Michael Graff Thresholds in the Finance-Growth Nexus QMF Sydney, 16 December 2005

1. Michael GraffThresholds in the Finance-Growth NexusQMF Sydney, 16 December 2005 Dr Michael Graff School of Economics University of Queensland and KOF ETH Z�rich m.graff@uq.edu.au graff@kof.ethz.ch

2. Overview Introduction: The Finance-Growth Nexus Empirical Analysis Conclusions References M. Graff (2005): Is there an Optimum Level of Financial Activity? KOF Working Paper No. 106, ETH Z�rich, August 2005 [kof.ethz.ch/pdf/wp_106.pdf]. M. Graff and A. Karmann (2005): What Determines the Finance-Growth Nexus? Empirical Evidence for Threshold Models, forthcoming in: Journal of Economics.

4. The problem Empirical studies that include proxies for �financial activity� (F) as explanatory variables in cross-country regressions of growth rates of per capita income on its supposed determinants have repeatedly reported a positive partial correlation between different indicators of F and growth rates of per capita income Only a small fraction of this literature devotes attention to possible breaks, poverty traps, imbalance effects and structural shifts in the finance-growth nexus Linear approach may not be appropriate �Balanced� financial development that is contingent on a country's general level of development Financial activity can be both too low and too high in terms of efficiently contributing to economic growth and development

6. Research design Fit data (referring to a panel of 90 countries from 1960�2000) to a standard growth equation including the usual growth regressors fixed country and period effects our focal variable: a proxy for financial activity Perform a battery of sensitivity tests to check for robustness of the financial activity regressor Infer a �balanced� financial development path from the data Determine deviations from balanced path Order the sample by this measure of financial imbalance Perform exploratory threshold regressions to identify possible locations for structural breaks of the finance regressor in the growth equation Run bootstrap test for significance of these structural breaks

7. Growth model and reduced form �Augmented� Cobb/Douglas aggregate production function, relating GDP in country i at time t to the factors of production and overall efficiency parameter Yi,t = Ai,t K?i,t L�i,t H?i,t g(Y/L)i,t = g(A)i,t + ? g(K/L)i,t + ? g(H/L)i,t + (?+�+?�1) g(L)i,t , g(A)i,t = a0 + a1 ln(Y/L)i,t�1 + ?j aj Xj;i,t�1 g(A)i,t = a0 + a1 ln(Y/L)i,t�1 + a2 Ft�1 g(Y/L)i,t = �0 + �1 g(K/L)i,t + �2 g(H/L)i,t + �3 ln(Y/L)i,t�1 + �4 Fi,t�1 Panel data set: 90 countries and 8 five-year-growth periods (N�=�720) g(Y/L)i,t = �0 + � i+ �t + �1 g(K/L)i,t + �2 g(H/L)i,t + �3 ln(Y/L)i,t�1 + �4 Fi,t�1 + ?i,t Reduced form is model based, parsimonious and closely in line with what is referred to in the prevailing research, easing interpretation and comparison with other studies.

8. Sample and data 90 countries from the Penn World Table 6.1 (1960, 1965, ..., 2000) L: number of people aged 15�64 K: estimated from I by the perpetual inventory method, using a depreciation rate of 10%. Human capital (H/L): mean years of schooling, population 15�65 Financial activity (F): first principal component of three indicators: share of the financial sector in GDP (UN National Account Statistics, �finance, insurance and business services�) share of labour employed in the financial system (ILO Yearbook of Labour Statistics, ISIC-2 �major division 8�: financial institutions, insurance, real estate and business services) M2/GDP This proxy for F to capture the share of resources a society devotes to run its financial system Advantages compared with traditional F-indicators: Less ambiguous, less sensitive to changes in regulations etc

10. Is there an optimal financial activity level ? Optimum development path F*t , contingent on other dimensions of social and economic development Dimensions of this contingency repeatedly stated in literature: real development Y/L highly qualified human capital TER F*t = f�[(Y/L)t�?, TERt�?)], ?, ? ? 0 Correlation Y/L - TER ? 1 F*i,t = ?0 + ?1 ln(Y/L)i,t + ?2 (Y/L)i,t + ?3 (Y/L)2i,t + ?i,t R��=�0.75, all regression coefficients significant, confirming non-linearity ?i,t�=�Fi,t���F*i,t : measure of financial underdevelopment (if negative), or excess financial activity (if positive)

11. Financial imbalance threshold regression 719 repeated regression with �4 set free across two subgroups group that scores low on?i,t and a corresponding high scoring group D(?)n = 0 if observation belongs to low-scoring subgroup, 1 otherwise n = rank position of the split variable ?i,t in ascending order. g(Y/L)i,t = �0 + �i + �t + �1 g(K/L)i,t + �2 g(H/L)i,t + �3 ln(Y/L)i,t�1 +�4 Fi,t�1+ �5 D ? Fi,t�1 + ?i,t �5 : point estimate of the difference of �4 between group Dn�=�0,1 t-statistics identifies location of split that is optimal in improving fit However, invalidates t-test for structural break, since sequential search for the optimum split negates null of �no structural beak� Hansen (1999, 2000): bootstrapping At a candidate location suggested by t-statistics: non-parametric test 1000 bootstrapped samples (820 draws with replacement) 1000 point estimates �5 for difference of coefficients for F between the sub-samples ? 5% and 95% percentiles to growth.

12. Threshold variable: random, point estimate and t-statistics for �5, traditional 95% confidence interval

13. Threshold variable: ?i,t(Y/L), full sample point estimate and t-statistics for �5, traditional 95% confidence interval

14. Threshold variable: ?i,t(Y/L), full sample point estimate bootstrap (n=1000) for �5 at observation No. 144

15. Threshold variable: ?i,t(Y/L), sample = observations 145�820, point estimate and t-statistics for �5, trad. 95% confid. interval

16. Threshold variable: ?i,t(Y/L); sample = observations 145�820, point estimate bootstrap (n=1000) for �5 at observation No. 465

17. Conclusions Finance-growth nexus is not linear Countries gain less from a given level of financial activity if the latter fails to keep up with or exceeds what would follow from an expansion path that is �normal�, given its overall state of development. Empirical support for notion of �balanced� financial development with a development specific optimum level of financial activity two thresholds relative financial underdevelopment wasteful excess levels of financial activity.