Financial Development, Trade Openness and Growth in the First Wave of Globalization.

D'Onofrio, Alexandra ; Rousseau, Peter L.

Introduction

It is now widely recognized that financial development is an important driver for long-run growth. (1) A large literature also relates trade openness and export orientation to growth. (2) But there is less emphasis on interactions between financial development, trade, and growth. Beck (2002) examines links between financial development and trade in manufactured goods in a cross-country regression analysis with recent data and finds a large impact of finance on exports and the trade balance of manufactured goods. At the same time, Baltagi et al. (2009) find that trade is a significant determinant of financial development. Bordo and Rousseau (2012) study a panel of now-developed economies from 1880 to 2010 and find cross-correlations between financial development and trade before 1914 that vanish after World War II.

The mixed findings for the postwar period likely reflect the complexity of modern global interactions. Yet the later nineteenth and early twentieth centuries are often viewed as an era of integration among the rapidly developing economies of Europe and the Americas and represent a period when global interconnections were simpler. Most of these economies also developed well-functioning financial systems by this time. In this paper, we conduct dynamic panel and individual time series analyses for 17 economies to investigate the nexus between financial development, trade, and output in a more detailed manner and over a longer historical period (1850-1929) than earlier studies. (3) We find that financial development was a key fundamental behind trade and growth, and that direct linkages between trade and growth, when present, tended to run in the opposite direction in a manner consistent with O'Rourke and Williamson (1999) and O'Rourke (2000).

The period from 1850 to 1929 was critical for many now-developed but then- emerging economies that unified politically, made significant reforms to their economic and industrial structures, and took strides in opening their markets. Bordo et al. (1999) and Obstfeld and Taylor (2001), for example, show that international capital markets were surprisingly well integrated by the end of the nineteenth century, with capital flowing flexibly across countries in response to rates of return. The world economy stepped back from rapid globalization after World War I, but by then most countries had built modern financial foundations and were experiencing sustained growth until the Great Depression put this progress on hold.

Empirics

To study historical relationships between financial development, trade, and growth in the first wave of globalization, we turn to a number of data sources. Population, M3, trade, and GDP and its deflator are from worksheets underlying Bordo and Jonung (1987), Obstfeld and Taylor (2000), and Rousseau and Wachtel (1998). Broadly defined money (M3) serves as our general measure of financial development, and the sum of exports and imports captures the extent of international trade and openness. We convert all variables to real per capita values using population and the GDP deflator and take natural logarithms prior to analysis.

Common in studies of the finance-growth relationship using even post-1960 data, M3 is the only broadly representative aggregate reflecting the amount of liquid liabilities created by the banking system for a sufficient number of countries from as early as 1850.4 Normalized by population, it reflects the intensity of banking services and is presumably related to their diffusion. In terms of the quantity identity, the dynamic relationship between broad money and output that we propose captures an adjustment process where increases in broad money lower velocity temporarily until output adjusts upward, with velocity achieving a point where financial depth (i.e., M3/GDP, the inverse of velocity) exceeds the starting point due to decreasing returns.

Specifically, we first consider a panel vector autoregression (VAR) to assess the size and direction of dynamic relationships between output, financial, development, and trade of the form:

[mathematical expression not reproducible] (1a)

[mathematical expression not reproducible] (1b)

[mathematical expression not reproducible] (1c)

where [y.sub.j,t] is GDP for country j at time t, [f.sub.j,t] is broad money, [r.sub.j,t] is the sum of exports and imports, k is the number of lags, [[eta].sub.j] is a country-specific fixed effect, [[phi].sub.t] are fixed effects for years, and [[epsilon].sub.j,t] is a random disturbance with a distribution that is approximately normal. (5)

We estimate the model using the least square dummy variable (LSDV) estimator and the system GMM estimator developed by Arellano and Bover (1995) and Blundell and Bond (1998). (6) After estimating the panel VAR, we compute F tests on the coefficient blocks to study the strength and direction of the dynamic relationships in the system.

Table 1 reports the coefficients and test statistics for the panel VARs with k = 2 selected via a series of nested likelihood ratio tests. The coefficients on broad money have a positive sum in the output Eq. (la), while those on trade have a negative sum. The F tests from the LSDV estimation indicate a leading role for both broad money and trade in GDP at the 10% level. There is no evidence of feedback from GDP to broad money or trade, while there is evidence of a positive and leading role for money in the trade Eq. (1c) at the 5% level. The GMM results are qualitatively similar except that trade is no longer significant in the output equation. Sargan tests do not reject the model's over-identifying restrictions, and there is no evidence of second-order autocorrelation in the residuals. Overall, the dynamic panels offer evidence consistent with a leading role for broad money in both output and trade; i.e., innovations in M3 move both positively and not the other way around. (7)

Figure 1 includes selected cumulative zero impulse responses from the LSDV panel for ten-year horizons with one standard error bands.8 Shocks to broad money affect both output and trade positively and persistently; the cumulative responses rise sharply for 2 years before gradually leveling off, and the lower standard error bands remain above zero for 3 years. The responses to unit shocks to trade, on the other hand, reveal a much more limited effect on both broad money and output, with the lower standard error band for the cumulative response of money never crossing zero and a negative response of output throughout the 10-year horizon. These are consistent with our interpretation of the panel Granger tests.

We now explore the underpinnings of the panel results by estimating VAR and vector error correction models (VECMs) separately for each of the 17 countries. The VAR specifications are similar to those in Eq. 1, but without the fixed effects and country subscripts. Sims et al. (1990) indicate that levels are appropriate for classical inference in tri-variate VARs when the individual variables have unit roots and the system has a single cointegrating vector. (9) We form the VECM by transforming the standard VAR as follows:

[mathematical expression not reproducible] (2a)

[mathematical expression not reproducible] (2b)

[mathematical expression not reproducible] (2c)

where the [[beta].sub.1], [[beta].sub.2], and [[beta].sub.3] are loadings that form a stationary linear combination of the data (the "error correction" term or ECT). A variable, say financial development, is said to be "weakly exogenous" or "forcing" for output in a long-run sense when the coefficient on the ECT (i.e., [[alpha].sub.2]) is not significant in the broad money Eq. (2b), but [[alpha].sub.1] is significant in the output Eq. (2a). The direction of the effect can be deduced from the signs of the [alpha] coefficients in each system and the [beta] loadings on the forcing variable. Equation-by-equation OLS is efficient for estimating since we do not impose restrictions on the coefficients and the equations in each system include the same predetermined variables on the right-hand side.

Table 2 summarizes the individual time series results. A plus sign (+) indicates that the sum of the coefficients for the directional effect specified in the column label is positive and statistically significant at the 10% level or less, while a minus sign (-) indicates a negative effect. The left panel presents these effects for Granger causality tests from levels VARs, while the right panel reports instances of weak exogeneity as described above. The remarkable feature of Table 2 is the consistency with which broad money acts as a forcing variable behind output (column 1) and trade (column 2), although of course not uniformly across all 17 countries. Broad money positively drives growth either in the Granger sense or via weak exogeneity (i.e., the long-run sense) in 11 countries, while broad money leads trade for 12 countries. We note that four of the countries (Argentina, Brazil, Portugal, and Spain) where broad money did not affect output had yet to begin a modern industrial transformation by 1929. Interestingly, there are seven countries where trade had a significant and negative effect on output (column 2). This is consistent with O'Rourke (2000) and Bordo and Rousseau (2012), who show that the standard positive correlation between trade and output across countries in the post-World War II period did not hold universally before 1929. For example, some countries in the sample, such as Canada, were protectionist while still growing quickly.

Discussion

Both the panel and individual time series results suggest that financial development matters positively and directly for trade and growth. Finance may affect growth through familiar channels such as easing external financing and improving the allocation of savings. Yet firms and industries with better access to financing arrangements can also scale up production to the point where they can become exporters (Beck 2002). This second effect is particularly important in the period we study when process and transportation improvements brought more and more diverse goods to world ports, where expanding trade credit and bills markets further encouraged trade. Of course finance at the country-level could itself have also promoted trade through the direct provision of resources for building ports and making other improvements in transportation. Once this finance-led push put the first wave of globalization into motion, trade disruptions after World War I could not shut down the mobilization of domestic resources that had begun.

At first, the negative and significant sign of trade on growth and the absence of mutually reinforcing effects between financial development and trade are surprising. While caution is necessary due to model uncertainty and noisy data for countries at these earlier stages of global trade and commerce, we may be capturing a temporary crowding out of domestic production along a transition path. This suggests that we should not fully dismiss the possibility of a positive role for trade in economic growth. Demetriades and Rousseau (2011) find in post-1960 data for 84 countries that link from trade to finance are strong in middle income countries, but not in lower and higher incomes ones. Given that our sample covers a mix of low and middle income countries by today's standards, it is perhaps not surprising that trade had yet to become a driver behind financial development.

Conclusion

We study interactions between broadly defined money, trade, and growth for 17 economies from 1850 to 1929 and find that broad money is a fundamental force behind both output and trade. Trade openness does not have direct positive effects on broad money or output. This evidence from the first globalization wave highlights the benefits that might follow from making financial development a priority in emerging countries seeking fuller integration with the new global economy.

References

Arellano, M., and O. Bover. 1995. Another look at the instrumental variable estimation of error-components models. Journal of Econometrics 68: 29-51.

Baltagi, B.H., P.O. Demetriades, and S.H. Law. 2009. Financial development and openness: Evidence from panel data. Journal of Development Economics 89: 285-296.

Beck, T. 2002. Financial development and international trade. Is there a link? Journal of International Economics 57: 107-131.

Blundell, R., and S. Bond. 1998. Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics 87: 115-143.

Bordo, M.D., B. Eichengreen, and D. Irwin. 1999. Is globalization today really different than globalization a hundred years ago? In Brookings trade policy forum, ed. S. Collins, and R. Lawrence, 1-50. Washington, DC: Brookings Institution.

Bordo, M.D., and L. Jonung. 1987. The long-run behavior of the velocity of circulation. New York: Cambridge University Press.

Bordo, M.D., and P.L. Rousseau. 2012. Historical evidence on the finance-trade- growth nexus. Journal of Banking & Finance 36: 1236-1243.

Demetriades, P.O., and P.L. Rousseau. 2011. Government, openness and finance: Past and present. Manchester School 79(s2): 98-115.

Dollar, D. 1992. Outward-oriented developing economics really do grow more rapidly. Economic Development and Cultural Change 40: 523-544.

Edwards. S. 1998. Openness, productivity and growth: What do we really know? Economic Journal 108: 383-398.

Johansen, S. 1991. Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica 59: 1551-1580.

King, R.G., and R. Levine. 1993. Finance and growth: Schumpeter might be right. Quarterly Journal of Economics 108: 717-737.

Levine, R., N. Loayza, and T. Beck. 2000. Financial intermediation and growth: Causality and causes. Journal of Monetary Economics 46: 31-77.

Lutkepohl, H. 1993. Introduction to multiple time series analysis, 2nd ed. New York: Springer.

Obstfeld, M., and A.M. Taylor. 2000. Global capital markets: Integration, crises, and growth. Japan-US Center Sanwa monographs on international financial markets. Cambridge: Cambridge University Press.

Obstfeld, M., and A.M. Taylor. 2001. Globalization and capital markets. In Globalization in historical perspective, ed. M.D. Bordo, A.M. Taylor, and J.G. Williamson, 121-187. Chicago: University of Chicago Press.

O'Rourke, K.H. 2000. Tariffs and growth in the late 19th century. Economic Journal 110: 456-483.

O'Rourke, K.H., and J.G. Williamson. 1999. Globalization and history: The evolution of a I9th century Atlantic economy. Cambridge, MA: MIT Press.

Phillips, P.C.B., and P. Perron. 1988. Testing for a unit root in time series regression. Biometrika 75: 335-346.

Rousseau, P.L., and P. Wachtel. 1998. Financial intermediation and economic performance: Historical evidence from five industrialized countries. Journal of Money, Credit and Banking 30: 657-678.

Sachs, J., and A. Warner. 1995. Economic reform and the process of global integration. Brookings Papers on Economic Activity 1: 1-118.

Schularick, M., and A.M. Taylor. 2012. Credit booms gone bust: Monetary policy, leverage cycles, and financial crises, 1870-2008. American Economic Review 102: 1029-1061.

Sims, C.A., J.H. Stock, and M.W. Watson. 1990. Inference in time series models with some unit roots. Econometrica 58: 113-144.

https://doi.org/10.1057/s41294-017-0048-y

Published online: 4 January 2018

(1) See, among many others. King and Levine (1993), Rousseau and Wachtel (1998), and Levine et al. (2000).

(2) See Dollar (1992), Sachs and Warner (1995) and Edwards (1998).

(3) The countries and starting years of data coverage are: Argentina 1884; Australia 1880; Brazil 1880; Canada 1870; Denmark 1850; Finland 1862; France 1851; Germany 1883; Italy 1872; Japan 1878; Netherlands 1850; Norway 1865; Portugal 1880; Spain 1875; Sweden 1870; United Kingdom 1870; and the United States 1851.

(4) Schularick and Taylor (2012) build an impressive dataset of bank loans that includes 15 of the countries we study starting as early as 1870, but the coverage is insufficient for our purposes. In particular, the loan sample is 18% smaller than ours for broad money, and coverage prior to 1900 is 35% smaller. Given our interest in time series analysis going well back into the nineteenth century, we proceed with broad money as our financial aggregate. At the same, we note that the loans and broad money series have correlations of more than 0.9 for all of the common countries in our samples measured across their overlapping observations. We thank Moritz Schularick for making an updated version of the loans data available to us.

(5) The model assumes that the errors are orthogonal to the fixed effects and lags of the system variables, have positive variance, and are uncorrected across cross-sectional units and time. We use country-fixed effects because they likely reflect country-specific omitted characteristics that are correlated with other explanatory variables.

(6) Given the disruptions to international trade after the start of World War I, we checked our results against those obtained when ending the panel in 1913, which reduced the number of observations by 23%. We found that broad money and trade continued to Granger cause output at the 5 and 10% levels, respectively, in the LSDV estimations and at the 5% level in the system GMM estimations. Broad money Granger causes trade at the 15% level, which is slightly weaker than the result obtained in the full sample.

(7) In any finite regression framework, the results of tests for statistical causality based on timing must be interpreted with caution given that they are predicated on inclusion of the full set of relevant information in the model. If this is violated, variables outside the system could exogenously move the variables of interest with different timing patterns, giving only the impression of one "causing" another. In this sense our results are only suggestive of the underlying economic forces that the data reveal.

(8) As Lutkepohl (1993, 43-18) describes, zero impulse responses assign the initial shock in a VAR system to the first variable in the investigator's ordering, with responses then computed in the standard manner. We order the variables in Fig. 1 according to their relative exogeneity as indicated by the panel Granger tests, with broad money first, trade second, and output last.

(9) For the 51 series that we consider, all but three (GDP for Denmark and M3 for Japan and Sweden) are unable to reject the null hypothesis of a unit root at the 5% level with the augmented Dicky-Fuller or Phillips and Perron (1988) test. Johansen (1991) tests indicate the presence of a single cointegrating vector for GDP, M3, and trade in all but two countries (Italy and Japan).

Alexandra D'Onofrio (1) * Peter L. Rousseau (2)

(1) Arcelli Center for Monetary and Financial Studies (CASMEF), Luiss Guido Carli University, Viale Romania, 32, 00197 Rome, Italy

(2) Department of Economics, Vanderbilt University, Box 1819 Station B, Nashville, TN 37235, USA

[mail] Peter L. Rousseau

Peter.L.Rousseau@Vanderbilt.Edu

Alexandra D'Onofrio

alexandra.donofrio@gmail.com

Caption: Fig. 1 Selected impulse responses from panel VAR. Note The graphs on the left report responses of real per capita trade and output to a one unit shock in real per capita broad money, while the graphs on the right show the responses of broad money and output to a one unit shock in trade. The ordering of variables as suggested by the block exclusions tests is broad money first, trade second, and output last. The solid lines are the mean impulse responses that result from 2500 random draws from the distribution of the VAR coefficients, and the dashed lines are one standard error bands Table 1 Panel LSDV and gmm estimates of VAR systems LSDV Variable equation GDP Broad Trade (1a) money (1b) (1c) [GDP.sub.-1] 0.924 *** 0.088 0.087 (0.045) (0.091) (0.161) [GDP.sub.-2] 0.003 -0.044 -0.101 (0.043) (0.085) (0.164) Broad [money.sub.-1] 0.052 ** 0.877 *** 0.172 ** (0.026) (0.074) (0.069) Broad [money.sub.-2] -0.040 0.018 -0.150 ** (0.026) (0.071) (0.070) [Trade.sub.-1] -0.014 0.014 0.660 *** (0.018) (0.029) (0.134) [Trade.sub.-2] -0.006 -0.005 0.195 * (0.015) (0.026) (0.113) Year effects Yes Yes Yes Country effects Yes Yes Yes Observations 985 985 984 [R.sup.2] 0.999 0.999 0.999 Sargan test AR(1) AR(2) P values F-GDP 0.000 0.293 0.810 F-M3 money 0.087 0.000 0.046 F-trade 0.089 0.801 0.000 System gmm Variable equation GDP Broad Trade (1a) money (1b) (1c) [GDP.sub.-1] 0 991 *** 0.130 0.195 * (0.072) (0.107) (0.112) [GDP.sub.-2] 0.002 -0.109 -0.177 (0.070) (0.099) (0.109) Broad [money.sub.-1] 0.053 0.945 *** 0.159 ** (0.033) (0.072) (0.069) Broad [money.sub.-2] -0.043 0.030 -0.149 ** (0.034) (0.070) (0.068) [Trade.sub.-1] -0.005 0.005 0.739 *** (0.016) (0.034) (0.107) [Trade.sub.-2] 0.002 -0.000 0.233 ** (0.017) (0.032) (0.101) Year effects Yes Yes Yes Country effects Observations 985 985 984 [R.sup.2] Sargan test 0.896 0.625 0.726 AR(1) 0.000 0.001 0.003 AR(2) 0.304 0.218 0.366 P values F-GDP 0.000 0.319 0.152 F-M3 money 0.005 0.000 0.095 F-trade 0.193 0.726 0.000 The table presents LSDV (left) and system GMM (right) estimates of panel VARs with two lags of the system variables and robust standard errors in parentheses. The columns correspond to Eqs. (1a)-(1c) in the text. P values of F tests for block exclusion of GDP, broad money and trade are reported in the lower part of the table along with P values of Sargen tests of the GMM model's over-identifying restrictions and tests for AR(1) and AR(2) errors. *, ** and *** denote statistical significance at the 10, 5, and 1% levels, respectively. Gaps in the data for France (1914-1920) and Germany (1914-1924) are treated as missing values Table 2 Summary of VAR and VECM results VAR Country (lag) M [right arrow] Y T [right arrow] Y Argentina (4) + Australia (2) + Brazil (4) - Canada (3) + - Denmark (2) + Finland (2) + - France (4) Germany (4) + - Italy (2) Japan (4) Netherlands (4) + - Norway (4) + - Portugal (4) - Spain (4) Sweden (2) + - UK (4) - USA (4) + Total (+, -) (9,2) (1,7) VAR Country (lag) M [right arrow] T T [right arrow] M Argentina (4) - + Australia (2) - Brazil (4) + - Canada (3) - Denmark (2) + + Finland (2) France (4) + Germany (4) + + Italy (2) + Japan (4) + Netherlands (4) + Norway (4) + - Portugal (4) - Spain (4) + - Sweden (2) + UK (4) USA (4) - - Total (+, -) (10, 3) (3,6) VAR Country (lag) Y [right arrow] M Y [right arrow] T Argentina (4) + Australia (2) - Brazil (4) Canada (3) - Denmark (2) - Finland (2) + France (4) + Germany (4) - Italy (2) + Japan (4) Netherlands (4) - Norway (4) Portugal (4) - Spain (4) - Sweden (2) + - UK (4) + + USA (4) + + Total (+, -) (4,2) (5,6) VECM- weak exogeneity Country (lag) M [right arrow] Y T [right arrow] Y Argentina (4) Australia (2) Brazil (4) + - Canada (3) Denmark (2) Finland (2) + - France (4) Germany (4) + - Italy (2) Japan (4) + - Netherlands (4) + - Norway (4) Portugal (4) - - Spain (4) Sweden (2) + - UK (4) USA (4) Total (+, -) (6,1) (0,7) VECM- weak exogeneity Country (lag) M [right arrow] T T [right arrow] M Argentina (4) + Australia (2) - Brazil (4) + Canada (3) + - Denmark (2) + + Finland (2) + France (4) + Germany (4) + Italy (2) - Japan (4) + Netherlands (4) + Norway (4) Portugal (4) - Spain (4) + - Sweden (2) + + UK (4) - USA (4) - - Total (+, -) (10,3) (3,5) VECM- weak exogeneity Country (lag) Y [right arrow] M Y [right arrow] T Argentina (4) + Australia (2) - Brazil (4) - Canada (3) - - Denmark (2) - + Finland (2) + France (4) + Germany (4) - Italy (2) + Japan (4) - Netherlands (4) - Norway (4) Portugal (4) - Spain (4) - - Sweden (2) + - UK (4) + USA (4) + + Total (+, -) (5,3) (4,9) The left-hand panel summarizes the statistically significant P values obtained in the levels VAR systems (Eqs. 1a-1c) by country. The lag lengths (k) of the VARs are reported in parentheses beside each country name; the VECMs (Eqs. 2a-2c) reported in the right-hand panel use (k - 1) lags in first differences. A plus sign (+) indicates that the directional link given by the column heading is positive and significant at the 10% level or less, while a minus sign (-) indicates a negative effect at the 10% level or less. The right-hand panel summarizes the statistically significant ECM terms from the VECM estimations. In the table, Y stands for GDP, M for broad money and T for trade openness