Economic Policy Uncertainty and Inflation Expectations

Theory and evidence suggest that in an environment of well-anchored expectations, temporary economic news or shocks should not affect agents' expectations of inflation in the long term. Our estimated structural VARs show that both long- and short-term inflation expectations are sensitive to policy-related uncertainty shocks. While economic activity contracts, long-term inflation expectations raise in response to such shocks. These results suggest that observed uncertainty about the stance and perceived effectiveness of policy raises concerns about future inflation and entails additional risks to central banks' hard-won inflation credibility.

ABSTRACT: Theory and evidence suggest that in an environment of well-anchored expectations, temporary economic news or shocks should not affect agents' expectations of inflation in the long term. Our estimated structural VARs show that both long-and short-term inflation expectations are sensitive to policy-related uncertainty shocks. While economic activity contracts, long-term inflation expectations raise in response to such shocks. These results suggest that observed uncertainty about the stance and perceived effectiveness of policy raises concerns about future inflation and entails additional risks to central banks' hard-won inflation credibility. Policy-related uncertainty is seen by academics and market participants as a prominent contributor to the overall economic uncertainty observed during the recent crisis.
In addition, a rapidly growing theoretical and empirical literature suggests that uncertainty has recessionary effects on economic activity. To date, the literature has focused on the effects of policy uncertainty on economic outcomes, such as output, investment, consumption and unemployment. However, less is known whether uncertainty caused by economic policy affects agents' expectations, and more specifically those about inflation. In this paper, we investigate this question empirically, by studying the dynamic relationship between policy-related uncertainty and measures of inflation expectations of professional forecasters.
We estimate structural Bayesian VARs, linking policy uncertainty with inflation expectations while accounting for a measure of economic activity and monetary policy, for the US and the euro area. We use the index of Baker, Bloom, and Davis (2012)  We find several effects of policy uncertainty. First, economic activity contracts given a surprise innovation to policy uncertainty. Second, inflation expectations are also responsive: short-term ones fall, reflecting the slack economy while long-term inflation expectations rise. Third, monetary policy-related uncertainty is not always the important factor behind the dynamics of inflation expectations. Fiscal policy-related uncertainty seems to also play an important role. Furthermore, monetary policy appears to face a trade-off between responding to the state of the economy and to long-run inflation expectations. Given an unexpected innovation to policy uncertainty, central banks lower interest rates strongly, resembling the response of a central bank that follows a typical Taylor rule, accommodating the economy in response to falling output and prices.
Theory and evidence suggest that in an environment of well-anchored expectations, temporary news or shocks to economic variables, should not have an effect on long-run inflation expectations. However, we show that they increase in response to policy uncertainty shocks. Furthermore, these shocks account for about 10 to 30 percent of their variation. This result is fairly robust when taking into account specific policy-related uncertainties, different measures of inflation expectations, different orderings of the variables in the structural VARs, and different periods covered by the sample. A rise of long-term inflation expectations at times of economic contraction suggests that heightened policy uncertainty indeed raises concerns about an increase in future inflation.
Overall, these results support our hypothesis that, in an environment of increased policy uncertainty, agents begin to question the ability and the commitment of policy makers to deliver on their promises. This result is of current relevance especially for central Our work contributes to the empirical literature on the macroeconomic effects of policy uncertainty. To our knowledge, our paper is the first to look at the effects of policyrelated uncertainty shocks on inflation expectations. Furthermore, we are the first to provide empirical evidence on the impact of different types of policy-related uncertainty on economic activity. We confirm previous findings that uncertainty shocks generate economic contractions, for the US and the euro area. Our finding that policy uncertainty shocks pose upward risks to the anchoring of long-term inflation expectations is new to the literature.

INTRODUCTION
Policy-related uncertainty is seen by academics and market participants as a prominent contributor to the overall economic uncertainty observed during the recent crisis. In the World Economic Outlook of October 2012, the IMF states: "The biggest factor weighing on the world economy was uncertainty among investors over whether policymakers in advanced economies will deliver on promises." These worries are grounded when taking into account a rapidly growing theoretical and empirical literature suggesting that uncertainty has recessionary effects on economic activity. 1 To date, the literature has focused on the effects of policy uncertainty 2 on economic outcomes, such as output, investment, consumption and unemployment (see among others Baker, Bloom, and Davis (2012) and Fernandez-Villaverde, Guerron-Quintana, Rubio-Ramirez, and Uribe (2011)).
However, less is known whether uncertainty caused by economic policy affects agents' expectations, and more specifically those about inflation. 3 In this paper, we investigate this question empirically, by studying the dynamic relationship between policy-related uncertainty and measures of inflation expectations of professional forecasters. Bearing in mind that inflation expectations contain important information on central bank credibility, a more general question is whether policy uncertainty undermines the credibility of policies and institutions. Academics and policy makers often state that the extent to which inflation expectations are anchored is the best measure of the credibility of monetary policy.
Our investigation comes at a time when central banks are still at the center stage to resolve the crisis. During the last years they resorted to standard and non-standard measures, aimed at providing support to the transmission mechanism of monetary policy. Some observers criticize them for this, some claim they go beyond their mandate and the general public shares the feeling they have not done enough to prevent the crisis. 4 Furthermore, regular surveys on public opinion and attitudes in Europe show a clear declining trend of trust and satisfaction with the way central banks have been doing their job (see Figure 9 in Appendix B). Given this situation, our hypothesis is that in the light of increased overall policy uncertainty (not necessarily only uncertainty about monetary 1 Theoretically, uncertainty is supposed to reduce hiring, investment and consumption of durables in presence of adjustment costs (Bernanke (1983), Dixit and Pindyck (1994) and Bentolila and Bertola (1990)), financial frictions (Arellano, Bai, andKehoe (2011), Gilchrist, Sim, andZakrajsek (2010) and Christiano, Motto, and Rostagno (2010)), managerial risk aversion (Panousi and Papanikolaou (2011)) and precautionary motives.
2 Policy uncertainty may be referred as policy risk as well. In this paper, we use the term policy uncertainty in line with the literature we follow. 3 There are examples of theoretical models with incomplete information where agents' beliefs and sentiments change in response to policy, see Eusepi and Preston (2010) and Bianchi and Melosi (2012) among others. 4 This is based on the results from the FT/Harris poll, conducted online among 6,237 adults in France, Germany, the UK, Spain, Italy and the US, April 2008, August 2008and February 2009 policy) agents begin to question the ability (expertise) of policy makers as well as their commitment to their promises (targets). With respect to monetary authorities, this means to question their credibility. Shedding light on this issue is of great importance, considering the role that credibility and reputation have for policy effectiveness. 5 We estimate structural Bayesian VARs, linking policy uncertainty with inflation expectations while accounting for a measure of economic activity and monetary policy, for the US and the euro area. We use the index of Baker, Bloom, and Davis (2012) as a measure of policy uncertainty. This index is supposed to capture uncertainty about what policy action the decision makers will undertake, uncertainty about the economic effects of current and future actions and/or inactions. This can be uncertainty about different economic policies altogether but in our estimations we provide evidence even for specific types of uncertainties related with fiscal and monetary policy. Regarding inflation expectations, we use short-and long-term survey-based inflation expectations of professional forecasters as measured by Consensus Economics and Survey of Professional Forecasters, of the Federal Reserve Bank of Philadelphia and of the European Central Bank.
We find several effects of policy uncertainty. First, economic activity contracts given a surprise innovation to policy uncertainty. Second, inflation expectations are also responsive: short-term ones fall, reflecting the slack economy while long-term inflation expectations rise. Third, monetary policy-related uncertainty is not always the important factor behind the dynamics of inflation expectations. Fiscal policy-related uncertainty seems to also play an important role. Furthermore, monetary policy appears to face a trade-off between responding to the state of the economy and to long-run inflation expectations. Given an unexpected innovation to policy uncertainty, central banks lower interest rates strongly, resembling the response of a central bank that follows a typical Taylor rule, accommodating the economy in response to falling output and prices.
Theory and evidence suggest that in an environment of well-anchored expectations, temporary news or shocks to economic variables, should not have an effect on long-run inflation expectations. However, we show that they increase in response to policy uncertainty shocks. Furthermore, these shocks account for about 10 to 30 percent of their variation. This result is fairly robust when taking into account specific policy-related uncertainties, different measures of inflation expectations, different orderings of the variables in the structural VARs, and different periods covered by the sample. A rise of long-term inflation expectations at times of economic contraction suggests that heightened policy uncertainty indeed raises concerns about an increase in future inflation. Overall, these results support our hypothesis that, in an environment of increased policy uncertainty, 5 The importance of reputation and credibility of central banks is well-recognized in theoretical works, starting with Kydland and Prescott (1977) and Barro and Gordon (1983). agents begin to question the ability and the commitment of policy makers to deliver on their promises. Our work is related to different strands of literature. First it relates to the literature on the macroeconomic effects of uncertainty in general (e.g. Bernanke (1983), Dixit and Pindyck (1994), Bentolila and Bertola (1990), Arellano, Bai, andKehoe (2011), Gilchrist, Sim, andZakrajsek (2010) and Christiano, Motto, and Rostagno (2010)) and of policy uncertainty in particular (e.g. Fernandez-Villaverde at al. (2011) and Born and Pfeifer (2011)). It also relates to the literature that studies changes in agents' beliefs in response to policy (e.g. Eusepi and Preston (2010) and Bianchi and Melosi (2012) among others).
We bridge these two strands of the empirical literature by highlighting a new channel through which policy uncertainty can affect the macroeconomy and providing empirical evidence for it. To our knowledge, our paper is the first to look at the effects of policyrelated uncertainty shocks on inflation expectations. Furthermore, we are the first to provide empirical evidence on the impact of different types of policy-related uncertainty on economic activity. We confirm previous findings that uncertainty shocks generate economic contractions, for the US and the euro area. Our finding that policy uncertainty shocks pose upward risks to the anchoring of long-term inflation expectations is new to the literature.
The structure of the paper is as follows. Section 2 presents the policy uncertainty measure and an overview of the recent developments in inflation expectations. Section 3 presents the empirical methodology (structural VAR estimations) and the discussion of results. Section 4 concludes.

ECONOMIC POLICY UNCERTAINTY AND INFLATION EXPECTATIONS: AN OVERVIEW
In this section we discuss how economic policy-related uncertainty and inflation expectations are measured. We also show their evolution throughout the years and point out main episodes associated with increases in policy uncertainty. In our estimations we use only the news-based 7 component of EPU for several reasons.
First, we want to avoid a potential link between the "disagreement" component of the index with the inflation expectations we use in our VAR estimation. They are based on expectations of professional forecasters, either from Consensus Economics or Fed's SPF. Furthermore, working with the news-coverage component allows us to distinguish between uncertainty coming from monetary, fiscal or labor market policies, for example.
In this way, we are able to study whether other kinds of policy uncertainty are affecting the beliefs of agents about inflation expectations and about the ability of central bankers to deliver on their mandates. However, the EPU index is a proxy variable and subject to measurement errors. For example, it is often questioned whether this index is just another measure of the state of the economy or whether it suffers from political slant. Certainly, policy uncertainty 6 For the US it has an additional component, the number of federal tax code provisions set to expire in future years. For more information, visit www.policyuncertainty.com. 7 Literature knows uses of narrative as variables: Romer and Romer (1989) and Romer and Romer (2004) to identify monetary policy shocks, Ramey and Shapiro (1998) and Ramey (2011) for fiscal policy shocks and Doms and Morin (2004) explore the linkages between media coverage of economic events, consumers' perceptions, and economic outcomes. is part of overall economic uncertainty and Baker, Bloom, and Davis (2012) show that at certain times, it is its main contributor. In response to potential measurement errors, they evaluate the index in several ways and argue that, although present to a small extent, these issues do not undermine the accuracy of the index. For example, they find a strong correlation between the computerized newspaper component 8 of the EPU index and a measure of what a human reader would call economic policy uncertainty. They also show that the EPU index is consistent with the frequency of the word "uncertain" in the FOMC Beige Book and with the responses of the stock market generated by policy news.
Moreover, the EPU index does not appear to be strongly affected by newspaper political slant. Especially in recent years, we have observed policies that have generated uncertainty about future inflation. For instance, there has been criticism towards the ECB potentially acting beyond its mandate through the bond-buying programs, first announced in 2010.
These programs raised concerns about the ECB being at risk of operating under fiscal dominance, thereby harming its independence. This, in turn, would lead to a difficulty for the ECB to ensure price stability. Policy uncertainty that might feed into expectations about future inflation has also arisen from the discussions about the exit strategies of the central banks that implemented quantitative easing. If not done carefully, exit from massive monetary stimulus could jeopardize future price stability. Moreover, uncertainty arising from fiscal pressures in the US, also raises concerns about the Fed being able to deliver price stability in the future. Therefore, it seems important to investigate whether in an environment of high policy-related uncertainty, these concerns have fed into agents perceptions regarding policy makers and their policies. 8 The newspaper component of the EPU index is based on automatic searches of specific terms related to economic uncertainty and policy in the largest newspapers for each country.  long-term BEIRs show a greater volatility throughout the whole sample (see Figure 10 in Appendix B). They are especially more responsive to news in the post-Lehman period, reflecting liquidity and risk premia concerns in financial markets. Several analyzes on the development of inflation expectations during the crisis show that long-term inflation expectations have become less firmly anchored, to a larger extent in the UK and in the US, relative to the euro area (see among others, Galati, Poelhekke, and Zhou (2011)).

POLICY UNCERTAINTY SHOCKS -A VAR ANALYSIS
In this section we study the effects of policy-related uncertainty shocks on inflation expectations using VAR techniques. We introduce the estimation methodology and the data used before discussing the results.
3.1. The model. To study our question of interest we employ VAR models that have the following standard representation: for t = 1, ..., T, where y t is a n × 1 vector of endogenous variables, z t is a m × 1 vector of exogenous variables, p is the number of lags and u t represents the reduced-form errors, Let k = np + m + 1 and define the k × n matrix of coefficients taking the following form: Θ = (A 0 , A 1 , ..., A p , B 0 ) . The VAR can be conveniently rewritten as a multivariate regression model in matrix notation: We have defined the T × n matrices Y and U and the T × k matrix X as: We employ Bayesian techniques for estimation. Our prior and posterior for (Θ, Σ) belong to the Normal-Whishart family, as in Uhlig (2005) 10 . Making use of all the definitions above, the likelihood function of (Θ, Σ) based on Y takes the following form: where we have used the Maximum Likelihood Estimators (MLE) of Θ and Σ, If the prior is described byΘ 0 , N 0 , S 0 and ν 0 , then the posterior is described byΘ T , N T , S T and ν T where: Following Uhlig (2005), we use a flat prior with N 0 = 0, ν 0 = 0, S 0 andΘ 0 being arbitrary. Then the posterior is characterized by N T = X X, ν T = T, S T =Σ andΘ T =Θ.
We estimate the above model (BVAR) for the US and the euro area. In addition to the individual BVARs, we also estimate a panel-VAR including both economies, in order to get more statistical power and to increase the precision of our estimates, given the relatively short data sample for the euro area (starting in 1999). This approach allows us to uncover common dynamic relationships for the US and the euro area while accounting for country-specific fixed effects.
The fixed-effects panel VAR is characterized by the same structure as the standard VAR presented above but the representation includes additionally a cross-sectional dimension. A panel VAR model with fixed effects and common slope coefficients can be written in the following form: for t = 1, ..., T, where i = 1, ..., I is the number of cross-sections, y i,t is a n × 1 vector of endogenous variables, z i,t is a m × 1 vector of exogenous variables, A 0i are unit specific intercepts that also include unit-fixed effects, and u i,t represents the reduced-form errors, 3.2. Data and Estimation. We estimate structural BVARs for the US and the euro area, with the following vector of endogenous variables: y t = (epu t , gdp t , π e long|t , π e short|t , i t ), with epu t being the news-based economic policy uncertainty, gdp t the real GDP, the π e long|t and π e short|t the long-and short-term inflation expectations, respectively and i t being the short-term interest rate. The same variables enter in the panel-BVAR as well.
The BVARs also include a constant and an exogenous variable, either the oil prices or the US industrial production, depending on the country-specific BVARs. This model allows us to study the impact of policy uncertainty on inflation expectations, while accounting for a measure of economic activity and monetary policy, in a parsimonious way.
As a proxy for policy uncertainty we use the corresponding country-specific, newsbased policy uncertainty index of Baker, Bloom, and Davis (2012). The overall policy uncertainty measure, epu t , incorporates uncertainty about different types of policy altogether, like fiscal, monetary, financial or any other type of regulatory policies. However, we are interested in studying the effects of uncertainty related with specific policies separately, as well. If the structural VAR estimations show that the overall policy uncertainty is significant for the dynamics of inflation expectations, being able to identify the specific policy responsible for these dynamics is important. On the other hand, specific-types of policy uncertainty could have a higher relevance for the dynamics of certain variables, even when the overall policy uncertainty does not. Our framework allows us to study these options.
In our analysis we also explore the effects of monetary and fiscal policy-related uncertainty, separately. Measures of monetary policy-and fiscal policy-related uncertainty are constructed by Baker, Bloom, and Davis (2012) and are currently available only for the US and Germany. 11 To our knowledge we are the first to use this novel data set in the empirical literature on the macroeconomic effects of uncertainty. In our estimations we use the measures for Germany as proxies for the euro area. One should keep in mind that these measures capture the policy uncertainty as discussed in the German media. This uncertainty is often related with important policies or developments outside the country as well. For the monetary policy-related uncertainty the approximation seems reasonable, given that there is a single monetary policy in the euro area. With respect to fiscal policy this match might appear weak at first sight. However, especially during the last years, fiscal issues across the euro area have been closely followed by the German public and heavily discussed in the media. Observing the evolution of the German index for fiscal policy uncertainty (Figure 8,  We use survey-based measures of inflation expectations from two sources, the Consensus Economics (CE) and the Survey of Professional Forecasters (SPF) of the Fed of Philadelphia and of the ECB, respectively. 12 These sources provide surveys of professional forecasters. In both, CE and SPF, the respondents are usually banks, universities, financial firms, consulting groups, and economic forecasters at large companies. Sometimes the respondents overlap between these two sources but in general the composition is different. Furthermore, the number of respondents in these surveys is different. CE surveys report the inflation expectations of about 240 respondents compared to about 90 respondents for the Fed's SPF and about 40 respondents for the ECB's SPF. In our estimations, the short-term inflation expectations refer to the expected inflation one year ahead, π e short|t = π e t+1|t . On the other hand, the long-term inflation expectations refer to the expected inflation five years ahead, π e long|t = π e t+5|t . Only in the case of the Fed's SPF, long-term inflation expectations refer to expectations over the next ten years. 13 That is, for the case of the US, the period for which long-term expectations are measured differs between the CE and the Fed's SPF, making the results not directly comparable.
Another option would be to also check for the effect of policy uncertainty shocks on the financial market-based inflation expectations, the so-called breakeven inflation rates (BEIRs) resulting from the difference between nominal Treasury bonds and Treasury inflation-protected securities. Although BEIRs are attractive because of their availability in high frequency, two issues arise. First, these rates do not reflect only inflation risks but also liquidity and risk premium concerns. Second, even though the literature offers methods to distinguish the inflation expectations component from the other two risks, there is still no consensus about the best way of doing this.
All BVARs are estimated at a quarterly frequency. Our variables of interest are available in different frequencies, monthly (policy uncertainty), quarterly (real GDP, SPF shortand long-term inflation expectations, CE short-term inflation expectations) and biannual (CE long-term inflation expectations). We use them all at quarterly frequency. 14 Biannual data are linearly interpolated to monthly frequency. Then, for all monthly series we use the end of quarter observation. The interpolation of the CE inflation expectations from semi-annual to quarterly frequency does not seem to be innocuous to our main results. In general, data aggregation or interpolation pose additional difficulties for the 12 More details on these data can be found in Appendix A. 13 Starting from 2005, Fed's SPF is also collecting the inflation expectations over the next five years. 14 We have estimated our BVARs in monthly and biannual frequency as well and main results are comparable. Results from these estimations are available upon request.
researcher that wants to identify structural shocks based on timing restrictions. The interpolation of CE long-term inflation expectations might hinder our identification strategy as a two-sided filter might destroy the temporal ordering. To check if our results are sensitive to this issue, we have also estimated our BVARs with CE long-term expectations at quarterly frequency constructed differently, with the value for the missing quarter being substituted with the value of the previous quarter.
Policy uncertainty and real GDP enter the estimation in log levels, and inflation expectations and interest rates in percent. For the panel-BVAR estimation, the period covered is 1999Q1-2012Q3, constrained by the availability of the data for the euro area. For US, the period covered is 1991Q4-2012Q3. We provide inference through the median response and its 68 percent posterior distribution, based on 2000 draws. 15 We also calculate the forecast error variance decomposition (FEVD) in order to asses the relative contribution of the policy uncertainty shock to fluctuations in our chosen variables. As described above, the VAR coefficients are drawn from a normal-inverse-Wishart distribution with flat prior. A flat prior allows us to use the benefits of the Bayesian techniques while having our results more data-driven, making them easily comparable with those in the related literature that do not use Bayesian methods for estimation. The optimal lag is selected based on the BIC information criteria and reported below each figure of results. Under our identification strategy, we relax the exogeneity restriction on policy uncertainty and allow it to respond with delay to other shocks through the lag polynomial.
Policy uncertainty, as measured by the EPU index, could arise not only from unexpected innovations to policy but also as a response to other shocks in the economy. For example, 15 The estimation methodology builds on Uhlig (2005). Calculation of posterior distributions are made following Reppa (2009). 16 Bachmann, Elstner, and Sims (2013) have used in addition long-term restriction to identify uncertainty shocks. a contractionary shock hitting the economy could also lead to a rise in policy uncertainty if the public does not know how policy will respond. In the benchmark BVAR we assume that uncertainty about policy is affected only with delay to such shocks. Given these assumptions, the relationship between the reduced-form errors, u t and structural shocks, t , is presented as below.
In our estimations, a policy uncertainty shock corresponds to an increase of two stan- In Figure 3 we observe that in response to a two standard deviation innovation in the EPU measure, the real GDP contracts, short-term inflation expectations fall following 17 We show the complete set of responses of all our variables to policy uncertainty shocks in Appendix D.
slack economy and interest rates decrease considerably in response to weak economy.
On the other hand, long-term inflation expectations rise. This response peaks around the third quarter (about 5 basis points) and dies out in two years. Similar patterns are observed even when looking at the responses of inflation expectations along our different dimensions. In all panels of Figure 4, an innovation in the respective measure of policy uncertainty induces an increase in the median response of long-term inflation expectations. This response peaks in about three quarters irrespective of the source, SPF or CE, although in magnitude the peak is higher for CE expectations. That is, for the SPF expectations, the responses are slightly muted when compared with those from the CE, but one should bear in mind that for the case of US, the SPF long-term refers to inflation expectations over a longer period (next 10 years). Along the shock dimension, the rise of long-term inflation expectations appears slightly stronger given a monetary policy-related uncertainty shock. On the other hand, the response of short-term inflation expectations to policy uncertainty shocks remains on the negative side. When compared with the response of longterm inflation expectations, they show a higher degree of responsiveness (up to 15 basis points) and volatility. One year ahead inflation expectations of professional forecasters appear to be more responsive to overall policy uncertainty and to fiscal policy uncertainty. As before, expectations from CE respondents appear to react stronger in magnitudes than the SPF ones.   around the zero line. In the case of the SPF expectations the reversion is slower and smoother. The immediate response of the SPF short-term inflation expectations to policy uncertainty shocks is negative. Differently, the CE short-term inflation expectations increase in response to such shocks. However, the increase of the median is short-lived, lasting only about two to three quarters. The response is slightly sharper given a monetary policy uncertainty shock.
With respect to other variables, we observe that in all estimations a policy uncertainty shock is associated with an economic contraction. The GDP contraction appears on impact and up to 1 percent in the case of the US, and delayed and muted for the euro area (see Appendix D, Figures 11 to 16). For US, the real GDP declines for about three quarters and the recovery phase lasts up to ten quarters. Specific policy uncertainty shocks produce qualitatively comparable responses for the real GDP. On the other hand, central banks in both economies respond with lowering interest rates strongly given a positive innovation to all types of policy uncertainty measures that we consider. If we take into account that short-term inflation expectations are highly correlated with actual inflation (about 60 to 70 percent in our sample), then this move resembles the response of a central bank that follows a typical Taylor rule, accommodating the economy in response to falling output and prices.
The GDP decline, immediate or not, and its relatively quick reversal seem to be in line with previous findings in both the theoretical and the empirical literature on the macroeconomic effects of uncertainty shocks. The magnitudes are also comparable. 18 The empirical finding on the effect of specific types of policy uncertainty is new to the literature.
We find that monetary-and fiscal policy-related uncertainty are equally harmful to economic growth. Different channels through which policy uncertainty affects economic activity could be at work, such as the precautionary saving motive or the "wait and see" 18 The majority of previous studies in the literature have considered the response to a two standard deviation innovation of such shocks, to approximate the level of uncertainty that was observed during the recent crisis. dynamics, the former affecting negatively aggregate consumption and the latter affecting investment. 19 With regard to long-term inflation expectations, theory and evidence suggest that in an environment of well-anchored expectations, temporary news or shocks to economic variables, should not have an effect on them. However, they appear sensitive to policy uncertainty shocks in our BVAR analysis. We observe that long-term inflation expectations increase. The magnitude does not seem high at first sight. However taking into account that uncertainty shocks in our sample have been up to 4 standard deviations, that induces a high enough rise on long term inflation expectations (about 10 basis point) to miss the ECB's "below, but close to, 2%" or Fed's 2% inflation objective.
In addition, the table below shows the forecast variance error decomposition of longterm inflation expectations to policy uncertainty shocks. We observe that such shocks account for about 10 to 30 percent of the variation of long-term inflation expectations. This contribution is not negligible having in mind the small variation of long-term inflation expectations in our sample and that the majority of it is explained by its own shocks. 20  The idea behind this concept is that in the presence of high uncertainty and adjustment frictions, firms pause hiring and investment, and wait for calmer periods to expand. Under these conditions, production falls but pick-ups quickly due to pent-up demand for production factors (Bernanke (1983), Dixit and Pindyck (1994), Bloom (2009) and Bloom, Floetotto, Jaimovich, Saporta-Eksten, and Terry (2012)). 20 As parallelism, the VAR literature on effects of monetary policy shocks has estimated not more than 10 and 20 percent contribution of monetary shocks on FEVD of output and prices, respectively. periods during which the emphasis is mainly on output stabilization (passive regime).
Deviations from low inflation policies are not penalized immediately because agents are "optimistic" that the deviation is short lasting. Once there is uncertainty about the duration of the passive regime, inflation expectations rise.
Moreover, the opposite directions of the responses of short-and long-term inflation expectations to a policy uncertainty shock provide us with further evidence on low probability events (i.e. policy regime switches) being taken into account when forming expectations. For example, agents might believe that there is a likelihood of switching to a high inflation regime, hence long-term inflation expectations rise. But because this regime has very low probability of occurring it is unlikely that we observe it in our data sample.
Therefore, short-term inflation expectations expectations do not rise on medium-term.
Overall, we argue that even though the commitment of central banks to a stable and low inflation has not changed, agents seem to perceive that it would be more difficult for central banks to achieve their targets. Such a scenario is likely when taking into account the unprecedented policies monetary authorities took in response to the recent crisis and the problems arising from large fiscal deficits; if they are not well-managed they risk fueling inflationary pressures.
3.5. Robustness checks. In the following we examine whether our main results are sensitive to issues related with the identification of the policy uncertainty shocks and to the sample selected for analysis. First we investigate if the results are sensitive to the ordering of the variables in the BVAR. As discussed above, the identification of uncertainty shocks is recent and a consensus on the best identifying restrictions is yet to be reached in the literature. In our benchmark identification, policy uncertainty responds immediately only to its innovations and with a delay to other shocks. Here we relax this assumption and the order of the variables in the BVAR is: gdp t , π e long|t , π e short|t , i t , epu t . Under this ordering, policy uncertainty is allowed to be affected immediately by all the shocks. In the first column of Figure 7 we show the results from the panel-BVAR with this alternative ordering.
Furthermore, in our benchmark estimations we include short-term interest rates to account for conventional monetary policy. However, starting from the fourth quarter of 2008 both the Fed and the ECB turned to unconventional policies. To account for this, we include a dummy variable in our BVARs that takes the value of one starting from the fourth quarter of 2008. We present the results related with this robustness check for the panel-BVAR in the second column of Figure 7. We also check if our main results hold when the period corresponding to the recent crisis is excluded. To this aim, we estimate our BVARs with the data sample ending in the fourth quarter of 2006. The After performing the sensitivity checks described above, we observe that our result from the benchmark analysis holds: innovations to policy uncertainty induce an increase in the median response of long-term inflation expectations. With respect to short-term inflation expectations, we observe that their response is sensitive to the sample period taken into account. When the recent crisis is excluded or when the period of unconventional policies is accounted for (by including the dummy variable), short-term inflation expectations increase in response to policy uncertainty shocks. However, they still fall given a fiscal policy uncertainty shock. With respect to the real GDP and interest rates, we generally observe the same qualitative results as in the benchmark analysis.

CONCLUDING REMARKS
We bridge the existing empirical literature on the macroeconomic effects of uncertainty shocks and the empirical literature on inflation expectations and provide first evidence on the effects of policy-related uncertainty, coming from extraordinary events and actions of decision makers, on beliefs and perceptions of agents towards them. Previous studies have shown that policy uncertainty is harmful for investments, consumption and employment. In this paper we show that the observed uncertainty about the stance and perceived effectiveness of policy should be troubling for central bankers 21 as it seems to entail additional risks to their hard-won inflation credibility.
Using BVARs we investigate whether policy-related uncertainty, as quantified by Baker, Bloom, and Davis (2012), has fed into inflation expectations in the US and the euro area.
We find that while economic activity contracts, long-term inflation expectations rise in response to a policy-related uncertainty shock. This finding is robust across the two countries, different specifications, more specific measures for policy uncertainty, different measures of inflation expectations, and different orderings of the variables in the BVARs.
Our results show that long-term inflation expectations of professional forecasters are not perfectly anchored and that policy-related uncertainty poses upside risks to them.
Given that well-anchored long-term inflation expectations reflect the credibility of monetary policy, we find support for the hypothesis that, in an environment of increased policy uncertainty, agents begin to question the ability and the commitment of policy makers to deliver on their promises. This result is of current relevance especially for cen- In this paper we study the expectations of professional forecasters, which are some of the most informed agents in the economy. However, it would also be interesting to study how general public's expectations and perceptions towards central banks are affected by policy uncertainty. Examining this issue is of interest for several reasons: the general public constitutes a large proportion of the agents in the economy, they are generally less financially literate than the professional forecasters, and they are more likely to be 21 Governor of Bank of Canada, Mark Carney, made such a remark on policy uncertainty in his speech " Uncertainty and Global Recovery" in October 2012, at Vancouver Island Economic Summit. influenced by the media. Particularly, one could study the role of policy uncertainty for the dynamics of trust in the ECB and of satisfaction with the way BoE is doing its job to preserve price stability, given the negative trends observed in these measures during the last years. We plan to investigate these issues in future works. APPENDIX A. DATA  'tax', 'spending', 'regulation', 'central bank', 'budget', and 'deficit'. Baker, Davis and Bloom (2012), data as in December 2012 Monthly

Monetary Policy Uncertainty
The number of news articles containing the terms 'uncertain' or 'uncertainty', 'economic' or 'economy', as

FIGURE 9. General public opinion towards central banks in Europe
Notes: Net trust in the ECB is defined as the difference between the share of respondents who state they tend to trust and the share of respondents who state they tend to not trust the ECB. Net satisfaction with Bank of England (BoE) is the share of satisfied minus the dissatisfied respondents when asked to assess the way the BoE is doing its job to set interest rates to control inflation. Sources: Standard Eurobarometer Survey, BoE.

FIGURE 10. Recent developments of inflation and long-term market-based inflation expectations
Notes: Actual inflation (HICP) and long-term (5 years ahead) Break Even Inflation Rates (BEIR). Source: Bloomberg .

APPENDIX C. PRIORS
As in Uhlig (2005), a proper Normal-Whishart distribution is parametrized by a mean coefficient matrixΘ of size k × n, a positive definite mean covariance matrix S of size n × n, a positive definite matrix N of size k × k and a degrees of freedom real number ν ≥ 0 to describe the uncertainty about (Θ, Σ) around (Θ, S). The Normal-Whishart distribution specifies that Σ −1 follows a Whishart distribution W n (S −1 /ν, ν) with E[Σ −1 ] = S −1 , and that, conditionally on Σ, the columnwise vectorized form of the coefficient matrix , vec(Θ), follows a Normal distribution N(vec(Θ), Σ ⊕ N −1 ). To draw from the the Whishart distribution W n (S −1 /ν, ν), a method is to calculate Σ = (RR ) −1 , where R is a matrix of size n × ν with each column an independent draw from the Normal distribution N(0, S −1 /ν).
If the prior is described byΘ 0 , N 0 , S 0 and ν 0 , then the posterior is described byΘ T , N T , S T and ν T where Using a weak prior with N 0 = 0, ν 0 = 0, S 0 andΘ 0 being arbitrary, then the posterior is characterized by N T = X X, ν T = T, S T =Σ andΘ T =Θ. Notes: The solid line in black denotes median impulse response from the panel-BVAR(2), with y t = (pu t , gdp t , ie long|t , ie short|t , i t ), a constant and log level of crude oil prices as an exogenous variable. The shaded area corresponds to the 68 percent credible set (posterior distribution). Policy uncertainty and GDP are in log levels. IE Long for the US corresponds to ten years ahead inflation expectations while for the euro area it represents five years ahead inflation expectations. In both cases, IE Short represents one year ahead inflation expectations, in percent. The period considered in the estimation: 1999Q1-2012Q3. Horizontal axis is lag horizon in quarters. We show the results of three robustness checks: (1) reordering the endogenous variables, y t = (gdp t , ie long|t , ie short|t , i t , pu t ); (2) including a dummy that takes the value of one starting from 2008Q4; (3) splitting the sample until 2006Q4.