New international evidence on interactions between bank size, industrial production and trade.

Wang, Zijun ; Lo, Wai-Chung ; Fung, Hung-Gay 等

Abstract

This study examines the influence of banking development and trade on industrial production for a total of 26 countries. Generalized impulse response functions based upon vector autoregression (VAR) models are utilized to provide insights into the dynamic relationship of the banking development, trade growth and the growth of industrial production. The results show that the three macro-series are related to each other for most countries, and tend to have positive impacts on each other. The relationships do not seem to be systematically-different between industrialized countries and newly industrialized or developing economies.

Key words: banks, industrial production, international trade and VAR model.

JEL Classification:C32, F14, and G21.

Introduction

Many economists view banks as traditional financial intermediaries, which collect deposits and grant loans and thus they are critical for allocating resources within an economy and for economic growth. This view is challenged in a competitive financial market, in which firms no longer rely solely on banks to raise capital for growth, leading to a situation of so-called financial disintermediation. The firms can either borrow directly from the financial market or issue securities (stocks or bonds) domestically or globally to raise capital. Given the rapid growth of financial markets, banks on one hand try to securitize their loan portfolios to maintain competitiveness, and on the other hand they become more involved in risk management for themselves and for their customers in light of large interest rate and exchange rate volatility [Mehta and Fung (2004)]. At the same time, banks are still subject to heavy regulations by government to maintain interest rate stability to achieve steady economic goals and inter-temporal smoothing of investment and savings [Allen and Gale (2000)].

The above discussion raises the question what roles the banks play in an open economy. It is important to assess the mechanism through which banks affect economic growth and how they affect international trade performance in a global environment. If banks directly affect economic growth, this will confirm the traditional view of banks. However, if we observe banks only indirectly affect economic growth, the result will lead support to the assertion that banks are changing their roles to become risk management and income smoothing in the economy. It is hypothesized that banks improve trade through assumption of risk and removal of information asymmetry for trade transactions across countries because of increasing currency risk and payment uncertainty globally.

Empirical evidence indicates that financial market development contributes to economic growth [see, for example, Levine and Zervos (1998); Rajan and Zingales (1998)]. The connection of financial development and economic growth is also widely studied [Levine, Loayza and Beck (2000); Rousseau and Wachtel (2000)]. We complement the current literature by studying the dynamics between the growth of the banking sector and the growth in the industrial production.

This study also examines the trade-led growth hypothesis, and the feedback/ interaction of economic growth and trade. Many studies such as Feder (1983), Ram (1987), Serletis (1992), and Fung et al. (1994) have supported significant relationship between trade and growth. We improve these studies by incorporating the role of banks in these relationships. We extend these studies using an impulse response analysis based on vector autoregressions (VAR models) that have played an important role in summarizing the dynamic interactions among the bank size, industrial production, and trade over time. Since the variables in a VAR model may be cointegrated [Johansen (1988, 1991)], we incorporate cointegration structure in our analysis to avoid misspecification problem, along with the generalized impulse response approach.

The paper provides several interesting findings. First, the banking sector tends to lead growth in industrial production at a relatively longer time horizon, while the growth in trade leads to the growth in industrial production at a shorter time horizon. Second, the development of the banking sector seems to be exogenous, but trade growth leads to increase in bank size in trade-oriented countries. Third, industrial production affects trade significantly, but we find that the impact of bank size on trade is significant in industrialized countries, implying that banks help facilitate trade growth by mitigating some of the risk factors. Finally, trade appears to react more responsively to banking development after the collapse of the Bretton Wood system, providing additional evidence that banks)May a more important role in trade transactions as Countries face more currency risk in a global economy.

The rest of the paper is organized as follows: Section 2 explains the VAR modeling and the calculation of generalized impulse responses; Section 3 briefly discusses the data; Section 4 reports and discusses impulse responses of each of the three variables (bank size, industrial production, and trade) to innovations in the other two variables. In this section, we also briefly investigate how the interactions between the variables may have evolved over time. The final section is the conclusion.

Empirical Framework

Researchers often use either the recursive Cholesky factorization pioneered by Sims (1980), or a non-recursive strategy suggested by Blanchard and Watson (1986), Bernanke (1986), and Sims (1986), among others. Both methods rely on economic theory or other prior knowledge to determine the ordering of variables in VAR models and provide information about the linkages between innovations. Different orderings may lead to quite different results depending on the degree of correlations between different shocks. A particular ordering implies that we impose a priori economic structure on the multivariate processes (when the structure itself is often the subject of study). Unfortunately, predictions of economic theory are often ambiguous.

Specifically, let [Y.sub.t] denote a (m x 1) vector of stationary processes under investigation, where m is three (bank size, industrial production, and international trade) in our case. The dynamic relationship among these processes can be modeled as a VAR of order k,

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], ... (1)

where [Y.sub.t] = ([Y.sub.1t], [Y.sub.2t], ..., [Y.sub.mt])', m is an (m x 1) vector of intercept terms, [F.sub.j] is (m x m) coefficient matrices, [[epsilon].sub.t] is an (m x 1) vector of innovations following a multivariate normal distribution with variance S. Furthermore, [[epsilon].sub.t] can be correlated only contemporaneously. Model (1) has an infinite moving average representation,

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]. ... (2)

Note that in (2) we omit the part corresponding to the intercept term in (1), since it is irrelevant in subsequent calculations. The commonly used orthogonalized impulse responses is defined as

[partial derivative][Y.sub.t + s]/[partial derivative][[delta].sub.j] = [[PSI].sub.s][Pe.sub.j], s = 1, 2, ..., ... (3)

where P is recursive-form Cholesky factor of [SIGMA], [e.sub.i] is the selection vector that has all elements equal to 0 except for the ith element being 1. As a (m x 1) vector, equation (3) measures the change of [Y.sub.t] at period t + s if [Y.sub.jt] were increased by one standard deviation, [[delta].sub.j], at period t, where [[delta].sub.j] is the squared root of the jth diagonal element of [SIGMA]. Under the recursive assumption, [Y.sub.1t] is contemporaneously exogenous. Furthermore, all other variables are allowed to be caused by innovations from variables preceding them, but not the ones behind them. Therefore, the traditional impulse responses in (3) critically depend on the ordering of [Y.sub.t] since P does.

We use a generalized impulse response function developed by Koop et al. (1996) and Pesaran and Shin (1998), which is invariant to the ordering of the variables. We measure the effect of a particular shock by integrating out the effects of other shocks to the VAR system using the distribution of the errors. When [[epsilon].sub.t] follows a multivariate normal distribution, Koop et al. (1996) show that the (scaled) generalized impulse responses is given by

[partial derivative][Y.sub.t + s]/[partial derivative][[delta].sub.j] = [[PSI].sub.s][SIGMA][e.sub.j]/[[delta].sub.j], s = 1, 2, ..., ... (4)

Comparing (4) with (3), it can be seen that the variance [SIGMA] in (4) replaces its orthogonalized factor, P, in (3). Therefore, the generalized decomposition is invariant to the ordering of variables in the VAR model. Recently Ewing et al. (2003) and Wang et al. (2003) apply this generalized method to study the effects of innovations in other macro variables or sectors/markets on stock returns.

To generate confidence intervals for the generalized impulse responses (4), we use the bootstrap method proposed by Runkle (1987). Specifically, we generate a complete set of artificial observations of. [Y.sub.t] based on model (1) by re-sampling from the estimated residuals. We calculate the generalized impulse responses based on the estimated parameters from the simulated data. Repeating this process, say, N times, we have an empirical distribution for the point estimate in (4). Then, a 90% confidence interval is constructed based on the 5th and 95th percentiles of the distribution. (1)

So far, we have assumed that the vector process [Y.sub.t] is stationary. However, it is now widely recognized that many macro series contain a stochastic trend, or they are nonstationary. Since the first difference of [Y.sub.t], [DELTA] [Y.sub.t] = [Y.sub.t] - [Y.sub.t-1], is stationary, [Y.sub.t] can be modeled as the following VAR in first differences:

[DELTA][Y.sub.t] = [mu] + [k-1.summation over (j = 1)] [[GAMMA].sub.j] [DELTA][Y.sub.t-j] + [[epsilon].sub.t]. ... (5)

It is possible that while the individual elements of [Y.sub.t] are nonstationary, they may have common trends. Alternatively, there exist some long run equilibrium relationships among them. Therefore, certain linear combinations of these nonstationary series may be stationary. In this case, the following error correction model is appropriate:

[DELTA][Y.sub.t] = [mu] + [PI][Y.sub.t-1] + [k-1.summation over (j = 1)] [[PI].sub.j] [[GAMMA].sub.j][DELTA][Y.sub.t-j] + [[epsilon].sub.t], ... (6)

where the new term [PI] [Y.sub.t-1] contains long-run equilibrium information. Clearly, when [Y.sub.t], is a cointegrated process, a VAR in first differences is misspecified. Note that model (6) nests model (1) and (5). Specifically, let the rank of P be r. If 1 < r < m, then [Y.sub.t] is cointegrated. If r = 0, then model (6) reduces to (5), a VAR in first differences. Finally, if r = m, or [PI] is of full rank, then (6) is equivalent to (1), a VAR in levels. This also implies that all elements in [Y.sub.t] are stationary by themselves.

After the error correction model (5) is estimated, it is straightforward to calculate the generalized impulse responses and varianee deeompositions following the steps described earlier in this section (for more details, see Pesaran and Shin, 1998).

Data

The variables used to estimate equation (1) include three variables: bank asset, industrial production (IP), and total trade balance. That is, we have [Y.sub.t] = (bank asset, IP, trade). Traditionally, researchers use the overall size of the banking sector to study the effects of banking development. Certain researchers choose other variables. For example, based on the argument that overall bank sizes include the assets of various banking institutions and hence does not focus on the credit provided by the banking sectors, Levine and Zervos (1998) use the value of the loans made by commercial banks and other deposit taking banks to the private depositors as a measure of banking development. Since our purpose is to study the dynamic relationships among the banking sector, the trade sector, and the industrial production, we use the overall size of the banking sector as a proxy of banking development.

We examine the dynamic relationships between these three variables for 26 industrialized, newly industrialized and developing economies. The choice of countries, data frequencies, and sample periods in the analysis are mainly due to the availability of data. All series are compiled from International Monetary Fund (IMF)'s International Financial Statistics (IFS) database. The series of bank asset is obtained by adding reserves, claims on government entities, claims on private sectors, etc., under the section Bank Institutions. The reported trade balances for Chile, Israel, Korea, Mexico, Uruguay are denominated in U.S. dollars. We convert them into local currencies using period average exchange rates (market rate). Depending on the nature of the variables, different methods are used to fill the missing values.

Appendix A summarizes the sample information for each country. In total we examine monthly observations of 21 countries and quarterly data of 22 countries. Following the convention, all variables used in the model are in logarithmic form. The period of study for each country varies depending on data availability from the IFS. Some data series start as early as 1958 and mostly end in 2002.

Empirical Results

To test the stationarity of each series, we first apply the univariate Augmented Dickey-Fuller test (Dickey and Fuller, 1979). The results are summarized in Appendix B for both monthly and quarterly observations. Among 63 monthly series, there are only 3 having t-statistic smaller than -3, appearing to be stationary. By the same standard, 6 quarterly series appear to be stationary out of a total of 66. Further stationarity tests indicate that the first differences of these series are all stationary. Given this evidence, we believe that all three variables, bank size, IP and trade, are I(1) nonstationary process with only a few exceptions.

We test the cointegration rank of [Y.sub.t] within the framework of (6). The optimal number of lags, k, for equation (6) is chosen based on minimization of Schwartz information criterion. While the maximum lag order is set 24 for monthly data and eight for quarterly data, we also impose a minimum of 12 and four lags for monthly and quarterly data, respectively.

Appendix C summarizes the chosen lag orders for 43 VAR models of 26 countries. For the monthly observations, 12 lags are appropriate for 11 countries, and 13 lags for the other 9 countries. The longest lag order is 14 for French data. Similarly, four. lags are required for 8 VARs with quarterly observations, and five lags for the other 12 VARs. We also find that seven lags are needed in the French quarterly data.

Having determined the lag order k, we consider the cointegration relations in the model. The cointegration rank is determined for each VAR by implementing Johansen's (1988, 1991) trace procedure. The results are reported in Appendix C. The second column shows that, for monthly observations, the numbers of VAR that has cointegration rank of 0, 1, 2, and 3 are 14, 3, 2 and 2, respectively. Alternatively, we model Bangladesh, Canada, Israel, Korea, and Netherlands as error correction models with one or two cointegration relations, Austria and Japan as VAR in levels. We model the remaining countries as VAR in first differences. For the 22 quarterly VAR we find that the numbers of countries that have cointegration rank of 0, 1, 2 and 3 are 13, 4, 1 and 4, respectively. Note that we model monthly observations of Japan, and quarterly observations of France, Japan, and Netherlands as VARs in levels, which agree with the univariate ADF test results in Appendix B, where we find that two or all three variables in these countries appear to be stationary.

Comparing column 2 with column 4, we can see that time aggregation (from monthly to quarterly observations) leads to changes in the number of cointegration relations for seven countries, but has no impact on the other nine countries. Conditional on the lag order and cointegration rank chosen above, we estimate model (1) by the maximum likelihood method (the amount of output is large, hence not reported here. But both the data and parameter estimates of the 43 VARs are available from the authors at request).

Based on equation (4), we calculate the generalized impulse responses of variable [Y.sub.i] to one standard deviation of innovation in variable [Y.sub.j] (i, j = bank size, industrial production, and trade). The responses are calculated for up to two years for out-of-sample periods (it is 1 up to 24 periods for monthly data, and 1 up to 8 periods for quarterly data). To save space, we summarize the results for s = 1, 6, 12, and 24 for monthly observations, and s = 1, 2, 4, and 8 for quarterly observations in Tables 1, 2, and 3. (Note: If 3 months is used in place of 1 month, all results reported in the paper hold.) For same reason, we also omit all variables' responses to their own innovations.

Impulse Responses of Bank Size

Table 1 summarizes the point estimates of generalized impulse responses of bank size to innovations in the variables of industrial production and trade for all 26 countries. Results based on monthly VAR are reported in the upper part of the table for 21 countries, and those based on quarterly VAR in the lower part for the remaining five countries. To test whether the point estimates reported in the table are significantly different from zero, we construct 90% confidence intervals using the simulation method briefly discussed earlier. (2)

The first four columns of Table 1 show the impulse responses of bank size to industrial production (IP). The results are mixed to innovations in IP; some have significant effect while others do not. The countries with insignificant IP effect are Austria, Denmark, Italy, Japan, Pakistan, Philippine, Spain, and U.K. for the monthly data and Australia, New Zealand, Switzerland, Uruguay and the U.S. for the quarterly data. Many of these countries are industrialized countries where the financial market is relatively well developed, implying that banks in these countries are exogenous and thus not responsive to the growth of the industrial production. Specifically, as financial markets are well developed in industrialized countries, banks are not the only source for financing. Thus, the growth of IP does not directly link to the growth in the bank sector. On the other hand, banks indeed play the vital role of channeling capital to firms in developing countries. Expansion of industrial production would lead to the growth in the banking sector in these countries. Our empirical results seem to be consistent with the observed pattern of financial disintermediation in the matured economies.

Columns 5 though 8 of Table 1 are the estimated changes in bank size following one unit shock in the trade series. It appears that a positive shock in trade significantly affect the bank size in 11 countries (Bangladesh, Canada, Denmark, India, Israel, Japan, Korea, Mexico, Pakistan, U.K., and the U.S.). The two developed countries, Japan and the U.S., which have negative effects of the impulse response, are worth discussion. The banking systems in these two countries are large and responsive to global competition and regulation. The negative impact of international trade on bank size may take two different routes. First, as the international trade grows, domestic banks expand more aggressively internationally instead of focusing only on the domestic market. The international expansion of banks at the expense of the domestic market is an attempt to capture growing international market and to follow their customers who invest abroad in other countries. As a result, the relative domestic size of the banking sector may suffer negatively.

Second, international trade entails risk of all types. Banks try to expand their services through various hedging devices (involving derivatives for risk management). The growth of the off balance items in banks has been growing in recent years as a result of derivatives transactions. This growth is not captured by the nominal size of the domestic banking sector. In addition, the securitization of assets further apparently makes the bank size appear smaller. In short, as a result of trade growth and financial disintermediaiton, the size of the banking sectors in the two countries will be affected negatively.

The mixed results of bank size to trade suggest that some countries are not responsive to trade growth, although in trade'oriented countries such as Korea and Mexico the bank size follows positively the growth in trade.

Impulse Responses of Industrial Production

Table 2 summarizes the dynamic impacts of bank size and trade on industrial production. Innovation in bank size generally has positive effects on IP (Denmark, India, Israel, Korea, Malaysia, Mexico, Spain, Australia, U.S.) However, the effects are significant and negative for Bangladesh, Norway, Switzerland, and Uruguay. The results are interpreted as indication that, in certain countries, the growth of the bank size is not efficient in stimulating growth of the economy, and thus may have an adverse effect on it. It is likely due to moral hazard problem, inefficiency, or a combination of the two.

The results in columns 5 through 8 of Table 2 clearly suggest that trade has a significant positive impact on the domestic industrial production, supporting the trade-led growth hypothesis in the literature. For example, at period 1, the responses are positive in all but three countries (Ireland, Netherlands, and Sweden).

Impulse Responses of Trade

The first four columns of Table 3 indicate that innovations in banking industry have had significant impacts on a nation's total trade balance in 13 countries. Experience in Korea is particularly interesting. The impact of banking industry on trade is not only positive and statistically significant, but also large in magnitude at all four horizons. Evidence from Mexico is similar except that the estimates are smaller in magnitude. These results suggest the banks in these countries are progressive in promoting trade growth.

Table 3 also provides strong evidence in support of influence of industrial production on trade. Growth in industrial production has positive and significant effects on trade growth in all 26 countries. Moreover, the influence seems to occur at the longer time-horizon. The magnitudes of the parameter estimates in the table also show that the impact of IP on trade is generally stronger in industrialized countries than in newly industrialized or developing countries.

Sensitivity Analysis

We conducted several sensitivity analyses to check the robustness of the results. First, it is possible that the relationships between the three variables may evolve as an economy goes through different development stages, the general economic environments of the economy and the international financial environment change. In order to examine this issue, we choose to look at the experiences of Korea and U.S. by estimating model (1) and calculating impulse responses for rolling windows of data (sub-samples). U.S. and Korea are chosen to represent developed and new industrialized economies, respectively. For the Korea data, we examine a sequence of 371 different sub-samples starting with 1962M1 to 1971M12, and ending with 1962M1 to 2002M10. Fixing the lag order at 13 (which is based all sample information, see Appendix 2), we re-estimate the cointegration rank for each sample using Johansen's (1988, 1991) method.

Figure 1 contains plots of estimated impulse responses of bank size, IP and trade to the remaining two variables in the VAR systems 12 months after shocks. Horizontal axes correspond to estimation end period of each sub-sample. Two interesting points are worth discussing. First, the collapse of the Bretton Wood system in 1971M8-1973M3 (a period further hit by the first off crisis) had significant impacts on all relationships among the three variables, although in different directions. The impact of IP on bank size increased while its impact on trade decreased following the crisis, reflecting the sensitivity of the export-oriented Korean economy to the outside environment. The impact of trade on bank size and IP also decreased. The positive impact of banking industry on industrial production had a one-time increase after the crisis, but has since remained. Second, the impacts of IP on bank size are most significant before early 1980s, a period when Korea completed its industrialization. The impact of bank size on trade was positive and increased slowly for the whole sample period, only briefly disrupted by the energy crisis.

Results for U.S. data are plotted in three panels of Figure 2. As in the case of Korea, the first energy crisis also appeared to have had significant impacts on U.S. economy. The influence of IP on bank size is negative in early periods, but has become less significant over time. Both bank size and IP have played increasingly important roles in trade growth.

[FIGURE 1 OMITTED]

[FIGURE 2 OMITTED]

To provide more evidence about the impact of the collapse of the Bretton Woods System on the dynamics of the three variables we divide the sample into two periods: before and after the collapse. The crisis period 1971M8-1973M3 (equivalently, 1971Q2-1973Q1) are excluded from the samples. Table 4 summarizes the impulse response estimates in both the pre- and post-crisis periods for The Korean and U.S. data. It can be seen from the table that the dynamic relationships between bank size, industrial production, and trade are in general strengthened after the crisis.

Researchers have found that temporal (time) aggregation may have consequences on many characteristics of usual interests to researchers, such as exogeneity, causality, cointegration, and common features (Pierse and Snell, 1995; Marcellino, 1999). For example, in application to a consumption function for the United Kingdom, Pierse and Snell (1995) find that cointegration of consumption and wealth is rejected with quarterly data but convincingly accepted with a longer span of annual data. We examine both monthly and quarterly data for the same countries for robustness. Generalized impulse responses of bank size, IP, and trade for 16 countries based on quarterly data are analyzed. Their results are quite similar with those in tables 1, 2 and 3 based on monthly observations for the same countries and thus they are not reported. The pattern suggests our results are robust and are not affected by temporal aggregation.

Summary and Conclusion

This study examines the dynamic relationships among bank size, industrial production, and trade for a total of 26 countries. Generalized impulse response functions based upon vector autoregressions are utilized to provide insights into the lasting issues. Our major findings can be summarized as follows.

First, we have a better understanding on the relationship of industrial production with bank size and trade. Although development in the banking sector in certain countries leads to the growth in industrial production, the pattern is not uniform across the sampled countries. It seems that the influence tends to be effective at a longer time horizon. On the other hand, there is strong evidence that the growth in trade leads the growth in industrial production and the influence occurs at a shorter time horizon.

Second, the development of the banking sector seems to be exogenous. This is particularly evident in developed countries. Both the growth in industrial production and trade do not show unambiguous effect on the growth of bank size. However, there are some indications that trade growth lead to increase in bank size in trade-oriented countries.

Third, the effect of industrial production on trade is significant and positive for all the sampled countries. The effect of bank size on trade is mixed, although the effect is more likely to be significant in industrialized countries. This result suggests that banks help facilitate trade growth by mitigating some of the risk factors by providing more services (such as hedging) in international trade by transactions.

Fourth, the relationship among the variables may evolve over time. We find that while the nature of the mutual influences (positive or negative) remains largely stable over time, the magnitude tends to vary during the sample periods.

Finally, the collapse of the Bretton Wood system that endorsed the fixed exchange rate regime seems to have impact on the dynamic relationship of the three macroeconomic series. Specifically, trade appears to react more responsively to banking development after the collapse of the Bretton Wood. This result provides some evidence that banks play a more important role in trade transactions as countries face more currency risk in a global economy.

APPENDIX A
Sample period

 Sample period

Country Monthly Quarterly

Australia 1959Q3-2002Q1 (a)
Austria 1958M12-1995M12 1958Q1-1996Q4
Bangladesh 1976M3-2002M3 (b)
Canada 1957M1-2002M10 1957Q1-2002Q2
Chile 1978M12-2002M11 1978Q4-2002Q3
Denmark 1974M1-2000M6 1968Q1-2000Q2
France 1962M1-1984M12 1957Q1-1998Q2
India 1963M1-2002M8 (c)
Ireland 1982M12-1996M12 (d)
Israel 1974M2-1996M6 (c) 1975Q1-2001Q3 (d)
Italy 1973M12-1998M12 1962Q1-1998Q4
Japan 1963M1-2002M10 1957Q1-2002Q3
Korea 1960M1-2002M10 (d) 1960Q1-2002Q2 (d)
Malaysia 1971M1-2002M9
Mexico 1970M1-2000M1 1981Q1-1999Q4
Netherlands 1957M1-1997M12 1957Q1-1997Q4
New Zealand 1977Q2-2002Q3
Norway 1960M1-2001M4 (f) 1960Q1-2001Q1
Pakistan 1977M7-1992M6
Philippine 1986M12-2000M8 1986Q4-1999Q3
Spain 1964M1-1998M12 1961Q1-1998Q4
Sweden 1969M12-2000M12 1969Q4-2000Q4
Switzerland 1964Q4-2002Q3
U.K. 1987M1-2002M12 1963Q1-2002Q3
Uruguay 1979Q1-2002Q3
U.S. 1957Q1-2002Q3

(a.) Values of import (1990Q1-Q4) are fitted values using an AR(4)
model. (b.) All variables are nominal values. (c.) Value of industrial
production in 1970M12 is fitted using an AR(12) model. (d.) The price
deflator is price of industrial products. (e.) The price deflator is
price of industrial products. The missing value of the index for
1990M2 is the simple average of those of 1990M1 and 1990 M3. Import.
(1989M10-1990M2) is fitted value using an AR(12) model. (f.) Bank
asset (1987M1-M8 and 1987Q1-Q2) is fitted value assuming the same
growth rate in this period.

APPENDIX B
Augmented Dickey-Fuller test results by country

 Monthly observations

Country Bank Industrial Trade
 size production

Australia
Austria -2.755 -1.533 -1.315
Bangladesh -3.297 1.918 -1.840
Canada -1.377 -2.020 -1.293
Chile -0.791 -0.587 -0.446
Denmark -0.216 -0.445 -0.217
France -1.263 -1.836 -1.219
India -0.646 0.874 0.658
Ireland 2.995 0.874 0.367
Israel -2.078 1.027 -0.271
Italy -1.166 -1.952 -2.271
Japan -1.986 -4.452 -2.444
Korea -0.649 -2.384 -2.553
Malaysia -1.544 -0.613 -0.971
Mexico -1.242 -1.692 -1.478
Netherlands -2.601 -3.419 -1.437
New Zealand
Norway -0.275 -2.265 -0.996
Pakistan -0.407 -2.170 0.988
Philippine -1.044 -1.306 0.406
Spain 0.312 0.496 -1.364
Sweden
Switzerland
U.K. 0.638 -1.278 -1.419
Uruguay
U.S.

 Quarterly observations

Country Bank Industrial Trade
 size production

Australia 0.204 -2.613 -0.657
Austria -2.586 -2.222 -1.838
Bangladesh
Canada -2.091 -2.319 -1.305
Chile -0.684 -0.068 -0.253
Denmark 0.182 0.075 -1.224
France -1.280 -2.913 -3.028
India
Ireland
Israel -2.349 -0.781 -0.217
Italy -2.207 -2.060 -2.326
Japan -3.283 -5.268 -2.626
Korea -0.763 -2.552 -3.616
Malaysia
Mexico -0.782 0.437 0.574
Netherlands -2.998 -3.379 -1.498
New Zealand -0.489 -0.691 -1.119
Norway -0.347 -2.359 -1.006
Pakistan
Philippine -1.086 -1.971 -0.210
Spain -0.780 -3.430 -1.527
Sweden 0.147 0.357 -0.807
Switzerland -1.151 -1.398 -1.608
U.K. -1.474 -1.399 -1.541
Uruguay -0.738 -2.390 -2.936
U.S. -1.376 -1.428 -1.260

Note: A drift term is included in the test. We select the number of
lagged first-differenced terms by minimizing the Schwarz information
criterion, where a maximum of 24 lags for monthly data for 8 for
quarterly data) is considered. We also impose a minimum of 12 lags for
monthly data, and 4 1ags for quarterly data.

APPENDIX C
Lag order and cointegration rank in VAR by country

 Monthly observations Quarterly observations

 Cointegration Cointegration
Country Lag order rank Lag order rank

Australia 5 1
Austria 13 3 5 0
Bangladesh 12 2
Canada 12 2 5 0
Chile 13 0 5 0
Denmark 12 0 5 0
France 14 0 7 3
India 13 0
Ireland 12 0
Israel 12 1 4 1
Italy 13 0 5 0
Japan 13 3 5 3
Korea 13 1 4 2
Malaysia 13 0
Mexico 13 0 4 0
Netherlands 12 1 5 3
New Zealand 4 0
Norway 12 0 5 0
Pakistan 12 0
Philippine 12 0 6 3
Spain 12 0 5 1
Sweden 13 0 5 0
Switzerland 4 0
U.K. 12 0 4 0
Uruguay 4 1
U.S. 5 0

Note: We select the optimal lag order in VAR models by minimizing the
Schwarz information criterion, where a maximum of 24 lags for monthly
data (or 8 for quarterly data) is considered. We set a minimum of 12
lags for monthly data, and 4 lags for quarterly data. The cointegration
rank is determined by the Johansen's (1988, 1991) trace test.

REFERENCES

Allen F. and Gale, D., (2000), Comparing Financial Systems, Massachusetts: MIT press.

Bernanke, B., (1986), Alternative explanations of the money-income correlation. Carnegie-Rochester Series on Public Policy 25, 49-99.

Blanchard, O.J., and M.W. Watson, (1986), Are business cycles all alike? In: Gordon, R. (Ed.), The American Business Cycle: Continuity and Change. University of Chicago Press, Chicago, 123-56.

Dickey, D.A., and Fuller, W.A., (1979), Distribution of the estimators for autoregressive time series with a unit root. Journal of American Statistical Association 74, 427-431.

Ewing, B. T., S. M. Forbes, and J. E. Payne, (2003), The effects of macroeconomic shocks on sector-specific returns. Applied Economics, 201-207.

Feder, G., (1983), On exports and growth, Journal of Development Economics, 59-73.

Fung, H.G., Sawhney, B., Lo, W.C., and Xiang, P., (1994), Exports, imports and industrial production: evidence from advanced and newly industrialized countries. International Economic Journal, vol. 8, 87-98.

Johansen, S., (1988), Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control 12, 231-254.

Johansen, S., (1991), Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models, Econometrica, 59, 1551-80.

Johansen, S., (2002), A small sample correction for the test of cointegrating rank in the vector autoregressions. Econometrica 70, 1929-1961.

Keep, G., Pesaran, M.H. and Potter, S.M, (1996); 'Impulse response analysis in nonlinear multivariate models, Journal of Econometrics, 74, 119-47.

Marcellino, Massimiliano, (1999), Some Consequences of Temporal Aggregation in Empirical Analysis. Journal of Business and Economic Statistics, 17, 129-36.

Metha, R. and Fung, H.G., (2004), International Bank Management, London: Blackwell Publishing Co.

Michaely, M., (1977), Exports and growth: an empirical investigation, Journal of Development Economics, 49-53.

Levine, R., Loayza N. and Beck T., (2000), Financial Intermediation and Growth: Causality and Causes, Journal of Monetary Economics, August, 46(1), 31-77.

Levine, R. and Zervos, S., (1998), Stock Markets, Banks, and Economic Growth, American Eeonomie Review, June 1998, 88(3), 537-58.

Pesaran, M.H. and Shin, Y., (1998), Generalized impulse response analysis in linear multivariate models, Economics Letters, 58, 17-29.

Pierse, R. G and Snell, A J., (1995), Temporal Aggregation and the Power of Tests for a Unit Root. Journal of Econometrics, 65: 333-45.

Rajan, R. G. and Zingales, L., (1998), Financial Dependence and Growth, American Economic Review, June 1998, 88(3), 559-86.

Rousseau P. L. and Wachtel, P., (2000), Equity Market and Growth: Cross-country Evidence on Timing and Outcomes, 1980-1995. Journal of Banking and Finance, December 2000,. 24(12), 1933-57.

Serletis, A., (1992), Export growth and Canadian economic development, Journal of Economie Development, 133-145.

Sims, C.A., (1980), Macroeconomics and reality. Econometrica 48, 1-48.

Sims, C.A., (1986), Are forecasting model useable for policy analysis? Federal Reserve Bank of Minneapolis Quarterly Review 10, 2-16.

Wang, Z., Kutan, A.M, and Yang, J., (2003), Information flows within and across sectors in the Chinese stock markets. Forthcoming in the Quarterly Review of Economics and Finance.

Zijun Wang, Private Enterprise Research Center, Texas A& M University, College Station, TX 77843, Email: z-wang@tamu.edu.

Wai-Chung Lo, School of Arts and Social Science, Open University of Hong Kong, 30 Good Shepherd Street, Homantin, Hong Kong, Email: wclo@ouhk.edu.hk.

Hung-Gay Fung, College of Business Administration, University of Missouri-St. Louis, 8001 Natural Bridge Road, St. Louis, MO 63121, Email: fungh@msx.umsl.edu.

NOTES

(1.) Both bootstrap and normal approximation are asymptotically valid methods to construct confidence intervals. However, in finite samples, they tend to produce wider confidence intervals than those of underlying true distributions, which should be kept in mind in interpreting the results presented later.

(2.) A point estimate is significant (marked by the symbol * in the table) if its corresponding confidence interval does not contain zero.

Table 1
Generalized Impulse Responses of Bank Size to Innovations
in industrial production (IP) and trade (s periods out of sample)

Country To innovations in IP

 Monthly data

 s = 1 s = 6 s = 12 s = 24

Austria 0.10 -0.10 -0.03 0.11
Bangladesh 0.01 -0.05 * 0.03 0.03
Canada 0.07 0.14 0.40 * 0.89 *
Chile -0.01 0.15 * 0.26 * 0.34 *
Denmark 0.03 0.02 -0.08 -0.08
France -0.81 0.75 * -0.16 0.10
India 0.02 0.14 * 0.07 0.08
Ireland 0.04 0.43 0.74 * 0.86 *
Israel 0.27 * 0.27 * 0.23 0.18
Italy -0.02 -0.11 -0.34 -0.38
Japan 0.00 0.03 -0.15 -0.22
Korea 0.10 * 0.21 * 0.29 * 0.31
Malaysia 0.03 0.11 * 0.10 0.12
Mexico 0.14 0.20 0.95 * 1.07 *
Netherlands -0.18 0.06 0.37 0.40 *
Norway -0.03 0.09 * 0.08 0.08 *
Pakistan 0.03 -0.06 -0.03 -0.04
Philippine 0.00 0.06 0.01 0.01
Spain 0.28 -0.06 -0.36 -0.59
Sweden 0.03 0.09 0.23 * 0.34
U.K. 0.24 0.00 0.60 0.82

 Quarterly data

 s = 1 s = 2 s = 4 s = 8

Australia 0.08 0.07 -0.12 -0.23
New Zealand 0.04 -0.08 -0.32 -0.32
Switzerland -0.02 0.01 -0.08 -0.05
Uruguay -0.17 0.02 -0.02 -0.01
U.S. 0.05 -0.02 -0.12 -0.53

Country To innovations in trade

 Monthly data

 s = 1 s = 6 s = 12 s = 24

Austria 0.03 0.05 -0.08 0.08
Bangladesh 0.03 * 0.02 0.03 * 0.04 *
Canada -0.02 0.04 0.07 * 0.09 *
Chile 0.04 -0.01 -0.04 -0.06
Denmark 0.08 * 0.02 -0.03 -0.01
France -0.07 0.21 0.06 0.14
India 0.04 * 0.01 -0.01 -0.02
Ireland 0.00 0.11 0.25 0.26
Israel 0.09 * 0.09 * 0.06 0.11
Italy 0.02 -0.07 -0.07 -0.08
Japan 0.01 -0.03 -0.10 * -0.14 *
Korea 0.05 * 0.08 * 0.09 * 0.20 *
Malaysia 0.00 0.01 -0.03 -0.01
Mexico 0.06 * -0.15 -0.27 -0.20
Netherlands 0.06 0.17 -0.01 0.16
Norway 0.00 0.02 0.03 0.03
Pakistan 0.05 * -0.05 -0.04 -0.03
Philippine 0.01 -0.04 -0.07 -0.06
Spain 0.02 -0.04 -0.09 -0.08
Sweden 0.00 0.01 -0.03 -0.02
U.K. 0.06 * -0.03 0.08 0.07

 Quarterly data

 s = 1 s = 2 s = 4 s = 8

Australia 0.02 0.02 -0.03 -0.05
New Zealand 0.13 -0.02 -0.04 -0.05
Switzerland 0.10 0.05 0.03 0.04
Uruguay 0.20 0.05 -0.11 -0.11
U.S. 0.01 -0.06 * -0.12 * -0.22 *

Table 2
Generalized impulse responses of industrial production (IP) to
innovations in bank size and trade (s periods out of sample)

Country To innovations in bank size

 Monthly data

 s = 1 s = 6 s = 12 s = 24

Austria 0.04 -0.02 -0.01 0.00
Bangladesh 0.08 -0.24 -0.43 * -0.38 *
Canada 0.02 0.09 0.08 0.16
Chile -0.02 -0.14 -0.29 -0.46
Denmark 0.03 0.20 * 0.27 * 0.23 *
France -0.05 0.03 0.04 0.02
India 0.02 0.04 0.10 * 0.12 *
Ireland 0.02 -0.03 0.07 0.00
Israel 0.24 * 0.11 0.16 0.14
Italy -0.01 0.09 0.11 0.15
Japan 0.00 0.28 * 0.82 * 1.34 *
Korea 0.14 * 0.19 * 0.34 * 0.54 *
Malaysia 0.23 0.38 * 0.30 0.25
Mexico 0.03 0.09 * 0.08 0.07
Netherlands -0.01 0.00 0.00 -0.03
Norway -0.09 -0.21 * -0.10 -0.18
Pakistan 0.30 0.06 0.35 0.22
Philippine -0.02 -0.38 -0.08 -0.25
Spain 0.02 0.03 * 0.07 * 0.09 *
Sweden 0.04 0.02 0.10 0.16
U.K. 0.07 0.02 -0.01 -0.03

 Quarterly data

 s = 1 s = 2 s = 4 s = 8

Australia 0.03 0.16 * 0.40 * 0.46 *
New Zealand 0.00 -0.03 -0.03 -0.04
Switzerland -0.03 -0.31 * 0.08 0.02
Uruguay -0.08 -0.13 * 0.04 0.05
U.S. 0.09 0.43 * 1.02 * 1.59 *

Country To innovations in trade

 Monthly data

 s = 1 s = 6 s = 12 s = 24

Austria 0.09 * 0.10 * 0.01 -0.06 *
Bangladesh 0.03 0.03 -0.05 * -0.02
Canada 0.04 * 0.03 0.04 0.03
Chile 0.16 * 0.05 0.04 0.07 *
Denmark 0.13 * 0.09 * 0.10 * 0.08
France 0.26 * 0.09 0.07 0.07
India 0.05 * 0.01 0.00 0.00
Ireland 0.11 0.14 0.17 0.16
Israel 0.06 * 0.02 0.03 0.01
Italy 0.02 0.05 0.04 0.04
Japan 0.08 * 0.09 * 0.04 -0.05
Korea 0.07 * 0.03 0.09 * 0.15 *
Malaysia 0.22 * 0.20 * 0.06 0.11
Mexico 0.09 * 0.05 * 0.04 0.05 *
Netherlands 0.03 0.03 -0.01 0.02
Norway 0.08 * 0.00 0.00 0.01
Pakistan 0.13 * 0.01 0.03 0.04
Philippine 0.12 * -0.04 0.03 -0.03
Spain 0.00 0.01 0.00 0.01
Sweden 0.08 * 0.09 * 0.09 0.11
U.K. 0.04 * 0.08 * 0.10 0.10

 Quarterly data

 s = 1 s = 2 s = 4 s = 8

Australia 0.08 * 0.12 * 0.09 0.10
New Zealand 0.13 * 0.17 * 0.11 0.13
Switzerland 0.55 * 0.24 * 0.36 * 0.34
Uruguay 0.26 * 0.16 * 0.19 * 0.31 *
U.S. 0.07 * 0.09 * -0.02 -0.16

Table 3
Generalized impulse responses of trade to innovations in bank
size and industrial production (IP) (s periods out of sample)

Country To innovations in bank size

 Monthly data

 s = 1 s = 6 s = 12 s = 24

Austria 0.05 0.06 0.10 0.13
Bangladesh 1.71 * 0.03 -0.72 * -0.32
Canada -0.17 0.19 0.34 * 0.49 *
Chile 0.43 -0.14 -0.03 -0.73
Denmark 0.55 * 0.13 -0.16 -0.06
France -0.02 0.02 0.10 0.09
India 0.38 * 0.09 0.25 0.24
Ireland 0.00 0.17 0.31 * 0.25
Israel 0.76 * 1.05 * 0.70 0.63
Italy 0.13 0.33 0.30 0.41
Japan 0.27 -0.53 -0.43 -0.13
Korea 1.17 * 0.84 * 0.68 * 0.98 *
Malaysia -0.06 0.12 0.65 0.62
Mexico 0.11 * 0.26 * 0.31 * 0.33 *
Netherlands 0.03 0.04 0.07 0.02
Norway 0.00 -0.20 -0.21 -0.14
Pakistan 1.69 * 0.51 0.13 0.25
Philippine 0.11 0.58 0.91 1.33
Spain 0.03 0.02 0.09 * 0.12 *
Sweden -0.01 0.06 0.10 0.20
U.K. 0.46 * 0.81 * 0.57 0.50

 Quarterly data

 s = 1 s = 2 s = 4 s = 8

Australia 0.06 0.12 0.48 * 0.64 *
New Zealand 0.06 0.09 0.19 0.13
Switzerland 0.17 -0.05 0.15 0.19
Uruguay 0.41 0.10 0.39 0.31
ITS 0.13 -0.23 0.70 1.80 *

Country To innovations in IP

 Monthly data

 s = 1 s = 6 s = 12 s = 24

Austria 0.33 * 0.56 * 0.12 0.04
Bangladesh 0.13 0.27 * 0.42 * 0.30 *
Canada 0.93 * 1.38 * 1.67 * 1.59 *
Chile 0.75 * 0.29 * -0.01 -0.12
Denmark 0.74 * 0.57 * 0.52 * 0.46
France 1.12 * 0.80 * 0.58 * 0.61
India 0.45 * -0.17 -0.36 * -0.34
Ireland 0.23 0.61 * 0.53 * 0.63
Israel 0.61 * 0.27 0.78 * 0.60 *
Italy 0.32 1.01 * 0.72 * 0.69
Japan 1.04 * 1.36 * 1.71 * 1.39 *
Korea 1.06 * 0.57 * 0.75 * 0.81 *
Malaysia 0.50 * 0.45 * 0.23 0.35
Mexico 0.83 * 0.65 * 0.22 0.39 *
Netherlands 0.23 0.44 * 0.65 * 0.52 *
Norway 0.49 * 0.27 * 0.33 * 0.21 *
Pakistan 0.39 * 0.04 0.11 0.11
Philippine 0.41 * 0.23 0.03 0.11
Spain 0.02 0.47 0.54 0.73 *
Sweden 0.71 * 0.53 * 0.52 0.78
U.K. 1.08 * 2.23 * 2.24 * 2.20

 Quarterly data

 s = 1 s = 2 s = 4 s = 8

Australia 0.75 * 1.26 * 1.56 * 1.14 *
New Zealand 0.86 * 1.07 * 1.28 * 1.22 *
Switzerland 0.65 * 0.76 * 0.37 0.42
Uruguay 1.09 * 0.41 0.08 0.17
US 0.64 * 1.56 * 1.46 * 1.41 *

Table 4
Generalized impulse responses of bank size, industrial production
(IP) and trade: before and after the collapse of Bretton Woods System

 To innovations in

Responses Bank size

Korea (before the collapse 1960M1-1971M6)

 s = 1 s = 12 s = 24

Bank size
IP -0.10 0.21 0.40
Trade 0.19 0.33 0.19

Korea (after the collapse: 1973M4-2002M10)

Bank size
IP 0.18 * 0.22 0.13
Trade 0.83 * 0.85 * 0.85 *

US (before the collapse: 1957Q1-1971Q2)

 s = 1 s = 4 s = 8

Bank size
IP -0.54 * 1.20 * 2.19 *
Trade 0.26 -0.59 0.48

US (after the collapse: 1973Q2-2002Q3)

Bank size
IP 0.24 * 1.28 * 1.73 *
Trade 0.20 1.05 * 1.67 *

 To innovations in

Responses IP

Korea (before the collapse 1960M1-1971M6)

 s = 1 s = 12 s = 24

Bank size -0.07 0.70 0.83
IP
Trade 2.10 * 0.66 0.67

 s = 12 s = 24 s = 1

Korea (after the collapse: 1973M4-2002M10)

Bank size 0.05 * 0.14 0.12
IP
Trade 0.81 * 0.76 * 0.81 *

US (before the collapse: 1957Q1-1971Q2)

 s = 1 s = 4 s = 8

Bank size -0.14 * -0.54 * -0.71
IP
Trade -0.12 0.54 0.02

US (after the collapse: 1973Q2-2002Q3)

Bank size 0.17 * 0.16 -0.27
IP
Trade 1.11 * 2.91 * 2.26 *

 To innovations in

Responses Trade

Korea (before the collapse 1960M1-1971M6)

 s = 1 s = 12 s = 24

Bank size 0.03 0.10 0.02
IP 0.05 * 0.07 * 0.06 *
Trade

 s = 12 s = 24

Korea (after the collapse: 1973M4-2002M10)

Bank size 0.04 * -0.11 * -0.10
IP 0.13 * 0.01 0.01
Trade

US (before the collapse: 1957Q1-1971Q2)

 s = 1 s = 4 s = 8

Bank size 0.01 -0.00 0.03
IP -0.02 -0.03 0.05
Trade

US (after the collapse: 1973Q2-2002Q3)

Bank size 0.02 -0.16 -0.36
IP 0.17 * 0.31 * 0.23
Trade