An Empirical Analysis of the Risk-Taking Channel of Monetary Policy in Turkey

Abstract

This article investigates the bank-specific characteristics of risk-taking behavior of the Turkish banking sector as well as the existence of risk-taking channel of monetary policy in Turkey. Using bank-level quarterly data over the period 2002–2012 a dynamic panel model is estimated. We find evidence that low short-term interest rates reduce the risk of outstanding loans; however short-term interest rates below a theoretical benchmark increase risk-taking of banks. This result holds for macroeconomic controls and external factors as well. Furthermore, in terms of bank-specific characteristics, our analysis suggests that large, liquid, and well-capitalized banks are less prone to risk-taking.

KEY WORDS:

• monetary policy
• transmission mechanisms
• risk-taking channel
• Turkey
• panel data

Acknowledgments

This article is based on Ekin Ayşe Özşuca’s PhD thesis prepared under the supervision of Elif Akbostancı at the Department of Economics, METU. The authors thank two anonymous referees and the editor, Ali M. Kutan, for helpful comments and insights.

Notes

1. The elements of the theory of risk-taking channel can be traced in the theoretical propositions of some previous studies such as; Keeley (1990), Gibson (1997), Allen and Gale (2000), Dell’Ariccia and Marquez (2006), Rajan (2006), and Matsuyama (2007). Although some of the mechanisms have been discussed previously, the term ‟risk-taking channel” of monetary policy first appeared in a paper written by Borio and Zhu (2008) in which they point to the potential relationship between low interest rates and increased bank risk-taking.

2. For further details on the mechanisms that monetary policy may influence bank risk-taking see, for instance, Campbell and Cochrane (1999), Dell’Ariccia and Marquez (2006), Rajan (2006), Farhi and Tirole (2009), Diamond and Rajan (2009), Adrian and Shin (2010).

3. The theoretical literature is still being developed and is rather limited for the time being. There are only a handful studies that present formal models where several mechanisms of the risk-taking channel act together. See, for instance, Dubecq et al. (2009), Dell’Ariccia et al. (2010), Agur and Demertzis (2010), Valencia (2011), Cociaba et al. (2011), Gonzalez-Aguado and Suarez (2011).

4. For studies that provide empirical evidence using micro level data see, for instance, Jimenez et al. (2009), Gambacorta (2009), Altunbaş et al. (2010), Brissimis and Delis (2010), Michalak (2010), Delis and Kouretas (2011), Delis et al. (2011).

5. Other studies on emerging markets are Iannidou et al. (2009) for Bolivia, Tabak et al. (2010) for Brazil, and Lopez et al. (2010) and Lopez et al. (2012) for Colombia.

6. See Özatay and Sak (2003), Özkan (2005), Akyürek (2006), among others for a detailed discussion of the crises.

7. We also experiment to calculate $\sigma (RO{A}_i)$ by using different number of quarters, but results are found to be very similar.

8. Interbank interest rates may be endogenous to general macroeconomic conditions. Moreover, causality may run in both directions between interest rates and bank-risk taking, if monetary authority takes interest rate decisions by considering credit market conditions. However, this is does not seem to hold exactly for Turkey, since the Central Bank of the Republic of Turkey (CBRT) did not systematically take into account banking sector conditions on its policy rate decisions. Furthermore, as stated by Aydın and Igan (2012) endogeneity of the policy is less of a problem as policies have been designed to act anchors following the 2000–2001 crisis. Nevertheless, we still choose to adopt a benchmark measure as an exogenous measure of monetary policy given the primary concern of our analysis.

9. First presented in Taylor (1993), the Taylor rule suggests a simple way to formulate monetary policy. It stipulates how the central bank should change its policy rate as output and inflation deviated from certain levels. Algebraically, it could be expressed as: $i_t$= $r^{\ast }+{\beta }_{\pi }$(${\pi }_t-{\pi }^{\ast })+{\beta }_y$($y_t-y_t^{\ast }$), where $i_t$ is the policy interest rate, $r^{\ast }$ is equilibrium real interest rate, ${\pi }_t$ is the inflation rate, ${\pi }^{\ast }$ is the target inflation rate and ($y_t-y_t^{\ast }$) is the output gap (the deviation of the actual GDP from its long-term potential level). Taylor (1993, 2001) proposed setting ${\beta }_{\pi }={\beta }_y=\mathrm{0.5.}$

10. See, among others, Gourinchas and Obstfeld (2012), Mendoza and Terrones (2012), Elekdag and Wu (2013).

11. Table 1 summarizes the data before corrupt observations are controlled.

12. Due to space constraints we do not present the specifications based on EDF. The specifications with EDF as our dependent variable are provided in Özşuca and Akbostancı (2012) available at: http://www.erc.metu.edu.tr/menu/series12/1208.pdf.

13. In what follows, we will modify the baseline model with the interaction effects. As stated in Delis and Kouretas (2011) “A problem with the inclusion of interaction effects is the severe multicollinearity between the multiplicative term and its constituents.” Hence, we deal with this problem by normalizing the bank-specific variables.

14. All empirical analyses in this study are done with STATA version 10.

15. Another benefit of the Blundell-Bond estimator is that it does not breakdown in the presence of unit roots as well. For proof; see Binder et al. (2003); Delis and Kouretas (2011).

16. However, this result should be evaluated cautiously since the related literature regarding the impact of the banking market structure on bank fragility posits that structural measures of competition like concentration ratios and nonstructural measures of competition, calculated from firm level data are different proxies and accordingly, measures different aspects of competition in the market. Therefore, results of the analysis might be sensitive to the market structure variable employed. However, as our primary concern is not on the bank market concentration–financial fragility nexus, this point is not critical from the standpoint of our analysis.

17. Among bank-specific characteristics, capitalization is not included as an explanatory variable in the regression where z-index is employed as dependent variable; since the ratio of equity to total assets is used to compute z-index as well.

18. For detailed results of the models using standard deviation of return on assets as dependent variable, see Özşuca and Akbostancı (2012) available at: http://www.erc.metu.edu.tr/menu/series12/1208.pdf.

19. In the models that we use EDF as our dependent variable, we have employed change in the industrial production index instead of the growth rate of GDP due to sample restrictions.

20. In the construction of the Taylor rule gap coefficients of 0.75 and 0.25 are used for inflation gap and output gap respectively, and real interest rate is taken as 10 percent. However, a proper estimation of the Taylor Rule would require different modeling and estimation techniques. For alternative techniques in the case of the Turkish case see Yazgan and Yılmazkuday (2007), Yılmazkuday (2007), Us (2007), and Akyurek et al. (2011).

21. For detailed estimation results with the Taylor rule gap see Özşuca and Akbostancı (2012) available at: http://www.erc.metu.edu.tr/menu/series12/1208.pdf.