Financial Reform:
An empirical assessment of its determinants
By
Eelco Zandberg
Thesis submitted in partial fulfillment of the requirements for the
Research Master in Economics and Business
August 2009
2
Table of Contents
1. Introduction
2. Literature Review
2.1 Discussion model, results and conclusions of Abiad and Mody (2005) 2.2 Other literature
3. Theory
3.1 Political economy of reform 3.2 Determinants of reform 4. Data
4.1 Data description
4.2 Analysis of various aggregation methods 5. Estimation Methods and Results
5.1 Bivariate relationships
5.2 Results estimation model Abiad and Mody (2005) with new dataset 5.3 Econometric model
5.4 Results estimation new model with dataset Abiad and Mody (2005) 5.5 Results estimation new model with new dataset
5.6 Sensitivity analyses 6. Summary and Conclusions References
3
1. Introduction
Financial reform can be defined as the movement of financial systems from government ownership or control towards greater private provision of financial services under fewer operational restrictions (Abiad and Mody, 2005). Examples of specific policy actions are the abolishment of credit controls and interest rate controls, privatization in the financial sector and the easing of restrictions on international financial transactions. In order to clarify terms it is good to note that financial reform refers to the process of moving towards more market-oriented financial systems, while the term financial liberalization denotes the particular state the financial system is in. So, increases in the level of financial liberalization are achieved by financial reforms. Although financial liberalization has been blamed by many as one of the causes of the worldwide financial crisis of 2008, it has also been welfare-improving because it has led to higher economic growth rates, by improving financial sector development (Bekaert et al., 2005; Klein and Olivei, 2008; Quinn and Toyoda, 2008). However, other studies point to the effects of financial liberalization on banking crises (Demirgüç-Kunt and Detragiache, 1998, 2000; Mehrez and Kaufmann, 2000). All in all, whether financial liberalization is a good thing as such has not been firmly established yet, and will also not be answered in this thesis.
So, although there exists a large literature that examines the consequences of financial liberalization, there are only a few studies that address the causes of these liberalization processes. It is very useful to establish what the economic effects of financial liberalization are, but maybe even more interesting to examine what has led to these reform efforts in the first place. Therefore, the research question of this thesis is what the main determinants of financial reform are and in which particular way they operate. In order to answer this question we perform several analyses. We will do this by building upon and improving Abiad and Mody (2005), the most important and comprehensive study on this subject until now. This thesis contains three main innovations, compared to Abiad and Mody (2005). First, we use a new dataset on financial liberalization that consists of 91 countries for the period 1973-2005. Second, we employ a different, and we believe, better methodological framework in order to disentangle the different forces at work in the financial reform process. Finally, we test the validity of several hypotheses from the literature that have not been tested before.
4
inflation or IMF programs. Contrary to other studies, we do not find evidence that it matters whether governments are in the first year of their term or whether world interest rates are high. Furthermore, political variables do not seem to play a role in the financial reform process as our results suggest that the level of democracy, government fractionalization, political ideology and the system of government (presidential or parliamentary) are not related to financial reform. Finally, legal origin and trade openness are also not important.
This thesis consists of six chapters. Chapter 2 provides a literature review that consists of two parts. The first part discusses Abiad and Mody (2005) critically and extensively, while the second part contains an overview of other important papers in the field. Chapter 3 describes the theory behind the political economy of reform and discusses the possible determinants of financial reform one by one. Chapter 4 describes all the data we use and analyzes various methods to aggregate the separate components of the financial liberalization index. Chapter 5 is the core of the thesis. It starts with the analysis of some bivariate relationships between possible determinants and financial reform. Subsequently, it shows estimations of the model of Abiad and Mody (2005) using a new dataset. The third section presents our econometric model and the fourth and fifth sections contain the results of the regression analyses of this new model using respectively the dataset of Abiad and Mody (2005) and the new dataset. The last section of this chapter provides results from sensitivity analyses. Finally, chapter 6 provides a summary and concludes.
2. Literature Review
2.1 Discussion model, results and conclusions of Abiad and Mody (2005)
5
the particular variable is systematically related to financial reform. These analyses show that new governments are more inclined to dislodge the status quo with either large reforms or large reversals. High U.S. interest rates are associated with less reform and IMF programs have a positive reform bias (note that the chi-square statistics are not significant for these last two variables). Balance-of-payments crises lead to large reforms while banking crises are associated with reversals. Recessions and periods of high inflation lead to reforms as well as reversals. As to the effect of the current state of liberalization on financial reform, it seems that countries with fully repressed or fully liberalized financial sectors tend to stay that way, while countries that are partially repressed or largely liberalized are most likely to carry out reforms. These results show that there is not a pure convergence effect at work in the sense that the current state of liberalization is negatively related to financial reform, but that reforms are most likely in an intermediate range of liberalization. Finally, political ideology of the government, system of government (parliamentary or presidential) and trade openness do not appear to be in any way related to financial reform.
In order to formally test several hypotheses they develop an empirical framework that accounts for domestic learning and regional diffusion effects. Domestic learning means that it might be the case that triggers for new reform episodes are provided by former reforms, so that once a country has started on a reform path it will continue. Regional diffusion would show up in interdependencies between countries from the same region in their reform efforts. Besides these two effects, they add several explanatory variables related to crises (dummies for banking crises, balance-of-payments crises, recessions and high-inflation periods), shocks (new government, U.S. interest rates, IMF program) and some other variables (government ideology and trade openness). Their benchmark specification is as follows:
6
The second term between brackets represents the regional diffusion effect; the further away a country is from the regional leader, in terms of its liberalization level, the larger this term. The regional leader is defined as the country in the region with the highest level of financial liberalization. They divide the world into five regions, namely Latin America, East Asia, South Asia, Africa/Middle East and the OECD countries.
The vectors SHOCKS, IDEOLOGY and STRUCTURE consist of exogenous variables such as several crisis indicators, shock variables (new government dummy, IMF program dummy and world interest rates), political ideology of the government and trade openness.
Besides the benchmark specification, they come up with a few other specifications as well. First, they drop the assumption that the desired level of financial liberalization equals 1, which leads to the following specification:
Second, the desired level of financial liberalization could depend on the country‟s level of economic development, which implies the following specification:
Finally, the effects of the other explanatory variables (including the regional diffusion effect) might be dependent on the level of financial liberalization and are thus interacted with FLi,t-1:
7
specification with interaction terms must be taken with great care. This can be shown by considering the following model:
The marginal effect of X on Y is β1+β3Z and the standard error of this marginal effect is . It might thus be the case that X is significant for
some values of Z and insignificant for others. This can simply not be inferred from the results that standard packages report.
Because of the discrete, ordinal nature of the dependent variable, AM use the ordered logit method to estimate equations 1-4. They estimate each specification with and without country fixed effects. Based on these regressions and the bivariate relationships presented before, they reach five specific conclusions:
1. Countries whose financial sectors were fully repressed were most likely to maintain the status quo. However, once initial reforms occurred, the likelihood of further reforms increased substantially.
2. Countries that were further away from the liberalization level of the regional leader liberalized their financial sectors faster; regional diffusion effects appear to have been important.
3. Several shocks affected the likelihood of reforms. New governments tended to alter the status quo, through both reforms and reform reversals, although reforms were more likely when the liberalization level was also low. Lower U.S. interest rates increased the probability of reform, while IMF programs had a strong positive effect when liberalization levels were low and a much smaller effect when liberalization levels were higher.
4. Crises also affected the chances of reform. Balance-of-payments crises increased the likelihood of reform while banking crises increased the likelihood of reversals.
8
As already pointed out by Huang (2009), the coding of the regional dummies of AM contains a small mistake; Singapore is classified as a country from Africa/Middle East and South Africa as an East-Asian country. These two should of course be interchanged. However, besides this minor error we believe that the model of AM contains several serious flaws. We will explain them one by one.
First, the regional diffusion effect is modelled rather ad-hoc. It is not clear to us why countries would only be influenced by other countries within their own region, for example. Furthermore, South Asia only contains five countries, which makes it difficult to speak of a true regional catch-up effect, as most of the region is simply not taken into account. The most important problem, however, is that the level of financial liberalization of the regional leader is almost time-invariant in some regions (especially for the OECD countries and East Asia), as can be seen in figure 1 that plots the level of financial liberalization for each regional leader against time:
Figure 1 Financial liberalization of regional leader plotted against time
This leads to the following problem: The term that should measure regional diffusion is
9
will not measure regional diffusion but instead a simple convergence effect. So, from the results one should merely conclude that countries with more repressed financial sectors are more induced to reform than countries with relatively liberalized financial sectors.
Second, we do not understand why AM call the term FLi,t-1(1-FLi,t-1) domestic learning. To us it is simply a term to capture possible non-linearities in financial reform efforts. A positive coefficient on this term would indicate that financial reform is most intensive at intermediate levels of liberalization, which is exactly what AM find.
Third, the inclusion of a large amount of interaction terms, as AM do in the final specification, is dangerous with respect to multicollinearity problems. Besides that, we do not really see the need of interacting all possible determinants with the initial level of financial liberalization.
Fourth, they describe several sensitivity analyses such as adding regional dummies, legal origin and the system of government (presidential or parliamentary). However, these sensitivity analyses can only refer to the regressions without country fixed effects, because these are all time-invariant variables. The inclusion of time-invariant variables and country fixed effects together is impossible as it would lead to perfect collinearity. Therefore, the sensitivity analyses of AM are only valid for the results without country fixed effects.
Finally, as Brambor et al. (2006) have pointed out, one should include all elements of an interaction term also separately in a regression analysis. However, in (3) the current state of liberalization is interacted with the level of economic development while the latter does not appear separately in the model. This can lead to wrong conclusions and should be avoided. In order to overcome these modelling problems we propose a different econometric model, which we will present in section 5.3. However, before that we will discuss other relevant literature in this field.
2.2 Other literature
10
All in all, some of the conclusions of AM are confirmed, but it also reaches a few distinct conclusions. One of them is that some of the findings of AM on the effects of crises and shocks can not be confirmed; it shows that the significant effects of AM of balance-of-payments crises and US interest rates are fragile. Furthermore, it suggests that policy change in a country is negatively rather than positively related to its liberalization level, and the liberalization gap from the regional leader appears less relevant than in AM. Finally, it observes a significant negative effect of democracy on financial reform. Although there is nothing wrong with this study, it is not really revolutionary, as the differences with AM are only minor.
Burgoon et al. (2008) re-examine the role that political factors play in the financial reform process. They employ the same methodological framework as AM, but explore the „political factor‟-hypotheses in much more depth. Their findings challenge the conclusion of AM that political variables do not play an important role in the financial reform process. They divide the financial reform process in an „input‟ and „output‟ side or „phase‟. The input side concerns processes such as the involvement of and acceptance of stakeholders of the decision to initiate (or continue with/reverse) financial reforms. The output side concerns the effects of these policy changes on the perceived interests of these stakeholders. As the input side is concerned, their empirical results suggest that shifts to left-wing governments decrease the chance of liberalization at low levels of democracy, but that this effect disappears at higher levels of democracy. International voter support for free-market internationalism, as opposed to anti-capitalist closure, is found to be conducive to financial reform. For the output side, some targeted social policies, such as health spending, and multilateral and bilateral aid are found to increase the chances of financial reforms taking place. One of the main problems with their paper is the interpretation of the statistical significance of variables that appear individually as well as in interaction terms in the regression model. For example, the variable „leftist government‟ is reported to be significantly negative at a 1% level, but also appears interacted with the level of democracy in the same regression equation. As Brambor et al. (2006) have pointed out, it is wrong to use individual standard errors to base a conclusion of whether a variable is significant or not upon, in the presence of interaction terms (for a more extensive discussion of this issue, see section 2.1).
11
Comparative Manifesto Project (CMP) (see Budge, 1992; Volkens, 1995; Kim and Fording, 2003). Besides that, they also use home country vote shares of the CP and the CMP Item 401 data to examine whether the domestic public opinion on capitalism influences international financial liberalization. Their dependent variable is the change in capital account regulation, so they focus on only one aspect of financial liberalization. The main conclusion from their analyses is that both domestic and global ideas are powerful influences of international economic policy. However, their conclusions seem trivial to us. We find it hard to see why it would not be the case that over longer time periods in democracies liberalization will take place if the public opinion supports it and will not take place if they oppose it.
Omori (2007) examines what the political determinants are of the magnitude of financial reforms in developing countries. His results suggest that IMF programs have a significant positive impact on the magnitude of financial reforms when the number of veto players in the government is small, but that this effect vanishes when the number of veto players becomes larger. Unfortunately, this conclusion suffers from the same problem with interpreting statistical significance of interaction variables, as described above. Other conclusions are that the magnitude of financial reforms is higher in countries with non-democratic governments and a large manufacturing sector.
Lora (2000) tests hypotheses about the determinants of structural reform for 20 Latin American countries for the period 1985-1995. The main conclusions with respect to the hypotheses about financial reform are that new governments, inflationary crises and capital flows to Latin America as a whole make financial reforms more likely.
3. Theory
3.1 The political economy of reform1
Economic reform programs have to cope with economic and political constraints. Economic constraints concern issues such as technical limits on what kind of reforms can be implemented reasonably and problems of incomplete information about the post-reform environment. The latter complicates policy-making because the policy maker will adopt different policy measures under incomplete information than under complete information.
12
Far more important, however, are political constraints. In order to implement reform measures governments need some kind of support from the general public. This is especially relevant in democracies, where governments run the risk of losing the elections after unpopular reforms have been carried out; but also autocratic regimes need some amount of public support in order to prevent violent resistance against the regime. Or, as Drazen (2000) puts it: for a
program of reform and transition to succeed, it must have the necessary political support at crucial decision stages. This relates to the concept of heterogeneity. If all consumers would
be the same, policy could be based on economic considerations alone and no specific problems would arise. However, in general consumers are characterized by heterogeneity of interests, which means that the political constraint plays a role.
First of all, uncertainty about the effects of a reform program can put a political constraint on the implementation of reforms. For example, as Fernandez and Rodrik (1991) show, there exists the problem of uncertainty about private benefits even if society as a whole is expected to gain. Another issue is that some consumers may experience uncertainty of a kind they have little or no experience with, for example in economies that used to be centrally planned. Secondly, asymmetric information may also put a political constraint on reform programs. This can arise when the government has to compensate individuals in exchange for political support. After the reforms have been carried out it becomes very difficult for the government to decide on the design of fair compensation packages, while individuals have an incentive to overstate their income loss. Therefore, feasible compensation programs are usually very difficult to design.
So, as long as the benefits of maintaining the policy status quo outweigh its costs, the status quo will persist (AM). Some kind of exogenous shock or event is thus needed in order to set the ball rolling. This can be several things: a financial or economic crisis, a new government, participation in an IMF program or reform efforts of other countries.
3.2 Determinants of reform
In this section we will introduce the possible determinants of financial reform one by one. We divide them in four groups, namely crises, shocks, political factors and other variables.
13
Pitlik and Wirth (2003), Heinemann (2004), Hoj et al. (2006) and Pitlik (2007), using different indicators for a crisis, finds evidence in favour of the crisis hypothesis. It should be noted that these studies all focus on general economic reform. A bad economic performance seems most likely to trigger general economic reforms. Examples are recessions and periods of high inflation. However, other types of crises, such as banking crises, debt crises, currency crises or balance-of-payments crises, might lead to predominantly financial reforms. As described in section 2.1 and 2.2, there is already some evidence that banking crises set back financial liberalization and that balance-of-payments crises lead to more financial reform efforts.
The second group of possible determinants are shocks. The honeymoon hypothesis states that new governments are more likely to carry out reforms as they want to realize the benefits of these reforms before the next election (Krueger, 1993). Another possible determinant in this category is IMF programs; countries that want to receive a loan from the IMF are usually forced to reform their economies (including financial sectors). Also, Bartolini and Drazen (1997) argue that access to cheap international capital increases the likelihood of financial reforms. Therefore, world interest rates, as a measure of the price of international capital, might matter.
14
AM do not consider the possible effects of the level of democracy on financial reform. However, according to Huang (2009), democracy is negatively related to financial reform. This is a remarkable finding in light of the empirical evidence of a positive relationship between democracy and general economic reform (Dethier et al., 1999; De Haan and Sturm, 2003; Pitlik and Wirth, 2003; Lundström, 2005; Pitlik, 2008). Therefore, we consider it useful to examine the possible relationship between democracy and financial reform.
Another political factor that might matter is government fractionalization. This is also ignored by AM, despite the fact that there are theoretical reasons to suppose it is important. Krueger (1993) argues that “strong” governments will react quickly to crises, while “fractional” governments will delay stabilizations. The more fractionalized a government is, the harder it will be to reach a consensus on possible controversial reform measures.
Finally, Persson (2002) argues that presidential forms of government will find it easier to make tough decisions than parliamentary systems as the latter will be more prone to problems related to conflicting interests.
In the category of „other variables‟, we consider trade openness and legal origin; AM find that trade openness has a significant positive impact on financial reform, especially at low levels of liberalization. A possible reason why trade openness would matter for finanical reform is that the more a country trades with the rest of the world, the higher might be the internal pressure from companies within the country for financial reform as it would facilitate their international business transactions considerably. Finally, La Porta et al. (1997) and Levine et al. (2000) find that legal origin matters for one measure of financial liberalization, namely the ratio of externally held equity market capitalization to GNP. Although AM do not find a significant effect of legal origin, we believe it is worthwhile to include the variable in our analysis.
4. Data
4.1 Data description
In this section we will describe all the data we use and provide summary statistics and data sources.
15
1. Credit controls and excessively high reserve requirements 2. Interest rate controls
3. Entry barriers
4. State ownership in the banking sector 5. Capital account restrictions
6. Prudential regulations and supervision of the banking sector 7. Securities market policy
Each component has been assigned a score between 0 and 3 with 0 representing the highest degree of repression and 3 full liberalization. These scores are then summed to obtain a value between 0 and 21 as a measure of overall financial liberalization. To facilitate interpretation, these values can be normalized to obtain values between 0 and 1.
The database covers 72 countries over the period 1973-2005 and 19 countries (mainly former-communist) over a shorter time span. It is an update of an earlier version that covered 36 countries over the period 1973-1996 and consisted of six slightly different categories of financial liberalization (it excluded prudential regulations and securities market policy, but included a measure of operational restrictions).
There are alternative measures of financial liberalization available. Williamson and Mahar (1998) constructed a dataset for 34 countries over 1973-1996 with six dimensions. Kaminsky and Schmukler (2003) provide an index with three components: domestic financial sector liberalization, capital account liberalization and the openness of the equity market to foreign investment. Their dataset covers 28 countries over the period 1973-1999. Finally, Bandiera et al. (2000) and Laeven (2003) composed datasets with six dimensions of financial liberalization, but with binary variables for each component, and a very small country coverage (8 and 13 resp.). All in all, we consider the dataset of Abiad et al. (2008) as the most comprehensive source of information on financial liberalization at this moment.
Table 1, that comes from Abiad et al. (2008), provides summary statistics for the financial liberalization index and its separate components:
Table 1 Summary Statistics for Financial Liberalization Components and Index
Number of Standard
Variable Observations Mean Deviation Minimum Maximum
16
Interest Rate Controls 2671 1.778 1.324 0 3
Entry Barriers 2671 1.769 1.179 0 3
Bank Regulation and Supervision 2671 0.776 0.958 0 3
Privatization 2671 1.248 1.187 0 3
Capital Account 2671 1.668 1.135 0 3
Securities Market 2671 1.490 1.129 0 3 Financial Reform Index 2671 10.321 6.333 0 21 Financial Reform Index (normalized) 2671 0.491 0.302 0 1
17
otherwise. These data come from La Porta et al. (2008). Table 2 provides summary statistics of all the variables:
Table 2 Summary Statistics Other Variables
Number of Standard
Variable Observations Mean Deviation Minimum Maximum BANK 2671 0.094 0.292 0 1 CUR 2671 0.134 0.341 0 1 DEBT 2671 0.042 0.200 0 1 RECESSION 2524 0.151 0.358 0 1 HINFL 2476 0.082 0.274 0 1 FIRSTYEAR 2495 0.196 0.397 0 1 IMF 2638 0.342 0.475 0 1 USINT 2671 5.895 2.868 1.01333 14.0775 RIGHT 2495 0.305 0.460 0 1 LEFT 2495 0.324 0.468 0 1 CENTER 2495 0.076 0.265 0 1 OPEN 2546 64.281 40.511 6.320343 447.296 DEMO 2621 3.390 7.005 -10 10 PARL 2488 0.397 0.489 0 1 PRESID 2488 0.603 0.489 0 1 GOVFRAC 2495 0.201 0.276 0 1
4.2 Analysis of various aggregation methods
18
factor, while principal components analysis is simply a method to reduce the dimensionality of multivariate data without assuming an underlying model (Lattin et al., 2003).
An underlying factor structure can only be present if the correlations between the variables are high enough. Table 3 shows the correlation coefficients between the components of the financial liberalization index. From the table, it is fair to say that the separate components are correlated substantially and therefore we think it is worthwhile to examine whether an underlying factor structure is present.
Table 3 Correlation matrix 7 components financial liberalization index
cred.cont. intr.contr. entrybar. bank.sup. privat. intlcapital sec.markets
creditcontrols 1.0000 intratecontrols 0.6510 1.0000 entrybarriers 0.5654 0.5500 1.0000 bankingsupervision 0.6084 0.5904 0.5650 1.0000 privatization 0.4942 0.4368 0.4353 0.4808 1.0000 intlcapital 0.5874 0.6055 0.5126 0.5782 0.5172 1.0000 securitymarkets 0.6237 0.6279 0.5447 0.6422 0.4923 0.6760 1.0000
Table 4 shows the eigenvalues of each factor, which are simply the variances. It becomes immediately clear from the table that the first factor already accounts for an enormous part of the total variance. Therefore, based on these results it seems logical to retain only the first factor. Figure 2 confirms this. It is called a scree plot and plots the eigenvalues against the number of factors. The idea of a scree plot is to look for an elbow in the figure and retain all factors to the left of the elbow, which is in this case only the first factor. Therefore, we conclude that only one factor should be retained from the factor analysis.
Table 4 Eigenvalues Factors
19
In order to calculate the contents of this factor we need so-called factor loadings. These are the correlations between the variables and the factor, but also represent how the variables are weighted for each factor. Table 5 gives an overview of the factor loadings for the first three factors. Based on the factor loadings for the first factor we can calculate the contents of the first factor.
Figure 2 Scree plot to determine the number of factors to retain
Table 5 Factor Loadings
Component Factor 1 Factor 2 Factor 3 creditcontrols 0.7912 0.0951 -0.0239 intratecontrols 0.7775 0.0276 -0.0946 entrybarriers 0.7033 0.1162 0.0430 bankingsuperv 0.7729 0.0500 0.0031 privatization 0.6199 -0.0125 0.1190 intlcapital 0.7845 -0.1502 0.0134 securitymarkets 0.8341 -0.1097 -0.0294
20
pairwise correlations are above 90 percent (just as in AM) and most of them are even very close to 100 percent. Based on the factor analysis as such and table 6 we do not see a reason why we can not simply add the separate components in order to obtain an overall index of financial liberalization. Therefore, all our analyses in section 5 will be performed with the normalized (between 0 and 1) sum of the separate components as an overall index of financial liberalization.
Table 6 Correlations among aggregation methods
sum ssq ssqr factor sum 1.0000
ssq 0.9847 1.0000
ssqr 0.9859 0.9423 1.0000
factor 0.9995 0.9841 0.9853 1.0000
5. Estimation Methods and Results
In this section we will describe all the analyses that we perform in order to get a clearer picture of the main determinants of financial reform. Section 5.1 provides an analysis of the bivariate relationships between financial reform and its possible determinants. Section 5.2 shows the results of estimating the model of AM with the new dataset. Section 5.3 explains the econometric model we use. Section 5.4 shows the results of estimating our model with the data of AM and section 5.5 shows the results of estimating our model with the new dataset. Finally, section 5.6 provides results from sensitivity analyses.
5.1 Bivariate relationships
21
emphasize that this does not prove causality and, more important, only shows the bivariate relationship. It might be the case that both financial reform and the particular variable are driven by a third variable, for example. However, we believe it is useful for gaining some insight in the data to take a closer look at these bivariate relationships before analyzing the determinants of financial reform in a more sophisticated econometric framework.
In table 7 we examine the relationship between reform and the current state of liberalization. Fully repressed corresponds to a value of 0 for the level of liberalization, partially repressed to 1, largely liberalized to 2 and fully liberalized to 3. A priori, one would expect some kind of convergence effect in the sense that countries with relatively repressed financial sectors are more likely to reform than countries with financial sectors that are already largely liberalized. The chi-square test shows that the current state of liberalization indeed matters, but a look at the table reveals that countries with fully repressed financial sectors are most likely to keep the status quo, while countries with financial sectors that are largely (but not fully) liberalized are most likely to reform. So, as AM already concluded from their limited dataset, reforms are most likely in an intermediate range of liberalization.
Table 7 Current state of liberalization
Fully Partially Largely Fully repressed repressed liberalized liberalized
Large reform 3.2 8.1 7.9 1.6 Reform 16.4 28.4 33.1 23.3 Status quo 77.8 56.2 53.3 70.1 Reversal 2.6 6.0 5.1 4.8 Large reversal 0.0 1.3 0.6 0.2 Total 100.0 100.0 100.0 100.0 Pearson Chi-sq: 151.14 Prob.: 0.00
22
clear that periods of high inflation shake the status quo, but it seems just as likely to lead to reforms as to reform reversals. The same holds for recessions, although the effect is not as strong as for high inflation periods.
Table 8 Table 9 Banking crisis? Currency crisis?
No Yes No Yes
Large reform 5.1 6.8 Large reform 4.7 8.8
Reform 24.1 29.6 Reform 23.9 29.5
Status quo 66.5 52.4 Status quo 67.0 53.3
Reversal 3.9 9.2 Reversal 4.0 7.1
Large reversal 0.3 2.0 Large reversal 0.4 1.4 Total 100.0 100.0 Total 100.0 100.0
Pearson Chi-sq: 37.51 Pearson Chi-sq: 35.61
Prob.: 0.00 Prob.: 0.00
Table 10 Table 11
Debt crisis? Recession?
No Yes No Yes
Large reform 5.4 1.8 Large reform 4.7 8.3
Reform 24.7 22.7 Reform 24.6 25.0
Status quo 65.3 62.7 Status quo 66.7 57.5
Reversal 4.2 10.0 Reversal 3.8 7.2
Large reversal 0.4 2.7 Large reversal 0.3 1.9 Total 100.0 100.0 Total 100.0 100.0
Pearson Chi-sq: 22.25 Pearson Chi-sq: 36.03
23 Status quo 66.6 47.2 Reversal 4.1 9.2 Large reversal 0.3 2.6 Total 100.0 100.0 Pearson Chi-sq: 88.84 Prob.: 0.00
Tables 13-15 show the results for the „shock‟-variables. World interest rates, that are proxied by US Treasury Bill rates, are defined as high when they are in the highest quartile. As table 15 shows, high US interest rates increase the likelihood of reform reversals. Furthermore, according to table 13 it does not seem to matter at all whether a government is new or not. On the other hand, IMF programs are associated with considerably more reform periods.
Table 13 Table 14 First year in office? IMF program?
No Yes No Yes
Large reform 5.3 5.8 Large reform 3.4 8.8
Reform 24.7 27.5 Reform 20.5 32.2
Status quo 65.1 61.5 Status quo 71.0 54.1
Reversal 4.6 4.4 Reversal 4.4 4.6
Large reversal 0.4 0.8 Large reversal 0.7 0.2 Total 100.0 100.0 Total 100.0 100.0
Pearson Chi-sq: 3.77 Pearson Chi-sq: 91.32
24
Prob.: 0.01
In tables 16-19 we examine the bivariate relationships between several political variables and financial reform. According to the chi-square tests, none of the political factors matters for financial reform. Table 16 reveals that democratic governments reform slightly more than their autocratic counterparts, but the difference is almost negligible. Government fractionalization and political ideology do not really seem to matter, although it should be noted that large reforms are more likely when there is a left- or right-wing government than a centrist government. Whether a country has a parliamentary or presidential system is also not important for financial reform.
Table 16 Table 17
Democracy? Government
Polity≤4 Polity>4 fractionalization Large reform 4.2 6.0 No fract. Govfrac>0 Reform 24.1 25.2 Large reform 5.0 5.6 Status quo 66.3 64.3 Reform 24.8 24.5 Reversal 4.5 4.4 Status quo 64.6 65.7 Large reversal 0.9 0.3 Reversal 4.9 3.9 Total 100.0 100.0 Large reversal 0.7 0.3 Total 100.0 100.0 Pearson Chi-sq: 8.72
Prob.: 0.07 Pearson Chi-sq: 3.76 Prob.: 0.44
Table 18 Table 19
Ideology System of
Left Center Right government
Large reform 6.1 3.9 6.3 Parl. Presid.
Reform 24.1 25.0 26.8 Large reform 5.2 5.5 Status quo 66.0 65.0 62.1 Reform 25.1 25.3
Reversal 3.5 5.6 4.4 Status quo 66.2 63.2
25
Prob.: 0.14 Pearson Chi-sq: 8.69
Prob.: 0.07
Tables 20 and 21 present the bivariate relations between legal origin and trade openness and financial reform. According to the chi-square test legal origin is not significantly related to financial reform, although table 20 reveals that countries with German legal origin carry out slightly more reforms than other countries. This might be explained by the fact that many former-socialist countries have German legal origins. However, these are precisely the countries with low initial levels of financial liberalization, which means that it is highly doubtful that German legal origin, in itself, is a determinant of financial reform. So, all in all, we conclude that legal origin does not really matter for financial reform. Finally, there does not seem to be a significant relationship between trade openness and financial reform.
Table 20 Legal origin
British French German Scandin. Large reform 4.0 5.8 6.1 4.7 Reform 22.9 25.3 26.8 22.7 Status quo 68.1 63.3 63.8 71.1 Reversal 4.7 4.8 3.3 1.6 Large reversal 0.3 0.8 0.0 0.0 Total 100.0 100.0 100.0 100.0 Pearson Chi-sq: 17.59 Prob.: 0.13 Table 21 Trade openness
26
Prob.: 0.59
These bivariate results give some preliminary indications of the relationships between several possible determinants and financial reform. However, in order to efficiently exploit the full information content of the data we need to analyze these possible relationships in a more sophisticated econometric model, which we will present in section 5.3.
5.2 Results estimation model of Abiad and Mody (2005) with new dataset
In this section we estimate the original model of AM with the new dataset that covers more countries over a longer time period to examine whether their results are robust against adding new data. The only difference with AM thus concerns the data here. The new dataset on financial liberalization that we use (Abiad et al., 2008) is much larger (more countries and more years) than the one used in AM. Furthermore, for some explanatory variables the data sources are different from AM. There is one variable in our dataset that differs from the one used by AM; they use BOP to indicate balance-of-payments crises, but due to data limitations we replace this variable by CUR that shows whether a country experienced a currency crisis (see section 4.1). The other explanatory variables are defined in exactly the same way as AM do. Tables 22-26 show the results from the estimations of equations 1-4 with the new dataset. For illustrative purposes the results of AM are also reported.
Table 22 Ordered logit estimates of equation 1 without country fixed effects
new data AM new data AM new data AM FLi,t-1(1-FLi,t-1) 5.031 4.001 4.829 4.652 4.435 4.188
(10.44)*** (4.14)*** (10.07)*** (4.81)*** (8.72)*** (4.25)*** REG_FLi,t-1-FLi,t-1 1.310 0.842 1.285 0.897 1.555 0.993
27 HINFLit 0.292 -0.161 0.056 -0.264 (1.26) (-0.40) (0.25) (-0.66) FIRSTYEARit 0.105 0.194 (1.05) (0.84) IMFit 0.586 0.326 (5.43)*** (1.75)* USINTit -0.027 -0.066 (-1.65) (-1.72)* LEFTit 0.326 0.242 (3.68)*** (1.00) RIGHTit 0.413 0.169 (3.61)*** (0.89) OPENit 0.002 -0.001 (1.34) (-1.05) Log L -3496.24 -762.66 -3095.46 -752.05 -2911.15 -747.20 Number of obs. 2580 805 2293 805 2138 805 Notes: The dependent variable is ΔFLit.
Numbers in parentheses are robust t-statistics adjusted for clustering by country. *, **, *** indicate significance at the 10%, 5% and 1% resp.
Table 23 Ordered logit estimates of equation 1 with country fixed effects
new data AM new data AM new data AM FLi,t-1(1-FLi,t-1) 5.972 6.316 5.900 6.932 5.638 6.295
(7.77)*** (4.18)*** (7.92)*** (4.61)*** (7.46)*** (3.99)*** REG_FLi,t-1-FLi,t-1 2.788 1.940 2.451 1.841 3.079 2.324
28 (-3.03)*** (-2.03)** LEFTit 0.382 -0.072 (2.62)*** (-0.20) RIGHTit 0.461 -0.196 (2.82)*** (-0.58) OPENit 0.000 0.006 (0.11) (0.01) Log L -3429.93 -749.10 -3042.06 -738.93 -2856.83 -734.11 Number of obs. 2580 805 2293 805 2138 805 Notes: The dependent variable is ΔFLit.
Numbers in parentheses are robust t-statistics adjusted for clustering by country. *, **, *** indicate significance at the 10%, 5% and 1% resp.
Table 24 Ordered logit estimates of equations 2 & 3
without country fixed effects
new data AM new data AM
FLi,t-1 4.572 4.161 5.114 4.314 (7.63)*** (4.27)*** (FLi,t-1) 2 -4.399 -4.324 -5.944 -6.091 (-8.60)*** (-4.11)*** (-9.19)*** (-4.41)*** FLi,t-1 x Yi,t-1 0.055 0.103
REG_FLi,t-1-FLi,t-1 1.714 0.896 1.532 0.568
29
OPENit 0.001 -0.001 0.001 0.000
(1.19) (-0.84) (1.03) (0.11) Log L -2910.98 -747.13 -2903.90 -744.94 Number of obs. 2138 805 2138 805 Notes: The dependent variable is ΔFLit.
Numbers in parentheses are robust t-statistics adjusted for clustering by country. *, **, *** indicate significance at the 10%, 5% and 1% resp.
Table 25 Ordered logit estimates of equations 2 & 3
with country fixed effects new data AM new data AM FLi,t-1 5.184 6.130 7.872 6.662
(6.25)*** (4.00)***
(FLi,t-1)2 -5.926 -6.701 -9.984 -9.763
(-7.60)*** (-3.28)*** (-8.02)*** (-3.70)*** FLi,t-1 x Yi,t-1 0.183 0.234
REG_FLi,t-1-FLi,t-1 2.329 1.902 3.621 2.070
30
Number of obs. 2138 805 2138 805 Notes: The dependent variable is ΔFLit.
Numbers in parentheses are robust t-statistics adjusted for clustering by country. *, **, *** indicate significance at the 10%, 5% and 1% resp.
Table 26 Ordered logit estimates of equation 4 new data AM new data AM
Without country With country fixed effects fixed effects FLi,t-1 8.462 3.824 5.741 4.486
(FLi,t-1) 2
-7.290 -3.644 -5.891 -2.930 REG_FLi,t-1-FLi,t-1 2.908 0.094 2.202 1.277
(REG_FLi,t-1-FLi,t-1) x FLi,t-1 -3.541 3.387 1.301 8.918
BOPit 0.834 0.796 BOPit x FLi,t-1 -0.901 -0.984 BANKit -0.431 -0.901 -0.409 -1.064 BANKit x FLi,t-1 0.430 -0.079 0.239 -0.006 CURit 0.424 0.418 CURit x FLi,t-1 -1.354 -1.175 RECESSIONit 0.107 -0.518 -0.081 -0.555 RECESSIONit x FLi,t-1 -0.029 1.224 -0.005 1.185 HINFLit 0.504 0.346 0.381 0.415 HINFLit x FLi,t-1 -1.595 -2.535 -1.424 -3.568 FIRSTYEARit -0.029 0.564 -0.016 0.607 FIRSTYEARit x FLi,t-1 0.278 -1.140 0.201 -1.041 IMFit 0.949 0.752 1.174 0.692 IMFit x FLi,t-1 -0.853 -1.502 -1.105 -1.762 USINTit -0.038 -0.073 -0.078 -0.088 (-1.79)* (-1.79)* (-3.06)*** (-2.03)** LEFTit 0.121 -0.117 0.138 -0.611 LEFTit x FLi,t-1 0.376 0.661 0.527 1.323 RIGHTit 0.292 0.366 0.195 0.151 RIGHTit x FLi,t-1 0.161 -0.397 0.542 -0.132 OPENit 0.004 0.003 0.010 0.029 OPENit x FLi,t-1 -0.004 -0.005 -0.011 -0.045 Log L -2899.49 -737.97 -2844.64 -718.92 Number of observations 2138 805 2138 805 Notes: The dependent variable is ΔFLit.
31
Note that we do not report t-statistics for the variables FLi,t-1 and FLi,t-1xYi,t-1 in columns 3 and 4 of tables 24 and 25 and for almost all variables in table 26 as it does not make sense for interaction variables. This issue is explained in section 2.1. However, we do show graphs with the marginal effects of the independent variables on financial reform over the whole range of financial liberalization. These graphs, which only represent the results of the new data, can be found in appendices 1 and 2. The middle line represents the marginal effect while the other two lines define a 95% confidence interval. The particular variable thus has a significant effect on financial reform whenever the upper and lower bounds of the confidence interval are both above (or below) the zero line.
32
openness has a positive effect on financial reform at low levels of liberalization. Our graphs suggest that it is never significantly different from zero.
5.3 Econometric model
We believe that there are two features of the data not carefully represented in the empirical literature until now, namely its dynamic nature and its spatial nature. For the panel dataset that AM use, n=32 and T=22. From an econometric point of view, n and T are large in this case. For the data that we use, n=62 and T=31, which is even larger. So, standard panel data models, such as fixed effects or random effects models, that are implicitly developed for a small T are not appropriate here. Furthermore, we think that spatial interdependence between countries can be modelled differently than the ad-hoc approach that AM use. Anselin (2001) distinguishes spatial dynamic models into four categories, namely “pure space recursive” if only a spatial time lag is included, “time-space recursive” if an individual time lag and a spatial time lag are included, “time-space simultaneous” if an individual time lag and a contemporaneous spatial lag are included, and “time-space dynamic” if all forms of dependence are included. We choose to estimate a model that falls in the last category, which is described in Yu et al. (2008). The model considered is
(6)
where Ynt = (y1t,y2t,…,ynt)‟ and Vnt = (v1t,v2t,…,vnt)‟ are n x 1 column vectors and vit is i.i.d. across i and t with zero mean and variance σ2
, Wn is an n x n predetermined spatial weights matrix that determines the spatial dependence between cross sectional units yit, Xnt is an n x k matrix of nonstochastic regressors, and cn is an n x 1 column vector of country fixed effects. The main features of this model are that it incorporates the concept of a spatial weights matrix to capture possible spatial effects and that it includes a lagged dependent variable.
33
In general, there are two main ways to model spatial effects. The first one is with a spatially lagged dependent variable, known as the spatial lag model. The second one is by modelling a spatial autoregressive process in the error term, known as the spatial error model. The spatial lag model posits that the dependent variable depends on the dependent variable observed in neighbouring units and on other exogenous variables. The spatial error model posits that the dependent variable only depends on exogenous variables but that the error terms are correlated across space (Elhorst, 2009). The model of Yu et al. (2008), as described above, is a spatial lag model.
For our particular purposes, the advantages of using a spatial weights matrix to model a spatial effect are that the spatial effect is not limited to a possibly small group of neighbouring countries, or countries from the same region, and that countries‟ reform efforts depend directly on the reform efforts of other countries (of this year and last year). At first sight, one would assume that countries are most influenced by countries that are geographically close to them. However, as Case et al. (1993) point out, the spatial links between countries might be based on economic rather than geographical interdependence. Therefore, we use two different weights matrices in our analyses: One is based on the physical distance between the capitals of the two particular countries and the other on the bilateral trade flows between them. In order to calculate the entries of the weights matrix based on physical distance, several issues have to be taken care of. First, the distances have to be inverted, so that a short distance corresponds to a large weight in the matrix and vice versa. Second, we replace all positive distances for countries with itself (the diagonal entries of the weights matrix) by zero. This has to be done because else the role of the magnitude of the financial reform efforts of the country itself would dominate the financial reform efforts in other countries. Furthermore, this would lead to serious multicollinearity problems with the lagged dependent variable.2 For both weights matrices, the weights have been row standardized, so that all entries of each row add up to 1. It is important to stress that the weights matrix is fixed and predetermined and is not estimated in any way, but forms together with the column vector Ynt the spatially lagged term. The data for the weights matrix based on physical distance between the capitals comes from the distance dataset of the CEPII (see http://www.cepii.fr/anglaisgraph/cepii/cepii.htm), while the weights matrix based on bilateral trade flows has been constructed by using data on imports and exports from the World Trade Organization (WTO) for the year 1998. The two standardized weights matrices can be found in appendix 3.
2We estimated all our models with the original positive entries as well and this makes, as expected, the lagged
34
The lagged dependent variable Ynt captures possible persistence effects; a positive coefficient for this term would mean that reforms are likely to lead to further reforms, reversals are likely to lead to further reversals and if no reforms or reversals take place, they are unlikely to take place in the future.
The model in (6) will be estimated by a quasi maximum likelihood (QML) procedure. The Matlab program that we use has been developed by Jihai Yu3 and is available upon request. All in all, we believe that this model is best suited to take possible spatial effects and the dynamic nature of the data into account.
5.4 Results estimation new model with dataset of Abiad and Mody (2005)
In order to disentangle the effects of adding more data and using a different econometric model, we first estimate (6) with the old dataset. Thus, results that differ from AM should be attributed solely to the model here, while different results in section 5.5 are a consequence of new data and a new model. Table 27 shows the results of estimating (6) using the dataset of AM, so 32 countries for the period 1975-19964:
Table 27 Quasi maximum likelihood estimates of (6) with old dataset
Weights matrix based Weights matrix based on distance on bilateral trade shares ΔFLi,t-1 0.110 0.105 0.099 0.116 0.110 0.105 (2.88)*** (2.77)*** (2.59)*** (3.05)*** (2.90)*** (2.76)*** W x ΔFLi,t 0.238 0.230 0.242 0.084 0.101 0.113 (2.91)*** (2.80)*** (2.98)*** (1.23) (1.50) (1.69)* W x ΔFLi,t-1 0.101 0.088 0.105 0.065 0.080 0.086 (0.85) (0.74) (0.87) (0.40) (0.49) (0.53) BANKit -0.022 -0.023 -0.023 -0.023 -0.024 -0.024 (-2.33)** (-2.41)** (-2.46)** (-2.43)** (-2.49)** (-2.55)** CURit 0.019 0.016 0.016 0.017 0.015 0.015 (2.16)** (1.83)* (1.81)* (1.99)** (1.71)* (1.68)* DEBTit -0.023 -0.025 -0.026 -0.027 -0.027 -0.028 (-1.63) (-1.73)* (-1.81)* (-1.89)* (-1.91)* (-1.98)** RECESSIONit -0.009 -0.010 -0.011 -0.010 -0.011 -0.012 (-0.98) (-1.18) (-1.23) (-1.10) (-1.27) (-1.31) 3
We thank Paul Elhorst for making this program available to us.
4As our model does not use exactly the same variables as AM we had to drop 3 countries due to data
35 HINFLit -0.006 -0.004 -0.003 -0.005 -0.003 -0.002 (-0.49) (-0.39) (-0.26) (-0.41) (-0.25) (-0.15) FIRSTYEARit 0.011 0.012 0.011 0.012 (1.70)* (1.75)* (1.65)* (1.69)* IMFit 0.022 0.022 0.022 0.022 (2.86)*** (2.92)*** (2.89)*** (2.92)*** USINTit -0.001 -0.001 -0.001 -0.002 (-0.77) (-0.89) (-1.34) (-1.47) LEFTit 0.002 0.000 (0.20) (0.02) RIGHTit -0.002 -0.002 (-0.20) (-0.17) DEMOit -0.001 -0.001 (-1.11) (-0.93) GOVFRACit 0.003 0.004 (0.21) (0.24) Log L 1146.0 1152.2 1153.2 1141.8 1148.7 1149.3 Number of obs. 704 704 704 704 704 704 Notes: The dependent variable is ΔFLit.
Numbers in parentheses are bias-corrected t-statistics. *, **, *** indicate significance at the 10%, 5% and 1% resp.
Columns 1-3 use the weights matrix based on physical distance between the capitals of the countries, while columns 4-6 use the weights matrix based on bilateral trade flows. In columns 1 and 4 we only include the crisis variables as exogenous variables, in columns 2 and 5 we add the shock variables and in columns 3 and 6 we estimate the full model. It should be noted that we do not include the variables PARL, PRESID and the legal origin variables. The reason is that the model, as decribed in (6), contains country fixed effects which makes the inclusion of time-invariant variables impossible. Also, OPEN is excluded from the model, because the estimation procedure requires a balanced panel, and the data on OPEN contains too many gaps. Furthermore, based on our earlier analyses we have no reason to expect that it is important in any way.
36
This suggests that countries are mainly influenced by countries that are geographically close to them and not so much by countries that they share economic linkages with. In accordance with AM, BANK is significantly negative in all specifications, so banking crises seem to set back financial liberalization considerably. The same holds for debt crises, while currency crises have a positive influence on financial reform.
For the shock variables, IMF programs and new governments appear to have a positive influence, although for the latter it is quite weak, while the US interest rate is not significant in any of the regressions. Finally, none of the political variables seems to be a significant predictor for financial reform. Especially the finding that democracy is insignificant is somewhat surprising because Huang (2009) found a strong negative effect of democracy on financial reform using the same dataset.
The main differences with the results of AM are that we demonstrate policy persistence and not so much learning, as AM call it, that IMF programs have a much stronger effect on financial reform than in AM and that new governments seem to be more conducive to reform efforts, contrary to the finding of AM that they do shake the status quo, but in both directions (reforms and reform reversals).
The model, as described above, assumes that the dependent variable is unbounded and continuous. However, the values of our dependent variable all lie between -1 and 1. Therefore, we calculate fitted values after each regression and check whether they lie between -1 and 1. In order to calculate these fitted values, we first have to calculate the country fixed effects manually as the Matlab program of Yu et al. (2008) does not provide them. We use the following equation, from Baltagi (2001), for caluculating these country fixed effects:
,
where ci is the country fixed effect of country i, yi denotes the average value of the
dependent variable of country i, yi(-1) is the average value of the lagged dependent variable
of country i, Wi denotes the ith row of the weights matrix W, and X i is the 1 x k row vector
of average values for the exogenous variables of country i. The coefficients λ, γ, ρ and β are estimates.
37
so we conclude that we can use this model safely, although it has not explicitly been developed for limited dependent variables.
5.5 Results estimation new model with new dataset
Finally, we want to estimate (6) with the extended dataset. Therefore, table 28 shows the results of estimating (6) using the new dataset5:
Table 28 Quasi maximum likelihood estimates of (6) with new dataset
Weights matrix based Weights matrix based on distance on bilateral trade shares ΔFLi,t-1 0.096 0.085 0.084 0.137 0.119 0.118 (4.18)*** (3.71)*** (3.67)*** (5.98)*** (5.20)*** (5.15)*** W x ΔFLi,t 0.381 0.357 0.355 0.285 0.307 0.303 (6.73)*** (6.18)*** (6.14)*** (6.64)*** (7.30)*** (7.18)*** W x ΔFLi,t-1 0.382 0.363 0.364 0.140 0.144 0.144 (5.08)*** (4.84)*** (4.85)*** (2.21)** (2.24)** (2.24)** BANKit -0.010 -0.012 -0.012 -0.008 -0.011 -0.011 (-2.31)** (-2.95)*** (-2.90)*** (-2.00)** (-2.73)*** (-2.67)*** CURit 0.010 0.008 0.008 0.012 -0.009 0.009 (2.80)*** (2.05)** (2.09)** (3.17)*** (2.29)** (2.32)** DEBTit -0.009 -0.012 -0.012 -0.016 -0.018 -0.018 (-1.47) (-2.07)** (-2.10)** (-2.76)*** (-3.09)*** (-3.11)*** RECESSIONit -0.002 -0.002 -0.002 -0.004 -0.003 -0.003 (-0.58) (-0.63) (-0.61) (-1.02) (-0.98) (-0.94) HINFLit 0.003 0.004 0.003 0.006 0.008 0.008 (0.50) (0.73) (0.66) (1.06) (1.48) (1.43) FIRSTYEARit 0.003 0.003 0.003 0.003 (0.94) (0.99) (1.06) (1.07) IMFit 0.016 0.016 0.018 0.018 (5.01)*** (5.06)*** (5.65)*** (5.67)*** 5
38 USINTit 0.000 0.000 -0.001 0.000 (0.67) (0.73) (-1.50) (-1.17) LEFTit 0.007 0.007 (1.64) (1.73)* RIGHTit 0.006 0.006 (1.43) (1.56) DEMOit 0.000 0.000 (0.17) (0.56) GOVFRACit -0.002 -0.001 (-0.42) (-0.24) Log L 3787.7 3802.0 3803.7 3752.5 3771.8 3773.9 Number of obs. 1922 1922 1922 1922 1922 1922 Notes: The dependent variable is ΔFLit.
Numbers in parentheses are bias-corrected t-statistics. *, **, *** indicate significance at the 10%, 5% and 1% resp.
Again, the persistence term (ΔFLi,t-1) is significantly positive in all specifications. Furthermore, the spatial effect pops up in all specifications as well. Contrary to the results in section 5.4, where the limited dataset is used, it is now significantly positive for both weights matrices. Besides that, it remains highly significant when the weights matrix is interacted with the lagged level of financial reform. All in all, these results provide evidence that countries‟ reform efforts are influenced by countries that are either geographically close to them or are an important trading partner.
Concerning the effects of crises, the results suggest that banking crises and debt crises set back financial liberalization, while currency crises lead to more financial reform. Recessions and periods of high inflation do not seem to matter. Of the shock variables, only IMF programs are, not surprisingly, strongly positively related to financial reform. The coefficients on new governments and US interest rates are insignificant in all cases. Finally, none of the political variables is significantly related to financial reform, although the dummy for left-wing governments is significantly positive at a 10% level in one specification.
39
0.35. As the level of financial liberalization is measured on a 0-21 scale, these numbers are not very large, but certainly not negligible.
Any differences between the results shown in tables 27 and 28 are due to differences in data coverage; the dataset that is used for the estimations in table 27 contains 32 countries for the period 1975-1996 while the estimations in table 28 are based on 62 countries for the period 1975-2005. One of the differences between these two analyses is that new governments are a determinant of financial reform in the limited dataset, although it is only significant at a 10% level, while they do not seem to matter anymore when we use the larger dataset. Another thing is that the spatial effect is much stronger in the second case. This could possibly be explained by the fact that the more countries are taken into account the more accurate the specification of the spatial weights matrix becomes. For the rest, the results are more or less comparable.
5.6 Sensitivity analyses
A careful inspection of the separate components of the financial liberalization index reveals that item 6, namely „prudential regulations and supervision of the banking sector‟ is somewhat different from the others. Whereas these other components measure the extent of liberalization in the financial sector, item 6 is the only component where a greater degree of government intervention is coded as a reform (Abiad et al., 2008). Furthermore, as we have seen during the global financial crisis of 2008, often financial liberalization and good regulations and supervision do not go hand in hand. Therefore, we think it is a worthwhile exercise to examine what happens when we leave this item out.
In order to test whether this would change our results we compose a new index of financial liberalization without item 6. The new index thus contains values between 0 and 18, which can again be normalized to obtain values between 0 and 1. Table 29 presents the results of estimating (6) with this alternative index of financial liberalization.
Table 29 Q.M.L. estimates of (6) with new dataset without item 6
Weights matrix based Weights matrix based on distance on bilateral trade shares ΔFLi,t-1 0.100 0.088 0.088 0.138 0.122 0.121
40 (6.71)*** (6.09)*** (6.09)*** (6.72)*** (6.98)*** (6.80)*** W x ΔFLi,t-1 0.376 0.355 0.355 0.147 0.151 0.151 (4.92)*** (4.67)*** (4.67)*** (2.13)** (2.11)** (2.11)** BANKit -0.011 -0.013 -0.013 -0.010 -0.013 -0.013 (-2.71)*** (-3.36)*** (-3.32)*** (-2.47)** (-3.20)*** (-3.15)*** CURit 0.010 0.007 0.007 0.012 0.009 0.009 (2.82)*** (2.04)** (2.09)** (3.23)*** (2.35)** (2.39)** DEBTit -0.009 -0.013 -0.013 -0.015 -0.018 -0.018 (-1.64) (-2.28)** (-2.30)** (-2.73)*** (-3.13)*** (-3.15)*** RECESSIONit -0.003 -0.003 -0.003 -0.004 -0.004 -0.004 (-0.84) (-0.91) (-0.90) (-1.13) (-1.11) (-1.08) HINFLit 0.004 0.005 0.004 0.007 0.009 0.009 (0.74) (0.97) (0.88) (1.42) (1.79)* (1.71)* FIRSTYEARit 0.003 0.003 0.003 0.003 (1.04) (1.11) (1.14) (1.18) IMFit 0.015 0.015 0.017 0.017 (5.06)*** (5.11)*** (5.64)*** (5.68)*** USINTit 0.000 0.000 0.000 0.000 (0.90) (0.90) (-0.76) (-0.56) LEFTit 0.006 0.007 (1.54) (1.69)* RIGHTit 0.005 0.006 (1.40) (1.61) DEMOit 0.000 0.000 (0.01) (0.23) GOVFRACit -0.003 -0.002 (-0.44) (-0.34) Log L 3872.2 3887.1 3888.7 3839.3 3857.7 3859.6 Number of obs. 1922 1922 1922 1922 1922 1922
41
the rest, the results of table 28 and 29 are almost the same. Therefore, we believe that our main results are robust against changing the contents of the financial liberalization index.
6. Summary and Conclusions
Taking together all the analyses that we have discussed in the last chapter, we think we are able to draw several robust conclusions. First of all, there is considerable policy persistence at work in the financial reform process; what happens this year is highly influenced by what happened last year. Countries that are reforming their financial sectors tend to keep doing that, while countries that are not will not do anything in the future, ceteris paribus. Therefore, some kind of trigger is necessary in order to start the financial reform process. One of these triggers can be actions of other countries. There is evidence that financial reform in a country is influenced by financial reform efforts in countries that are geographically close by or by countries that are important trading partners. Other triggers are crises: Banking crises and debt crises lead to reform reversals, while currency crises induce reform. Also periods of high inflation shake the status quo, albeit in both directions. On the other hand, recessions do not seem to be related to financial reform at all. Among shocks that can lead to policy change, only IMF programs increase the probability of financial reform; new governments and high US interest rates do not matter. The level of democracy, government fractionalization, political ideology and the system of government (presidential or parliamentary) do not seem to be important drivers of the financial reform process. Finally, legal origin and trade openness also do not matter for financial reform. So, although there is ample evidence that democracy is positively related to general economic reform, apparently this does not hold for financial reform.
42
democracy, we have no reason to believe that these factors matter. Findings from these papers that we agree with are the positive effect of IMF programs, the negative effect of banking crises and that countries‟ reform efforts are positively influenced by other countries‟ policy actions.
All in all, we think that this thesis provides an extensive and critical overview of the empirical literature on the determinants of financial reform. Furthermore, the analyses that we have done confirm some of the earlier findings while others are challenged.
References
Abiad, Abdul and Ashoka Mody (2005), „Financial reform: What shakes it? What shapes it?‟,
American Economic Review, 95 (1): 66-88.
Abiad, Abdul, Detragiache, Enrica and Thierry Tressel (2008), „A new database of financial reforms‟, IMF Working Paper No. 08/266.
Anselin, Luc (1988), „Lagrange multiplier test diagnostics for spatial dependence and spatial heterogeneity‟, Geographic Analysis, 20: 1-17.
Baltagi, Badi (2001), „Econometric analysis of panel data‟, John Wiley & Sons Ltd, Chichester, United Kingdom.
Bandiera, Oriana, Caprio, Gerard, Honohan, Patrick and Fabio Schiantarelli (2000), „Does financial reform raise or reduce saving?‟, Review of Economics and Statistics, 82: 239-263. Bartolini, Leonardo and Allan Drazen (1997), „When liberal policies reflect external shocks, what do we learn?‟, Journal of International Economics, 42 (3-4): 249-273.
Bekaert, Geert, Harvey, Campbell R. and Christian Lundblad (2005), „Does financial liberalization spur growth?‟, Journal of Financial Economics, 77 (1): 3-55.
Brambor, Thomas, Clark, William Roberts and Matt Golder (2006), „Understanding interaction models: Improving empirical analyses‟, Political Analysis, 14: 63-82.
Budge, Ian (1992), „ECPR party manifestos project [computer file]‟, 3rd edition, Colchester: ESRC Data Archive.
Burgoon, Brian, Demetriades, Panicos and Geoffrey Underhill (2008), „Financial liberalisation and political variables: a response to Abiad and Mody‟, Paper presented at the Research Workshop “Finance and Development”, Cyprus University of Technology, 8-9th
May 2008.
