International Transmission Channels of U.S. Quantitative Easing: Evidence from Canada

Transcription

1 International Transmission Channels of U.S. Quantitative Easing: Evidence from Canada Tatjana Dahlhaus Kristina Hess Abeer Reza Bank of Canada Bank of Canada Bank of Canada May 5, 214 PRELIMINARY AND INCOMPLETE DO NOT QUOTE Abstract This paper empirically assesses the spillover effects of the recent Quantitative Easing program undertaken by the U.S. Federal Reserve on the Canadian economy. The Federal Reserve responded to the 28 financial crisis and ensuing recession by reducing the policy rate to near zero, and subsequently conducting purchases of public and private assets in an unprecedented scale in an attempt to reduce long-term rates, effectively quadrupling the size of its balance sheet by the end of 213. We attempt to quantify the effects of these programs on the real Canadian Economy in a FAVAR framework, and assess the transmission channels for spillover. We construct a counter-factual scenario where the fed balance sheet grows at a rate prevalent prior to the crisis. We find that QE boosted Canadian output. The spillovers where transmitted mainly through the financial channel. The trade channel, however, provided a source of leakage for Canada, as the Canadian currency appreciated vis-à-vis the U.S., and net exports declined in comparison to the counterfactual scenario. The opinions expressed in this paper are the authors alone, and should not be attributed to the Bank of Canada. We thank Robert Lavigne, Christiane Baumeister, Michael Ehrmann, and seminar participants at the Bank of Canada for helpful comments and suggestions. All errors and omissions are our own. dahl@bankofcanada.ca hess@bankofcanada.ca reza@bankofcanada.ca. Corresponding author. 1

2 1 Introduction In the aftermath of the global financial crisis, many central banks turned to balance sheet expansion, or Quantitative Easing (QE), as a tool for providing monetary stimulus when their conventional policy interest rates became ineffective at their respective lower bounds. The U.S. Federal Reserve, in particular, responded to the crisis and ensuing recession by reducing the policy rate to near zero, and subsequently purchasing longterm assets in an unprecedented scale to bring down long-term interest rates, effectively quadrupling the size of its balance sheet by the end of 213 in the process. 1 Since then, the economic literature has focused on identifying the transmission channels of QE to the domestic real economy from a theoretical perspective (Bernanke [212], Woodford [212]), as well as quantifying their effects empirically (Lenza et al. [21], Chung et al. [212], Baumeister and Benati [213]). Given the global reach of the crisis that provoked such unprecedented policy action, however, it stands to reason that such policies would have large international spillover effects. Yet, there is a dearth of evidence on the international transmission channels of QE on the real economies of countries that did not adopt such programs. In this paper, we fill this gap by assessing the spillover effect of the Fed s QE program to the real economy of a non-qe country, namely, Canada. Specifically, we (a) quantify the effect of the Fed s long-term asset purchases on the Canadian real economy, and (b) identify the principal channels of policy spillover. We conduct our analysis in a two-country Factor-Augmented Vector Auto-Regression (FAVAR) framework, and consider a counter-factual scenario where the Fed s asset purchases from mid-28 followed a pattern prevalent before the crisis. Our research questions make FAVAR (Bernanke et al. [25], Boivin and Giannoni [27], Charnavoki and Dolado [214]) the natural framework to consider, as it allows dynamic interactions between a large number of financial and real variables that are necessary in assessing different spillover channels simultaneously. We choose Canada as a study case for assessing channels of spillovers because of its high trade and financial linkage with the U.S. As such, we expect spillovers to be strong and easily observable. Despite their close ties, however, the pre-crisis resilience of the Canadian financial sector prevented financial contagion during the crisis at a scale seen in other parts of the world. 1 Although the first wave of asset purchases were aimed at preventing the seizing up of credit markets and included short-term emergency lending to financial firms, purchases of agency debt and asset-backed securities, subsequent programs targeted long-term mortgage-backed securities and government bonds in an effort to bring down long-term interest rates. In this paper, we assess the effects of the later aspect of QE. For a discussion on the difference between the two aspects of balance sheet expansion, see Lenza et al. [21], and Kozicki et al. [211]. 2

3 As such, the Canadian recession following the crisis was milder in comparison to the one in the U.S., with less concerns for financial instability. Consequently, the Bank of Canada did not deem it necessary to implement QE, as the policy rate remained above its effective lower bound, albeit at historically low levels, till date. 2 Moreover, Canada is a small open economy with a floating exchange rate, with a long tradition of independent monetary policy, and availability of quality data. The latter is of particular practical importance when considering a FAVAR analysis. Finally, Canada is also a commodity exporter, allowing for a richer exploration of possible spillover channels of QE. In a closed economy context, the literature considers the transmission of QE to the domestic real economy to occur through the same channels as does conventional monetary policy stimulus (Bernanke [212], Woodford [212]). Large scale purchases of specific long-term assets by the Fed increases their prices and reduces their yields. As long as assets in an investor s portfolio are imperfectly substitutable, selling off long-term treasury bonds to the Fed increases her demand for other assets in search of higher yields. Through this portfolio balance effect, yields in all asset-classes fall, and their prices rise. A rise in asset prices boosts private consumption through wealth effects. Inter-temporal substitution will also provide incentives to economic agents to take advantage of this temporary fall in yields and increase consumption. At the same time, a rise in asset prices relative to the replacement cost of capital would, according to the traditional Q theory of investment, provide incentives for firms to invest in capital. To the extent that large-scale asset purchases act as a signaling channel for the Fed s willingness on keep interest rates low for an extended period of time (Bauer and Neely [212]), confidence would rise, and credit conditions would ease. Both of these channels would then amplify the positive effect on consumption and investment. In an open economy context, however, there can be substantive leakages through international trade channels. The Mundell-Flemming-Dornbusch overshooting model provides ambiguous implications of the effect of a monetary easing (Rogoff [22]). Through the expenditure switching effect, a monetary expansion in the U.S. would result in an appreciation of the future domestic exchange rate, and hence a depreciation of the home currency on impact. Terms of trade would then deteriorate, making home goods cheaper to foreigners. The resulting increase in net exports is beneficial to the domestic economy, but detracts from the real output of the foreign economy through a beggar-thy-neighbor 2 At the onset of the crisis, however, Canada readily provided term-liquidity facilities to financial market participants to stabilize the financial system and limit the repercussions of the crisis to the Canadian economy (Zorn et al. [29]). 3

4 effect. The income-absorption effect, however, implies that as long as expansionary monetary policy in the home country drives up domestic income, home demand for imports will rise, providing leakages for the home country through reduced net exports, and boosting the economy of the foreign country that is exporting to the home country. In short, whether Canada benefits from U.S. expansion through QE depends on which of the two effects dominate. A second layer of ambiguity stems from Canada s role as a commodity producer. Even when the income absorption effect dominates in the U.S., higher import demand for Canadian energy products may drive up the Canadian exchange rate, making tradable goods in other Canadian industries less competitive in a Dutch-disease-type mechanism (Charnavoki and Dolado [214]). In the presence of global financial market integration, however, the signaling and portfolio balance channels that reduce U.S. long-term spreads and increase U.S. asset prices may have similar effects on corresponding Canadian variables. As investors move away from U.S. long-term bonds to other assets, they may well choose to invest in Canadian long-term bonds and assets, driving prices up, and yields down in Canada. Indeed, Fratzscher et al. [213] find that QE boosted equity prices worldwide, including in Canada, while Bauer and Neely [212] show that long-term Canadian yields decreased following U.S. LSAP announcements. A reduction in Canadian interest rates would then improve Canadian consumption and investment through the same channels of transmission described above. We find that the implementation of QE had increased U.S. output by about 5 percent and Canadian output by a bit less than 5 percent by 213Q3, when compared with the counter-factual scenario. Because of QE, U.S. long-term spreads decline, and asset prices rise, boosting both investment and consumption, though the latter is not statistically significant at the 68%-level. The transmission to Canada occurs mostly through the financial market. Although Canada did not implement QE, long-term spreads in Canada falls, and asset prices in the Canadian stock market rise. The trade channel provides a source of leakage for Canada, as the Canadian exchange rate appreciates due to QE, reducing Canadian net exports world-wide. Exports in most product categories, however, were higher because of QE. The reduction in net exports can then be explained by a higher increase in imports than exports. This is consistent with a scenario where reductions in Canadian interest rates work as a monetary expansion from the Canadian perspective, and the income-absorption effect works as a leakage on the now-stronger Canadian economy. The finding that the financial channel dominates the trade channel in transmitting 4

5 positive spillovers from a U.S. monetary expansion is consistent with the transmission of conventional monetary policy documented by Kim [21] in small SVAR. Since the beginning of the foray into unconventional policy, a number of papers have attempted to quantify the real effects of QE on the aggregate real domestic economies of the central banks conducting such policies (cf. Lenza et al. [21] for an analysis on the euro area, Kapetanios et al. [212] for the U.K., Baumeister and Benati [213] for the U.S. and U.K., Schenkelberg and Watzka [213] for Japan, and Gambacorta et al. [212] for an international comparison). Considerable attention has also focused on the spillovers of QE to international financial markets (Neely [21], Bauer and Neely [212]), and the outflow of capital from the US to countries offering higher returns (Fratzscher et al. [213]). This has prompted a debate on the need for international policy co-ordination (Dedola et al. [213]), and the independence of central banks that do not participate in co-ordinated accommodation with the U.S. (Rey [213]). To the best of our knowledge, however, this is the only paper that attempts to assess the international spillovers of QE on a real scale. In this sense, it complements the literature on the real spillover of conventional monetary policy (Kim [21], Kazi et al. [213]). This paper also differs from the earlier studies on the effects of QE in two important ways. First, while Kapetanios et al. [212] and Lenza et al. [21] conduct their analyses using large-scale Bayesian VAR s, we estimate a two-country FAVAR with coefficient restrictions. This allows us to study the transmission of U.S. policy through dynamic interactions between a large number of financial and real variables, while at the same time restricting U.S. policy from being influenced by Canadian data. We believe this is the appropriate framework for studying relationships between a large country such as the U.S., and a small open economy, such as Canada (Charnavoki and Dolado [214]). Second, while earlier studies such as Kapetanios et al. [212] and Baumeister and Benati [213] estimate the effect of QE by constructing counter-factual scenarios that proxy U.S. policy through its intended, yet unobserved, effects on long-term spreads, we consider a counter-factual scenario that directly measure policy action through the observable expansion of the Fed s balance sheet. Our approach is closer to the definition of QE (Woodford [212]), and takes advantage of 19 quarters of data on measurable policy action. In this sense, our approach is closer to Gambacorta et al. [212] who consider shocks to central bank assets that produce no movement in their respective policy rates in a dynamic panel setting. The rest of the paper is organized as follows. Section 2 describes the two-country FAVAR setup, as well as the data and estimation process. Section 3 documents the 5

6 close ties between the U.S. and Canadian economies prior to the crisis, as well as the transmission of a conventional monetary policy shock from the U.S. to Canada in an attempt to set up expectations for the QE counter-factual exercise. Section 4 describes the counter-factual setup and the results of the exercise for both aggregate real and financial variables, as well as some disaggregated variables. Section 5 concludes. 2 Empirical Framework In this section we present the empirical framework which allows us to study the real effects of U.S. QE on Canada and the associated transmission channels. We first introduce the broad dataset containing U.S. as well as Canadian time series. Then, the specific FAVAR model used in this paper and its estimation are discussed. 2.1 Data We use quarterly data from 123 U.S. and 149 Canadian series ranging from 1982Q1 through 213Q3. The start date is restricted by the availability of Canadian disaggregated data. The U.S. series includes real data from national income, industrial production, employment by industry, as well as data on housing starts and sales, manufacturers orders and inventory, and a number of different price indexes. Movements in the financial sector is captured through stock prices, nominal and real exchange rates, interest rates of varying maturity, bond prices, monetary aggregates, and commercial bank balance sheet and lending conditions. We also include confidence indexes from the Conference Board and University of Michigan surveys on business sentiment. The QE variable represents the Federal Reserve s balance-sheet holding of Mortgage-backed securities and treasury bonds with maturities of more than 5 years. A more detailed discussion on the choice of this variable is provided in section 4 while describing the counter-factual setup. The Canadian series include real data on national accounts, Canadian exports and imports by product, and by regional destination, flows from the balance of payments account, and industry-level GDP and employment. We also include Canadian house prices and housing starts, a number of relative price and oil price indexes. Movement in the financial sector is captured through the Toronto Stock Exchange (TSX) price index, nominal and trade-weighted real exchange rates, interest rates of various maturities, as well as balance-sheet conditions from commercial banks and private industry. We capture Canadian business confidence by the Conference Board confidence index. 6

7 Most of our data series, both U.S. and Canadian, are downloaded via HAVER analytics. We construct the industry-level GDP for Canada by combining information from tables and from the Canadian Socioeconomic Information Management (CANSIM) database maintained by Statistics Canada. 3 The series are transformed to induce stationarity, and standardized before estimation. The appendix provides detailed descriptions, sources, and transformation codes for all series considered. 2.2 Empirical Model In order to shed light on the spillover effects of U.S. QE, we employ the FAVAR model introduced by Bernanke et al. [25]. A FAVAR model allows us to study the effects of U.S. policy on a broad set of U.S. as well as Canadian time series and, thus, to assess the importance of different international transmission channels. Similar to Boivin and Giannoni [27], Mumtaz and Surico [29] and Charnavoki and Dolado [214], our model contains two country-specific blocks. The first block summarizes the U.S. economy while the second block corresponds to the Canadian economy. The U.S. block contains unobservable as well observable factors, F US t = (F US,F t, Yt US ), where F US,F t denotes a is a M 1 vector of observable factors. K US 1 vector of unobservable factors and Yt US summarizes information about the U.S. economy and is extracted from our panel F US,F t of U.S. macroeconomic and financial time series. Yt US contains series measuring U.S. monetary policy, i.e., the Federal funds rate and our QE balance sheet measure. The second block consists of K CA unobservable factors, F CA t, extracted from our broad panel of Canadian macroeconomic as well as financial series. The U.S. and Canadian datasets and the factors are related in the following way: [ ] X US t = X CA t [ Λ US US Λ CA US Λ CA CA ] [ ] F US t F CA t [ ] u US t + u CA t where X US t and X US t are data for the U.S. and Canadian economies; Λ j US = (Λj US,F Λj US,Y ) are the loading matrices corresponding to the unobservable and observable U.S. factors and Λ CA CA denotes the loading matrix for the Canadian factors; and uus t and u CA t are vectors of idiosyncratic disturbances. The structure on the factor loadings allows us to include the U.S. factors in the Canadian block. 3 Statistics Canada discontinued table in 212Q4, where real industry-level GDP is measured in 22 Canadian dollars, and replaced it with table , where data is measured in 27 dollars and starts in 1997Q1. We construct our GDP series by taking data from the later table, and growing them out backwards using the growth rate of corresponding variables from the former table. (1) 7

8 Moreover, we model the dynamics of the factors as a restricted VAR (see e.g. Charnavoki and Dolado [214]): [ ] F US t = F CA t [ b 11 (L) b 21 (L) b 22 (L) ] [ ] F US t 1 F CA t 1 + e t, (2) where b ij (L) are lag polynomials of finite order p and e t denotes the reduced form residuals. The zero restriction on b 12 (L) allows us to model the assumption that Canadian factors have no impact on U.S. factors. This is reasonable assumption given that the Canadian economy is a small open economy Estimation Following Bernanke et al. [25], we estimate the FAVAR in two-steps using principal component (PC) analysis. First, the unobserved factors F US,F t and F CA t are estimated by extracting the largest principal components from our datasets X US t and X CA t, respectively. Second, we estimate a restricted VAR using the estimated factors from step 1 and the observed factors (see Equation (2)). We chose only 1 lag for the VAR to take into account our relatively small sample size. In order to control for the effects of the observable factors on the U.S. block, we regress F US t on Yt US and slow-moving factors extracted from slow-moving U.S. series to US obtain ˆΛ US,Y. US US The estimated U.S. factors are then computed as ˆF t = ˆΛ US,Y YUS t. 5 Moreover, we impose that the U.S. factors are included in the PC for the Canadian block in Equation (1). Therefore, we estimate the Canadian factors following the iterative procedure employed by Boivin and Giannoni [27] and Charnavoki and Dolado [214] in order to control for the effect of the U.S. factors in the Canadian block. Given an initial PC estimate of the Canadian factors, ˆFCA() t, we iterate through the following steps until convergence is achieved: 1. Regress X CA t 2. Calculate 3. Obtain on X CA(j) t ˆF CA(j+1) t ˆF CA(j) t = X CA t and the estimates of the U.S. factors, CA(j) ˆΛ US ˆF US t. as the first K CA principal components of US ˆF t, to obtain X CA(j) t. ˆΛ CA(j) US. 4. Back to step 1. The algorithm stops when the difference between ˆF CA(j) t is sufficiently small. 4 However, our main results presented below are not sensitive to this assumption. 5 See Bernanke et al. [25] for details. ˆF CA(j+1) t and 8

9 In order to decide on the number of factors in both blocks, we employ the criteria proposed by Bai and Ng [22] and scree plots of the eigenvalues. The different versions of the Bai and Ng criteria suggest between 6 and 1 factors for the U.S. block and 4 to 8 factors for the Canadian block. Taking into account the scree plots (see Figure 1), we decide to include 6 unobservable factors into each block. While 4 factors are adequate in explaining aggregate-level variables, we find that the 5th and 6th factors are important in the U.S. block in explaining movements in the exchange rate and financial variables such as the stock market index. On the other hand, the 5th and 6th factors in the Canadian block are important in explaining regional trade patterns and trade by product-category. Nevertheless, our results and conclusions are robust to any combination of 4 to 1 factors. 3 Economic ties between U.S. and Canada before the crisis To put the results of our counter-factual exercise in perspective, this section describes the strong trade and financial linkages between the U.S. and Canadian economies before the crisis (up until 27Q4). Table 1 shows the breakdown of Canadian exports by industry, as well as the proportion of exports in each industry that goes to the U.S. We see that on average, 83.9% of all Canadian exported goods went to the U.S. between 2 and 27. Other than for the non-energy mining industry, the U.S. is the final destination for almost all of Canadian exports. In particular, for the automotive industry, which accounts for 2 % of Canadian goods exports, and the Oil and Gas sector, which accounts for 16.5% of Canadian goods exports, more than 97 % of exports are destined for the U.S. Table 2 shows the same breakdown for Canadian imports. On average, 58.9 % of all Canadian imported goods came from the U.S. between 2 and 27. Table 3 shows Canada s international investment position, measured in book values. From 2 through 27, 45.8 % of Canada s total direct investment and 54.1 % of portfolio investment were held in the U.S. At the same time, 63.4 % of foreign direct investment, and 6.8 % of portfolio investments held by Canada came from the U.S. In particular, the U.S. share of foreign portfolio investments in Canadian money market securities and bonds rose from 47 % in to 55.9% in 2-7. It stands to reason, then, that business-cycle movements in the two countries will be strongly correlated, and policy implementation in the U.S. will have strong repercussions in Canada. 9

10 Table?? shows the correlation of a number of Canadian real and financial variables with corresponding measures in the U.S. in business cycle frequencies. Cyclically detrended Canadian GDP is strongly, and persistently correlated with U.S. GDP, with a contemporaneous correlation coefficient of.737. The same is true for the Canadian stock market, measured by the TSX index, and the U.S. market, measured by the S&P 5 index. Importantly, this strong and persistent correlation also holds for short-term treasury yields and the 1-year treasury spreads between the two countries. This suggests that monetary policy action in the U.S. that affects either short term interest rates or long-term spreads may have a strong spillover onto the Canadian economy through co-movements in the corresponding Canadian rates and spreads. 3.1 Transmission of a Conventional U.S. Monetary Shock The unconditional correlations discussed above suggest strong ties between U.S. and Canada through both trade and financial channels. It may be instructive, however, to look at the spillover channels for a conventional U.S. monetary policy shock before we try to explain spillovers from QE. With this objective, we estimate impulse responses in our FAVAR framework to a Cholesky-ordered shock to the Federal Funds rate, as in Bernanke et al. [25], and Boivin and Giannoni [27] with data up to 28:Q2. The transmission of conventional U.S. monetary policy shocks to Canada has been explored by Kim [21] in a small-scale structural VAR framework, who finds that the financial channel is the main source of spillovers for U.S. monetary policy to Canada. We add to this literature by allowing the analysis of multiple transmission channels in a generalized setting. Moreover, our two-country restricted FAVAR is the appropriate framework to consider spillovers to small open economies like Canada. Figure 2 provides impulse responses for a 1 basis point decline in the Fed Funds rate for U.S. variables, and figure 3 provides results for corresponding Canadian variables. A reduction in the Fed Funds rate results in a gradual increase in U.S. GDP, employment, consumption, investment, as well as business confidence. Interestingly, we find that the 1-year term spread increases on impact, while asset prices represented by the S&P 5 index declines, albeit without statistical significance. From a U.S.-Canada bilateral trade perspective, we find some evidence of expenditureswitching. The U.S. nominal exchange rate vis-à-vis Canada depreciates, while net exports to Canada increases. At the same time, U.S. terms of trade against the world deteriorates. From a world perspective, however, we do not find any evidence of the 1

11 expenditure-switching effect. The U.S. trade-weighted nominal exchange rate appreciates, while U.S. net exports decline. Most importantly, the decline in Canadian net exports to the U.S. fails to bring down Canadian output. In fact, Canadian output rises roughly in the same magnitude as output in the U.S., due to the strong linkages between the financial markets of the two countries. Both short-run interest rates, as well as long-run yields move in the same direction in Canada as in the U.S. Moreover, we now see a clear and statistically significant rise in asset prices in Canada, as represented by the TSX stock index. As a result, both consumption and investment rise. The income-absorption effect can now be seen from a Canadian perspective, as the demand for imports rise, providing leakages through reduced net exports to the U.S. Canadian terms of trade increase, in lock-step with oil prices, represented here by Brent prices. The Canadian currency vis-à-vis the U.S. appreciates, consistent with the idea that the Canadian dollar is a commodity currency. World-wide net exports, however, increase on impact, reflecting, possibly, a rise in global demand following a U.S. expansion. Turning to the industry-space in figure 4, we see that the rise in output is reflected in all industries, and not just mining, suggesting that there is no Dutch-disease effect present in the transmission of U.S. monetary policy. Charnavoki and Dolado [214] find that the Dutch-disease effect can be seen conditional only on a commodity price shock, and not after a global demand or supply shock. We supplement their result by showing that the Dutch-disease phenomenon is also absent following a conventional U.S. monetary policy shock. In summary, we see that a conventional monetary policy expansion in the U.S. increases GDP in Canada almost to the same extent as it does in the US. Moreover, the positive effects of the shock are transmitted across the border mainly through financial channels. The co-movement between oil prices, Canadian terms of trade, and the appreciation of the dollar suggest evidence of a commodity currency effect. However, the positive effects of increased GDP is felt in all industries, and not just in commodity-producing industries. 4 International Transmission of QE Having explored the transmission channels for the international spillover of a conventional U.S. monetary policy to Canada, we can now look at the spillovers of QE. To this end, 11

12 this section first describes the counter-factual exercise, then analyzes the results of this exercise by looking at the difference between the policy and no-policy scenario for a number of different variables. Following the collapse of Lehman Brothers, the Federal Reserve announced large scale purchases of private assets, as the target for the federal funds rate reached its effective zero lower bound. At first, purchases were geared towards keeping the financial system from melting down. In March 29, the Federal Reserve announced purchases of $1.25 trillion of MBS, and up to $3 billion of long-term treasury securities. This time, the aim was to bring down long-term rates for both treasury and mortgage securities. Further purchases of long-term treasury securities worth $6 billion was announced in November 21, and $ 4 billion announced in 211. The literature on the financial market impact of QE has estimated that asset purchases have reduced long-term yields anywhere between 3 to 1 basis points (Gagnon et al. [211], Krishnamurthy and Vissing-Jorgensen [211]). Studies on the real effects of QE conducted towards the beginning of the implementation of this novel policy paradigm usually take the estimated effect of asset prices on spreads from the above studies for granted, and construct counter-factual scenarios where longterm spreads are increased by the same amount in the absence of QE (cf. Baumeister and Benati [213] and Kapetanios et al. [212]). Instead of taking a counter-factual view on long-term spreads, we take advantage of the 19 quarters of Fed balance sheet data now at our disposal, and consider a counterfactual case where the Fed did not expand its balance sheet. Note, however, that the Fed s total balance sheet includes emergency loans and agency securities purchased in order to preserve financial market stability, and not aimed at reducing long-term yields. 6 We are only interested in the long-term assets (treasury securities higher than 5 years of maturity, and long-term mortgage-backed securities) in the Fed s balance sheet. Figure 5 shows the increase long-term treasury holdings in the Fed s balance sheet, and the corresponding decrease in long-term spreads, while the effective funds rate remains at the zero lower bound. For our counter-factual, we construct a scenario where the longterm asset holdings of the Fed grows at a rate prevalent to the crisis. Specifically, we use data until 28Q3 in the same FAVAR setup to grow out the path of long-term assets in the Fed s balance sheet till the end of 213Q3, and use that path as our counter-factual. In contrast to the Fed, however, the Bank of Canada did not undertake QE. Through financial market integration, however, long-term spreads in Canada also came down at 6 See Lenza et al. [21], and Kozicki et al. [211] for a discussion of the difference between these two goals of the Fed s policy. 12

13 the same time as the U.S. spread. This can be seen in figure 6, which charts the asset side of Bank of Canada s balance sheet, along with the Canadian target overnight rate, and the 1 year spread. Although the Bank of Canada had also expanded its balance sheet at the height of the crisis by issuing emergency loans to the financial market, there is virtually no change in the amount of its long-term asset holdings Counter-factual setup We construct a counter-factual scenario with a world with no QE as the expected value of our variables of interest conditional on a counter-factual path of the Fed s asset holdings. This approach is similar to Lenza et al. [21]. In contrast, Kapetanios et al. [212] and Baumeister and Benati [213] consider a counter-factual set of coincident structural shocks, identified through sign restrictions, that move spreads a certain amount higher than observed, while keeping the Fed Funds Rate at the zero lower bound. By conditioning on a particular path of a policy variable, we are spared from imposing a structural interpretation on the set of shocks that represent the Fed s observed policy action. We generate conditional forecasts for the policy, and the no-policy scenarios in the following steps. 1. Take data till 28Q4. Estimate the restricted FAVAR model given in equations 1 and 2: X t = Λ 1 F t + u t F t = B 1 (L)F t 1 + e t t = 1982Q1...28Q4 2. Starting from 29Q1, produce an unconditional forecast until 213Q3, and save the path of the QE variable x QE T +h X T +h = E T +h ( XT +h Λ 1, B 1 (L), F...F T ) for our no-policy scenario. 3. Take the full sample of data (till 213Q3) and re-estimate the restricted FAVAR 7 The increase in the other-asset category seen from 211 onwards reflects the Canadian Federal Government s debt management strategy. Specifically, in the 211 Federal Budget, the government decided to hold prudential assets in the form of cash and short-term treasury that will allow it to operate for one month in case its regular access to financial markets is hindered. For details, see the government s Prudential Liquidity Management plan delineated in the budget document. http: // 13

14 model. X t = Λ 2 F t + u t F t = B 2 (L)F t 1 + e t t = 1982Q Q3 4. Policy scenario: Produce a forecast from 29Q1 until 213Q2, using the model B 2 (L), Λ 2, and conditional on the actual path of both the QE variable x,qe t and the effective federal funds rate, x,f F t. ) ˆX T +h = E T +h (X T +h Λ 2, B 2 (L), F...F T, x,qe T +h, x,f F T +h for T + h = 29Q Q3. This is the policy scenario. 5. No-policy scenario: Construct a counter-factual forecast from 29Q1 until 213Q2, using the model B 2 (L), Λ 2, conditional on the actual path of the effective federal funds rate, x,f F T +h, and the counter-factual path of the QE variable,, constructed in step two. x QE T +h X T +h = E T +h (X t Λ 2, B 2, F...F t 1, x QE t... x QE t+h, x,f F t )...x,f F t+h for t = 29Q Q3. This is the no-policy counter-factual scenario. Then the difference between ˆX t and X t is then the effect of QE. 4.2 Results Figure 7 and 8 shows the effect of QE for a key set of variables for the U.S., and Canada, respectively. The solid line represents the differences between the point forecasts, and the dashed lines the 68 percentile residual bootstrap bands of the counter-factual exercise described in the steps above. Note that although the appearance of these figures resemble traditional impulse responses, they are not so. Instead, they represent the different between the policy scenario and the no-policy scenario. Unless otherwise noted, results are expressed in percentage deviation from the no-policy scenario. Figure 7 shows that the increase in long-term asset holdings in the Fed balance sheet has lowered the 1-year treasury spread by almost 1 basis points and consequently raised U.S. GDP by almost 5 % compared to the no-policy scenario by 213Q2. In 14

15 contrast, Chung et al. [212] estimate that QE lowered term premiums by 5 basis points, and raised real GDP by almost 2 % by early 212. Although our results are higher than these estimates, the ratio of spread decline to GDP rise is comparable. Turning to the domestic transmission of QE, we see that asset prices, represented by the S& P 5 index, rises, which, through the wealth effect channel, increases consumption. Central bank action successfully raises the confidence index, reduces uncertainty, as reflected in the drop in the VIX, and increases the banking sectors willingness to lend. The latter is reflected by the diffusion index from the Federal Reserve s senior loan officer survey (SLOS) capturing banks ease of lending, presented in the first column of the last row in figure 7. Consequently, corporate spreads (Baa - Aaa) decline by 2 basis points, and commercial and industrial loans increase. All of this results in a significant rise in U.S. domestic investment. Consistent with the expenditure-switching effect, unconventional monetary stimulus deteriorates U.S. terms of trade, and the trade-weighted nominal U.S. exchange rate depreciates in the beginning, and appreciates in the long run. Overall, however, U.S. net exports do not increase, suggesting, perhaps, that there is enough of an incomeabsorption effect that raises import demands to counteract, if not overturn, the rise in net exports suggested by the expenditure-switching theory. Figure 8 shows that the Fed s QE program resulted in Canadian GDP being higher than the counter-factual no-policy scenario by little under 5%, while Canadian 1 year treasury spreads declined by a similar amount as in the U.S. In short, the effect of U.S. monetary policy in Canada has almost the same magnitude of effect as it does in the U.S. This result is consistent with the results described in section 3.1, where Canadian variables respond roughly by the same magnitude as U.S. variables for a conventional U.S. monetary policy shock. That Canadian long-term rates declined following the Fed announcements of Q.E. purchases, has been documented by Bauer and Neely [212]. Our study shows that the 3-month treasury yields in Canada also drops by 5 basis points coincident with the implementation of QE. Because of financial market integration, Canadian asset prices, represented by the TSX index, also rise, in turn increasing Canadian consumption through wealth effects. Business confidence in Canada goes up, as does total bank loans. Consequently, investment rises significantly. Improved U.S. demand raises the price of oil, and the Canadian terms of trade improve, consistent with the commodity currency effects discovered by Charnavoki and Dolado [214]. The Canadian nominal exchange rate vis-à-vis the U.S. appreciates, and 15

16 total Canadian net exports decline by almost 3% of GDP. Figures 1 and 11 show the impact of QE on Canadian trade by product category. We see that exports of all products except industrial machinery, consumer goods and services increase. This suggests that a stronger U.S. economy resulting from the implementation of Q.E. benefited Canadian exports. The decline in net exports, then, can be explained by the fact that imports in many product categories increased by more than exports, resulting in no change, or a drop in net exports. In particular, industrial machineries, electrical equipment, consumer goods and services show the strongest decline in net exports all due to a strong rise in imports. The leakage from the export sector in Canada, then, follows the pattern of a strong income-absorption effect under Canadian monetary expansion. In other words, financial market integration resulted in Canadian interest rates and asset prices moving in the same direction as those in the U.S. The resulting rise in consumption and investment increased demands for imports, and the appreciation of the Canadian dollar made imports cheaper. As a consequence, net exports fell, despite the fact that our exports increased in most product categories, due to a stronger U.S. economy post-qe. Finally, figure 9 shows the impact of QE on Canadian GDP by industry. As a commodity exporting country, Canada generally benefits from a rise in global oil prices. At the same time, the commodity currency effect implies that increased demand for oil also appreciates the Canadian exchange rate, which may make other industries noncompetitive. Thus, the gains in the oil-producing industry might come at the expense of other industries. This trade-off is often referred to as the Dutch-disease phenomenon. As figure 9 shows, however, GDP in all industries, except arts and recreation, increased due to the implementation of QE. As such, we do not find any evidence of a Dutch-disease effect due to U.S. unconventional monetary policy. To summarize, we find that the implementation of QE in the U.S. benefited Canada almost to the same extent as it did the U.S. domestic economy. This spillover into Canada occurred mainly through the financial channel, as integrated financial markets made Canadian asset prices and interest rates move in the same direction as their U.S. counterparts. The Canadian exchange rate did appreciate, generating leakages through reduced net exports. However, these leakages seem to be a consequence of stronger Canadian demand for imports. Despite the coincident rise in oil prices and the Canadian exchange rate, we find no evidence of Dutch-disease-type phenomenon as a consequence of QE, as GDP in almost all industries shows gains. 16

17 5 Conclusion The U.S. Federal Reserve responded to the financial crisis and the ensuing great recession through balance sheet expansion, or quantitative easing. Aimed at lowering long-term rates when its conventional short-term policy instrument became stuck at its lower bound, the Fed announced and conducted purchases of long-term assets at an unprecedented scale, effectively quadrupling the size of its balance sheet by the end of 213. This paper assesses the impact of this program on Canada, and identifies the transmission channels through which this unconventional policy implemented by the U.S. Federal Reserve spilled over internationally to the Canadian economy. We consider a counter-factual scenario where the long-term treasury and mortgagebacked security holdings in the Federal Reserve s balance sheet expands at a pace prevalent before the crisis, instead of expanding eleven-folds, as it did between 28Q3 and 213Q3. We assess the likely outcome of this scenario in the presence of the realized path of the Federal Funds Rate in a two-country FAVAR framework, where parameters are restricted so that movements in Canadian factors do not influence movements in U.S. specific factors. We compare the model forecast conditional on this counter-factual with a forecast conditional on the realized values of both policy instruments of the Fed the Fed Funds Rate, which remained stuck at its lower bound, and the long-term asset holding in its balance sheet. We find that the Fed s quantitative easing program boosted Canadian GDP almost as much as it did the U.S. GDP. The transmission occurs primarily through the financial channel, as both short-term bond yields and long term spreads in Canada decline following the implementation of QE. These two features are common for both conventional monetary policy shocks, as well as spillovers of QE. Theoretically, there is a level of ambiguity regarding the effects of a monetary expansion in the U.S. on Canada. Under the expenditure switching effect, we expect to see the Canadian nominal exchange appreciate, and Canadian net exports decline. If the income-absorption effect dominates, however, we expect to see a higher overall U.S. final demand generate an increase in Canadian net exports to the U.S. In reality, the trade channel proved to be a source of leakage from the Canadian economy and the Canadian nominal exchange rate appreciated vis-à-vis the U.S., while total net exports declined. Finally, we note that the co-incident rise in oil prices, Canadian terms of trade, and the Canadian nominal exchange rate due to QE supports the commodity-currency effect faced by Canada. However, output in all sectors in Canada increases during QE, and we 17

18 find no Dutch-disease-type effect where the oil-producing sector benefits to the detriment of the remaining industries through a higher nominal exchange rate. References Jushan Bai and Serena Ng. Determining the number of factors in approximate factor models. Econometrica, 7(1): , 22. Michael D. Bauer and Christopher J. Neely. International channels of the Feds unconventional monetary policy. Technical report, 212. Christiane Baumeister and Luca Benati. Unconventional monetary policy and the great recession: Estimating the macroeconomic effects of a spread compression at the zero lower bound. International Journal of Central Banking, 9(2): , June 213. Ben Bernanke, Jean Boivin, and Piotr S. Eliasz. Measuring the effects of monetary policy: A factor-augmented vector autoregressive (favar) approach. The Quarterly Journal of Economics, 12(1): , January 25. Ben S. Bernanke. Monetary Policy since the Onset of the Crisis : a speech at the Federal Reserve Bank of Kansas City Economic Symposium, Jackson Hole, Wyoming, August 31, 212. Speech 645, Board of Governors of the Federal Reserve System (U.S.), August 212. URL Jean Boivin and Marc P. Giannoni. Global forces and monetary policy effectiveness. In International Dimensions of Monetary Policy, NBER Chapters, pages National Bureau of Economic Research, Inc, July 27. URL org/h/nbr/nberch/515.html. Valery Charnavoki and Juan J. Dolado. The Effects of Global Shocks on Small Commodity-Exporting Economies: Lessons from Canada. American Economic Journal: Macroeconomics, 6(2):27 37, April 214. URL aea/aejmac/v6y214i2p27-37.html. Hess Chung, JeanPhilippe Laforte, David Reifschneider, and John C. Williams. Have We Underestimated the Likelihood and Severity of Zero Lower Bound Events? Journal of Money, Credit and Banking, 44:47 82, URL mcb/jmoncb/v44y212ip47-82.html. 18

19 Luca Dedola, Peter Karadi, and Giovanni Lombardo. Global implications of national unconventional policies. Journal of Monetary Economics, 6(1):66 85, 213. URL Marcel Fratzscher, Marco Lo Duca, and Roland Straub. On the international spillovers of US quantitative easing. Working Paper Series 1557, European Central Bank, June 213. URL Joseph Gagnon, Matthew Raskin, Julie Remache, and Brian Sack. Large-scale asset purchases by the Federal Reserve: did they work? Economic Policy Review, (May): 41 59, 211. URL 17no.1.html. Leonardo Gambacorta, Boris Hofmann, and Gert Peersman. The Effectiveness of Unconventional Monetary Policy at the Zero Lower Bound: A Cross-Country Analysis. BIS Working Papers 384, Bank for International Settlements, August 212. URL George Kapetanios, Haroon Mumtaz, Ibrahim Stevens, and Konstantinos Theodoridis. Assessing the economywide effects of quantitative easing. Economic Journal, 122(564): F316 F347, November 212. Irfan Akbar Kazi, Hakimzadi Wagan, and Farhan Akbar. The changing international transmission of U.S. monetary policy shocks: Is there evidence of contagion effect on OECD countries. Economic Modelling, 3(C):9 116, 213. URL repec.org/a/eee/ecmode/v3y213icp9-116.html. Soyoung Kim. International transmission of U.S. monetary policy shocks: Evidence from VAR s. Journal of Monetary Economics, 48(2): , October 21. URL Sharon Kozicki, Eric Santor, and Lena Suchanek. Unconventional Monetary Policy: The International Experience with Central Bank Asset Purchases. Bank of Canada Review, 211(Spring):13 25, 211. URL v211y211ispring11p13-25.html. Arvind Krishnamurthy and Annette Vissing-Jorgensen. The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy. NBER Working 19

20 Papers 17555, National Bureau of Economic Research, Inc, October URL Michele Lenza, Huw Pill, and Lucrezia Reichlin. Monetary policy in exceptional times. Economic Policy, 25: , URL bla/ecpoli/v25y21ip html. Haroon Mumtaz and Paolo Surico. The Transmission of International Shocks: A Factor- Augmented VAR Approach. Journal of Money, Credit and Banking, 41(s1):71 1, URL Christopher J. Neely. The large scale asset purchases had large international effects. Technical report, 21. Helene Rey. Dilemma not trilemma: the global cycle and monetary policy independence. Proceedings - Economic Policy Symposium - Jackson Hole, pages 1 2, 213. URL Kenneth Rogoff. Dornbusch s Overshooting Model After Twenty-Five Years. IMF Working Papers 2/39, International Monetary Fund, February 22. URL http: //ideas.repec.org/p/imf/imfwpa/2-39.html. Heike Schenkelberg and Sebastian Watzka. Real effects of quantitative easing at the zero lower bound: Structural VAR-based evidence from Japan. Journal of International Money and Finance, 33(C): , 213. URL jimfin/v33y213icp html. Michael Woodford. Methods of policy accommodation at the interest-rate lower bound. Proceedings - Economic Policy Symposium - Jackson Hole, pages , 212. URL Lorie Zorn, Carolyn Wilkins, and Walter Engert. Bank of Canada Liquidity Actions in Response to the Financial Market Turmoil. Bank of Canada Review, 29(Autumn):7 26, 29. URL v29y29iautumn9p7-26.html. 2

21 Table 1: Goods trade from Canada (2-27 averages) 21 Canadian goods exports Canadian exports to US US imports from Canada (% GDP) (% goods exports) (% total Canadian exports) (% total US imports) All goods Agriculture Forestry Oil and gas Mining (ex. oil and gas) Consumer goods Chemicals Metals and minerals Machinery Electronics Autos Other transportation Other goods Note: Calculated based on nominal annual series in CAD (industry breakdown of Canadian exports) and USD (all other values). Percentages of total goods exports may not sum to 1 due to some omitted categories. Source: Calculations based on data from IMF Direction of Trade Statistics, WDI, US Census Bureau, Industry Canada