Prerequisites Almost essential: Design Contract DESIGN: TAXATION MICROECONOMICS Principles and Analysis Frank Cowell July 2015 Frank Cowell: Design-Taxation 1 The design problem The government needs to raise revenue • and it may want to redistribute resources To do this it uses the tax system • personal income tax • and income-based subsidies Base it on “ability to pay” • income rather than wealth • ability reflected in productivity Tax authority may have limited information • who have the high ability to pay? • what impact on individuals’ willingness to produce output? What’s the right way to construct the tax schedule? July 2015 Frank Cowell: Design-Taxation 2 A link with contract theory Base approach on the analysis of contracts • close analogy with case of hidden characteristics • owner hires manager • but manager’s ability is unknown at time of hiring Ability here plays the role of unobservable type • ability may not be directly observable • but distribution of ability in the population is known A progressive treatment: • outline model components • use analogy with contracts to solve two-type case • proceed to large (finite) number of types • then extend to general continuous distribution July 2015 Frank Cowell: Design-Taxation 3 Overview Design: Taxation Design basics Preferences, incomes, ability and the government Simple model Generalisations Interpretations July 2015 Frank Cowell: Design-Taxation 4 Model elements A two-commodity model • leisure (i.e. the opposite of effort) • consumption – a basket of all other goods Income comes only from work • individuals are paid according to their marginal product • workers differ according to their ability Individuals derive utility from: • their leisure • their disposable income (consumption) • Government / tax agency • has to raise a fixed amount of revenue K • seeks to maximise social welfare • where social welfare is a function of individual utilities July 2015 Frank Cowell: Design-Taxation 5 Modelling preferences Individual’s preferences • u = y(z) + y • u : utility level • z : effort • y : income received • y() : decreasing, strictly concave, function Special shape of utility function • quasi-linear form • zero-income effect • y(z) gives the disutility of effort in monetary units Individual does not have to work • reservation utility level u • requires y(z) + y ≥ u July 2015 Frank Cowell: Design-Taxation 6 Ability and income Individuals work (give up leisure) to provide consumption Individuals differ in talent (ability) t • higher ability people produce more and may thus earn more • individual of type t works an amount z • produces output q = tz • but individual does not necessarily get to keep this output? Disposable income determined by tax authority • intervention via taxes and transfers • fixes a relationship between individual’s output and income • (net) income tax on type t is implicitly given by q − y Preferences can be expressed in terms of q and y • for type t utility is given by y(z) + y • equivalently: y(q / t) + y July 2015 Frank Cowell: Design-Taxation A closer look at utility 7 The utility function (1) Preferences over leisure and income Indifference curves Reservation utility y u = y(z) + y yz(z) < 0 u≥u u 1– z July 2015 Frank Cowell: Design-Taxation 8 The utility function (2) Preferences over leisure and output Indifference curves Reservation utility y u = y(q/t) + y yz(q/t) < 0 u≥u u q July 2015 Frank Cowell: Design-Taxation 9 Indifference curves: pattern All types have the same preferences Function y() is common knowledge • utility level u of type t depends on effort z and payment y • but value of t may be information that is private to individual Take indifference curves in (q, y) space • u = y(q/t) + y • slope of given type’s indifference curve depends on value of t • indifference curves of different types cross once only July 2015 Frank Cowell: Design-Taxation 10 The single-crossing condition Preferences over leisure and output y High talent Low talent Those with different talent (ability) will have different sloped indifference curves in this diagram type b qa = taza qb = tbzb type a q July 2015 Frank Cowell: Design-Taxation 11 Similarity with contract model The position of the Agent • not a single Agent with known ex-ante probability distribution of talents • but a population of workers with known distribution of abilities The position of the Principal (designer) • designer is the government acting as Principal • knows distribution of ability (common knowledge) • the objective function is a standard SWF One extra constraint • the community has to raise a fixed amount K ≥ 0 • the government imposes a tax • drives a wedge between market income generated by worker and the amount available to spend on other goods July 2015 Frank Cowell: Design-Taxation 12 Overview Design: Taxation Design basics Analogy with contract theory Simple model Generalisations Interpretations July 2015 Frank Cowell: Design-Taxation 13 A full-information solution? Consider argument based on the analysis of contracts Given full information owner can fully exploit any manager • pays the minimum amount necessary • “chooses” their effort Same basic story here • can impose lump-sum tax • “chooses” agents’ effort — no distortion But the full-information solution may be unattractive • informational requirements are demanding • perhaps violation of individuals’ privacy? • so look at second-best case July 2015 Frank Cowell: Design-Taxation 14 Two types Start with the case closest to optimal contract model Exactly two skill types • ta > tb • proportion of a-types is p • values of ta , tb and p are common knowledge From contract design we can write down the outcome • essentially all we need to do is rework notation But let us examine the model in detail: July 2015 Frank Cowell: Design-Taxation 15 Second-best: two types The government’s budget constraint • p[qa - ya] + [1-p][qb - yb ] ≥ K • where qh - yh is the amount raised in tax from agent h Participation constraint for the b type: • yb + y(zb) ≥ ub • have to offer at least as much as available elsewhere Incentive-compatibility constraint for the a type: • ya + y(qa/ta) ≥ yb + y(qb/ta) • must be no worse off than if it behaved like a b-type • implies (qb, yb) < (qa, ya) The government seeks to maximise standard SWF • p z(y(za) + ya) + [1-p] z(y(zb) + yb) • where z is increasing and concave July 2015 Frank Cowell: Design-Taxation 16 Two types: model We can use a standard Lagrangian approach • government chooses (q, y) pairs for each type • subject to three constraints Constraints are: • government budget constraint • participation constraint (for b-types) • incentive-compatibility constraint (for a-types) Choose qa, qb, ya, yb to max p z(y(qa/ta) + ya) + [1-p] z(y(qb/tb) + yb) + k [p[qa - ya] + [1-p][qb - yb ] - K] + l [yb + y(qb/tb) - ub] + m [ya + y(qa/ta) - yb - y(qb/ta)] where k, l, m are Lagrange multipliers for the constraints July 2015 Frank Cowell: Design-Taxation 17 Two types: method Differentiate with respect to qa, qb, ya, yb to get FOCs: • pzu(ua)yz(za)/ta + kp + myz(za)/ta ≤ 0 • [1-p]zu(ub)yz(zb)/tb + k [1-p] + lyz(zb)/tb - myz(qb/ta)/ta ≤ 0 • pzu(ua) - kp + m ≤ 0 • [1-p]zu(ub) - k[1-p] + l - m ≤ 0 For an interior solution, where qa, qb, ya, yb are all positive • pzu(ua)yz(za)/ta + kp + myz(za)/ta = 0 • [1-p]zu(ub)yz(zb)/tb + k [1-p] + lyz(zb)/tb - myz(qb/ta)/ta = 0 • pzu(ua) - kp + m = 0 • [1-p]zu(ub) - k[1-p] + l - m = 0 Manipulating these gives the main results • for example, from first and third condition: • [kp - m ] yz(za)/ta + kp + myz(za)/ta = 0 • kp yz(za)/ta + kp = 0 July 2015 Frank Cowell: Design-Taxation 18 Two types: solution Solving the FOC we get: • - yz(qa/ta) = ta • - yz(qb/tb) = tb + kp/[1-p], • where k := yz(qb/tb) - [tb/ta] yz(qb/ta) < 0 Also, all the Lagrange multipliers are positive • so the associated constraints are binding • follows from standard adverse selection model Results are as for optimum-contracts model: • MRSa = MRTa • MRSb < MRTb Interpretation • no distortion at the top (for type ta) • no surplus at the bottom (for type tb) • determine the “menu” of (q,y)-choices offered by tax agency July 2015 Frank Cowell: Design-Taxation 19 Two ability types: tax design a-type’s reservation utility y b-type’s reservation utility b-type’s (q,y) incentive-compatibility constraint a-type’s (q,y) menu of (q,y) offered by tax authority ya Analysis determines (q,y) combinations at two points yb If a tax schedule T(∙) is to be designed where y = q −T(q) q qb July 2015 then it must be consistent with these two points qa Frank Cowell: Design-Taxation 20 Overview Design: Taxation Design basics Moving beyond the two-ability model Simple model Generalisations Interpretations July 2015 Frank Cowell: Design-Taxation 21 A small generalisation With three types problem becomes a bit more interesting • similar structure to previous case • ta > tb > tc • proportions of each type in the population are pa, pb, pc We now have one more constraint to worry about 1. participation constraint for c type: yc + y(qc/tc) ≥ uc 2. IC constraint for b type: yb + y(qb/tb) ≥ yc + y(qc/tb) 3. IC constraint for a type: ya + y(qa/ta) ≥ yb + y(qb/ta) But this is enough to complete the model specification • the two IC constraints also imply ya + y(qa/ta) ≥ yc + y(qc/tb) • so no-one has incentive to misrepresent as lower ability July 2015 Frank Cowell: Design-Taxation 22 Three types Methodology is same as two-ability model • set up Lagrangian • Lagrange multipliers for budget constraint, participation constraint and two IC constraints • maximise with respect to (qa,ya), (qb,yb), (qc,yc) Outcome essentially as before : • MRSa = MRTa • MRSb < MRTb • MRSc < MRTc Again, no distortion at the top and the participation constraint binding at the bottom • determines (q,y)-combinations at exactly three points • tax schedule must be consistent with these points A stepping stone to a much more interesting model July 2015 Frank Cowell: Design-Taxation 23 A richer model: N + 1 types The multi-type case follows immediately from three types Take N + l types • t0 < t1 < t2 < … < tN • (note the required change in notation) • proportion of type j is pj • this distribution is common knowledge Budget constraint and SWF are now • Sj pj [qj - yj] ≥ K • Sj pj z(y(zj) + yj) • where sum is from 0 to N July 2015 Frank Cowell: Design-Taxation 24 N + 1 types: behavioural constraints Participation constraint • is relevant for lowest type j = 0 • form is as before: • y0 + y(z0) ≥ u0 Incentive-compatibility constraint • applies where j > 0 • j must be no worse off than if it behaved like the type below (j-1) • yj + y(qj/tj) ≥ yj-1 + y(qj-1 /tj) • implies (qj-1, yj-1) < (qj, yj) • and u(tj) ≥ u(tj-1) From previous cases we know the methodology • (and can probably guess the outcome) July 2015 Frank Cowell: Design-Taxation 25 N+1 types: solution Lagrangian is only slightly modified from before Choose {(qj, yj )} to max Sj=0 pj z (y(qj / tj) + yj) + k [Sj pj [qj - yj] - K] + l [y0 + y(z0) - u0] + Sj=1 mj [yj + y(qj/tj) - yj-1 - y(qj-1 /tj)] where there are now N incentive-compatibility Lagrange multipliers And we get the result, as before • MRSN = MRTN • MRSN−1 < MRTN−1 • … • MRS1 < MRT1 • MRS0 < MRT0 • Now the tax schedule is determined at N+1 points July 2015 Frank Cowell: Design-Taxation 26 A continuum of types One more step is required in generalisation Suppose the tax agency is faced with a continuum of taxpayers • frequently used assumption • allows for general specification of ability distribution This case can be reasoned from the case with N + 1 types • allow N From previous cases we know • form of the participation constraint • form that IC constraint must take • an outline of the outcome Can proceed by analogy with previous analysis July 2015 Frank Cowell: Design-Taxation 27 The continuum model Continuous ability • bounded support [t,`t ] • density f(t) Utility for talent t as before • u(t) = y(t) + y( q(t) / t) Participation constraint is • u(t) ≥ u Incentive compatibility requires • du(t) /dt ≥ 0 SWF is t z (u(t)) f(t) dt t July 2015 Frank Cowell: Design-Taxation 28 Continuum model: optimisation Lagrangian is `t t z (u(t)) f(t) dt `t +k q(t) − y(t) − K] f(t) dt t + l [ u(t) − u] `t + t m(t) [du(t) / dt ] f(t) dt where u(t) = y(t) + y( q(t) / t) Lagrange multipliers are • k : government budget constraint • l : participation constraint • m(t) : incentive-compatibility for type t Maximise Lagrangian with respect to q(t) and y(t) for all t [t,`t ] July 2015 Frank Cowell: Design-Taxation 29 Output and disposable income under the optimal tax y t_ Lowest type’s indifference curve Lowest type’s output and income Intermediate type’s indifference curve, output and income _ t Highest type’s indifference curve 45° Highest type’s output and income Menu offered by tax authority q _ q_ July 2015 q Frank Cowell: Design-Taxation 30 Continuum model: results Incentive compatibility implies • dy /dq > 0 • optimal marginal tax rate < 100% No distortion at top implies • dy /dq = 1 • zero optimal marginal tax rate! But explicit form for the optimal income tax requires • specification of distribution f(∙) • specification of individual preferences y(∙) • specification of social preferences z (∙) • specification of required revenue K July 2015 Frank Cowell: Design-Taxation 31 Overview Design: Taxation Design basics Applying design rules to practical policy Simple model Generalisations Interpretations July 2015 Frank Cowell: Design-Taxation 32 Application of design principles The second-best method provides some pointers • but is not a prescriptive formula • model is necessarily over-simplified • exact second-best formula might be administratively complex Simple schemes may be worth considering • roughly correspond to actual practice • illustrate good/bad design Consider affine (linear) tax system • benefit B payable to all (guaranteed minimum income) • all gross income (output) taxable at the same marginal rate t • constant marginal retention rate: dy /dq = 1 - t Effectively a negative income tax scheme: • (net) income related to output thus: y = B + [1 - t] q • so y > q if q < B / t and vice versa July 2015 Frank Cowell: Design-Taxation 33 A simple tax-benefit system Guaranteed minimum income B y Constant marginal retention rate Implied attainable set Low-income type’s indiff curve Low-income type’s output, income 1-t High-income type’s indiff curve Highest type’s output and income “Linear” income tax system ensures that incentive-compatibility constraint is satisfied B q July 2015 Frank Cowell: Design-Taxation 34 Violations of design principles? Sometimes the IC condition be violated in actual design This can happen by accident: • interaction between income support and income tax • generated by the desire to “target” support more effectively • a well-meant inefficiency? Commonly known as • the “notch problem” (US) • the “poverty trap” (UK) Simple example • suppose some of the benefit is intended for lowest types only • an amount B0 is withdrawn after a given output level • relationship between y and q no longer continuous and monotonic July 2015 Frank Cowell: Design-Taxation 35 A badly designed tax-benefit system Menu offered to low income groups Withdrawal of benefit B0 Implied attainable set Low-income type’s indiff curve Low type’s output and income High-income type’s indiff curve y High type’s intended output and income ya High type’s utility-maximising choice yb The notch violates IC B0 causes a-types to masquerade as b-types q qb July 2015 qa Frank Cowell: Design-Taxation 36 Summary Optimal income tax is a standard second-best problem Elementary version a reworking of the contract model Can be extended to general ability distribution Provides simple rules of thumb for good design In practice these may be violated by well-meaning policies July 2015 Frank Cowell: Design-Taxation 37