Rethinking interest rate volatility as a macroprudential policy tool

Abstract

Along with most other central banks, Turkey's central bank has implemented unconventional policies since the 2007/2008 financial crisis. Financial stability has been one of the targets of these macroprudential policies. However, since Turkey is working toward this goal without increasing its inflation rate, tracking only short-term interest rates to measure this policy's effectiveness would be inefficient. In this paper, we provide empirical evidence from Turkey that interbank interest rate volatility can be an additional tool for monetary policy makers to help achieve the goal of financial stability. Impulse responses generated from the Vector Autoregressive models indicate that interest rate volatility increases interest rates, depreciates domestic currency and decreases credit growth and output. Its statistically insignificant effect on prices is open to interpretation.

Keywords:

• economics
• monetary policy
• interest rate volatility
• macroprudential policies

JEL Codes:

• E58
• E52
• E37

Acknowledgements

The authors would like to thank Rana Nelson and two anonymous referees for their valuable comments on the paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data for this article can be accessed here. http://dx.doi.org/10.1080/17938120.2016.1150009

Notes

1. See IMF (2012) for such macroprudential polices.

2. Central banks may also choose to decrease rather than increase interest rate volatility to mitigate obscured monetary policy signals, to promote a rapid and predictable monetary transmission mechanism and to help market participants better manage their economic and financial risks (see e.g. Gray and Talbot, 2006; Van’t dack, 1999). Here, we intend to document the effect of introducing interest rate volatility into the central bank policy arsenal. Interest rate volatility is a tool like any other policy tool that central banks have, with some promotional and some adverse effects on economic variables. We thank the anonymous referee for pointing the incentive that central banks may have to decrease interest rate volatility.

3. The ROM is given to banks to keep foreign currency and/or gold up to the officially announced level of domestic-currency-denominated reserve requirements. Banks are allowed (though not obliged) to keep 60% of their required Turkish Lira (TL) reserves as foreign exchange (in Euros and US Dollars (USD)) and an additional 30% as gold. As of May 2013, the first 35% of the allowed foreign exchange partition of the required domestic currency reserves is accepted by being multiplied by 1.4. The coefficients for each subsequent 5% increase are 1.7, 2.1, 2.4, 2.6 and 2.7, respectively. While the first 15% of the gold allowance is accepted as 1–1.4, the following three subsequent 5% increases are accepted with coefficients of 1.5, 2 and 2.5, respectively.

4. Developments in the current month also affect interest rate volatility. We address this issue by how we calculate interest rate volatility, taking the standard deviation of daily interest rates from the previous period for the current period. Note that this variable enters the specification as the first variable. Alternatively, we could take the standard deviation of daily interest rates within the same period and insert this variable into the VAR specification as the last variable in the Cholesky ordering. Thus, these two specifications employ the same identification assumptions. To obtain the identical impulse responses as in the former specification, it is necessary in the latter specification that all variables have two lags but that the volatility variable has three lags in the VAR specification. When we compare the impulse responses of the two specifications that have two lags, we see that our specification produces impulse responses with slightly lower confidence intervals.

5. If all series are I(0) or some series are I(0) and some are I(1) but the system is cointegrated, then VAR in its level is appropriate (see Lutkepohl & Reimers, 1992; Sims, Stock, & Watson, 1990). Even if we get mixed results for the unit root test across variables, we fail to reject the no-cointegrating vector. Thus we perform the analyses in levels. The results of these tests are not reported here to save space but are available from the authors upon request.

6. There are other statistical methods to calculate volatility, such as ARCH/GARCH-type models, Kalman filters, etc. Each of these captures a different component of uncertainty (see e.g. Berument & Yucel, 2005; Bomberger, 1996). However, using daily data allows us to capture the observed volatility rather than the perception of it (ARCH/GARCH-type models) or specification instability (Kalman filters).

7. The Euro was introduced in 1999; therefore for the period prior to 1999, we use the official convention between the Euro and the Deutsche Mark (DM) and use the basket as 0.5 USD + 0.974027 DM to calculate the exchange rate.

8. http://evds.tcmb.gov.tr/cbt.html.

9. The CBRT has been explicitly implementing interest rate volatility as a monetary policy tool only since 2007/2008. However, depending on capital in- and out-flows, de facto interest rate volatility has existed in Turkish money markets since the early 1990s. Thus, we can still observe and assess the effects of volatility on economic performance in Turkey even though it was not a CBRT policy choice before 2008. For this reason, we extended our data range back to the early 1990s.

10. Extending the time horizon up to 24, 36 and 48 periods does not change the basic conclusion of the paper. Thus we end the impulse responses at the eighteenth period.

11. Due to data availability for GDP the sample ends in September 2013.

12. The Turkish Statistical Institute (TurkStat) discounted calculating the WPI after 2005; therefore, for the post-2005 period we replace and update the series with the Producer Price Index (PPI).

13. Due to a low degree of freedom, the lag order is set to one.

14. There might be other specifications. One of them is that the relationship could be non-linear. As an exercise we tried a Threshold VAR (TVAR) specification of our benchmark model (not reported here to save space) and it seems that the basic results of the paper are robust.