• Nenhum resultado encontrado

Public debt sustainability and endogenous seigniorage in Brazil: time-series evidence from 1947-92: revised version

N/A
N/A
Protected

Academic year: 2017

Share "Public debt sustainability and endogenous seigniorage in Brazil: time-series evidence from 1947-92: revised version"

Copied!
30
0
0

Texto

(1)

❊♥s❛✐♦s ❊❝♦♥ô♠✐❝♦s

❊s❝♦❧❛ ❞❡

Pós✲●r❛❞✉❛çã♦

❡♠ ❊❝♦♥♦♠✐❛

❞❛ ❋✉♥❞❛çã♦

●❡t✉❧✐♦ ❱❛r❣❛s

◆◦ ✸✸✹ ■❙❙◆ ✵✶✵✹✲✽✾✶✵

P✉❜❧✐❝ ❉❡❜t ❙✉st❛✐♥❛❜✐❧✐t② ❛♥❞ ❊♥❞♦❣❡♥♦✉s

❙❡✐❣♥✐♦r❛❣❡ ✐♥ ❇r❛③✐❧✿ ❚✐♠❡✲❙❡r✐❡s ❊✈✐❞❡♥❝❡

❋r♦♠ ✶✾✹✼✲✾✷ ✭❘❡✈✐s❡❞ ❱❡rs✐♦♥✮

❏♦ã♦ ❱✐❝t♦r ■ss❧❡r✱ ▲✉✐③ ❘❡♥❛t♦ ❘❡❣✐s ❞❡ ❖❧✐✈❡✐r❛ ▲✐♠❛

(2)

❖s ❛rt✐❣♦s ♣✉❜❧✐❝❛❞♦s sã♦ ❞❡ ✐♥t❡✐r❛ r❡s♣♦♥s❛❜✐❧✐❞❛❞❡ ❞❡ s❡✉s ❛✉t♦r❡s✳ ❆s

♦♣✐♥✐õ❡s ♥❡❧❡s ❡♠✐t✐❞❛s ♥ã♦ ❡①♣r✐♠❡♠✱ ♥❡❝❡ss❛r✐❛♠❡♥t❡✱ ♦ ♣♦♥t♦ ❞❡ ✈✐st❛ ❞❛

❋✉♥❞❛çã♦ ●❡t✉❧✐♦ ❱❛r❣❛s✳

❊❙❈❖▲❆ ❉❊ PÓ❙✲●❘❆❉❯❆➬➹❖ ❊▼ ❊❈❖◆❖▼■❆ ❉✐r❡t♦r ●❡r❛❧✿ ❘❡♥❛t♦ ❋r❛❣❡❧❧✐ ❈❛r❞♦s♦

❉✐r❡t♦r ❞❡ ❊♥s✐♥♦✿ ▲✉✐s ❍❡♥r✐q✉❡ ❇❡rt♦❧✐♥♦ ❇r❛✐❞♦ ❉✐r❡t♦r ❞❡ P❡sq✉✐s❛✿ ❏♦ã♦ ❱✐❝t♦r ■ss❧❡r

❉✐r❡t♦r ❞❡ P✉❜❧✐❝❛çõ❡s ❈✐❡♥tí✜❝❛s✿ ❘✐❝❛r❞♦ ❞❡ ❖❧✐✈❡✐r❛ ❈❛✈❛❧❝❛♥t✐

❱✐❝t♦r ■ss❧❡r✱ ❏♦ã♦

P✉❜❧✐❝ ❉❡❜t ❙✉st❛✐♥❛❜✐❧✐t② ❛♥❞ ❊♥❞♦❣❡♥♦✉s ❙❡✐❣♥✐♦r❛❣❡ ✐♥ ❇r❛③✐❧✿ ❚✐♠❡✲❙❡r✐❡s ❊✈✐❞❡♥❝❡ ❋r♦♠ ✶✾✹✼✲✾✷ ✭❘❡✈✐s❡❞ ❱❡rs✐♦♥✮✴ ❏♦ã♦ ❱✐❝t♦r ■ss❧❡r✱ ▲✉✐③ ❘❡♥❛t♦ ❘❡❣✐s ❞❡ ❖❧✐✈❡✐r❛ ▲✐♠❛ ✕ ❘✐♦ ❞❡ ❏❛♥❡✐r♦ ✿ ❋●❱✱❊P●❊✱ ✷✵✶✵

✭❊♥s❛✐♦s ❊❝♦♥ô♠✐❝♦s❀ ✸✸✹✮

■♥❝❧✉✐ ❜✐❜❧✐♦❣r❛❢✐❛✳

(3)

Public Debt Sustainability and Endogenous

Seigniorage in Brazil: Time-Series Evidence from

1947-92

¤

Jo~

ao Vict or Issler and Luiz Renat o Lima

Graduat e School of Economics - EPGE

Get ulio Vargas Foundat ion

P. de Bot afogo 190 s. 1125-8

Rio de Janeiro, RJ 2253-900

Brazil

E-mail: jissler@fgv.br

First Version: June, 1997

T his Version: December, 1998

A bst r act

(4)

t he budget aft er shocks to eit her revenues or expendit ures? Third, are ex-pendit ures exogenous? T he result s show t hat (i) public de¯ cit is stat ionary (bounded asympt ot ic variance), wit h t he budget in Brazil being balanced almost ent irely t hrough changes in taxes, regardless of t he cause of t he ini-t ial imbalance. Expendiini-t ures are weakly exogenous, buini-t ini-tax revenues are not ; (ii) the behavior of a rational Brazilian consumer may be consist ent with Ricardian Equivalence; (iii) seigniorage revenues are crit ical t o restore intert emporal budget equilibrium, since, when we exclude t hem from t ot al revenues, debt is not sust ainable in econometric t est s.

1. I nt r oduct i on

At least since t he end of WWI I t he Brazilian economy has been plagued wit h chronic public de¯ cit s and relat ively high in° at ion - a typical example of

seigniorage-¯ nanced de¯ cit s. However, there have been very few episodes when a sharp increase in public debt was observed. Using nat ional account s dat a for t he revenue-GDP and t he expendit ure-GDP rat ios from 1947 t o 1992, t his pa-per st udies t hree cent ral issues in public ¯ nance. First , was t he pat h of public debt sust ainable during t his period? Second, if debt is sust ainable, how has t he government hist orically balanced t he budget aft er shocks t o eit her revenues or expendit ures were observed? For example, given an unpredict able increase in expendit ures, t here are two polar forms t o balance t he budget . One is t o in-crease t he present value of t axes and t he ot her is t o dein-crease t he present value of expendit ures. From t he point of view of a rat ional Brazilian t axpayer, it is import ant t o learn t o what ext ent t hese two forms of balancing t he budget have occurred. T hird, what are t he result s of exogeneity t est s for government expen-dit ures and t ax revenues? This last issue is mot ivat ed by t he fact t hat models of seigniorage-¯ nanced de¯ cit s such as Bruno and Fischer(1990), and t he ext ension in Ruge-Murcia(1995), assume respect ively t hat t he ¯ scal de¯ cit and t hat govern-ment expendit ures are exogenous. Moreover, in a more ° exible framework t han Barro(1974), Bohn(1992) shows t hat t he exogeneity of expendit ures is a necessary condit ion for Ricardian Equivalence. Given t hat post -war Brazil seems to ¯ t t he seigniorage-¯ nanced de¯ cit model, and t hat it is desirable t o discuss Ricardian Equivalence here, we ¯ nd it useful t o apply exogeneity t est s t o Brazilian ¯ scal ¯ gures. As far as we know, t his paper is t he ¯ rst work t o do it in t his cont ext .

(5)

are invest igated using unit -root t est s, coint egrat ion t est s, and calculat ing uncon-vent ional impulse-response funct ions based on Vect or Error-Correct ion Models (VECM's) where a balanced-budget rest rict ion is imposed. The search for a sen-sible t est of exogeneity has lead us t o use t he de¯ nit ions of weak, st rong and super exogeneity in Engle, Hendry and Richard(1983). T hey propose likelihood-based de¯ nit ions of exogeneity which could be t est ed, showing why t he previous con-cept s (pre-det erminedness and st rict exogeneity) were incomplet e or misleading. Exogeneity t est s were conduct ed using t he framework proposed in Johansen(1992, 1995).

T his paper has t hree main ¯ ndings. First , debt is sust ainable, wit h t he budget in Brazil being balanced almost ent irely t hrough changes in t axes, regardless of how t he init ial imbalance was generat ed. Expendit ures are weakly exogenous in t he sense of Engle, Hendry and Richard(1983). Second, t he behavior of a rational Brazilian consumer, wit h Ricardian preferences, is consist ent wit h t he Ricardian Equivalence result (Barro(1974)). Given, for example, a t ax break t oday, since his/ her best forecast for rest oring ¯ scal balance is t hat t he current imbalance will be fully reversed by fut ure t ax increases, consumpt ion and welfare will be unchanged1. Third, we show t hat seigniorage revenues are crit ical to rest ore int ert emporal budget equilibrium, since, when we exclude t hem from t ot al revenues, debt is not sust ainable in economet ric t est s. These result s mat ch t he convent ional wisdom about Brazilian public ¯ nance and are broadly consistent wit h t he t heoret ical model of opt imal seigniorage of Bruno and Fischer(1990), it s ext ension in Ruge-Murcia(1995), and wit h t he empirical ¯ ndings of Past ore(1995), Ruge-Murcia, and Rocha(1997).

T he paper is organized as follows: in Sect ion 2 t he methodology is present ed; in Sect ion 3 t he dat a set is discussed; in Sect ion 4 t he empirical result s are present ed and in Sect ion 5 we conclude.

2. M et hodol ogy

The economet ric t echniques used here t o t est whet her or not debt is sust ainable follow Hamilt on and Flavin(1986), and Bohn(1991). The calculat ions of t he \ un-convent ional impulse-response funct ion" follow Bohn(1991). We discuss a slight

(6)

caveat t o his approach. Exogeneity t est s are performed following t he typology in Engle, Hendry and Richard(1983); see also t he t est implement at ion in Jo-hansen(1992, 1995). A brief discussion of t hese t echniques is present ed here for t he sake of complet eness.

T he government budget const raint can be writ t en in t he following form:

Bt + 1 = Gt ¡ Tt + (1 + r )Bt + ²t+ 1; (2.1)

where Tt represent s ¯ scal revenues including seigniorage, Gt represent s ¯ scal

ex-pendit ures excluding debt -service payment s, r is t he \ real-int erest rat e"2on debt

(assumed ¯ xed), Bt is beginning of period public debt , and ²t + 1 is a st at ionary

measurement error inherit ed from assuming t hat rt = r for all t. All t ime series

are measured as a proport ion of GDP.

Wit hout loss of generality, we work with t he following version of equat ion (2.1):

Bt + 1 = G¤t ¡ Tt + Bt + ²t + 1; (2.2)

where G¤

t = Gt + r Bt is a broader version of government expendit ures, including

int erest paid on debt . Disregarding measurement error, and rearranging (2.2) we get :

Bt + 1¡ Bt = G¤t ¡ Tt = D eft; (2.3)

where D eft is t he public de¯ cit in period t. Equat ion (2.3) is t he basis for t he debt

sust ainability test of Hamilt on and Flavin(1986). It shows t hat whenever D eft is

not an int egrat ed series, Bt is di®erence-st at ionary. T hus, debt sust ainability is

t est ed viaa unit -root test on ¢ Bt.

A similar argument can be made from an int ert emporal perspect ive. The government int ert emporal budget const raint is:

Bt = 1 X

j = 0

½j Et h

Tt+ j ¡ G¤t + j ¡ ²t+ j + 1 i

; (2.4)

where ½ = (1+ r )1 is t he one-period discount rat e for fut ure t axes and expendi-t ures. Trehan and Walsh(1988) showed expendi-t haexpendi-t (2.4) holds whenever public debexpendi-t is

(7)

di®erence-stat ionary. From (2.3), t his last condit ion implies t hat G¤

t and Tt

coin-t egracoin-t e wicoin-t h coe± ciencoin-t (1; ¡ 1). T his is coin-t he coin-t escoin-t proposed in Bohn(1991) coin-t o check whet her or not debt is sust ainable. Under (rest rict ed) coint egrat ion, t he syst em on Xt = (G¤t; Tt)0, in error-correct ion form is3(Engle and Granger(1987)):

A(L )¢ Xt = ¡ ® ¯

0

Xt ¡ 1+ ¹t = ¡ ® D eft ¡ 1+ ¹t; (2.5)

where ¯ = (1; ¡ 1)0 is t he coint egrat ing vect or, ® is the adjust ment coe± cient vect or of t he error-correct ion t erm, and ¹t is a mult ivariat e whit e-noise process.

To simplify rat ional-expect at ion comput at ions of ¯ scal variables, we can rewrit e (2.5) as a ¯ rst-order syst em of equat ions as follows:

0 B B B B B B B @

¢ Xt

¢ Xt¡ 1

.. . ¢ Xt ¡ k + 1

D eft ¡ k 1 C C C C C C C A = 0 B B B B B B B @ A¤

1 A¤2 ¢¢¢ A¤k ¡ ®

I 0 ¢¢¢ 0 0

..

. ... ...

0 ¢¢¢ I 0 0

0 ¢¢¢ 0 ¯0 1

1 C C C C C C C A 0 B B B B B B B @

¢ Xt¡ 1

¢ Xt¡ 2

.. . ¢ Xt ¡ k

D eft ¡ k ¡ 1 1 C C C C C C C A + 0 B B B B B B B @ ¹t 0 .. . 0 0 1 C C C C C C C A , (2.6)

or, compact ly as:

Xt¤ = A¤Xt ¡ 1¤ + ¹¤t. (2.7)

where X¤

t = ³

¢ Xt0; ¢ Xt ¡ 10 ; ¢¢¢; ¢ Xt ¡ k+ 10 ; D eft ¡ k ´0

and ¹¤ t =

³

¹0t; 0; ¢¢¢; 0

´0 are nk + 1 vect ors, and A¤is t he [nk + 1] by [nk + 1] loading mat rix of X¤

t¡ 1.

Using (2.7) and (2.4), we can analyze t he e®ect s of unpredict able changes of G¤ t

and Tt on t heir respect ive present discount ed values. De¯ ne t he present discount ed

value of a generic variable z, P V (z)t = P

j ¸ 1

½j z

t + j, an innovat ion in z,zbt = zt ¡

Et ¡ 1zt, an innovat ion on t he present discount ed value of z,P V (z)d t = EtP V (z)t¡ 3Alt hough we are basically using X

t = (G¤t; Tt)

0

(8)

Et ¡ 1P V (z)t4, and t he inherit ed measurement error t erm, - t = Et[P V (²)t] ¡

Et ¡ 1[P V (²)t]. Consider Campbell's(1987) ident ity (1 ¡ ½) [zt + P V (z)t] = zt +

P V (¢ z)t, and t he fact t hat ¢Xct = Xct, t o obt ain:

¢cTt + P V (¢ T )d t = ¢Gc¤t + P V (¢ Gd ¤)t + r - t. (2.8)

From (2.8), for an unchanged debt value, and disregarding t he error t erm, any increase in expendit ures (not accompanied by an increase in t axes) would, in t he fut ure, eit her require a decrease in expenditures or an increase in t axes. In present -value t erms, t hese changes should o®set exact ly t he init ial change in expendit ures (a similar result applies for a change in t axes), i.e., ¢Gc¤

t = P V (¢ T)d t¡ P V (¢ Gd ¤)

t

must hold. Not ice that when we consider t he measurement error t erm, t his equa-t ion will noequa-t hold exacequa-t ly. The absoluequa-t e di®erence beequa-tween equa-t he lefequa-t - and equa-t he right -hand side is an increasing funct ion of r . Of course, it holds exact ly when r = 0 (i.e., ½= 1).

Calculat ing t he marginal impact of current innovat ions t o t axes and expendi-t ures on expendi-t he innovaexpendi-t ions of expendi-t he presenexpendi-t discounexpendi-t ed values can be easily done using (2.7). It is st raight forward t o show t hat :

d

P V (X¤)t = X

k ¸ 1

(½A¤)k¹¤t. (2.9)

T he part ial derivat ive of P V (Xd ¤)

t wit h respect t o ¹¤t is simply:

@P V (Xd ¤) t

@¹¤ t

= X

k ¸ 1

(½A¤)k = ½A¤(I ¡ ½A¤)¡ 1. (2.10)

Equat ion (2.10) is what we have labelled t he unconvent ional impulse-response funct ion. It depends only on ½and A¤. T he lat t er can be consist ent ly est imat ed

and t he former can be eit her est imat ed or calibrat ed. Equation (2.10) is \ uncon-vent ional" since it calculat es t he innovat ion present -value response t o shocks to t he syst em. It is t empt ing t o associat e t he l -t h element of t he vect or

hi n

½A¤(I ¡ ½A¤)¡ 1o

, where hi is a select ion vect or, wit h t he response t o a unit

impulse of t he l -t h element of ¹t. Alt hough Bohn(1991) claims t hat t his

impulse-response funct ion requires no ort hogonalizat ion of t he shocks, t he int erpretat ion

4T here is no cont radict ion between t he de¯ nit ion of bz

(9)

above embeds t he assumpt ion t hat no ot her element of ¹t changed when it s l -t h

element did. Of course, t his can only be t rue when t he covariance mat rix of t he shocks is diagonal.

T he exogeneity t est s conduct ed here follow t he typology in Engle, Hendry and Richard(1983); see Maddala(1992) for an int roduct ion, Ericsson and Irons(1994) for a more complet e review, and Ericsson and Irons and Johansen(1992, 1995) for t est ing procedures. There are t hree de¯ nit ions of exogeneity: weak, strong, and

super exogeneity. We only discuss t he ¯ rst two here5. T hey are relevant

respec-t ively respec-t o conducrespec-t condirespec-t ional inference and respec-t o perform condirespec-t ional forecasrespec-t ing. Engle, Hendry and Richard proposed de¯ nit ions of exogeneity t hat t ook into account t he paramet ers of int erest , somet hing missing from t he de¯ nit ions of predet erminedness and st rict exogeneity; see Engle, Hendry and Richard(p. 280) and Maddala(1992, p. 391). It does so by decomposing t he joint density of (yt; xt)0int o theproduct of t hecondit ional and marginal densit ies: f (yt jxt)¢f (xt).

Heurist ically, xt is said t o beweakly exogenousfor ¯ (a funct ion of t he paramet ers

of t he condit ional density) if f (xt) cont ains only nuisance paramet ers which are

irrelevant for conduct ing inference on ¯ . T hus, t o do inference on ¯ , we can \ discard" f (xt) and maximize t he likelihood using only t he t erms arising from

f (ytjxt). Weak exogeneity is a necessary condit ion t o have st rong and super

exogeneity. In addit ion, each of t hem require an ext ra condit ion. In part icular, st rong exogeneity requires t hat y does not Granger-cause x.

A key ingredient of t he weak exogeneity is t he \ separat ion" property for t he paramet ers associat ed wit h t he condit ional and marginal dist ribut ions. This is relevant for showing t hat innovat ions t o t he present discount ed value of x are not a®ect ed by shocks t o t he condit ional mean of y. Consider t he simple exam-ple in Ericsson(1994, pp. 22-24) where Xt = (yt; xt)

0

follows a Gaussian Vect or Aut oregression of order one, convenient ly reparamet erized as:

¢ yt = ¼11yt ¡ 1+ ¼12xt ¡ 1+ "1t

¢ xt = ¼21yt ¡ 1+ ¼22xt ¡ 1+ "2t;

where ("1t; "2t)0» I N (0; - ), \ » I N (¢)" reads independent and normally

dis-t ribudis-t ed random vecdis-t or, where - is a non-diagonal posidis-t ive-de¯ nidis-t e madis-t rix widis-t h

(10)

- = (! i j). If t here is one coint egrat ion vect or ¯0= (1; ¡ ±), where ± = ¡ ¼12=¼11 =

¡ ¼22=¼21, ¯0Xt ¡ 1 = (yt¡ 1¡ ±xt¡ 1), ®1 = ¼11, and ®2 = ¼21, t he error-correct ion

form is:

¢ yt = ®1(yt¡ 1¡ ±xt¡ 1) + "1t

¢ xt = ®2(yt¡ 1¡ ±xt¡ 1) + "2t: (2.11)

If we fact or t he joint density of the element s in Xt (given Xt ¡ 1) int o t he condit ional

density of yt given xt (and Xt¡ 1), and t he marginal density of xt (given Xt ¡ 1), we

get :

¢ yt = °1¢ xt + °2(yt ¡ 1¡ ±xt ¡ 1) + º1t

¢ xt = ®2(yt ¡ 1¡ ±xt ¡ 1) + "2t; (2.12)

where °1 = ! 12=! 22, °2 = ®1 ¡ (! 12=! 22) ®2, and º1t » I N (0; ¾2) and "2t »

I N (0; ! 22). First , not ice t hat t he two shocks are now independent , which did

not happen t o shocks in t he reduced form (2.11). Second, we can collect the pa-ramet ers of t he condit ional and marginal models in (°1; °2; ±; ¾2)

0

, and (®2; ±; ! 22)0

respect ively. Since ± is an element of bot h, and ®2 is in t he condit ional model

t hrough °2, t he paramet ers of t he condit ional and marginal models are not

sepa-rable in general.

However, if ®2= 0, xt is weakly exogenous for ¯ 0. In t his case, t he paramet ers

of t he condit ional and marginal models are respect ively (°1; °2; ±; ¾2) 0

and (!22);

t he ¯ rst equat ion is su± cient t o conduct inference on ± and t hus on ¯0= (1; ¡ ±);

t he second equat ion has only t he nuisance paramet er (! 22); and (2.12) simpli¯ es

t o:

¢ yt = °1¢ xt + °2(yt ¡ 1¡ ±xt ¡ 1) + º1t

¢ xt = "2t: (2.13)

Since "2t + i is independent of º1t, 8i , and P V (¢ x)d t is a funct ion of "2t + i,

8i > 1, it cannot be a®ect ed by changes in º1t. This makes weak exogeneity

relevant for inferring how t he government balances t he budget given a shock to eit her t ax revenues or expendit ures. For example, if we ¯ nd t hat expendit ures are weakly exogenous for t he coint egrat ion vect or, as in (2.13), P V (¢ Gd ¤)

t cannot

be a®ect ed by a shock t o t ax revenues (and vice-versa)6. Hence, aft er a shock to

t axes, int ert emporal equilibrium is rest ored via a change in P V (¢ T )d t alone.

(11)

Given t he discussion above, a t est for weak exogeneity was proposed by Jo-hansen(1992, 1995). It consist s of an exclusion test for ®2in (2.12). T his is exact ly

t he t est implement ed here.

3. T he D at abase

Sincewe apply unit -root and coint egration t ests t o invest igat e whet her public-debt is sust ainable, it is desirable t o work wit h dat a wit h a long span. Unfort unat ely, Brazil does not have long-span t ime-series dat a on debt . An alt ernat ive is t o use t he fact t hat , Bt + 1¡ Bt = G¤t ¡ Tt = D eft, relying on dat a on G¤t and Tt. FIBGE

provides annual nat ional-account dat a on expendit ures, which include payment s ofnominal int erest on public debt , and cannot be disaggregat ed furt her, and also annual dat a on revenues (excluding seigniorage). T hey cover t he period from 1947 t o 1992. Since G¤

t includes real-int erest expendit ures, not nominal, t here is

a mismat ch between t he dat a and t he t heoret ical framework.

Seigniorage, approximat ed by in° at ion t ax, is ext ract ed from Cysne(1995), and t hen added t o nat ional-account s t ax revenues t o form a series of t ot al t ax revenues7. T he former, and expendit ures, as described above, were divided by

GDP, and labelled Tt and G¤t respect ively.

According t o Ahmed and Rogers(1995), using nominal-int erest payment s in place of r Bt may bias t oward reject ion t he rest rict ed version of t he coint egrat ion

t est between G¤

t, and Tt. Indeed, we may get a coint egrat ing vect or di®erent from

(1; ¡ 1)0, possibly wit h t he coe± cient of G¤t being great er t han unity in absolut e value. This, however, is not t he only problem. Since t he nominal-interest rat e can be t hought of as t he sum of t he real-int erest rat e and t heex-anterat e of in° at ion, we may not get coint egrat ion at all if in° at ion is an int egrat ed process, which is a real possibility for Brazil. In order t o account for t hat problem wit h t he dat a, we increased t he signi¯ cance level of t he coint egrat ion t est from t he usual 5% or 10% levels (Johansen and Juselius(1990)), t o t he 20% level8.

7Seigniorage revenues, i.e., t he change in monet ary base used exclusively t o ¯ nance gov-ernment de¯ cit s, is not available for 1947-92, alt hough monet ary base is. Since high-powered money could change for reasons ot her t han t o ¯ nance government de¯ cit s, we chose not t o use t he ¯ rst -di®erence in monet ary base as a proxy for seigniorage.

8We only use t he 20% level when t est ing sequent ially for t he rank of coint egrat ion: H

(12)

4. Em pi r ical R esult s

4.1. U ni t -R oot and C oint egr at i on Test s

Before present ing t he result s of unit -root t est s it should be ment ioned t hat t he fact t hat our series are rat ios, t herefore lying in t he real int erval [0; 1], does not rule t hem out being int egrat ed processes; see Ahmed and Yoo(1989). Figure 1 shows t he dat a set used in t his paper. It is clear t hat all series are smoot h, indicat ing a high level of persist ence. The sample average (in percent age) of t he rat ios G¤ t

and Tt are very close, respect ively 24.8% and 24.4%. Despit e G¤t and Tt showing a

st eady increase since t he beginning of t he sample, G¤

t ¡ Tt shows mean reversion,

especially aft er t he expendit ure and t ax increases of t he mid-eight ies.

Table 1 shows t he result s of t he unit -root t est s performed: t he Augment ed Dickey-Fuller (ADF) and t he Phillips-Perron t est9. Regardless of t he t est type,

t he result s support one unit root for expendit ures and t axes, whereas t he de¯ cit (G¤¡ T ) is st at ionary even at 1%10.

T he next st ep is t o perform t he coint egrat ion t est between G¤

t and Tt. We used

t he likelihood-based coint egrat ion t est of Johansen(1991). It is well known t hat t he result s of coint egrat ion t est s using t his t echnique depend on t he det erminist ic component s included in t he VAR and on t he chosen lag lengt h. T herefore some pre-t est ing is done in order t o choose t hese. T he lag lengt h was select ed using two types of informat ion crit eria (Schwarz and Hannan-Quinn). To choose t he det erminist ic component s we used t he likelihood rat io t est discussed in Johansen. The pre-t est result s are present ed in Tables 2 and 3.

T he result s in Table 2a are based on a VAR est imat ed wit h an unrest rict ed const ant t erm and the result s in Table 2b on a VAR est imat ed wit hout a const ant . In bot h cases, using t he Schwarz crit erium, t he opt imal lag lengt h is two, but using t he Hannan-Quinn crit erium t he opt imal lag lengt h is t hree. Given t he Mont e-Carlo result s in Gonzalo(1994), we chose t o work wit h t hree lags in t he VAR11.

Test ing t he det erminist ic component s in t he VAR is more subt le, since it requires condit ioning on t he number of coint egrat ing relat ionships t o be performed; see

9T he result s of t he lat t er are especially relevant , since t he dat a clearly show signs of het eroskedast icity.

10Test s for two unit root s reject t his hypot hesis for all t hree series.

(13)

Johansen(1991). Table 3 shows t est result s when t he number of coint egrat ing vect ors is one. When t he model wit h an unrest rict ed const ant is t ested against t he one wit h a rest rict ed const ant t he t est st at ist ic is 1.64, dist ribut ed as a Â2

1. Thus,

we cannot reject t he rest rict ed model. The same result was obt ained when we t est ed t he rest rict ed const ant against t he no-const ant model, or t he unrest rict ed const ant versus t he no-const ant model. T hus, we used a VAR wit hout a const ant . T he result s of t he coint egrat ion t est are present ed in Table 4. As discussed above, we used t he 20% signi¯ cance level for t hecrit ical values of t he Trace and t he ¸max st at ist ics. At 20% we reject t he null of no coint egrat ing vect ors and cannot

reject t he null of one coint egrat ing vect or. Alt hough we used t he 20% level, it is wort h ment ioning t hat t he t est st at ist ics for t he null of zero coint egrat ing vect ors are very close t o t he crit ical value at 10%. The point est imat e of t he coint egrat ing vect or is (1; ¡ 0:94) (normalizat ion on expendit ures), which mat ches t he prior t hat using t he nominal-int erest payment s rat her t han t he real-int erest payment s would bias upwards t he absolut e value of t he expendit ures coe± cient . It is nat ural at t his point t o t est t he t heoret ical value of (1; ¡ 1) for t he coint egrat ing vect or. At usual signi¯ cance levels, t his t heoret ical vect or could not be reject ed. Hence, cointegrat ion t ests corroborat e our prior ¯ ndings from unit -root t est s t hat t he de¯ cit is st at ionary, and t hus debt is sust ainable.

4.2. Exogeneit y Test s

Table 5 shows t he result s of t he error-correct ion model est imat es. First , it seems t hat our choice of 3 lags for t he VAR was appropriat e. For t he system as a whole, ¢ G¤

t ¡ 2 is signi¯ cant , while ¢ Tt ¡ 2 is marginally signi¯ cant . Table 6 shows t he

re-sult s of t he weak exogeneity t est s for t he coint egrat ing vect or if t he coint egrat ing rank is one. At usual signi¯ cance levels we found t hat expendit ures are weakly exogenous for t he paramet ers of int erest in t he condit ional model of t ax-revenues, but t he converse is not t rue for revenues12. T he result s in Table 7 clearly

indi-cat e t hat expendit ures Granger-cause t axes and vice-versa at t he 5% signi¯ cance level. At 1%, however, t axes do not precede expendit ures. Therefore, alt hough expendit ures are weakly exogenous, t hey are not st rongly exogenous, since t hey are Granger-caused by t ax-revenues.

12T he fact t hat t he coe± cient of D ef

t ¡ 1on t he ¢ G¤t equat ion is not signi¯ cant led us t o t est t he joint hypot heses t hat t he coint egrat ing vect or and t he adjust ment coe± cient are respec-t ively (1; ¡ 1) and (0; ®). T he Â2

(14)

4.3. U nconvent ional I mpulse-R esp onse Funct ion

As discussed above, impulse-response result s are only int erpret able when t he variance-covariance mat rix of t he VECM errors is diagonal. T he correlat ion coef-¯ cient between t he two reduced-form residuals is 0.244. T he log-likelihood rat io st at istic t est ing for a diagonal covariance mat rix is 2.67, wit h a Â21 dist ribut ion.

The crit ical values are 3.84 and 2.71 at 5% and 10% respect ively. Thus, we do not reject the null of ort hogonal innovat ions, and no ort hogonalizat ion t echnique is used for reduced-form residuals.

Unconvent ional impulse-response funct ions were calculat ed using (1; ¡ 1) as t he coint egrat ing vect or and t he adjust ment fact or in t he form (0; ®)13. From t he VECM with two lags, insigni¯ cant regressors were delet ed based on t heir robust t-tests14. Based on t hese rest rict ed VECM est imates, and using (2.10), t he marginal

impacts of innovat ions t o G¤

t and Tt onP V (¢ Gd ¤)t andP V (¢ T)d t were calculat ed.

Table 8 shows t he impulse-response result s for t he following values of ½: ½= 0:97, ½ = 0:98, ½ = 0:99, and ½ = 1. To be able t o t est hypot heses on t he impulse-response paramet ers, we performed a Mont e-Carlo experiment wit h 1,000 replicat ions for each value of ½. The sample size used for each replicat ion is t he same as t hat of t he original dat a, and pre-sample observat ions were drawn to avoid dependence on t heir values. St andard errors of point -est imates were t hen comput ed and are included in Table 8.

Table 8 has two int erest ing result s. First , it is clear t hat expendit ures change very lit t le when we consider innovat ions in eit her expendit ures or revenues. For ½= 1, and t hus wit h no measurement error, 89% of expendit ure-generat ed de¯ cit s are eliminat edviaan increase in t axes, versus only 11% of decrease in fut ure ex-pendit ures. Not ice t hat t he former is not st atist ically di®erent from 100% and t he lat t er not st at ist ically di®erent from zero. Second, alt hough it is impossible t o infer consumer behavior from ¯ scal dat a, impulse-response result s are consis-t enconsis-t wiconsis-t h Ricardian-Equivalence behavior for Brazilian consumers (Barro(1974, 1989)) wit h t he appropriat e preferences. This is t rue since a change in t axes, not accompanied by a change in expendit ures, will be o®set 100% by a change in fut ure t axes. T hus, Ricardian consumers using VAR's t o infer t he meaning of a t ax break (hike) t oday will not change t heir opt imal consumpt ion allocat ion, since t hey expect t his t ax break (hike) t o be o®set by an equivalent increase (decrease)

13T he est imat e for ® is 0.25.

(15)

in fut ure t axes. T his result is a direct consequence of t he weak exogeneity of expendit ures for t he cointegrat ion vect or15.

T hese two basic result s are also obt ained for ½6= 1. For example, for ½= 0:99, point est imat es of @P V (¢ T )c t

@Gc¤ t

and @P V (¢ Gc ¤)t

@Gc¤ t

are 93% and -11% respectively, and

are st at ist ically equal t o 100% and zero respect ively. Moreover, @P V (¢ T )c t

@cG¤ t

is st ill st at istically equal t o -100%. For ot her values of ½6= 1, impulse-response est imat es are not very di®erent from t he result s above, alt hough not precisely est imat ed.

It is always appropriat e t o ask what is t he proper value of ½= (1+ r )1 t o use for t he sample period 1947-1992. Since ¯ scal variables are expressed as a proport ion of GDP, r should be int erpret ed (t o a logarit hmic approximat ion) as t he di®erence between t he real-int erest rat e on debt and t he growt h rat e of GDP. Due t o t he lack of long-span dat a on int erest rates, we can only conject ure here what is t he most appropriat e value of ½: during t his period, t he relat ively high in° at ion rat e, coupled wit h a relat ively high growt h rat e for GDP, would indicat e t hat r ' 0 is a reasonable value for r . Therefore, we propose using ½' 1.

Our last empirical t est checks whet her or not seigniorage was import ant for long-t erm ¯ scal balance. We perform a coint egrat ion t est wit h t he same revenue series as before wit h seigniorage revenues excluded, using, again, t he same signi¯ -cance level (20%). T he result s are present ed in Table 9. We ¯ nd no coint egrat ion in t his case. Not ice t hat t he Trace and t he ¸max st at ist ic are much smaller t han

t heir crit ical values at 20%.

T he evidence for Brazil suggest s t hat debt is sust ainable and t hat expendi-t ures are weakly exogenous. Thus, expendi-t he counexpendi-t ry followed much more closely a spend-and-t ax t han a t ax-and-spend policy. Indeed, our impulse-response result s suggest s t hat , given a shock t o expendit ures, t he present value of t axes will fully accommodat e t his imbalance rest oring long-run equilibrium. Given t hat seignior-age is crit ical for t he exist ence of equilibrium, it was probably t hrough seigniorseignior-age revenues t hat most expendit ure increases were ¯ nanced. In t his case Brazilian in-° at ion is simply a consequence of t he type of spend-and-t ax policy followed in t he post -war period.

It is int erest ing t o compare our result s t o t hose of ot hers who invest igat ed public-¯ nance issues in Brazil. Past ore(1995) t est s debt sust ainability using t he t echniques of Hamilt on and Flavin(1986) on a short -span ¯ scal dat a set . From

(16)

unit -root t est s, he ¯ nds t hat t he ¯ rst di®erence of public debt is st at ionary. Since t his series is equal t o D eft (see (2.3)), his result s and ours are ident ical.

Rocha(1997), also using a short -span dat a set , concludes t oo t hat debt is sust ain-able. Since all t hese result s are achieved using di®erent t echniques, di®erent series, and di®erent samples, t hey complement each ot her in con¯ rming t he convent ional wisdom about Brazilian public ¯ nance.

As far as we know, one of t he original cont ribut ions of t his paper is t o t est for exogeneity of ¯ scal dat a, showing t hat expendit ures are exogenous for a high-in° at ion count ry such as Brazil. This assumpt ion is implicit in t he seminal t heo-ret ical paper of Bruno and Fischer(1990) on opt imal seigniorage for high-in° at ion count ries16. T here, cent ral banks choose opt imal seigniorage condit ional on a

given de¯ cit . Of course, t his condit ional opt imizat ion relies on t he exogeneity of expendit ures, which was validat ed here for Brazil under proper economet ric t est s. When we consider t he fact t hat debt was only sust ainable when seigniorage was included as a government revenue, it becomes clear t hat t he fundament al charact er of Brazilian public ¯ nance was t hat of endogenous seigniorage used to accommodat e expendit ure increases. Indeed, t his conclusion is a common feat ure of t he work of Past ore(1995), Ruge-Murcia(1995), Rocha(1997), and ours.

5. Concl usi ons

This paper present s t est s on t he sust ainability of Brazilian public debt for t he post -war period - 1947-92. They show t hat debt is sust ainable only if seigniorage is included as a government revenue. In t his case, exogeneity t est s suggest t hat expendit ures are exogenous. Unconvent ional impulse-response result s show t hat , regardless of how an init ial ¯ scal imbalance is originat ed (shocks t o expendit ures or revenues), it is eliminat ed t hrough a change in t axes. This last result is con-sist ent wit h Ricardian-Equivalence behavior for Brazilian consumers wit h proper preferences. Based on t he evidence, we ¯ nd t hat Brazil has followed closely a spend-and-t ax policy in t he post -war era, wit h seigniorage having a crucial role in balancing t he budget . As a consequence, Brazilian in° at ion was consist ent ly high during t his period.

A brief re° ect ion on t he st at us of t he Real Plan is appropriat e. Since t he beginning of t he plan in July 1994, seigniorage revenues have decreased sharply.

(17)
(18)

R efer ences

[1] AHMED, S. and B.S. Yoo, 1989. \ Fiscal Trends and Real Business Cycles" , Working paper: Pennsylvania St at e University, University Park, PA.

[2] AHMED, S. and J.H. Rogers. \ Government Budget De¯ cit s and Trade De¯ cit s: Are Present Value Const raint s Sat is¯ ed in Long-t erm Dat a?" Jour-nal of Monetar y Economics, 36, pp. 351-374, 1995.

[3] BARRO, Robert J., \ Are Government Bonds Net Wealt h?" Journal of Po-litical Economy, Vol. 82, pp. 1095-1117, 1974.

[4] Barro, Robert J., \ The Ricardian Approach t o Budget De¯ cit s," Journal of Economic Perspectives, v3(2), 37-54, 1989.

[5] BOHN, Henning, \ Budget Balance Through Revenue or Spending Adjust -ment s? Some Hist orical Evidence for t he Unit ed St at es." Journal of Mone-tary Economics, 27, pp. 333-359, 1991.

[6] BOHN, Henning, \ Endogenous Government Spending and Ricardian Equiv-alence," Economic Jour nal, vol. 102, pp. 588-597, 1992.

[7] Bruno, M., and Fischer, S., 1990, \ Seigniorage, Operat ing Rules, and t he High In° at ion Trap," Quarter ly Jour nal of Economics, vol. 105(2), pp. 353-374.

[8] Campbell, J., \ Does Saving Ant icipat e Declining Labor Income? An Alt er-nat ive Test of t he Permanent Income Hypot hesis," Econometr ica, vol. 55(6), pp. 1249-73, 1987.

[9] CYSNE, Rubens Penha. \ O Sist ema O¯ cial e a Queda das Transferencias In° acion¶arias." Working paper: Graduat e School of Economics - EPGE, Get ulio Vargas Foundat ion, 1995.

[10] ENGLE, R.F., and Granger, C.W.J., 1987, \ Coint egrat ion and Error Cor-rect ion: Represent at ion, Est imat ion and Test ing," Econometr ica, vol. 55, pp.251-276.

(19)

[12] Ericsson, N.R., 1994, \ Test ing Exogeneity: An Int roduct ion," in Ericsson, N.R. and Irons, J.S., Edit ors,Testing Exogeneity. Oxford: Oxford University Press.

[13] ERICSSON, N.R. and IRONS, J.S., 1994, \Testing Exogeneity" , Oxford: Ox-ford University Press.

[14] GONZALO, J., 1994, \ Five Alt ernat ive Met hods of Estimat ing Long Run Relat ionships," Jour nal of Econometrics, vol. 60, pp. 203-233.

[15] HAMILTON, J. and M. Flavin, \ On t he Limit at ions of Government Borrow-ing: A Framework for Empirical Test ing." Amer ican Economic Review, 76, 808-819, 1986.

[16] JOHANSEN, S., 1991, \ Est imat ion and Hypot hesis Test ing of Coint egrat ed Vect ors in Gaussian Vect or Aut oregressions" , Econometr ica, vol. 59-6, pp. 1551-1580.

[17] Johansen, S.(1992), \ Coint egrat ion in Part ial Syst ems and t he E± ciency of Single-equat ion Analysis," Jour nal of Econometr ics, vol. 52(3), pp. 389-402.

[18] JOHANSEN, S., 1995, \Likelihood-Based I nference in Cointegrated Vector Auto-regressive Models," Oxford: Oxford University Press.

[19] Johansen, S. e Juselius, K ., \ Maximum Likelihood Est imat ion and Inference on Coint egrat ion - wit h Applicat ions to t he Demand for Money," Oxford Bulletin of Economics and Statistics, vol. 52, pp. 169-210, 1990.

[20] Lucas, Robert E., 1976, \ Economet ric Policy Evaluation: a Crit ique,"

Car negie-Rochester Conference Ser ies on Public Policy, vol. 1, pp. 19-46.

[21] MADDALA, G.S., 1992, "I ntroduction to Econometrics," Englewood Cli®s: Prent ice-Hall, 2nd. Edit ion.

(20)

[23] Rocha, Fabiana, \ Long-Run Limit s on t he Brazilian Government Debt ," Re-vista Brasileira de Economia, vol. 51(4), pp. 447-470, 1997.

[24] Ruge-Murcia, F., 1995, \ Government Expendit ure and t he Dynamics of High In° at ion," Working Paper: Universit¶e de Mont r¶eal.

(21)

Figure 1

Expenditures (Including Nominal Interest Paid on Debt) and

Revenues (Including

Seigniorage

) as a Proportion of GDP

Gt* =Gt +rBt

and

Tt

- 0 . 1 0 - 0 . 0 5 0 . 0 0 0 . 0 5 0 . 1 0 0 . 1 5 0 . 2 0

0 . 1 0 . 2 0 . 3 0 . 4 0 . 5 0 . 6

5 0 5 5 6 0 6 5 7 0 7 5 8 0 8 5 9 0

(22)

Table 1

Unit-Root Tests

Series

Lags

1

ADF Test

Phillips-Perron

Test

Tax Revenues ( T )

3

-3.03

-3.37

Expenditures

(G )*

4

-2.14

-2.54

Deficit

(G - T)*

4

-2.98**

-3.27**

Notes: (1) The number of lags applies only to the ADF test. The final choice was made

based on the t-test of significance of the last lagged first-difference. (2) The lag truncation chosen for the Bartlett kernel was 3. For (G )*

(23)

Table 2

VAR Lag Truncation

(a)

VAR

Order

Constant

Linear

Trend

Schwarz

Criterium

Hannan-Quinn

Criterium

1

unrestricted

no trend

-14.21

-14.36

2

unrestricted

no trend

-14.25

-14.52

3

unrestricted

no trend

-14.24

-14.61

4

unrestricted

no trend

-13.98

-14.45

(b)

VAR

Order

Constant

Linear Trend

Schwarz

Criterium

Hannan-Quinn

Criterium

1

no constant

no trend

-14.21

-14.31

2

no constant

no trend

-14.39

-14.60

3

no constant

no trend

-14.36

-14.68

(24)

Table 3

Deterministic Components Test (VAR(3))

Model VAR

Order

Constant

Linear

Trend

Restricted Vs.

Unrestricted

Model

Test

Statistic

1

3

unrestricted

no trend

2 versus 1

1.64

2

3

restricted

no trend

3 versus 2

2.02

3

3

no constant

no trend

3 versus 1

3.66

Notes: Results conditioned on the existence of one cointegrating vector. The critical value

(25)

Table 4

Johansen’s Cointegration Test (

Seigniorage

Included as Revenues)

Test Statistic

(Critical Value at the 20% Level)

Cointegrating

Vector

λmax

Trace

(Expend., Taxes)

Κ =0 Κ ≤1 Κ =0 Κ ≤1

9.185*

1.638

10.82*

1.638

(7.58)

(1.82)

(8.45)

(1.82)

(1.0, -0.94)

Cointegration Restriction Test

Restriction

( )

1,−1 χ2

( )

1 =150.

;

p-value

: 22.14%

(26)

Table 5

Vector Error Correction Model Estimates

EQ. 1

(

∆ G*t

)

EQ. 2

(

T

t

)

Regressor

Est. Coeff.

t

-stat

t

-prob

Est. Coeff.

t

-stat

t

-prob

∆ G*t1

0.17964

(0.19)

0.987

0.3298

0.23177

(0.09)

2.04

0.0474

∆ G*t2

-0.73603

(0.22)

-3.558

0.0010

-0.1790

(0.12)

-1.39

0.1720

T

t-1

-0.30270

(0.36)

-1.163

0.2521

-0.3066

(0.14)

-1.89

0.0657

T

t-2

0.61572

(0.43)

2.459

0.0186

0.15161

(0.12)

0.97

0.3362

Def

t-1

0.07624

(0.21)

0.503

0.6179

0.25661

(0.07)

2.72

0.0097

(27)

Table 6

Adjustment-Coefficient Weak Exogeneity Test

Null Hypothesis

Test Statistic

P-Value

Tt

is weakly exogenous for the

parameter of interest of the

G*

t

conditional model

7.36

0.0067**

G*

t

is weakly exogenous for the

parameter of interest of the

Tt

conditional model

1.11

0.29

(28)

Table 7

Granger-Causality Test

VAR(3)

Null Hypothesis

P-Value

T

t

does not Granger-cause

G*

t

0.023*

G*

t

does not Granger-cause

T

t

0.0000**

Notes: The symbols (*) and (**) represent rejection of the null at the 5 and 1%

(29)

Table 8: Unconventional Impulse-Response Function

(

ρρ

=0.97)

Impulse

T

T

G*

G*

Response

PV$

( )

T PV$

(

G*

)

PV$

( )

T PV$

(

G*

)

Pt-Estimate

SE

-1.22

(2.13)

-0.07

(0.47)

1.12

(2.13)

0.05

(0.51)

(

ρρ

=0.98)

Impulse

T

T

G*

G*

Response

PV$

( )

T PV$

(

G*

)

PV$

( )

T PV$

(

G*

)

Pt-Estimate

SE

-1.13

(1.05)

-0.04

(0.39)

1.01

(0.99)

-0.09

(0.42)

(

ρρ

=0.99)

Impulse

T

T

G*

G*

Response

PV$

( )

T PV$

(

G*

)

PV$

( )

T PV$

(

G*

)

Pt-Estimate

SE

-1.06

(0.09)

-0.02

(0.03)

0.93

(0.20)

-0.11

(0.18)

(

ρρ

=1.00)

Impulse

T

T

G*

G*

Response

PV$

( )

T PV$

(

G*

)

( )

PV$ ∆T PV$

(

G*

)

Pt-Estimate

SE

-1.00

(-)

0

(-)

0.89

(0.15)

-0.11

(0.15)

Notes: Standard errors calculated using a Monte Carlo, where the Data Generating

(30)

Table 9

Johansen’s Cointegration Test (

Seigniorage

Excluded as

Revenues)

Test Statistic

(Critical Value at the 20% Level)

Cointegrating

Vector

λmax

Trace

Κ =0 Κ ≤1 Κ =0 Κ ≤1

7.305

0.0454

7.351

0.0454

-(10.04)

(1.66)

(11.07)

(1.66)

Notes: The symbol (*) indicates rejection of the null at the 20% significance level. The

Imagem

Table 1 Unit-Root Tests
Table 8: Unconventional Impulse-Response Function ( ρρ =0.97) Impulse T T G* G* Response PV $ ( )∆T PV $ ( ∆ G * ) PV $ ( )∆T PV $ ( ∆ G * ) Pt-Estimate SE -1.22 (2.13) -0.07 (0.47) 1.12 (2.13) 0.05 (0.51) ( ρρ =0.98) Impulse T T G* G* Response PV $ ( )∆T

Referências

Documentos relacionados

Comparing to the surrounding trees of the same age, and older trees, a given tree has the highest growth rate of its total biomass simultaneously with its highest

Considering the importance of cultural systems in understanding people’s motivations, actions and subjective experiences we asked: in what ways do journalists experience

Mais especificamente, o estudo pretende: (1) descrever a composição da comunidade de pequenos pelágicos (riqueza e abundância de espécies) e suas variações em

Under certain conditions, it would be irrelevant how the deficits are financed, implying the assumption of the Ricardian Equivalence hypothesis, as stated already in

In order to assess the sustainability of budget deficits, co-integration tests between public expenditures and public revenues, allowing for structural breaks, are performed for the

No presente estudo, a análise foi voltada para verificar a ocorrência da relação entre o incômodo do zumbido e o exame de supressão de emissões otoacústicas, não chegando aos

− A incorporação das fibras de linho, independentemente do tipo de tecido utilizado, conduziu a um decréscimo da temperatura de fusão e das entalpias de fusão

Juliana Pinhata Sanches do Vale and Sílvio Hiroshi Nakao are the authors of the article titled Unconditional conservatism in Brazilian public companies and tax neutrality,