THE ROLE OF FINANCIAL LIBERALIZATION IN CONSUMPTION BOOMS

Transcription

1 THE ROLE OF FINANCIAL LIBERALIZATION IN CONSUMPTION BOOMS A Thesis submitted to the Faculty of the Graduate School of Arts and Sciences of Georgetown University in partial fulfillment of the requirements for the degree of Master of Public Policy By Zhi Wang, B.A. Washington, D.C. April 16, 2014

2 Copyright 2014 by Zhi Wang All Rights Reserved ii

3 THE ROLE OF FINANCIAL LIBERALIZATION IN CONSUMPTION BOOMS Zhi Wang, B.A. Thesis Advisor: Andreas Kern, Ph.D. ABSTRACT Consumption booms, defined as unusual upsurge in a country s real private consumption expenditures, are often perceived as dangerous macroeconomic events because of their detrimental effects on home goods prices, the balance of trade, and the resources available for investment. Using a large cross-country panel data set from , this paper assesses how financial liberalization explains the occurrence of consumption booms. There are two key findings: First, while financial liberalization is not responsible for triggering all the consumption booms across the sample, the booms have indeed arisen more readily in the circumstances of financial reform, particularly the liberalization of the capital account, holding other factors constant. Second, the correlation between the probability of booms occurring and financial openness is non-linear. As many emerging economies like India and China are considering fully liberalizing their financial markets, these countries need to be particularly cautious about potential consumption booms that might jeopardize the macroeconomic stability. iii

4 Many thanks to my professors at the McCourt School of Public Policy, particularly my thesis advisor, Andreas Kern. And a special thank you to my parents, whose uncompromising love and expectations for me make this research possible. iv

5 TABLE OF CONTENTS I. INTRODUCTION... 1 II. BACKGROUND AND LITERATURE REVIEW... 4 III. METHODOLOGY IV. DATA I. DEPENDENT VARIABLE CONSUMPTION BOOMS II. INDEPENDENT VARIABLES V. REGRESSION RESULTS VI. POLICY IMPLICATIONS AND CONCLUDING REMARKS VII. APPENDIX APPENDIX 1: DISTRIBUTION OF CONSUMPTION BOOMS APPENDIX 2. COUNTRY COVERAGE OF THE ANALYSIS OF CONSUMPTION BOOMS APPENDIX 3: CORRELATION MATRIX BETWEEN CONSUMPTION BOOM AND THE DEPENDENT VARIABLES APPENDIX 4: SUMMARY STATISTICS FOR THE FINANCIAL REFORM INDEX AND ITS COMPONENTS (83 COUNTRIES, ) APPENDIX 5: FRI COMPONENTS EFFECTS ESTIMATION FOR DIFFERENT INCOME GROUPS VIII. REFERENCE LIST v

6 I. INTRODUCTION Consumption booms, defined as unusual growth in a country s real private consumption expenditures (Montiel, 2000), were widely observed in recent years across the world before the Global Financial Crisis. Booms in household consumption are frequently associated with various macroeconomic phenomenons, including stabilization of high inflation, surges in capital inflows, the implementation of market oriented structural reforms (especially trade and financial liberalization), and favorable movements in the external terms of trade (Montiel, 2000, p. 457). Nowadays, developing countries like China are in transition to a liberalized financial market. A potential consumption boom-bust cycle that might accompany such financial deregulation could undermine these countries overall macroeconomic and financial stability. Thus consumption booms and their effects remain key concerns for authorities in these countries. Consumption booms can be dangerous to the whole economy and are often perceived as posing policy challenges because they are closely associated with surges in commodity prices, worsening of trade balance, and they might squeeze out the economic resources for investment. According to Montiel (2000), the emergence of a consumption boom may, for example, impede the goal of controlling inflation by driving up prices of domestic goods. Similarly, a consumption boom that arises during a period of capital inflows may significantly increase the country s current account deficit, which may, in turn, undermine the credibility of the prevailing exchange rate and contribute to capital flow reversals (Montiel, 2000, p. 457). Consumption booms that occur with capital inflows have been observed in several countries when they liberalized their capital account. For example, in Mexico, a financial and economic reform was introduced in Subsequently, between 1989 and 1994, the country experienced a consumption boom and 1

7 an excessively large current account deficit. As a result, in 1994, there was a sudden devaluation of the Mexican peso, which led to the notorious Tequila financial crisis (Gonzalez-Garcia, 2001). This takes us to the question whether financial liberalization systematically increases the likelihood of a consumption boom? In this paper, I hypothesize that financial liberalization, particularly the liberalization of the capital account, leads to greater credit availability, which results in enhanced household consumption. Based on the previous literature, this paper aims to reexamine the relationship between financial liberalization and consumption booms, controlling for all the variables considered to be relevant in former works. This study differs from the pervious research in at least three aspects. First, the dataset being analyzed includes a much bigger sample of 83 countries and regions, and more importantly, it covers a time span from As consumption booms are commonly identified across the world in recent decades, particularly around 2000s, I expect that those booms would allow new insights into the relationship. Second, this paper uses a de-trending method popularized by Hodrick and Prescott (1997), the HP filter, to identify consumption booms. Rather than former method of identifying a boom as a run of positive deviations of the ratio of private consumption to GDP from its deterministic trend (Montiel, 2000, p. 459), this paper applies a universal algorithm (Hodrick and Prescott s HP filter) and a relatively lower consumption boom threshold to include several episodes in Asia at the end of the millennium. A boom is identified only if the deviation of log real consumption expenditure from trend exceeds 1.55 times the standard deviation of the trend line itself. In addition, a lower threshold of one standard deviation is used to pin down the boom s start and end dates. It is worth noting that the thresholds are country-specific, ensuring that consumption booms are unusually large deviations relative to a country s own consumption path. Third, this paper uses a much more comprehensive 2

8 measure of the degree of financial liberalization, proposed by Abiad, Detragiache, and Thierry (2008). This new measurement considers the multi-faceted nature of financial reform and classifies financial policy changes along seven different dimensions: credit controls and reserve requirements, interest rate controls, entry barriers, state ownership, policies on securities markets, banking regulations, and restrictions on the capital account (Abiad et al., 2008, p. 3). The scores of the Financial Reform Index range from 0 to 21. Different from most former measures of financial development, which use binary dummy variables to identify financial liberalization, Abiad et al. s method provides a much better measure of the magnitude and timing of financial policy changes than was previously possible (Abiad et al., 2008, p. 4). The empirical results of this paper show that the relationship between the level of financial liberalization and the probability of a consumption boom to occur is non-linear. Holding other variables constant, countries at the intermediate stage of financial liberalization (e.g., China) are more likely to have consumption booms than countries with poor financial developments or with well-developed financial markets. Financial reforms significantly increase the chances of having consumption booms for these countries mainly through the channels of capital imports and wealth effects. The main policy implications for the countries are that they should carefully consider the timing of financial liberalization. And if they decide to implement financial reforms, a policy package that combines financial liberalization with measures that limit borrowing and enforce savings would effectively prevent the occurrence of consumption booms, and thus enhance macroeconomic stability. 3

9 II. BACKGROUND AND LITERATURE REVIEW During the 2000s until the outbreak of the Global Financial Crisis, most economies in the world embraced stable and rapid growth. Recent theoretical and empirical works had posed optimistic macro-prudential views of financial markets and claimed to have garnered a better understanding of the relationship between macroeconomic dynamics and financial developments. And policymakers did not engage in any major institutional reform because they extrapolated current performance over a long period: why should they announce possibly unpopular measures when their economies were in a good shape? The sophistication of the modern financial systems had provided rapid growth with low inflation (Boyer, 2012, p. 2). However, the collapse of Lehman Brothers and its aftermath were a wake up call for economists and policymakers to rethink how financial systems are associated with trade, investment, consumption and other macroeconomic dynamics. The severity and breadth of the economic crisis called for, in particular, a reassessment of how macroeconomic dynamics and financial markets are interconnected. One important dimension where the financial system and macroeconomic variables interact is how financial liberalization affects the magnitude of private consumption/saving (saving and consumption are directly negatively associated according to classic Macroeconomic theory). Household consumption expenditure, an important component of a country s output level, has been studied repeatedly by economists (Nelson, 1988). Over the past several decades, theories of private consumption behavior have been well developed from both a macro and micro perspective. However, few researchers have studied the occurrence of consumption booms in a systematic way. Consumption expenditure as a crucial component in any macroeconomic measurement of 4

10 a country s output performance usually exceeds 50% of its GDP (WDI, 2014). At the same time, consumption booms are often considered as policy and economic problems because of their impact on the prices of home goods, the foreign trade accounts, and the capital stock for investment, and hence hampers a country from economically sustainable development (Montiel, 2000). De facto, the repetition of consumption boom-bust cycles has been frequently associated with huge foreign capital inflows and sudden stops in these inflows. Furthermore, these boom episodes have been accompanied with speculative asset price bubbles and the build up of financial and macroeconomic imbalances, all of which are also common features of financial crises (Borio and Lowe, 2002). In this context, consumption booms seem to be a good early warning sign for the onset of a potential financial crisis. Despite the important influence of consumption booms, researchers have only recently started to systematically study the causes of consumption booms (e.g., Laibson and Johanna, 2010). However, Montiel (2000) is an important exception. In his comprehensive analysis, Montiel proposes four major categories of macroeconomic factors that may drive consumption booms: 1) Significant income redistribution due to populist policies. Such redistribution would tend to shift income from unconstrained well-off households to constrained poorer ones, (Montiel, 2000, p. 462), thereby triggering the rise of aggregate consumption. 2) Changes in intertemporal relative prices that encourage present consumption over saving for future, through several different channels: lower real interest rates driven by exchange rate-based stabilization measures in countries with chronic inflation (i.e., Rodríguez hypothesis), temporary real exchange rate appreciation as a result of incredible exchange rate-based stabilization (i.e., Dornbusch hypothesis), and higher nominal interest rates due to the use of cash-in-advance 5

11 technology for making transactions. 3) Wealth effects, where the private sector recognizes an escalation in its wealth and adjusts its consumption path accordingly. This could be the case when national wealth expands fast, the country s terms of trade permanently improves, or the policy regime of the country transforms into a form that is expected to expedite economic growth. In addition, Montiel (2000) states, private wealth may increase because of a change in fiscal policy that is perceived to reduce the present value of the private sector s future tax liabilities (i.e., Ricardian Equivalence, p. 464). 4) Fast expansion of credit supply to the private sector, triggered by a financial liberalization reform that relaxes credit constraints. This could happen either under the context of inappropriate domestic financial liberalization or large capital inflows from abroad (Montiel, 2000). With regard to how credit expansion associated with financial liberalization will boost consumption, researchers prior to Montiel had established a theoretical framework in which the increase of credit supply influences consumption/saving rates by removing liquidity constraints. For example, Campbell and Mankiw (1989) argue that before financial liberalization, not all households have access to credit markets, and hence, some households have no ability to smooth consumption over time. Thus, for the liquidity-constrained households, consumption decisions are entirely determined by current income. As a consequence, relaxation of liquidity constraints might be associated with a consumption boom and a decline in aggregate saving (p. 4). Furthermore, Crossley and Low (2012) argue, the more constrained are the households initially, the more significant the consumption growth can be expected. In addition, Reinhart, Carmen, Tokatlidis, and Ioannis (2005) point out that many of the past liberalization episodes unleashed a period of rapid growth in bank lending, asset price booms, and increases in consumption that often coincided with a decline in private saving rates. Many of those episodes also ended in a 6

12 full-fledged financial crisis (p.4). Hence, no analysis of consumption booms is complete without an examination of the extensiveness of liquidity constraints and an assessment of the pervasiveness of financial liberalization (Reinhart et al., 2005). However, Montiel s empirical results show that despite the fact that credit expansions have been important in a substantial minority of the consumption booms (2000, p. 474), there is no significant or systematic association between them across the sample. His study does not support the hypothesis that rapid credit expansion triggered by financial reforms drives consumption booms. Instead, he found that booms appear to be associated with wealth effects that are caused by substantial improvements in the terms of trade, as well as the appreciation of real exchange rate. Nevertheless, Montiel fails to carry out a comprehensive analysis of the relationship between financial liberalization and the occurrence of consumption booms, as the channels through which financial reforms could affect consumption might not be limited to credit expansions. Outside the channel of credit expansions, financial liberalization is also believed to stimulate consumption by driving up assets prices, a wealth effect, that Montiel has overlooked in his analysis. Laibson and Mollerstrom (2010) propose that assets bubbles, fueled by financial liberalization, drive up consumption expenditure. They argue that capital imports cause asset prices to increase rapidly, resulting in the bubbles in equity markets and real estate markets. These events create extra channels for households to significantly expand their wealth and hence change their consumption behaviors. In this case, capital inflows associated with financial reforms are mostly consumed instead of invested. And countries that experience such asset bubbles should also experience a consumption boom and a commensurate current account deficit (p. 28). Similarly, Greenspan and Kennedy (2008) have shown that wealth effects in 7

13 housing market play an important role. In the United States, the increases in residential real estate prices contribute substantially to consumption expansion via financial innovations such as home equity loans and mortgage refinancing cash-outs (Greenspan and Kennedy, 2008). Alternatively, Lehmussaari (1990) argues that financial liberalization could make the private sector more responsive to wealth changes, hence magnifying the impact of wealth effects on consumption. Motivated by the sharp decline in saving rates associated with financial reforms in all Nordic countries between 1984 and 1987, Lehmussaari examines the changes in household consumption and saving behaviors in those countries. He found that financial deregulation has induced a pattern change of consumption and saving. For all Nordic countries, it seemed that earlier structural relationships between consumption growth and wealth growth broke down after the deregulation. Wealth effects appeared to have played a more important role in determining consumption. After deregulation, the consumers responses to changes in real wealth are faster and more assertive. Prior to deregulation, credit control can partially mitigate the effects of low after-tax interest rates on consumption, whereas after the liberalization, with credit availability, consumers are able to respond quickly to any events (e.g., high GDP growth) where they perceive as an increase in private wealth. Quite opposite to the view that financial liberalization drives up consumption, a variety of theories imply that the volatility of consumption relative to that of output should go down as countries open up their financial market. Economists on this side of the question argue that financial liberalization provides better opportunities for countries to share macroeconomic risk and, thereby, smooth consumption. The basic idea here is that since output fluctuations are not perfectly correlated across countries, trade in financial assets can be used to delink national consumption levels from the country-specific components of these output fluctuations (Prasad, 8

14 Rogoff, Wei, and Kose, 2003, p. 38). Therefore, since developing countries are generally more specialized in terms of their outputs and factor endowment configuration, they can, in theory obtain even bigger gains than developed countries through the mechanism of international consumption risk sharing (Prasad et al., 2003, p. 9). However, recent studies have found a different picture. For example, Levchenko (2004) argues that consumption volatility relative to income growth increased, on average, for the developing countries in the 1990s, which was exactly the phase when these countries opened their capital account to the world market. In fact, the researchers discover a threshold effect. At low levels of financial integration, they conclude, an incremental increase in financial openness is associated with an increase in consumption volatility. However, once the level of financial integration crosses a threshold, the association becomes negative. In other words, for countries that are sufficiently open financially (often the features of developed economies), relative consumption volatility starts to decline when they further liberalize their financial markets. These findings suggest a non-linear relationship between financial integration and consumption volatility across countries with different initial levels of financial openness. In brief, previous studies on the relationship between financial liberalization and consumption booms are at most partial. Montiel (2000) examined the significance of the proposed determinants of consumption booms, yet failed to find systematic effects of financial liberalization. Economists have established frameworks in which financial liberalization might affect consumption. While some argue that financial reforms and integration will help smooth consumption in a country, others have found evidence that the financial deregulation actually increases consumption volatility for those countries at the intermediate stage of economic and financial development. 9

15 III. METHODOLOGY Based on the previous literature, this paper aims to reexamine the relationship between financial liberalization and consumption booms, controlling for all the variables considered to be relevant in former literature. The paper hypothesizes that financial liberalization is an important determinant of consumption booms. Whether it occurs in the form of an inappropriate domestic credit loosening, or a large capital account liberalization, or reforms of equity market, a financial liberalization will result in events such as rapid expansion of credit to the private sector, boosting wealth effects, or lower interest rates, which might significantly drive up consumption. The basic model is a Logistic regression model that uses the binary dummy variable Consumption Boom as the dependent variable, with value of 1 being a consumption boom episode and a zero being no such episode, and the Financial Reform Index (FRI) as the key independent variable; and that includes variables that have been emphasized in previous literature. The model is of the following form: P(Consumption Boom=1 Xs)= Λ (β0 + β1 Financial Reform Indexi,t + βk Zi,t) Where Z denotes all the control variables and for all real numbers x, Λ(x)=exp(x)/[1+exp(x)]. In this specification, the predicted value of β1 does not directly express the partial effect of financial liberalization on the probability of a consumption boom. A transformation of the coefficient is needed before interpretation. I use six control variables in most of my model specifications, the reason of which will be explained in the next section. P(Consumption Boom=1 Xs)= Λ (β0 + β1 Financial Reform Indexi,t-1 + β2 GDP Growth 10

16 Ratei,t-1 + β3 Current Account Balancei,t-1 + β4 Inflation Ratei,t-1 + β5 Δ Government Consumption-to-GDP Ratioi,t-1 + β6 Real Interest Ratei,t-1 + β7 Δ Exchange Rate Stability Index i,t-1) As households generally do not immediately adjust their consumption behavior due to changes in macroeconomic factors and financial markets, I take the lag for all the independent variables. For government consumption, and exchange rate stability index, I use lagged first difference values in order to account for changes in these variables over time. As is mentioned in the literature review section, some researchers believe that financial liberalization magnifies the impact of wealth effects on consumption. Hence, I incorporate an interaction term of FRI and GDP growth rate in the model to verify the validity of the theory. One of the unique features of this study is that I question whether the relationship between boom episodes and financial openness is linear. This is a reasonable question since economists have already found evidence that the relationship between consumption volatility and financial openness is non-linear and in fact, there is a threshold effect (Levchenko, 2004). Similarly, some researchers have also demonstrated that the relationship between economic growth and financial development is non-linear (Deidda, Luca, Fattouh and Bassam, 2002). In order to answer this question, I include a quadratic term of the FRI variable in the model specification. For the purpose of examining the sensitivity of the results, I implement several measures: 1) the study runs panel regressions of consumption boom on the FRI and controls using two different samples. The first sample only includes countries that have had consumption booms before, and the second sample considers all countries in the dataset. 2) I run several linear probability regressions as well to test the significance of the independent variables. In addition, to explore whether each of the factors above has a different effect for countries of different 11

17 backgrounds, particularly, different levels of economic development, I also run separate Logistic and OLS regressions across countries of different income groups. According to World Bank Atlas method, economies are divided into three income groups based on their GNI per capita. In 2012, the standard for low income is $1,035 or less and is $1,036 - $12,615 for middle income; for high income, the threshold is $12,616 (WDI, 2014). In the next step, to test for the robustness of the results, with reference to the method Mendoza and Terrones (2011) employed in their study of credit booms, I construct five-year event windows by including the macro aggregates centered on the peak of consumption booms. The windows show the cross-country means of GDP growth, private consumption, government expenditure, real interest rate and other variables in the model specification. I re-run the previous regressions over these 5-year-windows. If financial liberalization has lagged and has a gradual impact on consumptions, I expect the coefficients of these regressions to be greater and more significant. Finally, I break down the Financial Reform Index and test the effects of different components of FRI on consumption booms. According to the hypothesis of the paper, it is expected that capital account liberalization and credit loosening, in general, would be directly associated with consumption booms. The regressions on the FRI components are also implemented for different income groups as to determine whether the driving forces behind consumption booms are different. 12

18 IV. DATA I. Dependent Variable Consumption Booms A. Identifying Consumption Booms In their widely cited paper Rapid Credit Growth: Boon or Boom-Bust, Elekdag and Wu (2011) of the International Monetary Fund defined credit booms as episodes during which real credit to the private sector expands substantially faster than what has been observed in previous expansions (Elekdag and Wu, 2011, p. 6). Similarly, the present paper outlines consumption booms as episodes during which household real consumption expands substantially faster than historical expansion paths. To determine the occurrence of a credit boom, I use the pattern recognition algorithm that Elekdag and Wu implemented based on annual data going back to Specifically, I identify a consumption boom when the deviation of real private consumption expenditure from trend is in the upper tail of the distribution (Elekdag and Wu 2011, p.6). More precisely, a consumption boom occurs when the cyclical component of real consumption is larger than its standard deviation (Elekdag and Wu 2011, p. 7). While Elekdag and Wu used a threshold value of 1.55 to identify a credit boom, meaning that a credit boom occurs when the deviation of real credit of a country from its trend is 1.55 times larger than trend s own standard deviation, I use a value of 1.0 to determine the threshold. The reason to apply a lower threshold is that in comparison with real credit, the fluctuations in consumption are usually less intense. Real credit could increase as fast as 20 percent over a year due to certain external shocks; however, a growth of real consumption of more than 10% is generally considered uncommon (Sundaresan, 1989). Nevertheless, this paper also employs the threshold value of 1.55 to identify a strictly defined consumption boom and tests the sensitivity of the results. A value of 1.55 as threshold is corresponding to the case that deviation of real 13

19 consumption from trend is in the top 6 th percentile of the distribution (Elekdag and Wu, 2011). The trend of a country s consumption path is outlined by de-trending method, popularized by Hodrick and Prescott (1997), the HP filter. Two things need to be noticed about the method of identifying consumption booms: 1) the thresholds are country-specific, meaning that levels of consumption expenditure are only compared against each other across time within a same country; and 2) this paper uses real consumption instead of a consumption-to-gdp ratio, which Montiel employs in his study. This is because there are situations when both consumption and GDP fall, while the ratio increases because GDP falls more rapidly. This would lead to misrecognition of consumption booms when consumption expenditure is actually falling. B. An Illustration of the Identifying Method Figure 1: De-trending Method using China as an Illustration Source: author s own illustration based on the WDI database,

20 The case of China could serve as a useful illustration. Figure 1 above conveys the methodology very intuitively. The graph shows deviation of real consumption from trend the cyclical component (measured in percentage change) in China from In the example, some of the cyclical components exceed the cut-off threshold, identified by the dashed horizontal line and equal to 1.55 times the standard deviation of the cycle of the logarithm real consumption. A lower threshold using one standard deviation is also demonstrated by the 3-dot-dash line and is used to pin down the start and end dates of a boom identified with 1.55 threshold value (Elekdag and Wu, 2011). Thus, according to a threshold value of 1.00, two consumption booms have occurred in China since 1985 started in1987 and ended in 1989, with a peak in 1988, and the other one started in1999 and ended in But according to the stricter threshold value of 1.55, only one consumption boom from is identified. In addition, as shown in the figure, the cyclical components of consumption failed drastically after its peak in Therefore, the consumption boom in China during was part of a boom-bust cycle. C. Distribution of Consumption Booms Implementing the previously mentioned methods using a threshold value of 1.55 (strictly defined boom) on annual data for 83 countries going back to 1973, I identify 62 boom episodes, with an average length of slightly more than three years. 23 booms are in high-income countries, 27 in middle-income countries and 12 in low-income countries. A considerable number of consumption booms occurred immediately before the major economic or financial crises in 70s, 80s and 90s. Appendix 1 presents the distribution of booms across the 83 countries that I have studied. 15

21 II. Independent Variables A. Key Independent Variable the Financial Reform Index As the measurement of the degree of financial liberalization, this paper uses a very comprehensive index, the Financial Reform Index, proposed by Abiad, Detragiache, and Thierry (2008). This measurement recognizes the multi-faceted nature of financial reform and records financial policy changes along seven different dimensions: credit controls and reserve requirements, interest rate controls, entry barriers, state ownership, policies on securities markets, banking regulations, and restrictions on the capital account. Liberalization scores for each category are combined in a graded index that is normalized between zero and one (Abiad et al., 2008, p. 3). This measure is distinct from most former measures, which use dummy variables to encode an event of financial liberalization. The dataset for the Financial Reform Index covers a diverse range of countries, both in terms of regions and income levels. It includes a period of over 30 years, primarily from 1973 to As data for the financial reform index is not available after 2005, this paper has excluded episodes of consumption boom that have been identified after The study thus bases its estimates on this reduced sample. Summary statistics for the index and each of its components are in the Appendix 4. 16

22 B. Control Variables In principle, a consumption boom may be the result of a variety of macroeconomic factors. There is no reason to believe that booms share one cause or a set of causes. Nevertheless, controlling for the relevant variables that are believed to contribute to the appearance of a consumption boom would enable us to look at the partial effects of financial reform exert on boom episodes. As mentioned in the literature review section, Montiel (2000) categorizes the proposed causes of consumption booms from various theories into four different groups: Financial Liberalization, Wealth Effects, Changes in Intertemporal Relative Prices, and Significant Income Redistribution. Figure 2 below summarizes the variables on a macroeconomic level that most comprehensively and closely represent each category. Figure 2: Proposed Determinants of Consumption Booms Variables Expected Sign I. Degree of Financial Liberalization Financial Reform Index + Real Credit to the Private Sector + II. Changes in Intertemporal Relative Prices Real Interest Rate (%) - Nominal Exchange Rate (%) - Inflation Rate (%) + Real Exchange Rate (%) + Nominal Exchange Rate (%) + III. Wealth Effects GDP Growth (%) + Average Housing and Stock Prices + Terms of Trade + Current Account Balance (% to GDP) - Government Consumption (% to GDP) - IV. Significant Income Redistribution Change of the Gini Coefficient + 17

23 For the degree of financial liberalization, as mentioned previously, I have selected the Financial Reform Index proposed by Abiad et al. (2008), instead of credit expansion, which Montiel used in his study. As is shown in Table 1 below, the Financial Reform Index scores in my sample range from the minimum value to the maximum value, which is from 0 to 21. Overall, the scores have a mean of about 10.3 across all the sample countries over a 30-year period. For changes in intertemporal relative prices, I have selected real interest rate, inflation rate, and exchange rate stability index. I left out nominal interest rate because nominal interest is the direct result of the real interest rate and the expected inflation rate, and including it might create perfect co-linearity problem. As Table 1 demonstrates, inflation rates have extremely large variations across the whole sample. This generally helps us identify the effects, if any, of inflation on consumption. In addition, problems of multi-colinearity would occur if I included both the real and the nominal exchange rate. Thus, I use a measurement of exchange rate volatility called the exchange rate stability index (see, Aizenman, Ito, and Chinn, 2013), a score ranging from 0 to 1, to replace them. In terms of choosing variables to represent the role of wealth effects, I picked GDP growth rate, government consumption and current account balance. Average asset prices, including housing prices and stock prices, would also be good indicators of wealth effects and their sharp increase have indeed characterized a lot recent booms (Attanasio et al., 2009). Unfortunately, currently there are no datasets for average asset prices that provide enough information for a sufficient number of countries. As for the variable terms of trade, I excluded it because for one reason, the information that has been collected cannot cover all the years of the sample countries, and for another, it is strongly negatively correlated with current account balance. An improvement in the terms of trade would lead to a fall in exports and an increase in 18

24 imports. Therefore, the current account deficit is likely to be worsened. 1 To measure income redistribution, the best indicator available by far is the Gini coefficient. However, the changes of the index between consecutive years are generally too insignificant to run regression on. For this reason, I leave out the income-redistribution hypothesis from my estimation models. 2 Most of the data described above are gathered from the World Development Indicators, the primary World Bank database of development indicators, compiled from officially recognized international sources. The GDP data are collected from the World Saving Database. Summary statistics for all the independent variables are presented below in Table 1. Table 1: Summary Statistics for the Independent Variables (83 countries, ) 1 However, this relies on the Marshall-Lerner Condition. The Marshall Lerner condition states that if demand for exports and imports is relatively elastic then an increase in terms of trade will worsen the current account. 2 Montiel also excluded this hypothesis in his study due to data availability. (Montiel, 2000) 19

25 V. REGRESSION RESULTS A) Logistic Estimation The results of Logistic estimation and OLS estimation are presented in Table 2.1 and Table 2.2 below. Each column of the tables represents a model specification of which the name is shown at the top of the columns. As a simple procedure to reduce the potential effects of feedback from booms to the independent variables, equations for each model replace all explanatory variables with their lagged values or lagged first-differenced values. For example, the Financial Reform Index (FRI) of Mexico in 1990 is used to estimate its effects on the probability of the occurrence of a consumption boom in the To rule out the influence of trend effects and country-specific characteristics, year fixed effects are implemented in all models except for Logit I and country fixed effects are used in Logit IV and Logit V. One shortcoming of fixed effects Logit models is that they can t estimate the effects of the independent variables in which there are no variations. In other words, countries that have never had consumption booms will be excluded from the sample. Thus, in addition, I use country random effects in Logit III to complement time-specific and country-specific models. Overall, the coefficients of FRI suggest that changes in financial liberalization level have played an important role in generating consumption booms observed in the sample, no matter what threshold values are used to identify the boom episodes (1.55 for Table 2.1 and 1.00 for Table 2.2). The estimates of the fixed effects regressions and random effects regressions all approach statistical significance with the hypothesized positive signs. In Logit III, where 78 countries are observed, the coefficient of FRI is significant at 1% level. The model estimates that, 20

26 Table 2.1: OLS & Logit Coefficients (Standard Errors) for Models Predicting the Occurrence of Consumption Booms (1) (2) (3) (5) (7) (4) (6) (8) EQUATION VARIABLES Logit I Logit II Logit III Logit IV Logit V OLS I OLS II OLS III Consumption Boom Financial Reform Index (FRI) ** 0.117*** 0.227** * ** * (Threshold value: 1.55) (0.0228) (0.0649) (0.0363) (0.114) (0.116) ( ) ( ) ( ) FRI Squared ( ) ( ) ( ) ( ) GDP Growth (%) 0.172*** 0.257*** 0.169*** 0.257*** *** *** (0.0392) (0.0526) (0.0405) (0.0529) (0.0708) ( ) ( ) ( ) FRI*GDP Growth (%) *** * (0.0077) ( ) Current Account Balance (% of GDP) * *** ** *** *** *** *** ** (0.0294) (0.0440) (0.0305) (0.0441) (0.0447) ( ) ( ) ( ) Inflation (%) 2.45e * 4.90e * ** 1.07e e e-06 ( ) ( ) ( ) ( ) ( ) (1.14e-05) (1.14e-05) (1.14e-05) Real Interest Rate (%) *** *** *** *** *** * (0.0102) (0.0267) (0.0112) (0.0269) (0.0256) ( ) ( ) ( ) Exchange Rate Stability Index (0.437) (0.604) (0.496) (0.625) (0.633) (0.0301) (0.0303) (0.0303) Government Consumption (% of GDP) (0.491) (0.130) (0.0749) (0.131) (0.133) ( ) ( ) ( ) Constant *** *** (0.128) (0.131) (0.131) Observations 1, , ,510 1,510 1,510 R-squared Number of Countries Year Fixed Effects Country Fixed Effects Country Random Effects 21 Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 Source: author's own calculations. Note: a. Lagged values are used for all variables except the inflation rate and government consumption b. Lagged first-difference values are used for inflation rate and government consumption

27 Table 2.2: OLS & Logit Coefficients (Standard Errors) for Models Predicting the Occurrence of Consumption Booms (1) (2) (3) (5) (7) (4) (6) (8) EQUATION VARIABLES Logit I Logit II Logit III Logit IV Logit V OLS I OLS II OLS III Consumption Boom Financial Reform Index (FRI) *** 0.187** ** * (Threshold value: 1.00) (0.0183) (0.0491) (0.0248) (0.0868) (0.0894) ( ) ( ) ( ) FRI Squared * * * ( ) ( ) ( ) ( ) GDP Growth (%) 0.154*** 0.193*** 0.144*** 0.196*** *** *** (0.0310) (0.0385) (0.0311) (0.0389) (0.0673) ( ) ( ) ( ) FRI*GDP Growth (%) ** ** ( ) ( ) Current Account Balance (% of GDP) * *** * *** *** *** *** *** (0.0201) (0.0296) (0.0204) (0.0296) (0.0301) ( ) ( ) ( ) Inflation (%) 9.08e e e e-05 ( ) ( ) ( ) ( ) ( ) (1.30e-05) (1.30e-05) (1.31e-05) Real Interest Rate (%) ( ) (0.0113) ( ) (0.0114) (0.0115) ( ) ( ) ( ) Exchange Rate Stability Index (0.357) (0.467) (0.383) (0.478) (0.479) (0.0344) (0.0346) (0.0346) Government Consumption (% of GDP) (0.356) (0.0650) (0.0667) (0.0644) (0.0636) ( ) ( ) ( ) Constant *** *** (0.146) (0.149) (0.150) Observations 1,510 1,142 1,367 1,142 1,142 1,510 1,510 1,510 R-squared Number of Countries Year Fixed Effects Country Fixed Effects Country Random Effects 22 Standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1 Source: author's own calculations. Note: a. Lagged values are used for all variables except the inflation rate and government consumption b. Lagged first-difference values are used for inflation rate and government consumption

28 on average, a one-point increase of the score of FRI is associated with about a 2.3% increase in the probability of having a consumption boom, holding other variables constant. For example, India currently has FRI score around 11. If India fully liberalizes its financial system and raises the FRI score from 11 to 21, then the chances of that country having a consumption boom in the next year is predicted to be 23% more than without a financial liberalization, if other factors are held constant. The effects of GDP growth in generating consumption booms are also substantial. For the first four Logit models in both Table 2.1 and Table 2.2, the coefficients of GDP Growth are all significant at the 1% level. This implies that a wealth mechanism is operating through higher growth of real GDP. As the country s economic growth is accelerating and the private and public sectors share this increase in national wealth, households perceive an increase in their wealth and adjust their consumption paths accordingly. Therefore, a boom may result. However, when the interaction term of Financial Reform Index and GDP Growth is added to the regression (Logit V), the significance of GDP increase is washed away. Instead, the interaction terms become statistically significant at the 1 % level in both Table 2.1 and 2.2. This corresponds with Lehmussaari s view that financial liberalization might stimulate consumption by making the private sector more responsive to wealth changes, hence boosting the influence of wealth effects on consumption. With financial reform, households perceive that economic growth will bring about more wealth for them and thus expand their consumption. Apart from FRI and GDP growth, booms also tend to be strongly negatively associated with current account balances. Having buffered endogeneity issues by taking lagged values of the current account balance, the regressions show that large current account deficit actually finance the increased real consumption. However, while the current account balance is rather 23

29 significant in generating booms, the effects of real exchange rate changes are not obvious. This suggests that people don t dramatically increase their consumption expenditure in response to temporary real exchange rate appreciations, but are subject to other reasons that are also substantially driving the current account balance into deficit. A plausible cause might be huge foreign borrowing due to the liberalization of the capital account or of the security market. While a current account deficit must be accompanied by capital imports, it does not necessarily mean that domestic investment would be higher. 3 Instead, the domestic savings rate could decrease, so that consumption rises with investment staying flat. Klemm (2013) shows in his study that domestic investment may not increase as a result of capital imports, not even when the value of domestic assets increases. More precisely, the consequence of current account deficits could be that capital owned by residents is sold to foreigners, with no change in the stock of capital located in the economy. And residents reduce their net holdings of productive capital and consume the proceeds (Klemm, 2013, p. 7). Thus consumption booms occur. As for government consumption and inflation, the results provide little evidence that booms are systematically associated with them. The signs of the coefficients do not align with the proposed hypothesis by Montiel (2000) nor are they significant at any traditional levels. Despite the fact that reduced government consumption relative to GDP will lead people to perceive that their wealth is going to increase because of the reduced present value of their future tax liabilities (i.e., Ricardian Equivalence), it seems that households do not systematically change their consumption patterns. The results of the impacts of the real interest rate are mixed. While significant in Table 2.1, the significance is weakened if booms are identified with the threshold value of Nickell (2006) argues that, these effects could be huge in some countries, such as the UK, where the current account deficit is small compared to the stock of international assets and liabilities (p. 5). 24

30 Nevertheless, the signs are negative in both cases. Those results somewhat verify Rodriguez s hypothesis that consumption responds to the real interest rate. The fall of interest rates might trigger an increase in consumption spending as the private sector favors consuming over saving. B) OLS estimation The OLS estimations generate results that are very similar to the results of the logistic regressions. The coefficients of FRI all have theoretically appropriate signs and are statistically significant or marginally significant across all OLS models. In OLS I of Table 2.1, it is estimated that with a ten-point increase of the FRI score, the probability of a consumption boom occurring will increase by 6.6%. The predicted effect is not strong in magnitude, yet is still statistically significant at the 10% level. C) Non-linear relationship between FRI and the probability of a boom occurring Perhaps the most important finding of this paper from the empirical models is that when including a quadratic term of the Financial Reform Index in the models, the term has generally produced negative coefficients in almost all specifications. The non-linear effects are especially significant in Table 2.2 where booms are measured according to the threshold value of The positive coefficients of FRI across all model specifications are in line with this paper s hypothesis that as countries liberalize their financial markets and increase their FRI score, the probability of experiencing a consumption boom increases. However, the negative coefficients on the FRI squared term suggest that, if countries are already at the stage of a fairly-open or a well-developed financial markets, by further implementing financial reforms, the odds of a consumption boom occurring will actually be reduced. These results are in line with the consumption smoothing theory of financial liberalization (Kose, Prasad, and Terrones, 2003). D) Regressions Across Income Groups 25

31 To test if this non-linearity holds for developed and developing markets, I split the sample into income groups (low income, middle income, and high income) and run the models independently across the groups. As Table 3.1 and 3.2 demonstrate, the non-linear relationship between the level of financial depth and the chances of a consumption boom holds only for countries with high income per capita. With middle-income group, the results are mixed. In Table 3.2, the signs are in line with the case of high-income countries, but the coefficients are only marginally significant at about 20% level. In Table 3.1, the signs for the FRI term and the FRI squared term are reversed and insignificant. In countries with low income per capita, there is no significant relationship between financial depth and the occurrence of consumption booms. These observations are consistent with the story of the effects of financial liberation on economic growth across income groups (Deidda and Fattouh, 2002). The evidence suggests that for the chances of generating a consumption boom, the initial financial depth plays a significant role. So does the initial income level of the country. In fact, these two are closely associated. For low-income countries, their financial markets are generally underdeveloped, for many years the vast of majority of which get a FRI score under 7, one-third of the full score. Whereas, for high-income countries, the distribution of FRI scores have a much wider range, as they change significantly along the course of their development. As a result, the great variations of their FRI scores over the years support the sensitivity of the non-linearity. For the middle-income group, the results also indicate that high FRI scores are associated with a high probability of booms. However, if the level of financial liberalization passes a threshold, the chances of having booms will actually reduce. However, the significance of the non-linearity for mid-income countries might be weakened by big standard errors of the estimates. This is because 26