CHAPTER 12. The Quality Effect: Does Financial Liberalization Improve the Allocation of Capital?

Christopher Crowe, Simon Johnson, Jonathan Ostry, and Jeronimo Zettelmeyer
Published Date:
August 2010
  • ShareShare
Show Summary Details
Abdul Abiad, Nienke Oomes and Kenichi Ueda1 

12.1. Introduction

Is financial liberalization associated with improved capital allocation? This chapterfinds robust evidence of a “quality effect” of financial liberalization, using a newlydeveloped measure of efficiency in allocating capital across firms. The measure isessentially the variation in expected returns to investment. If financial liberalization is efficiency enhancing, this variation should be lower when markets—rather than governments—determine the allocation of credit. The hypothesis is that when government controls are reduced or removed, credit is reallocated from firms with low expected returns to firms with higher expected returns, raising expected returns for the former and reducing them for the latter.

We measure the variation in expected returns by the dispersion in Tobin’s Q, after controlling for industry, age, and leverage effects. We calculate this “Q-dispersion” for firms in five emerging-market economies—India, Jordan, Korea, Malaysia, and Thailand—from 1980 to 1994. Simple descriptive statistics show that Q-dispersion decreased in all five countries following financial liberalization. In panel regressions of Q-dispersion, we find that the coefficient on liberalizationis negative.

The regression results are robust, even when we include financial deepening asa control. Indeed, financial liberalization is strongly associated with improvedallocative efficiency but financial deepening is associated with lower allocative efficiency. Note that this distinction between financial liberalization and financialdeepening is not often made in the literature. Financial liberalization, on the one hand, refers to a reduction in the role of government, and an increase in the roleof the market. We measure this using the financial liberalization index of Abiadand Mody (2005), which summarizes de jure changes in credit controls, interestrate controls, entry barriers for banks, regulations, privatization, and restrictionson international financial transactions. Financial deepening, on the other hand, refers to an increase in the volume of financial activity and is typically measuredby indicators such as M2, credit to the private sector, or stock market capitalizationrelative to GDP. Although the two will tend to be related, they are not equivalent.2

Unlike this chapter, much of the literature on financial liberalization hasfocused on whether it has a positive “quantity effect,” as manifested in higherlevels of savings and investment. Theoretically, however, financial liberalizationcan improve the functioning of the financial sector without necessarily increasingsavings and investment. It is not surprising, then, that the empirical evidence ofa quantity effect is mixed.

A positive theoretical prediction, going back to McKinnon (1973) and Shaw (1973), is that higher interest rates, following the removal of interest rate ceilings, will generate greater savings. Higher rates of return may also result from betterinsurance against risk, which, as Obstfeld (1994) argues, may induce a shifttoward higher-risk, higher-return projects. Finally, a positive quantity effect oninvestment may be expected because increased competition among banks can leadfirms to internalize production externalities when making investment decisions (Ueda, 2006).

However, there are also reasons to expect a negative, or at least an ambiguous, effect on savings and investment. First, even if rates of return increase withimproved risk sharing or the removal of interest rate ceilings, the effect on savingswill depend on whether income or substitution effects prevail. Second, if liberalizedfinancial sectors provide for better insurance against risk, this may lower theneed for precautionary savings (Devereux and Smith, 1994).

The theoretical ambiguity of the quantity effect is reflected in empirical studies. Using both a cointegration and an augmented Euler equation approach, Bandiera and others (2000) show that, in a sample of eight developing countries, financial liberalization is not associated with an increase in savings. In fact, certainaspects of liberalization—those that reduce liquidity constraints for householdconsumption—are associated with a fall in savings. Jayaratne and Strahan (1996) find that the deregulation of bank branches in the United States in the 1970s didnot increase the volume of bank lending.

Thus far, only a few studies have attempted to estimate quality effects, and they have generally found positive results. In an early study, Cho (1988) finds thatfinancial liberalization in Korea led to a decrease in the variation in borrowingcosts, which he interprets as an improvement in allocative efficiency. 3Galindo, Schiantarelli, and Weiss (2007) also report a positive and significant effect ofliberalization on a measure of allocative efficiency, using firm-level data for 12developing countries. 4 In a somewhat different context, Chari and Henry (2003) find that capital account liberalization improves the allocation of capital acrosscountries, just as financial liberalization would improve the allocation of capitalwithin countries.

In our view, the main problem with existing studies on quality effects has beentheir definition of allocative efficiency. For example, Cho’s (1988) definition ofallocative efficiency as a reduction in the variation in borrowing costs is almosttautological, as this variation naturally decreases when governments eliminatedirected credit and interest rate controls. More importantly, even if all firms facedidentical borrowing costs, the allocation of capital would still not be efficient ifaccess to credit were determined by noneconomic factors. Galindo, Schiantarelli, and Weiss (2007) use a more sophisticated definition and argue that, if capital isallocated more efficiently after financial liberalization, more capital should flowto firms with a higher marginal product of capital. They test this hypothesis byassessing whether an investment-weighted average of ex post marginal returnsincreases relative to a naïve size-weighted average of ex post marginal returns. 5 However, a problem with this definition is that, as they themselves note, “themarginal product of capital of a perfectly efficient economy would be the same inall firms. Consequently, random allocations of capital would do as well as anyother allocation” (p. 12). In other words, in a perfectly efficient economy, theinvestment-weighted average should be equal to the size-weighted average. Butthis implies that the ratio of investment-weighted to size-weighted average ex postmarginal returns should converge to one, not diverge away from one, which is thecase for several countries in their sample. Moreover, although it is correct that exante (expected) marginal returns should be equal across firms in equilibrium, theeffect of financial liberalization on the ex post (realized) marginal returns is uncertain. As Obstfeld (1994) suggests, the improved availability of risk insurance maylead firms to adopt higher-risk, higher-return projects, thus creating a larger dispersionof ex post returns. 6

The rest of the chapter is organized as follows. Section 12.2 presents a simplemodel that explains our rationale for using Q-dispersion as our measure of variationin expected marginal returns. Section 12.3 discusses data sources and asimple bivariate analysis. Section 12.4 presents regression results and robustnes schecks and Section 12.5 concludes.

12.2. Financial Liberalization and Q-Dispersion: Theory and Measurement

We incorporate the classical Marshallian view that each firm has an optimal, industry specificoperating size. We thus write the profit function for a firm at time t as follows:

with a standard law of motion for capital:

where K denotes capital, L denotes labor, w is the real market wage, I is investment, and R is the gross interest rate. The function f is a constant-returns-toscale(CRS) production function with partial derivatives f1>0,f2>0,f11<0,f22<0,andf12>0. The function φ(It) measures the adjustment cost of investment, and satisfies φ>0andφ>0.7

Profit maximization gives the unique steady state optimal policy (K*, I*, L*) by

Also, the transition path of (K,L) to the steady state is uniquely determined. 8

In a fully liberalized financial sector, each firm faces the market interest rate, R, implying that the marginal returns to capital, given by (3), are equal acrossfirms. However, in a repressed financial sector, governments may impose pricecontrols (e.g., interest rate floors or ceilings) or quantity controls (e.g., directed credit) and both controls generate a variation in marginal returns across firms. Consider the case where a government controls interest rates and applies different rates to different firms. As is clear from (3), the variation in interest rates faced by firms generates variation in returns across firms. Alternatively, consider the case where interest rates are equal, but the investment amount I is determined by the government’either directly, via control of firms’ investment plans, or indirectly, via credit allocation. Let us denote this amount by Î . In this case, firms maximizetheir profit function (1) subject to (2) and the additional constraint I = Î . Letting λ denote the Lagrange multiplier associated with this constraint, the capital market condition (3) can then be rewritten as

If firms are constrained with respect to the amount they can invest (I^<I*), λ is positive. Conversely, if firms overinvest (I^>I*), as may happen, for example, when a government identifies specific industries for development or employment objectives, λ becomes negative.

Elimination of government controls leads to smaller variation of marginalreturns, as the market reallocates credit from the overinvesting firms to underinvestingfirms. This analysis can be generalized to a stochastic case, 9 where the realwage, the interest rate, and productivity are allowed to vary over time. In this case, our predictions apply to the ex ante, expected marginal returns to capital, ratherthan the ex post, realized marginal returns. As previously noted, the dispersion inex post marginal returns may actually increase after liberalization, if a betterfinancial system leads firms to select higher-risk, higher-return projects (Obstfeld, 1994).

An imperfect but nevertheless frequently used measure of expected marginal returns to capital is Tobin’s Q, which measures the discounted sum of expected future profits per asset. More precisely, the numerator of Tobin’s Q is the market value of its equity and debt, whereas the denominator is the replacement costs of tangible and nontangible assets. It should equal unity in perfectly functioning markets and in the absence of measurement errors. 10 We construct our measure of Tobin’s Q by making four approximations that are common in the literature. First, even though marginal Q (the ratio of the increment of market valuation tothe cost of the associated investment) provides the best estimate of the expected marginal return to capital (Hayashi, 1982), data constraints require us to proxymarginal Q by average Q. Second, because data on nontangible assets are notavailable, we follow the convention of using only tangible assets in the denominator (e.g., Blanchard, Rhee, and Summers, 1993; Bond and Cummins, 2001; and Chari and Henry, 2003). Third, data constraints require the use of book value rather than market value of debt. 11 Fourth, in the absence of data on the replacement cost of tangible assets, we approximate replacement costs by adjusting book values for cumulative inflation. 12

We attempt to correct for measurement errors (i.e., proxying marginal Q with average Q) and other factors that may affect the Q-dispersion by adjusting our estimates of Q for industry, age, and leverage effects. We adjust for industry effects to correct for the disparity between marginal and average Q that can arise from industry-specific production or adjustment cost functions—we thus focus on intra-industry allocation of capital. 13 In addition, controlling for industry effects allows us to correct for differences in Tobin’s Q because of differences in wages across industries. We also adjust for differences in the age of firms to correct for the factthat firms of different ages have different vintages of machines and factories. Controlling for age also allows us to correct for the possibility that younger firmsmay not yet be correctly valued in the stock market, as well as any measurementerrors in estimating capital stock resulting from accumulated differences in consumer-goods and investment-goods prices. 14 Finally, there may be measurement errorstemming from underinvestment as a result of debt overhang (Hennessy, 2004). 15 Specifically, if high leverage increases default probability, a firm’s managers (who are assumed to represent equity holders) will invest less than is optimal because the firmmay be taken over by creditors in the event of default. This creates an additional discrepancy between the observed average Q and the true marginal Q. Because the default probability may also be affected by liberalization, unlike the other sources of measurement error, the debt overhang problem may change the variance of themeasurement errors systematically before and after the liberalization.

In order to control for industry, age, and debt overhang effects, we run the following regression for each country and year:

where qh is the logarithm of Q for firm h = 1, …, H,16 Ageh is the differencebetween the current year and the year of establishment 17 of firm h, and Industry hj, is the binary variable, taking a value of 1 if firm h belongs to industry j, and avalue of 0 otherwise. 18 Both the linear and squared terms of the liability-assetratio are included to allow for the possibility that the default probability increasesnonlinearly with the leverage ratio. Running this regression gives us a residual

Running this regression gives us a residual, e h, which captures the componentof qh that is unexplained by the age, industry, and leverage effects, 19 and we constructan adjusted measure of Q for each firm:

Although it is unlikely that all the measurement errors of Tobin’s Q are eliminated, a change in our measured Q-dispersion will reflect a change in dispersion of true marginal product of capital so long as the distribution of any remainingmeasurement error is uncorrelated with financial liberalization.

Finally, we calculate the dispersion in q^h by using four inequality measures: the Gini coefficient, mean logarithm of deviations, Theil index, and the coefficient of variation. A comparison of these is useful because each index has different sensitivities to different ranges of the distribution. In particular, the Gini coefficient is mostsensitive to changes in q^h around the mean; the mean log deviation is most sensitive to changes in q^h at the bottom of the distribution; the coefficient of variation is mostsensitive to changes at the top end of the distribution; and the Theil index has constant sensitivity across the range of the distribution. The precise definitions of the four inequality indices are given in Appendix II of Abiad, Oomes, and Ueda (2007).

12.3. Data Description and Bivariate Analysis

Our measure of financial liberalization, described in detail in Abiad and Mody (2005), captures the various facets and gradations of financial reform. Specifically, the financial liberalization index takes as inputs the following six policy dimensions: credit controls, including directed credit toward favored sectors or industries, ceilings on credit toward other sectors, and excessively high reserve requirements; interest rate controls, including cases where the government directly controls interestrates or where floors, ceilings, or interest rate bands exist; entry barriers, including licensing requirements, limits to the participation of foreign banks, and restrictions relating to bank specialization or the establishment of universal banks; regulations, including operational restrictions (e.g., on staffing, branching, and advertising) which are considered repression, as well as prudential regulations, which are considered reforms; state ownership in the financial sector; and restrictions on international financial transactions, including restrictions on capital and current account convertibility, and the use of multiple exchange rates. 20

To compute Tobin’s Q, we use firm-level data from the International Finance Corporation’s (IFC) Corporate Finance Database, which is unique in that it covers emerging markets for most of the 1980s, during which much of the financial liberalizationtook place. From the original set of countries in the database, we eliminated countries that experienced hyperinflation, as this introduces large errors inbalance sheet data. We also eliminated countries with insufficient time coverage, particularly around the period of financial liberalization. Third, we dropped countriesthat lacked the data required to compute Tobin’s Q or our control variables. Finally, we dropped country-years with fewer than 10 firms. This left us with five countries: India, Jordan, Korea, Malaysia, and Thailand—the same five countries used by Chari and Henry (2003) for evaluating the impact of capital market liberalization. The data coverage for each country is summarized in Table 12.121

The fact that the IFC database only includes large publicly listed firms, creates abias against detecting a positive effect of liberalization on allocative efficiency and, therefore, would strengthen any finding of such a positive effect. Large firms aremore likely to be well connected and less likely to be financially constrained, evenunder financial repression. Hence, if we observe a decrease in Q-dispersion evenamong these large firms, then the efficiency gains are likely to be even larger if onecould measure Q-dispersion across firms of all sizes. Another advantage of focusing on publicly listed firms is that information on the activities of these firms is easy forinvestors to obtain, and therefore informational problems should be smaller.

Table 12.1International Finance Corporation (IFC) Corporate Financial Database Coverage (Number of firms per year)

The unbalanced sample also biases us against finding a decrease in Q-dispersion. A balanced sample contains only those firms that survived throughout the sample period; hence, biasing the sample toward firms that did not face financing constraints over sample periods. Using an unbalanced sample should eliminate this bias. And although financial liberalization may allow more marginal firms with severe credit constraints to enter the market, a better functioning financial systems hould decrease Q-dispersion even in the presence of new entrants.22

Let us now compare Q-dispersion before and after financial liberalization, using the liberalization dates specified by Demirgüç-Kunt and Detragiache (2001), which are based solely on interest rate liberalization. The liberalizationdates are 1991 for India, 1988 for Jordan, 1984 for Korea, 1987 for Malaysia, 23 and 1989 for Thailand. The results are supportive of a quality effect: in all cases, Q-dispersion declined following financial liberalization, although the degree ofdecline varied across countries. As Figure 12.1 shows, efficiency increased most strongly in Jordan and India. The Gini coefficient for Jordan, for example, dropped by 41 percent, whereas in India it decreased by 19 percent. Interestingly, the East Asian countries in our sample showed smaller gains following financial liberalization, with the Gini coefficient decreasing by 11 percent in Malaysia, 7 percent in Thailand, and by only 2 percent in Korea. The other dispersion measuresshow the same tendency.

Figure 12.1.Dispersion measures, pre- and postliberalization.

Source: Authors’ calculations.

12.4. Panel Regressions

12.4.1. Fixed Effects Regressions

In this section we employ both time and cross-country dimensions of the data on Q-dispersion and financial liberalization, and control for other factors that caninfluence the variation in expected returns. Our benchmark equation is as follows:

where i denotes country and t denotes time; Dit denotes Q-dispersion, FLIit1 the financial liberalization index with one year lag, 24 X it the vector of controlvariables, and εit the error term. Our hypothesis is that β is negative.

There are two obvious candidates that should be included in the set of control variables. The first is the ratio of stock market turnover to market capitalization, a measure of stock market liquidity. Although we have thus far assumed that markets are pricing stocks efficiently, it is not uncommon for stock prices to deviate from fundamentals, especially in thin markets typically observed in developing countries and emerging markets. At any point in time, this deviation creates an additional source of Q-dispersion across firms that does not reflect the underlying capital allocation. Hence, we need to distinguish this source of dispersion from the dispersion caused by improved efficiency in capital allocation for each country and each year. The second one is trade openness, the ratio of the sum of exports and imports to GDP. Exports and imports are directly affected by product market reforms that involve priceor quantity restrictions on goods and services. As such, trade openness acts as a proxyfor other reforms that may affect Q-dispersion through effects on firms’ profitability.

Table 12.2 reports the fixed-effect panel regression results, which shows that the coefficient on financial liberalization is negative and highly significant. This result is robust to using different measures of dispersion. Note that the effect ofstock market liquidity is negative as predicted and is generally significant. In addition, the effect of trade openness is negative and is almost always significant.

Now we try to separate the effects of financial liberalization and financial deepening, by including two different measures of financial deepening (FD)typically used in the literature. The first is bank credit to the private sector relativeto GDP, an indicator of the depth of the banking sector, and the second is stock market capitalization relative to GDP, an indicator of stock market development. Data for these indicators (as well as for stock market turnover) were taken from Beck, Demirgüç-Kunt, and Levine (2000). The regression is now expressed as

Table 12.3 presents these regression results, with one panel for each inequality measure. When the financial liberalization index and the two financial deepening indicators are included separately in the regressions (columns 1 through 3), financial liberalization is always correctly signed (negative) and strongly significant.

Table 12.2Fixed Effects Regressions
Fixed EffectsMemo: Random Effects
Financial Liberalization (t−1)30.19930.41930.91130.39830.05930.143230.535430.1259
Stock Market Turnover30.08430.17730.27230.15430.104130.21830.357630.2072
Trade Openness30.40330.75530.81630.87630.021930.04870.042430.0511
Number of countries55555555
Hausman Test Statistic:50.3237.4341.4351.75
Hausman Test p-value:
Absolute value of t statistics in bracketsIn the Hausman Test of FE vs. RE, the null hypothesis is that the Random Effects model is valid.
Table 12.3Fixed Effects Regressions With Financial Deepening Indicators
Dependent Variable: Gini Coefficient
Financial Liberalization (t-1)−0.199−0.254−0.263−0.275
Private Credit0.1030.3450.277
Stock Market Capitalization−0.0110.0840.042
Stock Market Turnover−0.084−0.110−0.095−0.121−0.097−0.120
Trade Openness−0.403−0.561−0.550−0.330−0.506−0.396
Number of countries555555
Dependent Variable: Theil Index
Financial Liberalization (t−1)−0.419−0.535−0.544−0.571
Private Credit0.2140.7240.607
Stock Market Capitalization−0.0320.1640.072
Stock Market Turnover−0.177−0.231−0.199−0.254−0.203−0.253
Trade Openness−0.755−1.089−1.047−0.601−0.956−0.714
Number of countries555555
Dependent Variable: Mean Log Deviation
Financial Liberalization (t31)−0.911−1.210−1.105−1.191
Private Credit0.7651.8611.921
Stock Market Capitalization−0.1190.255−0.037
Stock Market Turnover−0.272−0.422−0.308−0.470−0.312−0.470
Trade Openness−0.816−1.525−1.365−0.420−1.127−0.362
Number of countries555555
Dependent Variable: Squared Coefficient of Variation
Financial Liberalization (t31)−0.398−0.506−0.522−0.546
Private Credit0.1820.6740.544
Stock Market Capitalization−0.0300.1630.080
Stock Market Turnover−0.154−0.202−0.175−0.225−0.179−0.224
Trade Openness−0.876−1.194−1.153−0.733−1.075−0.858
Number of countries555555
Absolute value of t statistics in brackets.

The valuation effect, measured by stock market turnover, is also correctly signed(negative) and significant in almost all cases. However, the two financial deepeningindicators are insignificant.

When liberalization is combined with the two financial deepening indicators(columns 4 through 6), financial liberalization remains strongly significant and correctlysigned. Surprisingly, private credit now becomes significant, but with a positivesign, implying that private credit expansion without financial liberalizationworsens the efficiency of capital allocation. Although stock market capitalizationalone with financial liberalization also shows a positive sign, it is not significantwhen private credit is also added as a regressor (column 6). This is consistent withour view that capital allocation through credit is distorted under financial repression, and that credit booms without sufficient liberalization may harm an economy.

12.4.2. Allowing for Adjustment Lags

So far, we have assumed that the adjustment is quick and the capital level isadjusted to the optimal level within one year after the policy change. If adjustmentis slower, however, Q-dispersion may improve gradually over a few years. Inthis case, the current Q-dispersion should also be explained partially by the laggedvalues of Q-dispersion. To take this slow adjustment into account, the test equationneeds to be revised to

To estimate this revised test equation, we conduct a GMM dynamic panelestimation following Arellano and Bond (1991). 25 Our main results areunchanged for this specification. Table 12.4 shows that the coefficient on financial liberalization remains correctly signed in all the regressions and is statistically significant. Stock market turnover is also correctly signed and significant. Private credit is wrongly signed and often significant. As before, stockmarket capitalization is significant on its own but is not significant when alsocontrolling for private credit.

Table 12.4Arellano-Bond Dynamic Panel Regressions
Dependent Variable: Gini Coefficient
Financial Liberalization (t–1)−0.189−0.199−0.178−0.188
Private Credit0.2550.2840.215
Stock Market Capitalization0.1020.0920.071
Stock Market Turnover−0.077−0.076−0.054−0.094−0.070−0.084
Trade Openness−0.243−0.158−0.327−0.161−0.328−0.246
Lagged Dependent Variable0.4980.5060.5110.4480.4670.436
Number of countries555555
Second-order serial correlation0.450.420.360.460.400.44
Sargan test statistic (p -value):0.970.991.000.991.001.00
Dependent Variable: Theil Index
Financial Liberalization (t–1)−0.389−0.412−0.367−0.390
Private Credit0.5650.6320.493
Stock Market Capitalization0.2100.1910.142
Stock Market Turnover−0.167−0.167−0.120−0.205−0.152−0.185
Trade Openness−0.481−0.294−0.650−0.301−0.656−0.471
Lagged Dependent Variable0.4540.4580.4710.3940.4220.384
Number of countries555555
Second-order serial correlation0.760.940.970.520.690.53
Sargan test statistic (p -value):0.991.
Dependent Variable: Mean Log Deviations
Financial Liberalization (t–1)−0.766−0.886−0.708−0.851
Private Credit2.4812.6702.463
Stock Market Capitalization0.4800.4390.195
Stock Market Turnover−0.287−0.365−0.188−0.446−0.247−0.416
Trade Openness−0.871−0.201−1.245−0.197-1.245−0.415
Dependent Variable: Mean Log Deviations
Lagged Dependent Variable0.069−0.0440.092−0.1140.049−0.108
Number of countries555555
Second-order serial correlation0.540.660.560.560.500.54
Sargan test statistic (p -value):0.991.001.000.991.001.00
Dependent Variable: Squared Coefficient of Variation
Financial Liberalization (t–1)−0.385−0.403−0.363−0.382
Private Credit0.4870.5450.409
Stock Market Capitalization0.1990.1800.138
Stock Market Turnover−0.150−0.146−0.104−0.182−0.135−0.162
Trade Openness−0.533−0.362−0.690−0.376−0.699−0.542
Lagged Dependent Variable0.4940.5110.5140.4490.4660.439
Number of countries555555
Second-order serial correlation0.
Sargan test statistic (p -value):0.980.991.
Robust t statistics in brackets.

We also explored interactions between financial liberalization and financialdeepening to allow for the possibility of nonlinear effects. In particular, we tested whether both financial liberalization and financial deepening are required to realizea significant improvement in allocative efficiency. including these interactions, however, produced no interesting results.

12.4.3. Different Aspects of Financial Liberalization

Different components of financial liberalization may have different effects on Q-dispersion. To find out whether the results are driven by a specific type offinancial liberalization, we ran the regressions with each of the six subcomponents of the financial liberalization index. High correlations precluded the inclusion of all six at once, so the subcomponents are tried one at a time, with the other five components aggregated into a separate control (to avoid omitted variable bias).

Table 12.5 shows that each component is always correctly signed and often significant, even when controlling for the aggregate of the five other components, implying that all of the components seem to be associated with improved allocative efficiency. Specifically, interest rate liberalization and changes in regulations (both operational and prudential) are always significant, whereas removal of credit controls and liberalization of the capital account are significant for three of the four dispersion measures. Bank privatization and entry barriers seem to matter least for allocative efficiency, being significant for only two out of the fourdispersion measures. Moreover, the package of reforms also appears important, asthe coefficient on the aggregate of five other components is always correctly signed and statistically significant in most cases.

12.4.4. Other Robustness Checks

We conducted five robustness checks based on the Arellano-Bond regression specification. 26 The first robustness check we conducted was to drop one countryat a time, in order to investigate whether a single country was driving the results. This could have been the case because the sample contains only five countries, and Figure 12.1 seems to indicate that effects of financial liberalization werestronger in some countries than in others. We found, however, that our results held up in all cases.

As a second robustness check, we included a crisis dummy variable in our regressions to control for potential temporary effects on Q-dispersion. The crisisdummy was set equal to one if the country had experienced a currency crisis or banking crisis, based on the crisis database of Bordo and others (2001). Theoretically, the effect of a currency or banking crisis on allocative efficiency isunclear. The coefficient on the crisis dummy was found to be positive, suggesting that crises widen the variation in returns. However, the signs are almostalways insignificant, and including the crisis dummy does not change any of ourmain results.

Third, we tried interacting the financial liberalization index with countrydummies, i.e., allowing the coefficient on financial liberalization to vary acrosscountries; however, this did not generate any interesting patterns either. Thefinancial liberalization coefficient for India (which we used as the uninteractedcoefficient) was negative but insignificant; the coefficient for Jordan was significantly more negative than for India; the coefficients for Korea and Malaysia were in significantly more negative than for India; and the coefficient for Thailand was significantly more positive than for India.

Table 12.5Regressions Using Financial Liberalization Components
Dependent Variable: Gini CoefficientDependent Variable: Theil Index
FLI Component (t–1)−0.135−0.338−0.055−0.109−0.057−0.085−0.665−0.568−0.267−2.171−0.364−0.407
Remainder of FLI (t–1)−0.219−0.099−0.398−0.315−0.466−0.306−1.626−1.99−2.488−0.325−2.751−1.964
Trade Openness−0.327−0.266−0.313−0.289−0.316−0.331−1.119−1.093−1.442−1.116−1.469−1.55
Stock Market Turnover−0.044−0.064−0.068−0.057−0.063−0.04−0.176−0.259−0.277−0.271−0.227−0.114
Lagged dependent variable0.3950.4520.3940.4290.390.4340.0710.089−0.0520.098−0.0770.069
Number of countries444444444444
Dependent Variable: Mean Log DeviationDependent Variable: Squared Coeff. of Variation
FLI Component (t–1)−0.28−0.217−0.134−0.205−0.141−0.668−0.273−0.678−0.789−0.652−0.92−0.184
Remainder of FLI (t–1)−0.504−0.731−0.84−0.708−0.951−0.18−0.468−0.207−0.133−0.212−0.131−0.623
Trade Openness−0.65−0.543−0.646−0.585−0.661−0.685−0.685−0.571−0.66−0.616−0.67−0.699
Stock Market Turnover−0.098−0.138−0.143−0.127−0.132−0.087−0.075−0.112−0.118−0.101−0.11−0.067
[2.36]**[2.32]**[3.47]***[2.76]***[2.21]**[1.44][2.25]**[2.22]**[3.40]***[2.65]***[2.06]**[1.27 ]
Lagged dependent variable0.370.4260.3590.4060.3490.4090.3510.410.3550.3850.3530.388
Number of countries444444444444
Robust t statistics in brackets.

Fourth, when liberalizing the financial sector, a country might also improve closely related institutions such as law enforcement. Specifically, increased recovery of non performing loans implies a reduction of ex ante, equilibrium borrowing constraints (e.g., Albuquerque and Hopenhayn, 2004). Our regressions might pick up the effects of this potential omitted variable, so we added law enforcement measures from the International Country Risk Guide 27 to our set of controls. None of these measures were significant, and their inclusion did not change the financial liberalization effect found previously.

Lastly, we repeated all regressions while controlling for three-or five-year cumulative inflation to eliminate any measurement errors in Tobin’s Q, possibly remaining even after adjusting for inflation when constructing capital stock estimates. There are two possible off setting effects: on the one hand, across-the-board inflation “lifts all boats” and reduces inequality when measured in a log scale, though on the other hand, dispersion increases with heterogeneous inflation variation (in location and equipments), which in turn is likely correlated with across-the-board inflation. When we included inflation (the GDP deflator) in there gressions, we found that the coefficient on inflation was negative and occasionally significant, indicating that the first effect is stronger than the second. However, all key results remained the same.

12.5. Conclusion

Although recent studies have found little or no effect of liberalization on the level of savings and investment, we found robust evidence that liberalization isassociated with improved efficiency in allocating capital. With a simple generalequilibrium model, we predicted that financial liberalization, by equalizing access to credit, reduces the variation in expected returns across firms, which we measured by the dispersion in Tobin’s Q. In testing this prediction, we found that financial liberalization was negatively associated with Q-dispersion, and hence, positively associated with allocative efficiency. Inother words, the benefits of liberalization appear to be realized mainly throughits effect on the quality, not the quantity, of investment. In addition, we found that financial liberalization, rather than financial deepening, mattered the most for allocative efficiency. In fact, increasing private credit typically worsenedefficiency, suggesting that credit growth without liberalization may lead to a misallocation of credit.


    Abiad,Abdul, and Ashoka Mody,2005,“Financial Reform: What Shakes It? What Shapes It?”American Economic Review,Vol. 95,No. 1,pp. 6688.

    Abiad,Abdul,Nienke Oomes, and KenichiUeda,2007,“The Quality Effect: Does Financial Liberalization Improve the Allocation of Capital?”a revised version of IMF Working Paper 04/112. Available via the Internet:

    Albuquerque,Rui, and HugoA.Hopenhayn,2004,“Optimal Lending Contracts and Firm Dynamics,”Review of Economic Studies,Vol. 71,No. 2,pp. 285315.

    Arellano,Manuel, and Stephen Bond, 1991,“Some Tests of Specification for Panel Data:Monte Carlo Evidence and an Application to Employment Equations,”Review of Economic Studies,Vol. 58,No. 2 (April),pp. 27797.

    Bandiera,Oriana,Gerard Caprio,Patrick Hanohan, and Fabio Schiantarelli, 2000,“Does Financial Reform Raise or Reduce Saving?”Review of Economics and Statistics, Vol. 82,No. 2(May),pp. 23963.

    Thorsten,Beck,Asli Demirgüç-Kunt, and Ross Levine,2000,“A New Database on Financial Development and Structure,”World Bank Economic Review,Vol. 14,pp. 597605.

    Bekaert,Geert, and Campbell R. Harvey, 2000,“Foreign Speculators and Emerging Equity Markets,”Journal of Finance,Vol. 55 (April),pp. 565613.

    Blanchard,Olivier,C. Rhee, and Lawrence Summers,1993,“The Stock Market, Profit, and Investment,”Quarterly Journal of Economics,Vol. 108,No. 1,pp. 11536.

    Bond,Stephen and Jason Cummins,2001,“Noisy Share Prices and the Q Model of Investment,”Institute for Fiscal Studies Working Paper 01/22 (London: Institute for Fiscal Studies).

    Bordo,Michael D.,Barry Eichengreen, Daniela Klingebiel, and Maria Soledad Martinez Peria,2001,“Financial Crises: Lessons from the Last 120 Years,”Economic Policy: A European Forum,Vol. 32 (April),pp. 5182.

    Chari,Anusha, and Peter Blair Henry, 2003,“The Invisible Hand in Emerging Markets:Discerning or Indiscriminate?”University of Michigan Business School Working Paper.Available via

    Cho,Yoon Je,1988,“The Effect of Financial Liberalization on the Efficiency of Credit Allocation: Some Evidence from Korea,”Journal of Development Economics,Vol. 29 (July),pp. 10110.

    Demirgüç-Kunt,Aslí, and Enrica Detragiache,2001,“Financial Liberalization and Financial Fragility,”in Financial Liberalization: How Far, How Fast?ed. byGerardCaprio,PatrickHonohan, and Joseph E.Stiglitz(New York: Cambridge University Press).

    Devereux,MichaelB., and Gregor W.Smith, 1994,“International Risk Sharing and Economic Growth,”International Economic Review,Vol. 35 (August),pp. 53550.

    Djankov,Simeon,Caralee McLiesh, and Andrei Shleifer,2007,“Private Credit in 129 Countries,”Journal of Financial Economics,Vol. 84,No. 2,pp. 299329.

    Galindo,Arturo,Fabio Schiantarelli, and Andrew Weiss,2007,“Does Financial Liberalization Improve the Allocation of Investment? Micro Evidence From Developing Countries,”Journal of Development Economics,Vol. 83,No. 2,pp. 56287.

    Gomes,Joao,2001,“Financing Investment,”American Economic Review, Vol. 91 (December),pp. 126385.

    Hayashi,Fumio,1982,“Tobin’s Marginal q and Average q : A Neoclassical Interpretation,”Econometrica,Vol. 50,No. 1,pp. 21324.

    Hennessy,ChristopherA.,2004,“Tobin’s Q, Debt Overhang, and Investment,”Journal of Finance,Vol. LIX,No. 4,pp. 171742.

    Jayaratne,Jith, and PhilipE.Strahan,1996,“The Finance-Growth Nexus: Evidence from Bank Branch Deregulation,”Quarterly Journal of Economics,Vol. 111,No. 3,pp. 63970.

    Jovanovic,Boyan, and PeterL.Rousseau, 2002,“The Q-Theory of Mergers,”American Economic Review: AEA Papers and Proceedings,Vol. 92 (May),No. 2,pp. 198204.

    Laeven,Luc,2003,“Does Financial Liberalization Reduce Financing Constraints?”Financial Management,Vol. 21,No. 1,pp. 534.

    Love,Inessa,2003,“Financial Development and Financing Constraints: International Evidence from the Structural Investment Model,”Review of Financial Studies,Vol. 16,No. 2,pp. 76591.

    McKinnon,Ronald,1973,Money and Capital in Economic Development(Washington:Brookings Institution).

    Obstfeld,Maurice,1994,“Risk-Taking, Global Diversification, and Growth,”American Economic Review,Vol. 84,No. 5,pp. 131029.

    Sancak,Cemile,2002,“Financial Liberalization and Real Investment: Evidence from Turkish Firms,”IMF Working Paper 02/100 (Washington: International Monetary Fund).

    Shaw,EdwardS.,1973,Financial Deepening in Economic Development(New York: Oxford University Press).

    Ueda,Kenichi,2006,“Banks as Coordinators of Economic Development,”IMF Working Paper 06/264 (Washington: International Monetary Fund).

    Wurgler,Jeffrey, 2000,“Financial Markets and the Allocation of Capital,”Journal of Financial Economics,Vol. 58 (October/November),pp. 187214.

This chapter is a slightly revised version of an article that appeared in Journal of Development Economics, Vol. 87, No. 2 (2008): 270–82. Reprinted with permission from Elsevier.

We would like to thank Anusha Chari, Raymond Fisman, Xavier Gine, Nobu Kiyotaki, Aart Kraay, Inessa Love, Ashoka Mody, Jeremy Stein, Rene Stulz, Robert Townsend, and participants at the Annual Meeting of European Economic Association at the University of Amsterdam, the Sixth Jacques Polak Annual Research Conference at the IMF, and the Conference on Globalization and Financial Services in Emerging Economies at the World Bank, as well as the seminar participants atthe IMF and the World Bank for helpful comments and suggestions; and Tom Walter for excellenteditorial comments. The usual caveats apply.

For example, during the 1970s and up into the early 1980s, Japan and France had financiallydeep markets that were highly repressed. Conversely, several Latin American countries in the1990s (e.g., Peru, Argentina, and Brazil) had liberalized financial markets that were relatively shallow.

Bekaert and Harvey (2000) find that the average firm-level cost of capital declines in many emergingmarket economies after the opening of equity markets for foreign investors.

Wurgler (2000) finds that in countries with deeper financial sectors (but perhaps without liberalization), capital is better allocated in the sense that it tends to flow to growing industries. However, this is not necessarily evidence of a quality effect, because governments can artificiallystimulate growth in certain industries using directed credit and differential interest rates.

They proxy ex post marginal returns by the sales-to-capital ratio and the operating-profits-to-capitalratio, under the assumption of a constant-returns-to-scale technology.

A related strand of the literature has analyzed the effect of financial liberalization on firm-levelcredit constraints (e.g., Laeven, 2003, Love, 2003, and Sancak, 2002). However, from a general equilibrium perspective, Gomes (2001) argues that investment regressions on cash flows may notprovide meaningful measures of credit constraints both with and without Tobin’s Q as a controlvariable for potential growth opportunities. Tobin’s Q itself is affected by the presence of creditconstraints so that it should not be used as a control variable; however, without controlling for Tobin’s Q, there will be an omitted variable bias, as unobserved productivity shocks (e.g., growthopportunities) affect both cash flow and investment. Our approach is based on a general equilibriumframework and is not subject to Gomes’s critique. Moreover, our approach, by looking at Q-dispersion, accounts for the effects of liberalization not just on credit-constrained firms butalso on overinvesting (e.g., privileged) firms.

Allowing for the presence of low fixed-adjustment costs would not change the results.

We assume, for now, that these adjustments are quick (i.e., can be completed within a year) so that thesteady state values can be approximated with annual data. We relax this assumption in Section 12.4.

For a more thorough discussion of this case, see Appendix I of Abiad, Oomes, and Ueda (2007). The prediction would also follow from more complicated models with informational problems. That is, even with informational problems, the variation in marginal returns before financial liberalization would still reflect an inefficient allocation of capital, to the extent that governments allocate capital based on considerations beyond marginal productivity.

If the market valuation per unit of capital is different from the replacement cost of a unit of capital, firms have an incentive to adjust capital stock instantly. This discrepancy may also be quickly arbitraged away by investment firms looking for mergers or acquisitions (Jovanovic and Rousseau, 2002).

Although there is a standard approach to convert book values of debt to market values (Blanchard, Rhee, and Summers, 1993), this cannot be applied in our case because data on corporate bond ratesare not available for the relevant time period. According to Chari and Henry (2003), estimating the market value of debt would require further assumptions about unobservable corporate bond rates, which may be a cure worse than the ailment.

Specifically, if Kt is the reported value of tangible assets in year t, the inflation-adjusted value oftangible assets is given by Kt+(1δ).Kt1.(1+πt), where δ = 0.05 is an assumed depreciationrate and πt is the inflation rate. Here, we eliminate any additional measurement errors arising from inflation rate through different investment levels each year among firms, though we cannot correct the measurement errors in valuation at the initial year. As long as we look at changes in Q-dispersion, the initial measurement errors will not create any problem in panel regressions. All our empirical results hold when we do not adjust for inflation.

We ignore possible differences arising from patent holdings, because our focus is on developing countries, whereas most blueprints are produced by firms in industrial countries. Also, we ignore often observed, potentially bigger distortion of government interventions at industry level.

Although it is common to also control for firm size, this is not appropriate in our case because, according to our model, the firm size distribution depends directly on the extent to which financial sectors are liberalized.

We thank a referee for pointing this out.

We use the logarithm rather than the level of Q, because given a concave production function, the distribution of log (Q) better reflects the underlying distribution of capital than the distribution of Q itself. Indeed, in our data set, the distribution of Q itself is skewed to the right, whereas the distribution of the logarithm of Q is close to normal.

In the absence of data on the year of establishment for Thai firms, we measure their age as the difference between the current year and the year in which the firm was first listed at the Thai stock exchange.

We use 2-digit ISIC (rev. 2) classifications.

As the first-stage regressions are run for each country and each year, we do not report the results here.

In the theoretical analysis, we considered only the cases of credit control and interest control. However, the analysis can be easily extended to other areas of financial liberalization listed here. State ownership can be viewed as a more direct method of price and quantity control. Restrictions on international transactions can be viewed as intended to preserve effectiveness of the state control over the domestic financial system. Entry barriers may protect the monopolistic power of existing banks, possibly resulting in price and quantity distortions. Operational regulations can be viewed as a way to protect the existing system of allocating capital.

We used slightly fewer observations than mentioned in Table 12.1 because of the need to remove some outliers. We eliminated outliers by first taking the logarithm of Q, the distribution of which is close to normal, and then removing all observations further than three standard deviations from the mean. Our results were robust to using different procedures for removing outliers (e.g., excluding all observations with Q > 50 before calculating the standard deviations).

When more heterogeneous firms in production and cost functions enter after the liberalization, a possible increase in Q-dispersion could happen only if the technology is linear and heterogeneous across firms (e.g., yi = Aiki). However, as we assume, with more standard production function with decreasing marginal product of capital, the dispersion in marginal product of capital must decline bymore flexible adjustment of the size of capital, even with potentially more heterogeneous firms intotal factor productivity (e.g., yi=Aikiαli1α). Heterogeneity in the cost side can be similarly analyzed.

Interest rate decontrol in Malaysia occurred in October 1978, which predates both Demirgüç-Kunt and Detragiache’s sample and ours. However, Malaysia reimposed controls in 1985 before liberalizing them again in 1987.

We use the lagged index of financial liberalization, because of timing—some balance sheet information at the end of the firm-specific fiscal year may be dated before a regulatory change with in a calendar year. We obtained similar results using the contemporaneous financial liberalization index.

This estimation takes the first difference of the test equation and then applies GMM using instrumentalvariables, which are based on lagged dependent and independent variables. Specification tests rejectthe null hypothesis of second-order serial correlation, and the Sargan test does not reject the validityof the overidentifying restrictions. As such, the Arellano-Bond estimation of equation (10) is valid.

These robustness results are not reported but are available from the authors.

Specifically, we use “Law and Order,” “Investment Profile,” and “Repudiation (Risk) of Contractsby Government.” Measures of creditor rights from Djankov, McLiesh, and Shleifer (2007) were also considered, but show no variation for our countries over the sample period and hence get absorbed in the country fixed effect.

    Other Resources Citing This Publication