Liability dollarization, exchange market pressure and fear of floating: empirical evidence for Turkey

Abstract

The objective of this article is to examine the relationship between liability dollarization and the Exchange Market Pressure (EMP) in Turkey within an Autoregressive Distributed Lag (ARDL) and Granger causality framework using monthly data from 1991:12 to 2006:08. The findings suggest that there exists a long-term equilibrium relationship between EMP and liability dollarization, where liability dollarization Granger causes EMP both in the short- and long-run, with no evidence of reverse causality. This suggests that the predominance of foreign currency liabilities in the banks’ balance sheets in Turkey induces a selling pressure in the exchange market as well as a fear of floating.

Notes

¹ See Girton and Roper (1977, p. 537).

² The exchange rate is represented in terms of Turkish liras per US dollars so that an increase in the exchange rate represents the depreciation of the Turkish lira. The nominal exchange rate has been used following earlier studies. The possibility of real effective exchange rates has also been considered but since the correlation coefficient between nominal and real exchange rates during the sample period is 0.887, the choice of which series is to be used would clearly not affect the robustness of the empirical results.

³ See Balkan and Yeldan (1998) and Demir (2004).

⁴ The 3-month deposit rate has been used as a proxy for short-term interest rates for Turkey as 3-month T-Bill rates are not available.

⁵ Eichengreen et al. (1995) compare the domestic reserves with those of a reference country. However, why reserve changes in another country would be of interest in the case of Turkey is questionable. Clearly, this approach would render the EMP index subject to any idiosyncratic fluctuations in the reference country's international reserves. Therefore any reference country's reserves are not included in the present analysis.

⁶ The objective of principal component analysis is to reduce the dimensionality (number of variables) of the dataset but retain most of the original variability in the data. In econometrics, it has been used for reduction of large data collections into more manageable form, especially to deal with problems of multicollinearity and shortage of degrees of freedom (Klein and Ozmucur, 2003).

⁷ The Kaiser criterion suggests that only the factors with eigenvalues greater than 1.0 are retained (Kaiser, 1960).

⁸ See Weymark (1995, p. 279).

⁹ See Weymark (1995, 1997a, 1997b, 1998).

¹⁰ The summary of Weymark's (1995) model by Stavarek (2007) is used.

¹¹ Only if α ₁ = α ₂ = 1.

¹² The superscripts d and s represent demand and supply, respectively.

¹³  is held constant when exchange market pressure is imputed (Weymark, 1995, p. 279).

¹⁴ GDP is interpolated from quarterly series.

¹⁵ The Hodrick–Prescott filter is a smoothing method that is used to obtain a smooth estimate of the long-term trend component of a series (Hodrick and Prescott, 1997). Eviews 5.0 software uses 14 400 as the default smoothing parameter.

¹⁶ Weymark (1995) does not provide a detailed account of her estimation of the related parameters.

¹⁷ See Stavarek (2007).

¹⁸ Endogenous variables on the right-hand side of the equation are likely to correlate with the disturbance term. Thus, using the Ordinary Least Square (OLS) method would lead to biased estimates (Stavarek, 2007).

¹⁹ The variables used in the first stage of 2SLS to create the new variables are called instrumental variables, which are uncorrelated with the disturbance term and replace the problematic causal variables. This is accomplished using OLS regression with the problematic causal variable as the dependent and instrumental variables as the independents.

²⁰ See Stavarek (2007).

²¹ This result is not similar to the elasticity estimated by Weymark (1995) (as −3.089) for the Canadian economy. This can be attributed to the differences between the Canadian and Turkish economies. Indeed, the result is more consistent with the findings of Stavarek (2007) – 1.838, 0.906 and 0.899, respectively for Czech Republic, Hungary and Slovakia – which have more similar economies to that of Turkey.

²² Perron (1989) argues that the power to reject unit root decreases when the stationary alternative is true and a structural break is ignored.

²³ See Enders (2004, p. 207).

²⁴ Ben-David et al. (2003) cautions that ‘just as failure to allow one break can cause nonrejection of the unit root null by the ADF test, failure to allow for two breaks, if they exist, can cause nonrejection of the unit root null by the tests which only incorporate one break’ (Ben-David et al., 2003, p. 304).

²⁵ Lumsdaine and Papell (1997) extended the Zivot and Andrews (1992) model to accommodate two structural breaks. However, this test was criticized for the absence of the breaks under the null hypothesis of unit root as this could result in a tendency for these tests to suggest evidence of stationarity with breaks (Glynn et al., 2007).

²⁶ Perron–Vogelsang tests are based on the minimal value of t-statistics on the sum of the autoregressive coefficients over all possible breakpoints in the appropriate autoregression (Perron and Vogelsang, 1992, p. 303).

²⁷ See, Baum (2004).

²⁸ Choosing the dates exogenously introduces arbitrariness in the analysis. Even if sudden changes were observed in the data, the exact breakpoint affecting the parameters of the model may not obvious. Hansen and Johansen (1992) also suggest a parameter constancy test but they require the variables to be I(1). Also, they do not incorporate the short-run dynamics of a model into testing unlike CUSUM and CUSUMSQ tests.

²⁹ See Pesaran and Pesaran (1997, p. 117).