I. Introduction

In the last quarter of the twentieth century, financial sector reform was high on the agenda of policymakers. Though the trend worldwide was toward more liberalized financial systems, reform experiences differed considerably from one country to another, with their speed ranging from sluggish to swift and the magnitude of changes ranging from complete overhauls to minor tweaking. All countries experienced long stretches with no policy change and, occasionally, previous reforms were reversed. Our objective in this paper is to explain these variations. Using a newly constructed database, we document trends in financial sector reform and ask: when, by how much, and why did countries reform?

A large and technically sophisticated literature has examined the consequences of financial sector liberalization, but the causes of financial liberalization have received less attention. Levine (1997) summarizes the influence of financial development on growth. Demirguc-Kunt and Detragiache (1998) and Kaminsky and Reinhart (1999) discuss the causality from financial liberalization to economic crises. But what are the factors that stimulate financial sector reform? Do, for example, crises induce reforms? Country case studies have provided support for various economic and political theories of policy change.² Case studies, however, have obvious limitations and, as Drazen (2000, p. 449) notes, “it is striking how little statistical testing there has been.”³

A successful statistical examination of liberalization has to meet two challenges. First, policy changes tend to be episodic and the triggers for these episodes need to be identified. However, the episodes themselves—leading to both liberalizations and reversals—are embedded in a long-term process that, in the past quarter century, has trended toward greater liberalization. Thus, the second challenge is to identify a dynamic process that leads to these cumulative transformations.

We pursue a political economy perspective in explaining the timing, pace, and extent of financial sector reforms. The starting point of the analysis is an observed bias toward retaining the status quo, with the implication that established interest groups compromise to maintain the existing policy regime. From the literature, we identify three sources of reforms. First, reforms may be triggered by discrete events, or “shocks,” that change the balance of decision-making power. Of relevance are domestic events such as the “political honeymoon” following the formation of a new government, various types of crisis, and external influences, such as changes in global interest rates and leverage exercised by international financial institutions (Krueger, 1993). Second, “learning” may foster reform by revealing information that causes reassessment of the costs and benefits of the policy regime. Learning can also help resolve the impasse on account of uncertainty regarding the identity of winners (Fernandez and Rodrik, 1991). The term “learning” thus is shorthand for both discovery and the consequent realignments in relationships.⁴ Domestic learning may be supplemented by international “diffusion” as countries move to global or regional norms to compete for international capital (Simmons and Elkins, 2001). And, finally, reforms may be conditioned by political ideology of the ruling government (Alesina and Roubini, 1992; and Cukierman and Tommasi, 1998) and such structural features as openness to trade (Rajan and Zingales, 2002), legal system (La Porta and others, 1997), and form of government (Persson, 2002).

These sources of change—shocks, learning, and ideology and structure—have differing implications for the timing of liberalization. Following a shock, an immediate policy change may be undertaken. In contrast, learning allows for sustained changes. Where reforms are beneficial and/or alter the balance of power between competing interest groups, a learning process maintains a reform momentum, making further reforms more probable. Ideology and structure can act independently or can condition the response to events. This paper provides an integrated empirical framework to examine the various forces producing transitions to financial liberalization. The possibility of domestic learning implies a dynamic relationship between the level of financial sector liberalization and subsequent policy changes undertaken. The paper specifies such a relationship and, within that context, identifies the other factors that further strike at the status quo.

One constraint in analyzing the timing, direction, and size of policy changes has been the lack of a cross-country dataset on financial liberalization. Existing measures of liberalization refer either to a one-time change in rules (e.g., Bekaert and Harvey, 2000), which limits the investigation to episodes of liberalization, or to continuous proxies such as the level of financial development in an economy (e.g., Rajan and Zingales, 2002), which permit examination of long-term changes but not of discrete events. As the discussion above implies, liberalization is a mix of the episodic and the ongoing. Thus, an index is needed that combines both aspects of the reform—and reversal—process.

Our investigation uses a newly constructed financial liberalization index, covering 35 countries over the period 1973–96. The index allows a more precise determination of the timing and significance of various events cumulating in a financial liberalization process. The index recognizes the multifaceted nature of financial reform and is an annual aggregation of financial reform along six dimensions. The six facets that characterize the financial sector policy are: directed credit/reserve requirements; interest rate controls; entry barriers and/or lack of pro-competition policies; restrictive operational regulations and/or lack of prudential regulations; the degree of privatization in the financial sector; and the degree of controls on international financial transactions. On each dimension, a country is classified as being fully repressed, partially repressed, largely liberalized, or fully liberalized.

To allow for the discrete nature of liberalization, we use the ordered logit technique to analyze the determinants of reform. Our results suggest that shocks mattered, learning was crucial, and ideology and structure appear to have played only a modest role in the financial reform process. What types of economic and political events shocked the status quo? Among external influences, reforms were promoted by a decline in U.S. interest rates and participation in programs of the International Monetary Fund (though this effect was pronounced mainly in countries where financial sectors were highly repressed). Crises were liable not only to induce reforms but also reversals. A balance of payments crisis typically triggered financial sector reforms, but banking crises were often associated with policy reversals. The tendency toward status quo weakened during a chief executive’s first year in office, but changes included reversals as well as reforms. While these various events were influential, learning was essential to shaping and sustaining reforms. Countries with highly repressed financial sectors tended to stay that way but once initial reforms occurred, the process gained momentum and future reforms became more likely. Learning also occurred through observing regional reform leaders. Interestingly, ideology and structure seem to have had limited influence on the reform process. For example, left-wing governments were no less reform-oriented than right-wing governments, presidential and parliamentary regimes were equally inclined to reform, and legal systems did not come in the way of the move to liberalize. Greater trade openness, however, appears to have increased the pace of reform, especially where the level of liberalization was low.

The rest of the paper is organized as follows. In Section II, we describe the database of financial liberalization episodes and briefly highlight the salient features of financial reform over the past 25 years. Section III summarizes the relevant literature on the political economy of reform, focusing on the reasons for status quo bias and the conditions under which policy changes become more likely. In Section IV, we examine, one at a time, selected factors associated with policy changes. Section V presents multivariate analysis and includes several extensions to test the robustness of the results. Section VI concludes.

II. Financial Liberalization, 1973–96

The new database used in this study considers various financial sector policies in 35 countries over the 24-year period from 1973 to 1996. Six policy dimensions are inputs to the creation of an aggregate index of the degree of policy liberalization. These include:

• Credit controls, such as directed credit toward favored sectors or industries, ceilings on credit toward other sectors, and excessively high reserve requirements.

• Interest rate controls, including whether the government directly controls interest rates, or whether floors, ceilings, or interest rate bands exist.

• Entry barriers, such as licensing requirements, limits on the participation of foreign banks, and restrictions relating to bank specialization or the establishment of universal banks.

• Regulations and securities markets; included here are operational restrictions, such as on staffing, branching and advertising, and the establishment of securities markets.

• Privatization in the financial sector.

• Restrictions on international financial transactions, such as on capital and current account convertibility, and the use of multiple exchange rates.

Along each dimension, a country is given a score on a graded scale, with zero corresponding to being fully repressed, one to partially repressed, two to largely liberalized, and three to fully liberalized.⁵ Policy changes, then, denote shifts in a country’s score on this scale in a given year. In some cases, such as when all state-owned banks are privatized all at once, or when controls on all interest rates are simultaneously abolished, policy changes will correspond to jumps of more than one unit along that dimension. Reversals, such as the imposition of capital controls or interest rate controls, are recorded as shifts from a higher to a lower score. Given its detailed construction, the database thus allows a much more precise determination of the magnitude and timing of various events in the financial liberalization process.

Identifying the various policy changes included in our database was facilitated by the available surveys of financial liberalization experiences, such as Williamson and Mahar (1998), Fanelli and Medhora (1998), Johnston and Sundararajan (1999), De Brouwer and Pupphavesa (1999), and Caprio and others (2001). Nevertheless, frequent use of other resources, such as central bank bulletins and websites, IMF country reports, books, and journal articles was made when information was unclear or incomplete; references for each country are identified in the database itself.

Table 1 reports the correlations among the six components of financial liberalization. Some subcomponents show a higher correlation, indicating that the liberalizations along these dimensions tended to occur together. The three measures most frequently used as indicators of financial repression—credit controls, interest rate controls, and controls on international financial transactions—are all highly correlated with each other, with the correlations ranging from 0.76 to 0.82. Less correlated are the measures of financial liberalization relating to entry barriers and regulations. The measure of the extent of privatization in the financial sector has the lowest correlation with the other components, an indication that privatization docs not coincide with other reforms.

Table 1.

Correlations Among Financial Liberalization Components

Table 1.

Correlations Among Financial Liberalization Components

	Credit Controls	Interest Rate Controls	Entry Barriers	Regulations	Privatization	International Transactions
Credit controls	1.00
Intercut rate controls	0.82	1.00
Entry barriers	0.65	0.66	1.00
Regulations	0.69	0.68	0.58	1.00
Privatization	0.59	0.51	0.38	0.61	1.00
International Transactions	0.77	0.76	0.59	0.74	0.59	1.00

View Table

Table 1.

Correlations Among Financial Liberalization Components

	Credit Controls	Interest Rate Controls	Entry Barriers	Regulations	Privatization	International Transactions
Credit controls	1.00
Intercut rate controls	0.82	1.00
Entry barriers	0.65	0.66	1.00
Regulations	0.69	0.68	0.58	1.00
Privatization	0.59	0.51	0.38	0.61	1.00
International Transactions	0.77	0.76	0.59	0.74	0.59	1.00

The measures along the six dimensions can be aggregated to obtain an index of overall financial liberalization for each country in each year. Various aggregation methods produced very similar measures, so in this paper we simply use the sum of the individual components.⁶ Since each of the six indices can take on values between 0 and 3, the sum takes on values between 0 and 18. These aggregate measures of financial liberalization are tabulated for each country in the Data Appendix.

We draw four broad conclusions on the time profile of liberalization. First, despite stops, gaps, and reversals, financial sector liberalization advanced through much of the world in the last quarter century (Figure 1). Countries in all income groups liberalized, though higher-income economies remained more liberalized than lower-income economies throughout.

Financial Liberalization by Income Group, 1973-1996

Citation: IMF Working Papers 2003, 070; 10.5089/9781451849172.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Financial Liberalization by Income Group, 1973-1996

Citation: IMF Working Papers 2003, 070; 10.5089/9781451849172.001.A001

• Download Figure
• Download figure as PowerPoint slide

Financial Liberalization by Income Group, 1973-1996

Citation: IMF Working Papers 2003, 070; 10.5089/9781451849172.001.A001

• Download Figure
• Download figure as PowerPoint slide

Second, while trends appear smooth in these country averages, the reform process at the country level was typically characterized by long periods of status quo, or no change in policy. To facilitate the exposition, we classify policy changes for each country-year into five categories. A decrease in the financial liberalization measure by 3 or more points is classified as a large reversal; a decrease of 1 or 2 points as a reversal; an increase by 1 or 2 points as a reform; and an increase of 3 or more points is classified as a large reform. Finally, years in which no policy changes were undertaken are classified as status quo observations. Table 2 shows the distribution of various policy changes in the whole sample, as well as by region. Status quo observations represent the large majority of observations—over 76 percent of the whole sample. Reforms constitute another 15 percent of the sample, and large reforms account for another 5 percent. Reversals are relatively rare and large reversals even more so. There were two large reversals in the sample, and both occurred in Latin America—in Argentina in 1982 and in Venezuela in 1994.

Table 2.

Distribution of Financial Policy Choices, Full Sample and by Region

(In percent)

Table 2.

Distribution of Financial Policy Choices, Full Sample and by Region

(In percent)

		Full Sample	Latin America	East Asia	South Asia	Africa/ Middle East	OECD
Large reform		5.1	6.2	7.5	3.5	4.3	3.9
Reform		15.4	11.2	14.3	17.4	13.8	19.1
Status quo		76.2	78.3	73.9	75.7	79.7	74.3
Reversal		3.1	3.1	4.3	3.5	2.2	2.6
Large reversal		0.3	1.2	0.0	0.0	0.0	0.0
	Total	100.0	100.0	100.0	100.0	100.0	100.0

View Table

Table 2.

Distribution of Financial Policy Choices, Full Sample and by Region

(In percent)

		Full Sample	Latin America	East Asia	South Asia	Africa/ Middle East	OECD
Large reform		5.1	6.2	7.5	3.5	4.3	3.9
Reform		15.4	11.2	14.3	17.4	13.8	19.1
Status quo		76.2	78.3	73.9	75.7	79.7	74.3
Reversal		3.1	3.1	4.3	3.5	2.2	2.6
Large reversal		0.3	1.2	0.0	0.0	0.0	0.0
	Total	100.0	100.0	100.0	100.0	100.0	100.0

Third, there is evidence of regional clustering: countries within certain regions have tended to liberalize their financial sectors at roughly the same time, and in roughly the same way.⁷ With the exception of early reforms in Argentina and Chile in the 1970s, most of the reforms in Latin America were clustered in the late 1980s and early 1990s (Figure 2). The two exceptions, Chile and Argentina, also illustrate that reform is not a steady march forward: both countries reversed policy during the debt crisis of 1982–83. The process of financial liberalization in East Asia was much more gradual. Countries opened up their financial sectors in small steps in the early 1980s, with the whole reform process stretching over a decade in most cases. South Asian financial sectors remain at least partially repressed even at the end of our sample period. South Asia’s reforms occurred in the early to mid-1990s, with the exception of Sri Lanka, which undertook a major reform effort in 1978.

Financial Liberalization by Region, 1973-1996

Citation: IMF Working Papers 2003, 070; 10.5089/9781451849172.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Financial Liberalization by Region, 1973-1996

Citation: IMF Working Papers 2003, 070; 10.5089/9781451849172.001.A001

• Download Figure
• Download figure as PowerPoint slide

Financial Liberalization by Region, 1973-1996

Citation: IMF Working Papers 2003, 070; 10.5089/9781451849172.001.A001

• Download Figure
• Download figure as PowerPoint slide

Finally, four of the OECD countries (Germany, Canada, the United Kingdom, and the United States) already had liberalized financial sectors at the beginning of our sample period. The rest of the OECD countries in our sample started the period with relatively repressed financial systems but caught up and now have largely or fully liberalized financial sectors via a gradual process beginning in the late 1970s and early 1980s. Only New Zealand adopted a one-shot approach, undertaking most of its financial reforms in 1984–85.

III. The Factors That Drive Reform: Literature Review

The starting point for our analysis is that the status quo will persist so long as the benefits of maintaining it outweigh the costs for those who determine the timing and pace of policies. Theories of economic reforms fall into three broad categories: (a) “shocks” alter the balance of decision-making power, leading to both reforms and reversals; (b) perceived payoffs are updated via discovery or a reduction in uncertainty, which we simply call “learning”; and (c) ideology and political and economic structure condition the pace of reform. We discuss each possibility in turn. In undertaking this literature review, we draw on the broader literature on the political economy of reform rather than merely on studies focused on financial liberalization. While a variety of theories suggest different determinants of economic reform, the empirical conclusions from the case studies and the limited number of econometric analyses are ambiguous. The empirical ambiguity reflects the multifaceted nature of reform, the difficulty in identifying the occurrence and timing of reforms, and the difficulty also in defining determining events such as crises.

IV. Reform Determinants: Bivariate Relationships

In this section, we present selected bivariate relationships between policy changes and the conditions suggested by the theoretical and empirical literature reviewed above. These relationships have the benefit of greater transparency than the multivariate findings in Section V which corroborate these findings. Both in this section and the next, we focus on explaining movements in our aggregate index, rather than in the components. As noted above, the components are highly correlated. However, timing variations in the changes of particular components are of some interest and we discuss these in the sensitivity analysis.

Consider first the shocks that dislodge the status quo. The likelihood of reform, especially large reforms, increases significantly during an incumbent’s first year in office (Table 3, upper left-hand panel).¹⁴ The proportion of status quo observations drops from 78 percent to 70 percent. A Pearson chi-square test¹⁵ rejects the null hypothesis that the rows and columns are independent. Change, then is more likely during the first year in office, supporting the opportunistic politician theory and the importance of the electoral cycle. Note, however, that the effect is not unidirectional; reversals are also more frequent during the first year in office, as noted by Haggard and Webb (1993). Next, we follow Tornell (1998) and derive an index of drastic political change from the Polity IV dataset of Marshall and Jaggers (2000). The drastic political change measure is intended to capture periods involving substantial changes in the form of government such as coups d’etat and the imposition or lifting of martial law. The upper right-hand panel of Table 3 shows that the likelihood of financial reform is unaffected by drastic political change. In fact, there is a slight increase in the relative frequency of status quo observations, which suggests that financial reform is probably not high on the agenda during such periods.

Table 3.

Policies, Political Conditions and IMF Programs

(In percent)

Table 3.

Policies, Political Conditions and IMF Programs

(In percent)

				First Year in Office?						Drastic Political Change?
				No	Yes					No	Yes
Large reform				3.9	9.8	Large reform				5.2	3.9
Reform				15.4	15.3	Reform				15.4	15.4
Status quo				77.7	69.9	Status quo				76.0	78.9
Reversal				3.0	3.7	Reversal				3.2	1.9
Large reversal				0.0	1.2	Large reversal				0.3	0.0
	Total			100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		12.52				Pearson Chi-sq:		0.04
			Prob:	0.01					Prob:	1.00
				IMF Program?
				No	Yes
Large reform				4.6	6.6
Reform				14.4	18.3
Status quo				77.2	73.2
Reversal				3.6	1.9
Large reversal				0.3	0.0
	Total			100.0	100.0
		Pearson Chi-sq:		3.42
			Prob:	0.49

View Table

Table 3.

Policies, Political Conditions and IMF Programs

(In percent)

				First Year in Office?						Drastic Political Change?
				No	Yes					No	Yes
Large reform				3.9	9.8	Large reform				5.2	3.9
Reform				15.4	15.3	Reform				15.4	15.4
Status quo				77.7	69.9	Status quo				76.0	78.9
Reversal				3.0	3.7	Reversal				3.2	1.9
Large reversal				0.0	1.2	Large reversal				0.3	0.0
	Total			100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		12.52				Pearson Chi-sq:		0.04
			Prob:	0.01					Prob:	1.00
				IMF Program?
				No	Yes
Large reform				4.6	6.6
Reform				14.4	18.3
Status quo				77.2	73.2
Reversal				3.6	1.9
Large reversal				0.3	0.0
	Total			100.0	100.0
		Pearson Chi-sq:		3.42
			Prob:	0.49

The reform orientation of the government may also change under external conditionality as, for example, if under a program with the IMF. The lower left-hand panel of Table 3 shows that while the chi-square statistic is not significant, IMF programs have a positive reform bias: the presence of an IMF program is associated with a greater likelihood of reform and a smaller likelihood of reversal.

Table 4 relates financial reform to different types of crises. A distinction is made between balance of payments and banking crises, dates for which were compiled from the crisis database constructed by Bordo and others (2001). Other forms of adverse economic conditions considered include a recession, defined simply as negative GDP growth, and high inflation, defined as an annual inflation rate exceeding 50 percent. Under all four types of adversity, the relative frequency of status quo observations falls. However, the extent, and even the direction, of the policy change varies with the nature of the adversity. In response to a balance of payments crisis, the likelihood of a large reform increases from 3.1 percent to 9.7 percent. In contrast, when a country is in a banking crisis, the likelihood of a large financial reform falls from 5.5 percent to 2.6 percent and the possibility of reversals (big or small) increases from 2.3 percent to 9.5 percent. Finally, recessions and high inflation periods increase the likelihood of both reforms and reversals. For all four types of adversity, Pearson chi-square tests reject the null hypothesis that the distribution of policy changes is independent of the occurrence of a crisis, at the I percent level for balance of payments and banking crises, at the 5 percent level for high inflation cases, and at the 10 percent level for recessions.

Table 4.

Policies Under Crisis Conditions

(In percent)

Table 4.

Policies Under Crisis Conditions

(In percent)

				Balance of Payments Crisis?						Banking Crisis?
				No	Yes					No	Yes
Large reform				3.1	9.7	Large reform				5.5	2.6
Reform				15.2	15.8	Reform				15.8	13.0
Status quo				78.7	70.5	Status quo				76.4	74.8
Reversal				3.1	3.2	Reversal				2.3	7.8
Large reversal				0.0	0.8	Large reversal				0.0	1.7
	Total			100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		16.95				Pearson Chi-sq:		15.40
			Prob:	0.00					Prob:	0.00
				Recession?						High (>50%) Inflation?
				No	Yes					No	Yes
Large reform				5.1	4.9	Large reform				4.8	7.6
Reform				14.9	18.6	Reform				15.6	14.1
Status quo				77.1	69.6	Status quo				76.9	70.7
Reversal				2.8	4.9	Reversal				2.8	5.4
Large reversal				0.0	2.0	Large reversal				0.0	2.2
	Total			100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		8.79				Pearson Chi-sq:		10.22
			Prob:	0.07					Prob:	0.04

View Table

Table 4.

Policies Under Crisis Conditions

(In percent)

				Balance of Payments Crisis?						Banking Crisis?
				No	Yes					No	Yes
Large reform				3.1	9.7	Large reform				5.5	2.6
Reform				15.2	15.8	Reform				15.8	13.0
Status quo				78.7	70.5	Status quo				76.4	74.8
Reversal				3.1	3.2	Reversal				2.3	7.8
Large reversal				0.0	0.8	Large reversal				0.0	1.7
	Total			100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		16.95				Pearson Chi-sq:		15.40
			Prob:	0.00					Prob:	0.00
				Recession?						High (>50%) Inflation?
				No	Yes					No	Yes
Large reform				5.1	4.9	Large reform				4.8	7.6
Reform				14.9	18.6	Reform				15.6	14.1
Status quo				77.1	69.6	Status quo				76.9	70.7
Reversal				2.8	4.9	Reversal				2.8	5.4
Large reversal				0.0	2.0	Large reversal				0.0	2.2
	Total			100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		8.79				Pearson Chi-sq:		10.22
			Prob:	0.07					Prob:	0.04

To assess the relevance of the learning process, we ask if the distribution of policy changes varies with the level of financial liberalization in a country. One might surmise a negative relationship, since countries with highly repressed financial sectors have the most potential for liberalizing, while liberalized economies have less left to do. Table 5, which tabulates policy changes under different levels of financial liberalization, shows that this is not quite the case. Instead, we find an inverse U-shaped curve. Countries with highly repressed financial systems tend to stay that way, as evidenced by the much higher proportion of status quo observations. But if these countries liberalize somewhat and become only partially repressed, then the likelihood of further reforms increases substantially. Reforms are much more probable in countries whose financial sectors are in an intermediate range of liberalization—either partially repressed or largely liberalized. Finally, when financial sectors are liberalized (the fourth column), reforms again become less likely. It is interesting to note that, in this sample at least, the liberalized state seems to be an absorbing state; none of the countries that reach the liberalized state undertook reversals of previous reforms.

Table 5.

Policies and the Current State of Liberalization

(In percent)

Table 5.

Policies and the Current State of Liberalization

(In percent)

		Current State of Liberalization
		Fully Repressed	Partially Repressed	Largely Liberalized	Liberalized
Large reform		5.3	7.9	4.6	0.0
Reform		8.0	21.9	22.4	8.7
Status quo		85.9	64.1	67.2	91.3
Reversal		0.8	5.8	5.2	0.0
Large reversal		0.0	0.4	0.6	0.0
	Total	100.0	100.0	100.0	100.0
		Pearson Chi-sq:	58.54	Prob:	0.00

View Table

Table 5.

Policies and the Current State of Liberalization

(In percent)

		Current State of Liberalization
		Fully Repressed	Partially Repressed	Largely Liberalized	Liberalized
Large reform		5.3	7.9	4.6	0.0
Reform		8.0	21.9	22.4	8.7
Status quo		85.9	64.1	67.2	91.3
Reversal		0.8	5.8	5.2	0.0
Large reversal		0.0	0.4	0.6	0.0
	Total	100.0	100.0	100.0	100.0
		Pearson Chi-sq:	58.54	Prob:	0.00

The inverse U-shaped relationship between the level of financial liberalization and the incidence of reform supports the idea that “learning” creates a self-sustaining dynamic in the reform process. It also has an important implication for would-be reformers. Reforms do not have to be undertaken all at once—even small reforms are a victory, as they carry with them the seeds of future reforms.

Finally, we consider the role of ideology and political structure on policy reform. The upper left-hand panel of Table 6 shows the distribution of policy choices with the political orientation of the incumbent executive.¹⁶ The Pearson chi-square test statistic is insignificant and the table shows that left-wing governments are no less reform-oriented than right wing-governments. In fact, the former are slightly more likely to undertake large reforms, while the latter are slightly more likely to reverse previous reforms, which would tend to support the theoretical results of Cukierman and Tommasi (1998). Different regime types—presidential versus parliamentary forms of government—also do not have a significant effect on financial policy choices, as can be seen in the upper right-hand panel of Table 6. This contrasts with the strong findings of Persson (2002) regarding the influence of regime type on fiscal policy. Lastly, we look at whether countries which are more open to trade are more reform-oriented. We measure openness by the sum of imports and exports relative to GDP, and classify observations into quartiles. We find no significant relationship between openness and financial policy choice.

Table 6.

Policies, Ideology and Structure

(In percent)

Table 6.

Policies, Ideology and Structure

(In percent)

				Political Orientation?							System of Government?
				Left	Center/ Other	Right					Presidential	Parliamentary
Large reform				6.6	5.0	4.4	Large reform				0.0	0.5
Reform				15.5	13.6	17.1	Reform				3.2	3.0
Status quo				75.1	78.8	74.2	Status quo				74.3	77.0
Reversal				2.8	2.0	4.4	Reversal				16.8	14.7
Large reversal				0.0	0.7	0.0	Large reversal				5.6	4.8
	Total			100.0	100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		5.18					Pearson Chi-sq:		1.17
			Prob:	0.74						Prob:	0.88
				Trade Openness
				First Quartile	Second Quartile	Third Quartile	Fourth Quartile
Large reform				0.5	0.0	0.5	0.0
Reform				3.5	3.4	1.5	4.0
Status quo				77.4	73.5	76.6	77.1
Reversal				15.6	16.2	16.9	12.9
Large reversal				3.0	6.9	4.5	6.0
	Total			100.0	100.0	100.0	100.0
		Pearson Chi-sq:		5.00
			Prob:	0.96

View Table

Table 6.

Policies, Ideology and Structure

(In percent)

				Political Orientation?							System of Government?
				Left	Center/ Other	Right					Presidential	Parliamentary
Large reform				6.6	5.0	4.4	Large reform				0.0	0.5
Reform				15.5	13.6	17.1	Reform				3.2	3.0
Status quo				75.1	78.8	74.2	Status quo				74.3	77.0
Reversal				2.8	2.0	4.4	Reversal				16.8	14.7
Large reversal				0.0	0.7	0.0	Large reversal				5.6	4.8
	Total			100.0	100.0	100.0		Total			100.0	100.0
		Pearson Chi-sq:		5.18					Pearson Chi-sq:		1.17
			Prob:	0.74						Prob:	0.88
				Trade Openness
				First Quartile	Second Quartile	Third Quartile	Fourth Quartile
Large reform				0.5	0.0	0.5	0.0
Reform				3.5	3.4	1.5	4.0
Status quo				77.4	73.5	76.6	77.1
Reversal				15.6	16.2	16.9	12.9
Large reversal				3.0	6.9	4.5	6.0
	Total			100.0	100.0	100.0	100.0
		Pearson Chi-sq:		5.00
			Prob:	0.96

V. Reform Determinants: Multivariate Analysis

We investigate the robustness of the findings in the previous section by analyzing the various determinants of policy change in a multivariate context. Recall that our liberalization measure sums the six components of financial liberalization, each ranging from 0 to 3 and hence the overall measure takes on integer values between 0 and 18. To facilitate interpretation of the regressions, we divide the overall liberalization measure by 18 to get an index, FL_it, which ranges from 0 to 1, with 0 corresponding to a completely repressed financial sector and 1 corresponding to a fully liberalized financial sector. Our dependent variable is ΔFL_it, which measures policy changes. Given the discrete, ordinal nature of the dependent variable, we use the ordered logit method for our estimation.¹⁷

A. Benchmark Specification

As discussed in Section III, the status quo may be altered by shocks or events that alter the power balance among competing groups, domestic and regional learning processes, or ideological or structural factors. To model the domestic “learning” process, we specify a simple reduced-form dynamic process. Policy change is affected by the difference between the desired level of financial liberalization, $F{L}_{it}^{*}$, and the current level of financial liberalization, FL_i,t−1, so that

${\Delta }F{L}_{it}=\alpha (F{L}_{it}^{*}-F{L}_{i,t-1})+{\epsilon }_{it}$

The adjustment factor, α, is a measure of the status quo bias: the lower α is, the greater the status quo bias. Since $F{L}_{it}^{*}$, is not observable, we begin in our benchmark model with the assumption that the desired level of financial liberalization is $F{L}_{it}^{*}=1$. This assumption implies that financial liberalization is regarded by policymakers as welfare-enhancing and that full liberalization represents a “global norm.” Other possibilities, including country-specific measures of desired liberalization, are explored in the next section. Second, the adjustment factor, α, is likely to be time-varying, allowing for the possibility of learning. Following the theoretical literature, we assume that the resistance to reform is a function of the state of liberalization, i.e., α = k·FL_i,t−1, k > 0. The assumption here is that status quo bias is highest when financial sectors are repressed and the bias declines as the sector is liberalized. Such a dynamic would occur, for example, in a multi-stage version of the Fernandez-Rodrik model where earlier reforms help identify winners and losers. It is also consistent with the need to build technical and managerial expertise in reform implementation and with a strengthening of “outside” groups’ position relative to incumbents. We can thus rewrite equation above as:

${\Delta }F{L}_{it}={\theta }_{1}F{L}_{i,t-1}(1-F{L}_{i,t-1})+{\epsilon }_{it}$

Next, we allow for the possibility of regional diffusion. If such an influence was important, countries within a region would be induced to “catch up” with the highest level of liberalization reached within the region (the regional “norm”), either due to a reduction in uncertainty regarding the benefits of reform or due to competition for external capital flows.

The higher the gap between the maximum level of liberalization achieved in the region (REG_FL_i,t−1) and the level of a country’s state of liberalization (FL_i,t−1), the higher would be probability of further liberalization:

${\Delta }F{L}_{it}={\theta }_{1}F{L}_{i,t-1}(1-F{L}_{i,t-1})+{\theta }_2(\mathit{REG}\_F{L}_{i,t-1}-F{L}_{i,t-1})+{\epsilon }_{it}$

In addition, various shocks can dislodge the status quo and ideology and structure can influence the speed of reforms. We include in our empirical analyses a set of variables reflecting these influences (SHOCKS_it, IDEOLOGY_it, and STRUCTURE_it respectively). This implies the following specification:

$\begin{array}{l}{\Delta }F{L}_{it}={\theta }_{1}F{L}_{i,t-1}(1-F{L}_{i,t-1})+{\theta }_2(\mathit{REG}\_F{L}_{i,t-1}-F{L}_{i,t-1})\\ +{\theta }_3\prime {\mathit{SHOCKS}}_{it}+{\theta }_4\prime {\mathit{IDEOLOGY}}_{it}+{\theta }_5\prime {\mathit{STRUCTURE}}_{it}+{\epsilon }_{it}& & (1)\end{array}$

In the category of “shocks,” we first consider different types of crises, or economic adversity. These include, as discussed in Section III: a balance of payments crisis dummy (BOP_it), a banking crisis dummy (BANK_it), a recession dummy (RECESSION_it), and a high inflation dummy (HINFL_it).¹⁸ For the political “shock” variables, we examine the “honeymoon hypothesis” by including a dummy variable indicating the incumbent executive’s first year in office (FIRSTYEAR_it). The influence of IFIs on policy reform is proxied by an IMF program dummy (IMF_it). And to explore the influence of the global factors we include international interest rates (USINT_t).¹⁹ To assess the effect of political orientation on reform, we include dummy variables for left-wing and right-wing governments (LEFT_it, RIGHT_it); centrist governments are the omitted category. The only structural variable included here is trade openness (OPEN_it); in the sensitivity analysis below, we consider additional structural variables that may be relevant but are effectively time-invariant, such as legal origin and the system of government. In our discussion, we summarize also results obtained by interacting the ideology and structural variables with the “shock” variables, on the premise that ideology and structure may indirectly influence the speed of response following shocks. Descriptions of the variables used can be found in the Data Appendix.

Results in Table 7 are presented without and with country fixed effects. Columns 1 and 4 show the result of regressing ΔFL_it on the domestic learning dynamic, FL_i,t−1 (1 − FL_i,t−1), and on the regional diffusion variable, (REG_FL_i,t-1, − FL_i,t−1). The coefficient on FL_i,t−1 (1 − FL_i,t−1) is positive and significant at the 1 percent level, confirming the conjecture that status quo bias decreases as financial liberalization increases, and verifying the inverse U-shaped relationship between policy change and the level of liberalization suggested in Table 5. The regional diffusion effect is additional to domestic learning and is also strongly significant, suggesting that regional leaders exercise a strong pull towards liberalization.

Table 7.

Ordered Logit Estimates: Benchmark Specification (Equation 1)

Note 1: The dependent variable is the change in the Financial Liberalization Index, ?FL_it.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

Table 7.

Ordered Logit Estimates: Benchmark Specification (Equation 1)

	Without Country Fixed Effects			With Country Fixed Effects
FLi,t-1×(1-FLi,t-1)	4.001 *** (0.966)	4.167 *** (0.991)	4.188 *** (0.984)	6.316 *** (1.512)	6.388 *** (1.579)	6.295 *** (1.577)
REG_FLi,t-1FLi,t-1	0.842 *** (0.248)	0.948 *** (0.301)	0.993 *** (0.314)	1.940 *** (0.695)	2.319 *** (0.834)	2.324 *** (0.830)
BOPit		0.460 ** (0.189)	0.439 ** (0.190)		0.471 * (0.245)	0.488 ** (0.244)
BANKit		-0.985 *** (0.361)	-0.993 *** (0.361)		-0.959 *** (0.363)	-0.965 *** (0.358)
RECESSIONit		-0.059 (0.345)	-0.056 (0.338)		-0.063 (0.361)	-0.059 (0.366)
HINFLit		-0.209 (0.401)	-0.264 (0.399)		-0.353 (0.643)	-0.343 (0.652)
FIRSTYEARit		0.240 (0.222)	0.194 (0.231)		0.242 (0.262)	0.242 (0.270)
IMFit		0.295 (0.185)	0.326 * (0.186)		0.348 (0.269)	0.324 (0.261)
USINTt		-0.070 * (0.038)	-0.066 * (0.038)		-0.087 ** (0.042)	-0.089 ** (0.044)
LEFTit			0.242 (0.242)			-0.072 (0.366)
RIGHTit			0.169 (0.190)			-0.196 (0.338)
OPENit			-0.001 (0.001)			0.006 (0.009)
Log L	-762.662	-748.17	-747.20	-749.1	-734.56	-734.11
Wald test of joint significance (p -value)	0.000	0.00	0.00	0.000	0.00	0.00
Number of observations	805	805	805	805	805	805

Note 1: The dependent variable is the change in the Financial Liberalization Index, ?FL_it.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

View Table

Table 7.

Ordered Logit Estimates: Benchmark Specification (Equation 1)

	Without Country Fixed Effects			With Country Fixed Effects
FLi,t-1×(1-FLi,t-1)	4.001 *** (0.966)	4.167 *** (0.991)	4.188 *** (0.984)	6.316 *** (1.512)	6.388 *** (1.579)	6.295 *** (1.577)
REG_FLi,t-1FLi,t-1	0.842 *** (0.248)	0.948 *** (0.301)	0.993 *** (0.314)	1.940 *** (0.695)	2.319 *** (0.834)	2.324 *** (0.830)
BOPit		0.460 ** (0.189)	0.439 ** (0.190)		0.471 * (0.245)	0.488 ** (0.244)
BANKit		-0.985 *** (0.361)	-0.993 *** (0.361)		-0.959 *** (0.363)	-0.965 *** (0.358)
RECESSIONit		-0.059 (0.345)	-0.056 (0.338)		-0.063 (0.361)	-0.059 (0.366)
HINFLit		-0.209 (0.401)	-0.264 (0.399)		-0.353 (0.643)	-0.343 (0.652)
FIRSTYEARit		0.240 (0.222)	0.194 (0.231)		0.242 (0.262)	0.242 (0.270)
IMFit		0.295 (0.185)	0.326 * (0.186)		0.348 (0.269)	0.324 (0.261)
USINTt		-0.070 * (0.038)	-0.066 * (0.038)		-0.087 ** (0.042)	-0.089 ** (0.044)
LEFTit			0.242 (0.242)			-0.072 (0.366)
RIGHTit			0.169 (0.190)			-0.196 (0.338)
OPENit			-0.001 (0.001)			0.006 (0.009)
Log L	-762.662	-748.17	-747.20	-749.1	-734.56	-734.11
Wald test of joint significance (p -value)	0.000	0.00	0.00	0.000	0.00	0.00
Number of observations	805	805	805	805	805	805

Note 1: The dependent variable is the change in the Financial Liberalization Index, ?FL_it.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

The crisis/adversity dummies are added to the regression in columns 2 and 5. Consistent with Table 4, the coefficient on the balance of payments crisis dummy variable is positive, suggesting that these crises are an impetus to reform, while the coefficient on the banking crisis dummy is negative, confirming the earlier finding that reforms are less likely following banking crises and reversals more likely. The effects of the recession and high inflation dummy variables are insignificant, possibly because, as Table 4 shows, both reforms and reversals become more likely.

In columns 3 and 6 “shock” variables are included. The coefficient on the dummy variable for the first year in office is positive but not significant. Recall, the bivariate analysis (Table 3) showed change in both directions during the first year in office. The positive but insignificant coefficient possibly results from the increased likelihood of reform being (partially) offset by the increased likelihood of reversals. The IMF program dummy has a positive coefficient, indicating movement towards reform during periods of IMF programs. It is marginally significant when country dummies are not included, but becomes insignificant when country dummies are added, suggesting that the factors that lead to an IMF program also hasten financial reform. In line with Bartolini and Drazen (1997), a rise in U.S. interest rates is seen to slow down the pace of financial sector liberalization. Not surprisingly, since US interest rates vary only over time and not across countries, their effect is seen to be stronger when the within-country variation is isolated in the fixed effects regression.

Finally, consider the ideology and structure variables that display some variation over time (other variables are considered in the sensitivity analysis). The coefficient on the left-wing dummy variable is larger than that on the right-wing dummy variable, suggesting that left-wing governments tend to reform more than right-wing governments. However, the results indicate no statistical difference across party ideologies—a somewhat less spectacular result that that proposed by Cukierman and Tommasi (1998), but impressive and important nevertheless. Lastly, a country’s openness to trade, as measured by the sum of imports and exports relative to GDP, is not significant in these regressions.

Interacting the ideology and structural variables with the “shock” variables did not produce any additional significant results. Coefficient estimates were more positive for left-wing governments during balance of payments crises, and more negative for right-wing governments during banking crises, but as in the uninteracted case, these differences were not significant. Also, Rajan and Zingales (2002) argue that it is the combination of trade openness and availability of foreign capital that makes incumbents more willing to liberalize the financial sector. To capture this idea we interacted of OPEN_it with USINT_t (the latter proxying for the availability of foreign capital), but the results remained insignificant.

The coefficient estimates in Table 7 do not give the marginal effects of the right-hand side variables on the probabilities of reform, reversal, and status quo. Calculations (available from the authors) of these marginal effects on probabilities, evaluated at the means of the right-hand side variable, show that a balance of payments crisis increases the likelihood of reform by 8 percentage points; a banking crisis lowers it by 12 percentage points. An IMF program increases reform likelihood by 5 percentage points. The effect of a one-unit change in REG _FL_i,t−1 (i.e., the “reform leader” going from fully repressed to fully liberalized) increases reform likelihood by 35 percentage points.

To sum up, status quo bias declines as the financial sector becomes more liberalized— countries with highly repressed financial sectors tend to stay that way, but once reforms arc initiated, the likelihood of additional reforms increases. Regional diffusion effects arc strong. The easing of U.S. interest rates appears to lower costs of financial reform. IMF programs seem to be associated with faster reform, but the results caution that this association may reflect unobserved country features. While balance of payments crises tend to increase the likelihood of financial reforms, banking crises tend to increase the likelihood of reversals. The first year in office is associated with policy changes in both directions. Left wing and right wing governments are seen to operate similarly in similar situations, and openness to trade does not, on average, increase the pace of reform.

B. Alternative Specifications

Two important assumptions guided the benchmark empirical specification. We assumed that the desired level of financial liberalization is full liberalization. This may not be the case if, for example, institutional quality limits the prudent level of liberalization. We also assumed that the speed of adjustment to the desired level of liberalization is a function only of the current level of liberalization. Again, shocks of various kinds may induce policy reforms through changing the “adjustment” parameter. Here we examine the implications of relaxing both assumptions. The results reported above remain robust. However, interpretation is enhanced since the various interaction terms introduced in these more complex specifications suggest ranges of the state of financial liberalization where particular influences may be significant, even though they are not seen to be significant on average.

To allow for varying levels of “desired” financial liberalization, we relax the assumption that the desired level of financial liberalization is full liberalization, i.e., that $F{L}_{it}^{*}=1$ and explore two alternative specifications. The first alternative is that the desired level of liberalization is constant but not necessarily equal to 1, so that $F{L}_{it}^{*}=c$. Under this assumption, the coefficients on FL_i,t−1 and $F{L}_{i,t-1}^2$ are no longer constrained to be equal. Instead, when combined with Equation (1) and our earlier assumption that α = k·FL_i,t−1, we obtain the alternative specification

$\begin{array}{l}{\Delta }F{L}_{it}={\theta }_{1}F{L}_{i,t-1}+{\theta }_{2}F{L}_{i,t-1}^2+{\theta }_3(\mathit{REG}\_F{L}_{i,t-1}-F{L}_{i,t-1})\\ +{\theta }_4\prime {\mathit{SHOCKS}}_{it}+{\theta }_5\prime {\mathit{IDEOLOGY}}_{it}+{\theta }_6\prime {\mathit{STRUCTURE}}_{it}+{\epsilon }_{it}& & (2)\end{array}$

where θ₁ = kc is expected to be positive, and θ₂ = −k is expected to be negative. Testing whether θ₁ = −θ₂ would be equivalent to testing whether the assumption that $F{L}_{it}^{*}=1$ is valid.

A second alternative is that the desired level of financial liberalization increases with a country’s level of economic development. Less developed countries will tend to have poorer institutional quality that limits the capacity to undertake far-reaching reform; also, at lower levels of income, governments may perceive more opportunities to channel credit toward sectors that can generate positive externalities. This would suggest that $F{L}_{it}^{*}=a+b{Y}_{it}$, where Y_it is a measure of economic development, and would imply the following specification:

$\begin{array}{l}{\Delta }F{L}_{it}={\theta }_{1}F{L}_{i,t-1}+{\theta }_{2}F{L}_{i,t-1}^2+{\theta }_3(F{L}_{i,t-1}\cdot Y_{it})+{\theta }_4(\mathit{REG}\_F{L}_{i,t-1}-F{L}_{i,t-1})\\ +{\theta }_5\prime {\mathit{SHOCKS}}_{it}+{\theta }_6\prime {\mathit{IDEOLOGY}}_{it}+{\theta }_7\prime {\mathit{STRUCTURE}}_{it}+{\epsilon }_{it}& & (3)\end{array}$

where θ₁=ka and θ₃ = kb are expected to be positive, and θ₂ = −k is expected to be negative. The measure of the level of development used is GDP per capita in PPP terms from the Penn World Tables.

The estimation results for specifications (2) and (3) are reported in Table 8, again with and without fixed effects. The main results from the previous section are preserved in the alternative specifications, namely the inverse U-shaped relationship between ΔFL_it and FL_i,t−1, the positive and negative effects of BOP_it and BANK_it, respectively, and the influence of world interest rates. The term representing the regional gap remains positive in all estimations and significant in two of them. Once again the IMF dummy is significant without country fixed effects but becomes insignificant when fixed effects are included. All other variables remain insignificant as before.

Table 8.

Ordered Logit Estimates: Alternative Specifications (Equations 2 and 3)

Note 1: The dependent variable is the change in the Financial Liberalization Index.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

Table 8.

Ordered Logit Estimates: Alternative Specifications (Equations 2 and 3)

	Without Country Fixed Effects		With Country Fixed Effects
FLi,t-1	4.161 *** (0.974)	4.314 *** (0.973)	6.130 *** (1.532)	6.662 *** (1.583)
(FLi,t-1)2	-4.324 *** (1.053)	-6.091 *** (1.382)	-6.701 *** (2.043)	-9.763 *** (2.641)
FLi,t-1×Yi,t-1		0.103 ** (0.043)		0.234 ** (0.097)
REG_FLi,t-1-FLi,t-1	0.896 ** (0.426)	0.568 (0.487)	1.902 (1.291)	2.070 * (1.242)
BOPit	0.440 ** (0.190)	0.489 ** (0.193)	0.471 ** (0.233)	0.454 * (0.233)
BANKit	-0.988 *** (0.359)	-0.980 *** (0.357)	-0.942 *** (0.343)	-1.014 *** (0.344)
RECESSIONit	-0.043 (0.334)	-0.029 (0.333)	-0.060 (.0.368)	-0.035 (0.373)
HINFLit	-0.268 (0.394)	-0.215 (0.411)	-0.334 (0.647)	-0.429 (0.634)
FIRSTYEARit	0.194 (0.231)	0.153 (0.230)	0.238 (0.273)	0.250 (0.271)
IMFit	0.312 * (0.193)	0.400 ** (0.199)	0.331 (0.261)	0.370 (0.266)
USINTt	-0.067 * (0.039)	-0.071 * (0.039)	-0.093 ** (0.044)	-0.089 ** (0.044)
LEFTit	0.251 (0.239)	0.149 (0.226)	-0.031 (0.350)	-0.092 (0.354)
RIGHTit	0.187 (0.196)	0.183 (0.187)	-0.144 (0.333)	-0.091 (0.349)
OPENit	-0.001 (0.001)	0.000 (0.001)	0.007 (0.009)	0.010 (0.010)
Log L	-747.13	-744.94	-733.91	730.11
Wald test of joint significance (p -value)	0.00	0.00	0.00	0.00
Number of observations	805	805	805	805

Note 1: The dependent variable is the change in the Financial Liberalization Index.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

View Table

Table 8.

Ordered Logit Estimates: Alternative Specifications (Equations 2 and 3)

	Without Country Fixed Effects		With Country Fixed Effects
FLi,t-1	4.161 *** (0.974)	4.314 *** (0.973)	6.130 *** (1.532)	6.662 *** (1.583)
(FLi,t-1)2	-4.324 *** (1.053)	-6.091 *** (1.382)	-6.701 *** (2.043)	-9.763 *** (2.641)
FLi,t-1×Yi,t-1		0.103 ** (0.043)		0.234 ** (0.097)
REG_FLi,t-1-FLi,t-1	0.896 ** (0.426)	0.568 (0.487)	1.902 (1.291)	2.070 * (1.242)
BOPit	0.440 ** (0.190)	0.489 ** (0.193)	0.471 ** (0.233)	0.454 * (0.233)
BANKit	-0.988 *** (0.359)	-0.980 *** (0.357)	-0.942 *** (0.343)	-1.014 *** (0.344)
RECESSIONit	-0.043 (0.334)	-0.029 (0.333)	-0.060 (.0.368)	-0.035 (0.373)
HINFLit	-0.268 (0.394)	-0.215 (0.411)	-0.334 (0.647)	-0.429 (0.634)
FIRSTYEARit	0.194 (0.231)	0.153 (0.230)	0.238 (0.273)	0.250 (0.271)
IMFit	0.312 * (0.193)	0.400 ** (0.199)	0.331 (0.261)	0.370 (0.266)
USINTt	-0.067 * (0.039)	-0.071 * (0.039)	-0.093 ** (0.044)	-0.089 ** (0.044)
LEFTit	0.251 (0.239)	0.149 (0.226)	-0.031 (0.350)	-0.092 (0.354)
RIGHTit	0.187 (0.196)	0.183 (0.187)	-0.144 (0.333)	-0.091 (0.349)
OPENit	-0.001 (0.001)	0.000 (0.001)	0.007 (0.009)	0.010 (0.010)
Log L	-747.13	-744.94	-733.91	730.11
Wald test of joint significance (p -value)	0.00	0.00	0.00	0.00
Number of observations	805	805	805	805

Note 1: The dependent variable is the change in the Financial Liberalization Index.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

Next, since α measures the degree of status quo bias, an alternative econometric specification is for the shock, ideology and structure variables to enter via their effect on α, rather than directly on ΔFL_it. That is,

${\Delta }F{L}_{it}=\alpha (F{L}_{it}^{*}-F{L}_{i,t-1})+{\epsilon }_{it}$

where

$\alpha ={\theta }_1\cdot F{L}_{i,t-1}+{\theta }_2(\mathit{REG}\_F{L}_{i,t-1}-F{L}_{i,t-1})+{\theta }_3\prime {\mathit{SHOCKS}}_{it}+{\theta }_4\prime {\mathit{IDEOLOGY}}_{it}+{\theta }_5\prime {\mathit{STRUCTURE}}_{it}$

Below we present the case where $F{L}_{it}^{*}=c$.²⁰ This implies the following specification:

$\begin{array}{ccc}\begin{array}{l}{\Delta }F{L}_{it}={\theta }_{1}F{L}_{i,t-1}+{\theta }_{2}F{L}_{i,t-1}^2\\ +{\theta }_3(\mathit{REG}\_F{L}_{i,t-1}-F{L}_{i,t-1})+{\theta }_4(\mathit{REG}\_F{L}_{i,t-1}-F{L}_{i,t-1})\cdot F{L}_{i,t-1}\\ +{\theta }_5\prime {\mathit{SHOCK}}_{it}+{\theta }_6\prime {\mathit{SHOCK}}_{it}\cdot F{L}_{i,t-1}\\ +{\theta }_7\prime {\mathit{IDEOLOGY}}_{it}+{\theta }_8\prime {\mathit{IDEOLOGY}}_{it}\cdot F{L}_{i,t-1}\\ +{\theta }_9\prime {\mathit{STRUCTURE}}_{it}+{\theta }_{10}\prime {\mathit{STRUCTURE}}_{it}\cdot F{L}_{i,t-1}+{\epsilon }_{it}\end{array}& & (4)\end{array}$

We report the results from these estimations in Table 9. The inverse U-shaped relationship between ΔFL_it and FL_i,t−1 continues to hold. The interaction between the regional liberalization gap with the level of a country’s liberalization is positive and significant, suggesting that regional competitive spirit becomes more pronounced as the level of liberalization increases. The respective positive and negative effects of BOP_it and BANK_it on ΔFL_it remain, as before. The dummy for the first year in office is positive and significant at the 10 percent level, and the interaction term is negative and significant at 5 percent. The implication is that first year in office has a positive effect on reforms when the level of liberalization is low but that this effect tapers off as the level of liberalization rises. Similarly, the IMF program dummy is positive and significant at the 5 percent level, but the coefficient on the interaction term IMF_it · FL_i,t-1 is negative and is also significant at the 5 percent level.

Table 9.

Ordered Logit Estimates: Alternative Specifications (Equation 4)

Note 1: The dependent variable is the change in the Financial Liberalization Index.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

Table 9.

Ordered Logit Estimates: Alternative Specifications (Equation 4)

	Without Country Fixed Effects		With Country Fixed Effects
FLi,t-1	3.824	(1.445) ***	4.486	(2.085) **
(FLi,t-1)2	-3.644	(1.538) **	-2.930	(2.580)
REG_FLi,t-1-FLi,t-1	0.094	(0.569)	1.277	(1.378)
(REG_FLi,t-1-FLi,t-1)×FLi,t-1	3.387	(1.852) *	8.918	(2.509) ***
BOPit	0.834	(0.292) ***	0.796	(0.421) *
BOPit ×FLi,t-1	-0.901	(0.601)	-0.984	(0.870)
BANKit	-0.901	(0.539) *	-1.064	(0.590) *
BANKit ×FLi,t-1	-0.079	(1.071)	-0.006	(1.163)
RECESSIONit	-0.518	(0.461)	-0.555	(0.546)
RECESSIONit ×FLi,t-1	1.224	(0.890)	1.185	(0.977)
HINFLit	0.346	(0.496)	0.415	(0.778)
HINFLit ×FLi,t-1	-2.535	(1.356) *	-3.568	(1.520) **
FIRSTYEARit	0.564	(0.283) **	0.607	(0.319) *
FIRSTYEARit × FLi,t-1	-1.140	(0.621) *	-1.041	(0.717)
IMFit	0.752	(0.259) ***	0.692	(0.336) **
IMFit × FLi,t-1	-1.502	(0.674) **	-1.762	(0.891) **
USINTt	-0.073	(0.041) *	-0.088	(0.043) **
LEFTit	-0.117	(0.391)	-0.611	(0.530)
LEFTit ×FLi,t-1	0.661	(1.009)	1.323	(1.208)
RIGHTit	0.366	(0.304)	0.151	(0.381)
RIGHTit ×FLi,t-1	-0.397	(0.963)	-0.132	(1.210)
OPENit	0.003	(0.006)	0.029	(0.016) *
OPENit ×FLi,t-1	-0.005	(0.006)	-0.045	(0.018) **
Log L	-737.97		-718.92
Wald test of joint significance (p -value)	0.00		0.00
Number of observations	805		805

Note 1: The dependent variable is the change in the Financial Liberalization Index.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

View Table

Table 9.

Ordered Logit Estimates: Alternative Specifications (Equation 4)

	Without Country Fixed Effects		With Country Fixed Effects
FLi,t-1	3.824	(1.445) ***	4.486	(2.085) **
(FLi,t-1)2	-3.644	(1.538) **	-2.930	(2.580)
REG_FLi,t-1-FLi,t-1	0.094	(0.569)	1.277	(1.378)
(REG_FLi,t-1-FLi,t-1)×FLi,t-1	3.387	(1.852) *	8.918	(2.509) ***
BOPit	0.834	(0.292) ***	0.796	(0.421) *
BOPit ×FLi,t-1	-0.901	(0.601)	-0.984	(0.870)
BANKit	-0.901	(0.539) *	-1.064	(0.590) *
BANKit ×FLi,t-1	-0.079	(1.071)	-0.006	(1.163)
RECESSIONit	-0.518	(0.461)	-0.555	(0.546)
RECESSIONit ×FLi,t-1	1.224	(0.890)	1.185	(0.977)
HINFLit	0.346	(0.496)	0.415	(0.778)
HINFLit ×FLi,t-1	-2.535	(1.356) *	-3.568	(1.520) **
FIRSTYEARit	0.564	(0.283) **	0.607	(0.319) *
FIRSTYEARit × FLi,t-1	-1.140	(0.621) *	-1.041	(0.717)
IMFit	0.752	(0.259) ***	0.692	(0.336) **
IMFit × FLi,t-1	-1.502	(0.674) **	-1.762	(0.891) **
USINTt	-0.073	(0.041) *	-0.088	(0.043) **
LEFTit	-0.117	(0.391)	-0.611	(0.530)
LEFTit ×FLi,t-1	0.661	(1.009)	1.323	(1.208)
RIGHTit	0.366	(0.304)	0.151	(0.381)
RIGHTit ×FLi,t-1	-0.397	(0.963)	-0.132	(1.210)
OPENit	0.003	(0.006)	0.029	(0.016) *
OPENit ×FLi,t-1	-0.005	(0.006)	-0.045	(0.018) **
Log L	-737.97		-718.92
Wald test of joint significance (p -value)	0.00		0.00
Number of observations	805		805

Note 1: The dependent variable is the change in the Financial Liberalization Index.Note 2: Robust standard errors are in parentheses; adjusted for clustering by country. *** denotes significance at the 1 percent level; ** denotes significance at the 5 percent level; * denotes significance at the 10 percent level.

Table 10.

Goodness of Fit: Actual versus Predicted Policy

Note: Numbers in parentheses are the percentage of correctly predicted observations. For example, 59 percent (97 out of 165) of reform/large reform episodes are correctly predicted. Overall, (2+442+102) out of 805 or 68 percent of all observations are correctly predicted.

Table 10.

Goodness of Fit: Actual versus Predicted Policy

		Predicted Policy
Actual Policy		Reversal/ Large Reversal	Status Quo	Reform/ Large Reform	Total
Reversal/Large Reversal		2 (.07)	20	5	27
Status quo		16	442 (.72)	155	613
Reform/Large Reform		1	62	102 (.62)	165
	Total	19	524	262	805 (.68)

Note: Numbers in parentheses are the percentage of correctly predicted observations. For example, 59 percent (97 out of 165) of reform/large reform episodes are correctly predicted. Overall, (2+442+102) out of 805 or 68 percent of all observations are correctly predicted.

View Table

Table 10.

Goodness of Fit: Actual versus Predicted Policy

		Predicted Policy
Actual Policy		Reversal/ Large Reversal	Status Quo	Reform/ Large Reform	Total
Reversal/Large Reversal		2 (.07)	20	5	27
Status quo		16	442 (.72)	155	613
Reform/Large Reform		1	62	102 (.62)	165
	Total	19	524	262	805 (.68)

Note: Numbers in parentheses are the percentage of correctly predicted observations. For example, 59 percent (97 out of 165) of reform/large reform episodes are correctly predicted. Overall, (2+442+102) out of 805 or 68 percent of all observations are correctly predicted.

This suggests that the earlier ambiguity with respect to the role of IMF programs arose because these programs have the strongest effect on reform likelihood in countries that are still highly repressed, and most in need of reform, but the effect declines thereafter. The additional noteworthy finding is the positive effect of trade openness on the pace of reform at low levels of liberalization. As the negative coefficient on the interaction term indicates, this positive effect falls away as the level of liberalization increases.

As noted in Greene (2000), there is no widely accepted method for assessing goodness-of-fit in ordered logit models.²¹ In Table 9 we compare the actual policy implemented with the model’s predicted policy, where the predictions are generated using equation (4). To mitigate the effects of the extremely unbalanced data, we have grouped large reversals and reversals together, and reforms and large reforms together, although this still leaves the sample highly unbalanced, with reversals/large reversals comprising only 4 percent of observations. Only 7 percent (2 of the 27) of reversals are correctly predicted, but the two reversals that the model correctly predicts are the most significant reversals in our sample—Argentina in 1982 and Venezuela in 1994. The model does a much better job of predicting status quo observations (65 percent) and reform/large reform episodes (59 percent). Overall, the model correctly predicts 62 percent of observations.