are being charged penalties on your debt to IR? If you enter into an instalment arrangement, we’ll stop penalties of 1% per month . ”
The Responses
Testing Influences on Payment Choices Test group membership and risk factors in multinomial logit model • Multinomial logit : analyse discrete experiment response choices o compare marginal effects or ‘relative risk ratios’ (RRRs) o is RRR > or < 1 ? o RRRs relative to default of ‘no contact’ taxpayers o Test impact of variables likely to affect payment choice via p j ’ & r j • We are interested in: 𝑒 Pr (I) 𝑒 p j ′ 𝑒 Pr (I) 𝑒 Pr (I) 𝑒𝜍 j [Similarly for d 𝑄𝑠(P) /d X ] = 𝑒𝒀 + 𝑒 p j ′ 𝑒𝜍 j 𝑒𝒀 𝑒𝒀 where X is a vector of exogenous taxpayer characteristics ( X includes groups A, B, C).
Payment Choice Probabilities Pr (I) , Pr (P) Correl. With Correl. Partial derivative p j With r j ' ' Expected sign + Groups: A + + + B +[ r * ] + + C d Pr (I)/d p j ′ d Pr (P)/d p j ′ d Pr (I)/d r j d Pr (P)/d r j ≷ 0 ≥ 0 ≥ 0 < 0 Expected sign Pr (I) = probability of Instalment choice ⧭ Pr (P) = probability of Pay Now choice
Experiment Results Does contact &/or offer of penalty cessation make a difference to payment choices? (Based on multinomial logit model) Probability of selecting choice (logit model) Red bar = statistically significant difference from BAU
Experiment Results Does contact &/or offer of penalty cessation make a difference to payment choices? (Based on multinomial logit model) Marginal effect Red bar = statistically significant difference from BAU
Experiment Results Does contact &/or offer of penalty cessation make a difference to payment choices? (Based on multinomial logit model) Red bar = statistically significant difference from BAU
Experiment Results Does contact &/or offer of penalty cessation make a difference to payment choices? (Based on multinomial logit model) Red bar = statistically significant difference from BAU
Recommend
More recommend
