Reassessment of the finance-growth nexus in the presence of structural breaks

Abstract

This article investigates empirically the causality between financial development and economic growth for 11 sample countries in a structural break framework via minimum Lagrange multiplier unit-root test and Hsiao's causality test. We find most of the series can be more accurately characterized as a segmented trend stationary process around structural breaks as opposed to a stochastic unit-root process, which has important implications for policy-makers to formulate long-term development strategy and short-run stabilization policies as well as for academics to carry out cointegration analysis. Besides, we find different causality patterns in sample countries, which depart from those found in the existing literature, highlighting that it might be unreliable to apply only conventional unit-root test and inappropriate to take the first difference of the series to achieve stationarity, all of which might result in model misspecification. Our analysis also provides some evidence on the source of the mixed results on finance-growth nexus as they may differ due to the length of data span and the different econometric procedure.

‘One of the most important problems in the field of finance … is the effect that financial structure and development have on economic growth’.

Goldsmith (1969, p. 390)

Acknowledgements

We would like to thank the Editor Mark Taylor and an anonymous referee of this journal for constructive criticisms and detailed suggestions. We are also grateful to Li-qun Zhou, Jun-lu Ma, Yoshinori Shimizu, Kazumi Asako and Satoru Kanoh for their helpful comments and seminar participants at Hitotsubashi University and Nankai University. Part of this work was completed while Liang was visiting the Graduate School of Commerce and Management at Hitotsubashi University as a JSPS Post-graduate Fellow. He thanks the School and JSPS for their hospitality. Teng acknowledges financial support from Northeast Normal University, through the key projects of social sciences and philosophy–Study of internal and external balance under an open economy and study of China's finance-grwoth nexus. Liang also acknowledges financial support from a research grant (06JA790057). Any errors or omission are solely our own.

Notes

¹ These limitations include omitted variable bias, sample selection bias and inappropriate weighting of countries (Demetriades and Hussein, 1996), nonstationary property of the data (Christopoulos and Tsionas, 2004), unrealistic assumptions that both institutional features that determine economic growth across countries and marginal response of economic growth to each measure of financial development are invariant (Odedokun, 1996), violation of basic principle behind the averaging and pooling of cross-nation data due to the lack of balanced growth paths across countries (Quah, 1993), and heterogeneity of slope coefficients across countries (Pesaran and Smith, 1995). Moreover, correlation reveals nothing about causality.

² Chang (2002) finds no causal relationship between financial development and economic growth for Mainland China over the period 1987Q1 to 1999Q4.

³ This is also being stressed in Kapetanios (2005, p. 123).

⁴ This problem has also been noted previously by Nunes et al. (1997) and Vogelsang and Perron (1998), which might lead to spurious rejections and cause size distortions.

⁵ Since the minimum LM unit-root test assumes break(s) under the unit root null, this approach has gain popularity in unit-root test literature (e.g. Abras et al., 2004, Strazicich et al., 2004; Narayan, (2007).

⁶ Ng and Perron (1995) and Perron (1997) suggest that the general-to-specific procedure produces test statistics with better properties of stable size and higher power than that of information-based methods.

⁷ We thank Mark Strazicich for helpful suggestions on the model specification of de-trending segmented trend stationary series.

⁸ Responding to concerns about arbitrary lag selections and symmetric lags in Granger causality tests, Hsiao (1981) proposes estimating VARs in which the lag length on each variable in each equation can differ. Hsiao's approach is specifically designed to avoid the imposition of false or spurious restrictions on the model (e.g. Cheng and Lai, 1997; Hsiao, 1981; Bajo-Rubio and Montávez-Garcés, 2002).

⁹ All sample countries except China are also examined by Luintel and Khan (1999). Besides, four sample countries, including Korea, Philippines, Sri Lanka and Thailand are also investigated by Al-Yousif (2002).

¹⁰ Real per capita GDP is superior to total real GDP, because some of the errors inherent in the estimation of the level of GDP and population tend to be offsetting (Heston, 1994).

¹¹ Due to data availability, we use credit to all sectors by banking institutions divided by GDP to proxy financial development for China.

¹² Lumsdaine and Papell (1997) point out that the unit-root test results are very sensitive to the presumed number of structural changes. Hence, this incurs a basic inquiry that we have no reasonable foundation to think about that the indicator series of economic growth and financial development in the sample countries only have one or two permanent breaks. However, considering the focus of this article is not on the applicability of unit-root tests with more breaks as well as the time span in the dataset, we believe that the minimum LM unit-root test is a reasonable model specification, which can be justified as one of the most accurate way of capturing the dynamic behaviour of the series as well as the relationships between them.

¹³ The indicator of economic growth in Costa Rica and the traditional indicator of financial development (DLR) in Korea are tested to be nonstationary. Thus, it becomes difficult, if not impossible, for us to analyse a system with series integrated of both order zero and one using the present econometric method. Although this problem can be dealt with in an autoregressive distributed lag framework suggested by Pesaran et al. (2001), which is, however, not our main interest.

¹⁴ We admit that this is only a coarse comparison, since Al-Yousif uses different proxy measures of financial development, and Luintel and Khan analyse the finance-growth nexus in a multivariate VAR framework.

¹⁵ Since the data span utilized by Luintel and Khan (1999) does not cover the Southeastern financial crisis, the break points during this period cannot be identified.