Did Quantitative Easing affect interest rates outside the US? New evidence based on interest rate differentials

Transcription

1 Did Quantitative Easing affect interest rates outside the US? New evidence based on interest rate differentials Ansgar Belke (Universität Duisburg-Essen, CEPS Brüssel, IZA Bonn) Daniel Gros (CEPS Brüssel) Thomas Osowski (Universität Duisburg-Essen) Vorlesung, Empirie der internationalen Geld- und Finanzmärkte, 14. Juni, 2016

2 Motivation Until 2013, Fed was the most active central bank by implementing several non-standard measures (such as QE). Aim of QE: put pressure on long-term yields to stimulate economic activity and prevent deflation. Impact of QE depends on the design of the programme, the economic environment and state of financial markets and a country s economic structure (different EXR reactions in QE countries).

3 Motivation II How to evaluate effectiveness of QE? : Existing studies focus on developments within country undertaking QE, mostly event studies. We find that bond purchases in the US alone do not perturb co-movements of US and euro area interest rates. => Either: QE ineffective, or QE very effective as it depresses interest rates everywhere (but then impact on exchange rate should be zero).

4 Motivation III Literature emphasizes two main transmission channels: signalling and portfolio-balance. Signalling channel could work in reverse: QE could be a reaction to negative news about economy (stronger adverse effects of the financial crisis than previously expected). Portfolio balance channel: FED QE does not affect supply of euro denominated securities. Why should FED QE then affect euro interest rates? (US investors hold mostly USD denominated securities.) => Portfolio balance model cannot explain why US QE should affect euro and global interest rates.

5 Our argument would be: Details QE is supposed to work because it changes the amount of long term government bonds in the hand of the public. But this implies that a priori one should not expect QE in the US, which changes the amount of US federal government debt in the hands of the public, to have any impact on the market for (longer term) government debt elsewhere (for example in the euro area). QE by the Federal Reserve does not have any impact on the amount of euro area government debt in the hands of the public.

6 Details II Since QE by the Federal Reserve lowers the amount of long-term US government debt securities in the hands of the public, the portfolio balance channel implies that their price should go up. One can of course argue that the reduction of the supply of US government securities leads to other changes in portfolios, which could ultimately affect the demand for euro area government securities. But for most regulated entities, which usually have to hold a certain proportion of their portfolios in government securities, US and euro area government securities are not substitutes since these entities are usually required to hold their portfolio mostly in domestic currency.

7 Details III It thus remains difficult to explain, on the basis of portfolio balance considerations why QE by the Federal reserve should have a strong impact on the price or yield of euro area securities. In short, QE by the Federal Reserve changes the relative supplies of US Federal debt securities and those of euro area governments. One would thus expect, a priori, that QE in the US should affect the co-integration of interest rates across the Atlantic.

8 Stylised facts Motivation IV Interest rates fell trend-wise in most advanced economies, independently of whether QE was implemented or not. Our view: CB s QE programmes had a common underlying source is somehow compatible with implications of the signalling channel (QE as a sign that the crisis would be longer and more severe). Fall in interest rates might have occurred anyway. (expectation revisions of markets).

9 Motivation V This paper is one of the first that tries to test empirically whether QE has changed relationships on international financial markets. See also Thornton, D.L. (2014). Has QE been effective?, Federal Reserve Bank of St. Louis, Economic Synopses, 2014, Number 3. Employs cointegration approach to analyze whether the Fed s QE1 has caused a structural change in the US-European interest rate relationship.

10 QE in action - Bond purchase programmes in the USA November 2008: the Federal Reserve announced first round of QE (QE1) Worth $1.75 trillion between November 2008 and March 2010 October 2010: FOMC announced second round of QE (QE2) September 2011: Announcement, Operation Twist September 2012: Third round of QE (QE3) Overall, Fed balance sheet increased by about $3.5 trillion (20% of GDP)

11 Quantitative Easing and global financial markets - Purchase programmes in the USA Figure 1. Federal Reserve Balance Sheet ($ mil.)

12 Quantitative Easing and global financial markets - Purchase programmes in the US Different Fed s QE programmes not only differed in terms of their concrete design, but also with regard to the underlying economic environment during the time of their implementation Perception of systemic risk differed strongly over time and between the starting points of the QE programmes Highest amount of market uncertainty after the beginning of the financial crisis in September 2008 QE2 and QE3 were introduced by the Fed when financial markets had already stabilized ( QE1 most successful )

13 Quantitative Easing and global financial markets - Purchase programmes in the US Figure 2. CBOE Volatility Index (VIX)

14 Quantitative Easing and global financial markets Numerous papers on domestic effects of QE, but literature on international effects still small (but growing). General findings: Impact varies across countries. Depends on time and circumstances of implementation. US: QE1 was most effective in influencing financial markets, unemployment and inflation, while QE2 was far less effective. Overall, effects and magnitude of such shocks highly uncertain (QE as response to crisis: extremely difficult to disentangle QE effects, financial markets and macroeconomic conditions & estimation methods and models rely on strong assumptions about the transmission mechanisms of QE)

15 Quantitative Easing and global financial markets II Event studies find very large effects of QE compared to studies using different empirical frameworks. Drawbacks: by definition cannot test longer term impact of QE, strong assumptions about the identification of monetary policy shocks etc. Other studies (e.g. using VARs): QE (especially QE1) depressed medium and long-term yields over the medium to long term. Effects on international financial markets: most papers find cross-border effects, as well as effects on exchange rates (up or down, but less so for US).

16 Quantitative Easing and global financial markets III Standard event study methodology does not provide an estimate of the persistence of a monetary policy shock (Wright, 2011). See also Hamilton (2011) and Thornton (2013) for several critical remarks on measuring the effects of QE by using event studies. Wright, J.H. (2012), What does monetary policy do to long term interest rates at the zero lower bound?, The Economic Journal, Vol. 122(564), pp. F447-F466. Hamilton, J.D. (2011), Evaluating Quantitative Easing using event studies ( Thornton, Daniel L. (2013), An Evaluation of Event-Study Evidence on the Effectiveness of the FOMC s LSAP Program: Are the Announcement Effects Identified?, Working Paper No A, Federal Reserve Bank of St. Louis, October.

17 Eyeball econometrics: QE, interest rates, exchange rates and inflation Aim of large-scale asset purchases is to lower longterm interest rates (flattening yield curve). QE might also work by raising inflation expectations and thereby decreasing real interest rates. For the US it is difficult to detect any observable impact of the QE programmes on either exchange rate or inflation (see Figure 3) Exchange rate and inflation underwent large swings, which are unrelated to the various rounds of asset purchases by the Fed

18 Effects of the Fed s QE on interest rates and exchange rates and inflation Figure 3. US: Central bank balance sheet, exchange rate and inflation Difficult to find clear impact of QE on inflation (puzzle: inflation follows white noise process in spite of QEs)

19 Effects of the Fed s QE on interest rates, exchange rates and inflation Figure 4. Euro area: Central bank balance sheet, exchange rate and inflation

20 Effects of the Fed s QE on interest rates, exchange rates and inflation Lessons from Figure 4: For the euro area, the link between the central bank s balance sheet and both the exchange rate and inflation appears to be stronger and more persistent (see Figure 4). Surprising since there has been no QE in the euro area (until end of 2014), and ECB could influence the size of its balance sheet only indirectly via its offers of long-term lending to banks at favourable rates. In paper also some more detailed tables on this issue.

21 Global financial markets and national QE Up to now: national level! But financial markets in advanced countries are very open and highly integrated. In reality (long-term) interest rates have followed a common long-term trend across major currency areas. (Long-term) rates have been highly correlated across advanced economies, not only along a downward trend, but also during cyclical ups and downs (see Figure 5). Correlation is too tight and has lasted too long to be just a coincidence.

22 Global financial markets and national QE II Figure 5. Long-term interest rates in major currency areas since 1990

23 Global financial markets and national QE III Effectiveness of large asset purchases by the Federal Reserve should thus not be measured simply by the fall in US interest rates, but by a fall in the interest rate differential between the US and the euro area (or other major markets). Large asset purchases by the Fed have failed to have a differential impact on the US. Over most US QE periods, rates have declined only as much, sometimes even less than they have in areas where QE was not undertaken (see Table 3) No QE episode with change in differential > 0.1 %.

24 Global financial markets and national QE IV Figure 6. Transatlantic long-term interest rate differential from 2007 (prior: should become larger, but does not)

25 Global financial markets and national QE V Severity of the crisis and economic recession in developed economies led to (further) reductions in long-term interest rates across countries along the downward trend QE could be understood as a reaction to the crisis, but did not in itself reduce interest rates No independent, separate effect of the US QE on the US economy that cannot be related to the global downward trend => QE effects overestimated? Small and non-persistent impact of US QE on the interest rate differential and limited impact on exchange rates and inflation point in that direction as well.

26 Data and empirical approach (Left out: section on fiscal policy, debt management and the portfolio balance channel => fiscal sector s policy reaction has crowded out QE s impact.) Treasury bond yields for the US and German Treasury bond yields for the Euro Area as interest rate measures 10-year Treasury bond yields Focus on long-term interest rate measures Including nominal exchange rate (USD/Euro) Monthly end-of-period data from 2002:01 to 2014:12

27 Data and empirical approach II Econometric framework: cointegrated vector autoregressive (CVAR) model Model the impact of domestic interest rate shocks on foreign interest rates and the exchange rate, while taking care of the feedback between the variables Basic representation: X t = A 1 X t A k X t k + φd t + ε t, t = 1,, T (1)

28 Data and empirical approach III Reformulating the model in an error-correction form allows to distinguish between stationarity created by linear combinations of the variables and stationarity created by first differencing: ΔX t = ΠX t 1 + Γ 1 ΔX t Γ k ΔX t k 1 + φd t + ε t, t = 1,, T (2) QE might have had a separate, identifiable impact on long-term interest rates in the US, which should show up as a break in the long-run relationship (see section 2, separate = for which the global downward trend in interest rates does not account) Check for a potential structural break around the time of the Fed s announcements of QE1!

29 Data and empirical approach IV Large set of recursive techniques as proposed by Johansen & Juselius (2006) to check for structural changes in the relationship 1. Log likelihood test 2. Recursive tests based on the eigenvalues 3. Recursive tests of the cointegration space 4. Recursively calculated prediction tests Backward recursive estimation techniques are used as a complement.

30 Empirical Results - Unit Root Tests Order of integration to be determined Table 4. Unit root tests Test results propose that these are (trend-) stationary. I(1) property of interest rates controversial, but found by many authors such as Campbell/Shiller (1987); stochastic trend as a reaction to business cycle shocks; drawback of I(0) models: imply long-rates which are not volatile enough.

31 Empirical Results Unit Root Tests Empirical studies analysing exchange rate behaviour and monetary policy shocks usually adopt interest rates as I(1), at least in subsamples: Dickey, D.A., Jansen, D.W., Thornton, D.L. (1991): A primer on cointegration with an application to money and income, Federal Reserve Bank of St. Louis, March/April, pp Dees, S., di Mauro, F., Pesaran, M.H., Smith, L.V. (2005): Exploring the international linkages of the euro area : A global VAR analysis, CESifo Working Paper, No J.H. Stock and M.W. Watson (1988): Testing for common trends, Journal of the American Statistical Association, Vol. 83, pp and a lot more.

32 Empirical Results - Unit Root Tests Structural breaks might have strong effects on the results of unit root tests, sometimes leading to wrong implications generated by tests. To strengthen the robustness of our unit root tests, Zivot-Andrews tests are used. Zivot-Andrews tests allow for a single break in the intercept, the trend or both. Results of the Zivot-Andrews tests (see Table 5) support the findings of the ADF and PP tests. Use of cointegration approach legitimised.

33 Empirical Results - Unit Root Tests Table 5. Unit root test allowing for structural breaks - Zivot-Andrews tests

34 Empirical Results Cointegrated VAR estimations Focus on the relationship between long-term (ten-year) bond yields, since interest is in QE impact on financial markets. Our model contains the following variables: Include three lags in the unrestricted VAR Dummy variables are included in stationary part of VARequations at the dates: 2004:04, 2008:10 and 2008:12 Assumptions of the CVAR satisfied (Table 6) Only small evidence of autocorrelation and ARCH-effects

35 Empirical Results Cointegrated VAR estimations Table 6. Residual analysis diagnostic tests of the unrestricted VAR (3)- Model

36 Empirical Results Cointegrated VAR estimations Table 7. LR Trace Test for the unrestricted VAR (3)- model Results of trace tests clearly indicate the presence of one single cointegrating relationship.

37 Empirical Results Cointegrated VAR estimations Table 8. The just-identified long-run cointegration relations for r = 1 Long-run relationship in line with theory!

38 Empirical Results Cointegrated VAR estimations 1. Test for proportionality between the interest rate measures by restricting the coefficient of the exchange rate variable to zero Test of restricted model clearly rejects the imposed restriction. => Exchange rate important component of the estimated long-run relationship. 2. Check for a relationship among the interest rate differential between German and US yields and the exchange rate (see Table 9)

39 Empirical Results Cointegrated VAR estimations Table 9. The over-identified long-run cointegration relations for r = 1 Model is accepted!

40 Empirical Results Cointegrated VAR estimations According to our estimations, we obtain the following long-run relationship: (Ger10Y US10Y) = EXR Estimated relationship indicates that a reduction of the US (German) yield should lead to a depreciation (appreciation) of the US dollar vis-à-vis the euro. Movements of the exchange rate and the German yields thus reduce deviations from the long-run relationships.

41 Did QE1 cause a structural break? Two dates of particular interest WRT possibility of a structural break caused by QE1: Announcement of QE1 in 2008:11 FOMC s decision to buy Treasury bonds as well as the quantitative expansion of the QE programme in 2009:03 Log-likelihood test indicates some instability starting around the beginning of the subprime mortgage crisis in 2007 (Figure 9).

42 Did QE1 cause a structural break? Figure 9. Test for constancy of the log-likelihood

43 Did QE1 cause a structural break? No evidence of a structural break around the breakdown of Lehman-Brothers and the subsequent beginning of the global financial crisis. No clear impact of QE on test statistics => no evidence of significant parameter non-constancy around introduction of QE1 in 2008:11. Further evidence of a structural break starts to appear in the middle of Empirical realisation of the test statistic quickly increases, but fails to reject the null hypothesis of parameter constancy Checking the robustness by backward recursive tests No effect of QE on the structure of the relationship!

44 Did QE1 cause a structural break? Focus on the stability of the cointegrating relationship by checking the constancy of the eigenvalues (Figure 10) Some signs of instability in the estimated cointegration relation First sign of instability appears in mid Around announcement/implementation of QE1 cointegration relationship shows an almost linear development => no signs of structural changes caused by QE1. Additional evidence is generated by Figure 11, which illustrates the fluctuations of the eigenvalues. again showing no sign of instabilities.

45 Did QE1 cause a structural break? Figure 10. Recursively estimated Trace Test statistics

46 Did QE1 cause a structural break? Figure 11. Fluctuations of the Eigenvalue These tests do not differentiate between instability caused by α and β. Next step: focus on the components of our long-run relationship, β Test shows no significant evidence of a possible structural break in the long-run relation (Figure 12).

47 Did QE1 cause a structural break? Figure 12. Test of Beta Constancy

48 Did QE1 cause a structural break? Figure 13. Test of known beta Stability is not rejected until the start of the euro area debt crisis. No evidence of QE1 impact (Figure 13). Test results indicate that the long-run relationship may hav changed in comparison to the pre-crisis era.

49 Did QE1 cause a structural break? Final check: one step-ahead predictions of system (Figure 14) With regard to the introduction of QE1, the evidence of a break is again rather limited: exceptions are 2009:03 und 2009:04; announcement of US Treasury bond purchases? Figure 14. One step-ahead predictions of the full system

50 Results Some empirical evidence of potential structural changes. Limited evidence that there is some degree of instability prior to the outbreak of the global financial crisis No evidence that the announcement and conduct of QE1 generated a change in the relationship of risk-free bonds of the USA and Europe Highest amount of instability found in mid- to late-2010 (European debt crisis) May have estabilised the relationship between the nominal exchange rate and the interest rate differential Robustness checks: 5- and 7-year yields & VIX => Again no evidence that QE1 generated a structural change in transatlantic interest rate relationship.

51 Results II The euro crisis is widely believed to have generated a flight of capital from the euro area periphery to Germany, the socalled safe haven effect. This might have depressed German yields and could thus have led to impression that QE in the US had an equal impact on yields in the US and the risk free part of the euro area. However, the euro crisis did not coincide with the QE operated by the Fed. One could thus argue that if the safe haven effect had really been so important, it should have created a structural break in the co-movements of US and euro area or German interest rates.

52 Conclusions Estimation results of our CVAR analysis generated theoretically expected relationships between US and European yields /the nominal exchange rate No significant evidence of a structural break in the interest rate relationship caused by QE1 No evidence that QE had an effect on US interest rates that cannot be explained by the global downward trend in (long-term) interest rates While we cannot reject QE1 having no impact on the transatlantic interest rate relationship, we find evidence that the beginning of the European debt crisis had a more destabilising impact on the relationship

53 Conclusions II As our analysis focused on the effects of QE1, it is possible that QE2 or QE3 had different effects on the interest rate relationship. As the asset purchases during QE2 consisted solely of US Treasury bonds, there may have been a strong(er) effect on the transatlantic relationship of risk-free bonds. On the other hand, the effects of QE2 and also QE3 are generally considered to be even smaller compared to QE1. In this regard, we do not believe that QE2 and QE3 had a more pronounced effect on the interest rate relationship than QE1.

54 Conclusions III With regard to the interpretation of our empirical results, we cannot discriminate between the hypothesis that QE has been ineffective and the hypothesis that the impact of QE, possibly non-zero, has been the same on both sides of the Atlantic which might be interpreted as a perfect interest rate spillover.

55 Conclusions IV However, the latter hypothesis is not compatible with the way in which QE is supposed to work. While there does not exist a generally accepted framework of how QE is supposed to work, most existing models (e.g. Neely, 2010) comprise some elements of portfolio preferences and portfolio adjustment costs. The working mechanism of QE is generally supposed to be that it reduces the supply of longer-dated government bonds and forces the private sector to hold more shortdated central bank deposits. If one accepts this portfolio balance view, one can explain why QE should drive up the price of longer-dated bonds (and thus depress their return).

56 Conclusions V But QE by the Federal Reserve only reduces the supply of US government bonds. Given that the supply (and presumably the demand) of Euro denominated government bonds is not affected by US QE, there is no reason why US QE should have an impact on euro area longer-term interest rates in the same was as it affects US rates. Moreover, the observation that QE does not affect transatlantic interest rate differentials also makes it difficult to maintain the general presumption that US QE constituted a 'currency war' because it led to a depreciation of the US dollar.

57 Conclusions VI The observation that (US) QE has not affected transatlantic interest rate co-movements is compatible with the view that QE should be viewed as a result of negative shocks to global inflation and/or demand. In this view QE should be viewed as a (predictable) reaction of central banks to negative inflation and/or demand shocks. These shocks could be global in nature as suggested by the recent empirical work which shows that inflation has a strong global component (e.g. Ciccarelli/Mojon, 2010).

58 Conclusions VII But even if the shocks were national in nature (i.e. there could be stronger deflationary forces in the euro area than in the US), their impact on long-term interest rates could be uniform if they are long-term themselves, given that long-term shocks to demand in any large economy would tend to be distributed across the global economy.