Master Thesis Credit Growth Sensitivity to the Global Financial Cycle: a New Approach to Monetary Policy Independence

(1)

Credit Growth Sensitivity to the Global Financial Cycle: a New

Approach to Monetary Policy Independence

Master Thesis

International Economics and Business (Double degree, M.Sc. & M.A.) University of Groningen

Corvinus University of Budapest

Abstract

This thesis introduces a new approach to testing monetary policy independence in the context of the traditional monetary policy trilemma. Indeed, credit growth sensitivity to the Global Financial Cycle is used as a de facto measure of monetary policy independence. This approach builds on the assumption that monetary policy transmission works through both the interest rate channel and other transmission channels (i.e. credit channel). Credit growth is the effective outcome of the transmission, thus conveying information on both the direct effect of the interest rate channel and of the indirect amplification mechanism of other transmission channels. Although I find compelling evidence of a Global Financial Cycle of credit growth, sample limitations do not allow me to draw consistent conclusions about the validity of the trilemma. Moreover, the analysis suggests that credit to households and credit to non-financial corporations show different sensitivities to the Global Financial Cycle.

Key Words: Credit growth, trilemma, Global Financial Cycle, monetary policy independence

Name: Antonella Innocenzio Student Number: S3219054

(2)

(3)

i

1. Introduction

In an increasingly integrated financial world, evidence has emerged of the existence of a Global Financial Cycle, largely driven by the monetary policy of the United States (Miranda-Agrippino & Rey, 2015; Rey, 2013). As such, the monetary policy of a center country (i.e. the U.S.) determines the leverage of global banks, capital flows, and thus credit growth abroad. In turn, this has remarkable monetary policy implications for non-center countries: through capital flows, countries absorb the monetary conditions of the center economy despite flexible exchange rates. That is, in an integrated financial world, capital flows reduce the space for an independent monetary policy. Studying this effect, Rey (2013) provocatively sanctions the end of the traditional monetary policy trilemma. Since the transmission operates through capital flows, the trilemma morphs into a dilemma: an independent monetary policy is possible only amid capital controls, regardless of the exchange rate regime (Rey, 2013). Understandably, this conclusion spurred a debate among researchers, with many intervening to test the trilemma hypothesis. In this context, most of the literature focused on interest rate divergence with the financial centers as a measure of monetary policy independence (Aizenman, Chinn, & Ito, 2016; Kharroubi & Zampolli, 2016; Obstfeld, 2015; Ricci & Shi, 2016). As it is, long-term interest rates work in accordance with the trilemma and can be set more independently amid flexible exchange rates. Yet, short-term rates divergence is not enhanced by having flexible exchange rates.

(6)

2 This thesis aims at contributing to the line of research focusing on the consequences of the Global Financial Cycle on monetary policy independence by testing whether domestic credit in each country follows credit trends in the financial centers1. The assumption is that monetary policy in the financial centers works through both the traditional interest rate channel and the credit channel. In turn, shifts in the center countries’ conditions will spread internationally through the international credit channel, not just the interest rate channel. Thus, the Global Financial Cycle might cause disruptions in the domestic credit markets even in the absence of significant shifts in the domestic interest rates. In other words, changes in credit convey information on both the direct effect of the interest rate channel and of the indirect amplification mechanism of other transmission channels. As such, credit is an indirect measure of monetary policy shifts in the center countries and of their impact on every other country. Under these assumptions, I compute coefficients of sensitivity of credit growth in each country to credit growth in the financial centers. In a second phase, these coefficients are employed as a measure of monetary policy independence to test whether the traditional Mundellian hypothesis still holds. Finally, I test whether credit sub-components (credit to households and credit to non-financial corporations) show different sensitivities to credit growth in the non-financial centers. In fact, different sensitivities could prove material in determining the appropriate policy response. The research is innovative in two main components: i) the use of credit growth sensitivity to the Global Financial Cycle to estimate of monetary policy independence; and ii) the test of credit sub-components’ sensitivity to the Global Financial Cycle.

The rest of the analysis is structured as follows. Section 2 explores the most relevant literature and presents the formal hypothesis of this research. Section 3 describes the model specification, methodology and data in details. In section 4, I compute the coefficients of sensitivity. In section 5, the coefficients are used as dependent variable of monetary policy independence to test the trilemma hypothesis. Section 6 shows the results for credit sub-components. Finally, section 7 highlights research limitations and draws some key conclusions.

1_{The four countries considered as world financial centers in this thesis are China, Japan, the United Kingdom, and}

(7)

3

2. Literature review and hypotheses

This thesis builds on the recent line of research on financial integration and the Global Financial Cycle. For this reason, it is important to explore the origins of the concept of financial cycle and its main characteristics. In addition, I present in this section the work of Rey on the existence of a Global Financial Cycle and on its monetary implications, together with several authors’ tests related to the same hypothesis. Finally, I draw some key conclusions from the analyzed literature and present the formal hypotheses.

2.1 The (re)discovery of the Financial Cycle

“We thought we knew; we have since forgotten. It is high time we rediscovered the role of the financial cycle in macroeconomics” (Borio, 2013).

The concept of financial cycle is far from new to modern economists. In fact, Borio (2011) traces it back to the 19th_{century – the Age of the Gold Standard – when “financial crises were}

seen as a phase naturally linked to the business cycle” (p.22). For instance, Overstone & McCulloch (1857) introduce the “convulsion” phase of the business cycle, which effectively represents a stage of financial distress. Authors of the like of Wicksell (1936), Fisher (1932), Von Mises (1912) and Hayek (1933) all highlighted the role of money and credit expansions (or contractions) in the business cycle. However, since then and for most of the 20th_century,

traditional economic theory underplayed financial factors. According to Calvo (2013), mainstream macroeconomists in this phase adopted, perhaps unconsciously, the view of “finance as a veil”2_{(p. 1), thus ignoring crises that stem from the financial sector. Consequently,}

a possible malfunctioning of the credit market was also overlooked by policymakers. Things changed when the Global Financial Crisis of 2008 brought credit and finance back on the main stage. Soon, unconventional thinkers that had highlighted the role of credit were revived. Among these, first and foremost Minsky (1982), who had warned against the inherent financial instability of capitalist economies. Underplayed and criticized by his contemporaries, Minsky introduced a theory on the inevitability of financial bubbles and bursts that was understandably revived in the wake of the subprime mortgage crisis. In fact, the author had pointed out the pro-cyclical nature of asset prices, which rise amid rising demand (and vice versa) until an asset

2 _{Calvo (2013) draws a parallel with the classical view of ‘money as a veil’, which has however been debunked by}

(8)

4 bubble is formed. Inevitably, the euphoria surrounding the bubble eventually comes to an end. As soon as an adverse event pushes people to sell, panic explodes and asset prices plunge in the typical bubble burst. Liquidity and credit crunches had also been overlooked for years after the Great Depression. In fact, although severe liquidity crunches had characterized the 1930s financial panic, mainstream economists thought it unreasonable to think that it could happen again3. When liquidity crunches severely hit the world economy in the wake of the financial crisis, economists rediscovered their existence and the dangers related to them.

So where does the rediscovery of financial bubbles and liquidity crunches lead us? What have we learnt? In the attempt to answer this question, Borio (2012) has looked at the main existing research on the financial cycle. First, he extrapolates a practical definition of financial cycle: “self-reinforcing interactions between perceptions of value and risk, attitudes towards risk and financing constraints, which translate into booms followed by busts” (Borio, 2012, p. 2). Clearly, the focus is on pro-cyclicality, as Minsky himself had noted. In fact, there is a positive feedback loop between asset prices and credit growth. Credit expansions push asset prices. In turn, higher asset prices make for better balance sheets collaterals, feeding the credit boom itself (Borio, Furfine, & Lowe, 2001). Consistently, empirical studies have found domestic credit expansions to be the most robust predictor of financial crises (Gourinchas & Obstfeld, 2012; Schularick & Taylor, 2012). Moreover, empirical research has confirmed that in a boom-bust dynamic the composition of capital flows matters. That is, non-FDI capital flows (i.e. bank lending and portfolio flows) are much more volatile, thus increasing the likelihood of a sharp reversal (Wei, 2002). In this sense, Tong & Wei (2011) find evidence that the severity of credit crunches during the financial crisis was linked with non-FDI skewed composition of capital flows before 2008. Similarly, Calderon & Kubota (2012) found that certain components of credit inflows – i.e. private credit – are more likely to drive boom dynamics, especially in the housing market. This line of research suggests that certain components of capital flows and credit should be more thoroughly monitored when monitoring financial distress.

Given these empirical results, it appears clear that the importance of recognizing a financial cycle lies in the possibility of measuring it empirically to detect financial distress risks (Drehman, Borio, & Tsatsaronis, 2012). Drehman et al. (2012) also compare the financial cycle

3_{For instance, James Tobin (1989) deemed a liquidity crunch impossible to repeat. In his words, “[…] if a}

(9)

5 with the business cycle and note that the financial cycle has a much lower frequency. In fact, while the business cycle involves frequencies from 1 to 8 years, the average length of the financial cycle has been around 16 years since the 1960s4. There is, however, a link between the two cycles. According to Borio et al. (2001) the build-ups of the financial cycle amplify fluctuations in the real economy. Several empirical works confirm this hypothesis. Drehman et al. (2012), for example, find that “business cycle recessions are much deeper when they coincide with the contraction phase of the financial cycle” (p. iii). Similarly, Claessens, Kose, & Terrones (2012) analyze the interaction between business and financial cycles and find that recessions caused by financial factors tend to be longer and deeper.

2.2 Financial integration and the Global Financial Cycle

2.2.1 Financial integration

While rediscovering the importance of credit and of the financial cycle, economists were also faced with a new key evolution: financial markets integration. In fact, gross capital flows went through a spectacular expansion since the end of the 1990s, with both emerging markets and industrialized economies opening up to capital flows.

Figure 2.1. Total cross-border assets as a percentage of world GDP

Source: IMF International Investment Position Statistics, own calculations5_.

4_{Drehman, Borio, & Tsatsaronis (2012) use a sample of seven industrialized countries.}

5_{Countries classification follows Lane and Milesi-Ferretti (2003). Industrialized: AUS, BEL, CAN, CHE, DEN,}

FIN, FRA, GER, ISL, ISR, ITA, JAP, NLD, NOR, POR, SPA, SWE, GBR, USA. Emerging: ARG, BWA, BRA, CHL, CHN, COL, CZE, EST, HUN, IND, KOR, KGZ, LVA, LTU, MEX, MDA, NER, PAN, PRY, PER, POL, ROU, RUS, SEN, SVK, SVN, ZAF, THA, TUN, TUR, VEN.

(10)

6 Figure 2.1 reproduces a volume-based measure of international financial integration6 as constructed by Lane and Milesi-Ferretti (2003). The graph shows that, at a world level, accumulated cross-border assets as a percentage of GDP have doubled between 1995 and 2007, stabilizing at current levels since after the Global Financial Crisis. However, this expansion was much larger for industrialized countries than emerging ones. Intuitively, this level of financial integration creates interdependencies – especially among industrialized economies – contributing to the spreading of financial conditions worldwide.

2.2.2 The Global Financial Cycle

In the literature, authors have begun to notice the role of global factors as determinants of financial conditions since the 1990s. Notably, Calvo, Leiderman, & Reinhart (1996) suggested that the 1990s surge in capital flows to developing countries could not be due to solely domestic developments. In fact, the phenomenon was widespread and affected countries with very diverse characteristics and performance. In this sense, the authors distinguished between the global “push” factors, e.g. the cyclical movement of interest rates, and domestic “pull” factors due to specific country characteristics.

More recently, the focus has shifted on how credit and monetary conditions in financial centers, i.e. the United States, are transmitted to the rest of the world. Hereof, the financial cycle seems to be driven by the embedded interrelations of the monetary conditions of the center country, capital flows, and the banking sector leverage. This effect became evident during the Global Financial Crisis (IMF, 2011) and was analyzed by Bruno & Shin (2013) in their investigation of global factors associated with cross-border capital flows. The author’s findings confirm that the banking sector capital flows, which determine the transmission of financial conditions across borders, are mostly driven by global conditions rather than local ones.

On this topic, of particular importance is the work of Rey, who introduced the term “Global Financial Cycle” to describe the worldwide “large common movements in asset prices, gross flows and leverage” (Rey, 2013, p. 286). In her paper “Dilemma Not Trilemma” (2013), Rey builds on previous literature and offers compelling evidence on the existence, functioning and determinants of a Global Financial Cycle. Initial evidence suggests that there is a significant

6_{𝐼𝐹𝐼𝐺𝐷𝑃} 𝑖𝑡=

(𝐹𝐴𝑖𝑡+𝐹𝐿𝑖𝑡)

(11)

7 commonality in capital inflows and outflows across geographical regions and asset classes7. Moreover, Rey emphasizes the association of surges in capital flows – with the exception of FDI - with the lowering of the VIX, a phenomenon already tested by Bruno & Shin (2013). Increases in credit growth around the world and in leverages in all the main financial centers also move in parallel with the VIX. In this sense, the VIX works as a proxy for risk aversion and uncertainty in the financial centers. Furthermore, data shows that prices for a large cross-section of risky assets around the world are largely explained (25%) by one main global factor, i.e. the risk aversion of global banks, which is related to the VIX (Miranda-Agrippino & Rey, 2013). Thus, Rey concludes that not only there is a Global Financial Cycle driving capital flows and asset prices, but that the cycle is also strongly synchronized with fluctuations of the VIX. Additionally, Rey’s tests consistently find that credit growth is more sensitive to the Global Financial Cycle than other flows (i.e. equity). Thus, given the role of credit growth in the build-up of systematic risks (see section 2.1; Gourinchas & Obstfeld, 2011; Schularick & Taylor, 2012), its enhanced sensitivity to the Global Financial Cycle further highlights the importance of Rey’s findings.

To sum up, gross capital flows, asset prices, leverage and credit creation co-move in sync with fluctuations of the VIX. The next step is to analyze the determinants of the cycle. A two lag VAR analysis offers results that are consistent with the following interpretation: “When the federal funds rate [in the U.S.] goes down, the VIX falls (after about five quarters), European banks’ leverage rises, as do gross credit flows (after 12 quarters). A fall in the VIX leads to an increase in global domestic credit after four quarters” (Rey, 2013, p. 304). Moreover, expanding credit compresses the spread, thus translating into reduced uncertainty measured by a further decline in the VIX. This results link to research by Bekaert, Hoerova, & Lo Duca (2013) and Bruno & Shin (2013) showing an interaction between the federal funds rate and the VIX, through a surge in capital flows. Hence, the key take is that the monetary policy of the center country, i.e. the federal funds rate, is a crucial determinant of the Global Financial Cycle. In turn, this suggests that a loose monetary policy in the center state can spur credit growth around the world, and thus necessarily in countries where macroeconomic conditions would not favor

7_{Refer to Rey (2013) for heat maps of correlations of capital flows by asset classes. Although few correlations are}

(12)

8 such an expansion. In this sense, the existence Global Financial Cycle has clear monetary policy implications.

2.3 Monetary policy implications of the Global Financial Cycle

2.3.1 “Dilemma not trilemma”

Section 2.2.3 highlights what is perhaps the most important finding of Rey: the monetary policy of the center country is a key determinant of the Global Financial Cycle by ways of affecting the leverage of global banks, capital flows and thus credit growth. This effect has remarkable monetary policy implications, in that it shows that countries absorb the monetary conditions of the center country despite flexible exchange rates. That is, in an integrated financial world, capital flows reduce the space for an independent monetary policy. Thus, Rey (2013) provocatively sanctions the end of the monetary policy trilemma, according to which countries can only operate an independent monetary policy if they introduce capital controls or a flexible exchange rate regime (Mundell, 1963). In other words, Rey suggests that a floating exchange rate regime no longer insulates economies from the transmission of foreign monetary policy across borders. Since the transmission operates through capital flows, the trilemma morphs into a dilemma: an independent monetary policy is possible only amid capital controls and regardless of the exchange rate regime (Rey, 2013). Understandably, Rey’s (2013) provocative hypothesis of an “irreconcilable duo” (p. 287) dismissing the traditional Mundellian trilemma spurred a debate among researchers. In the next two sections, I explore the main fields of the current research on the effects of financial integration on monetary policy.

2.3.2 Interest rate independence and exchange rate regime

(13)

9 regimes (or amid capital controls). In other words, it seems that it may be too soon to dismiss the traditional monetary policy trilemma.

2.3.3 Beyond interest rates

As the debate on the standing validity of the monetary policy trilemma emerged, one issue became central: the definition of monetary policy independence. In fact, by testing interest rates, Aizenman et al.(2016), Kharroubi & Zampolli (2016), Obstfeld (2015), and Ricci & Shi (2016) all apply the strictest definition of monetary policy independence, i.e. independence in setting interest rates. Thus, their conclusions are limited to the effects of the cross-border transmission of monetary conditions through interest rates. However, evidence has emerged that interest rates may not be telling the whole story. For instance, since the crisis, developed countries have struggled to revive their economies and credit markets amid very low interest rates. Unconventional monetary policies such as quantitative easing, forward guidance and signaling effectively represent central banks’ attempt to operate monetary policy at a time when interest rates could not be lowered further. In this sense, policymakers have tried to leverage transmission channels other than the traditional interest rate channel. In fact, a vast literature has testified the existence of secondary channels of monetary policy transmission. In particular, as accurately surveyed by Minshkin (1996), monetary policy also acts through the so-called credit channel, which builds on the problem of asymmetric information in credit markets. That is, an expansionary monetary policy, which increases bank reserves and deposits, increases the quantity of loans available. Moreover, a monetary policy expansion causes a rise in equity prices, thus raising the net worth of firms, reduces nominal interest rates, generating an improvement of firms’ balance sheets, and improves their access to credit. Thence, the credit channel is an indirect amplification mechanism for monetary policy (Bernanke & Gertler, 1995). The existence of transmission channels other than the interest rate channel suggests that monetary policy independence does not strictly coincide with independence in setting interest rates. As a matter of fact, a broader definition of monetary policy independence should look at independence in monetary policy transmission through all possible channels.

2.3.4 Independent monetary policy transmission: mortgage spreads

(14)

10 retaining a certain independence in the setting of interest rates, countries might still inherit external monetary conditions through the credit channel. Indeed, Rey (2016) tests this hypothesis by looking at measures of the external finance premium, defined as the difference between the cost of raising funds externally and the opportunity cost of internal funds8 (Bernanke, 2007). Preliminary findings by Rey (2016) show that the external finance premium – as measured by mortgage spreads – responds to changes in the US monetary policy. That is, the cost of external finance reduces in response to cuts to the Federal Funds Rate, despite flexible exchange rates. This result suggests that there might be an “international credit or risk-taking channel of transmission” (Rey, 2016, p. 10) of US monetary policy. However, this findings cannot be considered definitive, as Rey (2016) only offers the results of four VAR analysis for as many developed countries (Canada, New Zealand, Sweden, United Kingdom). Moreover, the author suggests that further research should look at other measures of the external finance premium (i.e. corporate spreads, risk term premium). In fact, although mortgage spreads are of relevance due to the peculiar role of mortgages in boom-bust dynamics, the effects of the transmission channel as such cannot be confirmed by looking at only one type of credit. 2.4 Key takeaways and hypotheses

The relevant literature presented in section 2.1, 2.2 and 2.3 highlights four key takeaways that introduce and support this research. In this section, I offer a brief summary of these findings and introduce the formal hypotheses.

2.4.1 Key takeaways

The review of the literature presented so far highlights four main takeaways for the purposes of this research.

Takeaway 1a: Credit occupies a central role in determining financial stability.

In fact, section 2.1 highlights how the Global Financial Crisis has led to a rediscovery of the financial cycle, which is mostly led by credit dynamics. This process refers to the boom-bust cycle as described by Minsky (1998). Thus, the dangerous feedback loop between credit growth and asset prices is such that any research around financial cycles cannot but highlight the central role of credit.

8 _{In theory, external finance (raising funds from lenders) is more expensive than internal finance (using generated}

(15)

11 Takeaway 1b: Consumer and mortgage credit drive financial instability more often than other types of credit.

In a boom-bust dynamic the composition of capital flows matters in that certain components (i.e. mortgage and consumer credit) are subject to worse sudden stops (Tong & Wei, 2011) and determine “bad” credit booms (Calderon & Kubota, 2012).

Takeaway 2: Financial integration has generated an integrated Global Financial Cycle. The impressive accumulation of cross-border positions in capital markets has markedly changed the workings of the financial cycle. As such, a financial boom in a base country (typically, the United States) fuels cross-border capital flows and consequently credit supply abroad (Section 2.2). This effect is mediated by in a decline in volatility and in the perceived risk, which in turn feeds a self-reinforcing cycle leading to increased risk taking and more capital inflows (Bruno & Shin, 2013). Thus, financial markets are now characterized by a significant global co-movement of asset prices, gross flows and leverage (Rey, 2013).

Takeaway 3: Flexible exchange rates still guarantee countries’ ability to set short term interest rates independently.

The most significant finding of Rey’s (2013) work is that the monetary policy of a center country (i.e. the United States) is a significant determinant of such Global Financial Cycle. This raises concerns about the possibility of leading an independent monetary policy in such a financially integrated world. According to Rey (2013), countries absorb the monetary conditions of the center country despite flexible exchange rates. However, several authors (see paragraph 2.3.2) test this hypothesis and find that this effect is only true for long term interest rates, thus preserving the possibility of setting interest rates independently in the short run. As mentioned, this result confirms that monetary policy independence is partially preserved in its strictest definition: the setting of interest rates.

Takeaway 4: The Global Financial Cycle might be more effective in disrupting monetary transmission channels than the interest rates themselves.

(16)

12 external finance reduces in response to cuts to the Federal Funds Rate, despite flexible exchange rates.

2.4.2 Hypotheses

This thesis aims at contributing to the line of research focusing on the consequences of the Global Financial Cycle on monetary policy independence. As mentioned, Rey’s work has spurred a true debate in the literature, with authors of the like of Obstfeld (2015) intervening to test Rey’s hypotheses. However, most of the authors have found that a dismissal of the traditional monetary policy trilemma may be premature, in that exchange rate flexibility still guarantees some independence in setting short term interest rates. In her latest response, Rey (2016) explicitly broadens the definition of monetary independence to monetary transmission channels and introduces preliminary tests using mortgage spreads as a proxy for the effects of monetary transmission (Paragraph 2.3.4). This thesis builds on this extended approach to monetary policy independence. The aim is to test whether domestic credit in each country follows credit trends in the world financial centers.

The assumption behind this research question is that monetary policy in the financial centers works both through the traditional interest rate channel and the credit channel. In turn, shifts in the center countries’ conditions (i.e. credit) will spread internationally through the international credit channel, and not just through the interest rate channel. Thus, the Global Financial Cycle might cause disruptions in the domestic credit markets even in the absence of significant shifts in the domestic interest rates. In other words, changes in credit convey information on both the direct effect of the interest rate channel and of the indirect amplification mechanism of other transmission channels. As such, credit is a better estimator of monetary policy shifts in the center countries and of their impact on the other countries of the world.

Under these assumptions, exchange rate flexibility may not be sufficient to insulate countries from the international transmission channels – thus confirming Rey’s (2013) theory on the dismissal of the traditional monetary policy trilemma. At the same time, capital controls should prove to be the only effective tool in isolating domestic monetary conditions (Rey, 2013). Moreover, I expect that credit sub-components (credit to non-financial corporations and households) may show different sensitivities to the Global Financial Cycle. Interestingly, different sensitivities could prove material in determining the appropriate policy response, i.e. distinct capital restrictions.

(17)

13 Hypothesis 1: Credit growth in each country is determined by the Global Financial Cycle.

Domestic credit growth in ti is correlated with credit growth in the main financial

centers in ti-x (lagged effect). (H1)

Hypothesis 2: The Mundellian trilemma hypothesis is not adequate anymore: countries with flexible exchange rates are as sensitive to the Global Financial Cycle. Capital controls remain the only tool to isolate domestic monetary policy.

The sensitivity of domestic credit growth to credit growth in the main financial centers

is not correlated with the exchange rate regime. (H2a)

The sensitivity of domestic credit growth to credit growth in the main financial centers is positively correlated with capital account openness. (H2b) Hypothesis 3: Credit sub-components have different degrees of sensitivity to the Global Financial Cycle.

The difference between the growth rate of credit to households and credit to non-financial corporations is correlated with credit growth in the main non-financial centers in ti-x.

(H3a) Credit to households’ sensitivity to the Global Financial Cycle is larger than the sensitivity of total credit to the private non-financial sector.

(18)

14

3. Methodology and data

In this section, I explain the choice of methodology employed in this research, its formal specification and variable definition. Following Aizenman et al. (2016) and Forbes & Chin (2003), the analysis is developed in two steps. First, I compute a set of sensitivity coefficients of domestic credit growth to credit growth in the main financial centers, controlling for global and domestic factors (Step 1). These coefficients are used as a measure of monetary policy independence. Hence, I decompose the sensitivity coefficients into their explanatory components (Step 2). In other words, I test which conditions better explain such coefficients. In particular, this model allows me to disentangle the effect of the exchange rate regime and capital account openness on the sensitivity of credit growth (Aizenman et al., 2016). In addition to these two steps, I decompose credit into two sub-components. The specification and methodology for testing credit sub-components is explained in Section 6. Detailed data sources are listed in Appendix 2.

3.1 Choice of methodology

The debate on the monetary implications of financial integration has seen the adoption of two main methodologies: panel regressions and VAR analyses. Rey (2013, 2016) mostly uses VARs to highlight factor interdependencies and test for causality. However, many of the authors intervening in the debate use two-stage estimation procedures (Aizenman et al., 2016; Kharroubi & Zampolli, 2016; Obstfeld, 2015). The two-stage estimation procedure is used when the dependent variable of interest cannot be measured, but can be estimated as a coefficient in the first-stage regression. In this thesis, I adopt a two-stage estimation procedure loosely built on Aizenman et al. (2016). This model allows me to estimate the sensitivity of credit growth in each country to credit growth in the financial centers, thus effectively measuring the impact of international monetary transmission channels. The estimated coefficients are used as the dependent variable defining monetary policy independence in the second step of the analysis.

3.3 Model specification and definition of variables

3.3.1 Step 1: estimating sensitivity coefficients

In the first step, I estimate the sensitivity of domestic credit growth in each country to credit growth in the financial centers. For each country i, credit at each time t can be expressed as:

𝑪𝑹𝑬𝑫_𝑖𝑡 = 𝛼_𝑖 + 𝛽_𝑖𝑫𝑶𝑴_{𝑖,𝑡−1}+ ∑𝐶 𝛾_𝑖𝐶𝑭𝑪_𝑡−4𝐶

(19)

15 (1) where i denotes country and t denotes time, 𝛼_𝑖 is a country-specific effect, and 𝜀_𝑖𝑡 is a normally distributed error term.

𝑪𝑹𝑬𝑫 and 𝑭𝑪 measure year-on-year credit growth in each country and in the financial centers (c), respectively. 𝑭𝑪 is subject to one year (four quarters) lags. This lag was chosen as a likely candidate to show some transmission effect from the financial centers to the rest of the world, although the actual response of credit is bound to be heterogeneous for each country and gradual in time9. For the purposes of this research. I use four financial centers: China, Japan, the United Kingdom and the United States. Arguably, among these the United States has a predominant position as the financial center of the world10. As a matter of fact, research has confirmed that the Global Financial Cycle is driven in large part by changes in the United States (Rey, 2013, 2016). Consistently, I assume that, in a first phase, the transmission goes from the United States to the main financial centers, due to their larger financial markets and more direct linkages. Then, it extends from all financial centers to the world. Note that this effect is heterogeneous and gradual in time. While heterogeneity makes it difficult to pinpoint the exact time when the transmission channel shows its effects, the inherent gradualism increases the likelihood of capturing at least part of this effect within a reasonable time frame. Practically, this means that, by using the values of credit growth in the four financial centers at (t-4) as independent variable, the model is reasonably likely to capture some direct or indirect transmission of financial conditions. Moreover, China is included as a center country to capture the effect of the significant capital outflows from China to the Asian region, although the country is still formally applying capital account restrictions and is not an obvious financial center of the world.

𝑫𝑶𝑴 is a vector controlling for two domestic conditions:

9_{For instance, Rey (2016) shows that the response of mortgage spreads in developed countries to changes in the}

United States varies in time, from a couple of months (Canada) to six months (United Kingdom). The effects of this transmission are also expected to last in time, thus showing in later observations as well.

10_{Interestingly, there is strong collinearity between credit trends in the United States and in the United Kingdom}

(20)

16 a) Domestic short term interest rates (IR). These are average money market rates from the IMF International Financial Statistics or, when the former was not available, OECD11 database.

b) Inflation, computed using the Consumer Price Index (CPI) from the IMF International Financial Statistics database.

Both domestic variables are subject to a one-quarter lag to account for delays in the appearance of effects to changes in CPI and interest rates. An alternative specification of this step includes house prices as a domestic independent variable. These are computed using the real quarterly house prices index (base year = 2010) from the BIS database.

𝑮𝑳𝑶𝑩 controls for global variables, including:

a) The VIX index (logarithm), a measure of implied market volatility of the S&P 500 index options. The VIX is a proxy measure for global financial conditions.

b) The Bloomberg Commodity Returns Index (DJUBSTR), as a measure of real global conditions.

The main goal of step 1 is computing 𝛾𝑖, the coefficient of sensitivity of credit trends in each

country to credit trends in each financial center. The sensitivity coefficients thus computed are used as dependent variables in the second step.

3.3.2 Step 2: explaining the sensitivity coefficient

The second step of the analysis is a cross-sectional regression of the sensitivity coefficients 𝛾𝑖

(in absolute terms) on a number of country-specific variables, following Aizenman et al. (2016). The baseline equation for this second stage is the following:

𝛾_𝑖 = 𝜃 + 𝜇_𝑖𝐸𝑥𝑐ℎ𝑎𝑛𝑔𝑒 𝑅𝑎𝑡𝑒 𝑆𝑡𝑎𝑏𝑖𝑙𝑖𝑡𝑦_𝑖 + 𝜋_𝑖𝐶𝑎𝑝𝑖𝑡𝑎𝑙 𝐴𝑐𝑐𝑜𝑢𝑛𝑡 𝑂𝑝𝑒𝑛𝑛𝑒𝑠𝑠_𝑖 + 𝜎_𝑖 𝐹𝑜𝑟𝑒𝑖𝑔𝑛 𝐵𝑎𝑛𝑘𝑠_𝑖 + 𝜏_𝑖𝐵𝑎𝑛𝑘𝑖𝑛𝑔 𝐶𝑟𝑖𝑠𝑖𝑠 𝐷𝑢𝑚𝑚𝑦_𝑖

+ 𝜑𝑖 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝐶𝑜𝑒𝑓𝑓𝑖𝑐𝑖𝑒𝑛𝑡 𝐷𝑢𝑚𝑚𝑦𝑖+ 𝜔𝑖𝐷𝑒𝑣𝑒𝑙𝑜𝑝𝑒𝑑 𝐷𝑢𝑚𝑚𝑦𝑖 + 𝜀𝑖

(2) Exchange rate stability (ERS) is computed using the period average of the de-facto measure of exchange rate stability introduced by Aizenman et al. (2013)12, based on annual standard

11 _{The OECD measure is employed for all the Eurozone countries and for Colombia, Hungary, Israel, New Zealand,}

and Norway.

(21)

17 deviations of the monthly exchange rate between the home country and the base country. I employ period averages normalized between 0 and 1, where 1 is a more stable currency. Capital account openness (KAO) replicates the Chinn-Ito Index13 (Aizenman et al., 2013), which is based on the binary dummy variables that codify the tabulation of restrictions on cross-border financial transactions reported in the IMF Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER). I employ period averages normalized between 0 and 1, where 1 is a more open capital account.

The share of foreign banks as a percentage of the total is derived from the World Bank Global Financial Development Database. A better measure for this purpose would be the percentage of foreign banks’ assets in the domestic banking sector. However, this measure is not consistently available for every country of the sample during the considered time period. The crisis dummy takes value 1 in the presence of a banking crisis and is reproduced following the World Bank Global Financial Development Database. Two additional dummy variables are included to account for negative values of the sensitivity coefficients (NEG, which takes value 1 when the coefficient is negative), and to distinguish between developed and emerging countries (DVP, which takes value 1 for developed countries14_).

4. Step 1: computing the sensitivity coefficients

4.1 Step 1 - Descriptive Statistics

In the first stage of the estimation, I run regression (1) for the full sample (1990-2015) and for three non-overlapping sub-periods (1991-1998, 1999-2006, and 2007-2015). The full sample includes 35 countries (5 in America, 9 in Asia, 19 in Europe, and 2 classified as other). Appendix 1 to this thesis presents the full list of countries and their geographical localization. The three chosen sub-periods are non-overlapping. Note that each sub-period includes at least a full credit cycle, with credit growth peaking once in the early 1990s, then in the early 2000s, and finally before the 2008 financial crisis (Fig. 4.1). An alternative specification takes into

13_{Available on}_{http://web.pdx.edu/~ito/Chinn-Ito_website.htm}_.

14_{Developed: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Hong Kong, Ireland,}

(22)

18 consideration five shorter sub-periods, but without significant impact on the results. Among the three financial centers, China shows the most volatile credit growth. Correlation tables (see Appendix 3, Table A3.1) suggest that there is no excessive collinearity among the independent variables, with the exception of the United States and the United Kingdom, which constitute a more integrated financial center. The consequences of this correlation are further discussed in Paragraph 4.2

Figure 4.1. Credit to GDP in the main financial centers

Plotted time-series for the four independent variables of interest.

Overall, data is reasonably consistent through time, with a limited number of outliers. For what concerns the dependent variable, there are only two countries that noticeably deviate from the trend at a point in time: Indonesia and Ireland (Figure 4.2). In both Indonesia and Ireland credit growth booms in the late 1990s. With these exceptions, credit to GDP trends are linear throughout the sample. However, missing data and/or significant outliers in the secondary independent variables led to the exclusion of some relevant countries from the sample (e.g. Argentina, Switzerland). In the final specification, the interest rates and CPI of Brazil (until 2Q94, when they introduced the peg to the United States dollar) remain the most significant outlier. Missing data – especially for interest rates – significantly affects the first sub-period (1991-1998) and prevents the computation of sub-sample results for Colombia, Mexico and Russia. In the second sub-period, only India is dropped due to a gap in interest rate data.

(23)

19

Figure 4.2. Exceptional trends of credit to GDP: Ireland and Indonesia

Plotted time-series of credit to GDP in Indonesia and Ireland.

4.2 Step 1 - Data issues

The data used in step 1 of this research presents two main issues: collinearity and endogeneity. For one, as highlighted in section 4.1 and by Table A3.1 (Appendix 3), collinearity affects the independent variables computing credit growth in the United Kingdom and the United States. This result suggests that these two countries might actually form one integrated financial center. Collinearity makes it difficult to tease apart the individual effect of credit growth in each financial center. Thus, following Aizenman et al. (2016) and Forbes & Chinn (2003), I observe the joint significance of each set of collinear variables (i.e. cross-link, global, and domestic factors). Moreover, the impact of the collinearity between credit growth in the U.S. and U.K. is further tested as a robustness check in section 4.4, when the U.K. is dropped as a financial center. Results are robust to this alternative specification. Finally, in this model, collinearity does not pose a significant problem as I do not aim to interpret the coefficients individually. A second possible data issue is endogeneity, which might affect the variables included in the first step. In fact, it is impossible to exclude the existence of an unknown common factor jointly determining the dependent variable and one or more independent variables. In particular, there could be one common factor jointly determining credit growth in the four financial centers at (t-4), and later affecting credit trends in every other country at (t). While the significance of this effect cannot be dismissed, the use of lagged values for the independent variable of interest

(24)

20 allows me to at least exclude that the direction of causality may run from the dependent variable to the independent one.

4.3 Sensitivity coefficients computation

In the first step, I estimate regression (1) for each country of the sample in the 1991-2015 time period and in three sub-periods with heteroscedasticity robust standard errors. Figure 4.3 illustrates the share of countries for which the three groups of variables in (1) are jointly significant (with a p-value of 5% or less). Cross-country shows the joint significance of credit growth in the four financial centers; global factor shows the joint significance of the VIX and global commodity index; finally, domestic factor reflects trends in CPI and interest rate. The figure shows that credit growth in the financial centers (cross-country) is jointly significant for almost every country (90%) in the sample in the full period (1991-2015). The incurrence of joint significance for the cross-country factors is lower (56%) in the second sub-period (1999-2006), when both domestic and global factors become preponderant. Note that commodity prices in this specific sub-period rose very steeply. The exclusion of commodity prices from the specification of this period increases the significance of the cross-country factor, but reduces the goodness of the model. In the last period, the joint significance of the cross-country factor is above 90%, testifying that the effect of the Global Financial Cycle has become stronger in the last sampled period. Overall, credit growth in the financial centers is a good predictor of credit growth in each country of the sample. This result supports Hypothesis 1 of this research in that it suggests that credit growth is in large part determined by the Global Financial Cycle.

Figure 4.3. Proportion of significant F-tests

Jointly significance with a p-value of 5% or less. Cross-country = sensitivity to credit growth in the four financial centers; Global = VIX and global commodity index; Domestic = CPI and interest rate.

0.89 0.71 0.56 0.91 0.80 0.10 0.47 0.37 0.51 0.42 0.56 0.34 0 0. 2 0. 4 0. 6 0. 8 1 1991-2015 1991-1998 1999-2006 2007-2015

(25)

21

Overall, the high proportion of joint significance for cross-link factor reflects the widespread significance of the coefficients for each individual financial center (Table 4.1). However, note that the interpretation of the computed individual coefficients is not straight forward. For instance, credit trends in the American region are significantly correlated with China, the United Kingdom and Japan, but not with the United States – as it would be intuitively expected given their geographic location and financial integration. There are two main explanations behind these results. First, collinearity between the United States and the United Kingdom makes it hard to tease apart the individual effects of each predictor. Second, the transmission channel is heterogeneous in time among countries. As such, credit growth in Latin America might actually respond to changes in the United States much faster than the rest of the world exactly because of higher integration. In this case, the four quarter lag partially hides the direct transmission effect with the United States, but highlights the correlation with the other financial centers – which are also expected to respond to changes in the United States faster than most countries in the world. Note that both these effects inhibit the interpretation of the individual coefficients of Table 4.1. Yet, since the combination of this coefficient is carrying of the underlying information on countries sensitivity to credit trends in the financial centers, I can use them as dependent variable in the second step of this research.

Finally, Table A4.1 (Appendix 4) shows the individual coefficients for each sub-period. The analysis of these coefficients suggest that the sensitivity to credit growth in the financial centers has become increasingly significant in time in explaining domestic credit trends. In particular, sensitivities to credit growth in China have become more and more significant in the sample. 4.4 Robustness of the sensitivity coefficients

(26)

22

Table 4.1. Estimated cross-country sensitivity coefficients (1991-2015)

R2 _China _United

Kingdom

Japan United States

America Brazil 0.344 -0.358*** 0.211 0.712** -0.496 Canada 0.285 -0.173*** 0.152 0.330** 0.0488 Chile 0.837 -0.237*** -0.623** 0.133 0.816*** Colombia 0.508 -0.217* -0.674 1.203*** -0.112 Mexico 0.421 -0.304*** -0.749** 1.137*** -0.200 Asia Australia 0.620 0.0366 -0.102 -0.550*** 0.979*** Hong Kong 0.516 0.218*** -0.366** 0.556*** -0.168 Indonesia 0.380 0.380** -2.740*** -1.003 -1.826** India 0.470 -0.152* -0.0302 0.00362 0.197 Korea 0.584 -0.338*** -0.0399 0.612*** 0.123 Malaysia 0.429 -0.362*** -0.639*** 1.080*** 0.190 New Zealand 0.564 -0.0569 -0.192 -0.143 0.521*** Singapore 0.379 -0.423*** -0.372 0.155 -0.181 Thailand 0.730 -0.178** -1.228*** 0.723*** -0.966*** Europe Austria 0.343 -0.0116 0.0666 -0.0341 0.248*** Belgium 0.316 -0.0965 -0.0860 -0.285* 0.273 Czech Republic 0.341 -0.0666** -0.0503 0.381*** -0.0843 Denmark 0.550 -0.147** -0.0474 0.833*** -0.166 Finland 0.600 -0.0991*** 0.152 0.141 -0.0321 France 0.616 -0.127** 0.425*** -0.337** 0.152 Germany 0.690 -0.141** 0.557*** -0.259 0.514*** Greece 0.544 0.121* 0.535*** -0.318 0.512** Hungary 0.489 -0.0669*** 0.184*** -0.150* 0.00953 Ireland 0.676 0.0852 0.694*** -0.423** 0.0901 Italy 0.629 0.170 0.715*** -0.519 1.345*** Luxembourg 0.284 -0.125 -0.455 -0.0428 1.862*** Netherlands 0.551 -0.0601 0.618*** -0.322* 0.197 Norway 0.390 0.0307 0.0173 0.299** 0.0401 Poland 0.386 -0.0692 -0.372** -0.113 0.760*** Portugal 0.532 -0.303*** 0.00981 0.335 1.464*** Russia 0.540 -0.0694 0.284* -0.269 -0.255 Spain 0.190 0.332 1.242 -1.860* 0.724 Sweden 0.510 -0.180*** 0.266 -0.0924 0.140 Other _Israel _{0.376 -0.0653} _0.389*** _-0.140 _-0.133 South Africa 0.383 -0.135* -0.0789 -0.223 0.506*

(27)

23 Overall, house prices are somewhat informative in this research specification. However, the significant explanatory power of house prices in the last sub-period might be enhanced by the effects of the mortgage-led boom-bust dynamics of the 2008 Global Financial Crisis, making it a time fixed effect. As data on house prices is not consistently available for the whole time-period, I cannot exclude this latter effect nor include them as an independent variable in the main model.

Figure 4.4. Proportion of significant F-tests with and without house prices (2007-2015)

Jointly significance with a p-value of 5% or less. Cross-country = sensitivity to credit growth in the four financial centers; Global = VIX and global commodity index; Domestic = CPI, interest rate, and house prices (only on the right).

A second robustness check consists of dropping the United Kingdom as financial center in order to test for the impact of collinearity between credit growth in the United Kingdom and the United States. This alternative specification increases the occurrence of significant values for the United States, but it does not change the overall joint significance of the dependent variables. Moreover, excluding the United Kingdom as a financial center does not improve the goodness of fit of the model significantly nor consistently throughout the sampled countries. Finally, I compute the sensitivity coefficients for five smaller sub-samples covering five years each. This modification reduces number of observations in the coefficient computations. While the joint significance of the sensitivity coefficients remains prevalent in the sample, the small sample reduces the accuracy of the coefficient themselves.

0.91 _0.88 0.37 0.29 0.34 0.59 0 0. 2 0. 4 0. 6 0. 8 1

Without house prices With house prices

(28)

24

5. Step 2: the trilemma hypothesis

5.1 Step 2 - Descriptive statistics and data issues

Step 2 computes regression (2) using data for the whole period (1991-2015) and for four sub-period (1991-1998, 1999-2006, and 2007-2015). This phase consists of a cross-sectional analysis on an unbalanced panel. Table 5.1 provides summary statistics for the full period of analysis. In step 2, the absolute values of the sensitivity coefficients (sc) become the dependent variable. In fact, I use the absolute values as I am mainly interested in the size of the effect. Yet, I also include a dummy variable taking the value of 1 when the coefficient is negative to test for the significance of the direction. All independent variables except the share of foreign banks are normalized between 0 and 1. The average sensitivity coefficient is 0.5791, with the minimum being at 0.0666 (between Czech Republic and China) and the maximum being 2.74 (between Indonesia and China). Sensitivity coefficients are positively skewed, with most observations concentrated between 0 and 1. The index of Capital Account Openness (KAO) is normalized between 0 and 1, but with four times as many observations around the maximum than any other level. This latter issue is particularly problematic and suggests that the chosen sample of countries might not be representative enough of lower levels of capital account openness15_{. Exchange Rate Stability (ERS) values are registered at both extremes of its range,}

but there is a gap in frequency between 0.5 and 0.8 due to the composition of the index: countries tend to have either volatile exchange rates or pegged currencies, in a close to binary distribution. The crisis dummy takes value of 1 in the presence of a banking crisis. Almost three times as many countries went through at least one banking crisis in the period 1991-2015. The share of foreign banks ranges between 1.26% (Sweden) and 95.53% (Luxembourg), and it shows positive skewness towards the bottom, with a large share of countries having less than 20% foreign banks. The negative coefficients dummy takes value 1 about 35 times, as the coefficients are nearly evenly split between positive and negatives, while 20 out of 35 countries are classified as developed.

15_{The high incurrence of countries with open capital accounts is due to the chosen sample, which includes only}

(29)

25

Table 5.1. Step 2 summary statistics (1991-2015)

Variable Obs Mean Std. Dev. Min Max Skewness Kurtosis

Sensitivity coefficients (sc) 69 .5791 .5015594 .0666 2.74 1.962723 7.500173 Negative coefficients dummy 69 .5073 .5036 0 1 -.0290 1.0008 Capital Account Openness 140 .6682 .3331 0 1 -.6748 1.9977 Exchange Rate Stability 140 .4133 .3472 0 1 .4753 1.5251 Crisis dummy 140 .7714 .4214 0 1 -1.2928 2.6713 Foreign banks 140 32.64 25.8656 1.2632 95.5263 .7596 2.5744 Developed countries dummy 140 .5714 .4967 0 1 -.2887 1.0833

Correlations for the second step are reported in Table A3.2 (Appendix 3), showing there is no excessive collinearity among the independent variables. The only relevant correlation is between the two independent variables of interest – ERS and KAO - suggesting that countries with open capital accounts tend to have more pegged currencies. In this sample, this latter effect is overestimated due to the large prevalence of Eurozone countries, which have a pegged currency and very open capital accounts by design.

The results of step 2 are affected by some data issues, particularly a strong selection bias. In fact, the choice of sample was limited by the availability of credit data. In the existing dataset, 12 out of 35 sampled countries belong to the Eurozone. By definition, these countries have a perfectly pegged currency and tend to have very open capital accounts, hence the correlation between the two independent variables of interest. However, the Eurozone represents a peculiar case, in that the Euro as a whole is a flexible currency. Moreover, the construction of the ERS index presents some problems of its own. Indeed, the index, which is built on annual standard deviations of the monthly exchange rate, only takes value below 0.5 and above 0.8. Thus, the distribution of frequencies for ERS is skewed towards the extremes.

5.3 Step 2 – Results and robustness

(30)

26 The results of step 2 of this research are highly affected by the data limitations and issues presented in the previous section. As such, testing the trilemma hypothesis revealed problematic. The data from the full period 2015) and the first two sub-periods (1991-1998, 1999-2006) does not show significant correlation with the explanatory variables of the model. Only data from the third sub-period (2007-2015) offers some significant results (Table 5.2). In fact, exchange rate stability in the third period is significantly correlated with credit growth sensitivity. The interaction with the dummy variable for negative values shows that fixed exchange rates are correlated with higher credit growth sensitivity for both positive and negative coefficients. The positive significant link between ERS and sensitivity coefficients suggests that flexible exchange rates may still play a role, at least partially, in insulating countries from the Global Financial Cycle. The coefficient of Capital Account Openness is also significant for the last period. In this case, the relationship is negative for positive coefficients of sensitivity, and for developed countries. Note that this result is counterintuitive and in apparent contrast with both the dilemma and trilemma hypothesis. However, this result is most likely the product of dataset issues, as in the last period 20 out of 35 countries have maximum capital account openness (i.e. KAO=1).

(31)

27

Table 5.2. Selected results (2007-2015)

(1) (2) (3) (4) (5) (6) 2007-2015 2007-2015 2007-2015 2007-2015 2007-2015 2007-2015 ERS 0.196** 0.164*** 0.187* 0.203** (0.0925) (0.0590) (0.0965) (0.0917) KAO -0.263** -0.272** -0.264* -0.263* (0.128) (0.118) (0.130) (0.132) BANKS 0.00114 (0.00118) CRISIS 0.0518 (0.0550) 0.NEG#c.KAO -0.235 (0.138) 1.NEG#c.KAO -0.276** (0.128) 0.NEG#c.ERS 0.219** (0.101) 1.NEG#c.ERS 0.165* (0.0928) 0.DVP#c.KAO -0.229 (0.142) 1.DVP#c.KAO -0.269** (0.128) 0.DVP#c.ERS 0.196** (0.0758) 1.DVP#c.ERS 0.196* (0.103) Constant 0.316*** 0.270*** 0.317*** 0.319*** 0.307*** 0.316*** (0.0850) (0.0597) (0.0863) (0.0865) (0.0874) (0.0891) Observations 62 62 62 62 62 62 R-squared 0.239 0.275 0.248 0.249 0.247 0.239 Robust standard errors in parentheses

(32)

28

6. Credit to households and non-financial corporations

In this section, I test whether credit sub-components show different sensitivity to the Global Financial Cycle (Hypothesis 3a). In particular, it is interesting to evaluate whether credit to households is more sensitive than overall credit to the private non-financial sector (Hypothesis 3b). This is important due to the peculiar role of mortgage credit in boom-bust dynamics. In fact, higher sensitivity of credit to households could support the idea that capital controls are especially necessary to protect against the transmission of these dangerous financial conditions. 6.1 Credit sub-components - Model specification and descriptive statistics

The dependent variable of regression (1) – credit to the private non-financial sector – can be divided into two sub-components: credit to households (HH) and credit to non-financial corporations (NFC).

Hence, a possible modification of regression (1) employs credit sub-components as dependent variable:

𝑯𝑯_𝑖𝑡 = 𝛼_𝑖 + 𝛽_𝑖𝑫𝑶𝑴_𝑡−1𝐹 + ∑𝐶_𝑐=1𝛾_𝑖𝐶𝑭𝑪𝐶_𝑡−4+ ∑𝐺_𝑔=1𝛿_𝑖𝐺𝑮𝑳𝑶𝑩_𝑡𝐺 + 𝜀_𝑖𝑡

(3) 𝑵𝑭𝑪_𝑖𝑡 = 𝜃_𝑖 + 𝜇_𝑖𝑫𝑶𝑴_𝑡−1𝐹 + ∑𝐶_𝑐=1𝜋_𝑖𝐶𝑭𝑪𝐶_𝑡−4+ ∑𝐺_𝑔=1𝜌_𝑖𝐺𝑮𝑳𝑶𝑩_𝑡𝐺 + 𝜀_𝑖𝑡

(4) where 𝑯𝑯 is year-on-year growth in credit to households (namely, mortgages and consumer credit), while 𝑵𝑭𝑪 is year-on-year growth in credit to non-financial corporations. All independent variables are the same as in step 1. Appendix A2.3 describes the data sources in more details. The computation of regression (3) and (4) generates two more sets of sensitivity coefficients.

Hypothesis 3 states that the two sub-components show different sensitivities to credit growth in the global financial centers. In order to explore this hypothesis, I run the following regression:

𝑫𝑰𝑭𝑭_𝑖𝑡 = 𝜎_𝑖+ 𝜏_𝑖𝑫𝑶𝑴_{𝑖,𝑡−1}+ ∑𝐶_𝑐=1𝜑_𝑖𝐶𝑭𝑪𝐶_𝑡−4+ ∑𝐺_𝑔=1𝜔_𝑖𝐺𝑮𝑳𝑶𝑩_𝑡𝐺 + 𝜀_𝑖𝑡

(33)

29 where 𝑫𝑰𝑭𝑭 is computed as the difference between 𝑯𝑯 and 𝑵𝑭𝑪. A significant coefficient of cross-country sensitivity 𝜑_𝑖 would suggest that the two credit sub-components show different sensitivities to the Global Financial Cycle.

The dependent variable 𝑫𝑰𝑭𝑭, although fairly volatile, does not present significant data issues or outliers. For the sampled countries, 𝑫𝑰𝑭𝑭 takes values between (-0.53) and (0.87). Due to data unavailability, the time-period is reduced to 2000-2015, generating only two sub-periods (2000-2006, 2007-2015). Lack of data also prevents the computation of the coefficients for India and Malaysia in the first sub-period.

The analysis of credit sub-components is structured as follows. First, I run regression (3) and (4) to test whether the sub-components model specification is relevant and robust. Second, the computation of regression (5) on the difference between the two credit sub-components aims at confirming whether these show different sensitivities to credit growth in the financial centers. Finally, I explore whether the sensitivity of credit sub-components is larger than the sensitivity of total credit.

6.3 Credit sub-components – Computation and results

The model specification using credit sub-components as a dependent variable appears to be valid and robust (Figure 6.1). In fact, Figure 6.1 illustrates the share of countries for which the three groups of explanatory variables are jointly significant (with a p-value of 5% or less). As a matter of fact, both the cross-country factor and the domestic factor are significant for most countries in the sample. This share is even higher for both credit sub-components than for total credit. On the contrary, the significance of global factors is less prevalent.

Figure 6.1. Proportion of significant F-tests for total credit and credit sub-components

Jointly significance with a p-value of 5% or less. Cross-country = sensitivity to credit growth in the four financial centers; Global = VIX and global commodity index; Domestic = CPI and interest rate. Time period: 2000-2015.

0.86 0.94 0.94 0.80 0.35 0.35 0.63 0.71 0.79 0 0. 2 0. 4 0. 6 0. 8 1

Total credit to the private non-financial sector

Households Non-Financial corporations

(34)

30 Subsequently, I test whether the difference (𝑫𝑰𝑭𝑭) between credit growth for the two sub-components is significantly correlated with credit growth in the financial centers. Figure 6.2 plots the proportion of joint significant F-tests for the three groups of explanatory variables in regression (5) for the full period and two sub-periods. Credit growth in the financial centers (cross-country) is jointly significant for almost 70% of the countries in the sample using full period data. Thus, credit growth in the financial centers is a significant explanatory factor of the difference between credit to households and credit to non-financial corporations. The joint significance of the cross-country and domestic factors is confirmed in the two sub-period, while global factors are more significant for the full period. Moreover, the predominance of jointly significant results for the cross-country factor proves to be robust to the inclusion of house prices as a domestic factor. Overall, these results confirm that credit sub-components indeed show different sensitivities to credit growth in the financial centers.

Figure 6.1. Proportion of significant F-tests for regression (5)

Jointly significance with a p-value of 5% or less. Cross-country = sensitivity to credit growth in the four financial centers; Global = VIX and global commodity index; Domestic = CPI and interest rate.

Finally, I wish to compare the sensitivity of credit sub-components to the sensitivity of total credit. The aim is to gauge whether credit to households (or, vice versa, credit to non-financial corporations) is more sensitive than total credit to the Global Financial Cycle. The significance tests on the differences between the sensitivity coefficient of total credit and credit sub-components show different trends for non-financial corporations and households (Table 6.1). In fact, the coefficients of sensitivity tend to be larger for credit to non-financial corporations than total credit (70.8%), while the opposite is true for household credit, for which sensitivity is only higher in 39.5% of the cases. These results offer preliminary evidence that the sensitivity of firm credit is often greater than the sensitivity of total credit, while the opposite is true for

0.74 0.63 0.71 0.76 0.28 0.32 0.85 0.69 0.56 0 0. 2 0. 4 0. 6 0. 8 1 2000-2015 2000-2006 2007-2015

(35)

31 credit to households. Overall, given the peculiar role of mortgage credit in boom-bust dynamics, the lower sensitivity of credit to households to the Global Financial Cycle is reassuring against the dangerous effects of the transmission.

Table 6.1.Significance of beta coefficients and t tests

Non-financial

corporations Households (a) Number of betas significantly different from total credit betas 48 38 (b) Number of betas significantly larger than total credit betas 34 15

(c) Percentage of total (b/a) 70.8% 39.5%

6. Conclusions

This thesis inserts in the line of research on the monetary policy implications of the existence of a Global Financial Cycle. In order to do so, I introduce a new approach to estimating monetary policy independence in the context of the Mundellian monetary policy trilemma hypothesis. Building on the most recent contribution of Rey (2016), I assume that monetary policy transmission in the financial centers works through both the interest rate channel and other channels of transmission. In turn, shifts in the financial centers’ conditions spread internationally through the international credit channel, not just the interest rate channel. In other words, changes in credit convey information on both the direct effect of the interest rate channel and of the indirect amplification mechanism of other transmission channels. As such, I first compute coefficients of sensitivity of credit growth in each country to credit growth in the financial centers. In a second step, the coefficients of credit growth sensitivity are employed as a measure of monetary policy independence in testing the trilemma hypothesis. Finally, I explore whether credit sub-components have different sensitivities to the Global Financial Cycle.