財政赤字會造成通貨膨脹嗎？動態追蹤資料的分量迴歸分析 - 政大學術集成

全文

(1)國立政治大學經濟學系碩士論文指導教授: 林馨怡博士. 政治 Is Fiscal Deficit Inflationary? 大. 立. A Dynamic Panel Quantile Analysis. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i Un. 研究生: 朱浩榜中華民國九十九年七月. v.

(2) 致謝詞辛苦了一年、遭遇許多不曾想到的困難, 論文總算順利完成, 而研究所的學業也將告一段落。能夠寫完這篇論文, 首先要感謝林馨怡老師的指導及訓練, 讓我在這一年寫論文的過程中成長學習、並得以一窺學術研究的堂奧。其次, 感謝口試委員陳旭昇老師與林常青老師提供寶貴的意見及可能改進的方向, 使得論文內容更加充實、完整。此外, 感謝素昧平生的 Raj M. Desai 博士願意慷慨提供資料, 使我的樣本得以涵蓋更多的國家及更長的時間。感謝佩玗學姊及東慶學長在 cwTEX、 R 的程式碼、以及口試事宜等各方面的幫助與經驗的傳承, 給我在寫論文的過程中帶來不少的幫助。感謝侯俊宇, 因為一起奮鬥及勉勵, 一. 治政文電子郵件的王安中、教我解決過 stata 難題的簡國安、大以及其他幫助或鼓勵過我的研究立所同學們, 讓我這兩年的學業得以更加順利。感謝龔恩緯, 一直以來的好朋友, 無論發生什. 路走來才得以共同克服許多難關。感謝從碩一以來就互相照應的李宜謙、常常幫我訂正英. ‧ 國. 學. 麼事都可以討論分享、加油打氣。. 此外, 感謝伍芬婕與宜凌學姊, 願意花時間且耐性地校改我寫的英文, 讓我受益良多。. ‧. 若論文仍有文意表達或拼字文法上的錯誤, 當屬我的責任。其他還要感謝國高中的朋友、大學外交系的同學、政大崇德社的諸位, 以及其他每一位支持或鼓勵過我的人。. y. Nat. sit. 最重要的, 就是感謝我的父母與家人一直以來的支持與關懷、阿姨姨丈的照顧、以及各. er. io. 位親人的關心與鼓勵, 我才可以順利到達這個階段。最後也要謝謝上天保佑, 讓我能平安. n. 順利地完成學業。在政大待了六年 a , 現在終於到離開的時候了。v這一段里程的結束, 也標誌. l. i. C h , 再次由衷感謝每一位幫助過我的人著下一段里程的開始。在踏出校園之前 ! Un engchi. 朱浩榜民國 99 年 7 月 25 日.

(3) Abstract In economic theory, sustained fiscal deficits might cause inflation by means of money creation, and the economy in a higher inflation level would be more strongly impacted by an increase in deficits. Following the theoretical model of Catão and Terrones (2005), I scaled fiscal deficits by narrow money stock and examined the deficit-inflation relationship in 91 countries from 1960 to 2006. A dynamic panel quantile regression of Lin (2010) was employed, which can estimate the impact of fiscal deficits at various inflation levels and allows for a dynamic adjustment. The empirical results show that fiscal deficits will be more serious as inflation rises, and. 政治大 would be more effective in price 立 stabilization the higher the inflation. Moreover, the. weakly or not related to inflation if it is at a low level. Therefore, fiscal consolidation. ‧ 國. 學. results remain robust while taking other possibly inflation-related factors into consideration. Furthermore, the impact of fiscal deficits on inflation is generally greater in developing countries, particularly when inflation is at a high level. Finally, the. Nat. y. ‧. inflationary effect of deficits is not detected over 1990–2006.. n. al. er. io. panel data. sit. Keywords: Fiscal deficit; Inflation; Quantile regression; Price stabilization; Dynamic. Ch. engchi. i Un. v.

(4) Contents 1 Introduction. 1. 2 Literature Review. 4. 3 Econometric Methodology. 3.2. Quantile regression and endogeneity . . . . . . . . . . . . . . . . . . . 12 3.1.1. The model and estimation of quantile regression . . . . . . . . 12. 3.1.2. Endogenous problems in quantile regression . . . . . . . . . . 13. Quantile regression for panel data . . . . . . . . . . . . . . . . . . . . 15 Panel data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15. 3.2.2. 16. 3.2.3. 17. 學. 3.3. 治政 The model and estimation . . . . . 大 . . . . . . . . . . . . . . . 立 Large sample properties . . . . . . . . . . . . . . . . . . . . .. 3.2.1. ‧ 國. 3.1. 12. Dynamic panel quantile regression . . . . . . . . . . . . . . . . . . . . 18 The IVQR approach . . . . . . . . . . . . . . . . . . . . . . . 18. 3.3.2. The fitted value approach . . . . . . . . . . . . . . . . . . . . 20. ‧. 3.3.1. 22. sit. y. Nat. 4 Empirical Results. 4.2. Cross-sectional results . . . . . . . . . . . . . . . . . . . . . . . . . . 28. 4.3. Dynamic panel results . . . . . . . . . . . . . . . . . . . . . . . . . . 31. n. al. er. Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22. io. 4.1. Ch. engchi. i Un. v. 4.3.1. The baseline analysis . . . . . . . . . . . . . . . . . . . . . . . 31. 4.3.2. The extensive analysis . . . . . . . . . . . . . . . . . . . . . . 36. 4.3.3. The country group-specific analysis . . . . . . . . . . . . . . . 41. 4.3.4. The subsample period analysis and central bank independence 43. 5 Conclusions. 47. References. 75. A List of countries. 79 I.

(5) B Data sources and descriptions. 82. C List of country groups (1960–2006). 83. D List of countries with data of exchange rate regime (1960–2006). 84. E List of countries with data of central bank independence (1990– 2000). 85. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. II. i Un. v.

(6) List of Tables 1. Descriptive statistics of selected countries . . . . . . . . . . . . . . . . 23. 2. Descriptive statistics of specific income groups (1960–2006) . . . . . . 25. 3. Descriptive statistics of OECD and non-OECD countries (1960–2006) 26. 4. Cross-sectional results over 1960–2006 . . . . . . . . . . . . . . . . . . 48. 5. Dynamic panel results over 1960–2006 I (A) . . . . . . . . . . . . . . 50. 6. Dynamic panel results over 1960–2006 I (B) . . . . . . . . . . . . . . 51. 7. Dynamic panel results over 1960–2006 I (C) . . . . . . . . . . . . . . 52. 8. Dynamic panel results over 1960–2006 II (A) . . . . . . . . . . . . . . 54. 9. Dynamic panel results over 1960–2006 II (B) . . . . . . . . . . . . . . 55. 10. Dynamic panel results over 1960–2006 III (A) . . . . . . . . . . . . . 57. 11. Dynamic panel results over 1960–2006 III (B) . . . . . . . . . . . . . 58. 12. Dynamic panel results over 1960–2006 IV (A) . . . . . . . . . . . . . 60. 13. Dynamic panel results over 1960–2006 IV (B) . . . . . . . . . . . . . 61. 14. Dynamic panel results of high-income countries over 1960–2006 . . . . 63. 15. Dynamic panel results of middle- and low-income countries over 1960–. 立. 政治大. ‧. ‧ 國. 學. y. Nat. sit. 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64. 17. Dynamic panel results of non-OECD countries over 1960–2006 . . . . 67. 18. Dynamic panel results over 1970–2006. . . . . . . . . . . . . . . . . . 69. 19. Dynamic panel results over 1980–2006. . . . . . . . . . . . . . . . . 70. 20. Dynamic panel results over 1990–2006. . . . . . . . . . . . . . . . . . 72. 21. Dynamic panel results over 1990–2000. . . . . . . . . . . . . . . . . . 73. n. al. . . . . . . 66. er. Dynamic panel results of OECD countries over 1960–2006. io. 16. Ch. e n g c h i.. III. i Un. v.

(7) List of Figures 1. Cross-sectional results over 1960–2006 . . . . . . . . . . . . . . . . . . 49. 2. Dynamic panel results over 1960–2006 I . . . . . . . . . . . . . . . . . 53. 3. Dynamic panel results over 1960–2006 II . . . . . . . . . . . . . . . . 56. 4. Dynamic panel results over 1960–2006 III . . . . . . . . . . . . . . . . 59. 5. Dynamic panel results over 1960–2006 IV . . . . . . . . . . . . . . . . 62. 6. Dynamic panel results of high-income and middle- and low-income countries over 1960–2006 . . . . . . . . . . . . . . . . . . . . . . . . . 65. 7. Dynamic panel results of OECD and non-OECD countries over 1960–. 政治大. 2006 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 8. Dynamic panel results over 1970–2006 and 1980–2006 . . . . . . . . . 71. 9. Dynamic panel results over 1990–2006 and 1990–2000 . . . . . . . . . 74. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. IV. i Un. v.

(8) 1 Introduction In either economics or policy discussions, the topic of whether fiscal deficits cause inflation is intriguing. In macroeconomic theory, Sargent and Wallace (1981) argued that an economy might be dominated by monetary authority or fiscal authority. If an economy is dominated by monetary authority, then fiscal authority will face a budget constraint imposed by monetary authority when fiscal policy is formulated. Monetary policy can be independently implemented. Hence, money growth can be controlled and inflation will not be caused. However, if fiscal authority dominates an economy, the monetary authority cannot implement monetary policy independently. 政治大 creation, and inflation rises consequently. Therefore, in a “fiscal dominance” econ立 omy, sustained fiscal deficits will lead to inflation. Furthermore, Catão and Terrones. and would be forced to accommodate sustained fiscal deficits by means of money. ‧ 國. 學. (2005) argued that an economy in a higher inflation level would be impacted by an increase in deficits more strongly, because its inflation tax base is typically narrower.. ‧. They also interpreted that the government can allocate seigniorage intertemporally. y. Nat. by borrowing, so budget deficits do not have to induce inflation contemporaneously.. sit. Alternatively, the conventional view in terms of Keynesian aggregate demand. er. io. considered that an increase in government debt has a wealth effect on households, so. al. n. iv n C an increase in aggregate demand will price level and inflation. In addition, h eraise n gthe chi U the income will raise and the demand for goods and services will increase. Therefore,. a recently developed fiscal theory of the price level (FTPL) argued that the price level is jointly determined by fiscal and monetary policy, and equilibrium may not be as unique. For investigating the deficit-inflation relationship, I used a dynamic panel quantile regression in 91 countries from 1960 to 2006. There are two reasons for using quantile regression. The first motivation is that in the theoretical model of Catão and Terrones (2005), fiscal deficit is scaled by narrow money stock which stands for an inflation tax base. Consequently, given a change in the deficit-to-GDP ratio, 1.

(9) fiscal deficits would be more inflationary in a higher-inflation economy, because its inflation tax base is typically narrower. The second reason is that previous empirical studies discovered that a fiscal deficit is generally inflationary in high-average inflation countries, high-inflation periods and developing countries. Otherwise, deficits may play a weak or even non-role in the determination of inflation. However, highaverage inflation countries and high-inflation periods are classified arbitrarily in previous studies. Accordingly, quantile regression can estimate the inflationary effects of fiscal deficits at various inflation levels, and the inflation levels do not need to be arbitrarily classified. Panel data provides plenty of observations across countries over a long time. 政治大 regression of Lin (2010), which is a two-stage fitted value approach, is employed to 立 horizon and allows for intrinsic dynamic adjustment, and dynamic panel quantile. can be examined clearly and comprehensively.. 學. ‧ 國. estimate the deficit-inflation linkage. Accordingly, the deficit-inflation relationship. The findings of my study are that fiscal deficits will be more inflationary the. ‧. higher the inflation rate, and will weakly or not be related to inflation when inflation. y. Nat. is at a low level. Taking one or more lagged deference dependant variables as. io. sit. instruments will not change the results. Therefore, fiscal consolidation would be. er. more effective in price stabilization as inflation rises higher. In addition, scaling. al. n. iv n C the estimates become significant h as e inflation at ia U n g c h low level.. deficits by GDP and controlling money growth, the results are similar except that Secondly, the results. remain robust when taking other possibly inflation-related factors into consideration (growth of GDP per capita, oil price inflation, openness and exchange rate regime), so the estimated deficit-inflation relationship is stable. Thirdly, the impact of fiscal deficits on inflation is generally greater among developing countries (represented by middle- and low-income countries and non-OECD countries), especially as inflation is at a high level. These findings support the theoretical model of Catão and Terrones (2005) — the economy in a higher inflation level would be impacted by fiscal deficits. 2.

(10) more strongly — and are consistent with previous empirical studies (e.g. De Haan and Zelhorst, 1990; Fischer et al., 2002; Catão and Terrones, 2005). Finally, the deficit-inflation relationship does not notably change during 1960–2006, 1970–2006 and 1980–2006, but it is not detected over 1990–2006. The remainder of the thesis is organized as follows. Section 2 discusses the literature review. Section 3 presents the econometric model. The data description and empirical results are reported in Section 4, and Section 5 offers the concluding remarks.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 3. i Un. v.

(11) 2 Literature Review Whether fiscal deficits will raise inflation is an intriguing issue that may be discussed by economists. In theory, Sargent and Wallace (1981) proposed an analytical model to cover this topic. There are two different schemes in their framework: “monetary dominance” and “fiscal dominance.” The former indicates that monetary authority could implement monetary policy independently. The budget deficit is jointly determined by bonds sales to the public and seigniorage created by the monetary authority, so fiscal authority will face a budget constraint imposed by monetary authority when it formulates the fiscal policy. Therefore, monetary authority can. 政治大 fiscal dominance indicates that monetary authority is dominated by fiscal authority. 立 In this scheme, fiscal authority does not care budget balance when fiscal policy is. control the money supply and inflation rates. Contrary to monetary dominance,. ‧ 國. 學. formulated. However, the demand of government bonds is limited, and the interest rate of government bonds will increase when there are too many bonds for sale.. ‧. The interest rate could not be greater than the economic growth rate — otherwise. y. Nat. government debts would grow faster than real income and render the economy to. sit. become unstable. Therefore, even though the monetary authority wants to control. er. io. money growth, yet, it will still be forced to accommodate the bonds with additional. al. n. iv n C long run and the inflation rate willhrise consequently. e n g c h i U Accordingly, fiscal dominance base money. Ultimately, monetary authority cannot control money growth in the. supports the hypothesis that budget deficits lead to inflation, but monetary dominance does not. Furthermore, Catão and Terrones (2005) scaled fiscal deficits by narrow money which stands for an inflation tax base, and they argued that an economy in a higher inflation level would be impacted by an increase in deficits more strongly because its inflation tax base is typically narrower. They also interpreted that the deficitinflation relationship is dynamic under a fiscal dominance scheme. Because the government can allocate seigniorage intertemporally by borrowing, budget deficits 4.

(12) do not have to induce inflation currently. However, budget deficits play a key role in the present value of future money accommodation (for financing government bonds), so deficits can still ultimately lead to inflation. Therefore, deficits are inflationary in the long run, but not necessarily in the short run. Different from Sargent and Wallace (1981), the conventional view of debt provides another channel in terms of Keynesian aggregate demand to interpret why an increase in debt may cause inflation. In the main idea of the conventional view, Elmendorf and Mankiw (1999) concluded that an increase in debt has a positive wealth effect on households, so the demand for goods and services will raise and inflate the economy.. 政治大 level can be determined by fiscal policy (debt). In a “non-Ricardian” case, both 立 In addition, the fiscal theory of the price level (FTPL) also claims that the price 1. fiscal and monetary policy, which determine the government’s future primary sur-. ‧ 國. 學. pluses, are exogenously determined by the government itself. When the government adjusts the present value of its future primary surpluses lower than the real value of. ‧. the debt, the price level will rise to lower the real value of the debt. Then, solvency. y. Nat. at a new real value will be produced and the real debt will devaluate (Minford and. io. sit. Peel, 2002). Under the FTPL, fiscal policy is directly linked to the price level though. er. the present value budget constraint.. al. n. iv n C h e n gFor in specific country groups and periods. United States, c hthei U. Empirical results about the connection between fiscal deficits and inflation vary Hamburger and. Zwick (1981) examined the deficit-money linkage from 1954 to 1976. They concluded that budget deficits are broadly inflationary. In particular, the deficit-money linkage becomes stronger in the “Keynesian period” (1961–1974). This is due to an expansionary fiscal policy and a following interest-rate moderating monetary policy. Dwyer (1982) used quarterly data covering 1953–1978 to test the relationships 1. FTPL is developed by Leeper (1991), Sim (1994) and Woodford (1994, 1995). The recent developments of FTPL, see, for example, Woodford (2001), McCallum (2001), Cochrane (2001, 2005) and Leeper and Yun (2006).. 5.

(13) between debt, price and money with a vector autoregression (VAR) model. However, there is no evidence that debts held by the public and by the Federal Reserve play a role in determining the price level and other macroeconomics variables such as interest rates and the money stock. In his results, fewer deficits would not lower the inflation rates. Darrat (1985) investigated whether budget deficits and money growth will impact inflation. He took both budget deficits and money growth into consideration, because he regarded deficits as a non-monetary factor. He showed that both budget deficits and money growth are significantly inflationary from 1958 to 1979. Similarly, Ahking and Miller (1985) examined the relationships between deficits,. 政治大 treat all variables as endogenous. They separated the quarterly data (1947–1980) 立. money growth and inflation, which were estimated in a VAR framework so as to. into three decades (1950s, 1960s and 1970s) for comparison. Budget deficits were. ‧ 國. 學. uncovered which caused inflation during 1950s and 1970s. They provided evidence that the deficit-inflation linkage of the United States does exist during some specific. ‧. periods. However, this effect is independent of money growth, which implies that. y. Nat. inflation is not due to monetization.. io. sit. King and Plosser (1985) also investigated the deficit-seigniorage relationship in. er. terms of neoclassical macroeconomic models. They estimated the connection by. al. n. iv n C h e n g cInhaddition fiscal deficits and seigniorage in 1953–1982. i U to the United States, they. both ordinary least squares (OLS) and VAR, but found little connection between. also estimated the deficit-seigniorage connection of other 12 industrial and develop-. ing countries, but still failed to demonstrate that the relationship is broadly significant. Other than the United States, there are also many empirical research studies about other industrial countries. Other than the aforementioned King and Plosser (1985), Giannoros and Koulluri (1986) utilized data from 1950 to 1981 to examine whether deficits lead to money growth and inflation in 10 industrial countries. The. 6.

(14) results showed that the impact of budget deficits on money supply and inflation was insignificant. Like Giannoros and Koulluri (1986), Protopapadakis and Siegal (1987) also examined the debt-money and the debt-inflation connection for 10 major advanced countries during 1952–1987. They applied non-parametric and regression tests, and interpreted that there is no association between debt and money growth, and the association between debt growth and inflation is very weak. Barnhart and Darrat (1988) checked the causality between fiscal deficits and money growth across seven industrial countries from 1960 to 1984. Their study rejected the hypothesis that deficits Granger-cause an increase in the money growth.. 政治大 Therefore, there is no general relationship between deficits and 立. Reversely, an increase in money growth does not Granger-cause an increase in deficits, either.. money growth, and fiscal deficits are not inflationary.. ‧ 國. 學. Since the studies about industrial countries concluded that the inflationary effect of budget deficits was broadly insignificant, some economists considered that the ef-. ‧. fect may be more significant in developing countries. De Haan and Zelhorst (1990). y. Nat. did a search about developing countries and concluded that government debt may. io. sit. induce money creation in some channels. First, political pressures may force mon-. er. etary authorities to finance budget deficits, especially when the central bank is not. al. n. iv n C heng to a higher optimal level of seigniorage. Third, the c h i Utime-inconsistency theory sugsufficiently independent. Second, less efficiency and ability of taxation would lead. gests that the government has a motive to generate unexpected inflation in order to. decrease the real value of interest-bearing debt, and it implies that the government could get a capital gain from unexpected inflation. The final channel is the aforementioned “fiscal dominance” hypothesis proposed by Sargent and Wallace (1981). Among these theories, De Haan and Zelhorst (1990) considered that the “fiscal dominance hypothesis” is the most adequate case for developing countries. Empirically, they collected data from 17 developing countries during 1961–1985 and estimated. 7.

(15) the effect of deficits on money growth with a VAR model. Unfortunately, they did not provide strong evidence to support the fiscal dominance hypothesis. However, they discovered that budget deficits are positively related to inflation during acute inflation periods with a nonparametric method. Metin (1998) did a system cointegration analysis based on Turkish data during 1954–1986 and applied an error-correlation model to estimate the relationship between budget deficits and inflation. He uncovered that deficits lead to inflation directly in Turkey, and the current real income growth had a negative effect on inflation. Including fiscal imbalances, output gaps, supply-side cost shock and inflation. 政治大 in 53 developing countries from 1964 to 1998. They showed that the fiscal balance 立 persistence, Loungani and Swagel (2003) generally discussed inflationary factors. weakly correlates to inflation, but the correlation becomes stronger in higher-average. ‧ 國. 學. inflation countries. In addition, they found a non-linear relationship between a deficit and inflation. The impact of deficits on inflation is significant when the deficit-. ‧. to-GDP ratio is above 5%. Additionally, they used money growth and exchange rate. factors varies under different exchange regimes.. io. sit. y. Nat. movements to represent fiscal factors and found that the relative importance of two. er. For the specific groups of developing countries, there are also some empirical. al. n. iv n C h einflation nection between budget balances and monthly data of three transition i U n g c hwith. researches as follows. Komulainen and Pirttilä (2002) utilized VAR to test the con-. economies (Russia, Bulgaria and Romania). But generally, deficits did not play an inflationary role. Doma¸c and Y¨ ucel (2005) investigated the determining factors of high inflation episodes in 15 emerging markets from 1980 to 2001. Employing a pooled probit model, Doma¸c and Y¨ ucel discovered that government deficits are positively significant, so expansionary fiscal policy most likely launched inflation episodes. Moreover, Coll and Pedauga (2007) placed their focus on 18 Latin American countries during 1980–2004. They took institutional and economic structural. 8.

(16) factors to explain why inflation in Latin America declined steeply in the 1990s. Their dynamic generalized method of moments (dynamic GMM) results revealed indebtedness as an inflationary factor, although it was not very robust. Meanwhile, growth in GDP per capita was negatively related to inflation. Baldacci et al. (2004) researched “expansionary fiscal contraction” in low-income countries. Expansionary fiscal contraction means that a sustained decrease in fiscal deficits will increase the real income level. They examined the influence of reductions in fiscal deficits on other macroeconomic variables with a panel dataset in 39 low-income countries from 1990 to 2001. Their results presented that in indebted countries, fiscal deficits are inflationary and harmful to economic growth. However,. 政治大 Therefore, expansionary fiscal contraction is a useful policy only for high-deficit 立. in less indebted countries, fiscal deficits are insignificant to inflation and growth.. countries.. ‧ 國. 學. To investigate the impact of deficits generally, Karras (1994) applied a panel data model and GLS method in 32 developed and developing countries during 1950s–. ‧. 1980s. In his conclusion, the expansionary effect of fiscal deficits on money growth. y. Nat. is insignificant. It means that deficits are not be monetized. In addition, the main. io. sit. determinant of inflation was money growth rather than fiscal deficits, and an increase. er. in deficits lead to a reduction in output growth and investment.. al. n. iv n C U However, he provided no h efrom cross-sectional data of 90 countries n g1971 c htoi 1990.. Checking the deficit-seigniorage linkage, Click (1998) used OLS for estimating. evidence which indicated that an increase in domestic debt will cause seigniorage to rise. Cottarelli et al. (1998) broadly discussed some non-monetary determinants of inflation in 47 countries from 1993 to 1996. They pointed out that based on revenue motives, the fiscal budget is a determinant of inflation. It induces the central bank to create seigniorage or inflation tax, especially when the budget is imbalanced or the financial market is less developed. Additionally, the past inflation rates influence. 9.

(17) current inflation due to its own persistence and inertia, so it used dynamic GMM as an empirical model. The result showed that fiscal deficits play a significant role in inflation, particularly when the countries’ securities markets do not develop well. Fischer et al. (2002) generally investigated the relationship between inflation, money growth, seigniorage and fiscal deficits. Their dataset is large and contains 94 countries from 1960 to 1995. According to the cross-sectional results, the fiscal deficit is positively significant to seigniorage and inflation. It indicates that deficits give the government an incentive to create seigniorage and ultimately inflate the economies. For observing shot-run effects, they employed a fixed effect model to analyze panel data. In addition, they classified countries into two groups according to. 政治大 flation countries, can fiscal deficits play a significant role in seigniorage and inflation. 立. their long-run average inflation rates. They discovered that only in high-average in-. Otherwise, fiscal deficits have no effect. Furthermore, they selected high-inflation. ‧ 國. 學. episodes in high-average inflation countries and found that in high-average inflation countries, fiscal deficits are positively related to inflation during high-inflation. ‧. episodes, but otherwise they are insignificant.. y. Nat. Catão and Terrones (2005) modeled a new approach (fiscal deficit is scaled by. io. sit. narrow money) to theoretically prove that persistent fiscal deficits will lead to infla-. er. tion, and an economy in a higher inflation level will be impacted by an increase in. al. n. iv n C the econometric model they used h is e a mean group n g c h i U(or a pooled mean group) esti-. deficits more strongly. They collected data from 107 countries over 1960–2001, and. mator, which could model a non-linear relationship, allow for heterogeneity across countries, and reveal the short- and long-run influences. They showed how budget deficits can positively relate to inflation, but the inflationary effect of deficits depend on the market’s financial depth, inflation tax bases and the credibility of monetary authorities. Additionally, a fiscal deficit is inflationary in developing and high-inflation countries, but not in low-inflation and developed countries. Kwon et al. (2009) examined the debt-inflation connection with a forward look-. 10.

(18) ing fiscal-monetary model of inflation, and their model nested the quantity theory of money and the theory of Sargent and Wallace (1981). Because the model takes into account forward looking expectations, multiple equilibrium paths can coexist. They used the data of 71 countries from 1962 to 2004 and a dynamic GMM model to measure the effect of debt on inflation. The results showed that debt growth is strongly inflationary in indebted developing countries, and less strong in other developing countries. In the case of advanced countries, debt growth is less inflationary. As an important determinant of inflation, the budget deficit is also considered in many inflation related research studies. For example, Desai et al. (2005) analyzed the relationship between inflation and inequality in 120 countries covering 1960–2000. 政治大 In their result, the fiscal deficit is positively associated with 立. through a political structure channel, and they considered that fiscal balance is an important variable.. inflation except in advanced countries. Alfaro (2005) studied the role of openness in. ‧ 國. 學. inflation, and she also controlled the effect of fiscal deficits. Although the robustness of deficits is not strong, the coefficient is always positive.. ‧. In light of aforementioned theoretical and empirical researches, I conclude that. y. Nat. a fiscal deficit is generally inflationary in high-average inflation countries, high-. io. sit. inflation periods and developing countries. Otherwise, deficits may just play a weak. er. or even nonexistent role in the determination of inflation. This means that high-. al. n. iv n C h e n g chypothesis. idence for supporting the “fiscal dominance” hi U. inflation periods, high-average inflation and developing countries provide strong ev-. 11.

(19) 3 Econometric Methodology In this section, quantile regression and how to deal with endogenous problems will be introduced. This will be followed by a discussion of quantile regression for panel data. Finally, the estimation of a dynamic quantile regression for panel data is described.. 3.1 Quantile regression and endogeneity 3.1.1 The model and estimation of quantile regression Quantile regression is an econometric technique which can estimate the parameters. 政治大. at a specific quantile of the population.2 Contrary to OLS method, quantile re-. 立. gression provides different estimates at various quantiles, and OLS only provides an. ‧ 國. 學. average estimate of the population.. A classic quantile regression model can be written as follows: 0. ‧. 0. yi = xi β(τ ) + εi (τ ) or Qyi (τ |xi ) = xi β(τ ),. Nat. sit. y. where yi is a dependent variable, Qyi (τ |xi ) is the τ th quantile of yi conditional on. al. er. io. xi , εi (τ ) is an error term and Qεi (τ |xi ) = 0, xi is an explanatory variable, and β(τ ). n. is the interesting parameter at the τ th quantile.. Ch. i Un. v. Then, we minimize the following objective function at a given τ , and the estiˆ ) can be obtained. mator β(τ X. engchi. {i:yi ≥x0i β}. =. n X. X. τ |yi − x0i β| +. (1 − τ )|yi − x0i β|. (1). {i:yi <x0i β}. ρτ (yi − x0i β),. i=1. where ρτ (yi − x0i β) = (yi − x0i β)(τ − 1{yi −x0 β<0} ) is a check function, and 1A is a i indicator function. If condition “A” holds, then 1A is equal to 1. If “A” does not 2. Quantile regression is proposed by Koenker and Bassett (1978).. 12.

(20) hold, then 1A is equal to 0. Thus, if yi − x0i β ≥ 0, then ρτ (yi − x0i β) = τ (yi − x0i β); if yi −x0i β < 0, then ρτ (yi −x0i β) = (τ −1)(yi −x0i β). In equation (1), residual terms are positive in the former because observations are greater than estimates, and given a weight τ . In the latter, residual terms are negative because observations are smaller than the estimates, and are given a weight (1 − τ ). The large sample properties of quantile regression can be shown as √. ˆ ) − β(τ )) → N(0, J −1 SJ −1 ), n(β(τ. where J = ∇β(τ ) IE[ϕ(xi , yi , β(τ ))] = −IE[xi x0i fεi (τ )|xi (x0i β(τ ))] and. 政治大. S = IE[ϕ(xi , yi , β(τ ))ϕ(xi , yi , β(τ ))0 ] = τ (1 − τ )IE(xi x0i ),. 立. where fεi (τ )|xi (.) is the conditional probability density function of the error term. ‧. ‧ 國. Therefore,. 學. εi (τ ).. J −1 SJ −1 = τ (1 − τ )IE(xi x0i fεi (τ )|xi (x0i β(τ )))−1 IE(xi x0i )IE(xi x0i fεi (τ )|xi (x0i β(τ )))−1 (2) .. y. Nat. sit. If fεi (τ )|x (.) = fεi (τ ) (.), this means that the probability density function of the error. n. al. er. io. term εi (τ ) is independent of xi , and equation (2) can be simplified as. Ch. J −1 SJ −1 =. i Un. v. τ (1 − τ ) IE(xi x0i )−1 . [fεi (τ ) (0)]2. engchi. 3.1.2 Endogenous problems in quantile regression How can an endogenous problem generated in the quantile regression be interpreted at first. Consider an endogenous regression model: 0. 0. 0. 0. yi = di α0 + xi β0 + (di α1 + xi β1 )εi (τ ),. (3). where di is an endogenous variable, xi is an exogenous variable, and εi (τ ) is an error term. α0 , α1 , β0 and β1 are vectors of parameters. Then, the τ th conditional 13.

(21) quantile function of equation (3) would be 0. 0. Qyi (τ |di , xi ) = di α(τ ) + xi β(τ ),. (4). −1 −1 where α(τ ) = α0 + α1 F−1 εi (τ |di , xi ), β(τ ) = β0 + β1 Fεi (τ |di , xi ), and Fεi (τ |di , xi ). refers to the τ th conditional quantile of εi . Therefore, equation (4) indicates the τ th conditional quantile of yi . Furthermore, if τ is a random variable with uniform distribution, yi can be rewritten as 0. 0. yi = di α(ui ) + xi β(ui ), where ui |di , xi ∼ Uniform(0, 1) and ui represent unobserved factors which have an impact on yi .. 立. 政治大. Assume that di = δ(xi , zi , vi ), where δ is a unknown function, vi is an unob-. ‧ 國. 學. served disturbance which is correlated with ui , and zi is a valid instrument which is independent of xi and zi . Accordingly, vi influence di through δ because vi is. ‧. correlated with ui . Thus, there is an endogenous problem in di , and the estimators of yi on (di , xi ) would be biased and inconsistent. The instrumental variable zi can. sit. y. Nat. be utilized to solve endogenous problems.. io. er. The “fitted value” approach which consists of two stages is a solution, and it is developed by Amemiya (1982) and Powell (1983). Assume that zi is a valid. n. al. Ch. i Un. v. instrument variable for di , and we can run OLS of di on zi to get its OLS fitted value dî as the first step. After getting dî , we can consider a new quantile regression. engchi. model 0 0 yi = dî α(τ ) + xi β(τ ) + εi (τ ).. (5). Then, we run a quantile regression of yi on (dî , xi ) in equation (5), and the consistent estimators α ˆ f v (τ ) and βˆf v (τ ) can be obtained. In addition, the instrumental variable quantile regression (IVQR), which is proposed by Chernozhukov and Hansen (2005, 2006), is another solution to deal with endogenous problems for quantile regression. The procedure of IVQR is as follows. 14.

(22) First, we run OLS of di on (zi , xi ), and we will get its least squares projection φî . Consider the following model:3 0 0 0 yi − di α(τ ) = xi β(τ ) + φî γ(τ ) + εi (τ ).. (6). Second, define a grid of αj where j = 1, . . . , J and plug it back into equation (6). 0. 0. Next, take (yi − di αj ) as a new regressand, run a quantile regression of (yi − di αj ) on (xi , φî ), and search the α ˆ j which makes kˆ γ (ˆ αj )k minimized. And then, the estimators α ˆ CH,j (τ ) and βˆCH (ˆ αCH,j (τ ), τ ) are the results of what we are looking for. Let θˆCH (τ ) = (ˆ αCH (τ ), βˆCH (ˆ αCH (τ ), τ )),. 政治大. Chernozhukov and Hansen (2006) showed large sample properties of IVQR as. 立. n(θˆCH (τ ) − θCH (τ )) → N (0, JCH (τ )−1 SCH (τ )JCH (τ )−1 ).. (7). 學. ‧ 國. √. In equation (7), they suggested that JCH (τ ) and SCH (τ ) can be estimated as n 1 X 0 1(|ˆεi (τ )|≤Hn ) [φî , x0i ]0 [d0i , x0i ] and 2nHn i=1. al. sit. io 0. n X. 0 0 [φî , x0i ]0 [φî , x0i ],. i=1. er. Nat. 1 SˆCH (τ ) = τ (1 − τ ) n. y. ‧. JˆCH (τ ) =. v. n. where εî (τ ) = yi − di α ˆ CH (τ ) − x0i βˆCH (τ ), and Hn is a kernel bandwidth.. Ch. en. gc 3.2 Quantile regression for panel data. hi. i Un. 3.2.1 Panel data The panel data is a dataset where cross-sectional observations are observed over multiple time periods, and hence panel data contains properties of both cross-section and time-series. A typical panel data model is yit = ηi + x0it β + εit , 3. We can also use zi to substitute the projection φî in equation (6).. 15. (8).

(23) where i = 1, . . . , n and t = 1, . . . , T . i represents different observations, and T represents different time periods. yit is an dependent variable which belongs to an individual i at time T , xit is a vector of explanatory variables, and εit is an independent and identically distributed error term. ηi is a time-invariant individual effect, and how to deal with individual effect is an important econometric issue. If ηi is a fixed term, equation (8) is a fixed effect model. If ηi is a random term, equation (8) is a random effect model. 3.2.2 The model and estimation In a quantile regression for panel data, if an individual effect ηi is fixed, Koenker. 政治大. (2004) proposed a method for eliminating fixed effects. Consider a panel quantile. 立. regression model 0. 0. ‧ 國. 學. yit = ηi + xit β(τ ) + εit (τ ) or Qyit (τ |ηi , xit ) = ηi + xit β(τ ). Like equation (1), the objective function is. Nat. sit. k=1 t=1 i=1. (9). y. 0. ωk ρτk (yit − ηi − xit β(τk )).. ‧. q T X n X X. er. io. Where k = 1, . . . , q, and k represents various quantiles. ρτk (.) is a check function. al. n. iv n C quantile. However, when the dimensions U q are too large, solving for h e n gofcn,h Ti and equation (9) is difficult.. as in equation (1). ωk is a weight that controls the relative influence of the τk th. Koenker (2004) proposed a shrinkage method to eliminate fixed effects ηi , and he considered the panelized version of equation (9): q n T X X X. 0. ωk ρτk (yit − ηi − xit β(τk )) + λ. k=1 t=1 i=1. n X. |ηi |.. i=1. When λ → 0, the panel quantile regression estimators can be obtained.4 4. λ is a tuning parameter, and λ is chosen as 1 in practice.. 16. (10).

(24) 3.2.3 Large sample properties First, we impose conditions A1–A3 as follows. A1 : yit is independent of the conditional distribution functions Fit with differentiable conditional densities 0 < fit < ∞, and the derivatives fit0 are bounded at 0. ξit (τ ), where ξit (τ ) = ηi + xit β(τ ). A2 : D0 and D1 are positive definite:   √ 0 0 0 0 ω ΩωI I ω ΩW ⊗ I X/ n  and D0 = lim T −1  √ 0 0 n,T →∞ W Ωω ⊗ X I/ n W ΩW ⊗ X X/n √ 0 0 ωk I Φ k I ω1 I Φ1 X/ n √ √ 0 0 ω1 X Φ1 I/ n ω1 X Φ1 X/ n .. .. . . √ 0 0 ωq X Φq I/ n.  P. 政治大. 立. 學. ‧ 國.    D1 = lim T −1  n,T →∞  . √  0 ωq I Φq X/ n   ... 0  , .. ...  .  √ 0 . . . ωq X Φq X/ n. .... y. kxit k < M.. io. sit. Nat Second, let. max. 1≤i≤n,1≤t≤T. er. A3 :. ‧. where Ω is a q × q matrix with elements τk ∧ τl − τk τl , and Φj = diag(fit (ξit (τj ))). N I = In 1lT , 1lT = (1, . . . , 1)0 , and I is the matrix which identifies individual effects.. al. n. i√v √ n C VnT (δ) = ωk ρh τ (yit − ξit (τk ) − z it δo / T − xit δk / nT ) U engchi k=1 t=1 i=1 q T X n X X. 0. 0. k. n X √ − ρτk (yit − ξit (τk )) + λT | ηi − δoi / T | − | ηi |, i=1. where     ˆ δ=  . δˆ0 δˆ1 .. . δˆq. . .       =    . √. T (ˆ η − η) √ ˆ 1 ) − β(τ1 )) nT (β(τ .. . √ ˆ q ) − β(τq )) nT (β(τ. 17.     .  .

(25) √ Hence, under conditions A1–A3 and given λT / T → λ0 , na /T → 0, and a > 0, δˆ can minimize VT n and VT n has a limiting distribution 1 0 0 0 V0 (δ) = −δ Bg + δ D1 δ + λ0 δs . 2 0. 0. 0. Bg denotes a Gaussian vector with a zero mean and covariance D0 , s = (s0 0pq ) , and s0 = (sgn(ηi )).. 3.3 Dynamic panel quantile regression In a dynamic panel model, dynamic terms may raise biases of estimators. Conventionally, taking dynamic terms as endogenous variables and employing lagged (or. 政治大 can help us reduce the dynamic bias. The following are two approaches for dynamic 立 lagged differenced) dependent variables as instrument variables is a method which 5. panel quantile regression.. ‧ 國. 學. 3.3.1 The IVQR approach. ‧. According to Chernozhukov and Hansen (2005, 2006), Harding and Lamarche (2009). y. Nat. proposed an IVQR method to deal with endogenous problems for panel data. Sim-. n. al. er. io. a dynamic panel model. sit. ilarly, Galvao (2008) proposed an IVQR method for dynamic panel data. Consider. Ch. i Un. v. Qyit (τ |ιit , yit−1 , xit ) = ι0it η(τ ) + α(τ )yit−1 + x0it β(τ ),. engchi. (11). where yit is a dependent variable, xit is an exogenous variable, yit−1 is the lag of a dependent variable, η(τ ) = (η1 (τ ), . . . , ηn (τ )) is a vector of individual effects, and ιit is an indicator variable which identifies the individual effects. α(τ ) and β(τ ) are parameters. However, an individual effect η(τ ) in equation (11) is different from equation (9). In the model of Koenker (2004), an individual effect will not change with τ , but an individual effect of Galvao (2008) is a dummy variable which will change with τ . 5. See, for example, Anderson and Hsiao (1981) and Arellano and Bond (1991).. 18.

(26) If zit is a valid instrument variable, we consider a new objective function n X T X. ρτ (yit − ι0it η(τ ) − α(τ )yit−1 − x0it β(τ ) − zit0 γ(τ )).. (12). i=1 t=1. Unlike Chernozhukov and Hansen (2006), Galvao (2008) uses zit to replace the projection of yit−1 on (zit , xit ). The estimation procedure consists of two steps. First, define a grid of αj where j = 1, . . . , J and plug αj into αj (τ )yit−1 respectively. Second, take (yit − αj (τ )yit−1 ) as a new dependent variable, run a quantile regression of (yit − αj (τ )yit−1 ) on (ιit , xit , zit ), and search the α ˆ j among j = 1, . . . , J which makes kˆ γ (ˆ αj , τ )k minimized, and the estimators α ˆ G (τ ) and βˆG (ˆ αG (τ ), τ ) can be obtained.. 政治大. Galvao (2008) presented large sample properties as the following. Consider a. 立. closed ball with the center α(τ ), radius πn , and πn → 0 going slowly. For any. √ √ √ √ 0 0 0 ρτ (yit − ξit (τ ) − ιit δη / T − yit−1 δα / nT − xit δβ / nT ) − zit δγ / nT. ‧. t=1 i=1. ‧ 國. VnT (δ) =. T X n X. 學. αn (τ ) → α(τ )(δα → 0), rewrite objective function equation (12) as. Nat. 0. y. − ρτ (yit − ξit (τk )), 0. sit. n. al. er. io. where ξit (τk ) = ηi (τ ) + α(τ )yit−1 + xit β(τ ) + zit γ(τ ), and    √  δˆη T (ˆ η (αn , τ ) − η(τ ))      δˆ   √nT (α (τ ) − α(τ ))  n  α    δˆn =  = √ .  δˆβ   nT (β(α ˆ n , τ ) − β(τ ))      √ δˆγ nT (ˆ γ (αn , τ ) − γ(τ )). Ch. engchi. i Un. v. Under some conditions,6 let ψ(u) = (τ − 1{u<0} ) and Ψ = ψτ (u)(yit − ξit (τ )) so that Galvao    δˆα     δˆβ  =     ˆ δγ 6. (2008) derives   −[Jα0 J¯γ AJ¯γ Jα ]−1 [Jα0 J¯γ AJ¯γ (X 0 MI Ψ)] min δˆγ (δα )0 Aδˆγ (δα )   0 ¯ −1 0 ¯ 0  ¯ ¯ ¯ J¯β [−(Z 0 MI Ψ) − Jα δα ]   =  −Jβ [(I − Jα [Jα Jγ AJγ Jα ] Jα Jγ AJγ )(X MI Ψ)] J¯γ [−(Z 0 MI Ψ) − Jα δα ] −J¯γ [(I − Jα [Jα0 J¯γ AJ¯γ Jα ]−1 Jα0 J¯γ AJ¯γ )(X 0 MI Ψ)]. For more details, see Galvao (2008).. 19.   . .

(27) J¯γ Jα is invertible, so δˆγ = 0 + Op (1) + op (1). Define θˆG (τ ) = (ˆ αG (τ ), βˆG (ˆ αG (τ ), τ ), γG (τ )), and hence the large sample properties can be shown as √ nT (θˆG (τ ) − θG (τ )) → N ((K 0 , L0 )S(K 0 , L0 )), where S = τ (1 − τ )IE[(Z 0 , MI )(Z 0 , MI )0 ], K = (Jα0 J¯γ0 A[α(τ )]J¯γ Jα )−1 Jα0 J¯γ0 A[α(τ )]J¯γ ), and L = J¯β (I − Jα K). Empirically, (K 0 , L0 ) and S can be estimated as JˆG (τ ) and SˆG (τ ): JˆG (τ ) =. 政治大. n X T X 1 1(|ûit (τ )|≤HnT ) [zit0 , x0it , ι0it ]0 [yit−1 , x0it , ι0it ] 2nT HnT i=1 t=1. 立. ‧ 國. 學. n T 1 XX 0 0 0 0 0 0 0 ˆ [z , x , ι ] [z , x , ι ], SG (τ ) = τ (1 − τ ) nT i=1 t=1 it it it it it it. ‧. 0 where uît (τ ) = yit − ιit ηˆG (τ ) − yit α ˆ G (τ ) − x0it βˆG (τ ), and HnT is a kernel bandwidth.. sit. y. Nat. 3.3.2 The fitted value approach. io. er. Lin (2010) solved the endogenous problems by using a fitted value and utilized the shrinkage method proposed by Koenker (2004) to eliminate an individual effect.. n. al. Consider the following model,. Ch. engchi. i Un. v. Qyit (τ |ηi , yit−1 , xit ) = ηi + α(τ )yit−1 + x0it β(τ ).. (13). Contrary to equation (11), individual effects ηi are fixed, which means ηi will not change with τ such as in Koenker (2004). yit is an dependent variable, yit−1 is a dynamic term, and xit is a covariate. α(τ ) and β(τ ) are interesting parameters. Similarly, assume zit is a valid instrument variable, and we can estimate the parameters by a two-step procedure. First, run OLS of yit−1 on zit , and we can. 20.

(28) obtain the fitted value of yit−1 : yît−1 . Substitute yît−1 for yit−1 in equation (13), and the model will become Qyit (τ |ηi , yît−1 , xit ) = ηi + α(τ )ˆ yit−1 + x0it β(τ ).. (14). Second, we have to solve the following objective function for estimating parameters in equation (14): q T X n X X. 0. ωk ρτk (yit − ηi − α(τ )ˆ yit−1 − xit β(τk )) + λ. n X. k=1 t=1 i=1. |ηi |.. i=1. As in equation (9), k = 1, . . . , q and k represents various quantiles, ρτk (.) is a check function, and ωk is a weight which controls the relative influence of the τk th quantile. Again, when λ → 0, the dynamic panel quantile regression estimators (ˆ αL (τ ), βˆL (τ )). 政治大 can be obtained. The fitted value approach of Lin (2010) is applied in my thesis, 立 because a fixed effects model is a common choice for macroeconomists (Judson and. ‧ 國. 學. Owen, 1999).. ˆ ), bootstrapping is utilized For estimating the variance-covariance matrix of β(τ. ‧. here.7 Bootstrapping is a re-sampling method, which can help us obtain properties. y. Nat. of an estimator from an approximating distribution. In practice, we could sample. sit. from observations {yi , xi , i = 1, . . . , n} according to i, and hence a new sub-sample. n. al. er. io. {yi∗ , x∗i } can be obtained. Then, we run a quantile regression of yi∗ on x∗i , and the estimator βˆ∗ (τ ) can be obtained. Next, we resample and run the regression as above, and we can get a number of βˆ∗ (τ, b), where b = 1, . . . , B and B is a number. Ch. engchi. i Un. v. of re-sampling times. For example, if we do a re-sampling B times, we can obtain estimators βˆ∗ (τ, 1), βˆ∗ (τ, 2), . . . , βˆ∗ (τ, B). And then, the variance-covariance matrix ˆ ) can be estimated as of β(τ B. ˆ )) = Vd ar(β(τ. 1 X ˆ∗ 0 ¯ ¯ (β (τ, b) − βˆ∗ (τ ))(βˆ∗ (τ, b) − βˆ∗ (τ )) , B − 1 b=1. P ¯ ˆ∗ where βˆ∗ (τ ) = B −1 B b=1 β (τ, b). 7. Bootstrapping is proposed by Efron (1979). The application of bootstrapping to quantile regression, see Buchinsky (1995, 1998).. 21.

(29) 4 Empirical Results This section is organized as follows. At first, data descriptions, sources and characteristics are summarized. And later on, average long term data is used in crosssectional analysis. And finally, panel data is used to estimate empirical results which consist of baseline, extensive, country group-specific and period specific analysis.. 4.1 Data The main dataset consists of a panel of 91 countries (see Appendix A) from 1960 to 2006, and the main sources are the IMF’s International Financial Statistics (IFS). 政治大 World Bank’s World Development Indicators (WDI), Desai et al. (2003), 立. and the Penn World Table version 6.3 (PWT 6.3). Some gaps are filled with the 8. Mitchell. (2007a–c) and the United Nations’ National Accounts Statistics database.. ‧ 國. 學. Inflation is measured by the annual change rate in the consumer price index. Fiscal deficits are nominal central government deficits scaled by narrow money stock. ‧. (M1) and nominal GDP, so I calculated the deficit-to-money ratio as fiscal deficits. y. Nat. over M1 and the deficit-to-GDP ratio as fiscal deficits over nominal GDP. The money. io. sit. growth rate is the annual change in the money stock (M1). The growth rate of real. er. GDP per capita is the annual change in the real GDP per capita, which represents. al. n. iv n C currency, and oil price inflation ishitseannual h i U The benefit of measuring the n g cchange. real economic growth. The oil price is the average crude price of petroleum in local. oil price in local currency is that each country could face various oil prices. Finally, openness is measured by the average of the import- and export-to-GDP ratio. (For detailed data sources and descriptions: see Appendix B.) Table 1 provides a summary of the characteristics of the original data of selected countries over the period from 1960–2006 (panel A), 1970–2006 (panel B), 1980–2006 (panel C) and 1990–2006 (panel D). We can see that from 1960–2006 to 1990–2006, 8. I thank Dr. Raj M. Desai for generously sharing their dataset.. 22.

(30) Table 1: Descriptive statistics of selected countries mean. quantile. median. 0.25. quantile. standard. 0.75. deviation. minimum. maximum. number of countries. (A) 1960–2006 inflation rate (%). 27.27. 2.69. 6.16. 12.88. 307.09. -100.00. 10945.70. 91. deficits/money (%). 23.83. 5.67. 18.08. 33.93. 39.35. -180.64. 1056.96. 91. deficits/GDP (%). 3.36. 0.98. 2.80. 4.92. 5.61. -22.24. 204.56. 91. money growth rate (%). 29.50. 6.73. 13.17. 21.66. 275.45. -99.90. 11673.40. 91. growth of real GDP per capita (%). 2.39. -0.01. 2.47. 4.92. 5.87. -42.95. 68.87. 91. oil price inflation (%). 84.64. -0.54. 4.87. 24.36. 3273.49. -63.42. 213153.20. 91. openness (%). 34.22. 19.30. 27.81. 42.43. 24.68. 0.15. 228.47. 91. inflation rate (%). 31.05. 3.29. deficits/money (%). 25.82. 5.96. deficits/GDP (%). 3.63. money growth rate (%). 31.19. 7.15. (B) 1970–2006. deficits/GDP (%). 42.67. -180.64. 1056.96. 96. 2.95. 5.28. 6.00. -21.98. 204.56. 96. 14.11. 23.35. 274.36. -62.55. 11673.40. 96. 2.29. 4.75. 5.92. -41.11. 60.37. 96. -1.62. 11.42. 31.01. 3591.80. -63.42. 213153.20. 96. 36.72. 21.08. 30.04. 46.49. 25.01. 0.15. 228.47. 96. 39.89. 2.73. 6.67. 13.96. 392.18. -100.00. 10945.70. 101. 26.92. 5.62. 19.17. 38.57. 48.69. -221.44. 1056.96. 101. 3.57. 0.96. 2.85. 5.37. 6.57. -22.66. 204.56. 101. 38.78. 6.53. 13.50. 23.19. 322.70. -62.55. 11673.40. 101. 1.87. al. n. growth of real GDP per capita (%). 36.34. -0.34. io. money growth rate (%). 19.08. 2.20. oil price inflation (%). 176.93. openness (%). 38.23. -0.45 20.34. C 22.11h. (D) 1990-2006. 2.08. 5.39. -36.18. 56.40. 101. n 49.84 U 25.26 i e 31.97 h ngc. -19.34. 301488.60. 101. 0.15. 228.47. 101. 30.74. 4.37. y. Nat. deficits/money (%). 96. sit. inflation rate (%). 10945.70. ‧. (C) 1980-2006. -100.00. er. openness (%). 336.16. 101.68. ‧ 國. oil price inflation (%). 1.04. 14.44. 學. growth of real GDP per capita (%). 立. 政治大 7.44. 46.94. iv. 5789.44. inflation rate (%). 24.89. 2.44. 5.70. 11.82. 276.81. -13.85. 7485.49. 98. deficits/money (%). 23.66. 3.33. 15.47. 35.00. 44.10. -180.64. 551.36. 98. deficits/GDP (%). 2.83. 0.62. 2.38. 4.54. 6.51. -21.98. 204.56. 98. money growth rate (%). 27.45. 6.80. 13.32. 23.09. 230.26. -29.67. 6724.82. 98. growth of real GDP per capita (%). 2.26. 0.08. 2.38. 4.60. 5.14. -36.18. 56.40. 98. oil price inflation (%). 12.86. -6.43. 10.24. 32.10. 24.84. -40.27. 104.42. 98. openness (%). 40.64. 24.55. 34.01. 53.22. 25.09. 0.15. 228.47. 98. Source: the International Financial Statistics, Mitchell (2007a–c), the Penn World Table 6.3, Desai et al. (2003), the World Development Indicators and the United Nations’ National Accounts Statistics database.. 23.

(31) the average inflation rates are 27.27%, 31.05%, 39.89% and 24.89%, and the standard deviations of the inflation rate are 307.09, 336.16, 392.18, and 276.81 respectively. Inflation tends to be higher and more volatile since 1960, and it becomes lower and more stable after 1990. Compared with the median of inflation rate, 6.16%, 7.44%, 6.67% and 5.70% respectively, the average inflation rates are higher. This means that the average is prone to be affected by extreme observations. Quantile regression can avoid the estimated outcomes affected by extreme observations. The deficit-to-money ratio is similar. From 1960–2006 to 1990–2006, the average deficit-to-money ratios are 23.83%, 25.82%, 26.92% and 23.66%, and the standard deviations are 39.35, 42.67, 48.69 and 44.10 respectively. On the other hand, the. 政治大 deviations are 5.61, 6.00, 6.57 and 6.51 respectively. As we can see, both the deficit立 average deficit-to-GDP ratios are 3.36%, 3.63%, 3.57% and 2.83%, and the standard. after 1990, and yet the volatility is still large after 1990.. 學. ‧ 國. to-money ratio and the deficit-to-GDP ratio tend to be larger and become smaller. The average money growth rates are 29.50%, 31.19%, 38.78% and 27.45% from. ‧. 1960–2006 to 1990–2006 respectively, and the standard deviations are 275.45, 274.36,. y. Nat. 322.70 and 230.26. Similarly, both the growth rates and volatility reach a peak in. io. sit. the 1980s, and decline after 1990. About the other controlled variables, the growth. er. rate of the real GDP per capita declines until 1980–2006, and becomes rapid after. al. n. iv n C and sharply drops in 1990–2006. h Finally, e n gopenness c h i Uis consistently growing higher, 1990. The oil price inflation rises higher until reaching a peak during 1980–2006,. and volatility is stable.. Next, for comparing countries in various development levels, I classify country groups according to the income level and OECD membership which are based on the World Bank list of economies (July 2009) (see Appendix C). However, the classification “OECD” consists of countries which are not only OECD members but also in a high-income level. Some OECD members in middle- or low-income levels are excluded such as Turkey, and the group of high-income countries contains OECD. 24.

(32) Table 2: Descriptive statistics of specific income groups (1960–2006) mean. quantile. median. 0.25. quantile. standard. 0.75. deviation. minimum. maximum. number of countries. (A) high-income countries inflation rate (%). 7.02. 2.15. 3.95. 7.71. 16.22. -20.63. 373.82. 33. deficits/money (%). 19.65. 3.16. 12.47. 28.16. 38.14. -110.26. 613.14. 33. deficits/GDP (%). 2.73. 0.67. 2.22. 4.55. 4.49. -22.24. 26.74. 33. money growth rate (%). 12.90. 5.43. 10.02. 15.42. 23.02. -76.85. 430.17. 33. growth of real GDP per capita (%). 2.95. 0.98. 2.80. 4.90. 4.27. -21.97. 27.81. 33. oil price inflation (%). 15.34. -2.84. 1.98. 19.49. 47.00. -63.42. 405.31. 33. openness (%). 40.91. 23.27. 33.98. 49.08. 30.09. 4.63. 228.47. 33. inflation rate (%). 38.80. 3.54. deficits/money (%). 26.20. 7.91. deficits/GDP (%). 3.72. money growth rate (%). (B) middle- and low-income countries. 384.01. -100.00. 10945.70. 58. 20.26. 36.13. 39.84. -180.64. 1056.96. 58. 1.15. 3.12. 5.02. 6.13. -21.98. 204.56. 58. 38.94. 8.14. 15.46. 25.00. 344.25. -99.90. 11673.40. 58. 2.07. -0.78. 2.09. 4.96. 6.59. -42.95. 68.87. 58. 124.07. 0.00. 7.11. 30.13. 4099.92. -60.09. 213153.20. 58. 30.42. 17.42. 25.10. 37.61. 20.03. 0.15. 122.23. 58. 立. ‧. openness (%). 16.47. ‧ 國. oil price inflation (%). 8.04. 學. growth of real GDP per capita (%). 政治大. n. al. er. io. sit. y. Nat. Source: the International Financial Statistics, Mitchell (2007a–c), the Penn World Table 6.3, Desai et al. (2003), the World Development Indicators and the United Nations’ National Accounts Statistics database.. Ch. i Un. v. classification. Therefore, high-income or OECD countries represent economies in. engchi. higher development, and middle- and low-income or non-OECD countries represent developing economies. Table 2 is a summary of the characteristics of the original data in a high-income country group (panel A) and a middle- and low-income country group (panel B). We can see that in high-income countries, the average inflation is 7.02% and its standard deviation is 16.22. In middle- and low-income countries, the average inflation rate is 38.80% and 384.01. Accordingly, inflation is lower and more stable in high-income countries.. 25.

(33) Table 3: Descriptive statistics of OECD and non-OECD countries (1960–2006) mean. quantile. median. 0.25. quantile. standard. 0.75. deviation. minimum. maximum. number of countries. (A) OECD countries inflation rate (%). 6.36. 2.33. 4.12. 7.97. 7.06. -13.85. 84.22. 24. deficits/money (%). 14.99. 2.80. 10.56. 22.81. 22.38. -72.42. 160.32. 24. deficits/GDP (%). 2.56. 0.63. 2.01. 4.12. 3.95. -22.24. 20.79. 24. money growth rate (%). 11.59. 5.60. 9.47. 15.28. 12.87. -62.55. 192.09. 24. growth of real GDP per capita (%). 2.90. 1.16. 2.74. 4.61. 3.27. -13.56. 21.37. 24. oil price inflation (%). 14.65. -2.49. 2.70. 19.42. 44.95. -63.42. 292.31. 24. openness (%). 30.02. 20.79. 28.28. 36.24. 14.64. 4.63. 92.15. 24. inflation rate (%). 34.76. 3.03. deficits/money (%). 26.99. 7.72. deficits/GDP (%). 3.65. money growth rate (%) growth of real GDP per capita (%) oil price inflation (%) openness (%). 政治大 7.31. 15.15. 357.58. -100.00. 10945.70. 67. 20.36. 37.50. 43.43. -180.64. 1056.96. 67. 1.14. 3.10. 5.08. 6.07. -21.98. 204.56. 67. 35.91. 7.61. 14.63. 24.27. 320.69. -99.90. 11673.40. 67. 立. 學. ‧ 國. (B) non-OECD countries. 2.21. -0.64. 2.24. 5.11. 6.54. -42.95. 68.87. 67. 109.71. -0.07. 5.96. 28.44. 3814.75. -60.09. 213153.20. 67. 35.73. 18.74. 27.59. 46.28. 27.24. 0.15. 228.47. 67. ‧. er. io. sit. y. Nat. Source: the International Financial Statistics, Mitchell (2007a–c), the Penn World Table 6.3, Desai et al. (2003), the World Development Indicators and the United Nations’ National Accounts Statistics database.. al. n. iv n C money ratio is 19.65% and 38.14 inhhigh-income i U and 26.20% and 39.84 in e n g c h countries, Scaling by money stock, the average and standard deviation of the deficit-to-. middle- and low-income countries. Then, scaling by GDP, the average and standard deviation of the deficit-to-GDP ratio are 2.73% and 4.49 in high-income countries, and 3.72% and 6.13 in middle- and low-income countries. Obviously, whether scaling by money or GDP, the fiscal deficit is more critical in middle- and low-income countries. The average money growth rate is 12.90% in high-income countries and 38.94% in middle- and low-income countries. Its standard deviation is 23.02 in high-income. 26.

(34) countries and an astounding 344.25 in middle- and low-income countries. Apparently, money growth gets better control in high-income countries. The average growth rate of real GDP per capita and its standard deviation are 2.95% and 4.27 in high-income countries, and 2.07% and 6.59 in middle- and low-income countries. The long-term economic growth is higher and more stable in high-income countries. The average oil price inflation is 15.34% in high-income countries, but is 124.07% in middle- and low-income countries. It might be because exchange rates devaluate in middle- and low-income countries greater than in high-income countries. And finally, high-income countries are more open and have a higher dependence on trade. Table 3 provides a summary of the descriptive statistics of the original data in. 政治大 characteristics of variables do not change a lot. The average inflation rate and its 立. OECD countries (panel A) and non-OECD countries (panel B). We can see that the. standard deviation are 6.36% and 7.06 in OECD countries, and 34.76% and 357.58. ‧ 國. 學. in non-OECD countries. This means that inflation is at a higher level and more volatile in non-OECD countries.. ‧. The average fiscal deficits scaling by money are 14.99% and 26.99% in OECD. y. Nat. and non-OECD countries respectively, and the standard deviations are 22.38 and. io. sit. 43.43. Scaling by GDP, the average fiscal deficits are 2.56% and 3.65% in OECD and. er. non-OECD countries, and the standard deviations are 3.95 and 6.07 respectively.. al. n. iv n C h emoney About other variables, the average h i Urate is 11.59% in OECD counn g cgrowth. Similarly, a fiscal deficit is a more critical problem in non-OECD countries.. tries and 35.91% in non-OECD countries and its standard deviation is 12.87 in high-income countries and 320.69 in non-OECD countries. Money growth is well. controlled in non-OECD countries. The average growth rate of real GDP per capita and its standard deviation are 2.90% and 3.27 in OECD countries, and 2.21% and 6.54 in non-OECD countries. Similarly, the long-term growth rate is higher and more stable in OECD countries. The average oil price inflation is 14.65% in OECD countries and 109.71% in non-OECD countries. And finally, openness in OECD. 27.

(35) countries is higher than in non-OECD countries, and this means that OECD countries are more dependent on trade. In general, we can find that from 1960–2006 to 1980–2006, the macroeconomic performance and related variables deteriorate and become more volatile, and they develop into better and more stable entities after 1990. Openness is the only variable that consistently goes higher. On the other hand, whether classifying countries by income level or OECD membership, macroeconomic variables, including inflation and fiscal deficits, perform better and are more stable in countries in higher development, and perform worse and are more unstable in countries in lower development.. 政治大 First, the deficit-inflation linkage 立 is examined with the long-term average data over 4.2 Cross-sectional results. ‧ 國. 學. the period covering 1960–2006. Given a specific quantile τ , the empirical model of quantile regression is. 0. (15). ‧. πi = xi β(τ ) + εi (τ ),. where πi is the logarithm of the average of annual inflation (log(inflation)), and. Nat. sit. y. inflation is measured by the percent change in the CPI index. Taking the logarithmic. al. er. io. transformation could normalize the data and reduce the effect of the extreme data.. n. xi is a vector of explanatory variables, including intercepts and fiscal deficits. β(τ ). Ch. i Un. v. is a vector of parameters at a specific quantile τ , and εi (τ ) is an error term.. engchi. The cross-sectional results are shown in Table 4. Table 4 reports the coefficients at various quantiles from 0.1 to 0.9, and also reports the results of OLS for comparison. For calculating a confidence interval, bootstrapping is employed to estimate stand errors, and the number of re-sampling times is 1000. We can find that the deficit-inflation relationship is weak and not robust in the long-term average data, and this result is consistent with some previous empirical studies, such as those observed by Click (1998) and Kwon et al. (2009). First, the theoretical model of Catão and Terrones (2005) is followed. They mod28.

(36) eled the deficit-inflation relationship with scaling deficits by narrow money which stands for the size of the inflation tax base. The results are reported in panel (A) of Table 4 and plotted on the upper part of Figure 1. The black solid and black dotted lines indicate coefficients at various quantiles and the 95% confidence interval respectively. The grey solid line is the OLS estimate. We can see that from quantiles 0.1 to 0.9, the coefficients of the average deficit-to-money ratio are 0.5287, 0.5869, 0.5803, 0.7813, 0.7457, 0.6716, 0.8603, 0.8144 and 1.7967. From quantiles 0.1 to 0.5, the coefficients are significant at the 1% level, and significant at the 5% level at quantile 0.6 and 0.7. However, the coefficient is insignificant at quantile 0.8. Although it seems that the long-term average deficits become inflationary as long-. 政治大 other hand, the OLS coefficient is 0.8687 at the 1% level of significance; it indicates 立. term average inflation rises, it is not significant enough at the top quantile. On the. that average deficit-to-money ratios have a positive impact on average inflation.. ‧ 國. 學. Second, the standard specification of scaling fiscal deficits by GDP rather than money is estimated. The results are shown in panel (B) of Table 4 and plotted in on. ‧. the middle part of Figure 1. The same black solid and the black dotted lines indicate. y. Nat. coefficients at various quantiles and the 95% confidence interval respectively. The. io. sit. grey solid line is the OLS estimate. We can see that the coefficients of the average. er. deficit-to-GDP ratios are 4.7201, 4.2520, 3.7353, 3.2610, 2.9507, 2.1895, 1.7003,. al. n. iv n C level from quantiles 0.1 to 0.3 andhatethe 5% level n g c h i atU quantile 0.4, but insignificant 3.3068, 4.5772 and 8.1045 from quantiles 0.1 to 0.9. They are significant at the 1%. from quantiles 0.5 to 0.9. The OLS coefficient is 0.8687 at the 1% level of significance.. Compared with scaling deficits by money, the deficit-to-GDP ratio is significant at fewer quantiles and shows that the long-term relationship is weak. However, the OLS estimator is still positively significant. Finally, with the standard specification of scaling fiscal deficits by GDP, the average money growth rates are controlled in equation (15), such as Kwon et al. (2009). As inflation, the average money growth is transformed into a logarithmic. 29.

(37) form. The results are shown in panel (C) of Table 4 and the lower part of Figure 1. The meanings of the black solid line, the black dotted line, and the grey solid line are defined above. From quantiles 0.1 to 0.9, the coefficients of the average deficitto-GDP ratio are 2.5915, 1.1392, 1.0395, -0.1321, 0.1123, 0.5190, 0.6264, 0.5740 and -0.1386, and the coefficients of the average money growth are 0.6631, 1.2118, 1.2456, 1.2775, 1.2526, 1.2569, 1.2467, 1.3152 and 1.3279. All of the coefficients of the deficit-to-GDP ratio are insignificant, and all of the coefficients of the money growth are significant at a 1% level except quantile 0.1, where the coefficient is significant at the 10% level. The OLS estimates of the deficit-to-GDP ratio and the money growth are 0.6031 and 1.1724 respectively. The same, the deficit-to-GDP. 政治大 Compared with previous studies, Click (1998) discovered that domestic debt has 立. ratio is insignificant and the money growth is at a 1% level of significance.. no effect on seigniorage with long-term average data, and therefore fiscal variables. ‧ 國. 學. are not inflationary. Kwon et al. (2009) found a weak relationship between public debt and inflation over the long term. On the other hand, Fischer et al. (2002). ‧. showed that fiscal deficits positively affect inflation in the long-term average data in. y. Nat. a full sample.. io. sit. Accordingly, only when the long-term average money growth is not controlled,. er. the OLS estimators are consistent with Fischer et al. (2002). Otherwise, the esti-. al. n. iv n C h ewith the fiscal deficit is weakly associated and not robust in the long-term n ginflation chi U mated results are weak or nonexistent. Hence, my cross-sectional results show that. average data, so it tends to be in line with Click (1998) and Kwon et al. (2009).. As Kwon et al. (2009) illustrated, in the long run, debt must be solved with a fiscal surplus or be monetized ultimately, and which one is chosen is determined by the policy regime (Sargent, 1982). However, the policy regime could be different in each country and change over time, so it is difficult to find a statistical linkage of fiscal variables and inflation in long-term average data.. 30.

(38) 4.3 Dynamic panel results 4.3.1 The baseline analysis Since long-term average cross-sectional data cannot clarify the relationship between fiscal deficits and inflation, the panel data provides another approach to estimating the deficit-inflation relationship in the long-run. Panel data estimation contains individual time-invariant terms for controlling country-specific effects, and could allow for an intrinsically dynamic adjustment to distinguish the long-run effect. Empirically, given a specific quantile τ , the dynamic panel quantile regression model is. x β (τ ) + ε (τ ), 政治大. πit = ηi + α(τ )πit−1 +. 立. p X. 0. it−j j. it. (16). j=0. where πit is the logarithm of one plus annual inflation (log(1+inflation)), and in-. ‧ 國. 學. flation is measured by an annual change in the CPI index. ηi is a time-invariant. ‧. individual effect, xit−j refers to current and lagged fiscal deficits, and εit (τ ) is an P error term. α(τ ) and βj (τ ) are the parameters to be estimated, and pj=0 βj (τ ) is what we are concerned with. Dynamic panel quantile regression of Lin (2010),. Nat. sit. y. which is a two-stage fitted value approach, is employed to estimate equation (16),. io. er. and lagged differenced dependant variables are taken as instruments. Lagged inflation is included on the right-hand side to capture persistence and. n. al. Ch. i Un. v. dynamic adjustment. The fiscal deficit is scaled by narrow money, which is modeled. engchi. by Catão and Terrones (2005). They scaled the deficit by narrow money rather than by GDP, because the former (narrow money) stands for the size of the inflation tax base. Thus, given a change in the deficit-to-GDP ratio, the economy in higher inflation would be impacted by deficits more strongly, because its inflation tax base is typically more narrow. In addition, the fiscal deficits do not necessarily impact on inflation contemporaneously since the government can borrow and allocate seigniorage intertemporally. Therefore, the fiscal deficit is considered to be a distributed-lag due to the dynamic relationship. Accordingly, the deficit-inflation relationship is nonlin-. 31.