IslamicMarkets uses cookies.
About our cookie policy.
  of  

or
Sign in to continue reading...

Identifying Credit Supply Shocks in Turkey

Tayyar Buyukbasaran
By Tayyar Buyukbasaran
3 months ago
Identifying Credit Supply Shocks in Turkey

Create FREE account or Login to add your comment
0 Comments


Transcription

  1. Identifying Credit Supply Shocks in Turkey Tayyar B ÜYÜKBAŞARAN Gökçe KARASOY CAN Hande KÜÇÜK February 2019 Working Paper No: 19/06
  2. © Central Bank of the Republic of Turkey 2019 Address: Central Bank of the Republic of Turkey Head Office Structural Economic Research Department İstiklal Caddesi No: 10 Ulus, 06050 Ankara, Turkey Phone: +90 312 507 80 04 Facsimile: +90 312 507 78 96 The views expressed in this working paper are those of the author(s) and do not necessarily represent the official views of the Central Bank of the Republic of Turkey. The Working Paper Series are externally refereed.
  3. IDENTIFYING CREDIT SUPPLY SHOCKS IN TURKEY Tayyar B üyükbaşaran, Gökçe Karasoy Can, Hande Küçük1 Abstract This paper aims to identify credit supply shocks and analyse their macroeconomic effects in Turkey. For this purpose, we use a Bayesian Structural Vector Autoregression (SVAR) with sign and zero restrictions. We focus on the impact of credit supply shocks on real GDP growth and highlight how the size of this impact changes when we explicitly account for the effects of capital inflows on credit conditions. Hence, our results confirm the importance of external finance for credit supply in Turkey. Our main findings are robust to some alternative data choices, prior selections as well as some alternative identifying restrictions. Keywords: Credit supply shocks, SVAR, Bayesian VAR, sign and zero restrictions JEL Classification: C11, C32, E52, F41. 1 Central Bank of the Republic of Turkey. Research and Monetary Policy Department. Istiklal Cad. No.10 06100 Ulus/Ankara,TURKEY. E-mails: tayyar.buyukbasaran@tcmb.gov.tr, gokce.karasoy@tcmb.gov.tr, hande.kucuk@tcmb.gov.tr 1
  4. Non-technical Summary Turkey has experienced a notable credit deepening process since early 2000s thanks to the comprehensive structural reform agenda put in place after the 2001 financial crisis . Credit growth accelerated further due to strong capital inflows in the aftermath of the global financial crisis. The remarkable rise in credit depth and the synchronization between financial cycles and business cycles require a thorough understanding of the effects of credit market developments on the real economy. Accordingly, this paper aims to identify credit supply together with other macro shocks for Turkey and analyse their macroeconomic effects. The identification strategy assumes that credit supply shocks lead to an opposite movement between credit spread and credit growth whereas credit demand shocks move credit spread and credit growth in the same direction. Results indicate that credit supply shocks have a significant effect not only on credit spread and credit growth but also on other macroeconomic variables. Accordingly, a credit supply shock has a significant but temporary effect on output growth whereas it has a more limited and uncertain effect on inflation. Moreover monetary policy responds to credit supply shock. This paper also discusses the role of domestic and global financial conditions on credit supply in a small open emerging economy. We show that a broader definition of a credit supply shock which captures not only domestically driven changes in credit supply, but also the effects of global liquidity and capital inflows on credit conditions has a larger and more significant impact on growth compared to a narrower (more domestic) definition of a credit supply shock. The identified credit supply shock seems to be compatible with major shifts in domestic financial conditions, e.g. macroprudential measures, and global financial conditions throughout the sample. For example, The Lehman crisis of September 2008 is reflected as large negative credit supply shocks in 2008Q4 and 2009Q1. Recently, the historical shocks show that the policies regarding to improve the access of corporate sector to finance (credit guarantee fund policy) contributed positively to both credit supply and credit demand shocks during the first and second quarter of the 2017. 2
  5. I . INTRODUCTION Turkey has experienced a notable credit deepening process since early 2000s thanks to the comprehensive structural reform agenda put in place after the 2001 financial crisis. Credit growth accelerated further due to strong capital inflows in the aftermath of the global financial crisis. As a result, Turkey’s domestic credit-to-GDP ratio, which had been well-below the emerging market average for a long period of time, steadily increased at a faster pace throughout the 2000s, catching up with the emerging market average in 2015 (Figure 1). Figure 1: Credit Deepening in Turkey relative to Advanced and Emerging Economies (Domestic Credit Stock/GDP, Annual, Percent) 140 120 100 High income (OECD countries) Turkey Emerging Markets 80 60 40 20 2014 2008 2002 1996 1990 1984 1978 1972 1966 1960 0 Source: World Development Indicators (WDI). High income (OECD) is original WDI classification, Emerging markets include Chile, Indonesia, Russia, Hungary, Mexico, India, South Africa, Brazil, Poland, Malaysia, Peru, Bulgaria, Colombia, Romania, Thailand. The remarkable rise in credit depth and the synchronization between financial cycles and business cycles require a thorough understanding of the effects of credit market developments on the real economy. Given that credit growth and other macroeconomic variables such as output growth can be affected simultaneously by a variety of shocks that are exogenous to the credit market, identifying changes in credit growth that stem from credit supply developments is crucial. Accordingly, this paper aims to identify credit supply shocks for Turkey and analyse their macroeconomic effects using a structural VAR (SVAR) framework. For identification of shocks, we embed sign and zero restrictions consistent with economic theory and use Bayesian methodology to estimate the model (Arias et al, 2014 and Dieppe et al, 2016). We work with a sample that covers the post-2001 crisis financial deepening period of the Turkish economy using quarterly data between 2003Q2 and 2018Q2 to estimate the model. Endogenous variables included in the SVAR are the growth rate of the real stock of credit extended by banks, credit spread defined as the spread between the commercial loan rate and 3
  6. the deposit rate , real GDP growth, inflation, monetary policy rate and the exchange rate. Given that Turkey is a small open emerging economy, we also include data on capital inflows to emerging markets as an exogenous variable to control for the effects of global liquidity and global risk appetite. We use a set of timing, zero and sign restrictions on impulse response functions to identify six structural shocks in the six variable SVAR system. These shocks include an aggregate demand shock, an aggregate supply shock, a monetary policy shock, a capital inflow shock, and finally a credit supply shock and a credit demand shock – the last two shocks constituting the focus of this paper. A credit supply shock is defined as a shock that moves credit-deposit rate spread and credit growth in opposite directions on impact, while a shock that affects these variables in the same direction on impact is defined as a credit demand shock. In line with the existing literature, both shocks are assumed to have no significant contemporaneous impact on macroeconomic variables such as inflation, economic growth and policy rate (Barnett and Thomas, 2014 among others).2 Other structural shocks are defined by similar restrictions. Since we are imposing restrictions only on the initial quarter values of the impulse response functions, how these variables respond to these shocks in the consecutive quarters is obtained from the estimation of the SVAR. Impulse response functions obtained from the estimated SVAR indicate that credit supply shocks have a significant effect not only on credit spread and credit growth but also on other macroeconomic variables. Accordingly, a credit supply shock has a significant but temporary effect on output growth whereas it has a more limited and uncertain effect on inflation. The identified credit supply shock seems to be compatible with major shifts in domestic financial conditions, e.g. macroprudential measures, and global financial conditions throughout the sample. The historical variance decompositions reveal that credit supply and credit demand shocks, which are the main drivers of the deviations of credit spread and credit growth from their historical trends, seem to have a smaller role in accounting for the historical deviations in GDP growth. Similar analyses carried out for more advanced economies (such as Busch et al. (2010) for Germany, Barnett and Thomas (2014) for the UK, and Halvorsen and Jacobsen (2014) for Norway and the UK), find a much bigger role for credit supply shocks in accounting for the 2 We allow for an alternative identification scheme that allows credit supply and demand shocks to affect macroeconomic variables on impact as a robustness check, which leads to similar impulses responses as in the baseline identification. 4
  7. historical deviations of GDP growth from its trend – especially during (and after) the global financial crisis. Our results are more in line with Tamási and Világi (2011) where credit supply shocks did not play a dominant role in the 2008 recession in Hungary, which like Turkey, had a much lower credit-to-GDP ratio compared to advanced economies at the time. Apart from being the first paper to identify credit supply shocks for Turkey from a macro perspective, i.e. differentiating credit supply shocks from credit demand shocks and other structural shocks, this paper also contributes to the empirical literature by discussing the role of domestic and global financial conditions on credit supply in a small open emerging economy. We show that a broader definition of a credit supply shock which captures not only domestically driven changes in credit supply, but also the effects of global liquidity and capital inflows on credit conditions has a larger and more significant impact on growth compared to a narrower definition of a credit supply shock. The outline of the paper is as follows: Section II introduces the data and methodology. In section III, we present our main empirical findings including impulse response functions and historical variance decompositions. Section IV displays some additional specifications and robustness checks. Section V summarizes the findings and concludes. II. DATA AND METHODOLOGY To identify different shocks that shift the supply and demand for credit and to analyse their impact on other macroeconomic variables, we use a structural vector autoregression (SVAR) analysis applied to a small set of variables. In a general unrestricted VAR, each endogenous variable is regressed on the lagged values of itself and all the other endogenous variables as well as some exogenous variables. An unrestricted VAR can be represented as: P(L)Et = x + εt (1) where Et includes all endogenous variables, x (or in the case of exogenous time series xt ) includes all exogenous terms, and εt are multivariate normally N(0, I) distributed unrestricted error terms or reduced-form residuals of the VAR. P(L) = I + P1 L + P2 L2 + ⋯ + PN LN is a lagged polynomial, where N is the lag length of the VAR. The underlying structural economic shocks are γt = M −1 εt defined as a linear combination of unrestricted error terms. Therefore, reduced-form residual can be written as εt = Mγt where M is the impact matrix of each structural shock. In order to identify the structural macroeconomic 5
  8. shocks one needs to identify each shock uniquely by applying a set of restrictions to the matrix M . The VAR specification and the identifying restrictions are discussed in more detail in Section II.2. II.1. Data We work with a sample that spans the period between 2003Q2 and 2018Q2, which is undoubtedly a much smaller sample compared to the samples used in similar studies for the US and other advanced economies. One of the main reasons for the use of this sample is that the Turkish economy has gone through a major structural change following the 2001 crisis, which transformed the banking sector as well as the fiscal and monetary policy frameworks. The financial intermediation capacity of banks was quite limited before this period given the high share of government debt in bank assets stemming from the high public borrowing requirement. Monetary policy was not oriented towards achieving an explicit inflation-target but instead involved intermediate targets on exchange rates and monetary base. Inflation targeting regime was adopted in 2002, with the official implementation of inflation targeting starting in 2006. Data limitations also play an important part in our choice of sample as we do not have detailed credit and credit spread data prior to 2002. Given the short sample, we tried to build the most concrete model that can capture the effects of credit supply and demand shocks with a minimum number of variables. The endogenous variables included in the SVAR are the total stock of credit extended by banks, credit spread defined as the spread between the commercial loan rate and the deposit rate, real GDP growth, inflation, monetary policy rate and the exchange rate. Given that Turkey is a small open economy, international capital flows play an important part for macroeconomic dynamics. The exchange rate is included in the SVAR to capture the effects of capital inflows. We also include data on capital inflows to emerging markets as an exogenous variable in the SVAR to control for the effects of external factors such as global risk appetite towards emerging economies. Details of the data are as follows:  We obtain data on the total stock of credit extended by banks from the database of the Central Bank of the Republic of Turkey. Total credit stock includes credit extended to households and to non-financial firms by banks as well as total credit card expenditure.3 The bulk of total loans are loans extended to firms, with an average share of 71 percent 3 Non-bank financial intermediation is negligible in Turkey except for a few items such as loans for vehicles. 6
  9. in total credit throughout the sample . Consumer loans are denominated mostly in Turkish lira due to regulations that prohibit household borrowing in FX4, while a significant portion of firm loans are in FX, i.e. around 47 percent in total firm loans throughout the sample. When summing up domestic and foreign currency loans to arrive at the total credit stock figure, FX loans are converted into domestic currency using a fixed exchange rate to prevent changes in exchange rate to inflate the credit series and make them too volatile. 5 We deflate nominal credit data using the Consumer Price Index (CPI, 2003=100). The series are seasonally adjusted. Given that total real credit series has a unit root, we use the quarterly difference of logged total real credit in the SVAR.  Credit spread is defined as the difference between the commercial loan rate and the deposit rate averaged over a quarter. This is an indicator that reflects the relative price of loans as well as being a measure of overall tightness in credit conditions.6 The majority of commercial loans and deposits are of relatively short maturity in Turkey, while consumer loans are typically extended for longer maturity. Hence, the spread between the commercial loan rate and the deposit rate reflects less of a maturity mismatch compared to the spread between the consumer loan rate and deposit rate. Besides, commercial credit constitutes the largest part of total credit. We carry out robustness checks using a weighted average of consumer and commercial loan rates when calculating the spread. The credit spread series is stationary, hence no transformation is used when including in the SVAR.  Given the change in the monetary policy framework in the aftermath of the global financial crisis, we combine different policy rates that were relevant in the pre- and post2010 periods. Prior to 2010 the financial system was in a net liquidity surplus and the overnight borrowing rate was the policy rate. As of May 2010, the CBRT started providing weekly funding to the financial system through one-week repo auctions in addition to overnight lending. In this period, the amount of funding provided through each liquidity instrument varied according to the intended monetary policy stance, and the BIST overnight rate fluctuated within a wide interest rate corridor.7 In order to account for the periods where the BIST overnight rate significantly differed from the 4 By law in October 2009, household’s foreign currency denominated credit usage is prohibited. The average exchange rate for the currency basket (weighted average of euro and USD credit, with 30 and 70 percent weights, respectively) is fixed at the average value of the currency basket between October 2008 and April 2011. 6 See Box 5.2 in Central Bank of Turkey Inflation Report 2016 IV. 7 See Küçük, Özlü, Talaslı, Ünalmış and Yüksel (2016) for a discussion about the role of liquidity policy in explaining the spread between the BIST overnight rate and the CBRT average funding rate in this period. 5 7
  10. CBRT average funding rate due to the liquidity policy of the CBRT , we take the BIST overnight rate as the policy rate for the period after 2010Q2.8 Figure 2: Data used in the baseline analysis 30 20 7 Nominal Exch. Rate 25 Total Real Credit 15 20 15 6 Credit Spread 5 10 10 4 5 5 3 0 0 2 -5 -5 5 1 -10 8 CPI-D 4 45 Real GDP 6 40 Nominal Policy Rate 35 4 3 0 2003q2 2004q2 2005q2 2006q2 2007q2 2008q2 2009q2 2010q2 2011q2 2012q2 2013q2 2014q2 2015q2 2016q2 2017q2 2018q2 2003q2 2004q2 2005q2 2006q2 2007q2 2008q2 2009q2 2010q2 2011q2 2012q2 2013q2 2014q2 2015q2 2016q2 2017q2 2018q2 -15 2003q2 2004q1 2004q4 2005q3 2006q2 2007q1 2007q4 2008q3 2009q2 2010q1 2010q4 2011q3 2012q2 2013q1 2013q4 2014q3 2015q2 2016q1 2016q4 2017q3 2018q2 -10 30 2 25 2 0 20 -2 15 0 -4 10 -1 -6 5 0 2003q2 2004q2 2005q2 2006q2 2007q2 2008q2 2009q2 2010q2 2011q2 2012q2 2013q2 2014q2 2015q2 2016q2 2017q2 2018q2 2003q2 2004q1 2004q4 2005q3 2006q2 2007q1 2007q4 2008q3 2009q2 2010q1 2010q4 2011q3 2012q2 2013q1 2013q4 2014q3 2015q2 2016q1 2016q4 2017q3 2018q2 2003q2 2004q2 2005q2 2006q2 2007q2 2008q2 2009q2 2010q2 2011q2 2012q2 2013q2 2014q2 2015q2 2016q2 2017q2 2018q2 1 15 Global Liquidity Indicator 10 5 0 -5 2003q2 2004q1 2004q4 2005q3 2006q2 2007q1 2007q4 2008q3 2009q2 2010q1 2010q4 2011q3 2012q2 2013q1 2013q4 2014q3 2015q2 2016q1 2016q4 2017q3 2018q2 -10 Note: All series except for the credit spread and the nominal policy rate are logged and first differenced.  Considering the sensitivity of financial and macroeconomic variables to capital flows and the exchange rate in Turkey, we include the nominal exchange rate in the SVAR. Nominal exchange rate is not only a broad proxy for capital inflows but also reflects the domestic component of the risk premium as well as the general sentiment in the economy.9 We use quarterly average of USD/TRY exchange rate (a rise in USD/TRY exchange rate is a depreciation of Turkish lira) logged and first differenced to render a stationary series (Table 1). As a robustness check we also estimate the model using the CPI-based real effective exchange rate (2003=100) published by the CBRT. 8 We carry out robustness checks using the CBRT average funding rate. Results are not much affected but the BIST rate provides narrower credibility intervals especially for impulse responses to a monetary policy shock. 9 Karasoy and Yüncüler (2015) show that consumer confidence is strongly correlated with the USD/TRY exchange rate in Turkey. 8
  11.  In addition, we added a global liquidity indicator as an exogenous variable to our SVAR model in order to control for the effects of global conditions (push factors) on capital inflows. This is important since we know that credit conditions in Turkey are closely linked to global liquidity conditions through their effects on capital flows. We used “Global Liquidity Indicators” dataset published by BIS to calculate the change in total US and euro credit flows to emerging countries. This variable is also logged and first differenced. As an alternative global liquidity indicator, we employed total balance sheet size of the Fed and the ECB and checked that the results are quite similar. We also carried out estimations with oil prices included as an exogenous variable but the results were affected only very little. Given the short sample, we decided not to include this variable in the baseline SVAR.  The source of real GDP (2009=100) and the CPI (2003=100) is Turkstat. We use a measure of core CPI, i.e. CPI excluding unprocessed food, alcoholic beverages and tobacco (CPI-D), which accounts for nearly 84 percent of the headline CPI. This choice is due to the fact that unprocessed food and tobacco prices exhibit short-term noisy movements and have the highest unexpected volatility among the CPI subcomponents in Turkey. Both series are seasonally adjusted, logged and first differenced. The data we have used in the analysis can be seen in Figure 2. II.2. Identification of Structural Shocks A set of timing, magnitude (zero) and sign restrictions is used to identify six structural shocks in the six variable SVAR system. As for the timing of the restrictions, all restrictions apply to the initial period, the quarter in which the shock occurs. Shocks are identified with the zero and sign restrictions imposed within the initial quarter. Where there are no contemporaneous restrictions imposed (where the cells in Table 2 are left blank), impulse responses are determined agnostically, i.e. determined by the estimated model. This is also true for all impulse responses following the initial quarter. Four of these shocks, aggregate supply, aggregate demand, monetary policy and capital inflow (exchange rate) shocks are the standard aggregate shocks for a small open economy. Additional two shocks, credit supply and credit demand shocks, are the shocks related with the credit market and financial system. Special focus will be given to these last two shocks in this paper. The underlying restrictions, which are summarized in Table 2, can be described as follows: 9
  12.  An aggregate supply shock is assumed to be a shock which moves inflation and real GDP growth in opposite directions within the same period. For example, a negative aggregate supply shock might reflect a fall total factor productivity or a rise in oil prices.  An aggregate demand shock is assumed to be a shock which moves inflation and real GDP growth in the same direction contemporaneously. Moreover, it is assumed that monetary policy responds to an aggregate demand shock within the same quarter, i.e. policy rate is raised following a positive aggregate demand shock that raises both inflation and output growth.  A negative shock to the policy rate (a surprise fall in the policy rate) is a shock that depreciates Turkish lira, leads to a rise in inflation and a fall in real GDP growth within the same quarter.  A negative (positive) capital inflow shock is a shock that depreciates (appreciates) the Turkish lira, reflected as a rise (fall) in USD/TRY exchange rate. Furthermore, in light of the evidence regarding the exchange rate pass-through to inflation in Turkey (Kara and Öğünç, 2005) it is assumed that a nominal depreciation (appreciation) leads to a rise (fall) in inflation within the same quarter.  A credit supply shock would typically lead to an opposite movement between credit spread and lending. Thus, a negative credit supply shock rises the credit spread while reducing credit growth within the same quarter. Moreover, it is assumed that the contemporaneous effect of the credit supply shock on inflation, growth and policy rate are negligible within the same quarter. In other words, the identification given in Table 2 imposes that credit supply shocks affect inflation, growth and policy rate with a onequarter lag. A negative credit supply shock may be due to an unanticipated fall in lending appetite, which may result from tighter macroprudential policies that increase the cost of lending for banks, or banks’ heightened risk perceptions regarding the economic outlook or collateral values, or higher liquidity constraints of banks etc.  A credit demand shock is assumed to move credit spread and credit growth in the same direction, i.e. a positive credit demand shock leads to a rise in credit spread as well as an increase in credit growth. As in the case of the credit supply shock, it is assumed that a credit demand shock affects inflation, growth and policy rate with one-quarter lag. Given that we have separate shocks like aggregate demand and aggregate supply which 10
  13. might also affect credit demand ; it might not be easy to imagine what would constitute a credit demand shock. Given that Turkey is still in a relatively early phase of financial development, we interpret this shock as a change in credit seekers’ access to credit. This might stem from changes in the search technology of credit seekers (e.g. increase in the number of people that have a credit card) or macro-micro financial regulations that affect access to credit (e.g. reduction in the penalty rate of overdue credit payments or an increase in the number of instalments on a credit card) may increase credit demand. Search technology and financial regulations may also affect credit supply as well. This is one of the reasons why we identify both credit supply and credit demand shocks. Table 2: The shocks and restrictions in SVAR Shocks Variables* Capital Inflow shock Exchange Rate - Credit demand shock Credit supply shock Aggregate Aggregate demand supply shock shock Monetary policy shock + Credit + - Spread + + 0 0 + + + 0 0 - + + 0 0 + - Inflation Growth Policy Rate   - *Exchange Rate: Quarterly change in USD/TRY (negative implies nominal appreciation of Turkish lira). Credit: Quarterly growth rate of real, seasonally adjusted total credits. Spread: Quarterly average spread between the commercial loan rate and deposit rate. Growth: Quarterly GDP growth. Policy Rate: CBRT Policy Rate All restrictions apply to initial period of the shock (t=0). (+) values imply that shock will affect the corresponding variable positively, (-) values imply that shock will affect the corresponding variable negatively. (0) means there will be no simultaneous effect of the shock on the corresponding variable. Blank cells imply that there are no restrictions for the particular shock-variable combination. The baseline restrictions in Table 2 may be challenged on various grounds, in particular on the assumptions required to identify the credit market shocks, which are the focus of this paper. One issue is about our assumptions regarding the effects of capital inflow shocks on the credit market. Given that external finance is an important source for the banking sector in Turkey, it might be reasonable to assume that a capital inflow shock that appreciates the nominal exchange rate lowers credit spreads and expands credit growth at the same time, as would a positive shock to credit supply do. However, imposing these extra sign restrictions on credit variables would imply that the capital inflow shock would also act as a credit supply shock and that the definition of the credit supply shock would be narrowed down to a “domestic credit supply shock”. Our view is that both identifications offer interesting insights. We prefer to stick to the “broader definition” of credit supply shocks in the baseline identification depicted in Table 2, which would include the effect of capital flows on credit conditions as well. There is an even broader definition than we provide in the baseline, where we remove the exogenous global liquidity variable from the model, which in turn implies that changes in global liquidity are also attributed 11
  14. to a credit supply shock. We explore this alternative identification of ‘narrower’ and ‘broader’ credit supply shocks in more detail in Section III.2 to highlight the role of external financing conditions on credit supply. A second challenge to the set of restrictions in Table 2 might be related to our assumption of no contemporaneous effect of credit market shocks on real GDP, inflation and policy rate. These identification restrictions are widely used in this line of research as in Barnett and Thomas (2014). These contemporaneous zero restrictions are critical to identify credit supply and demand shocks in a way that differentiates them from aggregate demand and supply shocks which might affect credit spread and credit growth for reasons unrelated to the credit market. Although crucial for identification, these contemporaneous zero restrictions might also imply that we are identifying credit supply and demand shocks that are not large enough to trigger a financial accelerator mechanism as in Bernenke, Gertler and Gilchrist (1999). In that sense the identified credit market shocks in this set-up do not capture large credit market shocks that have an immediate significant effect on the total number of credit restricted firms or households and hence on growth, inflation and policy rate.10 II.3. Some Details of the Methodology In order to impose a combination of zero and sign restrictions necessary to identify the shocks, we use standard and well-established techniques and employ the BEAR toolbox (Dieppe et al 2016). In particular, the procedure introduced by Arias, Rubio-Ramirez and Waggoner (2014) is used. Briefly11, this methodology first picks an impact matrix