1
Effects of United States quantitative easing
policies on capital flows to emerging market
economies
Raphaël S. Kleinherenbrink Student number: 10431241
Faculty of Economics and Business
Study programme: BSc Economics and Business Specialization: Economics and Finance
Thesis supervisor: Stephanie Chan
6 July 2015
Abstract
At the initial stage of the financial crisis in 2008, in an attempt to keep interest rates low with no conventional monetary transmission mechanisms at hand, the Federal Reserve (FED) started to engage in the unconventional monetary policy (UMP) practice of quantitative easing. Following the expansionary UMP by the United States (US), many emerging market economies (EMEs) witnessed large increases in their capital inflows. Using panel data fixed effects models, this paper examines the relationship between quantitative easing (QE) in the US and capital flows to EMEs based on a sample of five EMEs; Brazil, India, Indonesia, Mexico and Turkey, over the period between the first quarter of 2005 and the last quarter of 2014. Three
transmission channels between QE and capital flows to EMEs are identified: portfolio, liquidity and confidence channels. Of these, the portfolio channel was found to have transmitted a positive effect of QE on capital flows to EMEs. Also, measured through dummy variables, a significant effect unobserved through the transmission channels was found during QE1.
Keywords: EMEs, capital flows, QE, UMP, portfolio rebalancing, LSAP, FED
Acknowledgement
I would like to express my gratitude for the valuable insights and support I have enjoyed from my thesis supervisor Stephanie Chan. Views expressed in this paper are exclusively the author’s and do not represent the views of any other institution.
2
Statement of originality
This document is written by Raphael S. Kleinherenbrink who declares to take full
responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is
3 Table of contents 1. Introduction 1.1. Research method 2. Theoretical background 2.1. Transmission channels
2.1.1. Portfolio rebalancing channels 2.1.2. Liquidity channels
2.1.3. Confidence channels 2.2. Standard determinants
3. Data and empirical model 3.1. Research methodology 3.2. Variable presentation 3.3. Model presentation 3.4. Diagnostic tests 4. Results and discussion
4.1. Descriptive statistics 4.2. Diagnostic test results 4.3. Regression results 5. Conclusion References Appendix 4 5 6 7 8 8 9 9 10 10 10 13 14 14 14 19 19 21 23 25
4
1. Introduction
At the initial stage of the global financial crisis in 2008, the Federal Reserve (FED) announced its first quantitative easing (QE) program. QE is the practice carried out by a country’s central bank (CB), which includes pumping money into the economy by obtaining securities from the market. The tool is generally used in exceptional economic circumstances such as economic and financial crises when short-term interest rates are approaching their zero lower bound (Bouraoui 2015, 1563).
The first QE programme, called QE1, included the outright purchases of mortgage-backed securities and Treasury bills totally worth $2.1 trillion, carried out between November 2008 and March 2010 (Bouraoui 2015, 1563). At the time of announcement, the FED, in an attempt to counter economic turmoil, had already pushed the nominal short-term interest rate to its zero lower bound using conventional monetary policy measures: decreasing key CB interest rates to stimulate economic activity (Lenza et al. 2010, 299). When the nominal short-term interest rate has
reached its zero lower bound, however, conventional monetary policy tools, which are the short-term interest rate, the discount lending rate and reserve requirements, can no longer be used to achieve long-term price stability. This was the initial reason for the implementation of unconventional monetary policy (UMP). After the European debt crisis caused new instability in financial markets, the FED announced its second QE program, called QE2. This program included $600 billion worth of purchases of Treasury securities between November 2010 and June 2011. And, in September 2012, the FED launched QE3, a third round of asset purchases: an open-ended operation, which included $85 billion worth of purchases of fixed-income securities on a monthly basis (Bouraoui 2015, 1562).
The QE programs involving large-scale asset purchases (LSAP), were initially meant to improve financial market functioning during economic turmoil, but ended up to be used with the objective of post-crisis economic recovery as well. In emerging market economies (EMEs), large concern about possible spillover effects of the prolonged LSAP developed. Between mid-2009 and the first quarter of 2013, during QE1, QE2 and QE3, cumulative gross financial flows to the developing world increased from $192 to $598 billion, compared to an increase of $182 between mid-2002 and the first quarter of 2006 (Lim et al. 2014, 6). As a result, many EMEs saw a significant increase in their real exchange rate (Bouraoui 2015). Meanwhile, on 22 May 2013, the FED officially announced the possibility of reducing its QE program. This was seen as an immediate threat to EMEs and caused large capital outflows from the developing world (Bouraoui 2015, 1563; Lim et al. 2014, 6). With the tapering announcement, in contribution to earlier concerns from EMEs about appreciating exchange rates and increasing asset prices, this raised concerns about disorderly capital reversals as a reaction to a change in expectations about United States (US) monetary policy (Tillman 2014, 1; Morgan 2011, 17; Lim et al. 2011, 6).
Both Barroso et al. (2013), Lavigne et al. (2014) and Chen et al. (2012) identify capital flows as the most or one of the most important transmission channels between US UMP and key economic variables in EMEs such as interest rates, asset prices and
5
exchange rates. Currently, there is a considerable amount of existing research on the macroeconomic effects of QE and its effects on financial markets. There is, however, a much smaller amount on the effects of US QE on EMEs and especially non-Asian EMEs. Even though, according to both studies by Chen et al. (2012) and Barroso et al. (2013), effects of US QE on Latin America appear to be stronger as compared to Asia.
Findings in preceding literature include Tillman (2014), which finds US QE to have considerable effects on EMEs' capital flows, equity prices and exchange rates but does not aim to provide an explanation of capital flows. Furthermore, Morgan (2011) finds QE to have increased capital flows to EMEs, but focuses purely on Asian EMEs. According to Ahmed and Zlate (2014), US QE is only one of many
determinants of capital flows to US QE. And, according to Bowman et al. (2014), if every country-specific characteristic is taken into account, US UMP may not even have outsized effects on asset prices in EMEs. On the other hand, however, Lavigne et al. (2014) finds QE to have increased capital flows to EMEs, but is unclear on whether the effects of QE on capital flows differ from the effects of conventional monetary policy.
Even tough capital flows are in literature widely acknowledged to be one of the key determinants of key macroeconomics variables in EMEs, the subject is still met with a lot of ambiguity in academics in the explanation of the total effects and composition of cross border capital flows. In modern day worldwide economic turmoil and uncertainty about sovereign debt bailouts and further expansionary or contrary unconventional monetary policy measures, research on this topic could possibly offer serious outcomes for future economic policy and have near-future welfare implications for EMEs. Therefore, in an attempt to contribute to the literature concerning the effects of unconventional monetary policy on capital flows to EMEs, this paper conducts its research around the following question:
What are the effects of quantitative easing by the Federal Reserve on capital flows to emerging market economies?
1.1 Research method
In order to come to an answer to the main question, this paper performs a regression analysis on a model for gross financial flows similar to the model used by Lim et al. (2014). The latter offers a model to evaluate the effects of QE policies by the FED through both observable transmission mechanism variables and variables accounting for unobserved effects over and above the observed.
Section two offers the theoretical background of the research on the
transmission of UMP and in specific QE to cross-border capital flows. Moreover, it will also give background information on research on the standard determinants of capital flows for the sake of the model tested further on in this paper. In general, three observable transmission channels between QE and capital flows to EMEs are
identified: portfolio, liquidity and confidence channels.
In section three, the models to be estimated and the dependent and
6
Lim et al. (2014) and Nier et al. (2014). The dependent variable used is gross financial flows. This variable is made up of the relevant EME’s changes in foreign holdings of direct investment, portfolio holdings and cross-border bank lending. Independent variables included are the yield curve, interest rate differential, growth differential, PMI, 3-month Treasury bill rate (T-bill rate), M2, VIX, GDP growth, overall risk rating, amount of trade, external debt, credit and exchange rate. These variables include both transmission mechanism variables of QE to capital flows and standard determinants of capital flows. Moreover, to account for effects of QE above the effects observed through transmission channels, QE indicator variables will be added. Which leads to three different model estimations: one with a single QE indicator variable, one with distinct QE variables for every QE episode and one which adds another indicator variable to the latter for the QE tapering announcement by the FED in May 2013.
The tests were performed on variables of Brazil, India, Indonesia, Mexico and Turkey, observed over the same period of time. As this paper is merely interested in variables that vary over time, the tests performed were based on fixed effects models. This way, all observations are taken into account, but the test controls for countries’ individualities. Moreover, to control for time-specific effects, time indicator variables are included in the model.
The results are presented in section four. On a 5% significance level, the portfolio channel proved significant through the yield curve. On a 10% significance level, the liquidity channel proved significant as well through the 3-month T-bill rate. Also the QE indicator proved significant, and when broken up into three different episodes, only the first QE episode showed effects above the observed effects through the mentioned transmission channels. The higher effect of QE1 as compared with later episodes is in line with findings in previous research (Cúrdia and Ferrero 2013; Krishnamurthy and Vissing-Jorgensen 2013, in: Lim et al. 2014, 15).
2. Theoretical background
Much research on the effects of quantitative easing policies on worldwide capital flows has preceded this paper. There has, however, been substantially less research on the effects of QE policies in the United States on capital flows to EMEs, especially on EMEs outside of Asia (Bouraoui 2015). This section aims at identifying possible explanatory transmission channels between US QE policies and capital flows to EMEs and standard determinants of capital flows.
Morgan (2011) investigates the effects of US increased liquidity due to UMP on gross private capital flows to Asian economies. The paper finds excess capital flows to Asian economies attributable to US QE present, especially portfolio flows, but small in size. It argues that these excess capital flows to Asian economies could have been easily been absorbed by sterilization policies by domestic CBs. Cho and Rhee (2013) and Chen et al. (2012) also investigate the effect of US QE on Asia. In line with Morgan (2011), the study finds modest effects over the QE period: QE1 is found to have a significant effect on capital flows, while during QE2 and QE3, no large
7
effects could be found. On the other hand, research by Tillman (2014) found US QE to cause excess capital inflows to EMEs, exchange rate appreciation, interest rate depreciation and a boom in equity prices, just as Barroso et al. (2013). A similar result is also found by Lim et al. (2014), which found a positive relationship between US QE and gross financial flows to developing economies using vector autoregression (VAR). A possible explanation for the difference in findings is offered by Chen et al. (2012). Also through VAR, the latter study finds the effects of CB policies on both advanced and developing economies in Asia and Latin America to be asymmetric and much stronger in Latin America as compared to emerging Asia, with the strongest effects in Brazil and Hong Kong. Possible explanations offered by Chen et al. (2012) include differences in stage of development, financial policy frameworks, economic and financial interconnectedness with other economies, exchange rate regimes and many other factors. Moreover, Ahmed and Zlate (2014, 223) also names loose policy in advanced economies, local fundamentals and economic growth prospects.
2.1. Transmission channels
Conventional monetary policy is transmitted through the standard transmission
channels of monetary policy: the interest rate and other asset price channels (including exchange rate, equity prices and the credit channel). In the case of UMP (which includes QE), the case is that these conventional transmission channels are either ineffective, unavailable or weak. Hence, LSAP programs by the central bank are justified (Lim et al. 2014).
In an attempt by Lim et al. (2014) to document the effect of US QE on gross financial flows to developing economies, three transmission channels are identified for this form of UMP: liquidity, portfolio and confidence channels. This is supported by Chen et al. (2012), Tillman (2014) and Fratzscher et al. (2013). According to Lim et al. (2014), the flows appear to be heterogeneous: portfolio channels seem to be driving the results, while foreign direct investment (FDI) appears to be insensitive to US QE. The latter is supported by Ahmed and Zlate (2014) who also find a positive effect of US UMP on EMEs capital inflows and in specific on portfolio inflows. On the contrary, Bouraoui (2015), who performs a regression analysis to investigate whether the depreciation of currencies following QE announcements is a result of capital flows, finds a different result. The paper finds that discrepancies in the depreciation of currencies as a result of the announcement by the FED on the
'tapering' of QE are the result of discrepancies in resulting capital outflows. The paper, however, concludes that both inward and outward FDI appear to be the most common determinant of capital flows and exchange rate fluctuations as a result of
announcements about the LSAP programme of the FED. Lavigne et al. (2014), which also states US QE has likely increased capital flows to EMEs, also identifies portfolio and confidence channels as the transmission channels between US QE and capital flows to EMEs. In addition to those channels, however, it identifies an exchange rate and a trade-flow channel. The latter two, however, may not be relevant for gross capital flows to EMEs because as mentioned in Lavigne et al. (2014), it concerns
8
demands for US production and the trade-flow channel is there to partly offset this effect. Hence, it does not concern capital flows into EMEs. For the purpose of investigating the effect of US QE of gross capital flows, therefor, this study sticks to the transmission channels identified by Lim et al. (2014, 8): portfolio, liquidity and confidence channels.
2.1.1. Portfolio rebalancing channels
QE involves LSAP of longer-duration assets such as mortgage-backed securities. As financial assets are imperfect substitutes with distinct characteristics, as result of LSAP programs, demand and yield of assets may change. When engaging in outright asset purchases, the FED changes the supply of relatively low-risk long-term assets it purchases. This happens through the replacement of longer-duration assets for long-term government bonds. The new composition of investors' portfolio's resulting from this practice thus bears less risk and therefor holds a lower risk-adjusted return. Investors, in turn, will start to turn to more risky assets, which offer a higher expected return. This process is called portfolio rebalancing. EMEs are rich with these sorts of investment opportunities, which explains portfolio rebalancing as a transmission channel between US QE and gross capital flows to EMEs. This process leads to lower risk premiums, higher asset prices and lower yields in EMEs (Lavigne et al. 2014, 25; Chen et al. 2012; Lim et al. 2014, 8; Fratzscher et al. 2013).
Lim et al. (2014, 8) expects the portfolio balance channel to be observable through increased demand for temporal (longer duration) and spatial (developing country) assets. Which comes as a result of investors' need to rebalance their portfolio after their risk exposure (and therefor their expected return) has decreased. This was also observed by Fratzscher et al. (2013) in the case of liquidity injections and LSAP announcements. The latter paper found the portfolio rebalancing channel across countries was the main transmission channel of FED liquidity injections and LSAP announcements to capital flows as they induced large capital flows to EMEs. In the case of FED treasury purchases, on the contrary, the paper found the portfolio
rebalancing transmission channel to function mostly across asset classes, from bonds to equity, both for the US and EMEs.
2.1.2. Liquidity channels
When the FED engages in LSAP, long-term assets purchased from private sector investors through QE are replaced by increased reserves on bank's balance sheets. These reserves are more liquid than the long-term assets they replace. As a result, banks face less liquidity constraints and are able to extent more credit to investors (Lim et al. 2014; Fratzscher et al. 2013). Since financial markets are currently highly integrated, a decrease in borrowing costs also leads to increased bank lending to EMEs, which explains increased capital flows to EMEs. Also, as a result of the decrease in the liquidity premium, investors will start to engage in carry trades in a search for yield. This is the process in which investors buy a currency with higher interest rates to capture the difference (Chen et al. 2012, 237).
9
2.1.3. Confidence channels
By engaging in LSAP, the FED may give off the signal that it will commit to keeping interest rates low even after achieving its initial target: economic recovery. The reason why it gives this signal is because a premature increase in interest rates may decrease the value of assets held by the central bank (Lavigne et al. 2014, 25; Lim et al. 2014, 8). As a result, the risk-free component of bond yields may decline (Fratzscher et al. 2013).
As a result of an increase in information about the future state of the economy, the confidence of households and business improves, which is the confidence
signaling channel. Concerns about deflation risk as a result of a premature increase in the interest rate are reduced and regular CB intervention has the potential to reduce volatility and hence uncertainty about future economic conditions. As a result, investment and consumption increases, both domestically and in foreign countries (Lim et al. 2014, 8; Tillman 2014, 1). The information about the future state of the economy may also have an unpredictable effect on capital flows depending on investor's the interpretation of the policy influencing its risk appetite. If an
announcement by the FED, for instance, gives of the sector to investors worldwide that the US has very significant economic recovery ahead of it, it may lead to large capital flows to the US (Joyce 2010; Wright 2011, in: Fratzscher 2013).
Another form of transmission trough the confidence channel is mentioned by Bowman et al. (2014). The paper mentions the possibility that markets may interpret the FED's announcements in the way that their policy will change other countries’ monetary policy. As a change in US interest rates may lead to changes in exchange rates, investors may start to expect changes in other country's monetary policy as a reaction.
2.2. Standard determinants capital flows
In Ahmed and Zlate (2014, 223), key determinants of capital flows to EMEs are discussed. The paper names, apart from US interest rates and loose policy in advanced economies (such as QE), also the growth prospects of an EME. Moreover, the paper also discusses the country-specific attractiveness as an investment destination and interest rate determinants as important determinants, which is supported by Barroso (2013). The study also found the sensitivity capital flows have to interest rate differentials to have increased in post-crisis period. Differences in the effects of US QE on countries are also found by Chen et al. (2012). The latter study observes differences in the way country-specific variables react to changes in the US term spread and notes the impact of QE in the US is much more influential on EMEs as compared to advanced economies. Bowman et al. (2014) identifies countries who typically have capital flows that are sensitive to US QE policies as having high interest rates, credit default swap spreads, inflation rates, current account deficits and relatively vulnerable banking systems. Studies by both Nier et al. (2014) and Chen and Kahn (1997), focusing on the cost of financing aspect as a determinant of capital flows, state that the financial development in the EME can affect both the size of capital inflows as well as the composition of capital inflows. The latter paper
10
demonstrates how capital flows to EMEs are generated by a combination of financial market development in the recipient’s country and the country's growth (potential). Moreover, the paper also found capital inflows to be determined by the interaction between financial market development and growth (potential).
A study by Mody et al. (2001) examines the interaction between so called country-specific 'pull' factors and global 'push' factors that drive capital flows to EMEs. It found that country-specific pull factors have a more significant effect on capital flows to EMEs than global push factors. Short-term dynamics, however, were found to be largely influenced by push factors. A similar result is found by Carlson and Hernández (2002, 6). This paper found ‘push’ factors to be overall determinant of capital flows to the developing world, but ‘pull’ factors to eventually determine which of the developing countries receive these capital flows. Relevant country specific factors mentioned in the latter study include creditworthiness, monetary policy and regional location. According to Carlson and Hernández (2002), the factor often found to be determinant in the short-term is the state of the economy in the recipient country.
3. Data and empirical model
3.1. Research methodology
The main focus of this research is on the investigation of the effects of US QE on capital flows to EMEs among others through the transmission channels as mentioned before: portfolio channels, liquidity channels and confidence channels. Above the three channels, however, also unobserved effects of US QE on capital flows will be accounted for. In order to come to a conclusion, the relationship between capital flows and the transmission channels mentioned above, the standard determinants of capital flows, and unobserved effects are tested.
In order to estimate the effects of QE on capital flows to EMEs, a linear
regression will be performed on a fixed effects model based on panel data on a sample of 5 EMEs: Indonesia, Turkey, Brazil, Mexico and India. This sample of countries is compromised of the five EMEs showing the largest near future growth prospects, taking into account as main growth drivers: growth in physical capital stock, labour force, quality of labor and technological innovation (Hawksworth and Cookson 2008). The period of observations taken on a quarterly basis is ten years, from the first
quarter of 2005 to the fourth quarter of 2014, with three missing observations, therefor leading to 197 observations of capital flows in total. As the tests involve variables of five countries observed over the same period of time, a regression is performed on a fixed effects model based on panel data to account for the countries’ individualities. The models include many variable proposed by Lim et al. (2014, 9) and Nier et al (2014).Furthermore, the models include time indicator variables to capture time-specific effects. In this research, three models will be used for estimation. The next section presents the variables used.
3.2. Variable presentation
11
Gross financial flows (GFF)
The dependent variable in the regression model is made up of three pieces of data together compromising the gross financial flows to a country. The dependent variable consists of portfolio flows, FDI, both corrected for disinvestment, and changes in bank lending. Portfolio flows and FDI are subtracted from the International Financial Statistics database and the data on changes in bank lending is subtracted from the Bank of International Settlements (BIS).
The independent variables
Yield curve (YLD)
As mentioned in the theoretical background, portfolio rebalancing can occur both across asset classes and across countries. The yield curve is the term spread between 10-year US bonds and 3-month Treasury bills. This is a global push variable. It is a primary measure of the effect that QE can have on long-term yields and thus on portfolio rebalancing across asset classes, the increased demand in temporal assets (both in the US and cross borders in EMEs). The data on the 3-month Treasury bill was derived from Datastream and the data on the 10-year US bond was derived from the Federal Reserve Economic Data (FRED).
Interest rate differential (INT)
This is the second primary measure of the portfolio rebalancing channel. This variable captures the spatial rebalancing due to its country-specific nature, the rebalancing across countries and increased demand for developing country assets (Lim et al. 2014). The interest rate differential calculation:
Interest rate differential = yield 10-year government bond EME - yield 10-year T-bill The interest rate differential is therefore determined by both monetary policy in the US and in the EME, and therefor captures the relative attractiveness of investing in EMEs as compared to the US due to higher interest rate returns (Nier et al. 2014).
Growth differential (GRT)
This variable is a secondary measure of the portfolio balance channel. It is calculated as the percentage growth of the real GDP from the previous of the EME minus the percentage growth of the real GDP of the US from the previous quarter.
Growth differential = % growth real GDP EMEt-1 - % growth real GDP USt-1 This measure is used to capture growth prospects across EMEs and long-term return differentials. In previous research, the lagged growth differential has been proven a significant determinant of capital flows (IMF 2013 and Ahmed and Zlate 2013, in: Nier et al 2014, 6; Chen and Kahn 1997).
Purchasing Managers Index (PMI)
12
index for the manufacturing sector worldwide is a measure of manufacturing activity based on surveys by Markit. The PMI is used as a measure for growth prospects.
3-month Treasury bill (TBL)
As a primary indicator for the liquidity channel, the 3-month Treasury bill rate is used, as this is a clear reflection of liquidity constraints faced due to interest rates. Also, changes in the 3-month Treasury bill rate as a result of QE are measured. Whether a change in this variable translates into a change in money stock is uncertain because it gives of a price signal.
M2
The second measure for the liquidity channel is the lagged money supply in billions of USD. This is a direct measure of the quantity of liquidity in the economy in the
previous period.
The Volatility Index (VIX)
VIX is a ticker symbol for the Chicago Board Options Exchange Market Volatility Index, which is a popular measure of the implied volatility on S&P500 index options and an indicator of the expected volatility of stock market volatility for the next 30-day period. The VIX is a measure designed to capture market sentiment for investing in risky assets and is sometimes used as a measure for financial market uncertainty (Lim et al. 2014, 12; Nier et al. 2014, 8). Hence, it is used as a measure for the confidence channel.
GDP growth
As a standard determinant for the baseline capital flows, the lagged growth of the gross domestic product of the EME is used. This is the growth of the GDP at current prices. Quarterly GDP growth data was derived from the EME's CBs.
Overall risk rating (RISK)
As mentioned in the theoretical background, several studies have found the
investment climate of the EME is determinant of the capital flows to the developing country. As a measure of investment climate, this paper uses the relevant country's investment risk rating. This is a variable composed by The Economist Intelligence
Unit. It is based on the analysis and forecasts of credit risks and is regularly reviewed.
The measure also includes political, economic policy and economic structure risks.
Trade (TRD)
This variable is the sum of the EME's import and export as a percentage of the country's GDP at current prices. It is a good proxy for trade relations between the EME and the rest of the world. Previous studies have observed a positive effect on capital flows (Barroso 2013)
13
Total amount of foreign financing of the EME as a percentage of its GDP at current prices. This measures the EME's financial integration with the global financial system and its dependence on foreign financing. Bowman et al. (2014, 11) found a positive relation with an EME’s sensitivity of capital flows to US UMP.
Credit (CRD)
Total amount of credit provided to the residential public by domestic banks in EMEs. The amount of credit as a percentage of GDP presents a proxy for financial market development, which has been found to be a determinant of both size and composition capital flows to EMEs (Chen and Kahn 1997; Mody et al 2001).
Exchange rate (XRT)
The amount of foreign currency units worth one USD. The rationale behind including exchange rates as a variable lies, among others, in how an appreciation of
depreciation of a countries exchange rate leads to the amount of credit risk the
borrowers from foreign lenders in that country hold. If the currency of an EME or any other country appreciates, the residents that have borrowed in foreign currency have an increased balance sheet denoted in USD and are therefore more likely to pay off their loans. This in turn will lead to a decline in financing constraints, which could lead to increased capital flows to the country (Nier et al. 2014).
The possible reverse causality, however, is that capital flows could lead to an appreciation of the currency (Nier et al. 2014; Tillman 2014). This in turn could lead to a decrease in demand for products denoted in the EME’s currency, putting
downward pressure of the EME (Lavigne et al. 2014).
Quantitative easing (QE)
In addition to variables measuring the transmission channels between US QE and capital flows to EMEs and standard determinants, indicator variables are added to control for effects of QE of capital flows to EMEs unobserved and above the effects observed through the transmission channels. As presented in the next section, three models will be estimated. The first model includes one indicator variable (QE) for QE capturing all effects above the observed in all three QE episodes (2009Q1-2010Q3, 2010Q4-2011Q2 and 2012Q4-2014Q2, Lim et al. 2014, 38). The second model being estimated includes separate indicator variables for each QE episode (QE1, QE2 and QE3) and in the third model a variable (denoted QE_taper) is added indicating the tapering announcement in May 2013 (2013Q2) to capture possible effects it may have had not measured by the transmission channel variables.
3.3. Model presentation
The formal specification of the first model estimated is as follows:
GFFit = GFFi,t-1 + β1*YLDt + β2*INTit + β3*GRTit + β4*PMIt + β5*TBLt + β6*M2t + β7*VIXt + β8*GDPit + β9*RSKit + β10*TRDit + β11*EDBTit + β12*CRDit + β13*XRTit + β14*QEt +αi +Tt + uit
14
To control for time-specific effects the model includes a time-indicator variable denoted Tt in the model specification. The variable αi exhibits a fixed effects model and controls for country-specific effects. i refers to the country being observed and t refers to the time being observed.
The model in general is able to observe the effects of QE observed through measurements of its transmissions channels mentioned before. But through adding indicator variables for the periods of QE, it is also useful to observe the effects not observed through the portfolio, liquidity and confidence channels. Also, it includes standard determinants of capital flows as control variables.
The formal specification of the second model estimated is:
GFFit = GFFi,t-1 + β1*YLDt + β2*INTit + β3*GRTit + β4*PMIt + β5*TBLt + β6*M2t + β7*VIXt + β8*GDPit + β9*RSKit + β10*TRDit + β11*EDBTit + β12*CRDit + β13*XRTit + β14*QE1t + β15*QE2t + β16*QE3t +αi +Tt + uit The second model differs from the first model in that it measures whether there were effects of specific QE periods on gross capital flows than from the QE policies all together.
The formal specification of the third model to be estimated is:
GFFit = GFFi,t-1 + β1*YLDt + β2*INTit + β3*GRTit + β4*PMIt + β5*TBLt + β6*M2t + β7*VIXt + β8*GDPit + β9*RSKit + β10*TRDit + β11*EDBTit + β12*CRDit + β13*XRTit + β14*QE1t + β15*QE2t + β16*QE3t +
β17*QE_tapert +αi +Tt + uit
This model adds QE_taper to the equation which is also an indicator variable that measures the unobserved effects through transmission mechanism variables due to the tapering announcement by the fed in 2013Q2.
3.3. Diagnostic tests
The initial fixed effects regression is performed under the assumptions of
homoscedasticity and no presence of autocorrelation or cross sectional dependence among residuals. If these assumptions do not hold, this could lead to misspecification of the model. To test the homoscedasticity of the residuals, a test is performed based on a modified Wald statistic (Baum 2001). To test for autocorrelation between residuals in panel data, a Woolridge test will be performed (Woolridge 2010). To test for cross sectional dependence, test statistics proposed by Pesaran (2004) are used. Under the null hypothesis of the diagnostic tests, homoscedasticity and non-auto correlated residuals are assumed.
4. Results and discussion
15
In table 1, a summary of the descriptive statistics of the quarterly data for the entire period between 1 January 2005 and 31 December 2014 is presented. The average (median) amount of gross financial inflows in the five EMEs investigate is around 5181.084 (5777) million USD. The averages (medians) for the entire period and all five countries investigated are 5.472453% (6,8105%) for the interest rate differential, -1,33156% (-0,83%) for the growth differential, 19,679 (16,54) VIX points for VIX and 2817516 (2922371) million USD for the assets on the balance sheet of the FED.
In table 2 and 3, a distinction is made between the periods in which the FED executed QE policies and the periods in which it did not. Classifying the periods in which the FED engaged in QE, this research follows Lim et al. (2014, 38). According to the latter paper: QE1 occurred in the period 2009Q1-2010Q3, QE2 in the period 2010Q4-2011Q2 and QE in the period 2012Q4-2013Q3. In the period in which the FED engaged in QE, the average (median) gross financial flows to EMEs was 5647.41 (6037.42) million USD, compared to 4984.92 (5585.01) million USD in the period in which the FED did not engage in QE. Also, interest rate differentials turned out to be higher in the period with QE and growth differentials appear to have been smaller on average in the periods with QE. Investors’ appetite for risky investments seems to have been higher during QE periods, with the VIX at an average (median) of 24,13538 (21,68) in the period with QE compared to 17,5337 (15,4) in the period without QE.
Graph 1, 2, 3 and 4 present a graphical illustration of how gross financial flows, the interest differential, the 3-month T-bill rate and VIX behave between 2005Q1 and 2014Q4. As can be seen, from the start of QE1 in 2009Q1, gross financial flows experience a sharp increase in all sample countries, especially in Brazil. At the start of QE2, a similar but less strong movement can be observed. Again gross financial flows to Brazil experience the largest effect. At the start of QE3, a substantial increase in gross financial flows to all sample countries can be observed. At the end of 2013, however, around the time of the tapering announcement by the FED, the gross financial flows all seem to decrease again.
In table 2, the movements of the interest rate differential are shown over time. The overserved reaction to QE1 is modest, just as with the case of QE2. Interest rate differentials do not seem to react substantially, while according to the logic found in previous academic research, one would expect it to increase in response to QE
programs. The graph of the three-month T-bill rate shows a sharp decline between the start of 2007 and QE1, likely as a result of expansionary conventional monetary policy. From the start of QE1, the three-month T-Bill rate is maintained low, but the graph does not show any substantial responses to QE. The VIX, on the other hand, which is shown in graph 4, shows an increase at the start of all QE programs, of which in the first case a notable substantial increase and the latest a modest increase.
16 GFF 3-month T-bill rate M2 Yield curve PMI Interest differential Growth differential VIX GDP growth Risk rating Trade External debt Credit Exchange rate Mean 5181.08 1.37925 8667.17 1.95775 52.45 5.472435 23472.23 19.679 1.85317 44.5017 0.40588 0.88139 1.37953 1975.55 Std error 650.546 0.13094 111.050 0.07901 0.35424 0.35083 2006.974 0.62917 0.22969 0.48880 0.00976 0.03387 0.04595 279.785 Median 5777 0.15 8470.7 2.14 53 6.8105 16792 16.54 2.1 43.33 0.41270 0.70692 1.29430 13.0102 Std dev 9177.09 1.85190 1570.49 1.11730 5.00969 4.961492 26851.49 8.89780 3.24023 6.91261 0.13796 0.47774 0.64979 3956.75 Sample variance 842188 3.42951 246642 1.24836 25.0969 24.6164 7.21E+08 79.1707 10.4990 47.7841 0.01903 0.22823 0.42222 156558 Range 59422.93 4.88 5169.3 3.82 23.8 18.21 181318 34.3 14.7 31 0.53277 1.81677 2.29069 12383.8 Minimum -17646 0.02 6403.1 -0.26 35.5 -4.9 -37511 11.15 -8 32 0.14796 0.23830 0.35361 1.17257 Maximum 41776.5 4.9 11572.4 3.56 59.3 13.31 143807 45.45 6.7 63 0.68073 2.05507 2.64431 12385 Table 1: descriptive statistics entire period
GFF 3-month T-bill rate M2 Yield curve PMI Interest differential Growth differential VIX GDP growth Risk rating Trade External debt Credit Exchange rate Mean 5647.40 0.11307 8950.90 2.88538 52.9 7.202953 -2.24937 24.1353 1.365 43.9129 0.38555 0.84441 1.43575 1899.40 Std error 1240.53 0.00783 92.1379 0.06821 0.81947 0.483306 0.380511 1.11624 0.50298 0.76781 0.01633 0.05789 0.07572 472.845 Median 6037.41 0.12 8632.6 3.11 55.2 7.74 -2.25 21.68 2.26 43.33 0.39118 0.69182 1.44061 12.6703 Std dev 10001.5 0.06316 742.840 0.54996 6.60681 3.896538 3.044093 8.99947 4.02388 6.19035 0.13168 0.46678 0.61052 3812.20 Sample variance 100030 0.00399 551811. 0.30246 43.65 15.18301 9.266504 80.9906 16.1916 38.3204 0.01734 0.21788 0.37274 145328 Range 56786.9 0.19 2297 1.68 23.8 13.52 13.46 28.7 14.7 26 0.48456 1.69084 2.02058 11548.5 Minimum -15010. 0.02 8296.5 1.88 35.5 -0.21 -9.99 13.58 -8 32 0.14796 0.36422 0.54041 1.46166 Maximum 41776.5 0.21 10593.5 3.56 59.3 13.31 3.47 42.28 6.7 58 0.63252 2.0550 2.56099 11550 Table 2: descriptive statistics of QE period
17
Table 3: descriptive statistics of non-QE period GFF 3-month T-bill rate M2 Yield curve PMI Interest differential Growth differential VIX GDP growth
Risk rating Trade External debt Credit 3-month T-bill rate 0.07656 M2 -0.0289 -0.77526 Yield curve -0.04993 -0.88616 0.509893 PMI 0.160369 0.00873 0.185341 0.001555 Interest differential 0.005749 -0.55459 0.458418 0.505594 0.008885 Growth differential -0.26396 0.027835 0.177672 -0.16418 0.177807 -0.21634 VIX -0.08843 -0.44278 0.024184 0.493054 -0.57257 0.235404 -0.25206 GDP growth -0.07442 0.081447 0.018736 -0.12525 0.249781 -0.06076 -0.25739 -0.20755 Risk rating 0.255285 -0.39512 0.47154 0.279754 0.082131 0.139189 0.161709 0.049744 -0.12743 Trade -0.152 -0.0402 0.085563 -0.00399 0.048738 -0.34618 0.135715 -0.05892 0.450573 0.268694 External debt -0.49435 -0.01415 0.047687 -0.01083 -0.00345 0.033675 0.111999 -0.03014 0.191957 -0.71466 0.071013 Credit -0.18934 -0.35652 0.452741 0.235588 0.063393 0.581676 0.241052 0.023878 -0.35408 0.077823 -0.60753 0.119997 Exchange rate 0.002186 -0.0174 0.030191 0.012433 -0.00829 0.261135 -0.52748 3.98E-05 0.594229 -0.38806 0.045962 0.332005 -0.26134 Table 4: Correlation matrix describing statistics over entire observed period
GFF 3-month T-bill rate M2 Yield curve PMI Interest differential Growth differential VIX GDP growth Risk rating Trade External debt Credit Exchange rate Mean 4984.92 2.06462 8445.02 1.46577 52.1192 4.510038 -0.92015 17.5780 2.08476 44.7846 0.41521 0.89673 1.33671 1999.90 Std error 769.313 0.17399 158.643 0.09103 0.35684 0.458824 0.208721 0.71828 0.24651 0.64141 0.01235 0.04262 0.05810 351.961 Median 5585.08 1.615 7624.35 1.605 52.8 5.1535 -0.55 14.965 2.05 43.335 0.42394 0.75137 1.22008 13.0559 Std dev 8737.71 1.98386 1808.81 1.03798 4.0686 5.231408 2.379795 8.18969 2.81070 7.31328 0.14082 0.48408 0.66253 4012.97 Sample variance 76347716.55 3.93572 3271797.11 1.07741 16.5542 27.36763 5.663427 67.0711 7.90008 53.4841 0.01983 0.23434 0.43894 16104000.63 Range 0.37497 -1.6672 -1.3551 -1.2104 6.82072 -0.98842 -0.12633 4.70106 0.60845 -0.7819 -0.8617 -0.7247 -1.0772 0.62374 Minimum 0.20370 0.23639 0.52312 -0.1609 -2.1734 -0.28283 -0.54602 2.26231 -0.5983 0.37769 -0.0483 0.59694 0.34026 1.57959 Maximum 49628.9 4.88 5169.3 3.45 20.7 18.07 11.5 34.3 13.9 30 0.50932 1.78570 2.29069 12383.8
18
Graph 1 Graph 2
19
4.2. Diagnostic test results
As shown in table 6 below, both the modified Wald statistic and the Pesaran test for
autocorrelation show significant outcomes on a 1% significance level. To adjust the model for heteroscedasticity and cross sectional dependence, the regressions are performed using robust standard errors.
Specification
Modified Wald test for groupwise heteroscedasticity χ 2 P-value
50.28 0.0000 Woolridge test for autocorrelation F
P-value
0.509 0.5149 Pesaran’s test for cross sectional independence Test-statistic
P-value
-4.583 0.0000
Table 5: Diagnostic test results for the fixed effects regression model 4.3. Regression results
The estimates found in the adjusted fixed-effects models are shown in table 6. The full regression results can be found in the appendix of this paper.
Model 1 Model 2 Model 3
Constant 119480 119480 119480 (48015.54) (48015.54) (48015.54) Yield curve -17060.62** -17060.62** -17060.62** (4833.709) (4833.709) (4833.709) Interest differential 261.7519 261.7519 261.7519 (246.0601) (246.0601) (246.0601) Growth differential -284.4648 -284.4648 -284.4648 (387.7702) (387.7702) (387.7702) PMI 191.38 191.38 191.38 (255.0862) (255.0862) (255.0862) 3-month T-Bill -13125.71* -13125.71* -13125.71* (4517.681) (4517.681) (4517.681) M2 -7.078767 -7.078767 -7.078767 (3.92031) (3.92031) (3.92031) VIX -224.3713 -224.3713 -224.3713 (172.4175) (172.4175) (172.4175) GDP growth -82.66538 -82.66538 -82.66538 (362.1358) (362.1358) (362.1358) Risk rate -1284.39 -1284.39 -1284.39 (1658.343) (1658.343) (1658.343) Trade 661.1311 661.1311 661.1311 (23255.31) (23255.31) (23255.31) External debt -4688.456 -4688.456 -4688.456 (7479.842) (7479.842) (7479.842) Credit -501.7134 -501.7134 -501.7134 (5436.944) (5436.944) (5436.944) Exchange rate .4010742 .4010742 .4010742 (1.529591) (1.529591) (1.529591) All QE episodes 10784.66*** 4042.187 QE1 10784.66*** 10784.66*** (4042.187) (4042.187) QE2 4890.572 4890.572 (5799.445) (5799.445) QE3 -2851.519 -2851.519
20 (5427.191) (5427.191) QE tapering 5883.534 (5524.359) R2 (within) 0.3206 0.3206 0.3206 R2 (between) 0.0017 0.0017 0.0017 R2 (overall) 0.2324 0.2324 0.2324
Table 6: regression results. Note: numbers in parenthesis denote std. error values;*** denotes significance at 1%, ** denotes significance at 5% and * denotes significance at 10%
Using a significance level of 5%, of the transmission mechanism variables, it is possible to conclude that QE has had effect on gross financial flows merely the portfolio transmission mechanism. On a 10% significance level, however, it can also be inferred that QE has had an effect through the liquidity channel. Both the coefficient of the yield curve and the 3-month T-bill rate are significant in determining gross financial flows to EMEs on the last mentioned
significance level. The yield curve, which represents the change in temporal demand, the portfolio channel both within the US and to foreign countries has a negative coefficient. This is consistent with the expected workings of the mechanism. QE is likely to decrease long-term yield as the demand for long-term yield is increased. As a result investors will rebalance their portfolio to assets with a higher risk-adjusted return, both domestically and in foreign markets. Which explains the negative coefficient.
A possible explanation for the larger size of the coefficient of the yield curve as compared to the interest rate differential and growth rate differential as transmission mechanisms for the portfolio channel is offered by Fratzscher et al. (2013). According to the latter, FED liquidity injections and LSAP announcements cause foremost portfolio rebalancing across countries. Whereas outright repurchases by the FED mainly cause portfolio rebalancing across classes. All QE episodes in the observed period concern asset repurchases.
Also the significant negative coefficient of the 3-month Treasury bill rate is according to the functioning of the transmission mechanism. As a result of increased liquidity, the 3-month Treasury bill rate decreased, as shown in graph 3. Consequently, the financing constraints decrease and gross financial flows to EMEs increase.
A remarkable result is that both transmission mechanism variables that are found to have had significant effects (on a 10% significant level) are global ‘push’ factors not influenced by country-specific factors. The stronger effect of ‘push’ factors compared to ‘pull’ factors is in accordance with the research by Lim et al (2014), Fratzscher et al. (2013) and Morgan (2011). It, however, contradicts earlier findings by Mody et al. (2001) and Carlson and Hernández (2002).
Also from the results on the QE indicators, several conclusions can be drawn. First of all, the dummy variable representing the entire QE period enters with significance. This means that above the effects of QE through the observed transmission channels, on a 1% significance level, it can be concluded that there have also been unobserved effects of QE on gross financial flows over and above the effects observed through the transmission channels.
Also the findings when breaking up the QE dummy variable into three dummy variables, denoting three different QE episodes, are notable. These measures of unobserved effects display a diminishing effect of QE programs over time, with the QE2 having a decreased and insignificant
21
effect and QE3 even showing a negative, although also insignificant, effect. One possible
explanation includes the fact that the sizes of the QE3 and QE2 programs were notably smaller as compared to QE1: $2.1 trillion in QE1 compared to $600 billion and $85 billion in respectably QE2 and QE3. The finding that the first two QE episodes had a larger effect compared to the last one is consistent with findings in preceding literature (Cúrdia and Ferrero 2013; Krishnamurthy and Vissing-Jorgensen 2013, in: Lim et al. 2014, 15). Including the dummy variable to account for effects of the tapering announcement in 2013Q2 showed no significant results.
Based on the model results, it can be concluded that the control variables have not been significant determinants of gross financial flows to our sample in the period observed. This result, however, is contradictory to much preceding literature (Lim et al. 2014; Nier et al. 2014; Ahmed and Zlate 2014). It is important, however, to take into account that the periods investigated and the number of observations across the amount of countries has often been much higher in
preceding research. Lim et al. (2014) for instance, observes 60 countries from 2000Q1 to 2013Q2, also using quarterly data. Moreover, Ahmed and Zlate (2014) covers 12 countries between
2002Q1 and 2013Q2 and also uses quarterly data. Also, samples used previously are not identical to the one used in this paper. Therefore, there is the possibility of a difference in the existing relationship between capital flows and determinants.
5. Conclusion
This study was conducted with the purpose of examining the effects of US QE on capital flows to EMEs and being able to observe how these effects were transmitted. In previous research, much attention has been dedicated to the transmission of US UMP to EMEs through capital flows and on the standard determinants of capital flows to EMEs. Narrowing down, the transmission channels could be categorized in three categories: portfolio, liquidity and confidence channels. Even though a rather large amount of attention has been dedicated, there still was a notable amount of ambiguity among researchers about the overall effects and composition of effects of QE on EMEs and their capital inflows. This study attempted to contribute to the existing
literature on the transmission of UMP in advanced economies to EMEs. The variables mainly of interest were the transmission mechanism variables: yield curve, interest rate differential, growth differential, 3-month Treasury bill, M2 and VIX. Above the transmission mechanism variables used to capture the effects of QE through transmission channels, QE dummy variables accounting for both the entire QE period, different QE episodes and different QE episodes and the tapering announcement were added to capture the effects above the observed through transmission
channels. The results present significant evidence for effects of US QE on capital flows to EMEs both through the observed transmission channels and the unobserved over and above the latter. The transmission channels at work were observed to be the portfolio and liquidity channel. Contrary to previous literature, however, the amount of transmission mechanism variables that turned out to enter with significance were low (two out of seven).
In concluding on this study, it should be noted that the sample size of 197 observations was small relative to much previous research conducted. This could explain both the low amount
22
of transmission mechanism variables found to be significant and the insignificance of the control variables that are widely acknowledged in academics. Recommendations for further research include the workings of standard determinants of capital flows in times of US UMP and the transmission channels of the up till now unobserved effects of US QE on EMEs over and above the observed effects through the transmission channels.
23
References
Ahmed, S., and Zlate, A. 2014. “Capital flows to emerging market economies: A brave new world?” Journal of international money and finance 48:221-248
Barroso, R.B.J.B., Pereira da Silva, L.A., and Soares Sales, A. 2013. “Quantitative Easing and Related Capital Flows into Brazil: measuring its effects and transmission channels through a rigorous counterfactual evaluation” Bank of Brazil Working Paper 313 Baum, C.F. 2001. “Residual diagnostics for cross-section time series regression models” Stata
Journal 1(1):105-116
Bouraoui, T. 2015. “The effect of reducing quantitative easing on emerging markets” Applied
Economics 47:1562-1573
Bowman, D., Londono, J.M., and Spariza, H. 2014. “U.S. Unconventional Monetary Policy and Transmission to Emerging Market Economies”, International Finance Discussion Papers No. 1109
Carlson, M., and Hernández, L. 2002. “Determinants and Repercussions of the Composition of Capital Flows”, International Finance Discussion Paper No. 717
Chen, Z. and Kahn, M.S. 1997. “Patterns of Capital Flows to Emerging Markets: A Theoretical Perspective”, International Monetary Fund Working Paper No. 97/13
Chen, Q., Filardo, A., He, D., and Zhu, F. 2012. “International spillovers of central bank policies”, Bank of International Settlements Papers No. 66
Cho, D and Rhee, C. 2013. “Effects of Quantitative Easing on Asia: Capital Flows and Financial Markets” Asian Development Bank Working Paper Series, No. 350
Fratzscher, M., Lo Duca, M., and Straub, R. 2013. “Quantitative Easing, Portfolio Choice and International Capital flows” European Central Bank, Working Paper Series, No. 1557 Hawksworth, J. and Cookson, G. 2008 “Beyond the BRICs: a broader look at emerging market
growth prospects” The World in 2050 March 2008
Lavigne, R., Sarker, S. and Vasishtha, G. 2014. “Spillover Effects of Quantitative Easing on Emerging-Market Economies” Bank of Canada Review Autumn 2014:23-33
Lenza, M., Pill, H., and Reichlin, L. 2010. “Monetary policy in exceptional times”, Economic
Policy April 2010:295-339
Lim, J.J., Mohapatra, S., and Stocker, M. 2014. “Tinker, Taper, QE, Bye? The effect of Quantitative Easing on Financial Flows to Developing Countries”, World Bank Policy Research Working Paper Series No. 6820
Mody, A., Taylor, M.P., and Kim, J.Y. 2001. “Modeling Fundamentals for Forecasting Capital Flows to Emerging Markets” International Journal of Finance and Economics 6:201-216 Morgan, P.J. 2011. “Impact of US Quantitative Easing Policy on Emerging Asia” Asian
Development Bank Institute Working Paper Series No. 321
Nier, E., Sedik, T.S., and Mondino, T. 2014. “Gross Private Capital Flows to Emerging Markets: Can the Global Financial Cycle Be Tamed?” International Monetary Fund Working Paper 14/196
24 Cambridge Working Papers in Economics 0435
Taylor, M.P, and Sarno, L. 1997. “Capital Flows to Developing Countries: Long- and Short-Term Determinants” The World Bank Economic Review 11(3):451-70
Tillman, P. 2014. “Unconventional Monetary Policy Shocks and the Spillovers to Emerging Markets” Hong Kong Institute for Monetary Research Working Paper No. 18
Wooldridge, J.M. 2010. “Econometric analysis of cross section and panel data” MIT press books 1
25
Appendix
Variable Data source(s)
Gross financial inflows IFS, BIS
Yield curve Datastream, IFS
Interest rate differential IFS
Growth differential Datastream, OECD
PMI Datastream
3-month Treasury bill Datastream
M2 Datastream
VIX Datastream
GDP growth Datastream, FRED
Trade/GDP IFS, BIS, WDI, local CBs, FRED
External debt/GDP Datastream, BIS, local CBs, FRED
Credit/GDP BIS, local CBs, FRED
Exchange rate Datastream
26
Model 1 Model 2 Model 3
Constant 119480** 119480** 119480** (48015.54) (48015.54) (48015.54) Yield curve -17060.62*** -17060.62*** -17060.62*** (4833.709) (4833.709) (4833.709) Interest differential 261.7519 261.7519 261.7519 (246.0601) (246.0601) (246.0601) Growth differential -284.4648 -284.4648 -284.4648 (387.7702) (387.7702) (387.7702) PMI 191.38 191.38 191.38 (255.0862) (255.0862) (255.0862) 3-month T-Bill -13125.71*** -13125.71*** -13125.71*** (4517.681) (4517.681) (4517.681) M2 -7.078767* -7.078767* -7.078767* (3.92031) (3.92031) (3.92031) VIX -224.3713 -224.3713 -224.3713 (172.4175) (172.4175) (172.4175) GDP growth -82.66538 -82.66538 -82.66538 (362.1358) (362.1358) (362.1358) Risk rate -1284.39 -1284.39 -1284.39 (1658.343) (1658.343) (1658.343) Trade 661.1311 661.1311 661.1311 (23255.31) (23255.31) (23255.31) External debt -4688.456 -4688.456 -4688.456 (7479.842) (7479.842) (7479.842) Credit -501.7134 -501.7134 -501.7134 (5436.944) (5436.944) (5436.944) Exchange rate .4010742 .4010742 .4010742 (1.529591) (1.529591) (1.529591) All QE episodes 10784.66*** 4042.187 QE1 10784.66*** 10784.66*** (4042.187) (4042.187) QE2 4890.572 4890.572 (5799.445) (5799.445) QE3 -2851.519 -2851.519 (5427.191) (5427.191) QE tapering 5883.534 (5524.359) R2 (within) 0.3206 0.3206 0.3206 R2 (between) 0.0017 0.0017 0.0017 R2 (overall) 0.2324 0.2324 0.2324 F 1.46 1.46 1.46 Prob>F 0.0479 0.0479 0.0479
Appendix B: regression results using default standard errors. Note: numbers in parenthesis denote std. error values;*** denotes significance at 1%, ** denotes significance at 5% and * denotes significance at 10%
27
Appendix C: STATA output modified Wald test for groupwise heteroscedasticity
Appendix D: STATA output Wooldridge test for autocorrelation in panel data
Appendix E: STATA output Pesaran’s test for cross sectional independence Prob>chi2 = 0.0000
chi2 (5) = 50.28
H0: sigma(i)^2 = sigma^2 for all i in fixed effect regression model
Modified Wald test for groupwise heteroskedasticity . xttest3
Prob > F = 0.5149 F( 1, 4) = 0.509 H0: no first-order autocorrelation
Wooldridge test for autocorrelation in panel data > E
> PMI TBILL3MONTH M2 VIX GDP_GROWTH RISK_RATING TRADEGDP EXTERNALDEBTGDP CREDITGDP EXCHANGE_RATE Q . xtserial GFF1_alternative YIELDCURVEDIFFERENCEBETWEEN INTEREST_DIFFERENTIAL GROWTH_DIFFERENTIAL
Average absolute value of the off-diagonal elements = 0.278
Pesaran's test of cross sectional independence = -4.583, Pr = 0.0000 . xtcsd, pesaran abs
28
Appendix F: STATA output regression model 1 using robust standard errors rho .14311671 (fraction of variance due to u_i) sigma_e 7959.2581 sigma_u 3252.7967 _cons 119480 64796.9 1.84 0.139 -60424.98 299385.1 40 5671.233 11126.29 0.51 0.637 -25220.31 36562.77 39 6906.256 10512.44 0.66 0.547 -22280.96 36093.47 38 19524.76 9851.25 1.98 0.119 -7826.698 46876.21 37 19929.65 14348.03 1.39 0.237 -19906.86 59766.16 36 14254.15 10101.16 1.41 0.231 -13791.18 42299.48 35 15165.41 10221.89 1.48 0.212 -13215.1 43545.92 34 -7752.645 8524.909 -0.91 0.415 -31421.59 15916.3 33 -13636.18 8158.964 -1.67 0.170 -36289.1 9016.738 32 -23410.06 4912.459 -4.77 0.009 -37049.23 -9770.887 31 0 (omitted) 30 -16765.87 6093.081 -2.75 0.051 -33682.97 151.2363 29 2627.612 3805.246 0.69 0.528 -7937.445 13192.67 28 0 (omitted) 27 -6278.691 3230.044 -1.94 0.124 -15246.73 2689.35 26 -6085.353 5035.267 -1.21 0.293 -20065.49 7894.789 25 -5894.088 4112.147 -1.43 0.225 -17311.24 5523.062 24 -11045.57 8408.852 -1.31 0.259 -34392.28 12301.15 23 -4422.943 4817.843 -0.92 0.411 -17799.42 8953.534 22 -4340.416 6529.473 -0.66 0.543 -22469.14 13788.31 21 0 (omitted) 20 -3810.35 4452.484 -0.86 0.440 -16172.43 8551.728 19 610.3564 7289.174 0.08 0.937 -19627.64 20848.35 18 5600 6806.1 0.82 0.457 -13296.76 24496.76 17 0 (omitted) 16 0 (omitted) 15 4530.435 6299.02 0.72 0.512 -12958.45 22019.32 14 4297.764 3653.995 1.18 0.305 -5847.351 14442.88 13 -2893.447 5676.157 -0.51 0.637 -18652.99 12866.09 12 946.8935 1590.553 0.60 0.584 -3469.188 5362.975 11 5478.004 2507.828 2.18 0.094 -1484.844 12440.85 10 4867.258 4252.79 1.14 0.316 -6940.38 16674.9 9 0 (omitted) 8 4062.36 3233.004 1.26 0.277 -4913.899 13038.62 7 6578.255 7117.138 0.92 0.408 -13182.09 26338.6 6 -4012.91 3299.273 -1.22 0.291 -13173.16 5147.34 5 -810.0988 898.848 -0.90 0.418 -3305.701 1685.503 4 -4516.223 5009.111 -0.90 0.418 -18423.75 9391.299 3 -5564.716 6736.559 -0.83 0.455 -24268.4 13138.97 2 -9190.68 8715.482 -1.05 0.351 -33388.74 15007.38 QUARTER QE 10784.66 1781.97 6.05 0.004 5837.119 15732.2 EXCHANGE_RATE .4010742 .6422458 0.62 0.566 -1.382086 2.184234 CREDITGDP -501.7134 5921.291 -0.08 0.937 -16941.85 15938.43 EXTERNALDEBTGDP -4688.456 3972.15 -1.18 0.303 -15716.91 6340 TRADEGDP 661.1311 14305.67 0.05 0.965 -39057.77 40380.03 RISK_RATING -1284.39 1427.787 -0.90 0.419 -5248.564 2679.783 GDP_GROWTH -82.66538 107.9563 -0.77 0.487 -382.4002 217.0695 VIX -224.3713 136.7289 -1.64 0.176 -603.9916 155.249 M2 -7.078767 5.210393 -1.36 0.246 -21.54514 7.387604 TBILL3MONTH -13125.71 5510.156 -2.38 0.076 -28424.35 2172.94 PMI 191.38 107.8141 1.78 0.151 -107.96 490.72 GROWTH_DIFFERENTIAL -284.4648 331.6349 -0.86 0.439 -1205.231 636.3012 INTEREST_DIFFERENTIAL 261.7519 157.2733 1.66 0.171 -174.9087 698.4124 YIELDCURVEDIFFERENCEBETWEEN -17060.62 5848.892 -2.92 0.043 -33299.74 -821.489 GFF1_alternative Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust
(Std. Err. adjusted for 5 clusters in COUNTRY_CODE) corr(u_i, Xb) = -0.5123 Prob > F = .
F(4,4) = . overall = 0.2324 max = 40 between = 0.0017 avg = 39.4 R-sq: within = 0.3206 Obs per group: min = 39 Group variable: COUNTRY_CODE Number of groups = 5 Fixed-effects (within) regression Number of obs = 197
29
Appendix G: STATA output regression model 2 using robust standard errors rho .14311671 (fraction of variance due to u_i) sigma_e 7959.2581 sigma_u 3252.7967 _cons 119480 64796.9 1.84 0.139 -60424.98 299385.1 40 5671.233 11126.29 0.51 0.637 -25220.31 36562.77 39 6906.256 10512.44 0.66 0.547 -22280.96 36093.47 38 19524.76 9851.25 1.98 0.119 -7826.698 46876.21 37 19929.65 14348.03 1.39 0.237 -19906.86 59766.16 36 14254.15 10101.16 1.41 0.231 -13791.18 42299.48 35 15165.41 10221.89 1.48 0.212 -13215.1 43545.92 34 5883.534 8881.812 0.66 0.544 -18776.33 30543.4 33 0 (omitted) 32 -9773.881 10950.49 -0.89 0.423 -40177.32 20629.55 31 0 (omitted) 30 -16765.87 6093.081 -2.75 0.051 -33682.97 151.2363 29 2627.612 3805.246 0.69 0.528 -7937.445 13192.67 28 0 (omitted) 27 -6278.691 3230.044 -1.94 0.124 -15246.73 2689.35 26 -191.2643 8394.195 -0.02 0.983 -23497.29 23114.76 25 0 (omitted) 24 -5151.478 11733.8 -0.44 0.683 -37729.72 27426.77 23 -4422.943 4817.843 -0.92 0.411 -17799.42 8953.534 22 -4340.416 6529.473 -0.66 0.543 -22469.14 13788.31 21 0 (omitted) 20 -3810.35 4452.484 -0.86 0.440 -16172.43 8551.728 19 610.3564 7289.174 0.08 0.937 -19627.64 20848.35 18 5600 6806.1 0.82 0.457 -13296.76 24496.76 17 0 (omitted) 16 0 (omitted) 15 4530.435 6299.02 0.72 0.512 -12958.45 22019.32 14 4297.764 3653.995 1.18 0.305 -5847.351 14442.88 13 -2893.447 5676.157 -0.51 0.637 -18652.99 12866.09 12 946.8935 1590.553 0.60 0.584 -3469.188 5362.975 11 5478.004 2507.828 2.18 0.094 -1484.844 12440.85 10 4867.258 4252.79 1.14 0.316 -6940.38 16674.9 9 0 (omitted) 8 4062.36 3233.004 1.26 0.277 -4913.899 13038.62 7 6578.255 7117.138 0.92 0.408 -13182.09 26338.6 6 -4012.91 3299.273 -1.22 0.291 -13173.16 5147.34 5 -810.0988 898.848 -0.90 0.418 -3305.701 1685.503 4 -4516.223 5009.111 -0.90 0.418 -18423.75 9391.299 3 -5564.716 6736.559 -0.83 0.455 -24268.4 13138.97 2 -9190.68 8715.482 -1.05 0.351 -33388.74 15007.38 QUARTER QE3 -2851.519 9527.586 -0.30 0.780 -29304.34 23601.3 QE2 4890.572 3439.7 1.42 0.228 -4659.566 14440.71 QE1 10784.66 1781.97 6.05 0.004 5837.119 15732.2 EXCHANGE_RATE .4010742 .6422458 0.62 0.566 -1.382086 2.184234 CREDITGDP -501.7134 5921.291 -0.08 0.937 -16941.85 15938.43 EXTERNALDEBTGDP -4688.456 3972.15 -1.18 0.303 -15716.91 6340 TRADEGDP 661.1311 14305.67 0.05 0.965 -39057.77 40380.03 RISK_RATING -1284.39 1427.787 -0.90 0.419 -5248.564 2679.783 GDP_GROWTH -82.66538 107.9563 -0.77 0.487 -382.4002 217.0695 VIX -224.3713 136.7289 -1.64 0.176 -603.9916 155.249 M2 -7.078767 5.210393 -1.36 0.246 -21.54514 7.387604 TBILL3MONTH -13125.71 5510.156 -2.38 0.076 -28424.35 2172.94 PMI 191.38 107.8141 1.78 0.151 -107.96 490.72 GROWTH_DIFFERENTIAL -284.4648 331.6349 -0.86 0.439 -1205.231 636.3012 INTEREST_DIFFERENTIAL 261.7519 157.2733 1.66 0.171 -174.9087 698.4124 YIELDCURVEDIFFERENCEBETWEEN -17060.62 5848.892 -2.92 0.043 -33299.74 -821.489 GFF1_alternative Coef. Std. Err. t P>|t| [95% Conf. Interval] Robust
(Std. Err. adjusted for 5 clusters in COUNTRY_CODE) corr(u_i, Xb) = -0.5123 Prob > F = .
F(4,4) = . overall = 0.2324 max = 40 between = 0.0017 avg = 39.4 R-sq: within = 0.3206 Obs per group: min = 39 Group variable: COUNTRY_CODE Number of groups = 5 Fixed-effects (within) regression Number of obs = 197
