GOVERNMENT BOND YIELDS AT THE EFFECTIVE LOWER BOUND: INTERNATIONAL EVIDENCE.
Following the international financial crisis of 2007-2009, a small, but systemically important, group of central banks in advanced economies reduced their policy rates to the zero lower bound (ZLB). As bank lending and economic activity remained weak in many countries, some central banks even introduced negative policy interest rates. As a result, the literature has begun to refer to the effective lower bound (ELB) for policy rates. While prior to the crisis, only the Bank of Japan kept the interest rate indicative of its policy stance at the ZLB, since 2008, several other central banks have followed suit, including the Bank of Canada (BoC), the Bank of England (BoE), the European Central Bank (ECB), the Swedish Riksbank, the Swiss National Bank (SNB), and the U.S. Federal Reserve. Central bankers in these economies have at the same time become increasingly vocal about the ability of monetary policy to remain effective after the ELB is reached (e.g., Bernanke 2012; Carney 2013; Svensson 2010; Yellen 2014). Of course, many of these same central banks have also turned to unconventional monetary policies (UMPs), partly to counteract this impression that monetary policy is ineffective when rates have reached the ELB.
Indeed, UMPs have helped dispel the impression that monetary policy could not be effective after reaching the ELB. The works of Christensen and Rudebusch (2012), Gagnon et al. (2011), Krishnamurthy and Vissing-Jorgensen (2011), Joyce et al. (2011), and De Rezende, Kjellberg, and Tysklind (2015) represent only a small sample of the significant literature that investigates the impact of UMPs on sovereign bond yields and other assets involved in various asset purchase programs.
Alongside the implementation of UMPs, central banks have increasingly utilized the timing and content of announcements as an additional element of their monetary policies. The increased reliance on such announcements has also contributed to changing attitudes about the ability to influence the stance of monetary policy when the ELB on nominal policy rates is reached. Indeed, a small number of central banks began to resort to forward guidance--communicating conditional commitments to future policy paths--to further influence expectations and the effectiveness of monetary policy. Nevertheless, due to continuous doubts regarding their effectiveness, these strategies remain controversial (e.g., Ball et al. 2016; Filardo and Hofmann 2014). However, while many recognize the importance of verbal interventions of this nature, few have actually conditioned their findings on the impact of central bank communication on sovereign bond yields, as is done in the present study.
UMPs are aimed at influencing yields across the term structure through direct and indirect channels. Of course, it has long been known that news, in the form of macroeconomic announcements reflecting underlying economic conditions, influence yields on financial instruments with different maturity structures. This was true even before UMPs were implemented. However, if we are to evaluate the effectiveness of monetary policy at the ELB, the challenge is to properly identify whether market participants are less responsive to news because central banks are at, or near, the ELB. In this connection, it then becomes policy relevant to estimate whether and how responsive yields of varying term structures are to such news and announcements. This paper provides international evidence of the data dependence of yields for select short- and long-term government bonds. We address this issue by looking at the reactions of sovereign bond yields to news events and central bank communications for samples that precede and follow the twin financial crises of 2007-2009 and the 2010-2012 euro area sovereign debt crisis.
In our methodology, the responsiveness of yields to central bank announcements at the ELB depends on the selection of a benchmark period when market participants' responses to such news was considered "normal." In other words, the extent to which monetary policy is constrained is quantified in relation to a period when policy rates were sufficiently far away from the ELB that it would not affect markets' reactions to news. Swanson and Williams (SW 2014a, 2014b) propose such a methodology to empirically assess the constraint posed by the ZLB for the United States, the United Kingdom, and Germany. This study relies on a modified version of their approach.
Our study departs from SW (2014a, 2014b) in three important respects. First, our metric for news effects is significantly broadened to include variables that quantify the content of press releases and meeting minutes published by central banks. In so doing, we rely on an algorithm that converts vectors of words into a numerical indicator that can be used in econometric estimation. Second, we extend their earlier analyses by adding data for Canada, Sweden, Switzerland, and the euro area. This allows for international evidence about the impact of the ELB across a wider spectrum of monetary policy strategies. Third, unlike SW (2014a, 2014b), our estimates are also conditioned on international spillover effects. Although we focus on the content of central bank communication on domestic yields, spillover effects should not be ignored (also see Appendix S1, Supporting Information).
To preview our results, we find that the sensitivity of yields to news falls at the ELB. Most medium- to long-term yields become constrained around 2012, that is, 3 years after most central banks in our study reached the ELB, but there are some exceptions. For example, 10-year government bond yields in Sweden and the United Kingdom retain "normal" responsiveness to news--relative to a benchmark period where "news response" is deemed normal (i.e., consistent with precrisis types of responses)--while yields in the United States and Canada retain some responsiveness to news relative to their benchmark periods. There is also evidence that Swedish and Swiss yields became more responsive to news after their respective central banks introduced negative policy rates.
While sensitivity of yields to news generally decreased at the ELB, yields need not be less responsive to all forms of news. Indeed, we find that government bond yields became more responsive to shocks in the content of central bank communication at the ELB, especially at the longer end of the yield curve. Therefore, our results provide further evidence that central bank communication is an important element of monetary policy making when the interest rate tool loses efficacy.
The rest of the article is organized as follows. The next section provides an overview of the relevant literature on the ELB and the associated monetary policy challenges. Section III is devoted to the methodology and data. Section IV covers a few stylized facts and describes our main findings. The article concludes with a brief summary of the results and the implications for the effectiveness of monetary policy at the ELB.
II. LITERATURE REVIEW
Under "normal" circumstances, news moves markets. (1) Furthermore, while news does arrive in several forms, the extant literature relies exclusively on macroeconomic and financial announcements (e.g., employment reports, CPI inflation releases, etc.). Gurkaynak, Sack, and Swanson (2005a, 2005b), for instance, have made the case that such announcements represent an incomplete representation of the sources of surprises in monetary policy. This idea is especially evident as central banks increasingly rely on verbal announcements that accompany policy rate decisions, or the eventually publicized minutes of the policy committee meetings, to steer monetary policy. These channels provide additional, potentially relevant sources of news.
Moreover, there is a large literature that explores the impact of monetary policy decisions and news events on yields along the term structure. But other than for Japan, where the Bank of Japan has kept its policy rate near zero for well over a decade, or Switzerland, which has also experienced bouts of low inflation and interest rates over the past several years, there are relatively few studies that ask whether monetary policy can retain its influence when policy rates effectively fall to zero or below. (2) As recent as a decade ago, only small samples were available to assess financial market reactions at the ELB, which brought Bernanke, Reinhart, and Sack (2004, 2) to comment that "systematic empirical evidence on the potential efficacy of alternative policies is scant." (3)
Surveying the empirical research on the financial market effects of quantitative easing (QE), Gagnon (2016, Table 1) identifies that 18 of 24 are event studies, while only a handful (3-6) perform time series analysis. But, now that several more central banks have driven policy rates near the ELB and over a longer period, there is much greater scope for a time series exploration of the impact of news at very low interest rates.
Event studies have their challenges (e.g., MacKinlay 1997), but time series analyses are not without problems of their own. Perhaps the greatest advantage of the time series approach is that it does not require the investigator to take a stand on the window of time over which a news event or policy announcement is thought to influence yields. Indeed, the selection of the window limits the possibility of a delayed market reaction either because investors have yet to digest existing information or because they revisit and revise previously held beliefs. (4) However, the time series approach is complicated by the possibility that estimates might be contaminated by the omission of certain factors or events that influence yields. The methodology adopted in this study is suitable under the circumstances as it allows for an examination of a wider array of determinants of policy decisions over time as opposed to judiciously chosen events. It also allows for a quantitative measurement of the time-varying sensitivity of yields at the ELB. In any event, we do take steps to limit the risks that arise from a time series approach as they relate to the issues investigated. Moreover, our methodology permits a more direct comparison with the recent evidence (e.g., SW 2014a, 2014b).
Also related to these issues is the choice of the sampling frequency. Some research considers the financial market impact of UMPs at the intradaily sampling frequency (e.g., Gilchrist,
Lopez-Salido, and Zkrajsek 2014; Rogers, Scotti, and Wright 2014). (5) One important reason for observing daily effects is that monetary policy announcements and reactions potentially cover several time zones. Indeed, there are reasons to expect these effects to linger for longer because of the phenomenon of rational inattention (e.g., Sims 2003; Wiederholt 2012). Accordingly, daily data are capable of adequately capturing the impact of new information on financial markets, including cross-border effects of news announcements that originate in a different part of the world.
If the ELB is problematic for the effectiveness of monetary policy, then how can the constraint be overcome? Suggestions were already being made several years ago (e.g., Blinder 2000; Buiter 2009; Goodfriend 2000; McCallum 2000). Some central banks attempted to influence medium- to long-term yields; however, several observers were skeptical that this would be an effective strategy (e.g., Blinder 2000; Ueda 2005). The experience at the Bank of Japan also casts a shadow on the power of these kinds of policies (e.g., Bernanke, Reinhart, and Sack 2004; Kuttner and Posen 2004).
Forward guidance is another policy tool used to prevent monetary policy from being constrained at the ELB. These verbal tools work by communicating when the policy stance could be changed or reversed. But, as there is a critical element of conditionality about these promises, the effectiveness of such policies may diminish, especially if the commitment is not viewed as being credible (Woodford 2012). (6)
Despite extensive research on UMPs, few studies have specifically examined the extent to which yields were constrained during the ELB period. SW (2014a, 2014b) offer an empirical methodology that identifies whether or not yields are constrained. On the experience in the United States, they report that yields on Treasuries were "surprisingly" responsive to news until 2011 when the Fed began issuing various forms of forward guidance. Therefore, while the federal funds rate reached the ELB in late 2008, monetary policy continued to be effective. They also reach a similar conclusion using data on the United Kingdom and Germany; just as Winkelmann, Bibinger, and Linzert (2016) do in their study of the euro area.
But this literature omits an important source of news: new information in central bank communications. (7) This type of news is important because, although monetary policy interest rates may be constrained, monetary conditions continue to change over time. Several approaches have been applied to measure the content of central bank announcements. One frequently used method is to manually construct variables aimed at capturing the intentions of policy communications. An early application of this approach was performed by Romer and Romer (1989, 2004) who used a "narrative approach" to complement quantitative data to extract U.S. Federal Open Market Committee (FOMC) intentions for the federal funds rate. Typically, these studies assign a negative (positive) value to statements that are perceived to be dovish (hawkish) based on their own interpretation of the central bank's communication (e.g., Gerlach 2007) or the interpretation of the communication by news reports (e.g., Ehrmann and Fratzscher 2007; Jansen and De Haan 2010). This approach to coding policy intentions captures asymmetry in monetary policy communications; however, it is a subjective measurement that is bounded by a predetermined, ordinal scale. Another approach frequently used in the literature is to derive the interpretation of monetary policymakers' intentions from financial market reactions using principal component analysis (e.g., Gurkaynak, Sack, and Swanson 2005a, 2005b). While this approach generates a continuous variable that is not clouded by hindsight, similar to event study methodologies, it requires defining a time period within which financial market volatility is considered relevant to the announcement.
More recent research applies text analysis methods to quantify the content of central bank communications. De Carvalho, Hsu, and Nechio (2016), for example, quantify the content of U.S. FOMC statements by calculating a ratio of the number of times "hawkish" and "dovish" appear in news articles surrounding an announcement. We also rely on semantic-based measurements, and use a "dictionary" technique to quantify the content of central bank communications (see Hart, Childers, and Lind 2013; Loughran and McDonald 2011). Instead of coding central bank statements based on a narrow list of words (i.e., hawkish/dovish), the dictionaries define a collection of words that aim to convey meaning along specific dimensions of central bank statements. As Meade, Burk, and Josselyn (2015) argue, U.S. policymakers view the minutes "... as providing insight about the breadth of views." Furthermore, since meeting participants provide input before the final draft is published (Danker and Luecke 2005), the intention and, therefore, the content of the document ought to reflect the views of FOMC members.
III. DATA AND ECONOMETRIC SPECIFICATIONS A. News
Our sample includes countries that have reached the ELB in policy rates. The economies that have, in recent years, come closest to the ELB at one time or another include Canada, the euro area (our sample includes euro area yields, (8 )Germany and France), Sweden, Switzerland, the United States, and the United Kingdom. All the economies in the sample have, at least in de jure terms, flexible exchange rates and open capital markets. (9) Our sample period begins in January 2001, ends in March 2015, and consists of daily data. Since the financial crisis in the United States began some time in 2007, while the sovereign debt crisis in Europe begins in 2010, our sample also covers a significant span of time when financial markets operated more or less normally. (10) Table 1 provides details about the time span where the ELB was in operation in the economies considered in this study."
Financial markets are forward looking; therefore, anticipated economic activity should be priced into assets (Kuttner 2001). Our methodology hypothesizes that markets should only react to the unexpected or "surprise" components of each data release. As such, our study is concerned with the sensitivity of yields to two sets of surprise news variables, conditional on other control variables (see below). The first set is a vector of the surprise component of macroeconomic news announcements. Using a methodology similar to that adopted by SW (2014a, 2014b), and several others, surprise macroeconomic announcements are constructed by subtracting a consensus forecast from the actual data release and normalizing the data by the full sample standard deviation. (12) Estimation of news effects relies on taking a stand on expectations. Of course, expectations are not unique. Forecasters can disagree (e.g., Siklos 2013, and references therein) and this can have implications for studies such as ours. Therefore, there is the potential for mis-measurement. There is a large literature supporting the notion that combining forecasts yields superior forecasts (e.g., Armstrong 2001), though the reason remains unclear. Nevertheless, as long as the forecasts used are generally unbiased, any mis-measurement is likely to be noise and would not materially impact the conclusions. We believe this assumption is valid for the data set employed in the present study. (13) The indicators selected for the analysis are typically considered to be "market-moving," that is, they are believed to have a same day impact on domestic and global yields; they are also similar to those employed by SW (2014a, 2014b).
The second set of variables is more novel under the circumstances. We evaluate the changing content of monetary policy communications as an additional source of news. To quantify central bank communication, we rely on an algorithm that measures the content of central bank press releases and meeting minutes based on a large dictionary of terms. The dictionaries we use include a combination of those constructed by Loughran and McDonald (2011) based on financial texts, and DICTION (see Hart, Childers, and Lind 2013), originally created using political texts. Specifically, three dictionaries are defined, grouping words in categories that aim to capture meaningful aspects of central bank communication: optimism, pessimism, and certainty. These dictionaries were augmented with language commonly used in central bank communication. In addition, words that are more ambiguous in the context of central banking, such as "risk," "unemployment," "surprise," and so on, were removed from the dictionaries. In general, the optimism dictionary aims to capture language that signals improved economic conditions or outlook, while the pessimism dictionary includes language that may indicate that economic conditions have deteriorated or that the outlook has worsened or even turned negative. Importantly, we argue that the content variables are crucial for capturing the monetary policy committee's interpretation of economic conditions and their potential impact on future monetary policy decisions. For this reason, the content variables offer additional information beyond numerical economic forecasts or macroeconomic data releases.
The third content variable is labeled "certainty." This variable brings together two meaningful dimensions of central bank communication: it subtracts language that conveys uncertainty and/or disagreement from language that conveys agreement and/or certainty. Central bank communication primarily aims to achieve two goals: to improve the effectiveness of monetary policy by delivering clear information about current policy decisions and the potential future policy path, and to reduce uncertainty in financial markets (e.g., Blinder et al. 2008). The uncertainty variable aims to capture the latter; this dictionary includes words that are used to express doubt or a lack of confidence in statements. Dissent (or agreement) improves information on the former, and has proved to be useful for forecasting future monetary policy decisions (Riboni and Ruge-Marcia 2014). Analyzing the content of the texts may indeed be preferable to simply tallying the vote count for monetary policy decision, as voting patterns have been found to be a poor measure of actual level of disagreement for the U.S. FOMC (Meade 2005); in addition, several central banks do not reveal voting information (e.g., ECB) or do not take a formal vote (e.g., BoC). We use DICTION 6.0 software to calculate the percentage of words that convey optimism, pessimism, and certainty as a percentage of the total words of the text; the "content" variables are therefore continuous, and are equal to zero on days where the central bank does not release a communication. Appendix S1 elaborates on the content and construction of the dictionaries.
B. Econometric Estimates of News Effects
The standard model to evaluate the sensitivity of yields to surprise news events (e.g., Giirkaynak, Sack, and Swanson 2005a, 2005b) posits that the unanticipated component of macroeconomic data releases is the principal determinant of changes in yields. An extended version of their model can be written as follows:
(1) [DELTA][y.sub.t] = [alpha] + [gamma][Z.sub.t] + [beta][X.sub.t] + [delta][C.sub.t] + [[epsilon].sub.t]
where [DELTA][y.sub.t] is the daily change in a government bond yield, Z, is a vector of dummy variables capturing monetary policy news events, and X, is a vector of surprise components of macroeconomic data releases. Some news events, such as days when a monetary policy committee meets, are not surprises. However, even if the timing of these events is known in advance there can be a surprising outcome to the meeting, either because of changes in the verbal content of the announcements or because a policy change is decided. This is why it is important to extend the standard specification by adding a vector capturing the verbal content of central bank announcements via a proxy for the content of policy committee press releases and meeting minutes (C,). (14)
The sample period starts in January 2001 for most events in the United States and March 2003 for the United Kingdom and the euro area economies. For the United States, X, includes only domestic surprises in economic data releases. For the United Kingdom and the euro area, both domestic and U.S. surprises are used; and for France and Germany, domestic, euro area, and U.S. data release surprises are included. For Canada, only U.S. surprises are used; and for Switzerland and Sweden, both U.S. and euro area surprise data are used. (15) The fact that domestic data for macroeconomic releases are unavailable for the smaller economies in our sample is not as great a disadvantage as one might think, since news releases often overlap with releases for major economies. Moreover, announcements from the major economies potentially have an outsized influence on the sovereign yields of small-open economies. Indeed, global capital markets are responsive to U.S. economic conditions and yields respond to U.S. news events (e.g., Lombardi, Siklos, and St. Amand 2016). Lack of sensitivity to U.S. news events would therefore suggest domestic yields are constrained when policy rates are at the ELB.
The content variables are also standardized to capture the impact of a "surprise" in the language of a release. It is for this reason that press releases and meeting minutes are included in the analysis, but not central bankers' speeches. Press releases and meeting minutes offer a consistent means of communication, that varies due to deliberate changes in communication (i.e., press releases) or changes in the conversation (i.e., minutes); a "surprise" element can therefore be extracted based on changes in the language used in the communication. The surprise element is measured by the standardized variable--mean of zero and standard deviation equal to one. (16)
To isolate the impact of the content of central bank press releases, Z, is a set of dummy variables for the dates of key monetary policy communications and important UMP announcements. One variable captures the dating of monetary policy committee meeting press releases, another captures the release date of monetary policy committee meeting minutes (where available), and a final dummy is used to account for announcements of UMPs for the United States, the United Kingdom, and the euro area. For the United States, there is an additional dummy variable to capture the winding down of the U.S. Fed's QE policy, or "tapering" announcements, since it is plausible that the QE and tapering announcements may have asymmetric effects on domestic as well as on foreign yields. These dummy variables are equal to one the day of and day after an event. (17) Further information on specific variables, sources of data, and summary statistics is relegated to Appendix S1.
SW (2014a, 2014b) generalize Equation (1), which serves as a benchmark of sorts, and consider nonlinear extensions. In most cases, however, they conclude that |3 is constant. Yet, the prior is that the impact of news (and additionally here, the content of central bank announcements) on asset prices can be time-varying. Moreover, at the ELB, there is a strong possibility that the response is nonlinear. (18)
We chose a specification that permits nonlin-earity to enter via a behavioral expression that captures a common determinant of the sensitivity of yields to news. Accordingly, we report results based on regressions of the form:
(2) [mathematical expression not reproducible]
where f([[THETA].sub.t]) denotes a function that relates variables believed to capture yield sensitivity. More precisely:
(3) [mathematical expression not reproducible]
Equations (2) and (3) are jointly estimated. We assume that [[THETA].sub.t] is a vector consisting of the first principal component of the available domestic yields at various terms to maturity, the (10-day) rolling standard deviation of the first principal component, and a measure of global uncertainty, captured by the (10-day) rolling standard deviation of the first principal component of U.S. treasury yields at various terms to maturity. As identified by SW (2014b), the sensitivity of yields may depend on the actual level of yields. News can impact yields across the term structure; it is therefore appropriate to use the first principal component, which extracts common movements in yields, as a representative of the level of yield. The volatility of the first principal component is meant to capture the effects of uncertainty on yield movements. (19) We do not impose sign restrictions on cp in Equation (3) because it is unclear, a priori, whether uncertainty increases or decreases the sensitivity of yields.
Like SW (2014b), we normalize f([[THETA].sub.t]) to unity during a chosen benchmark period so that the sensitivity of yields at the ELB period can be more easily identified. In selecting the benchmark period, our first and most important criterion was that policy rates not be constrained by the ELB. SW (2014a, 2014b) chose a benchmark period in which they assumed yields responded normally to news events (1990 to 2001 for the United States; and 1993 to 2006 for the United King-don and Germany). We maintain that in periods of uncertainty yields may still respond normally to surprise data, as long as they are not constrained. For this reason, we set f([[THETA].sub.t]) to an average of unity when central bank policy rates are 2% or higher. The normalization to unity ensures ease of interpretation such that any value above one suggests yields are more sensitive to news than normal, and any value below one suggests yields are more constrained relative to when the ELB constraint is not in play. The 2% threshold is also a quantitative rule that prevents some arbitrariness about when the ELB is reached. (20)
IV. EMPIRICAL RESULTS A. A Few Stylized Facts
Table 2 provides summary statistics for the pessimism, optimism, and certainty variables for monetary policy committee press releases and meeting minutes (where available). The table shows how the content of central bank communications changed in the aftermath of the crisis. In general, the use of language conveying certainty increased during the postcrisis period. Interestingly, variation in content from meeting to meeting also decreased after the crisis, particularly when conveying certainty and optimism. This may reflect the fact that central bankers became more deliberate in their choice of words during crises to avoid triggering financial volatility, as well as near the ELB, in order to use changes in communication as an effective policy tool. Similarly, the use of pessimistic language did not increase in most central bank communications, as one might reasonably expect during a period of economic turmoil; but rather only did so in less than half of the communications in our sample. These stylized facts suggest that there is scope for asymmetric effects in quantifying the content of central bank statements. This possibility is especially important at the ELB, where yields may be less responsive to negative surprises than to positive surprises. (21)' (22)
Optimism and pessimism were also analyzed against economic, financial and policy indicators to get a better understanding of what the variables are capturing. In general, pessimism is negatively correlated with the gross domestic product growth outlook, and positively correlated with the Chicago Board Options Exchange Volatility Index (VIX) and indicators of domestic and global economic policy uncertainty (Baker, Bloom, and Davis 2016). These relationships align with our expectations: negative tone increases with uncertainty, financial volatility and poor economic prospects. Similarly, optimism is negatively correlated with the inflation outlook and measures of financial volatility and policy uncertainty. These relationships are generally weaker for optimism than for pessimism (correlation matrix is available in Appendix S1). The strength of these correlations suggests that the language used in central bank communications conveys data on the real economy, uncertainty, and financial volatility, among other relevant information. We argue that the additional information transmitted by central bank communication includes the sentiment of the monetary policy committee about how current and future projected economic circumstances relate to the expected path of the monetary policy setting. That is, variables that evaluate the content of central bank communication are interpreted as a signal about the economy and its interaction with future policy decisions.
It is more difficult to link the "certainty" content variable with trends in economic and financial conditions. It is possible that this sentiment captures different phenomena for each central bank. We suggest that the certainty indicator could be capturing the level of agreement on current decisions and/or the outlook for the future policy path, as well as the degree of confidence in economic projections. In addition, the stylized facts indicate that the communication of certainty has changed in form in the postcrisis periods: most central banks increased the use of language expressing certainty and decreased the variability of this language. It is plausible that the impact of this variable on financial market behavior may also have changed in the postcrisis period.
Figure 1 plots the 2-year (top graph) and 10-year (bottom graph) sovereign bond yields for the economies in our sample. Based on the movements in these yields, it appears that 2-year yields have been relatively stable once the dramatic fall during the financial crisis is factored in, and therefore potentially constrained (if the range of yields is anything to go by). For 10-year yields, however, there has been considerable movement up to and including the beginning of 2015. Although the overall global trend is toward lower long-term yields, cross-country differences and the range of yields observed do not give the impression that they are constrained in any fashion. A crude analysis of yield movements, however, is not sufficiently informative about whether yields are constrained or whether they remain responsive to news events. Therefore, we turn to econometric estimation.
B. Econometric Estimates
Table 3 only includes the coefficient estimates and standard deviations on the domestic tone variables (C,) for the benchmark and ELB periods of the 3-month, 2-year, and 10-year yields for Equation (1). A case is made in the present study that how central bankers communicate their policies has a separate influence on yields, and we are interested in assessing whether this influence becomes constrained at the ELB. The other estimates are available in Appendix S1, but are not reported as they are not the focus of the analysis.
The most important observation is that the impact of the content of communication increases during the ELB period (relative to "normal" times), both in magnitude and statistical significance. One notable exception to this pattern is that the impact of the content of BoE communications decreases at the ELB. This result is consistent with the view that central bank communication can be an important policy tool once the monetary policy interest rates approach the ELB.
Turning to the impact of content, optimism generally increases sovereign yields. During the benchmark period, optimism in BoE and Riksbank policy board meeting minutes and the ECB president's introduction to the press conference move medium-to-long term sovereign yields. Despite these effects, language conveying optimism appears to have a larger impact on yields at the ELB. This is observed for the United States, Sweden, and the euro area economies. In the United States, optimistic content in FOMC communications increases yields at the longer-end of the yield curve during the ELB period. Similarly, in the euro area, optimism increases short- to medium-term yields at the ELB. Optimism, however, can also have a negative impact on sovereign bond yields; this is apparent at the longer end of the yield curve in the euro area and Canada, and along the yield curve in Sweden. Our indicator of optimism does not identify whether this is applied to future inflation or real economic growth. It is quite conceivable however, that optimism about lower inflation might lead to a reduction in sovereign yields in some economies while greater optimism about future economic growth would naturally produce the opposite result. (23)
With respect to pessimistic language, higher content reduces yields during the benchmark period in the United States, the United Kingdom, Sweden, Switzerland, and the euro area economies. Pessimistic language affects yields all along the curve, except for the United States where it only affects short-term yields. At the ELB, a similar result is observed, pessimistic language appears to be able to impact both the short and long end of the yield curve. Generally, pessimism has a negative impact on long-term yields, as observed for U.S. FOMC minutes and Riksbank communications. This undoubtedly reflects the negative effect of a downgrade in the economic outlook. At the short end of the yield curve, pessimistic language has a positive effect, as observed in the United States and Canada. This likely captures the connection between pessimism and perceived risk suggested earlier (also see Appendix S1). In the euro area economies, pessimism has a positive impact along the yield curve at the ELB; this is a plausible reaction to heightened uncertainty about the future of the Economic and Monetary Union. Importantly, our results suggest that yields can be just as responsive to negative shocks as positive shocks in the content of central bank communication when policy rates are near the ELB.
Looking at the impact of certainty clearly shows how the impact of content can change over time. The sentiment of certainty decreases yields in the United Kingdom, Switzerland, and the euro area in the benchmark period. During this period, certainty may have transmitted to market behavior through the risk channel. That is, an increase in content of language expressing certainty, could lead to a decrease in perceived risk, or in uncertainty in future policy decisions or the economic outlook. During the ELB period, however, certainty has a positive impact in the United Kingdom, the United States, Sweden, and the euro area economies. Apart from Sweden, the impact appears to be larger on the short-to-medium end of the yield curve. This may reflect more certainty in the economic outlook which signals that a change in monetary policy stance will occur in the near-term.
Turning briefly to spillovers, the magnitude of the impact of the content of U.S. FOMC communication increased yields in the euro area at the ELB, but decreased U.K. gilt yields. In all other countries, the impact of the content of FOMC statements and meeting minutes remains similar to the benchmark period. Generally, greater optimism in FOMC communications increased yields, and pessimism decreased yields except for a few instances where it increased yields on short-term bonds. Certainty is shown to have decreased yields in the euro area economies, while increasing yields in Canada and Sweden. For Canada and Sweden, greater certainty emanating from the United States could reflect higher future real economic growth. The United States, of course, is Canada's largest trading partner by far. An explanation for the euro area response is less obvious; good economic news for the United States may have been interpreted as bad news for the euro area which was still in the midst of the sovereign debt crisis while the United States was emerging from its own financial crisis. Finally, the content of ECB press conferences is not found to have any notable spillovers to Sweden or Switzerland at the ELB.
Next, we examine the estimates based on Equation (2). In addition to being based on a nonlinear functional form, the specification permits us to evaluate the time-varying sensitivity of yields--conditional on both market and central bank news--to two forms of market uncertainty and the level of the yield, as previously defined (see Equations (2) and (3)). Figure 2 plots estimates of the time-varying sensitivity of 2- and 10-year yields to changing uncertainty. The Figure is subdivided into two parts. Part (A) plots the sensitivity indicator for 2-year yields on the left and for the 10-year yields on the right for the United States, the euro area, the United Kingdom, and Switzerland. The same structure is used in Figure 2B where data for France, Germany, Canada, and Sweden are plotted. Table 4 provides the numerical average of the sensitivity of yields, that is, f([[THETA].sub.t]), during the ELB period; as it is set to unity during the benchmark period. A value below one represents a decrease in sensitivity to news and a value of zero indicates that an asset price is fully constrained.
Paralleling the results found by SW (2014b), we see that the sensitivity of 2-year U.S. Treasuries dropped as the financial crisis developed (top left corner of Figure 2A). By 2012, the 2-year yields in all the countries in the sample except Switzerland became constrained or unresponsive to news events. Indeed, during the ELB period for each country (defined in Table 1), yield sensitivity (f([[THETA].sub.t])) is reduced to a range of 0.03 in France to 0.49 in Canada and Sweden. However, the 10-year yields generally became more responsive to news events during the crisis. While some 10-year sovereign bond yields became constrained over time, others remained responsive to news through most of the crisis. And indeed, while some long-term sovereign yields are clearly restrained during the ELB period (including the euro area, France, and Switzerland), other sovereign yields may have become more responsive to news, including Sweden and the United Kingdom.
In addition, some 10-year yields, though less sensitive, appear to remain somewhat responsive to news, including the United States, Canada, and Germany. This may be related to the fact that these are safe and heavily traded assets. Canada is an interesting case because bond yields remained less responsive to news even after the policy rate was lifted from the ELB in mid-2010. These results reveal how yields can be constrained, or at least less responsive to news, even when policy rates are not at the ELB (i.e.,
Canada), while they may also remain sensitive at the ELB in some circumstances, particularly for long-term yields (i.e., United States and United Kingdom). (24)
The long-term yields in the European countries (Figure 2B) appear to be much more sensitive to news events after the crisis than in Canada and the United States, as well as the United Kingdom. This may be due to the possibility that these yields became more vulnerable to uncertainty as the euro area crisis evolved. This argument is further underscored by the fact that yield sensitivity declined for these yields after key policy events in the euro area such as the securities market program announced in May 2010 and the passages of the fourth and fifth Greek austerity packages by narrow margins in June and October 2011.
Another interesting pattern, although with too few observations to make a firm conclusion, is that the announcement of negative interest rates in Sweden in February 2015 and in Switzerland in December 2014, appear to have increased the sensitivity of yields. Similarly, short-term yields in the euro area became more sensitive to news when the ECB introduced extraordinary UMP measures in early 2015. Due to the limited sample size for the period when policy rates are negative and the ECB was engaged in a full-scale sovereign bond purchase program, it is of course difficult to reach a definite conclusion about the change in the responsiveness of yields owing to these events.
This study has empirically considered the potential constraint that the ELB poses for central banks seeking to retain the effectiveness of monetary policy. We examine the responsiveness of yields across the term structure of sovereign bonds to news in eight economies where nominal policy rates hit the ELB following the international financial crisis of 2007-2009.
Using a version of the methodology recently implemented by SW (2014a, 2014b), we extend their approach to analyze the impact of the content of central bank communications, in addition to the news content of various UMPs and macroeconomic data releases, on the term structure of interest rates. We also explicitly allow for crosscountry spillovers from the United States and the euro area.
We find that most medium- to long-term yields become constrained around 2012--approximately 3 years after most central banks approached the ELB. There are, however, some notable exceptions, including 10-year government bond yields in Sweden, the United States, and the United Kingdom, which were only slightly less responsive to news relative to a benchmark period when their "news response" was deemed to be normal (i.e., consistent with precrisis types of responses). For example, Canada's record illustrates that yields can be constrained even if the policy rate has lifted off from the ELB. Since Canada is a small-open economy whose economic fortunes are closely tied to the United States, our findings also suggest that spillover effects, which we consider but SW (2014a, 2014b) did not, may also play a role in the data dependence of yields. There is also early evidence that yields may once again become responsive to news events even when the zero boundary has been breached, as it was in Sweden at the beginning of 2015. For small-open economies, the other variable that is ignored in our analysis is the exchange rate which no doubt also influences some of our findings.
A general reduction in the sensitivity of yields need not, however, imply that monetary policy becomes less effective. We conclude that sovereign bond yields become more sensitive to the content of central bank communications for longer maturities after the policy rate reached the ELB. This suggests that monetary policy remains an effective tool at least for a period of time even after the ELB is reached, as longer-term sovereign bond yields are not immediately constrained by the ELB. Nevertheless, it is also likely that it will be more difficult for central banks to maintain the effectiveness of monetary policy since, as demonstrated in this paper, the responsiveness of asset prices will likely eventually taper off.
Our empirical analysis can be extended in a number of directions. First, optimism, pessimism, and certainty are not the only sentiments that can be extracted from verbal or written forms of central bank communication. Next, for small-open economies, interest rate and exchange rates are jointly determined. Since news can affect both, there is scope for a more sophisticated analysis of the kind conducted here. Finally, if central banks ever return to "normal" interest rates we could then investigate whether a "new normal" has emerged in how yields across the term structure respond to news from various sources and whether the resulting spillovers would also change. All of these ideas are left for future research.
Amirault, D., and B. O'Reilly. "The Zero Lower Bound on Nominal Interest Rates: How Important Is It?" Bank of Canada Working Paper No. 2001-6, April 2001.
Armstrong, J. S. "Combining Forecasts." in Principles of Forecasting: A Handbook for Researchers and Practitioners, edited by J. S. Armstrong. Norwell, MA: Kluwer Academic Publishers, 2001, 1-19.
Baker. S. R., N. Bloom, and S. J. Davis. "Measuring Economic Policy Uncertainty." Quarterly Journal of Economics, 131(4), 2016, 1593-636.
Ball, L., J. Gagnon, P. Honohan, and S. Krogstrup. "What Else Can Central Bank Do?" Geneva Reports on the World Economy No. 18, September 2016.
Bernanke, B. S. "Monetary Policy Since the Onset of the Crisis." Speech delivered at the Federal Reserve Bank of Kansas City Economic Symposium, Jackson Hole, WY, August 31, 2012.
Bernanke, B., V. Reinhart. and B. Sack. "Monetary Policy Alternatives at the Zero Lower Bound." Brookings Papers on Economic Activity, 2, 2004, 1-100.
Blinder, A. S. "Summary Panel--Monetary Policy at the Zero Lower Bound: Balancing the Risks." Journal of Money, Credit and Banking, 32(2). 2000. 1093-99.
Blinder, A., M. Ehrmann, M. Fratzscher, J. De Haan, and D. -J. Jansen. "Central Bank Communication and Monetary Policy: A Survey of Theory and Evidence." NBER Working Paper Series No. 13932, 2008.
Buiter, W. H. "Negative Nominal Interest Rates: Three Ways to Overcome the Zero Lower Bound." North American Journal of Economics and Finance, 20(3). 2009, 213-38.
Carney, M. "Monetary Policy After the Fall." Speech delivered at the Eric J. Hanson Memorial Lecture, University of Alberta, Edmonton, AB, May 1, 2013.
Christensen, J., and G. Rudebusch. "The Response of Interest Rates to US and UK Quantitative Easing." Economic Journal, 122, 2012, F385-414.
Chung, H., J.-P. Laforte, D. Reifschneider. and J. C. Williams. "Underestimated the Likelihood and Severity of Zero Lower Bound Events." Journal of Money, Credit and Banking, 44(1), 2012,47-82.
Danker, D. J., and M. M. Luecke. "Background on FOMC Meeting Minutes." Federal Reserve Bulletin, Spring. 2005, 175-79.
Davis, J. S., and M. A. Wynne. "Central Bank Communications: A Case Study." Federal Reserve Bank of Dallas Globalization and Monetary Policy Institute Working Paper No. 283,2016.
De Carvalho, C. V, E. Hsu. and F. Nechio. "Fed Communication and the Zero Lower Bound." FRBSF Economic Letter, 21,2016.
De Rezende, R., D. Kjellberg, and O. Tysklind. "Effects of the Riksbank's Government Bond Purchases on Financial Prices." Economic Commentaries No. 13, Sveriges Riksbank, 2015.
ECB. "The New Euro Area Yield Curves." ECB Monthly Bulletin, February, 2008, 95-103.
Ehrmann, M., and M. Fratzscher. "Communication by Central Bank Committee Members: Different Strategies, Same Effectiveness?" Journal of Money, Credit and Banking, 39(2/3), 2007, 509-41.
Filardo, A., and B. Hofmann. "Forward Guidance at the Zero Lower Bound." BIS Quarterly Review, March, 2014, 37-54.
Fratzscher. M., M. Lo Duca, and R. Straub. "On the International Spillovers of US Quantitative Easing." The Economic Journal, 2016: https://doi.org/10.1111/ecoj.12435.
Gagnon, J. "Quantitative Easing: An Underappreciated Success." Peterson Institute for International Economics Policy Brief, 2016. PB16-14.
Gagnon, J., M. Raskin, J. Remache, and B. Sack. "The Financial Market Effects of the Federal Reserve's Large-Scale Asset Purchases." International Journal of Central Banking, 7(1), 2011, 3-43.
Gerlach, S. "Interest Rate Setting by the ECB, 1999-2006: Words and Deed." International Journal of Central Banking, 3(3), 2007, 1-45.
Gilchrist, S., D. Lopez-Salido, and E. Zkrajsek. "Monetary Policy and Real Borrowing Costs at the Zero Lower Bound." FEDS Working Paper No. 2014-39, May 2014.
Goodfriend, M. "Overcoming the Zero Lower Bound on Interest Rate Policy." Journal of Money, Credit and Banking, 32(4:2), 2000, 1007-35.
Gurkaynak, R., B. Sack, and E. Swanson. "The Sensitivity of Long-Term Interest Rates to Economic News: Evidence and Implications." American Economic Review, 95(1), 2005a. 426-36.
--. "Do Actions Speak Louder Than Words? The Response of Asset Prices to Monetary Policy Actions and Statements." International Journal of Central Banking, 1(1), 2005b, 55-93.
Hart, R. P., J. P. Childers, and C. J. Lind. Political Tone: How Leaders Talk and Why. Chicago, IL: University of Chicago Press, 2013.
He, Z. "Evaluating the Effect of the Bank of Canada's Conditional Commitment Policy." Discussion Paper No. 11, 2010.
Jansen, D.-J., and J. De Haan. "Has ECB Communication Been Helpful in Predicting Interest Rate Decisions? An Evaluation of the Early Years of the Economic and Monetary Union." Applied Economics, 41(16), 2010, 1995-2003.
Joyce, M., A. Lasaosa, I. Stevens, and M. Tong. "The Financial Market Impact of Quantitative Easing in the United Kingdom." International Journal of Central Banking. 7(3), 2011, 113-61.
Krishnamurthy, A., and A. Vissing-Jorgensen. "The Effects of Quantitative Easing on Interest Rates: Channels and Implications for Policy." Brookings Papers on Economic Activity, Fall, 2011, 215-87.
Kuttner, K. "Monetary Policy Surprises and Interest Rates: Evidence from the Fed Funds Futures Market." Journal of Monetary Economics, 47(3), 2001, 523-44.
Kuttner, K., and A. Posen. "The Difficulty of Discerning What's Too Tight? Taylor Rules and Japanese Monetary Policy." North American Journal of Economics and Finance, 15(1), 2004, 53-74.
Levin, A., D. Lopez-Salido, E. Nelson, and T Yun. "Limitations on the Effectiveness of Forward Guidance at the Zero Lower Bound." International Journal of Central Banking, 6(1), 2010, 143-89.
Lombardi, D., P. L. Siklos, and S. St. Amand. "Exchange Rates, Central Bank News and the Zero Lower Bound." Applied Economics Letters, 24(4), 2016, 269-72.
Loughran. T, and B. McDonald. "When Is a Liability Not a Liability? Textual Analysis. Dictionaries, and 10-Ks." Journal of Finance, 66(1), 2011. 35-65.
MacKinlay, A. C. "Event Studies in Economics and Finance." Journal of Economic Literature. 35(March). 1997, 13-39.
McCallum. B. "Theoretical Analysis Regarding a Zero Lower Bound on Nominal Interest Rates." NBER Working Paper No. 7677, 2000.
Meade, E. E. "The FOMC: Preferences, Voting, and Consensus." Federal Reserve Bank of St. Louis Review. 2005, 93-102.
Meade. E. E., N. A. Burk, and M. Josselyn. "The FOMC Meeting Minutes: An Assessment of Counting Words and the Diversity of Views." FEDS Notes No. 2015-26, May 2015.
Riboni, A., and F. Ruge-Marcia. "Dissent in Monetary Policy Decisions." Journal of Monetary Economics, 66, 2014, 137-54.
Rogers, J. H., C. Scotti, and J. H. Wright, "Evaluating Asset-Market Effects of Unconventional Monetary Policy: A Cross-Country Comparison." International Finance Discussion Paper No. 1101, 2014.
Romer. C, and D. Romer. "Does Monetary Policy Matter? A New Test in the Spirit of Friedman and Schwartz," in NBER Macroeconomic Annual 1989, Vol. 4. edited by O. J. Blanchard and S. Fischer. Cambridge, MA: National Bureau of Economic Research, 1989, 121-84.
--. "A New Measure of Monetary Policy Shocks:
Derivation and Implications." American Economic-Review, 94(September), 2004, 1055-84.
Siklos, P. L. "Sources of Disagreement in Inflation Forecasts: An International Empirical Investigation." Journal of International Economics, 90, 2013, 218 - 31.
Siklos, P., and A. Spence. "Faceoff: Should the Bank of Canada Release Its Projections of the Interest Rate Path? The Cases for and Against." CD. Howe Institute Commentary. 2010.
Sims, C. "Implications of Rational Inattention." Journal of Monetary Economics, 50(April), 2003, 665-90.
Summers, L. "Panel Discussion: How Should Long-Term Monetary Policy Be Determined?" Journal of Money, Credit and Banking, 23(3), 1991, 625-31.
Svensson. L. E. O. "Estimating and Interpreting Forward Interest Rates: Sweden 1992-1994." NBER Working Paper No. 4871, 1994.
--. "Monetary Policy and Financial Markets at the Effective Lower Bound." Journal of Money, Credit and Banking, 42(sl), 2010,229-42.
Swanson, E., and J. C. Williams. "Measuring the Effect of the Zero Lower Bound on Yields and Exchange Rates in the U.K. and Germany." Federal Reserve Bank of San Francisco Working Paper No. 2013-21, August 2014a.
--. "Measuring the Effect of the Zero Lower Bound on Medium and Longer-Term Interest Rates." American Economic Review, 104(October), 2014b, 3154-85.
Tsay, R. Analysis of Financial Time Series. 3rd ed. New York: John Wiley & Sons, 2010.
Ueda, K. "The Bank of Japan's Struggle with the Zero Lower Bound on Nominal Interest Rates: Exercises in Expectations Management." International Finance, 8(2), 2005. 329-50.
Wiederholt. M. "Dispersed Inflation Expectations and the Zero Lower Bound." Working Paper. Northwestern University, February 2012.
Winkelmann, L., M. Bibinger, and T. Linzert. "ECB Monetary Policy Surprises: Identification through Cojumps in Interest Rates." Journal of Applied Econometrics. 31 (4), 2016,613-29.
Woodford, M. "Methods of Policy Accommodation at the Interest-Rate Lower Bound." in The Changing Policy Landscape. Federal Reserve Bank of Kansas City Economic Policy Symposium. Kansas City, MO: Federal Reserve Bank of Kansas City, 2012, 185-288.
Yellen, J. L. "Monetary Policy and Economic Recovery." Speech delivered at the Economic Club of New York. New York, April 16,2014.
Additional supporting information may be found online in the Supporting Information section at the end of the article.
Appendix S1. Annex A: content and construction of the dictionaries.
Annex B: specific variables and sources of data.
Annex C: dates of UMP announcement included in the announcement dummies.
Annex D: summary statistics for macroeconomic surprises.
Annex E: correlation coefficients of content variables and key macroeconomic and financial indicators.
Annex F: full regression results from Equation (1).
DOMENICO LOMBARDI, PIERRE L. SIKLOS and SAMANTHA ST. AMAND (*)
(*) This is a CIGI-sponsored research project. The opinions in the paper are those of the authors and not the institutions that supported this research. Previous versions of this paper were presented at meetings of the Canadian Economics Association. the Asian Finance Association, the World Finance Conference, and the Symposium of the Society for Nonlinear Dynamics and Econometrics. Comments by the Editor and an anonymous referee are gratefully acknowledged. Lombardi: Director of the Global Economy Program, Centre for International Governance Innovation, Waterloo, Canada. Phone +1 519 885 2444 ext. 7227, Fax +1 519 885 5450. E-mail firstname.lastname@example.org
Sikhs: Senior Fellow. Centre for International Governance Innovation, Waterloo, Canada; Professor, Department of Economics, Wilfrid Laurier University, and Balsillie School of International Affairs, Waterloo, Canada. Phone + 1 519 884 0710 ext. 2559, Fax+1 519 884 0201. E-mail email@example.com
St. Amund: Senior Research Associate, Centre for International Governance Innovation, Waterloo, Canada. Phone + 1 519 885 2444 ext. 7232, Fax +1 519 885 5450, E-mail firstname.lastname@example.org
BoC: Bank of Canada
BoE: Bank of England
CI: Confidence Interval
ECB European Central Bank
ELB: Effective Lower Bound
FOMC: Federal Open Market Committee
MRO : Marginal Refinancing Operation
QE: Quantitative Easing
SNB Swiss National Bank
SW: Swanson and Williams
UMP : Unconventional Monetary Policy
ZLB: Zero Lower Bound
(1.) By "normal," we are referring to interest rate levels that are sufficiently above the ELB so that the concerns that are the subject of this study are no longer relevant. Of course, what is considered normal can be period-specific. s we shall see, however, our results are robust to the selection of reasonable thresholds above which yields are deemed to be "normal."
(2.) There is an implicit assumption, informed by the well-known Fisher equation, that nominal interest rates and (expected) inflation move in the same direction. One worry that emerged during the 1990s concerning the prospect of low inflation is that it might increase the chances of hitting the ZLB (see, e.g., Summers 1991). But the two need not go hand in hand. For example, the 1950s and early 1960s were a period of relatively low inflation but the risks of the ZLB were low. However, recent evidence suggests that previous research understated the threat of reaching the ZLB (Chung et al. 2012).
(3.) Given the small number of observations and historical examples that combine low inflation with low interest rates, many of the pre-2007 studies rely on model simulation results. For a comprehensive early review of the literature for the United States, Canada, and Japan, see Amirault and O'Reilly (2001).
(4.) Whether this is the result of a form of rational inattention is unclear. Nevertheless, the notion that not all agents react to news at the same time is a possibility that is now routinely admitted. We do not, however, pursue this line of investigation any further.
(5.) The use of intradaily data raises some technical challenges. For example, daily data are often nonstationary in levels but stationary in first differences; the time series properties of intradaily data can be rather different (Tsay 2010).
(6.) Filardo and Hofmann (2014) is a recent survey of the experience with forward guidance in the United States, Japan, the United Kingdom, and the euro area. For the Canadian case, see Siklos and Spence (2010) and He (2010). For a more critical view of the earlier U.S. experience, see Levin et al. (2010).
(7.) Two recent studies have analyzed the impact of U.S. FOMC communications at the ELB (Davis and Wynne 2016; De Carvalho, Hsu, and Nechio 2016). These studies use a different approach to measuring communication than the present study. They find that U.S. FOMC communications continued to impact long-term yields during the period where the federal funds rate was at the ELB.
(8.) The data for the euro area yield curve are retrieved from the ECB. The terra structure is estimated by imposing a functional form between interest rates and time to maturity, using an approach proposed by Svensson (1994). Refer to ECB (2008) for more information.
(9.) Switzerland typically operates with a flexible exchange rate; however, from 6 September 2011 to 15 January 2015, the SNB imposed a minimum exchange rate with the euro of 1.20 Swiss Francs. Denmark also lowered its policy rate to the ELB; but it is not part of this study because it maintains a fixed exchange rate with the euro.
(10.) See one of several crisis timelines. For example, https://www.stlouisfed.org/Financial-Crisis (Federal Reserve Bank of St. Louis).
(11.) Japan is excluded because the sample period does not extend far enough to create a significant benchmark where yields were not constrained by the ELB.
(12.) It is conceivable that the magnitude of surprise macroeconomic announcements experienced significant changes throughout the period owing to the 2007-2009 international financial crisis. However, there were few statistically significant differences in macroeconomic surprises between periods (see Appendix S1). We performed sensitivity analyses to verify whether normalizing the variable using different sample periods, versus normalizing using the full sample, influenced the regression results, and found no statistically significant differences.
(13.) We are grateful to a referee for pointing out this issue.
(14.) To reduce noise, we drop observations on dates where a news event does not occur; that is, when [Z.sub.t] [X.sub.t] and [C.sub.t] are all equal to zero. However, we also ran the regression on the full data set and the results were unchanged.
(15.) Some news is released after financial markets in other time zones have closed; for example, FOMC press releases and meeting minutes are released at 2 p.m. EST when markets in Europe are closed (8 p.m. CET). These events were appropriately adjusted to represent a shock on the following trading day.
(16.) Two other measures of surprise were used to test the sensitivity of the results. One alternative measurement is the change in the content variable from one communication to the next. The second subtracts the precrisis median; the content of communication before the crisis could be taken as a sort of benchmark, where a deviation from this level could be interpreted as a shock. The results remain unchanged.
(17.) Other variants were considered with no impact on the conclusions. This coding is consistent with the literature (Fratzscher, Lo Duca, and Straub 2016), and is preferable for a sample that spans several time zones.
(18.) As explained by SW (2014a) there are relatively few macroeconomic announcements per year. In addition, there are typically only 4 to 12 central bank announcements or press releases per year. Hence, rolling or recursive estimates would not be terribly informative.
(19.) Equation (2) allows for time variation as explained by Equation (3). However, the estimated coefficients inside the brackets in Equation (2) are mean responses that are not time-varying. Some of the more important surprises are likely already incorporated in Equation (3). Hence, omitting them from X is intended to reduce the possibility of mis-specification. This occurred for U.S. nonfarm payroll surprises on U.S. and Canadian 2- and 10-year yields. German and euro area 2-year yields, and Swiss 10-year yields; for U.S. retail sales on U.S. 2- and 10-year yields, and French and Swedish 2-year yields; for U.S. manufacturing index on 2-year Canadian and Swiss yields; durable goods orders on 2-year Swiss and Swedish yields; and existing home sales and consumer credit on Swiss 2-year yields.
(20.) There are several reasons why identifying the ELB is not straightforward. Most importantly, the ELB rate differs by country. In addition, the identified ELB may change overtime (refer to Table 1). Finally, since the central banks in our data set have typically adopted a 2% inflation target, the chosen threshold also has the advantage of approximately capturing the dividing line between positive and negative real returns. We tested the sensitivity of this level by setting a 3% threshold on interest rates. Our results remain unchanged.
(21.) This is somewhat of an exaggeration since negative rates are possible (see below). However, it is unclear how far below zero central banks are prepared to go.
(22.) We performed a sensitivity analysis of the potential asymmetric impact of macroeconomic news surprises on yields, and found no statistically significant differences between the impact of positive and negative shocks on bond yields.
(23.) In Canada, at least, the period of the ELB at the time coincided with a sharp increase in energy prices which may have played a factor in the BoC abandoning its conditional commitment a couple of months before forward guidance was due to expire. In Sweden, it was rising housing prices that led the Riksbank to raise the policy rate, only to reverse its decision later.
(24.) In a similar analysis applied to exchange rates, Lombardi, Siklos, and St. Amand (2016) find that the exchange rates of small-open economies (Canadian dollar and Swedish krona) remain highly responsive to news event at the ZLB. Although yields are constrained, the exchange rate mechanism may still act as an effective shock absorber.
TABLE 1 How the Effective Lower Bound Is Defined in Different Economies Economy ELB Sample Period ELB Rate (a) Canada 21 April 2009 to 31 -0.50% target overnight May 2010 lending rate Euro area 4 July 2014 to 31 0% target MRO rate March 2015 (b) Switzerland 3 August 2011 to 31 -1.25 to 0.25% target range for March 2015 (b) the 3-month Libor United Kingdom 5 March 2009 to 31 0.5% target bank rate March 2015 (b) United States 16 December 2008 to 0.00 to 0.25% target range for the 31 March 2015 (b) federal funds rate Sweden 2 July 2009 to June 30 -0.50% repo rate 2010; 9 July 2014 to 31 March 2015 (b) Economy Notes Canada The press release of the interest rate announcement on 21 April 2009 stated that the Bank judges 0.25% to be the ELB. Revised in December 2015 to -0.50% (Bank of Canada 2015). Euro area Decreased MRO rate to 0.15% on 4 June 2014, which Draghi indicated "for all practical purposes, we have reached the lower bound" but indicating further adjustments are possible. The target MRO rate was decreased further in September 2014 and December 2015; and lowered to 0% in March 2016. It remains unclear what the ELB is. Switzerland The Libor rate was 0.00 to 0.25% (aiming for as close to zero as possible) between 3 August 2011 and 17 December 2014. It was lowered to -0.75 to 0.25%, aiming for negative territory from 18 December 2014 to 14 January 2015. On 15 January 2015, it was further lowered to -1.25 to -0.25%. United Kingdom United States Sweden The policy rate was 0.25% from 2 July 2009 to 30 June 2010. It was lowered back to 0.25% on 9 July 2014 followed by further cuts. Note that the Riksbank never stated that 0.25% was the ZLB; in fact the committee meeting minutes indicate that they did not believe the ZLB was static and it could in fact be negative > (a) As of February 2017. (b) The ELB end date of 31 March 2015 represents the end of the sample period of the analysis. Source: Central banks. TABLE 2 Uncertainty, Optimism, and Pessimism Sentiment in Monetary Policy Communication Certainty Source Precrisis Postcrisis BoC Press Release 3.97(1.20) 4.13(1.16) ECB Introductory Statement to the 3.50 (*) (0.80) 4.08 (*) (0.76) Press Conference Swedish Riksbank Press Release 3.51 (1.09) 3.12(1.15) Swedish Riksbank Minutes of the 0.58 (*) (0.89) 1.75 (*) (0.36) Monetary Policy Meeting SNB Monetary Policy Assessment 2.15(1.88) 2.73(1.55) BoE Press Release 7.89 (*) (2.47) 8.88 (*) (2.11) BoE Minutes of the Monetary Policy 1.12 (*) (0.59) 1.77 (*) (0.67) Committee Meeting U.S. FOMC Monetary Policy 3.94 (*) (1.94) 7.80 (*) (1.66) Statement U.S. FOMC meeting minutes 2.23 (*) (0.80) 3.06 (*) (0.44) Optimism Source Precrisis Postcrisis BoC Press Release 1.67 (*) (0.81) 2.49 (*) (0.74) ECB Introductory Statement to the 2.67(0.71) 2.86 (0.49) Press Conference Swedish Riksbank Press Release 1.12 (*) (0.49) 1.56 (*) (0.67) Swedish Riksbank Minutes of the 1.38 (*) (0.27) 1.53 (*) (0.24) Monetary Policy Meeting SNB Monetary Policy Assessment 1.32(0.62) 1.11 (0.40) BoE Press Release 1.15(0.56) 1.08 (0.37) BoE Minutes of the Monetary Policy 0.95 (*) (0.21) 1.26 (*) (0.31) Committee Meeting U.S. FOMC Monetary Policy 1.87(1.06) 2.22 (0.59) Statement U.S. FOMC meeting minutes 1.99 (*) (0.42) 1.64 (*) (0.32) Pessimism Source Precrisis Postcrisis BoC Press Release 1.32 (*) (0.74) 2.36 (*) (0.79) ECB Introductory Statement to the 1.34 (*) (0.45) 1.65 (*) (0.55) Press Conference Swedish Riksbank Press Release 1.34(0.68) 1.63(0.82) Swedish Riksbank Minutes of the 1.81 (0.47) 1.75(0.34) Monetary Policy Meeting SNB Monetary Policy Assessment 1.46 (*) (0.98) 2.14 (*) (0.78) BoE Press Release 1.63(0.92) 1.16(1.10) BoE Minutes of the Monetary Policy 1.58 (*) (0.48) 1.90 (*) (0.48) Committee Meeting U.S. FOMC Monetary Policy 1.67(1.02) 1.66 (0.87) Statement U.S. FOMC meeting minutes 1.98(0.62) 1.75(0.56) Notes: Values represent the number of positive or negative words as a percentage of the total word count. Standard deviations are in parenthesis. Precrisis period is 1 January 2001 to 30 September 2008; postcrisis period is 1 October 2008 to 31 March 2015. (*) Statistically significant difference in means in pre-and postcrisis periods at the 99% confidence level. Source: Authors' calculations. TABLE 3 Results from Equation (1) (Benchmark and ELB Period) Benchmark 3-Month 2-Year United States FOMC Press Release Certainty -4.79 0.22 (2.96) (2.55) FOMC Press Release Optimism 2.57 1.47 (1.82) (1.76) FOMC Press Release Pessimism -4.04 (***) 0.01 (1.33) (1.18) FOMC Minutes Certainty -0.30 0.64 (1.53) (1.37) FOMC Minutes Optimism 1.10 0.53 (0.94) (1.19) FOMC Minutes Pessimism 0.66 0.95 (1.68) (1.08) N 645 646 United Kingdom BoE Press Release Certainty 0.22 -1.49 (*) (0.29) (0.83) BoE Press Release Optimism 0.96 2.53 (0.67) (1.60) BoE Press Release Pessimism -3.61 (***) -5.04 (***) (0.63) (1.25) BoE Minutes Certainty -1.07 (*) -0.56 (0.55) (0.79) BoE Minutes Optimism 1.13 2.47 (**) (0.75) (1.06) BoE Minutes Pessimism -0.88 (**) -1.59 (**) (0.40) (0.71) N 920 920 Canada BoC Press Release Certainty -0.22 0.40 (0.83) (0.92) BoC Press Release Optimism 1.59 1.39 (1.10) (1.07) BoC Press Release Pessimism -0.90 -0.98 (1.05) (1.04) N 1,101 1,102 Euro Area ECB Introduction to the Press -0.86 (*) -0.61 Conference Certainty (0.51) (1.03) ECB Introduction to the Press 0.17 0.15 Conference Optimism (0.58) (0.80) ECB Introduction to the Press -0.17 -1.27 (*) Conference Pessimism (0.64) (0.74) N 667 667 Germany ECB Introduction to the Press -1.11 -1.06 Conference Certainty (1.01) (1.04) ECB Introduction to the Press -1.55 -0.49 Conference Optimism (1.01) (0.75) ECB Introduction to the Press -1.60 (*) -1.43 (*) Conference Pessimism (0.91) (0.75) N 964 964 France ECB Introduction to the Press -0.62 -0.62 Conference Certainty (0.49) (0.84) ECB Introduction to the Press -0.31 -0.94 Conference Optimism (0.49) (0.82) ECB Introduction to the Press -0.95 (**) -2.09 (**) Conference Pessimism (0.39) (1.02) N 946 946 Sweden Riksbank Press Release 1.53 0.66 Certainty (1.63) (1.39) Riksbank Press Release -0.03 -0.99 Optimism (1.55) (1.77) Riksbank Press Release -4.07 (*) -2.02 Pessimism (2.35) (1.65) Riksbank Minutes Certainty 0.24 0.17 (0.24) (0.66) Riksbank Minutes Optimism 0.13 1.38 (**) (0.36) (0.68) Riksbank Minutes Pessimism -0.31 0.48 (0.24) (0.79) N 839 839 Switzerland SNB Press Release Certainty -0.23 3.83 (13.67) (4.33) SNB Press Release Optimism -2.16 0.86 (3.57) (1.61) SNB Press Release -1.67 -1.32 Pessimism (5.24) (1.54) N 290 290 10-Year 3-Month United States FOMC Press Release Certainty 0.52 -0.03 (1.94) (0.25) FOMC Press Release Optimism -0.04 0.34 (1.14) (0.25) FOMC Press Release Pessimism 0.11 0.10 (0.95) (0.19) FOMC Minutes Certainty 0.33 0.72 (**) (1.28) (0.28) FOMC Minutes Optimism 1.41 -0.14 (0.99) (0.33) FOMC Minutes Pessimism 1.63 (*) 0.55 (*) (0.96) (0.32) N 646 935 United Kingdom BoE Press Release Certainty -0.75 0.43 (***) (0.62) (0.16) BoE Press Release Optimism 1.75 -0.21 (1.21) (0.20) BoE Press Release Pessimism -2.15 (**) -0.51 (**) (1.04) (0.21) BoE Minutes Certainty -0.57 0.03 (0.62) (0.08) BoE Minutes Optimism 1.89 (**) -0.03 (0.80) (0.08) BoE Minutes Pessimism -1.62 (***) 0.02 (0.54) (0.07) N 920 1,018 Canada BoC Press Release Certainty 0.19 -1.24 (0.56) (0.83) BoC Press Release Optimism -0.01 1.29 (0.58) (1.13) BoC Press Release Pessimism 0.19 10.21 (**) (0.69) (4.23) N 1,102 155 Euro Area ECB Introduction to the Press -0.18 1.63 (***) Conference Certainty (0.46) (0.54) ECB Introduction to the Press 0.99 (*) 1.74 Conference Optimism (0.58) (1.46) ECB Introduction to the Press -0.53 0.89 Conference Pessimism (0.40) (0.57) N 667 104 Germany ECB Introduction to the Press -0.11 -0.08 Conference Certainty (0.55) (0.54) ECB Introduction to the Press -0.11 4.32 (***) Conference Optimism (0.57) (1.62) ECB Introduction to the Press -1.26 (***) 3.34 (***) Conference Pessimism (0.41) (1.13) N 110 110 France ECB Introduction to the Press -0.41 2.99 (***) Conference Certainty (0.68) (0.50) ECB Introduction to the Press -0.86 5.24 (***) Conference Optimism (0.68) (0.30) ECB Introduction to the Press -1.82 (**) 6.15 (***) Conference Pessimism (0.71) (0.74) N 946 104 Sweden Riksbank Press Release -1.46 -0.24 Certainty (0.96) (1.05) Riksbank Press Release -0.35 -1.71 (*) Optimism (1.36) (1.02) Riksbank Press Release -0.55 -4.27 (***) Pessimism (1.12) (1.54) Riksbank Minutes Certainty 0.69 -0.21 (0.61) (0.43) Riksbank Minutes Optimism 0.87 -0.88 (***) (0.80) (0.29) Riksbank Minutes Pessimism 0.83 -3.39 (***) (0.91) (0.78) N 839 299 Switzerland SNB Press Release Certainty -6.89 (**) -3.50 (3.10) (6.06) SNB Press Release Optimism -2.18 (*) -7.73 (1.23) (10.16) SNB Press Release -2.96 (**) 5.56 (***) Pessimism (1.25) (1.78) N 290 644 ELB 2-Year 10-Year United States FOMC Press Release Certainty 0.02 0.20 (0.78) (1.59) FOMC Press Release Optimism 2.51 (**) 4.72 (**) (1.13) (2.28) FOMC Press Release Pessimism 0.61 0.72 (0.98) (1.90) FOMC Minutes Certainty 1.50 (**) 1.19 (0.70) (1.13) FOMC Minutes Optimism 1.12 (*) 1.48 (0.59) (0.99) FOMC Minutes Pessimism 0.33 -1.91 (**) (0.58) (0.87) N 935 935 United Kingdom BoE Press Release Certainty 0.08 0.12 (0.81) (1.34) BoE Press Release Optimism -0.43 1.86 (1.47) (2.65) BoE Press Release Pessimism 0.20 -1.45 (0.83) (1.44) BoE Minutes Certainty 0.50 0.66 (0.42) (0.65) BoE Minutes Optimism -0.33 -0.33 (0.56) (0.71) BoE Minutes Pessimism -0.06 -0.83 (0.44) (0.68) N 1,018 1,018 Canada BoC Press Release Certainty -0.97 1.89 (1.86) (1.17) BoC Press Release Optimism -0.86 -2.18 (**) (1.90) (1.08) BoC Press Release Pessimism 10.33 -0.45 (6.95) (5.34) N 155 155 Euro Area ECB Introduction to the Press -0.62 5.83 (**) Conference Certainty (0.53) (3.22) ECB Introduction to the Press 2.62 (**) -5.52 Conference Optimism (1.06) (4.96) ECB Introduction to the Press 1.66 (***) -2.88 Conference Pessimism (0.59) (3.65) N 104 104 Germany ECB Introduction to the Press 0.08 -0.33 Conference Certainty (0.42) (1.47) ECB Introduction to the Press 1.40 -4.83 (**) Conference Optimism (1.30) (2.24) ECB Introduction to the Press 0.52 4.62 (**) Conference Pessimism (0.65) (2.17) N 110 964 France ECB Introduction to the Press 0.67 (**) 1.90 (*) Conference Certainty (0.27) (1.06) ECB Introduction to the Press 1.04 (***) 0.81 Conference Optimism (0.36) (0.69) ECB Introduction to the Press 2.40 (***) 5.10 (***) Conference Pessimism (0.46) (1.40) N 104 104 Sweden Riksbank Press Release 3.04 3.94 (***) Certainty (1.87) (1.37) Riksbank Press Release -4.39 (***) -4.19 (***) Optimism (1.60) (1.20) Riksbank Press Release -0.86 -4.26 (**) Pessimism (2.27) (1.97) Riksbank Minutes Certainty -0.36 0.17 (1.43) (1.30) Riksbank Minutes Optimism -1.00 0.10 (2.05) (1.00) Riksbank Minutes Pessimism -5.12 -3.76 (6.27) (2.55) N 299 299 Switzerland SNB Press Release Certainty -1.27 2.40 (2.25) (1.50) SNB Press Release Optimism 2.05 -1.13 (1.54) (1.25) SNB Press Release 2.88 -1.81 Pessimism (1.90) (1.22) N 644 644 Notes: Shown are coefficients and standard deviation in brackets for estimates of Equations (1) for the benchmark and ELB periods for each country. All regressions use Newey-West standard errors. N is the number of observations; dates when there are no news events are excluded from the sample. (***) Significant at 1% level; (**)significant at 5% level; (*)significant at 10% level. TABLE 4 Value of the Constraint to News at the ELB (f([[THETA].sub.t])) GBR USA CAN SWE 2-Year 0.30 (*) 0.28 (*) 0.49 (*) 0.49 (*) yield (-0.01; 0.62) (-0.25; 0.82) (-0.73; 1.71) (-1.87; 2.85) 10- 1.24 (*) 0.79 (*) 0.65 (*) 1.28 (*) Year (0.34; 2.15) (-0.33; 1.91) (-0.31; 1.61) (0.09; 2.48) yield CHE DEU FRA EUR 2-Year 0.38 (*) 0.29 (*) 0.03 (*) 0.21 (*) yield (-0.56; 1.33) (-0.03; 0.62) (-0.32; 0.38) (-0.18; 0.59) 10- 0.17 (*) 0.41 (*) -0.19 (*) -0.15 (*) Year (-0.18; 0.51) (-0.39; 1.21) (-0.56; 0.19) (-0.68; 0.40) yield Notes: Displayed are the values of the level of constraint to news at the ELB (value of f([[THETA].sub.t])) and the 95% confidence interval in brackets. As f([[THETA].sub.t]) is set to unity during the benchmark period--a period where yields are expected to respond normally to news--a value greater than one suggests higher than normal responsiveness to news and a value below one represents lower than normal sensitivity to news. (*) Mean value of f([[THETA].sub.t]) during the ELB period is statistically different than one.
|Printer friendly Cite/link Email Feedback|
|Author:||Lombardi, Domenico; Siklos, Pierre L.; Amand, Samantha St.|
|Publication:||Contemporary Economic Policy|
|Date:||Jan 1, 2019|
|Previous Article:||DYNAMIC ENTRY WITH DEMAND AND SUPPLY SIDE SPILLOVERS.|
|Next Article:||REAL ESTATE AND THE GREAT CRISIS: LESSONS FOR MACROPRUDENTIAL POLICY.|