Japan and the World Economy 23 (2011) 178189

Contents lists available at ScienceDirect

Japan and the World Economy

journal homepage: www.elsevier.com/locate/jwe

Ination target and debt management of local government bonds,

Hiroshi Fujiki a, Hirofumi Uchida b,*
a b

Monetary Affairs Department, Bank of Japan, 2-1-1 Nihonbashi-Hongokucho, Chuo-ku, Tokyo 103-8660, Japan Graduate School of Business Administration, Kobe University, 2-1 Rokkodai, Kobe, Hyogo 657-8510, Japan

A R T I C L E I N F O

Article history: Received 6 September 2010 Received in revised form 6 May 2011 Accepted 6 July 2011 JEL classication: E52 H11 H7 Keywords: Fiscal decentralization Government bond management Externality Ination targets Fiscal discipline

Abstract: We show that the optimal ination target imposed on a discretionary central bank varies with the extent of scal decentralization. Our analysis compares two scal regimes for local government bond management: the partially decentralized (PD) regime where the central government determines the amount of local bond; and the fully decentralized (FD) regime where each local government determines the amount of local bond. In both regimes, an ination target has two effects: it harnesses surprise ination; and it induces excess issuance of local bonds. Due to externality in determining the level of local government bond, however, the second effect, and thereby the optimal level of the ination target, are smaller in the FD regime than in the PD regime. We also nd that even if scal decentralization in its isolation deteriorates social welfare, we may be able to improve social welfare by introducing an ination target when scal decentralization measures are adopted. 2011 Elsevier B.V. All rights reserved.

1. Introduction The aim of this paper is to theoretically investigate the impact of scal decentralization on the design of monetary policy. The proponents of scal decentralization argue that each public service should be provided by the jurisdiction having control over the minimum geographic area that would internalize benets and costs of such provision (Oates, 1972). In line with this argument, many economies including Canada, China, Colombia, Indonesia, Italy, and Spain have recently put forward scal decentralization. In Japan, the movement towards scal decentralization began in June 2003 when the Japanese Council of Economic and Fiscal Policy (CEFP) announced the Trinity Reform Package that aims to decentralize Japanese scal policy.1

Any views expressed herein are solely those of the authors and do not reect those of the institutions to which they belong. An earlier version of this paper was presented at the Contract Theory Workshop and the Monetary Economics Workshop. The authors would like to thank Hiroshi Osano, Shinsuke Kambe, Katsuya Takii and Tatsuya Kikutani for their helpful comments. The second author thanks the Inamori Foundation and the Grant-in-Aid for Scientic Research (No. 22330096) for nancial support. E-mail address: hiroshi.fujiki@boj.or.jp. * Corresponding author. Tel.: +81 78 803 6949; fax: +81 78 803 6949. E-mail addresses: hiroshi.fujiki@boj.or.jp (H. Fujiki), uchida@b.kobe-u.ac.jp (H. Uchida). 1 See Section 2 for more institutional background of the scal decentralization in Japan.

Although scal decentralization is an important change in policy implementing structure, monetary economists did not pay, to the best of our knowledge, much attention to the design of optimal monetary policy under scal decentralization. We can indeed evaluate the welfare impact of scal decentralization in its isolation by assuming that the structure of monetary policy bodies is given. However, as scal policy and monetary policy are closely interacted with each other, it is important to investigate the impact of scal decentralization on the interaction of scal and monetary policy. For example, in economies like Japan where scal decentralization is put forward within a country, we need to investigate the interaction among a monetary authority and multiple scal authorities, including a central government and local governments. In this paper, we theoretically consider the impact of scal decentralization on monetary policy. More specically, we investigate how scal decentralization of public bond issuance affects the conduct of monetary policy. To do so, we consider the interaction of three factors, an ination target, scal discipline, and scal decentralization of public bond issuance. Our modeling of the rst factor, the ination targeting scheme, is based on the literature on the time-inconsistency problem of discretionary central banks. Kydland and Prescott (1977) show that without commitment to a predetermined policy and given the expectations of private agents, the central bank has an incentive to create surprise ination, thereby boosting the economy. Given rational expectations of agents, surprise ination increases the

0922-1425/$ see front matter 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.japwor.2011.07.001

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

179

equilibrium ination rate, but fails to raise equilibrium output. This time-inconsistency problem is important because, in practice, it is difcult for a central bank to commit to a predetermined policy. A solution to this problem is to introduce an ination target, as suggested by Walsh (1995). As for the second factor, we follow Beetsma and Bovenberg (2001) and model scal discipline as reducing socially wasteful expenditure by local governments that only meets specic local interests. The consideration of scal discipline is important because it directly affects the determination of the amount of local bonds. In practice, introducing scal discipline may involve reducing the benets of local special interest groups, which may impose the costs of regional unrest and tension. However, neither the central bank nor the central government considers the imposition of scal discipline costly. Under the framework that integrates these two factors, we investigate the impact of the third factor, a change in the management of local government debt due to scal decentralization. We examine the change by comparing the following two scal regimes. First, we consider a partially decentralized scal policy (PD) regime, which best represents Japans situation before scal year 2006. In this regime, the central bank controls the ination rate, the central government determines the amount of local bonds issued and the local tax rate, and local governments determine the level of local government expenditure. Second, we consider a fully decentralized scal policy (FD) regime, which represents Japans scal policy after scal year 2006. In the FD regime, the central bank controls the ination rate, the central government determines the local tax rate, and local governments determine the amount of local bonds to be issued and the level of local government expenditure. Under the setting explained above, we rst demonstrate that in the PD regime, the central government determines the amount of local bonds outstanding by taking into account two marginal effects of issuing an additional local bond. Firstly, the local bond compensates for the adverse effects of wasteful scal expansion by the local governments (the compensation effect of local bonds). Secondly, it affects the ination-setting behavior of the central bank, and thus increases the magnitude of surprise ination and social loss in the future (the cave-in effect of local bonds). The amount of local bonds outstanding is determined by balancing these two effects. The direct effect of imposing an ination target in this setting is to prevent surprise ination, which has been already recognized in the literature. We call this direct effect the monetary discipline effect of ination targets. In addition, however, there is another effect. By harnessing surprise ination, an ination target indirectly weakens the cave-in effect of local bonds, and thereby increases the equilibrium amount of local bonds. This is the scal expansion effect of ination targets. The optimal level of the ination target balances these two offsetting effects. We then proceed with the case in which scal decentralization occurs (i.e., the FD regime). In this case, yet another distortion, referred to as the externality effect, appears in the determination of local bonds. The local governments, which now control the amount of local bonds, put a greater weight on output gap minimization in their own region, and do not fully take into account the effect of their behavior on the national ination rate. This reduces the compensation effect of local bonds relative to the cave-in effect, and more local bonds are issued in the FD regime. When an ination target is introduced in this regime, it has the same two effects, the monetary discipline effect and the scal expansion effect, on the amount of local bonds as in the PD regime. However, the latter effect becomes smaller in the FD regime than in the PD regime because the local governments issue ceteris paribus more local bonds than in the PD regime due to the externality effect explained above. It is thus optimal to set a more conservative

(i.e., smaller) ination target in the FD regime than in the PD regime. The most important policy implication obtained from this analysis is that policy makers should pay attention to the degree of scal decentralization when they impose an ination target on a discretionary central bank. Svensson (1997) shows that setting a constant ination target eliminates ination bias associated with surprise ination. This rule is undesirable once we take into account the scal policies by central and local governments. We further show that as scal decentralization advances, the optimal target level should be conservative. Finally, we evaluate the welfare levels in the alternative scal regimes. In our model, on the one hand, scal decentralization which is characterized as a change from the PD regime to the FD regime increases social loss because of the externality effect. On the other hand, the introduction of an ination target decreases social loss, both in the FD regime and in the PD regime. An interesting implication derived from these results is that accompanying scal decentralization with introduction of an ination target achieves a lower level of social loss. Our model is an example of the conict between scal decentralization and macroeconomic policy that is pointed out in the scal decentralization literature. Fiscal decentralization may lead to a conict between local governments pursuing an expansionary scal policy and the national government pursing a contractionary scal policy. The situation is likely to be relevant if the local government faces a soft budget constraint (see Tanzi, 2001 for details). In our model, we additionally consider the presence of the monetary authority and thus scal decentralization puts pressure on both the central bank and the central government. Although we take Japan as an example of an economy pursuing scal decentralization of public bond issuance, the present analysis is applicable to other economies which are moving towards scal decentralization as well as an introduction of an ination targeting framework. Methodologically, the present paper is closely related to two studies by Beetsma and Bovenberg (1997, 2001). Beetsma and Bovenberg (2001) consider the role of a scal transfer system, taking into account the lack of scal discipline. There are three main differences between their model and that presented in the current analysis. First, they do not consider scal decentralization because in their model, national, rather than local, government implements decentralized scal policy. Second, their model is static whereas we consider debt issuance in a two-period model. Third, their primary interest is the role of monetary policy and the public transfers system as shock absorbers for region-specic economic shocks. This is of no concern in this paper. With regard to the relationship between time-inconsistent monetary policy and public-debt policy, our model is similar to Beetsma and Bovenberg (1997). However, their analysis focuses only on centralized scal policy, and they do not focus on the issue of scal decentralization and scal discipline. Furthermore, they do not investigate ination targets in their model.2 With regard to the relation between our theoretical institutional setup of ination targeting and actual examples of such schemes, Heenan et al. (2006) and Svensson (2011) recommend a small but positive rate of ination target as a practically desirable target, because in practice we need to take into account issues such as measurement bias of price indexes, rigidity of nominal wages, and the reduction of the probability of hitting zero bound of

2 In a similar setting, Fujiki et al. (2004) compares alternative institutional arrangements including ination targets to achieve the optimal outcome, and Uchida and Fujiki (2005) examine the optimal state-contingent ination target under uncertainty. However, these studies also do not consider the issue of scal decentralization.

180

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

nominal interest rate and of falling into deation. We abstract from these issues because we focus on a theoretically neglected interaction between scal decentralization and the optimal level of ination target, but the issues raised above must be taken into account when we consider the real-world application of our model. The remainder of the paper is organized as follows. The next section reviews brief institutional background of scal decentralization in Japan. In Section 3, we present our model. Section 4 considers the second-best allocation. We analyze a partially decentralized scal policy regime without commitment in Section 5, and a fully decentralized scal policy regime without commitment in Section 6. In Section 7, we extend the analysis on social welfare. Section 8 concludes the paper by summarizing policy implications obtained from the analysis.

soundness of a local government in those measures worsens beyond a threshold value, the central government can stop the issuance of the local bonds by the local government.5 In sum, the scal decentralization evoked by the Trinity package made local governments more independent and accountable to the issuance of local bonds. 3. Model 3.1. Supply function There are n regional economies in a country. These are indexed by i = 1, . . ., n. There are four types of agent: private agents and a local government for each regional economy, a central government, and a central bank. We consider a two-period model (t = 1, 2) in order to include local bonds.6 The behavior of private agents in each economy is described by the following supply function: xit npt pe t it : t (1)

2. Fiscal decentralization in Japan The movement towards scal decentralization in Japan began in June 2003 when the Japanese Council of Economic and Fiscal Policy (CEFP) announced the Trinity Reform Package that aims to decentralize Japanese scal policy. The package encompasses three primary measures to achieve this aim: rst, it proposes to redistribute more tax revenue to regional governments. Second, it advocates cutting the local allocation tax grant. The third measure recommends reconsidering the role of national government disbursements.3 In accordance with the spirit of this package, the institutional setup for funding local government bond has changed dramatically since scal year 2006. Before scal year 2006, the Ministry of Internal Affairs and Communications (MIC) approved the issue of local bonds. After scal year 2006, local governments can issue local bonds, even if the MIC disagrees. More precisely, after scal year 2006, local governments need to consult with the MIC or prefectural governments only on their bond issuance plans. Even if the MIC or prefectural governments disagree, the local governments can still issue the local bonds, as long as the local parliament approves.4 Furthermore, to make the sprit of the Trinity package clear, the Law for the Promotion of Decentralization Reform was enforced in April 2007. The law changes the responsibility between the central government and local governments for many areas of public services. The changes include the enhancement of nancial independence of local governments from the central government. The independence accompanied the accountability on the nancial position of local governments. Looking at the nancial distress in several local governments in recent years, the Law relating to the Financial Soundness of Local Government was passed in 2007, and enforced from April 2009. Under this new law, the central government checks from 2009 the nancial soundness of local governments using ve nancial indexes, including a real decit ratio or a real debt service ratio. In particular, if the nancial
3 Behind the announcement of Trinity Reform Package lays the fact that, the main sources of revenue for local governments were local taxes, local allocation tax grants, national government disbursements, and local bonds. These sources of revenue were under the strict control of the central government in those days. First, local governments did not have the right to determine taxes or tax rates; these were determined by the Local Tax Law, which was a national law. Second, the local allocation tax grant, a public nance transfer from the central government, was under the control of the Ministry of Internal Affairs and Communications (MIC). Third, the national government disbursement was under the discretion of the central government. Finally, local government bonds were issued upon approval by the MIC. 4 Note that there are two special occasions in which the MIC or prefectural governments still needs to approve the issuance of local government bonds. They are cases in which the revenue of the local government is insufcient relative to its expenses and in which it is insufcient relative to the amount of local government debt outstanding. See details in a document of Japan Local Government Bond Association (http://www.chihousai.or.jp/english/04/overview12.html).

In the above equation, xit is the natural logarithm of supply, n is a constant, and pt is the ination rate which is the same for all regional economies. In addition, pe is the ination rate, which t private agents rationally expect, and tit is the tax rate.7 We suppose that pt is set by the central bank in each regime. The determination of tit is explained later. 3.2. Budget constraint We use a simple framework to model government budget constraints and to focus on the basic mechanism behind the determination of the amount of local bonds issued. The local governments face the following budget constraint:

t i1 di g ei ; t i2 g 1 rdi :

(2) (3)

di(! 0) is the amount of local bonds, g is the level of local government expenditure, and r is the real interest rate. For simplicity, g and r are assumed to be constant, which implies that ti1 and ti2 are residually determined when ei and di are chosen.8 ei(! 0) is the level of wasteful expenditure, which represent scal discipline explained below. This specication follows that employed in Beetsma and Bovenberg (2001). Note that ei is wasteful and has no role other than to increase local benets. For example, the central government may be reluctant to promote a new airport in a local prefecture, while a local government would like its own new airport, even if it were unprotable. As shown below, scal discipline is best achieved when ei is zero.
5 See White Paper on Local Public Finance, 2007, MIC for details on those institutional changes. 6 Extending the model to include more than two periods does not materially change the main results of this paper. 7 This simplication is justied as follows. Private agents care only about their target wages, the logarithm of which is normalized to zero. Under rational expectations, the log of the wage is equal to the log of the expected price level. Output is produced according to a standard CobbDouglas production function, in which the only input is labor. For a more detailed derivation, see Beetsma and Bovenberg (1997). Note that setting the productivity coefcient of the Cobb Douglas production function to 0.5 yields Eq. (1) with n = 1, as shown by Beetsma and Bovenberg (2001). 8 In reality, local governments in Japan do not have the right to determine taxes or tax rates. Rather, the rates are determined by the Local Tax Law, which is a national law. In this sense, the present model considers a situation where local governments have more inuences on the level of local tax rates than they actually do. This assumption is, however, for simplication of the analysis. The model can be extended to allow for an explicit determination of tax rates by the central government, although it complicates the analysis.

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

181

Note that ei enters into the budget constraint in the rst period only. This is because second-period discipline is inessential to the analysis and its absence simplies the model. We further assume that ei is not veriable and that it is impossible to write a contract or target based on the value of ei. This is because, in practice, it is difcult to know the nature of the changes in public spending. Some may interpret an increase in ei as a subsidy to a local group, while others may believe that the increase reects a tax increase due to a positive shock. ei could be modeled as subject to uncertainty, although we do not explicitly model it to avoid complexity. In Japan, another important issue between the central and local governments is public nance transfers from the central government: the local allocation tax grant and disbursement by the national government.9 We use a simpler model that focus on the basic mechanisms behind the issuance of local bonds and do not take account of the transfers. Nonetheless, our model is sufcient to shed light on problems arising from the decentralized issuance of local bonds and its implications for the conduct of monetary policy. Incorporating the effects of public nance transfers from the central government is an interesting and important exercise, but because this makes the model very complicated we would like to defer this exercise to a future study. 3.3. Loss function Suppose a local loss function in each region that is determined by the rate of national ination and the output gap, which is dened as the difference between actual and desired output. LL LL bLL ; i i1 i2 where i 1h 2 fpt xit x2 ; 2

optimal level of the ination target. We can extend the model to take account of public spending in the loss function, but it makes it difcult to understand the main points of this paper. The local governments have their own loss function, which reects a preference for (or lack of) scal discipline. The function is represented by: LLG LLG bLLG ; i i1 i2 where 1 2 fpt x2 at dei ; it 2

LLG it

t 1; 2:

This formulation of scal discipline follows that of Beetsma and Bovenberg (2001). In this equation, d is the coefcient representing the magnitude of the benet accruing to the local government due to wasteful expenditure. at is an index to include ei only in the rstperiod budget; that is, a1 = 1 and a2 = 0. This loss function is designed to represent the following situation. In practice, scal discipline may be achieved by cutting benets to local special interest groups, and this discipline reduces the need for nance on a one-to-one basis. However, enforcing scal discipline is costly for the local governments because greater scal discipline, represented by a decrease in ei, leads to regional unrest or tension. Thus, the enforcement of stricter scal discipline increases the loss to the local governments, but does not increase that of the central government. The central bank has the following loss function, which captures the effects of the ination target set by the central government in the absence of policy commitment: LCB LCB bLCB ; 1 2

LL it

t 1; 2:

where !

In this equation, b is a discount factor that takes a value between zero and unity, f is a weight that represents the relative importance attached to stabilizing ination and output, and x is the target level of output. For expositional simplicity, we assume that x 0 throughout. Therefore, xit below represents not only the output level but also the output gap. We assume that the efciency of policy regimes (centralization vs. decentralization of local bond issuance and the setting of an ination target) is judged by the following social loss function of the country, which is the average loss across regions. 1 S LL n i i 1 1 S LL b Si LL ; i2 n i i1 n  LC bLC ; 1 2

LCB t

1 1 fpt p 2 S j x2 t jt 2 n

t 1; 2:

LC

We model the central bankers discretionary policy making where p is zero. If the government imposes an ination target, we t assume that the government announces the target level of ination rate p , which is to be achieved by the central bank. The t government chooses p to minimize LC before the beginning of t the rst period. Note that the central bank only cares about national ination, rather than regional ination. This assumption is consistent with the mandate of the Bank of Japan. Moreover, the ination differential among regions in Japan are very small. For example, annual changes in consumer price index (general, excluding fresh food) in 2007 by regional break down differ only 0.5 percentage point across regions. 3.4. Policy regimes

where ! 1 1 fp2 S j x2 ; LC t t jt 2 n

t 1; 2:

Many papers in this area assume that governments care about public spending in addition to ination rate and output. Following Beetsma and Bovenberg (1997) we do not take account of public spending so as to focus on complex interactions between the degree of scal decentralization on the bond issuance and the
9 Beetsma and Bovenberg (2001) consider the situation in which the budget constraints of local governments include public nance transfers from the central government.

Policy and choice variables are pt, di, and ei (t = 1, 2, i = 1, . . ., n).10 pt is determined by the central bank, and the local governments determine ei of each region. However, depending on the structure of scal policy regimes, different authorities choose di. We consider ve policy regimes. In the partially decentralized (PD) regime, the central government determines di as in Japan before 2006. In the fully decentralized (FD) regime, it is the local governments that choose di, even if, in principle, the central government disagrees as in Japan after 2006. Fiscal decentralization is modeled as a move from
10

g is exogeneous and ti1 and ti2 are residually determined.

182 Table 1 Summary of the results. Regimes SB PD PDIT FD FDIT

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

p2
0
1 fS 1

p1
0

xi2 ng
1 nS d 1 nS d

xi1 ng 1d n 1d n 1d n
2 2

ei 0
11rS n2 1rS 11rS n2 1rS 11rS0 n2 1rS0

fd

d 11r g 1r d 11r g 1r d 11r g 1r d 11r g 1r p 2

(NA) (NA) 0 (NA)
nn1 n2 n2 f d

fS d fS0 fS0
1 1

0
fd
2

d
n d nn2 n2f f

n1 0 d S n1 0 d n 2fnn2n S n f

0 di 0
1 n2 1rS 1 n2 1rS 1 n2 1rS0 1 2 2 n n f n2 1rS0 1 d 1r g

1d n

n2 n1 11rS0 2 2 n n f n2 1rS0

Regimes SB PD PDIT FD FDIT

p 1
(NA) (NA)
d f

1 1r g

1 d 1r g

(NA)
d f

n2 n1

1 d 1r g

the PD regime to the FD regime. To isolate the direct impact of scal decentralization that is the most relevant to the conduct of monetary policy, we intentionally abstract other aspects of scal decentralization of public debt issuance. Incorporating those aspects is important, but complicates the analysis, which is left for a future study. In the context of the two scal regimes, we introduce an ination target that is exogenously imposed by the central government to examine the effects of scal policy on the central banks optimal ination target. These regimes are referred to as the PDIT and FDIT regimes, respectively. A large number of studies have argued that the Bank of Japan should adopt an ination target to manage ination expectations formed by private agents. To our knowledge, this paper is the rst to simultaneously consider the adoption of an ination target and scal decentralization in a unied framework.11 These scal regimes are to be compared with a benchmark case, the second-best allocation (SB). This is the case in which a social planner determines the levels of the policy variables and is committed to predetermined policies.12 3.5. Assumptions Before starting the analysis, we explain two assumptions that are made throughout the analysis. The rst assumption is as follows: 1 b1 r 0: (4)

as di becomes zero in the SB regime. Qualitatively similar results can be obtained, even if we assume instead that 1 b(1 + r) > 0. However, the derivation is then more complicated. The second assumption is as follows:

d > n2 Sg;
2

(5)

where S  1 + n /f. This ensures that di is nonnegative when ei > 0 in the PD regime. This assumption is sufcient for di > 0 in the FD regime as well. Note that this assumption also ensures:

d!

1 1 rnS ng; 1 1 rS

which is the necessary and sufcient condition for the local benet of increasing ei to be so large that it is optimal for the local governments to set a positive ei in the equilibrium of the PD regime. This inequality is also a sufcient condition for ei to be nonnegative in the FD regime. 4. Second-best resource allocation In this section, we discuss the properties of the second-best allocation. We assume that a social planner commits to the optimal allocation for both monetary and scal policies. Specically, the social planner sets pt, di, ei (t = 1 and 2, i = 1, . . ., n) to minimize LC at the beginning of period 1.13 Note that in this regime, as the central bank can commit to a predetermined ination rate, we take the condition pt pe as given. We assume a t symmetrical solution. The results in the SB regime are presented in Table 1. Appendix A.1 contains a brief derivation of these results. We offer their economic interpretations for the rest of this section. Note that for ease of comparison, Table 1 also presents the results in the other (i.e., PD, PDIT, FD, and FDIT) regimes that will be discussed and compared in the subsequent sections. 4.1. The ination rate In the second-best equilibrium, we obtain a zero ination rate for both periods. This is because there is no surprise ination
13 The results are unchanged, even if policies are determined at the beginning of each period.

This implies that the real rate of interest is equal to the inverse of the discount factor, and that agents discount the loss in the second period by the same rate as the interest rate on local bonds. This assumption is solely for analytical simplicity,
11 Note that in some sense it might be the FDIT regime that resembles the situation in Japan after scal 2006. On March 10, 2006, the Bank of Japan released a document The Banks Thinking on Price Stability. This document does not explicitly mention an ination target, but states an understanding of medium- to long-term price stability. In particular, it states that by making use of the rate of year-on-year change in the consumer price index to describe the understanding, an approximate range between zero and two percent was generally consistent with the distribution of each Board members understanding of medium- to long-term price stability. As of April 28, 2011, the understanding expressed in terms of the year-on-year rate of change in the CPI falls in a positive range of 2 percent or lower, centering around 1 percent. 12 Following Beetsma and Bovenberg (2001), we term this case the second best case because in the presence of lump-sum taxes the welfare level can be further higher, which is the rst-best case.

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

183

because of the lack of commitment to the ination rate chosen in this policy regime. As pt is already known when expectations are formed, pe is equal to pt. Eq. (1) implies that it is impossible to increase t output by increasing the rate of ination, taking advantage of a given level of expected ination. Hence, the ination rate pt only affects the rst term of LC , and pt = 0 is optimal. t 4.2. Fiscal discipline We obtain the result that eSB = 0 in the second-best equilibrium. This is a very natural result for the social planner, because ei is purely waste and increases social loss. To elaborate on this point, the marginal effect of ei on social loss is:

a benchmark for other cases. Extension to the case of sequential decision-making is an intriguing future research topic. We begin by obtaining a solution for each period for the central bank by minimizing LCB . Then, we analyze decision-making by the t central and the local governments. In Section 5.1, we obtain the results for the PD regime, in which there is partial decentralization with no ination target, by simply setting p to zero. In Section 5.2, t we obtain the results for the PDIT regime, in which there is partial decentralization with an ination target, by taking account of p . t Section 5.3 discusses policy implications of these results. 5.1. PD regime: partial decentralization without an ination target The results in the PD regime are as summarized in the second row of Table 1. The detailed derivation is found in Appendix B.1. We discuss the economic intuition behind these results below. 5.1.1. Ination rate Comparing the ination rates in Table 1, we can see that the ination rate is higher in the PD regime than in the SB regime. This is the standard surprise ination result. Because expected ination rates are formed when the central bank determines the level of pt, the rst-order condition for the central bank is:

@LC 1 @xi1 @t i1 x @ei n i1 @t i1 @ei

i 1; . . . ; n;

(6)

which is positive in equilibrium. 4.3. Local bond Due to the simplifying assumption in Eq. (4), the equilibrium level of local bonds becomes zero. The rst-order condition for d is then as follows: 1 1 nng ei di b n1 rng 1 rdi 0: n n (7)

@LCB 1 @x fpt Si xit it 0: n @pt @pt

(8)

The rst term represents the marginal effect of di on LC through ti1, 1 and the second term represents the effect on LC through ti2. As 2 ei = 0, balancing these effects yields: 1 b1 r ng; n1 b1 r2

dSB i

i 1; . . . ; n:

In contrast to the case of policy commitment where pe = p, the discretionary central bank can boost the economy, represented by the second term in this equation. The bank thus sets a higher ination rate than that in the SB case, although the behavior of the bank is rationally expected before forming the expectation, which results in pe = p in equilibrium. 5.1.2. Fiscal discipline The lack of scal discipline creates an incentive for the local governments to increase ei in the PD regime. For a local government, the rst-order condition for ei is as follows:

We conrm that assumption (4) leads to the optimality of a zero amount of local bonds. Under this assumption, the marginal effects of d on LC and on LC cancel out. Note again that the assumption is 1 2 made for the purposes of analytical simplicity. The main results of this paper are qualitatively the same, even if we assume that 1 b(1 + r) > 0 and investigate the case in which dSB > 0. In i Appendix A.2, we provide an intuitive account for the bond determination in the SB regime. 5. Partial decentralization without commitment We next consider the PD regime, where dis are determined by the central government, but eis are determined by the local governments. In the PD regime, expectations about the rst-period ination rate, pe , are formed at the beginning of the rst period. 1 Following the formation of the ination expectations, the local governments determine ei, the central government chooses di, and the central bank determines p1, respectively. The decisions of these agents are made simultaneously. In the second period, ination expectations pe are formed, and the central bank subsequently 2 determines p2. By assumption we abstract from any scal choices in the second period. Note that in the PD regime, the central bank cannot be committed to a predetermined ination rate. Therefore, the bank can determine pt that differs from pe , and is not subject to the t constraint pt pe . In equilibrium, however, rational expectation t ensures pt pe . t As for the timing of the decision-making by the policy makers, we argue that modeling simultaneous decision-making by the policy makers is realistic, because it is hard to strategically change a policy decision just in response to decisions by other policy makers. Also, simultaneous decision-making is analytically tractable and serve as

@LLG @x @ti1 xi1 i1 d 0; @ei @ti1 @ei

i 1; . . . ; n:

(9)

Note the difference between this equation and Eq. (6). Increasing ei benets the local government, which is represented by the inclusion of d in Eq. (9). Assumption (5) ensures that d is sufciently large, so that the benet of increasing ei to enjoy local benets outweighs the losses that arise from wasteful expenditure. Therefore, ei is positive. 5.1.3. Local bond In the PD regime, the local bond level is determined by the following rst-order condition:

@LC 1 1 nng ei di p2 n2 n @di n 1 nng 1 rdi 0;

n

(10) i 1; . . . ; n:

Eq. (10) appears almost the same as Eq. (7). However, there are two differences between the two equations. First, the second term of Eq. (10) reects the fact that the amount of local government bonds affects the second-period ination rate. The central banks benet from boosting the economy through surprise ination increases as the amount of local bonds increases, but this increases the secondperiod social loss. This is the additional marginal cost of increasing di. We call this cost the cave-in effect of local bonds. Second, ei in the rst term of Eq. (10) becomes positive, while it was zero in Eq. (7). Thus, compared with the SB regime, the marginal

184

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

benet of issuing a new local bond in the rst period (represented by the rst term) increases. To put it differently, let us note that a marginal increase in di from the second best level (=0) affects the rst-period loss because (i) the increase in di decreases the rstperiod tax rate, (ii) the reduced tax rate increases the rst-period output, and (iii) the increased output reduces the rst-period loss. Because of a positive ei, effect (iii) is larger in the PD regime than in the SB regime, which makes it attractive to increase di. We refer to this additional marginal benet of di in the rst period as the compensation effect. The optimal amount of local bonds balances the additional marginal benet (the compensation effect) and the additional marginal cost (the cave-in effect). Under assumption (5), the former dominates the latter. The equilibrium level of di in the PD regime is positive, rather than zero as in the SB regime. A more detailed illustration of the determination of di is given in Appendix B.2. One important implication of our nding here is that we must take into account the compensation effect, as well as the cave-in effect, when we investigate the association between the ination rate and government bonds. The mechanism of the cave-in effect, i.e., the reduction of the level of government bonds from the optimal level so as to inuence the future ination rate, has already been pointed out by Beetsma and Bovenberg (2001) in a slightly different setting.14 We nd that the cost of the cave-in effect should be weighed against the benet of the compensation effect. If the compensation effect dominates the cave-in effect, a correct policy recommendation would allow the central government to increase the level of government bonds. 5.2. PDIT regime: PD with an ination target Now we consider the situation in which the central government announces the target ination level p . The central government t chooses p to minimize LC before the beginning of the rst period. t We demonstrate that it is optimal for the central government to set the following levels of ination targets:

Similar to the rst-period target, the direct effect of introducing a negative second-period ination target is the monetary discipline effect, i.e., it decreases the second-period ination rate (the rst term of Eq. (11)). This is the marginal benet of introducing a negative target, which aids the prevention of surprise ination. In addition to the monetary discipline effect, however, another offsetting effect arises from a second-period target (the second term of Eq. (11)). A negative ination target in the second period increases the amount of local bonds and thereby the second-period tax rate. This makes the negative second-period output gap (see Table 1) deviate further from zero. This is the marginal cost of introducing a negative target. We call this the scal expansion effect of a negative ination target. The latter effect is unique to the second-period ination target. Similar to the rst-period target, the direct effect from introducing a negative second-period target (the monetary discipline effect) is to change the central banks ination-setting behavior in the same period, because the target only enters in the rst term of the loss function and the rst-order condition for ei or d is free from the target level. However, in equilibrium, the second-period target indirectly affects the level of d. This is because the change in the level of p2 due to the introduction of the second-period target changes the rst order condition for di. More specically it changes the magnitude of the cave-in effect as demonstrated by the second term of Eq. (10). A negative target decreases p2(> 0) and thus weakens the cave-in effect of di. Because the compensation effect then dominates, and the equilibrium level of di increases. This is detrimental to social welfare, as it expands the second-period output gap. Note again that the rst-period target does not affect the level of d, since it only affects the behavior of the central bank in setting the rst-period ination rate. The optimal level of the second-period ination target balances its marginal benet (the monetary discipline effect) and its marginal cost (the scal expansion effect). We demonstrate that a zero ination target in the second period balances these two effects. On the one hand, Eq. (11) can be rewritten as: ' & @LC 1 @LC 1 2 Si 2 n 0: S @p2 n @x2i (12)

d pPD ; 1 f pPD 0: 2
The equilibrium in the PDIT regime given these ination targets are characterized in the third row of Table 1, a derivation of which is outlined in Appendix B.3. The intuition underlying the negative and zero ination targets for the rst and the second periods, respectively, is as follows. As for the rst-period target, it only affects the behavior of the central bank in setting the rst-period ination rate, because p enters the 1 rst term of LCB only. All the target can then do is to x the 1 distortionary behavior of the central bank, i.e., surprise ination, in the rst period. As pPD > 0, it is optimal to set a negative target to 1 induce p1 = 0. We call this the monetary discipline effect of introducing a negative ination target. In contrast, the optimal second-period target plays a more complicated role. Interestingly, its optimal level is zero, which implies that imposing an ination target in the second period is not desirable. The reasoning behind the irrelevance of the second-period target is as follows. The rst-order condition for the determination of p is: 2

On the other hand, we can show that the rst-order condition for the ination-setting behavior of the central bank in the second period is now:

fp2 p Si xi2 n 0; 2
and thus Eq. (12) becomes

1 n

(13)

1 fp 0: 2 S This implies that, without a second-period target, the relative magnitude of the marginal benet and the marginal cost of the second-period ination rate with respect to LCB is the same as the relative magnitude of the marginal benet and the marginal cost of the second-period target with respect to LC. It should be noted that the offset of the marginal benet and the marginal cost relies on the assumption of a loglinear supply function and a quadratic loss function. If different functional forms are adopted, the optimal ination target would not be zero. Nevertheless, even in a more general specication, p has a similar 2 indirect effect on LC because it determines the magnitude of the cave-in effect, and thereby affects the level of di. Such an effect is expected to move the optimal level of the ination target towards zero.

@LC @p2 1 @LC @x2i @t2i @di 2 2 S 0: @p2 @p n i @x2i @t2i @di @p 2 2

(11)

14 This effect is called the credibility effect in Beetsma and Bovenberg (2001). As the setting is somewhat different, the interpretation also differs.

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

185

5.3. Discussion The PD regime approximates the situation in Japan before 2006 in two respects. First, the MIC retains control of the amount of local government bonds and gradually cuts public nance transfers before making signicant changes to the local tax system. Second, the Bank of Japan does not adopt an ination targeting framework. According to the results obtained in Section 5.1, the behavior of the central government can be interpreted as follows. Taking into account the impact of bond issuance on future ination (the cavein effect), the central government has an incentive to reduce local bonds below the second-best level. However, as the local governments prefer to spend wastefully, local scal discipline is absent. The central government is thus obliged to offset the increase in local spending by increasing local bonds (the compensation effect). If the discussion in Section 5.2 offers any guide, it is that the design of an ination target requires an attention to scal policy if the wasteful expenditures of local governments are unavoidable. The ination target that only focuses on the prevention of surprise ination (the monetary discipline effect) is very costly. An ination target determines the extent of the cave-in effect of local bonds as well because the target directly determines the level of the ination rate. A negative ination target thus has an adverse effect on social welfare (the scal expansion effect). The loss from this adverse effect may completely offset the benecial effect of preventing surprise ination if the wasteful expenditures of local governments are unavoidable. As the difference between the equilibrium ination rate in the PD and SB regimes is constant, one might expect that a central bank with a conservative ination target, as suggested by Svensson (1997), can benet society in both periods 1 and 2. However, in the determination of pPD , we took into account the fact that p affects 2 2 not only p2 but also di. One can indeed equate pPD and pSB by 2 2 setting p as follows: 2

second term of Eq. (10)) differs in the PD regime and in the FD regime. That is why the equilibrium level of the second-period ination rate differs between the regimes. 6.1.2. Fiscal discipline The level of ei is also determined in the same way as that in the PD regime, as in both regimes it is determined by the local governments. The rst-order condition for the determination of scal discipline is the same and is represented by Eq. (9). Hence, the decision rule for the local governments about scal discipline remains the same under both the PD regime and the FD regime. However, the equilibrium level of scal discipline differs in the PD regime and in the FD regime. We can conrm that eFD > ePD . i i Under the FD regime, an additional distortion, which is explained later, affects the amount of local bonds and thus the level of scal discipline differs. 6.1.3. Local bond We obtain the result that dFD is larger than dPD . The rst-order i i condition for di is:

nng ei di p2 n2 nng 1 rdi 0;

i 1; . . . ; n:

1 n

(14)

p 2

fb1 rS

d:

However, as this distorts the determination of di, it is not optimal. 6. Full decentralization without commitment We now consider the FD regime, where dis and eis are determined by the local governments. This regime approximates the situation in Japan after the scal year 2006. The sequence of decision-making is the same as in the FD regime, but it is now the local governments that determine di. 6.1. FD regime: full decentralization without an ination target The equilibrium in the FD regime is represented in the fourth row of Table 1. In this table S0  (1 + n2/nf) < S. These results are derived following the same procedure as in the case of the PD regime (Appendix B.1) but taking into account the fact that it is the local governments that determine dis.15 6.1.1. Ination rate The determination of the ination rate is the same as in the PD regime, and is represented by Eq. (8). The central bank has an incentive to boost the economy and thus to create surprise ination. Under rational expectations, the equilibrium level of the ination rate is positive. However, as it is not the central government but rather the local governments that determine the level of di in the FD regime, the extent of surprise ination (the

The difference between Eqs. (14) and (10) is that the rst and third terms are not multiplied by 1/n in Eq. (14). This difference arises because the local governments, rather than the central government, determine di. Under the PD regime, di indirectly affects LC through xit because its effect is transmitted through one of the n regions budgets, while in the FD regime, dis directly affect LLG , and no multiplication factor 1/n exists in the i rst and third terms of Eq. (14). Thus, in addition to the distortions observed in the PD regime, there is an additional distortion that arises because the local governments only consider the trade-off between their regional outputs and national ination. This is termed the externality effect of scal decentralization. Compared with the rst and the third terms, the second term of the rst-order condition for di is relatively smaller in the case of Eq. (14) than in the case of Eq. (10). That is, due to the externality effect, the cave-in effect is relatively smaller, or the compensation effect is relatively larger, in the FD regime than in the PD regime. As demonstrated in Section 5.1, the cave-in effect decreases the amount of local bonds, while the compensation effect increases it. Accordingly, the marginal cost of issuing additional local bond decreases, resulting in larger issues of local bonds.16 Our result reproduces the risk of scal decentralization with soft budget constraints suggested in the literature: the local governments waste ei in both the PD regime and the FD regime. As they can determine the amount of local bonds in the FD regime, they have access to a laxer nancing source, or a softer budget, and issue more bonds to waste more expenditure. 6.2. FDIT regime: FD with an ination target We assume that the central government announces the target level of ination rate p , which is to be achieved in the FD regime. t The central government chooses p to minimize LC before the t beginning of the rst period. We can demonstrate that the optimal target levels in the FD regime are as follows:
15

For a detailed derivation, see Fujiki and Uchida (2011).

186

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

LC

d pFD ; 1 f nn 1 pFD 2 d: 2 n n2 f
Note that the optimal ination target is now negative in the second period. The equilibrium in the FD regime with these ination targets are summarized in Table 1(the fth row).17 The derivation of the rst-period target is the same as that under the PD regime. The rst-period target has a monetary discipline effect only. The target is thus set to prevent surprise ination in the rst period. A negative ination target pFD d=f 1 induces a socially efcient ination rate in the rst period. The difference between the PDIT regime and FDIT regime is with respect to the second-period ination target. The optimal level of the second-period ination target is now negative. Based on the discussion in Section 5.2, this is a very natural result. Because of the externality effect in the FD regime, the relative magnitude of the second term of Eq. (14) is smaller than that in Eq. (10). The smaller the cave-in effect, the smaller the indirect effect (the scal expansion effect) of the second-period ination target on the equilibrium local bond level. Therefore, the benet of the monetary discipline effect of a conservative target, i.e., the prevention of second-period surprise ination, outweighs the cost of the scal expansion effect of a conservative target, which works through the determination of d in the FDIT regime. As a result, we obtain a smaller target with a larger absolute value. 6.3. Discussion The FD regime approximates the situation in Japan after scal year 2006 in two respects. First, local governments determine the amount of local government bonds. Second, at the time of writing, the Bank of Japan has not adopted ination targeting framework. According to the results obtained in Section 6.1, there is an additional distortion, the externality effect, which shows up in the determination of the amount of local bonds by the local governments in the FD regime. This additional effect arises because the local governments put a greater weight on output gap minimization in their own regions, ignoring their effect on the national ination rate, and increase the amount of local bonds outstanding. The most important implication obtained in Section 5 was that, when we design an ination targeting framework, we must weigh the indirect negative effect of a conservative target, which increases the amount of local bonds, and the direct positive effect of a conservative target, which prevents the central bank from creating surprise ination. The results in Section 6.2 imply that the relative magnitude of the two effects vary depending on the extent of scal decentralization. Due to the externality effect, the positive effect of imposing a conservative negative ination target dominates the negative effect in the PDIT regime. This is why the optimal level of the ination target becomes smaller. 7. Welfare comparison Having focused on the determination of the ination target, we now turn to a welfare comparison. Fig. 1 briey summarizes how the equilibrium welfare levels, as measured by the equilibrium values of the loss functions, differ across policy regimes.18 The
16 Fujiki and Uchida (2011) provide more intuitive account for the determination of di in the FD regime. 17 See Fujiki and Uchida (2011) for a more detailed derivation of the results.

FD

FDIT PD PDIT SB

Fig. 1. Welfare comparison between the ve regimes. The value of the social loss function, LC, is illustrated for each regime. Where there is one region (n = 1), the values are the same under PD and FD, and under PDIT and FDIT. As the number of regions increases, the social loss increases under the FD and FDIT regimes. Linearity is assumed for the purpose of expositional simplicity.

gure shows that scal decentralization, which is represented by a shift from the PD regime to the FD regime, increases the social losses of economic agents if n > 1. This is because the FD regime imposes an additional distortion in the form of the externality effect in determining d.19 The most important welfare implication in this paper is that even if scal decentralization in its isolation deteriorates social welfare, we may be able to improve social welfare by introducing an ination target when scal decentralization measures are adopted. As shown in the gure, as long as there are few regions or as long as the externality effect is small, the level of the social loss under the PD regime is larger than that under the FDIT regime

8. Conclusion Many studies have argued for the benets of an ination target to alleviate the problem stemming from time inconsistent monetary policies. However, these studies have not taken into account the fact that the design of scal policy may inuence the benets of an ination target. Our analysis shows that if a country moves to a scally decentralized regime that allows local governments to decide the level of local bonds, policy recommendations under a centralized scal policy may not be desirable. Ination targets should be designed to comply with scal policy, which requires more than the mere setting of a numerical target. Therefore, the benet of an ination target may not be warranted, as it is in typical models of ination targets that regard scal policy as given.

18 For expositional simplicity, linearity is assumed in the gure. See Appendix C for details of the derivation. 19 Note that this is a direct consequence of our modeling of scal decentralization as a delegation of authority to determine the level of local bonds from the central to the local governments. To focus on the direct effect of scal decentralization on the conduct of monetary policy, we abstract from other aspects of scal decentralization. To introduce those aspects is complicated, and is an interesting issue for future research.

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

187

Appendix A. Second best A.1. Equilibrium of SB regime The rst-order conditions for the social planner are summarized as follows:

B.1.2. First-period ination rate Second, we obtain the solution to the central banks determination of p1 in the rst period. The problem for the central bank is: min LCB ;
p1

bfp2 0; fp1 0;
1 @xi1 @ti1 1 @x @t x b xi2 i2 i2 ldi 0; n i1 @t i1 @di n @ti2 @di 1 @xi1 @ti1 x lei 0; n i1 @t i1 @ei i 1; . . . ; n; i 1; . . . ; n;

given (16), pe , di, and ei, and subject to (1) for i = 1, . . ., n and t = 1. 1 The rst-order condition is:

fp1 Si xi1

1 n

@xi1 0: @p1

This leads to:

where ldi and lei are the Lagrange multipliers for the constraints di ! 0 and ei ! 0. Under assumptions (4) and (5), we obtain the equilibrium represented in Table 1. A.2. Illustration of the determination of the amount of local bonds in the SB regime Rearranging (7) yields:

n pPD t 1 nt 1 : 1 f

(17)

B.1.3. Fiscal discipline The central banks choice of ination rate is rationally expected by the central and the local governments. The problem for the local government in region i is: min LLG ;

di g

n2
n

f1 rdi gg:

(15)

ei ji1;...;n

The left- and right-hand sides of this equation respectively represent @LC =@di and b@LC =@di . 1 2 Focusing on @LC =@di (the left-hand side of the equation), when di 1 is small, the derivative is negative so that an increase in di is desirable. Conversely, when di is large, it is positive so that an increase in di is undesirable. The intuition is as follows. An increase in di reduces ti1 and thereby increases xi1. On the one hand, if di < g, this is desirable because, in this region, the output gap xi1 is negative and the increase in xi1 is desirable. On the other hand, if di > g, the output gap xi1 is positive and the increase in di makes the gap larger. Hence, increasing di is not desirable in this region. With regard to the effect of di on LC , b@LC =@di is negative. This 2 2 implies that an increase in di unambiguously increases the secondperiod loss. In the second period, ti2 equals g + (1 + r)di and is positive. This implies that the second-period output gap xi2 is negative. Hence, an increase in di, which reduces xi2, is undesirable. The level of di is determined by balancing the two marginal effects of di on LC; that is, the marginal decrease in LC and the 1 marginal increase in LC . Because of assumption (4), dSB 0. 2 i Appendix B. Partial decentralization regimes B.1. Equilibrium of PD regime B.1.1. Second-period ination rate First, we examine how the central bank determines p2. The problem for the central bank in the second period is as follows: min LCB ; 2
p2

subject to (2), (3), and ei ! 0 for i = 1, . . ., n, (1) for i = 1, . . ., n and t = 1, 2, and given di. The rst-order condition is: xi1

@xi1 @ti1 d lei 0; @ti1 @ei

i 1; . . . ; n;

(18)

where lei is the Lagrange multiplier for the constraint ei ! 0. B.1.4. Local bond Now we turn to the central governments problem. The central government solves the following problem: min LC ;

di ji1;...;n

subject to (2), (3), di ! 0 for i = 1, . . ., n, (1) for i = 1, . . ., n and t = 1, 2, and given ei. The rst-order condition is:

@p2 t 2 ; p @t 2 1 @x @t 2 xi1 i1 i1 bfp2 t 2 ; p 2 n @ti1 @di @t 2 @di 1 @x @t b xi2 i2 i2 ldi 0 i 1; . . . ; n: n @t i2 @di

(19)

In this context, ldi is the Lagrange multiplier for the constraint di ! 0. B.1.5. Equilibrium Under assumptions (4) and (5), in equilibrium, we obtain ePD > 0 (lei = 0) and dPD > 0 (ldi = 0). Hence, these two assumptions i i support the existence of the equilibrium in which we are interested. The equilibrium levels of ePD and dPD are obtained as follows. i i The rst-order conditions, (18) and (19), are rewritten as: nng ei di d 0; i 1; . . . ; n; 1 1 1 nng ei di p2 t 2 n2 nng 1 rdi 0; n n n i 1; . . . ; n:

given pe , di, and ei, and subject to (1) for i = 1, . . ., n and t = 2. 2 The rst-order condition is:

fp2 Si xi2

1 n

@xi2 0: @p2

This condition leads to:

n pPD t 2 nt 2 ; 2 f
where an overline denotes average values over all regions.

(16)

Assuming symmetry and solving for ei and di leads to the results in Table 1.

188

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

B.2. Illustration of the determination of the local bond level in the PD regime Here we discuss how the two additional effects, the compensation effect and the cave-in effect, arise and change the determination of di in the PD regime. Given symmetry, rearranging Eq. (10), eliminating ei by introducing its rst-order condition (i.e., Eq. (9)), and eliminating p2 by introducing the rst-order condition for p2 (Eq. (8)) yield the equation to determine di in equilibrium:

B.3.2. The optimal target The optimal target level is obtained as follows. For the rstperiod target, its optimal level is determined by the following equation:

@LC @p1 1 fp1 0: @p1 @p 1

The optimal target is thus p d=f. 1 As for the second-period target, as Eqs. (24) and (25) show that ei di does not depend on p , the target only affects LC . The rst2 2 order condition for the determination of p is: 2

d
n

n2
n

Sf1 rdi g g:

(20)

First, we nd the compensation effect in the left-hand side of Eq. (20) that represents @LC =@di . Compared with Eq. (15), @LC =@di in 1 1 the PD regime is smaller due to a positive value of ei. This increases di from the second-best level, ceteris paribus. Note that the local governments set ei to offset the change in di, which can be conrmed by rewriting (9) as: nng ei di d 0; i 1; . . . ; n: (21)

@LC @p2 1 @LC @x2i @t2i @di 2 2 S 0: @p2 @p n i @x2i @t 2i @di @p 2 2

As @p2 =@p @t 2i =@di @di =@p 1=S; this equation leads 2 2 to,

The rst term of Eq. (10) is thus constant as shown in Eq. (20). Second, the cave-in effect affects the magnitude of b@LC =@di . 2 Because of this effect, the right-hand side of Eq. (20) is S times greater than that of Eq. (15). Thus, the cave-in effect reduces di, ceteris paribus. The cave-in effect can also be interpreted in the following manner. An increase in di increases ti2 and thereby increases the second-period ination rate through the second term of Eq. (8). This is not desirable because, without an ination target, the second-period ination rate is positive because of the lack of commitment (see Table 1). The effect thus increases the marginal cost of di and contributes to reducing the amount of local bonds. Under assumption (5), the left-hand side of Eq. (20) (=@LC =@di ) is 1 smaller than the right-hand side (=b@LC =@di ) at di = 0. Thus the 2 central government chooses a positive level of di. Note that the result dPD > dSB crucially depends on assumption (5). If i i 1 > b(1 + r), then dSB > 0 and dSB can be greater than dPD when d i i i is small and the cave-in effect dominates the compensation effect. Therefore, the main point of our result is not that dPD > dSB but that i i di has two opposing effects on LC in the PD regime. B.3. Equilibrium of PDIT regime

@LC 1 @LC @x2i 2 2 S 0; @p2 n i @x2i @t 2i

or 1 n

fp2 Si x2i n 0:

(26)

However, the rst-order condition for the determination of p2 in the second period, which leads to (22), is:

@LCB 1 @LCB @x2i 2 2 S 0; @p2 n i @x2i @p2

or 1 n

fp2 p Si x2i n 0: 2
Eqs. (26) and (27) imply that:

(27)

pPD 0: 2
Appendix C. Details of welfare comparison

B.3.1. The ination rate and the ination target By introducing an ination target in both periods, we obtain the following rst-order conditions:

n pPDIT t 2 ; p nt 2 p ; 2 2 2 f n PDIT p1 t 1 ; p1 nt 1 p : 1 f

(22) (23)

We then obtain the following levels of di and ei in equilibrium: ePDIT p 2 i 1 1 rS 1 1 r 1 d g p ; 1r 1 rS 2 n2 1 rS 1 (24)

This appendix explains the derivation of Fig. 1. We begin with the welfare comparison for the special case of n = 1. In this case, regimes FD and FDIT are equivalent to regimes PD and PDIT, respectively. Specically, using the fact that S = S0 , it is easy to establish that the rst-period output gap, x1, the second-period output gap, x2, and the second-period ination rates, p2, are the same for the PD, PDIT, FD, and FDIT regimes. With regard to the rst-period ination rate, we can conrm that: pPD pFD d=f > pPDIT pFDIT 0. Therefore, 1 1 1 1 the following relationship is established: LPDITC LFDITC < LPDC LFDC for n 1;

dPDIT p 2 i

n2 1 rS

1 1 g p : 1r 1 rS 2

(25)

These equations determine the levels of t 1 and t 2 , and recursively, the rates of ination as follows: 1 1

pPDIT 2 pPDIT 1

d p ; S 2 fS 1 d p : 1 f

where L(j)C denotes the equilibrium value of the social loss function of the country under regime j. We then consider what happens if n increases from unity. First, we can conrm that under regimes PD and PDIT, the output gaps and ination rates for both periods are the same as those when n = 1. This is because it is the central government that determines the amount of local government bonds by considering their average effect on national ination. There is no disagreement about the marginal benet of additional ination caused by an increase in the amount of bonds issued in the regions. Thus, the externality

H. Fujiki, H. Uchida / Japan and the World Economy 23 (2011) 178189

189

effect does not arise, even if there is more than one region. Hence, the values of the social loss function are the same when n = 1 and n > 1 in PD and PDIT. Second, under regimes FD and FDIT, the output gap and the ination rate in the rst period are the same as those when n = 1. However, the values of xFD , xFDIT , pFD and pFDIT depend on n. This is 2 2 2 2 because the cave-in effect becomes larger, or the externality effect 2 2 2 increases, as n increases. We can show that xFD , xFDIT , pFD , 2 2 2 FDIT 2 and p2 are all increasing in n. Hence, the equilibrium social loss is increasing in the number of regions, n, under both the FD and the FDIT regimes. However, as an ination target reduce the loss, @LFDC =@n > @LFDITC =@n. This implies that the greater the 2 2 number of regions, the greater the benet of imposing an ination target under the FD regime. References
Beetsma, R.M.W.J., Bovenberg, A.L., 1997. Central bank independence and public debt policy. Journal of Economic Dynamics and Control 21, 873894.

Beetsma, R.M.W.J., Bovenberg, A.L., 2001. The optimality of monetary union without a scal union, Journal of Money. Credit and Banking 33, 179204. Fujiki, H., Osano, H., Uchida, H., 2004. Optimal contracts for central bankers and public debt policy. Japanese Economic Review 55, 372400. Fujiki, H., Uchida, H., 2011. Ination Target and Debt Management of Local Government Bonds. Available at SSRN: http://ssrn.com/abstract=926974. Heenan, G., Peter, M., Roger, S., 2006. Implementing ination targeting: Institutional arrangements, target design, and communications. IMF Working Paper, WP/06/ 278. Kydland, F.E., Prescott, E.C., 1977. Rules rather than discretion: the inconsistency of optimal plans. Journal of Political Economy 85, 473492. Oates, W.E., 1972. Fiscal Federalism. Harcourt, New York. Svensson, L.E.O., 1997. Optimal ination targets, Conservative central banks and linear ination contracts. American Economic Review 87, 98114. Svensson, L.E.O., 2011. Ination targeting. In: Friedman, B., Woodford, M. (Eds.), Handbook of Monetary Economics, Vols. 3a and 3b, Elsevier. Tanzi, V., 2001. Pitfalls on the road to scal decentralization. Carnegie Endowment Working Papers, No. 19. Uchida, H., Fujiki, H., 2005. Optimal ination target under uncertainty. Japan and the World Economy 17, 470479. Walsh, C.E., 1995. Optimal contracts for central bankers. American Economic Review 85, 150167.