The effects of foreign capital inflows on exchange market pressure in Turkey

(1)

The Effects of Foreign Capital Inflows on

Exchange Market Pressure in Turkey

Faculty of Economics and Business

Amsterdam School of Economics

MSc. Economics Master Thesis

International Economics and Globalization Track

Berk Bulut

July 15, 2018

Email: berk.bulut@student.uva.nl Student Number: 11594721

(2)

Statement of Originality

This document is written by Berk Bulut who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

If we compare Turkish Lira’s (TL) value at the end of June 2018 to what it was exactly 5 years ago at the and of June 2013, it can be seen in Figure 3 that it has depreciated by around 137% vis-à vis U.S. Dollar (USD) over the 5 years. In addition to the nominal value loss of TL, “weighted average funding cost” (WAFC) which will be our short-term interest rate proxy for Turkey increased by 1266 Basis Point (BPS) and foreign reserve of Central Bank of the Republic of Turkey (CBRT) decreased by 30 billion USD in the same 5 year period. The stress in the Turkish foreign exchange (FX) market possibly indicates a structural problem of the Turkish economy if we take into account of its rapidly increasing FX debt level and dependency on foreign capital.

According to the CBRT Inflation and Price Stability Booklet (2013), since 2001 the primary objective of the CBRT is to achieve and maintain price stability. Thus, one of the concern of CBRT should be the exchange rate pass through (ERPT) degree (i.e. elasticity of domestic prices with respect to the changes in the nominal exchange rate). Although EPRT level has decreased after adopting inflation targeting (IT) regime first loosely in 2002 and then strictly in 2006, it is structurally at significant level for Turkish economy and ought not to be overlooked for implementation of a successful monetary policy (Yazgan&Zer-Toker, 2010). In addition to that one should take into account that in the years between 2002 and 2012 TL had not depreciated considerably vis-à-vis USD (Figure 3). In the view of this these facts, Arslaner&Karaman&Arslaner&Kal indicate that lower short-term ERPT degree argument belonged to the period during which the TL was appreciating and ERPT has been increasing to significant levels since then (2014). Therefore, it seems reasonable for CBRT to avert significant and drastic changes in the currency rate by intervening through different channels.

As announced in 2017, CBRT has been working on “Systemic Risk Data Tracking Model” to monitor cash flows or maturity and currency mismatches which will be used to manage risk. With the materialization of the model it is aimed to increase the use of hedging mechanism and decrease the foreign currency exposure of those who have serious currency mismatches in their balance sheets (Gençay, 2017). Managing these mentioned risks would certainly have a positive impact on financial stability and price stability which is the mail objective of CBRT. Thus, it would not be wrong to state that exchange rate is very crucial for Turkish economy.

(5)

It is known that CBRT uses its policy instruments to manage exchange rate from time to time, therefore nominal change in the value of TL does not fully reflect the pressure in the FX market. EMP Index has been used widely in order to capture the required change in the nominal value of the currency to cancel out the excess supply or demand in the FX market if policy makers would refrain from actions to offset. Although it was first introduced by Girton&Roper (1977) to measure required intervention to offset disequilibrium in the domestic source of the monetary base to achieve an exchange rate target, the idea was developed to its modern concept by other contributions (Eichengreen&Rose&Wyplosz (1995a, 1995b), Weymark (1995), Klaassen&Jager (2011)). As leaving the reasons to be provided in the Section 2, throughout the paper Klaassen&Jager (2011) EMP measure will be used to quantify the pressure in the FX market from the early moments of 2000-2001 crisis in Turkey till today, specifically between September 2000 and June 2018.

The aim of this paper is to analyze EMP level in Turkey by using Klaassen &Jager (2011) EMP measure due to its improvements over the previous research and compare the recent depreciation of TL to crisis times of early 2000s and investigate whether the dependency of Turkish economy on foreign capital has an effect on EMP in the short run (SR) and in the long run (LR) by using autoregressive distributed lag (ARDL) bounds testing approach to cointegration developed by Pesaran&Shin (1998) and Pesaran&Shin&Smith (2001).

This section will be followed by Section 2 giving more information about Turkish economy, EMP and ARDL bounds testing approach in the literature and their applications to Turkey. Section 3 provides insights on how EMP measure will be built for Turkey and related data, and in Section 4 ARDL Model will be constructed and related data will be presented. Section 5 and Section 6 present empirical applications and results of EMP measure in Turkey and ARDL bounds testing approach respectively, and finally Section 7 concludes.

(6)

2. Literature Review

First of all, in Section 2.1. the situation and evolution of Turkish economy after 2000-2001 crisis will be detailed. Section 2.2. provides further information on approaches to EMP and EMP measures in the literature and Section 2.3 addresses the previous studies on EMP in Turkey. Section 2.4. and Section 2.5. include brief review of literature on ARDL Bounds Testing Approach and its applications to Turkey respectively.

2.1. Turkish Economy After 2000-2001 Crisis

Turkish economy experienced a quick rebound after the 2000-2001 financial crisis resulting the collapse of the pegged exchange rate regime. The crisis had fiscal policy misalignments and its distorted relationship with the already fragile banking sector in the origin (Ozatay&Sak, 2002). Thanks to macroeconomic and structural adjustment programs and a new set of economic policies, both fiscal consolidation and transformation to rule-based fiscal, financial and monetary institutions can be seen the main reason behind the high growth rate of Turkish economy after the 2000-2001 crisis. Following the real GDP decline by 5.96 percent in 2001, the real growth rate increased to 7,14 on average for the following 6 years between 2002-2007 (Figure 1). In addition to that, after implementation of IT regime the 35 years of double-digit inflation rate period came to an end in 2007 (World Bank Open Data, Inflation, consumer prices (annual %), n.d.).

Increased global liquidity during 2000s coupled with perceptions of preferable macroeconomic conditions in emerging markets (EM) and in Turkey, has resulted significant capital inflows to Turkey due to relatively high interest rates and stability of TL (Karimova&Caliskan&Karimov, 2015). As Figure 1 shows capital flows were primarily in the form of volatile Foreign Portfolio Investments (FPI), only with a limited share of Foreign Direct Investment (FDI) in overall financial account (Onaran, 2007). Onaran (2007) also states that Turkish economy has been dependent on the foreign capital inflows to fuel domestic investments and boost the growth rate due to low ratio and short maturity of savings. It can be seen from Figure 1 that especially after 2003 there seems to be a positive correlation between FPI inflows and GDP growth rate. Furthermore, EM crises showed that capital inflows to EMs are often pro-cyclical, hence, increasing when the economies are in a good state and decreasing in times of slowing down (Kaminsky&Reinhart&Vegh, 2005). This feature of capital flows further increases the imbalances and feeds the risk financial crisis. Additionaly, Figure 2 shows

(7)

that the average current account deficit (CAD) of 0.59% of GDP between 1986 and 2001 for 16 years has increased to 4.57% of GDP on average between 2002 and 2017 for again 16 years. This indicates that long lasting CAD has grown into a serious structural problem of Turkish economy which also makes it dependent on foreign capital inflows to finance CAD. 33% appreciation of TL against USD between 2003 and 2008 (Figure 3) was one of the reasons of the rise in import demand and consequently expansion of CAD led by hot money financing (Yeldan&Ünüvar, 2016). According to Vural&Zortuk, mainly due to Turkey’s large CAD and heavy reliance on capital inflows to finance it, Turkish economy is vulnerable to sudden shifts in investor sentiment (2011). Therefore, volatile FPI could also lead to volatility in exchange rate which can increase the risk of financial crisis (Erdal&Pinar, 2015).

Figure 1 – FDI, net (BoP, Current US$), FPI, net (BoP, Current US$), Real GDP growth (annual %)

Due to increasing risk of financial crisis with significant increase in amount of short-term capital inflows especially right after 2007-2008 global financial crisis (GFC), “soft landing” program implementing alternative macroprudential policy measures was introduced in 2011 (Bakir&Ertan, 2018). These macroprudential policies in Turkey (e.g. imposing some regulations on FX borrowing, bank reserve requirements and restrictions on private consumption credits) have brought about a soft landing and improved sustained growth

(8)

prospects by lowering the sensitivity of economic activity to capital flow volatility considerably at the expense of higher GDP growth rates for the following years (Kenc, 2016).

Figure 2 – Current Account Balance of Turkey (% of GDP)

Although, Turkey now has floating exchange rate regime the negative gap between firms’ dollarized assets and liabilities has been increasing considerably (Karimova et al., 2015). If corporate debts are denominated in dollars while the value of corporate assets depends on local currency (i.e. liability dollarization), sharp and unexpected upward currency movements would have a deteriorating effect on financial and external stability which gives rise to “fear of floating” (Calvo&Reinhart, 2002; Chang&Valesco, 2006). Total FX-denominated debt reached 69.5% of GDP in 2017 due to long-running structural imbalance between domestic savings and investment in Turkey increasing stress as TL continues to depreciate (Johnson, 2018, May 11). Chui&Kuruc&Turner (2016) stresses the significance of currency mismatches of private sector but especially non-financial sector due to its higher vulnerability (i.e. less naturally hedged comparing financial sector) in the EM which have risen since 2010. Considering the rise of net open FX position of non-financial sector in proportion to GDP in Turkey from 3 to 27 very quickly and immature currency futures and options market in Turkey causing most of the market participants’ currency risk to stay unhedged, Turkish economy can not be considered as immune to sudden changes in exchange rate (Gençay, 2017). One of the reason to that is

CBRT’s tendency to support real appreciations and tendency to intervene real depreciations with FX sales and increasing domestic interest rates (i.e implicit asymmetric exchange rate peg policy) which has increased imbalances and mismatches to hit inflation target with help of

(9)

TL especially after 2014 by 113% would have considerable consequences in Turkish economy after taking a glance at Figure 3.

After Fed signalled tightening of monetary policy in 2013, volatility of capital flows and currencies in EMs has increased (Société Générale, 2016, November). Stress following this increase in volatility and where it originates can be explained by the “Original Sin Hypothesis” of Eichengreen&Hausmann (1999). They explain that inability to issue domestic currency denominated debt in international capital markets or in the long-term even for domestic investment as confirmed by Onaran (2007) leads to unavoidable financial fragility because of currency and/or maturity mismatch in the private sector balance sheet. It can be inferred that high level of open FX position in balance sheet leads to detererioration of the balance sheet by increasing the burden of debt service in case of domestic currency depreciation contrary to appreciation strengthening the balance sheet. Consequently, changes in exchange rate affecting spending, production, and investment decisions of private sector via balance sheet constraint and exacerbate the effects of shocks on the economy (Bernanke&Gertler&Gilchrist, 1996). This effect known in the literature as “open economy Bernanke-Gertler-Gilchrist effect” which corresponds to higher exchange rate exposure and vulnerability of firms can be used to explain the positive correlation between FPI inflows and highly volatile GDP growth rate in Turkey (Figure 1). Current situation of Turkey reminds of 1997-1998 Asian financial crisis with a significantly depreciated currency and accumulated FX debt. In this case, however, non-financial sector holds bigger share of the total FX debt instead of non-financial sector as in Asian example (Société Générale, 2016, November). In addition, Krugman (1999) discovers the causes of the Asian financial crisis and proposes a 3rd_{generation currency crisis model} benefiting from previous research and indicates the importance of self-fulfilling panic because of balance sheet vulnerability to self-validating collapses in confidence.

Political turmoil started with Gezi Park protests and peaked with failed coup attempt, has been catalyzing capital flight and leading to unstable economic environment in Turkey considering the fragility as mentioned in the previous paragraph. Even more leveraged corporate balance sheets, especially non-financial sector having considerable mismatches in their balance sheet, after GFC by means of exchange rate appreciations or stability are subject to significant deleveraging pressure and liquidity risk when their TL depreciates (Kalemli-Ozcan&Liu&Shim, 2018, March). The problem is amplified by lack or high costs of hedging, especially for small and medium-sized enterprises (SME) and moral hazard as increased expectations of non-financial sector to be bailed out in this case by policies aiming to curb

(10)

depreciation pressures (Kliatskova&Mikkelsen, 2015, December). Kliatskova&Mikkelsen (2015, December) also explain that moral hazard causes non-financial sector not to internalize their risks and even borrow more in foreign currency therefore this suggests that the FX debt of non-financial corporates is indeed more important than that of the government and banks in affecting the policy reaction to exchange rate movements. Thus, Krugman’s (1999) third generation currency crisis model may explain the aggravating effects of exchange rate depreciations leading causing payments problems and creating further upward pressure on EMP in Turkey (Pienkowski, 2017).

Figure 3 – Spot USD/TRY (End-of-Month) 2.2. Exchange Market Pressure (EMP) Measures

In order to construct an appropriate EMP measure for Turkey, two basic parts of EMP have to be taken into account and reviewed which are components of it and their weights. The literature of EMP consists of four major articles on this issue which are Girton&Roper (1977), Eichengreen et al. (1995b), Weymark (1995) and Klaassen&Jager (2011).

The first EMP measure was developed by Girton&Roper by using a monetary model to measure required intervention to offset disequilibrium in the domestic source of the monetary base to achieve an exchange rate target in Canadian Dollar vis-à-vis USD (1977). Their definition of EMP can be summarized as sum of the rate of appreciation or depreciation of home currency relative to foreign currency and negative of the percentage change in the ratio of foreign reserves valued in domestic currency to the total supply of domestic base money.

(11)

Although it was the first EMP measure, lack of interest rate component hinders EMP measure to reveal the actual EMP. Girton&Roper (1977) EMP measure is as follows:

EMPt = st – (rt) (1)

where st denotes the percentage change of TL per USD and rt denotes the percentage change in the ratio of domestic value of foreign reserves change to the total supply of base money within prespecified time period.

Eichengreen et al. (1995a, 1995b) develops a model-independent EMP measure on the contrary to Girton&Roper (1977) and weights its components by the ratios of the standard deviation (SD) of exchange rate change to the SD of the corresponding component to restrain from a volatile component to dominate EMP measure. The reason it is called model-independent is that none of the components or weights are derived within a model, but raw data is used to calculate EMP. Starting from pegged currency countries example, they state that increasing interest rate to meet speculative sales to defend currency can actually decrease capital flight and in turn prevents domestic currency to depreciate further (1995a). Therefore, as also the case of Turkey suggests interest rate component should be included to EMP measure in addition to Girton&Roper’s (1977). Eichengreen et al. (1995a) measures EMP of 22 (mostly OECD) counries against Deutsche Mark, additionaly suggests and applies the idea of using EMP measure to define crisis times by using its mean and standard deviation to specify outliers (i.e. currency crisis) (Eichengreen et al.,1995b). Their EMP measure follows as:

EMPt = st + st

ln((it− it∗))(ln(it – it*)) – st

rt (rt) (2)

where ln(it – it*) is the natural logarithm of the difference between short-term domestic and foreign interest rates, thus ln(it – it*)) denotes the percentage change of the mentioned interest rate difference and percentage change of foreign exchange reserves as a proportion of first lagged narrow money supply (M1). st, ln(it – it*) and rt are SDs of corresponding components.

Weymark (1995) introduces general definition of EMP as the exchange rate depreciation (appreciation) that would have been required to remove the excess supply of (demand for) a currency (i.e. especially for Canadian Dollar in her research) in the absence of exchange market intervention by central bank (CB). She also suggests using the foreign reserve elasticity of exchange rate derived within a model as weight of reserve change component to

(12)

convert reserve changes into exchange rate unit equivalent for EMP measure (Weymark, 1995). Therefore, Weymark’s (1995) EMP measure consists of weighted reserve intervention component and nominal change in the exchange rate,

EMPt = st + (rt) (3)

where the only coefficient unkown  is the foreign reserve elasticity of exchange rate (−st rt). In spite of all the improvements there has been, to the author’s knowledge, no definition-consistent suggestions made to measure of interest rate component of EMP except the methodology of Klaassen&Jager (2011). In the EMP literature, measures are either lack of interest rate instrument or include it as the first-difference of interest rate or domestic and foreign interest rate differential. However, first it is a wide known fact that countries and in our case Turkey broadlys uses short-term interest rates to smoothen exchange rate fluctuations and secondly as it can be seen from Figure 5 using interest rate changes can be misleading in case of a steady interest rate period following a drastic change in interest rate (e.g. 6 months of unchanged O/N interest rate period after the increase of O/N borrowing rate by 425 BPS and O/N lending rate by 450 BPS on January 29 2014 to reverse capital flights following a political turmoil) would incorrectly undercalculate EMP in the following period. Klaassen&Jager (2011) proposes a solution to this problem by using Taylor Rule (1993) to develop a new approach. Their approach aims to reveal how interest rate differs from what it would be without the motive of defending domestic currency of CB (i.e. difference between interest rate and counterfactual interest rate of passive CB in exchange market) to measure the exchange rate change in the absence of CB intervention which is consistent with Weymark’s definition of EMP. Besides Klaassen&Jager (2011) EMP measure has been proved to derive superior results of EMP than others’ and the generalized EMP measure in the literature (Equation (4) to specify crisis events for both 1992-1993 Exchange Rate Mechanism (ERM) crisis and 1997-1998 Asian crisis. Therefore, their methodology will be used throughout the paper to construct EMP Index for Turkey and detailed in Section 3.

EMPt = st + wiit + wc (− 𝑅t

(13)

2.3. EMP in Turkey

Girton&Roper (1977) EMP measure was applied to Turkish economy by Parlaktuna (2005) for the period between 1993 and 2004. Her main conclusions are; there are statistically significant and positive relationship between money multiplier and EMP, and negatively statistically significant relationship between domestic credit as a proportion of base money. Parlaktuna (2005) also concludes that there is not statistically significant effect of foreign price level and real income on EMP. Coefficients are obtained by ordinary least squares (OLS) regression and the effectiveness of foreign reserves to absorb EMP shocks between 1993 and 2000 found to be successful.

Goksu&Kadioglu&Kucukkocaoglu (2015) apply the methodology of Eichengreen et al. (1995a, 1995b) to Turkey between 1998 and 2013 by using first-difference of short-term interest rate as interest rate intervention proxy variable. They state that Turkey experienced three periods in which EMP exceeds the forestated threshold level of 1.5 σEMP + μEMP where the σEMP and μEMP are sample SD and sample average of EMP respectively. Two of these events occured during 2000-2001 crisis of Turkey and one after 2007-2008 GFC, in the beginning of 2009.

Feridun (2012) shows that Eichengreen et al. (1995a, 1995b) EMP measure outperforms Weymark’s (1995) and Girton&Roper’s (1977) approach to determine crisis periods in Turkey between 1992 and 2006 by using domestic and foreign interest rate difference as interest rate intervention proxy variable to calculate EMP.

2.4. ARDL Bounds Testing Approach

In order to investigate the long-run and short-run relationship of EMP with net foreign capital inflow we will apply a cointegration technique and transform our model into error correction representation. Engle&Granger (1987) and Johansen (1988) cointegration tests can be applied if all variables are integrated in the same order. However, our preliminary investigation indicates that our variables are a mixture of I(0) and I(1). Thus, following Pesaran&Shin (1998) and Pesaran et al. (2001) ARDL Bounds Testing approach to cointegration can applied irrespective of whether variables are either I(0) or I(1). Another advantage of ARDL Bounds Testing approach is it performs much better in small samples and yields consistent estimators of short-run and long-run coefficients (Pesaran&Shin, 1998). In order to test the long-run cointegration unrestricted error correction model (UECM)

(14)

transformation will be applied to our base ARDL model (Pesaran&Shin, 1998). Another advantage of ARDL bounds testing approach is it is possible to obtain long-run elasticities between our variables by using UECM coefficients (Tsounis&Vlachvei, 2017). Cointegration is going to be tested by using Wald test for joint statistical significance on long-run coefficients (i.e. bounds test) in UECM and compare the F-statistic with asymptotic critical values (Pesaran et al., 2001). If cointegration is confirmed by rejecting the null hypothesis in case of F-statistic exceed the critical bound, we will continue to form the vector equilibrium (or error) correction model (ECM) to estimate short-run elasticities (Pesaran et al., 2001). ECM also reveals the speed of adjustment to equilibrium to ensure that our model is dynamically stable (Pesaran&Shin, 1998). ARDL bounds testing approach is based on OLS estimation of our models. Considering our relatively short 15-year time span data and different order of integration of the variables, ARDL Bounds Testing approach to cointegration allowing short-run and long-short-run dynamics between variables is appropriate for our analysis.

ARDL bounds testing approach has been used widely in the literature and some of the examples are given following in the paragraph. Morley (2007) examines the effect of stock market on exchange rate in the context of the monetary model of exchange rate determination and stated a significant short-term effect with different signs in different lags in the UK. De Vita&Abbott (2004) tests impact of real exchange rate volatility on export volume to ssome elected countries and rest, and they found that volatility has a statistically significant impact on U.S. exports to most of the markets although the sign and magnitude of the effect varies. Atkins&Coe (2002) examines the empirical validity of the LR Fisher effect in Canada and states that in the LR nominal interest rate response to change in the inflation rate is close to unity supporting Fisher hypothesis.

2.5. Applications of ARDL Bounds Testing Approach to Turkey

Feridun (2012) examines the relationship between liability dollarization by applying ARDL bounds testing approach in Turkey. He founds that liability dollarization level of banks puts statistically significant upward pressure on EMP by using Eichengreen et al. (1995a, 1995b) measure (Feridun, 2012). On the contrary to his analysis we would like to investigate the relationship of non-financial sector due to its more fragile financial structure. In addition to that, Feridun (2012) uses two proxies to measure liability dollarization which are the ratio of foreign liabilities to foreign assets and the ratio of foreign liabilities to total assets. However,

(15)

net open FX position of non-financial sector in proportion to GDP will be used in our model as the variable corresponding currency mismatch.

Feridun (2009) and Katircioglu&Feridun (2011) also investigate the relationship between all budget balance to GDP ratio, current account balance to GDP ratio, domestic credit to GDP ratio, excessive monetary expansion, real exchange rate overvaluation, decrease of foreign reserves to M2 ratio and increase of banking sector fragility index with EMP. They found a positive and statistically significant relationship between all variables with EMP by using ARDL bounds testing approach for analysis of the relationship of variables individually with EMP (Feridun, 2009; Katircioglu&Feridun, 2011).

(16)

3. EMP Measure for Turkey

This section starts with description of EMP components for Turkey, Section 3.2. continues with the corresponding weight selection and calculation, in Section 3.3. data used to measure EMP for Turkey will be detailed.

3.1. Componenets of EMP in Turkey

Klaassen and Jager (2011) proposes the EMP measure below, EMPt = st + wi (it – itd_{) + wc ct} ₍₅₎

where  is the first-difference operator and st is the natural logarithm of the nominal spot exchange rate (the price of one unit of foreign currency in terms of TL), hence st linearly approximates to the percentage nominal exchange rate change. it is the short-term interest rate in Turkey at t, itd _{is the counterfactual interest rate in Turkey at t which is the interest rate if} CBRT had no motive to remove the excess supply or demand in the exchange rate market through interest rate channel. Thus, difference of them is used as interest rate intervention proxy for our measure. ct is the foreign reserve intervention by the CBRT scaled by M1 of the previous period (t-1). Following the previous research counter country for the EMP measure was choosen as U.S. because of its convenience. Also, CBRT holds foreign reserves mostly in USD to use as a policy instrument to narrow down the fluctuation margin of the exchange rate. From time to time it intervenes to the market by using its foreign reserves or through increasing interest rates to attract foreign capital to relieve foreign currency shortage. Thus, merely considering nominal change of TL vis-à-vis USD does not reflect actual EMP.

First component is the percentage change of nominal exchange rate. Because TL has depreciated significantly, using the nominal change instead of rate of change in its value would incorrectly yield larger changes in our EMP measure even though change is smaller in percentage (Figure 4). In order to prevent this misspecification, change in the natural logarithm of the nominal spot exchange rate, st, which linearly approximates to the percentage exchange rate change will be used.

The second component is the interest rate intervention component. In order to calculate the counterfactual interest rate, as Klaassen and Jager (2011) suggests, generalized Taylor Rule will be used. Literature suggests that using expected future values instead of real time values

(17)

would increase the effectiveness of Taylor Rule to define the domestic interest rate (Clarida&Galí&Gertler, 1998). Therefore, the Taylor Rule that will be followed throughout the study is given as,

itd_{= rt}eq_{+ t}e_{+ gt + t} ₍₆₎

where rteq_{is the equilibrium real interest rate at time t, t}e_{is the expected inflation rate using} information available at time t, gt is the column vector of output gap forecast and inflation gap. Output gap defines the deviation of the GDP from its potential and inflation gap defines the deviation of expected inflation from inflation target.  is the policy preference coefficient regarding the gaps and t is the omitted monetary policy determinants.

According to Bernanke, The Fed has been following the Taylor Rule closely most of the time (2015, April 28). Therefore, it would be convenient to approximate U.S. interest rate with the following monetary policy rule,

it*_{= rt}eq*_{+ t}e*_{+ }*_gt*_{+ t}* ₍₇₎

If we assume that due to high integration of financial markets domestic and foreign omitted monetary policy determinants t and t* _{overlaps, by combining equations (2) and (3)} we get,

itd_{= it}*_{+ (t}e_{- t}e*_{) + (gt - }*_gt*_{) + (rt}eq_{- rt}eq*₎ ₍₈₎

Further in line with Klaassen and Jager (2011), we assume that domestic and foreign equilibrium real interest rates rteq_{and rt}eq*_{are equal and foreign policy preference coefficient} regarding the gaps, and *_{are both equal to 0.5. Thus, counterfactual interest rate equation for} Turkey is,

itd_{= it}*_{+ (t}e_{- t}e*_{) + (gt - gt}*₎ ₍₉₎

And the final component of our EMP measure is the scaled foreign reserve intervention. It is the negative value of the change of CBRT foreign reserves divided by M1t-1 which is the preivous end-of- month money supply in Turkey in USD. The logic of scaling the foreign reserve intervention by narrow money supply is to take into account of increasing FX market turnover. As the market size increases the same amount of intervention by CBs would become less and less effective to affect exchange rate. Due to data inavailability of FX market turnover,

(18)

the change of net foreign reserves of CBRT is scaled by M1t-1 following Jager and Klaassen (2011).

3.2. Weights of EMP Components in Turkey

As Jager and Klaassen (2011) suggest Eichengreen et al. (1995b) approach will be followed to calculate weights of EMP measure componenets for Turkey. It suggests using weights:

wi = st

(it− itd) and wc = st

ct (10)&(11)

where st is the SD of the monthly percentage nominal exchange rate change of TL vis-à-vis USD, (it – itd_{) is the SD of the difference between the nominal interest rate and the} counterfactual interest rate and ct is the SD of the scaled foreign reserve intervention. The reason we follow Eichengreen et al. (1995b) methodology is to restrain a volatile component to dominate EMP measure and its plausible results although its simplicity to calculate.

3.3. Data Collection Methods and Description

In this section data used to proxy the variables for our EMP measure and their sources will be detailed. Because we will measure EMP on monthly basis, the data was collected accordingly covering from the end-of month values of August 2000 till the end of June 2018.

First component of EMP Model, exchange rate component requires to obtain the end-of-month spot nominal exchange rate rate of TL vis-à-vis USD. This data was collected through Datastream. Obtained monthly nominal spot change and percentage change of TL vis-à-vis USD monthly given in Figure 4. Due to recent depreciation of TL, percentage exchange rate change will be used instead of nominal change as it is explained before.

Interest rate component requires monthly data for nominal interest rate, expected inflation rate, output gap forecast and inflation gap for both Turkey and U.S.

Instead of of taking 3-months deposit rate which is considerably lower than or O/N lending rate of CBRT which does not completely reflects the interest rate policy and deviates , from ST market interest rate, we propose to use WAFC as interest rate proxy because it reflects different channels of funding by CBRT. WAFC is calculated by taking the weighted average realized average overnight (O/N) rate in interbank money market, weekly repo rate

(19)

Figure 4 – Nominal Spot Change and Percentage Change in USD/TRY

(i.e. policy rate) and late liquidity window (LLW) rate. LLW is a relatively less limited lending facility of CBRT as a lender of last resort to prevent liquidity problems of banks at the end of the day because of limited interbank money market (e.g. CBRT changed regulation on Turkish Lira Liquidity Management by decreasing borrowing limit on interbank money market to zero to fund banks with late liquidity window instrument (CBRT Press Release, 2017). Therefore, it can be said that CBRT affect credit interest rates, exchange rates, interest rates on deposits, credit amounts by changing the weight of funding instruments which have considerably different rates and using WAFC would be more appropriate than the alternatives under these circumstances. However, CBRT EDDS only provides data for WAFC from January 2011 till now. Despite to this problem O/N interbank money market rate (i.e. Weighted Average of Overnight Simple Interest Rate) data is accessible and it is included to our dataset. Taking into account that those two variables approximates in normal times closely but due to relatively high rate of LLW that CBRT uses widely to fund the market since the start of the year 2017 those differ considerably since the beginning of 2017. Considering that LLW was used rarely before 2017 and they are different mainly due to CBRT started to implement more complex monetary policy to fund the market in 2010, we will use O/N interbank money market rate for the time period between September 2001 and October 2010(IMF, 2013). The reason November 2010 and December 2010 are left out is that in those months that O/N interbank money market rate was unusually low also due to unorthodox monetary policy of CBRT creating positive spread between market rates an done week repo rate and because CBRT decreased O/N borrowing rate

(20)

to an unusual low level very suddenly due to abrupt increase international capital inflow (IMF, 2013; Tchaidze&Hesse, 2013, April 2).

Instead Weighted Average Interest Rate for Deposits in TL Up to 1 Month was used which also approximates WAFC. O/N interbank money market rate and Weighted Average Interest Rate for Deposits in TL Up to 1 Month also approximates to each other and move together most of the time especially before 2010 due to new set of monetary policy implementation as it can be seen in Figure 6. All monthly data was collected from CBRT EDDS.

Figure 5 – CBRT Weighted Average Cost of Funding, Overnight Borrowing Rate and Overnight Lending Rate

Figure 6 – Weighted Average of Overnight Simple Interest Rate, WAFC, Weighted Average Interest Rate for Deposits in TL Up to 1 Month

(21)

Following Klaassen and Jager (2011) and Clarida et al. (1998), 12 months ahead expected inflation rate will be used to construct the counterfactual interest rate for Turkey. This data was collected from CBRT EDDS. However, it poses a challenge because this dates back to August 2001 on the database. Albeit to the problem CBRT Monetary Policy Applications Presentation (January 2002) contains 12 months ahead inflation expectations graph. In order to materialize the information, the graph has been digitized to obtain approximated inflation expectations data required for the timespan between September 2000 and August 2001.

In addition to inflation expectations, forecast of output gap also included following Clarida et al. (1998). Quarterly data for output gap forecast in Turkey and U.S. was collected through Datastream. This poses another problem for our monthly measure of EMP. In order to obtain proxy variables for monthly output gaps forecasts Data Frequency Conversion feature of EViews was used. Some of interpolation techniques to obtain monthly proxy data which has been used in the literature are cubic spline interpolation, linear interpolation and quadratic interpolation. For cubic spline interpolation and linear interpolation to be useful we require data points interpolated to specify the exact values at a start and an end date. However, because quarterly data reflects the average of 3 months period, not specific values of months to interpolate in between, the quadratic interpolation method with matching feature allowing to take quarterly data and interpolates to 3 relevant months accordingly to make 3 aforementioned months’ average equal to the quarterly data was used to obtain monthly proxy variables for output gap forecast in Turkey and U.S.

(22)

The other gap component is inflation gap. This is the difference between the expected inflation and inflation target. As it is mentioned before CBRT has started to use inflation target regime first loosely and then strictly in 2002 and 2006, respectively. Data for yearly inflation target was collected from CBRT website by following the menu Core Functions>Monetary Policy>Price Stability and Inflation> Inflation Targets. Inflation target of CBRT is specified and set as the year-end inflation rate calculated as the previous 12 months change in the CPI. Therefore, it would be appropriate to take yearly inflation target for the specified year also as the aforementioned month’s inflation target. Another problem is that our EMP measure also includes the time between September 2000 and January 2002. Although CBRT had no official inflation targeting regime, Anti-Inlation Program of 2000 had the inflation target of 25 that it has failed to achieve (CBRT Monetary Policy Report 2001, 2001, May 15). In addition to that Letter of Intent written by State Minister in charge of the Economy and CBRT President explains the reasons of why inflation target of 25% has not been met and indicates that the inflation target for December 2001 is 12% (Onal&Ercel, 2000, December 18). However, because Turkish economy was hit by February 2001 crisis inflation target for December 2001 was reset to 52.5% (CBRT Monetary Policy Report 2001, 2001, May 15). Further, it was reset to 58 with an additional Letter of Intent (Dervis&Serdengecti, July 31, 2001). Therefore, although CBRT had no official inflation target regime, as it is stated monetary policy reports and Turkey State Planning Organization’s Development Plans (1999, 2000) which also contain the same previous inflation targets for 2000 and 2001, CBRT and government collaborated to meet the inflation targets. Thus, these values are also taken as inflation target to be able to construct counterfactual interest rate for Turkey.

Analogously, monthly Federal Fund Rate and 12 months aheadh expected inflation rate for U.S. has been obtained through Federal Reserve Economic Data (FRED) St. Louis FED.

In 2012, Federal Open Market Committee (FOMC) declared that inflation at the rate of 2 percent, as measured by the annual change in the price index for personal consumption expenditures, is most consistent over the longer run with the Federal Reserve's statutory mandate for price stability and maximum employment.(The Fed Press Release, 2012, January 25) In addition to that seven months before the press release Fed stated that although The Federal Reserve has not established a formal inflation target, policymakers generally believe that an acceptable inflation rate is around 2 percent or a bit below (Current FAQs – Informing the public about the Federal Reserve, 2011, June 22). Additionaly it is stated in the statement

(23)

personal consumption expenditures (PCE) to change over the longer run, the FOMC can then implement monetary policy to help maintain an acceptable inflation rate; which is neither too high nor too low and as of June 22, 2011, PCE inflation projections for the longer run ranged from 1.5 percent to 2 percent. In addition, President of the Federal Reserve Bank of Boston, Eric S. Rosengren stated that although The Federal Reserve had no explicit inflation targeting regime before January 2012, Federal Reseve Monetary Policy can be observed as having an implicit inflation target of 2 percent between 1993 and 2012 (2013, April). Therefore, for both convenience and simplicity inflation target of U.S. will be taken as 2 percent to counstruct counterfactual interest rate for Turkey.

Interest rate of both Turkey and U.S. and calculated counterfactual interest rate in Turkey is presented in Figure 8.

Figure 8 – Interest Rate and Counterfactual Interest Rate in Turkey, Interest Rate in U.S.

Third component is the scaled foreign reserve intervention component. Klaassen and Jager (2011) suggests to proxy ct by -Rt /1, where Rt is total reserves minus gold and M1t-1 is lagged money supply MM1t-1. Total foreign reserve changes of CBRT is given in Figure 9. The data for foreign reserve of CBRT was collected from CBRT EDDS and monthly data for M1 was collected from CBRT EDDS for the time between December 2005 and today and the time between September 2000 and December 2005 was collected through Datastream.

(24)

(25)

4. ARDL Bounds Testing Approach for Turkey

Purpose of Chapter 4 is to first build an ARDL model and apply ARDL bounds testing model within UECM and ECM to investigate the short run (SR) and long run (LR) relationship between EMP with current account deficit and net FX position of non-financial sector and total net inflows of FDI and FPI. First, Section 4.1. constructs our empirical model within ARDL bounds testing approach. Next, data used to estimate the regressions will be detailed in Section 4.2.

4.1. Econometric Model

One of the aims of this research is to investigate whether elevated net open FX position of non-financial companies, current account deficit and foreign investment have a considerable effect on EMP. Pienkowski (2017) states that currency mismatches have an important effect on EMP. Also considering the possibility of self-fulfilling currency crisis we will use net open FX position of non-financial companies as an independent variable in our regression (Krugman, 1999). Yeldan&Ünüvar (2016) indicates the fragility associated with both the level and the external debt induced financing of the CAD relying on hot money inflow. Thus, we are using both CAD1 and NetFI as independent variables in order to take into consideration of both level and significance of financing it by foreign investment (FI) primarily composed of volatile FPI rather than stable FDI in Turkey.

In order to analyze the relationship, we first build our base ARDL model. The ARDL model is given by the following regression equation,

EMP_t= ∑_1𝑖EMP_t−i 𝑝 𝑖=1 + ∑_1𝑖NetFX_t−i 𝑞 𝑖=0 + ∑_1𝑖NetFI_t−i 𝑟 𝑖=0 + ∑_1𝑖CAD_1t−i 𝑠 𝑖=0 +₁₀+₁₁𝑇 +₁₂𝐷₁+₁₃𝐷₂+_1𝑡 (12)

where NetFXt is the net open FX position of non-financial companies divided by GDP, NetFIt is the total net inflows of FDI and FPI divided by GDP and CAD1t is the CAD divided by GDP, 0 is the constant term, T is the time trend considering decreasing trend of EMP after 2001 crisis due to the decrease of very high interest rates gradually through 2007 with decrasing inflation as it will be discovered in the next section, D1 is the first dummy variable taking 1 in October 2008 in which the depreciation effect of GFC on TL materializes and 0 otherwise, D2 is the second dummy variable taking 1 in May 2006 in order to capture the effect of brief turbulence

(26)

in financial markets especially in EM and 0 otherwise (BIS, 2006), and t is the error term. In order to estimate the model, the lag combination of the variables should be determined. Lag values of p, q, r and s stand for lagged values of all variables determined by Akaike Information Criterion (AIC) following Pesaran et al. (2001). AIC produces more lags than Schwarz Bayesian Information Criterion (SBC) and as we have monthly data having more lags will strengthen our model.

Dependent and independent variables will be tested for stationarity, whether they are I(0) or I(1), by using Augmented Dickey Fuller (ADF) test and Phillips-Perron (PP) test following Belke&Beckmann&Verheyen (2013) and Feridun (2009, 2012) due to their improvements over Dicket-Fuller unit root test suggested by Pesaran et al. (2001). These unit root tests are testing the null hypothesis that the time-series data integrates with the order of 1 against the alternative hypothesis that it is stationary. Lag lengths and the bandwidths are selected with Akaike Information Criterion (AIC) and the Newey–West Bartlett kernel, for ADF and PP tests respectively (Feridun, 2012). Results of ADF and PP tests are depicted in Table 1.

(27)

Values in table correspond to t-statistics and corrected t-statistics of tested coefficients for ADF and PP tests respectively. If t-statistic is lower than the corresponding critical value null hypothesis of non-stationarity is rejected in favor of the alternative hypothesis. If all variables are stationary OLS estimate of our base ARDL model can be used.However if there is any non-stationary variable, regression might lead to nonnormal distribution of t-statistics, coeffients that are biased to zero or spurious regression. (Stock&Watson, 2012). Thus, is important to know the level of integration of our variables.

According to Table 1 it can be concluded that it is appropriate to use ARDL bounds testing approach. Therfore, following Pesaran&Shin (1998) and Pesaran et al. (2001) UECM version of our base ARDL model will be used to test for cointegration. UECM to test cointegration is given below,

EMP_t=₁𝐸𝑀𝑃_𝑡−1+₂𝑁𝑒𝑡𝐹𝑋_𝑡−1+₃𝑁𝑒𝑡𝐹𝐼_𝑡−1+₄𝐶𝐴𝐷_{1 𝑡−1}+ ∑_2𝑖EMP_t−i 𝑝−1 𝑖=1 + ∑_2𝑖NetFX_t−i 𝑞−1 𝑖=0 + ∑_2𝑖NetFI_t−i 𝑟−1 𝑖=0 + ∑_2𝑖CAD_1t−i 𝑠−1 𝑖=0 +₂₀+₂₁𝑇 +₂₂𝐷₁+₂₃𝐷₂+_1𝑡 (13)

where 1, 2, 3 and 4 are the coefficients of the first lagged level variables indicating long-term and 2i,2i,2i,2i are coefficients of first-differenced variables indicating short-long-term effects. It is anticipated that an increase of the NetFX and CAD1 to have a positive effect, an increase of NetFI to have a negative effect on EMP in the LR. We also expect NetFX and NetFI to have negative effect but CAD1 to have a positive effect on EMP in the SR. 1t is the error term.

The null hypothesis of no integration (H0: 1 = 2 = 3 = 4 = 0) against alternative hypothesis of cointegration (H1: 1  2  3  4  0) will be tested by applying Wald test to calculate the F-statistic and comparing it with the corresponding table in Pesaran et al. (2001). Table contains two bounds which are upper bound for I(1) and lower bound for I(0). When the value of F-statistic is higher than the upper bound the null hypothesis is rejected and we can confirm the cointegration, but if the F-statistic is lower than the lower bound the null hypothesis cannot be rejected so long-run relationship between variables is rejected and lastly if the value falls between the upper and lower bounds the result will be inconclusive (Pesaran et al., 2001).

(28)

Assuming that UECM equilibrium is stable in the LR (i.e. EMP = NetFX = NetFI = CAD1 = 0) we can extract LR elasticities of EMP on independent variables (Pesaran&Shin, 1998). By algebraic operations we can obtain LR elasticities of EMP on independent variables which are −2

₁,

−

₃

₁ and

−

₄

₁ for NetFX, NetFI and CAD1 respectively and test for their

statistical significance by using Wald test.

After confirming the cointegration relationship between variables by applying this methodology, the following step is the estimation of the ECM of ARDL model given below,

EMPt = ∑_3𝑖EMPt−i 𝑝−1 𝑖=1 + ∑3𝑖NetFXt−i 𝑞−1 𝑖=0 + ∑3𝑖NetFIt−i 𝑟−1 𝑖=0 + ∑_3𝑖CAD_1t−i 𝑠−1 𝑖=0 + 𝐸𝐶𝑇𝑡−1+30+31𝑇 +32𝐷1+33𝐷2+2𝑡 (14)

where



2t is the error term, and the only unknown independent variable is ECT, the “error correction term”, which is the first lag of the residual series, z(t), of the following LR “cointegration equation” (Pesaran et al., (2001),

EMPt = − ₂ ₁𝑁𝑒𝑡𝐹𝑋𝑡− ₃ ₁𝑁𝑒𝑡𝐹𝐼𝑡− ₄ ₁𝐶𝐴𝐷1𝑡+ 𝑧(𝑡) (15)

As Stock&Watson suggests (2012) we need to check whether the error terms are serially independent because serially dependent error terms would cause standard errors to be invalid and consequently t-statistics to be invalid too. In order to test for autocorrelation LM test and Durbin-Watson (DW) statistics, will be used.

Additionaly, Breusch-Pagan-Godfrey (BPG), Jarque-Bera (JB) and Ramsey RESET tests will be implemented in order to check for heteroskedasticity, residual non-normality and functional misspecification problems.

Lastly, the cumulative sum of the recursive residuals, CUSUM and cumulative sum of squares of recursive residuals, CUSUMSQ, tests will be applied to check the stability of the coefficients.

(29)

4.2. Data Collection Methods and Description

In order to run our regressions, monthly data of net FX position of non-financial sector, net FDI, net FPI and CAD were obtained from CBRT EDDS and quarterly data of GDP of Turkey was obtained via Datastream. Due to data availability of net FX position of non-financial sector, sample period is taken as December 2012-March 2018.

As we have encountered before quarterly data poses a problem for the calculations. Thus, as it was explained before because quarterly data reflects the sum of 3 months period of GDP, not specific values of months to interpolate in between, in order to obtain monthly proxy variables for GDP we will use the quadratic interpolation method by matching data with their sum which allows to take quarterly data and interpolates to 3 relevant months accordingly to make 3 aforementioned months’ sum equal to the quarterly data. After obtaining all relevant monthly data we proceed to calculate independent variables.

NetFXt is the net open FX position of non-financial companies at month t, divided by the total of the last relevant 12 months GDP (i.e. aforementioned month and the previous 11 months), NetFIt is the total net inflows of FDI and FPI of the last 12 months divided by the total of the same 12 months GDP and CAD1t is the sum of CAD of the last 12 months divided by the total of the same 12 months GDP. As all these variables are ratios of two observed variables, they will be used directly instead of being transformed into their natural logarithm form.

As it can be seen in Figure 10, NetFX is the most changed independent variable from 2003 until the early 2018 although it seems stable around 5% through 2003 till 2007. However, one should also take into consideration that NetFX is a stock variable while NetFI and CAD1 are flow variables. TL conserving its nominal value vis-à-vis USD from 2003 until mid of 2011 on average, increased availability of low-cost credits from international financial markets due to interest rate spread. Quick recovery of Turkish economy after the negative effects of GFC tails off, combined with historically low interest rates in international capital markets compared to Turkey and caused demand for FX credits to increase even further (Yörükoğlu&Atasoy, 2010). In spite of the warnings by academics (Karaman, 2015) and increasing depreciation risk of TL since 2013, due to inability of Turkish economy to create LR resources for private sector, net FX position has reached around one quarter of GDP in the beginning of 2018. Figure 11 supports this claim by demonstrating that increase of net FX liabilities is mainly driven by the long-term liabilities rather than short term.

(30)

Figure 10 – Net FX Position of Non-Financial Sector divided by GDP (NetFX), Net Foreign Investment divided by GDP (NetFI), CAD divided by GDP (CAD1)

Figure 11 – Net Open FX Position of Non-Financial Sector (US$mn), Long-Term Liabilities of Non-Financial Sector (US$mn)

It is possible to infer from Figure 10 that periods of decrease and slowed down of net FI inflows coincide with upward movements of NetFX. This supports Onaran (2007) stating the

dependency of Turkish economy to foreign capital for investment and to boost the GDP growth. Another remark would be the positive difference between CAD1 and NetFI since third quarter

of 2013 coincides with Gezi Park protests, following political turmoil until now and significant depreciation of TL vis-à-vis USD. It strenghtens our choice of adding NetFI together with CAD1 in our regression as an improvement over Katircioglu&Feridun (2011).

(31)

5. EMP Analysis in Turkey

In this section first, EMP indicated by components of nominal exchange rate change, interest rate intervention and foreign reserve intervention will be examined in Sections 4.1., 4.2. and 4.3. respectively. Section 4.4. identifies the weights of our EMP measure. After indicating and analyzing each component and calculating the weights, this section concludes in Section 4.5. with the aggregate monthly EMP measure and its analysis for Turkey between September 2000 and March 2018.

5.1. Nominal Exchange Rate Change Component

As it is specified in Section 3.1. the first component of our EMP measure, st, is the monthly percentage nominal exchange rate change of TL vis-à-vis USD. As Weymark’s (1995) general definition of EMP suggests st is the the part of EMP which is cleared in the exchange rate market through (percentage) exchange rate depreciation (appreciation) to remove the excess supply of (demand for) TL other than the exchange market intervention by the CBRT. A positive value of st reflects a depreciation pressure on TL; hence st is the natural logarithm of the price of one unit of USD in terms of TL. Conversely a negative value of st reflects an appreciation pressure on TL. The monthly percentage nominal exchange rate change of TL vis-à-vis USD, st, is indicated in Figure 12.

Figure 12 - Monthly Percentage Change in Nominal Spot USD/TRY

Acording to the Figure 12 there are 3 months in which st exceeds 0.10 clearly except recent depreciations, these are February 2001, May 2006, and October 2008 reflecting 2001

(32)

crisis in Turkey, brief financial turmoil in 2006 and 2007-2008 GFC. This basic inference can be that although TL used to demonstrate more volatile behavior with higher variance of movements, it has been depreciating faster since 2013 although this period does not exhibit instantaneous depreciation pressure on TL vis-à vis USD but rather gradual. In order to support our suggestion, we can examine the nominal value of TL over years. After TL depreciates by 35% within a month during February 2001 crisis it only depreciates by 21% although it exhibits significant volatility until it was hit by GFC on October 2008. Again, after depreciation of TL by 18.6% within October, it only lost its value by 17% almost over 5 years until Gezi Park protests in May 2013. Although TL exhibited volatility it can be stated that expectations used to be wider on TL to turn back to its nominal value in the LR following a depreciation if we compare after 2013 period. Thus, it would not be wrong to consider the depreciation of TL for the last 5 years as unusual experience for Turkey.

5.2. Interest Rate Intervention Component

As it can be seen in the equation (1), interest rate intervention is the difference between the nominal interest rate and the counterfactual interest rate. Our EMP measure following Klaassen and Jager (2011) takes this differenced value on the contrary to the previous research taking the first-difference of interest rate as interest rate intervention component. Figure 13 demonstrates the significant difference between the nominal interest rate and the counterfactual interest rate especially before 2010. The reason period between September 2000 and March 2001 is not included in the graph is that the it – itd_{of 162.50% and it of 119.50% on December} 2000 and the it – itd_{of 384% and the it of 393.80% on February 2001.}

The positive value of interest rate intervention reflects the fact that the nominal interest rate is higher than the counterfactual interest rate of passive CB in exchange market. It can be interpreted as having higher interest rate to defend currency in our model. Thus, while positive values of it – itd_{increase EMP, negative values decrease EMP.}

(33)

Figure 13 – Interest Rate Intervention Component (it – itd)

As Klaassen and Jager (2011) explains, the effect of one time increase or decrease of interest rate to defend currency and then remaining constant in case of a resistant pressure, would untruly induce the subsequent months’ EMP to seem lower than their actual level if nominal interest rate continues to deviate from its counterfactual value. For this reason, we have used it – itd_{instead of it. There is also another improvement of this alteration in the interest} rate intervention proxy. Periods having higher interest rates usually followed by periods of declining interest rate. A negative it has a negative effect on EMP although it might be the case that even after the decrease of the interest rate it – itd_{can be positive and has a positive} effect on EMP. An explicit example of this can be May 2001 which witnesses interest rate decrease of 35400 BPS which arrive at conclusion of very significant negative EMP within the month incorrectly.Yeldan&Ünüvar (2016) explains this high interest rate intervention period as the result of orthodox strategy of CBRT following IMF programme to maintain overvalued exchange rate, attract external savings to boost GDP growth rate. It can be concluded that after this period due to very low levels of advanced-country interest rates and higher EM interest rates, CBRT did not have to intervene through interest intervention instrument to attract foreign capital.

We can see that after 2010 interest rate intervention decreases to even zero and hasn’t been increasing much since then except the recent sharp increase to defend the currency in May 2018. According to Yeldan&Ünüvar (2016) it has its source in almost deemed tobe failed complex monetary policy of CBRT has been started to implement in 2010.

(34)

5.3. Foreign Reserve Intervention Component

Foreign reserve intervention by CBRT is taken in our model as the change of foreign reserves (Rt) scaled by by M1t-1. As it can be seen by looking at Figure 9 that although foreign reserve intervention level has increased in terms of its nominal USD value one should also take into account of the increase of the foreign exchange market turnover. A decrease of the foreign reserves reflects the sale of USD from foreign reserves to create an extra demand for TL by CBRT to defend the value of TL which should have a positive effect on EMP. Therefore, negative value of the scaled foreign reserve change (-Rt /M1t-1) is taken to proxy foreign reserve intervention. Positive values of scaled foreign reserve intervention component, ct, indicate an increase of EMP as a result of a decrease of foreign reserves, and negative values of ct indicate a decrease of EMP as a result of an increase of foreign reserves.

Figure 14 – Scaled Foreign Reserve Intervention Component (ct)

Figure 14 demonstrates that after the abrupt effects of 2000-2001 crisis has been eliminated extreme values of ct tail off after the second period of 2002. The value of ct actually has not been exceeding 0.05 since the end of 2011. Thereby we can conclude that foreign reserve intervention is not as significant as it used to be anymore for CBRT considering the rise of daily FX market turnover from 1 billion USD on April 2001 to 22 billion USD on April 2016 as an indicator (BIS, 2016). One other remark is that CBRT has increased its foreign reserves from 18 billion U.S. $ to 110 billion $ by the help of appreciated TL which is depicted as the negative values of ct in Figure 12.

(35)

5.4. Weights of EMP Components

As we have all three components, we can calculate the weights of interest rate intervention and scaled foreign reserve intervention components as it is explained in Section 3.2. However, as it is mentioned before skyrocketed interest rates of 198.95% on December 2000 and 436% on February 2001 causes SD of the interest rate intervention component to be more than three times higher than its value on Table 2. Because they are the only monthly it – itd_{values not within ±1.5*(it – it}d_{) band and actually 5.60 and 13.20 (it – it}d_{) far from zero} respectively, we discard them and calculate (it – itd_{) again to prevent the weight of the interest} rate intervention component to be very low and thus the interest rate intervention effect on EMP to be considerably lower. The SD of the interest rate intervention component that will be used from now on is indicated by (it – itd_{)* in Table 2.}

Table 2- SDs of EMP Components

After obtaining SDs of the monthly percentage nominal exchange rate change of TL vis-à-vis USD, the interest rate intervention component and the scaled foreign reserve intervention component we can continue to calculate the relevant weights. Calculated weights are shown in Table 3.

Table 3 – Weights of Interest Rate Intervention (it – itd) and Foreign Reserve Intervention (ct) Components

5.5. EMP Measure in Turkey

After calculating each three components and their weights individually we can continue to obtain monthly EMP in Turkey by following equation (1). Our EMP measure reflecting the pressure on TL is depicted in Figure 15. As it is the case in Figure 12, period between September 2000 and March 2001 is not included in the graph. The reason for this is the very high levels of EMP with 1.08 on December 2000 and and 3.15 on February 2001.

(36)

Figure 15 – EMP in Turkey

After reaching astronomical degrees as a consequence of 2000-2001 crisis, EMP continues to be positive most of the time until 2009. The high and volatile EMP until the beginning of 2006 can be attributed to the high levels of the difference between the nominal interest rate and the counterfactual interest rate and relatively extreme successive positive and negative movements of the exchange rate respectively. Alper&Kara&Yörükoğlu (2013) indicates that new policy strategy of CBRT since 2010 aims to reduce the exchange rate volatility associated with volatile cross-border capital flows. The new policy measures, gradual decrease of interest rate after succesfully bringing inflation under control, increase of investor confidence and positive macroeconomic indicators were behind the decrease of EMP through 2006 and stays relatively within an acceptable band until Gezi Park protests. The importance of recent political risks on EMP can also be concluded by looking at Figure 15. EMP measure exceeds 0.10 following a corruption scandal on December 2013. Also, during May 2018 discussions on Erdogan challenges CBRT independence and upcoming elections caused EMP increase to almost 0.16 which is composed by 11% depreciation, interest rate intervention corresponding 2.5% depreciation pressure offset by CBRT and foreign reserve intervention corresponding 2.3% depreciation pressure offset by CBRT. The subsequent month has almost the same magnitude of interest rate intervention component and more than doubled effect of reserve intervention although TL vis-à-vis USD depreciated less than 1%. It is also an example of improvement of Klaassen&Jager (2011) EMP measure during persistent stress periods in FX market.

(37)

of TL offset by CBRT via intervention instruments are included in our EMP measure which causes the nominal percentage change of TL and EMP measure to differ in stress months in which CBRT intervenes. It can be confirmed on Figure 15 that political risks which has been increasing since 2013 led EMP measure to have an upward trend during the same period.

We can again state our inference about nominal exchange rate movements on EMP mentioned in Section 5.1. It can also be confirmed by EMP movements that although TL exhibited volatility it can be stated that expectations used to be wider on TL to turn back to its nominal value in the LR following a depreciation if we compare after 2013 period. Thus, it would not be wrong to consider the depreciation of TL for the last 5 years as an unusual experience for Turkey.

(38)

6. Empirical Analysis of ARDL Bounds Testing Model

This chapter starts with empirical application of our ARDL bounds testing model in Section 6.1. and Section 6.2. provides details on diagnostic tests results of the regressions. Finally, Section 6.3. analyses the empirical results.

6.1. Regression Results

Results of ADF and PP Tests in Section 4.1. demonstrated the convenience of applying ARDL Bounds Testing approach of Pesaran&Shin (1998) and Pesaran et. al (2001). This approach will be implemented in order to test the existence of cointegration between variables. We need to apply a Wald test on long-term coefficients in UECM to investigate whether the coefficients of LR (i.e. first lagged) variables have statistically significant effect on EMP. Our null hypothesis of no cointegration (H0: 1 = 2 = 3 = 4 = 0) against alternative hypothesis of cointegration (H1: 1  2  3  4  0) can be rejected if the relevant Wald F Test statistic is larger compared to critical value of the relevant ARDL model in Table 7 which is a reproduction of the table in Pesaran et. al (2001). Table 4 shows the F-statistics regarding 5 different model choices with “no constant and no trend”, “restricted constant and no trend”, “unrestricted onstant and no trend, “unrestricted constant and restricted trend” and “both unrestricted constant and trend”. In addition to F-statistics, t-statistics on the lagged level dependent variable having non-standard distribution from the 3 models which does not include restricted variables are also given in Table 4. They are also compared to their corresponding critical bounds given in Table 5 to strengthen our conclusion (Pesaran et al., 2001). In order to reject the null hypothesis as Section 4.1. explains F-statistics should be higher than the upper bound which corresponds to pure I(1) situation of all our variables.

Table 4 – Cointegration Coefficient Results of F and t statistics

The effects of foreign capital inflows on exchange market pressure in Turkey