BIS Working Papers
Do SVARs with sign restrictions not identify unconventional monetary policy shocks?
by Jef Boeckx, Maarten Dossche, Alessandro Galesi, Boris Hofmann and Gert Peersman
Monetary and Economic Department
JEL classification: D22, D84, E31
Keywords: inflation expectations, firms' survey, new information.
This publication is available on the BIS website (www.bis.org).
© Bank for International Settlements 2019. All rights reserved. Brief excerpts may be reproduced or translated provided the source is stated.
ISSN 1020-0959 (print)
ISSN 1682-7678 (online)
Do SVARs with sign restrictions not identify unconventional monetary policy shocks?
by Jef Boeckx, Maarten Dossche, Alessandro Galesi, Boris Hofmann and Gert Peersman
A growing empirical literature has shown, based on structural vector autoregressions (SVARs) identified through sign restrictions, that unconventional monetary policies implemented after the outbreak of the Great Financial Crisis (GFC) had expansionary macroeconomic effects. In a recent paper, Elbourne and Ji (2019) conclude that these studies fail to identify true unconventional monetary policy shocks in the euro area. In this note, we show that their findings are actually fully consistent with a successful identification of unconventional monetary policy shocks by the earlier studies and that their approach does not serve the purpose of evaluating identification strategies of SVARs.
JEL classification: D22, D84, E31
Keywords: inflation expectations, firms' survey, new information.
Assessing the macroeconomic effects of unconventional monetary policies implemented by the major central banks in the wake of the Great Financial Crisis (GFC) represents a formidable challenge for empirical research. Unlike conventional monetary policy implemented by adjusting a short-term interest rate, unconventional monetary policies were implemented through changes in the size and composition of the central bank’s balance sheet and targeted risk premia in money and capital markets. Identifying the effects of these policies on output and inflation required novel identification schemes to be devised.
Several studies have estimated the effects of unconventional monetary policies using structural vector autoregressive (SVAR) models that are identified with sign restrictions, often in combination with zero restrictions (e.g. Gambacorta et al. (2014); Weale and Wieladek (2016); Haldane et al. (2016); Panizza and Wyplosz (2016); Boeckx et al. (2017); Burriel and Galesi (2018); Hesse et al. (2018); Feldkircher et al. (2019); Lewis and Roth (2019)). In general, these studies find that policy measures which expand the balance sheet of the central bank had a significant positive impact on economic activity and consumer prices. However, in a recent study by the Netherlands Bureau for Economic Policy Analysis that has received considerable media attention, Elbourne and Ji (2019) challenge these findings for the euro area. They conclude, based on empirical analysis that revisits the findings for the euro area of Boeckx et al. (2017) and Burriel and Galesi (2018), that "these findings result from an identification strategy that does not successfully uncover unconventional monetary policy shocks".
The conclusion of Elbourne and Ji (2019) is based on three observations:
Based on these findings, Elbourne and Ji (2019) conclude: "Not only is it highly implausible that the identification scheme of Gambacorta et al. (2014), Boeckx et al. (2017) and Burriel and Galesi (2018) recovers the true unconventional monetary policy shocks without the information contained in the size of the ECB's balance sheet, the resultant shocks bear no resemblance to other credibly identified monetary policy shocks in this period. As such, the logical conclusion is that the SVAR models are not identifying unconventional monetary policy shocks."
This conclusion has potentially important implications as it casts doubt on benchmark empirical findings which suggested that unconventional monetary policy was effective in stimulating the economy. In this paper, we revisit the analysis of Elbourne and Ji (2019) in order to better understand what drives their findings and whether they really imply that previous empirical analysis on the subject is seriously flawed.
In Section 2 we reproduce their results using the model of Boeckx et al. (2017), which is the benchmark study for the euro area that uses SVARs identified with a combination of zero and sign restrictions to estimate the effects of unconventional monetary policies, in particular credit support policies that were introduced by the ECB to stimulate bank lending in the period after the GFC. In doing so, we revisit key aspects of the analysis in Elbourne and Ji (2019), specifically the choice of the sample period and model specification, which differ from the targeted benchmark studies, and the lack of statistical tests used to derive conclusions. In addition, several impulse responses of variables included in the VAR are not reported in their study.
Our analysis yields the following main findings (in chronological order of the three observations as listed before):
Overall, the approach and findings of Elbourne and Ji (2019) are not informative about the successfulness of identification strategies and do not support the notion that studies evaluating the macroeconomic effects of central bank unconventional monetary policy based on VARs identified through zero and sign restrictions are flawed. As an alternative, in Section 3, we evaluate whether the shocks identified through the model of Boeckx et al. (2017) are consistent with the narrative of the unconventional monetary policy measures launched by the ECB since 2007. The time series of shocks suggests that the expansionary measures are indeed reflected in expansionary unconventional monetary policy shocks identified through the SVAR, while periods of more restrictive balance sheet policies were associated with contractionary shocks. Even though we are well aware that any identification strategy has its drawbacks and needs to be interpreted carefully, the fact that the shocks capture the most important non-standard monetary policy measures during the sample period indicates that the identification strategy is plausible.
In order to assess whether SVAR models are able to plausibly recover unconventional monetary policy shocks in the euro area, Elbourne and Ji (2019) reproduce some of the existing studies for the euro area; that is, Gambacorta et al. (2014), Boeckx et al. (2017) and Burriel and Galesi Nonetheless, there appear to be some important differences, which we discuss in the first subsection. In the following, we reproduce the model of Boeckx et al. (2017), which is the benchmark study for the euro area. Using this model, we replicate the analysis of Elbourne and Ji and discuss their approach.
1. The VAR model of Boeckx et al. (2017)
Boeckx et al. (2017) was the first paper that used the identification strategy based on zero and sign restrictions to estimate the effects of the ECB's balance sheet policies on the euro area economy in the aftermath of the financial crisis. The approach in Boeckx et al. (2017) elaborates on Gambacorta et al. (2014), who have estimated a panel VAR for eight industrialized countries. Burriel and Galesi (2018), in turn, extend the analysis of Boeckx et al. (2017) to individual euro area countries by estimating a global VAR that takes into account cross-country interdependencies among euro area members. We therefore base the analysis in this note on Boeckx et al. (2017). Notice, however, that the results reported in the following also apply to these other studies.
Boeckx et al. (2017) estimate a monthly SVAR model for the euro area containing output (real GDP interpolated using industrial production), consumer prices (HICP), the ECB’s main refinancing operations (MRO) rate, the size of the balance sheet of the ECB (total assets), an indicator of financial stress (CISS-indicator of Hollo et al. (2012)), and the spread between the overnight (EONIA) and the MRO rate. These variables are shown in Table 1, which also includes the restrictions that have been imposed on the SVAR in order to disentangle exogenous balance sheet shocks (for a given policy rate) that are caused by unconventional monetary policy measures of the ECB (UMP shock) from other economic shocks. For a discussion of these policies, we refer to Boeckx et al. (2017).
First, there is no immediate impact of the shock on output and consumer prices. This assumption, which is plausible for monthly estimations, allows to disentangle the UMP shocks from real economy disturbances such as aggregate supply and demand shocks.
Second, UMP shocks that increase the balance sheet of the ECB do not increase financial stress (CISS-indicator). This restriction disentangles such innovations from the endogenous response of the balance sheet to financial stress. It follows as a complementary restriction from the assumption that central bank assets typically increase in response to a rise in the CISS-indicator. The latter reflects the idea that (i) the ECB may react to increased financial stress by expanding its balance sheet, and (ii) due to the fixed interest rate with full allotment (FRFA) policy during the sample period, the balance sheet of the ECB could rise endogenously when financial market uncertainty increases.
Third, an expansionary UMP shock does not increase the EONIA-MRO spread. The motivation of this restriction is that an expansionary balance sheet shock that is the consequence of an unconventional monetary policy action typically increases the liquidity surplus, which exerts downward pressure on the EONIA and thus the spread with the policy rate. The decline in the spread also allows to disentangle supply-driven changes in the ECB balance sheet from changes in the demand for central bank liquidity.
Fourth, since the aim is to estimate the effects of innovations to the ECB's balance sheet that are orthogonal to shifts in the policy rate, the identified shocks have a zero contemporaneous impact on the MRO-rate. This restriction guarantees that the identified balance sheet shocks do not capture conventional interest rate shocks. Elbourne and Ji (2019) do not exactly reproduce the Boeckx et al. (2017) model as they exclude the MRO-rate and its restriction. If anything, this would imply that unconventional and conventional policy shocks cannot be disentangled.
Finally, all sign restrictions are imposed on impact and the first month after the shock and implemented in a weak form; that is, as smaller/larger than or equal to zero. This allows for the possibility that an unconventional monetary policy measure, for example, influences the CISS- indicator immediately but central bank assets only with a lag. Hence, it accommodates for the fact that some measures were announced (1-2 months) before they were implemented. On the other hand, this implies that balance sheet policies that were announced more than 1-2 months in advance are captured by the other innovations in the VAR model.
By imposing these restrictions, it is possible to isolate unconventional monetary policy measures that have increased or decreased the balance sheet of the ECB in the aftermath of the financial crisis, and estimate their macroeconomic consequences. The sample period for the analysis is 2007M1-2014M12. In particular, the identification scheme is not valid for the period covering the ECB’s Expanded Asset Purchase Program (EAPP) since January 2015 because the volumes of the purchases under this program were announced several months in advance (and were thus not unanticipated), which can be considered as a structural break in the balance sheet reaction function. The choice of the sample period and its importance for the validity of the analysis is explained in detail in Boeckx et al. (2017).
2. Sign switch of the balance sheet restriction
We start by re-estimating the SVAR of Boeckx et al. (2017), switching the sign restriction that is imposed on the balance sheet (ECB total assets) such that an expansionary unconventional monetary policy shock reduces the size of the balance sheet. This exercise corresponds to Shock B in Table 1. Specifically, all restrictions are the same as the original identification scheme, except the one imposed on ECB Total Assets. Based on an inspection of the impulse responses of output and prices, Elbourne and Ji (2019) conclude that they find "statistically indistinguishable impulse responses and time series of supposed unconventional monetary policy shocks". The impulse responses of the other variables that are included in the SVAR are, however, not reported nor discussed in the paper. In addition, they do not conduct any formal test to come to this conclusion; that is, this conclusion is based on the fact that the median impulse response functions of Shock B fall within the confidence intervals of the responses of the UMP Shock.
The impulse responses to Shock B are shown in the right column of Figure 1, while the left column shows the impulse responses to the original unconventional monetary policy shock (UMP shock). Consistent with Elbourne and Ji (2019), we find that there is a rise in output and consumer prices, while the confidence intervals overlap. This finding in itself is, however, not surprising since a shock that is associated with a reduction in the stress indicator, which captures for example a reduction in economic uncertainty and an improvement of macroeconomic conditions, and a decline in the EONIA can be considered as an expansionary aggregate demand shock. More generally, there are numerous types of (demand) shocks that move output and prices in the same direction. It would be surprising if two (different) shocks originating in financial markets and being both associated with improved financial market conditions and a decline in the EONIA would have an impact on output and prices that is considerably different.
Notwithstanding the overlapping characteristics of both shocks, the results reveal that there are relevant differences between the dynamic effects. For example, the peak impact of the (original) UMP Shock on output is a couple of months later. The output effects are also much stronger for Shock B; that is, the (median) peak effect of the UMP Shock is 0.096%, compared to 0.155% for Shock B. Relatedly, it appears that the relative effects of both shocks on output versus prices are very different. This is illustrated in the third row of Figure 1, which shows the impact of the shocks on the output-price ratio based on the VAR-model. Specifically, the output response of UMP Shocks is significantly smaller than the price response the first month after the shock, while the impact on output is significantly larger than the price response during roughly one year after Shock B. Clearly, these differences are economically important and policy relevant.
Another remarkable difference is the impact on the EONIA-MRO spread, shown in Part II of Figure 1. After the UMP Shock, the spread gradually returns to its baseline, which is reached after six months. In contrast, despite the negative restriction on impact and one month after the shock, the spread immediately returns to the baseline after Shock B, in order to become significantly positive during more than one year. This pattern suggests that the (negative) sign restriction on the balance sheet (central bank liquidity) and the spread in the same direction to identify an expansionary policy shock is not supported by the data. This is not surprising since the spread reflects the presence of excess or shortage of liquidity in the overnight market for central bank liquidity.
Are the differences between the impulse responses statistically significant? An overlap of the error bands does not imply that the differences are statistically insignificant. To address this issue, we re-estimate the SVAR model and identify an UMP Shock and Shock B simultaneously. By nesting the estimation of both shocks within the same VAR-model, it is possible to calculate the differences between the impulse responses and the confidence bands of the differences. Furthermore, to account for the fact that the variance of both shocks may be different (a one- standard-deviation UMP Shock is not the same as a one-standard-deviation Shock B), we normalize both shocks based on their contemporaneous impact on the CISS-indicator; that is, the
impulse responses (and differences) correspond to a 1bps decline in the CISS-indicator induced by the shock.
Note: impulse responses with 16th and 84th percentiles error bands. UMP Shock is original shock of Boeckx et al. (2017),
The differences between the impulse responses obtained from this estimation are shown in Figure 2, together with confidence intervals. As can be observed in the figure, several impulse responses are statistically different from each other. For example, the effects on economic activity and output-price ratio are significantly stronger for Shock B compared to the UMP Shock (supported by more than 95% of the posterior draws), while also the impact on the EONIA-MRO spread is significantly different.
From these results, we can conclude that the sign restriction that is imposed on the balance sheet of the ECB is important for the estimation results. In other words, the balance sheet does contain relevant information for the estimation of the consequences of unconventional monetary policy shocks.
Notice that an implicit assumption of this exercise is that both shocks exist in reality. In particular, if Shock B does not exist or is not properly identified, the identification of the UMP Shock is distorted by identifying both shocks simultaneously. However, there are good reasons to believe that Shock B does not exist in reality. In particular, the ECB did react to financial stress shocks by lowering the interest rate during the sample period (e.g. in response to the collapse of Lehmann). Hence, the zero restriction on the MRO-rate is inconsistent with financial stress shocks. More generally, assuming that the ECB does not immediately react to shocks that have a significant impact on output and inflation is controversial (unless it is a monetary policy shock). Hence, it is more likely that the shocks captured by Shock B are a linear combination of several fundamental shock that are forced by the algorithm into a single series of shocks (i.e. shocks that offset each other in order to fulfill the sign and zero restrictions, which may even include monetary policy shocks).
The key result of Elbourne and Ji (2019) is that replacing the size of the balance sheet with a time series of independent random numbers produces very similar impulse responses to those reported in the original studies. In Figure 3, we reproduce this exercise for the SVAR-model of Boeckx et al. (2017). While the original results are again shown in the left column, the impulse responses obtained from the VAR-model with random numbers are shown in the right column. These are indeed very similar.
This finding is, however, not surprising. More specifically, like in any regression analysis, including random numbers as a variable in a VAR-system does not affect any coefficient of the system; that is, the coefficient of a series of random numbers that is included in a linear regression is equal to zero, nor does it affect the estimated coefficients of the other variables in the equations. As a consequence, including a series of random numbers in a regression analysis is pointless and the same as estimating the model without the series of random numbers (besides the fact that it adds some noise to the regressions).
This exercise is thus the same as estimating the VAR without the balance sheet and without the sign restriction on the balance sheet, which corresponds to the identification of Shock C in Table 1. In essence, the balance sheet becomes an unobservable variable in the system, which still contributes to the data generating process of all the other variables in the VAR. As a consequence, Shock C is a linear combination (weighted average) of the original UMP Shock and a shock that has the same effects except that it decreases instead of increases the central bank balance sheet (Shock B). This can be observed in Table 1: the UMP Shock and Shock B both fulfill all the (sign) restrictions of Shock C, while no other shock does. More specifically, Shock C can be represented as follows:
What determines the size of the weight a? This can easily be explained by the estimation algorithm. More specifically, when sign restrictions are used to identify the shocks, the algorithm draws a large number (e.g. 1,000,000) of possible series of shocks and impulse responses that are consistent with the data. If a draw satisfies the restrictions, the draw is kept. Otherwise, the draw is rejected by giving it a zero prior weight. The draws that are kept are then used to report (the distribution of the) impulse responses to the shocks. Since both the UMP Shock and Shock B fulfill all the restrictions of Shock C, both shocks will be accepted by the algorithm, and a will be the percentage of draws of UMP Shocks and 1- a the percentage of Shock B accepted by the algorithm.
When we return to the original estimations of Boeckx et al. (2017), it turns out that for all draws that fulfill the sign restrictions of Shock C (i.e. all sign restrictions except the one on the balance sheet), 67% have a positive impact on the balance sheet (i.e. are UMP Shocks) and 33% have a negative effect on the balance sheet (i.e. are ShockB). Put differently, a=0.67, and the correlation between Shock C and the UMP Shock should be approximately 0.67. Thus, the correlation of 0.61 reported by Elbourne and Ji (2019) is perfectly normal, and a mechanical consequence of the number of draws that fulfill the sign restriction of Shock B and the UMP Shock, respectively. Obviously, this also implies that the impulse responses of this exercise will be much closer to the UMP Shock than Shock B: 67% of the draws to construct the impulse responses are UMP Shocks, while only 33% of the draws are Shock B.
Overall, the relative strong correlation is silent about the successfulness of the identification scheme, it only reflects the much higher number of draws (i.e. double) that are consistent with an increase of the balance sheet; that is, these draws are much more supported by the data. In fact, this finding is not a surprise since Shock B does probably not exist in reality, as we have argued above.
Finally, Elbourne and Ji (2019) document that the shocks that are identified based on SVAR models with a combination of zero and sign restrictions are uncorrelated with alternative credible monetary policy shocks identified from futures rate surprises by Jarocinski and Karadi (2018) and Corsetti etal. (2018). More specifically, they obtain a correlation of -0.06 and 0.10, respectively, and conclude based on this observation: "Not only is it highly implausible that the identification scheme of Gambacorta et al. (2014), Boeckx et al. (2017) and Burriel and Galesi (2018) recovers the true unconventional monetary policy shocks without the information contained in the size of the ECB's balance sheet, the resultant shocks bear no resemblance to other credibly identified monetary policy shocks in this period. As such, the logical conclusion is that the SVAR models are not identifying unconventional monetary policy shocks."
The low correlation between both shock series is, however, not surprising since both shocks are fundamentally different. More specifically, the unconventional monetary policy shocks that are identified in Gambacorta et al. (2014), Boeckx et al. (2017) and Burriel and Galesi (2018) are unconventional monetary policy measures that expand the central bank balance sheets and are orthogonal to changes in the policy rate (conventional monetary policy). The latter is reflected in the zero restriction on the policy rate. Overall, the shocks reflect liquidity measures for a given policy rate to influence impaired financial markets and support bank lending. These policy measures are typically transmitted to the real economy via interest rate spreads and risk premia. For example, Boeckx et al. (2017) find that expansionary shocks lead to a decline in the spread between Euribor and Overnight Index Swap (OIS) rate and between peripheral and German sovereign bond yields, while bank lending rates decline. There is even a rise in (risk-free) German sovereign bond yields on impact and the policy (MRO) rate after a couple of months.
The studies of Jarocinski and Karadi (2018) and Corsetti et al. (2018), in contrast, identify shortterm (risk-free) interest rate surprises. More specifically, both studies first construct an instrumental variable using high-frequency surprises around policy announcements. Corsetti et al. (2018), for example, construct an instrument series based on changes in the 1-year EONIA swap rate (i.e. the OIS rate) around a narrow window of policy announcements. The underlying idea is that any surprises occurring within the window are most likely the consequence of monetary policy shocks. In particular, while expectations about economic activity and future policy rate changes are already priced in, unexpected changes in the policy rate should shift the swap instantly when the announcement is made. Similarly, any policy action that affects expectations about future rate movements should have an immediate impact on the swap.
Is it possible that the instrument of Jarocinski and Karadi (2018) and Corsetti et al. (2018) captures a mixture of interest rate and unconventional monetary policy announcements? Even though the ECB also announced several credit support measures at its press conferences in the period after the financial crisis, there is no reason why the latter should affect the 1-year EONIA swap rate, which is the average of the overnight rates for the subsequent year. This can be observed in Figure 4, which shows the impact of an expansionary unconventional monetary policy shock on the EONIA for the Boeckx et al. (2017) VAR-model. While the EONIA decreases on impact (as a result of the sign restriction on the spread), the impact on the EONIA very quickly becomes positive due to the endogenous rise in the policy rate. Hence, there is no reason to expect a (meaningful) decrease in the 1-year or 3-month swap rates in response to the shocks that are identified in Boeckx et al. (2017). Put differently, unexpected changes in the swap rates are not valid instruments for ECB balance sheet shocks. The shocks are fundamentally different, which is fully consistent with the (near) zero correlation between the series. In fact, a high correlation would be problematic for the identification strategy.
Finally, notice that even a low correlation between the same type of shocks does not tell whether an identification scheme has been successful or not, or whether the entire methodology is flawed. Sims (1998), in his comment on Rudebusch (1998), explained that the low correlation of (conventional) monetary policy shocks identified through different VARs at that time did not mean that any of the models was not successful or that VARs do not successfully identify monetary policy shocks more generally. Instead, Sims (1998) showed that it reflected differences in the specification of models. Furthermore, the series of policy surprises constructed by Jarocinski and Karadi (2018) and Corsetti et al. (2018) are only an instrumental variable for a monetary policy shock while it is well-known that the correlation between an instrumental variable and the underlying shock has been shown to be very often quite low (see e.g. Montiel Olea et al. (2016) and Peersman (2018)). In contrast, Boeckx et al. (2017) directly identify the UMP shock. A proper comparison would therefore need to be based on the monetary policy shock series, rather than on the instruments from one study and the shocks from another.
Do SVARs identify unconventional monetary policy shocks in the euro area? A narrative perspective
The analysis of the previous section shows that the observations of Elbourne and Ji (2019) do not imply that previous studies have not successfully identified unconventional monetary policy (or any other type of structural) shocks. That said, this does not mean that the identification strategy of SVAR models cannot be questioned. A conventional approach to justify any identification scheme should follow two steps. First, identification restrictions should be grounded on predictions of standard theoretical models, and possibly supported by other relevant empirical works in the literature. Thus, an identification strategy should be evaluated by the plausibility of the restrictions, as has been done in Section 2.1.
Second, the identified shocks should be consistent with the narrative of the conduct of policy over the period under investigation. In our specific case, the identified unconventional monetary policy shocks should capture major measures undertaken by the ECB between 2007 and 2015. In this respect, Boeckx et al. (2017) have already documented that the estimated time series of shocks captures fairly well these measures, as shown in Figure 5.
Consider, for example, the collapse of Lehman Brothers in September 2008. Elbourne and Ji (2019) argue that the identified expansionary (restrictive) unconventional monetary policy shocks are in practice expansionary (restrictive) financial stress shocks. If so, one would expect a considerable restrictive shock in September 2008. However, the SVAR of Boeckx et al. (2017) identifies a shock in the month of the collapse which dampened financial stress (and increased the balance sheet of the ECB). This is the opposite of an unfavorable financial stress shock. On the other hand, such a shock is consistent with the ECB’s full allotment decision and easing of collateral requirements in September 2008. Overall, the chart shows that major balance sheet policies launched by the ECB between 2007 and 2015 were followed by expansionary UMP shocks, while periods of more restrictive balance sheet policies or policy action were associated with contractionary UMP shocks, which indicates that the identification based on a combination of zero and sign restrictions is plausible.
In a recent paper that has received widespread international media attention, Elbourne and Ji (2019) have argued that studies based on SVAR models identified using a combination of zero and sign restrictions fail to plausibly identify true unconventional monetary policy shocks in the euro area. Their conclusion is based on the observation that they find indistinguishable impulse responses and time series of unconventional monetary policy shocks when they replace the balance sheet in the SVAR with random numbers, and when they impose the opposite restriction on the balance sheet in the identification scheme. In addition, they find that the shocks of these studies are uncorrelated with shocks that are obtained using high-frequency data in other studies.
In this note, we come to different conclusions. First, we show that finding qualitatively similar macroeconomic effects when switching the sign of the restriction of the balance sheet is not surprising since both shocks are associated with improved conditions in financial markets. At the same time, we show that important qualitative and statistical differences exist between this shock and the original unconventional monetary policy shock, implying that the sign restriction on the central bank balance sheet is important when identifying unconventional monetary policy shocks. Second, we have demonstrated that finding impulse responses and shock series that are very similar to those of an UMP shock when the balance sheet is replaced by random numbers is also not surprising. In particular, the thus identified shock is a linear combination of the original UMP shock and another shock that eases financial conditions, both shocks that would be expected to have expansionary macroeconomic effects. We further show that the high similarity between the shocks identified in a model with random numbers instead of the ECB’s balance sheet reflects the greater importance of the unconventional monetary policy shock in the data relative to the other shock during the sample period. Finally, the low correlation between unconventional monetary policy shocks with shocks obtained from high-frequency data reflects the fact that the two shocks capture different dimensions of monetary policy, balance sheet policies in the case of the former, and conventional monetary policy and forward guidance in the case of the latter.
Boeckx J., Dossche M. and G. Peersman (2017), "Effectiveness and Transmission of the ECB's Balance Sheet Policies", International Journal of Central Banking, 13(1), 297-333.
Burriel P. and A. Galesi (2018), "Uncovering the Heterogeneous Effects of ECB Unconventional Monetary Policies across Euro Area countries", European Economic Review, 101(C), 210-229.
Corsetti G., Duarte J. and S. Mann (2018), "One Money, Many Markets - A Factor Model Approach to Monetary Policy in the Euro Area with High-Frequency Identification", Cambridge Working Papers in Economics, No 1816.
Elbourne A. and K. Ji (2019), "Do Zero and Sign Restricted SVARs Identify Unconventional Monetary Policy Shocks in the Euro Area?", CPB Discussion Paper, No. 391.
Feldkircher, M., Gruber, T. and F. Huber (2019), "International Effects of a Compression of Euro Area Yield Curves", Journal of Banking and Finance, in press.
Gambacorta L., Hofmann B. and G. Peersman (2014), "The Effectiveness of Unconventional Monetary Policy at the Zero Lower Bound: A Cross-Country Analysis", Journal of Money, Credit and Banking, 46(4), 615-642.
Haldane, A., M. Roberts-Sklar, T. Wieladek and C. Young (2016), "QE: The Story so Far", Bank of England, Staff Working Papers, no 624.
Hofmann B. and G. Peersman (2017), "Monetary Policy Transmission and Trade-offs in the United States: Old and New", BIS Working Papers, No. 649.
Hollo D., Kremer M. and M. Lo Duca (2012), "CISS - A Composite Indicator of Systemic Stress in the Financial System", ECB Working Paper Series, No 1426.
Jarocinski M. and P. Karadi (2018), "Deconstructing Monetary Policy Surprises: the Role of Information Shocks", ECB Working Paper Series, No 2133.
Lewis V. and M. Roth (2019), "The Financial Market Effects of the ECB's Asset Purchase Programs", Journal of Financial Stability, forthcoming.
Montiel Olea, J., Stock J. and M. Watson (2016) "Inference in Structural VARs with External Instruments", Harvard University, mimeo.
Panizza, U. and C. Wyplosz (2016), "The Folk Theorem of Decreasing Effectiveness of Monetary Policy: What Do the Data Say?", paper prepared for the 17th J. Polak Annual Research Conference of the IMF, November 2016.
Peersman G. (2005), "What Caused the Early Millennium Slowdown? Evidence Based on Vector Autoregressions", Journal of Applied Econometrics, 20(2), 185-207.
Peersman G. (2018), "International Food Commodity Prices and Missing (Dis)Inflation in the Euro Area", CESifo Working Paper Series, No. 7338.
Rudebusch, G. (1998), "Do Measures of Monetary Policy in a VAR Make Sense? ", International Economic Review, 39 (4), 907-931.
Sims, C (1998), "Comment on Glenn Rudebusch’s “Do Measures of Monetary Policy in a VAR Make Sense? "", International Economic Review, 39 (4), 933-41.
Uhlig H. (2005), "What Are the Effects of Monetary Policy on Output? Results from an Agnostic Identification Procedure", Journal of Monetary Economics, 52(2), 381-419.
Weale, M. and T. Wieladek (2016), "What Are the Macroeconomic Effects of Asset Purchases?", Journal of Monetary Economics, 79, 81-93.