1
29 June 2015 Group 2
R.J. Döttling MSc Year 2014/2015
Bachelor’s Thesis Finance & Organization Semester 2 period 2
The effect of quantitative easing on the price and volatility
in the Pound-euro spot market:
A research on the effect of the MPC’s ‘Asset Purchase Programme’, financed by the Bank of England, on the Pound-euro spot market
Arend Griffioen 10086692
Abstract
This paper explores the effect of quantitative easing on the price and volatility in the Pound-euro spot market by applying an AR-GARCH model. The research uses weekly intervention size data as reported by the Bank of England and weekly data on the Pound-euro exchange rate from the period 2008-2013. One conclusion is that QE in the UK appreciated the pound relative to the euro and decreased the Pound-euro volatility in the whole sample. Additional tests with a dummy variable are done to investigate whether QE had a different effect in different parts of the sample. Weak evidence is found that QE only appreciated the pound at the onset of the ‘Asset Purchase Programme’ and actually depreciated it thereafter.
2
Statement of Originality
The author of this thesis hereby declares the originality of this thesis. The topic of the thesis is chosen by the author itself as a follow up on existing literature.
3
1. Introduction
The recent global financial crisis that started in 2007 led several monetary policy makers in developed economies to adopt unconventional policy measures, such as quantitative easing (QE) (Voutsinas & Werner, 2011). The Bank of Japan (BoJ) was the first among central banks to conduct such a policy to fight deflationary pressures in the 1990s (Joyce et al., 2012a, p. 274). The primary goals of the policy are to stabilize financial conditions and support aggregate demand in conditions of an effective lower bound on the nominal short-term policy rate (Nelson, 2013, p. 92).
In the UK, the Monetary Policy Committee (MPC) announced to use QE on 5 March 2009, financed by the Bank of England (BoE) (Joyce et al., 2012b). The goal of their ‘Asset Purchase Programme’ (APP) was to mitigate the undershooting of the consumer price index (CPI) inflation target of 2% (2012b, p. 348). For this reason, the MPC began to purchase public and private assets using central bank money, thereby increasing the money supply (Benford et al., 2009, p. 90).
Macroeconomic theory states that an increase in the money supply lowers the value of a currency (Mankiw, 2013). The expectation is thus that the increase in the money supply will depreciate the currency (2013, pp. 131-161). Previous literature also reports evidence of the link between QE and exchange rates. This thesis therefore investigates the effect of the APP on the Pound-euro exchange rate.
The contribution of this thesis is that it uses real intervention size data on a weekly frequency as reported by the BoE. Data on the Pound-euro exchange rate is used from the period 2008-2013. For the analyses, an AR-GARCH model will be used. Additional tests with dummy variables show whether QE had a different effect in different parts of the sample.
The main question of this research is: what is the effect of quantitative easing by the BoE on the price and volatility in the Pound-euro spot market? The main findings of this paper are that QE appreciates the pound relative to the euro and decreases the Pound-euro volatility over the whole sample period. On the other hand, this paper finds weak evidence that QE appreciated the pound only in the period before 28 Jan 2010 and actually depreciated it thereafter. This indicates that QE had a stabilizing effect on the overall economy in the period just after the collapse of Lehman Brothers. Another finding is that the volatility is lower in the period after 28 Jan 2010 than it was before that date, although no evidence is found that the impact of QE on the volatility was different between those two periods. As will be discussed, some of the findings in this paper contradict conventional monetary theory and earlier studies on central bank interventions.
4
Section 2 starts by explaining QE based on literature and through which channels it works and concludes with the findings of earlier studies. Section 3 describes the data and methodology used for this research in detail. Section 4 describes the results and analyses. Finally, Section 5 discusses the results and main drawbacks of this research.
2. Literature
The failure of Lehman Brothers in September 2008 induced a sharp decline of the confidence in the world economy (Joyce et al., 2011, p. 200), (see also Figure 2 in the Appendix for its effect on the Pound-euro exchange rate). Amongst other central banks, the BoE loosened their monetary policy to boost spending and inflation (Benford et al., 2009). By March 2009, the official Bank Rate was at 0.5%, its effective zero lower bound (ZLB) (2009, p. 90). Despite reaching the ZLB, the MPC felt additional measures were needed to reach its 2% inflation target in the medium run (Kapetanios et al., 2012, p. 316). Therefore, the MPC announced to purchase private- and public-sector assets using central bank reserves (Nelson, 2012, p. 93). As Table 1 shows, gilts comprised almost 98% of the net asset purchases.
Table 1: Net asset purchases in millions of pounds from 2008-2013
Asset Net purchases Share
Commercial paper 8,207 2.2%
Gilts 366,575 97.8%
Bonds 150 0.0004%
Total 374,932 100%
Excerpted from the Bank of England’s balance sheets.
When central banks purchase assets from banks, it simply credits the bank’s reserve account with the additional funds, thereby increasing the monetary base (Benford et al., 2009). In contrast, purchases from non-banks (as is the case with gilt purchases) credit the reserve account of the seller’s bank with the funds and so increase the monetary base and broad money supply at the same time (2009, p. 91-92).
Economic theory and empirical evidence show a link between money growth, inflation rates and exchange rates. The quantity theory of money shows a positive one-for-one
relationship between the rate of money growth and the rate of inflation (Mankiw, 2013). The Fisher-effect shows a positive correlation between inflation and interest rates (2013, pp. 100-111). Furthermore, there is a positive link between the inflation rates of countries and their exchange rates (2013, pp. 155-156). Because the APP increases the money supply, it is
5
of the pound relative to the euro. On the one hand, this thesis finds strong contradicting evidence for the whole sample. The results show that the pound actually appreciated during the whole QE period. On the other hand, evidence shows that this appreciation is mainly caused by the period before 28 Jan 2010, just after the collapse of Lehman Brothers. Weak evidence shows that QE actually depreciated the pound in the period after 28 Jan 2010.
The next subsections will first review some theoretical channels through which QE may work together with some related empirical results. Subsequently, some of the results of previous literature on QE will be discussed.
2.1. Channels
Joyce et al. (2010 and 2011) discuss some channels through which QE in the UK and in general may affect asset prices. This is important because changing asset prices means changing interest rates, which in turn affect exchange rates (see also Dropsy, 1996, p. 55). The main channels are: the signaling channel and the portfolio balance channel. This subsection outlays these channels to get a clearer view of the impact QE can have on the financial markets.
2.1.1. Signaling channel
In their papers, Joyce et al. state that this channel captures news announcements about
expected future policy rates. This affects gilt yields by changing their discount rates and feeds through into other asset prices. QE might signal lower policy rates in the short term and simultaneously signal higher inflation in the future, thus leaving the net effect on asset prices ambiguous (2010, p. 117). They also argue that asset purchases can be a useful tool to support spending and help ensure inflation expectations remain anchored to the target (2011, p. 201).
2.1.2. Portfolio balance channel
The portfolio balance channel reflects the direct impact on asset prices of investors
rebalancing their portfolios (Joyce et al., 2010). Because QE involves the central bank buying assets, it decreases the quantity of assets supplied. If assets are not perfect substitutes, then a change in the quantity of assets leads to a change in its relative expected rate of return. Thus, imperfect substitutability gives the BoE a mechanism to affect asset prices by inducing sellers to rebalance their portfolios (2010, p. 117). While the signaling channel affects expected policy rates, the portfolio balance effects work by reducing the spreads of longer-term interest rates and risk premia more generally (Joyce et al., 2011, pp. 201-202). Breedon, Chadha, and
6
Waters (2012) find supporting evidence for the existence of the portfolio balance effect in the UK (2012, p. 727).
2.1.3. Other channels
One other channel that Joyce et al. discuss is, for example, the liquidity premia channel. QE may affect the risk premia on asset prices, because central bank purchases provide extra liquidity in the financial markets. They state that asset prices may therefore increase through lower premia for illiquidity (2011, p. 201). Because this channel depends on the flow of purchases for its effect, they expect that it would be temporary and limited to the duration of the QE program (2010, p. 118).
Two other effects they discuss are confidence effects and bank lending effects. Joyce et al. argue that confidence effects may arise because asset purchases boosts consumer confidence and their willingness to spend. Bank lending effects arise because purchases from non-banks gains the banking sector both new reserves at the BoE and a corresponding
increase in customer deposits. This higher level of liquid assets could then encourage banks to give out more loans than they otherwise would. However, the MPC expects little impact through this channel due to increasing pressures on banks to reduce the size of their balance sheets (2011, p. 202). The study of Butt et al. (2014) on QE and bank lending in the UK supports this expectation. They find no evidence for increased bank lending due to QE (2014, p. 37). While there is no evidence for the bank lending effect, QE may have a significant effect on asset prices through the other channels.
2.2. Previous findings
This subsection outlays some relevant results of earlier studies on central bank interventions. Some studies find evidence that interventions depreciate the domestic currency and increase the exchange rate volatility, while others find only weak or no evidence. In contrast to these findings, this thesis reports strong evidence that the pound appreciated during QE in the UK and that the Pound-euro volatility decreased. On the other hand, this thesis also finds weak evidence that QE only appreciated the pound before 28 Jan 2010 and actually depreciated it thereafter.
Bonser-Neal and Tanner (1996) test the effectiveness of central bank intervention on the ex-ante volatility of the U.S. dollar/Deutsch Mark and U.S. dollar/Japanese Yen between 1985 and 1991. The use of ex-ante volatility means that it incorporates market participants’ expectations of volatility. Their conclusion is that central bank intervention is generally associated with either an increase in exchange rate volatility or no change at all.
7
Kim and Sheen (2006) investigate the effect of central bank intervention on the volatility of exchange rates in Japan. One of their main findings is that large and infrequent BoJ interventions after 1995 successfully influence exchange rate movements. However, the interventions cause more market disorderliness in the form of higher market volatility (2006, p. 3192-3193).
Frenkel, Pierdzioch, and Stadtmann (2005) also find that BoJ interventions increased the volatility of the Yen-dollar exchange rate during the period of 1993-2000 (2005, p. 38). Likewise, Fratzscher (2006) finds evidence that QE increases exchange rate volatility in the US, Eurozone, and Japan (2006, pp. 164-165). In contrast, Aguilar and Nydahl (2000) find no significant evidence of interventions influencing the price and volatility of the Swedish krona/U.S. dollar exchange rate (2000, p. 321).
More recently, Glick and Leduc (2012) find evidence that QE announcements in the US and the UK led to depreciations of the U.S. dollar and British Pound on those days (2012, p. 2078). Likewise, Kenourgios, Papadamou, and Dimitriou (2015) find a direct negative impact around QE announcements on the British Pound and no effect on its volatility (2015, p. 110).
Other papers study the effects of QE in the US and the UK on other macroeconomic variables (e.g. Kapetanios et al. (2012) and Christensen & Rudebusch (2012)). The
contribution of this thesis to the existing literature is that it uses real intervention size data to investigate the effect of QE on the price and volatility in the Pound-euro spot market. As mentioned before, the findings of an appreciating pound and decreasing volatility in the whole sample is in contrast with macroeconomic theory and the results of previous studies. On the other hand, the weak indication that QE depreciated the pound in the period after 28 Jan 2010 is actually in line with the theory and some of the literature.
3. Research method and data
This Section will describe the research method and data used for this research in detail. To measure the effect of QE on the Pound-euro exchange rate, the generalized autoregressive conditional heteroscedasticity (GARCH) model will be used. The GARCH is an extension of the ARCH model and is used to measure the time-varying volatility of asset returns (Stock & Watson, 2012, p. 703). Thus, the first test is to examine whether there is clustering of
volatility in the data of the Pound-euro. Figure 1 below plots the volatility of the Pound-euro exchange rate in the sample period. The Figure shows some visual evidence of volatility clustering (especially around the year 2009).
8
Figure 1: Volatility Clustering
The next step is to test whether there actually is an ARCH effect in the data. This is done by predicting the error of the Pound-euro exchange rate in the sample and testing it with the Lagrange Multiplier (LM) test for autoregressive conditional heteroscedasticity. Table 2 shows the output of this test, obtained with the help of statistical software.
Table 2: LM test for autoregressive conditional heteroscedasticity (ARCH)
Lags Chi2 Degrees of Freedom Prob>Chi2
1 208.489 1 0.000***
*** indicates statistical significance at the 1% level.
The Chi-squared statistic is significant at the 1% level, which means that there is enough statistical evidence to reject the null hypothesis of no ARCH effects in the first lag. This evidence validates the usage of the model for this research. The model used for this research is the first-order GARCH(1,1) model, because it uses weekly observations over a time span of six years. Higher-order GARCH(r,s) models with additional lag terms are useful when a long span of data is used, like several decades of daily data or a year of hourly data (Engle, 2001, p. 166).
To test whether to include AR(p) (autoregressive) terms to the model, the Akaike information criterion (AIC) is used (see Table 3 in the Appendix for the output of the first six lags). The Table shows that including the first lag of the Pound-euro has the best fit to the data. Therefore, an AR(1) term is added to the model. Further, to test whether to include MA(q) (moving average) terms to the model, Bartlett’s formula is used at the 95% confidence level. Figure 2 in the Appendix shows that the first six lags lie within this confidence band, indicating to include no moving averages.
9 Conditional mean equation
𝛥ln(𝑃𝐷𝐸𝑈𝑅)𝑡= 𝛽0+ 𝛽1𝐼𝑁𝑇𝑆𝐼𝑍𝐸𝑡+ 𝛽2𝛥ln(𝐷𝑂𝐿𝐸𝑈𝑅)𝑡+ 𝛽3𝑈𝐾𝐶𝑃𝐼𝑡+ 𝛽4𝐸𝑍𝐶𝑃𝐼𝑡 + 𝛽5𝑈𝐾𝐺𝐷𝑃𝑡+ 𝛽6𝐸𝑍𝐺𝐷𝑃𝑡+ 𝛽7𝛥ln(𝐹𝑇𝑆𝐸)𝑡+ 𝛽8𝛥ln(𝑃𝐷𝐸𝑈𝑅)𝑡−1+ 𝜀𝑡
Conditional variance equation
𝜎𝜀2𝑡 = 𝛼0+ 𝛼1𝜀𝑡−12 + 𝛾1𝐼𝑁𝑇𝑆𝐼𝑍𝐸𝑡+ 𝜑1𝜎𝜀2𝑡−1
𝛥𝑙𝑛(𝑃𝐷𝐸𝑈𝑅)𝑡 first difference of the log of the Pound-euro exchange rate 𝐼𝑁𝑇𝑆𝐼𝑍𝐸𝑡 net asset purchases by the BoE, in billions of Pounds 𝛥𝑙𝑛(𝐷𝑂𝐿𝐸𝑈𝑅)𝑡 first difference of the log of the Dollar-euro exchange rate
𝑈𝐾𝐶𝑃𝐼𝑡 unexpected CPI inflation rate changes in the UK
𝐸𝑍𝐶𝑃𝐼𝑡 unexpected CPI inflation rate changes in the Eurozone
𝑈𝐾𝐺𝐷𝑃𝑡 unexpected GDP growth changes in the UK
𝐸𝑍𝐺𝐷𝑃𝑡 unexpected GDP growth changes in the Eurozone
𝛥ln(𝐹𝑇𝑆𝐸)𝑡 first difference of the log the of FTSE 100 index
𝛥𝑙𝑛(𝑃𝐷𝐸𝑈𝑅)𝑡−1 first difference of the log of the Pound-euro exchange rate at time t-1
(the ‘AR term’)
𝜀𝑡 disturbance term
𝜎𝜀2𝑡 conditional variance of daily exchange rate changes.
The estimators of the GARCH coefficients are computed by the method of maximum likelihood estimation (MLE) with the help of statistical software.
The data set of this research comprises of 306 weekly observations on eight variables in the period from January 2008 to December 2013. The main variables are the log of the Pound-euro exchange rate closing prices and the intervention size in billions of pounds, as reported by the BoE in their data base. To avoid problems caused by nonstationarity in the data, the first difference of the log of Pound-euro is used for the analysis (see also Stock & Watson, 2013, pp. 588-610). The data on the Pound-euro exchange rate and the intervention size are plotted in Figure 3 and Figure 4 in the Appendix.
The first difference of the log of the Dollar-euro exchange rate is used as a control variable to sort out its spillover effects on the Pound-euro. This data is obtained from the data base of the European Central Bank. Further, as in the paper of Frenkel et al. (2005), the first difference of the log of the London Stock Exchange FTSE 100 index will be used to control for the impact of economic events. This data is obtained from the Yahoo finance website.
10
The other control variables are unexpected GDP growth and unexpected CPI inflation rate in the UK and Eurozone. These are macroeconomic announcement figures and are included for the following reasons. First, according to the flexible-price monetary model, GDP levels influence exchange rates (Pilbeam, 2013). This is because when income rises, the transactions demand for money increases. Assuming constant money stock and interest rates, the increased demand for money can only come about through a fall in domestic prices. Assuming purchasing power parity (PPP) between countries, the fall in domestic prices thus requires an appreciation of the currency to maintain PPP (2013, p. 152-156). Second, the ‘Fisher effect’ shows that there is a positive link between countries’ inflation rates and their exchange rates (Mankiw, 2013, pp. 100-111). Finally, the Efficient Market Hypothesis (EMH) states that all available information is reflected in the current price of assets (Bodie, Kane, & Marcus, 2011, pp. 371-401). Thus, only unexpected GDP growth and CPI inflation rates ought to matter for exchange rate movements. Therefore, the control variables
unexpected GDP growth unexpected CPI inflation rate in the UK and Eurozone are added to the equation. The actual and forecast figures of GDP growth and CPI inflation rate are comprised by market watcher Forexfactory. To obtain the unexpected growth and inflation rates, the forecast figures are subtracted from the actual figures. Fatum and Scholnick (2008) find supporting evidence that only the unexpected part of macroeconomic news influences exchange rates (2008, pp. 1084-1085).
Figure 4 in the Appendix shows the net asset purchases by the BoE in the sample period. The Figure shows there were no interventions in the period of Feb 2010 to Oct 2011. Two tests are performed to investigate whether the effect of QE was different before and after this period. For this purpose, the dummy variable, 𝑇𝐼𝑀𝐸𝐷𝑈𝑀𝑀𝑌𝑡, is created. In the first test, the dummy takes on the value of 1 after Oct 2011 and 0 otherwise. In the second test, the dummy takes on the value of 1 after Feb 2010 and 0 before that date. Specifically, the following equations are estimated for the two tests separately:
Conditional mean equation
𝛥ln(𝑃𝐷𝐸𝑈𝑅)𝑡= 𝛽0 + 𝛽1𝐼𝑁𝑇𝑆𝐼𝑍𝐸𝑡+ 𝛽2𝛥ln(𝐷𝑂𝐿𝐸𝑈𝑅)𝑡+ 𝛽3𝑈𝐾𝐶𝑃𝐼𝑡+ 𝛽4𝐸𝑍𝐶𝑃𝐼𝑡 + 𝛽5𝑈𝐾𝐺𝐷𝑃𝑡+ 𝛽6𝐸𝑍𝐺𝐷𝑃𝑡+ 𝛽7𝛥ln(𝐹𝑇𝑆𝐸)𝑡+ 𝛽8𝛥ln(𝑃𝐷𝐸𝑈𝑅)𝑡−1
11 Conditional variance equation
𝜎𝜀𝑡
2 = 𝛼
0+ 𝛼1𝜀𝑡−12 + 𝛾1𝐼𝑁𝑇𝑆𝐼𝑍𝐸𝑡+ 𝛾2𝑇𝐼𝑀𝐸𝐷𝑈𝑀𝑀𝑌𝑡+ 𝛾3(𝐼𝑁𝑇𝑆𝐼𝑍𝐸 ∗ 𝑇𝐼𝑀𝐸𝐷𝑈𝑀𝑀𝑌)𝑡
+ 𝜑1𝜎𝜀𝑡−1
2
𝑇𝐼𝑀𝐸𝐷𝑈𝑀𝑀𝑌𝑡 dummy variable that takes on value of 1 after Oct 2011
or Feb 2010
(𝐼𝑁𝑇𝑆𝐼𝑍𝐸 ∗ 𝑇𝐼𝑀𝐸𝐷𝑈𝑀𝑀𝑌)𝑡 interaction term.
The other variables are defined as in the previous model. As before, the method of MLE will be used to estimate the GARCH coefficients.
4. Results
This Section will briefly outlay the results of this thesis. In the first part, the results of the conditional mean equation are discussed. The second part shows the results of the conditional variance equation. The third part discusses whether QE had a different effect in different parts of the sample.
4.1. Conditional mean equation
Table 4 on the next page shows the MLE estimators of the GARCH model without dummy variables. The base specification for the model is presented in column (1). Regressions that include the control variables are presented in columns (2) – (5). Adding control variables to the regression increases the p-value of the intervention size coefficient. Additionally, the AR term becomes statistically significant at the 5% level when the control variables are added. The ARCH specifications are given in the middle of the Table and the summary statistics at the bottom. For convenience, the coefficients on intervention size in the conditional mean equation are multiplied by 1,000 in the discussion.
The regression in column (1) shows the estimated coefficient on intervention size. This coefficient is positive (0.516) and statistically significant at the 5% level. The positive sign indicates that intervention size tends to appreciate the pound in this sample.
In column (2), the first difference of the log of the Dollar-euro exchange rate is added. The coefficient on this variable is statistically significant at the 1% level, which indicates that it is a good control variable. The coefficient on intervention size changes from 0.516 to 0.557. The p-value improves from 0.038 to 0.007 and the log pseudo-likelihood increases from 935.20 to 971.23. This indicates an improvement of the model by adding this control variable.
12
Table 4: GARCH model specifications
Dependent variable: Δlog(pound-euro). Period: 01/2008 - 12/2013. n = 305.
[1] [2] [3] [4] [5]
Conditional mean equation
intervention size 5.16E-04 5.57E-04 5.75E-04 5.83E-04 6.04E-04 (0.038)** (0.007)*** (0.004)*** (0.004)*** (0.003)*** Δlog(dollar-euro) 0.3480 0.3566 0.3570 0.3407 (0.000)*** (0.000)*** (0.000)*** (0.000)*** unexp. inflation UK -0.0158 -0.0161 -0.0166 (0.002)*** (0.002)*** (0.002)*** unexp. inflation EZ -0.0080 -0.0095 -0.0088 (0.269) (0.218) (0.262) unexp. GDP growth UK 0.0069 0.0073 (0.024)** (0.016)** unexp. GDP growth EZ -0.0068 -0.0075 (0.610) (0.560) Δlog(FTSE 100) -0.0370 (0.200) Intercept -0.0004 -0.0006 -0.0001 -0.0001 -0.0001 (0.524) (0.231) (0.242) (0.280) (0.257) AR L1. -0.0742 -0.1501 -0.1625 -0.1485 -0.1490 (0.277) (0.021)** (0.011)** (0.028)** (0.029)** Conditional variance equation
intervention size -0.0583 -0.2662 -0.3491 -0.3559 -0.3561 (0.695) (0.040)** (0.011)** (0.012)** (0.009)*** ARCH L1. 0.1638 0.1201 0.1310 0.1373 0.1352 (0.022)** (0.019)** (0.013)** (0.014)** (0.015)** GARCH L1. 0.7884 0.8442 0.8469 0.8432 0.8445 (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** Intercept -11.73 -12.22 -12.57 -12.63 -12.60 (0.000) (0.000) (0.000) (0.000) (0.000) Summary statistics and joint tests
Wald Chi2 5.82 92.47 124.12 133.27 139.84
(0.0546)* (0.000)*** (0.000)*** (0.000)*** (0.000)*** Log pseudo-likelihood 935.20 971.23 976.24 977.94 979.22
The p-values are given in parentheses. The coefficients are statistically significant at the *10%, **5%, or ***1% level.
Column (3) adds unexpected GDP growth in the UK and Eurozone. This UK control variable is statistically significant at the 1% level; the Eurozone control variable is
insignificant. The intervention size coefficient changes from 0.557 to 0.575. The p-value changes from 0.007 to 0.004 and the log pseudo-likelihood increases from 971.23 to 976.24. Thus, adding these control variables also improve the model.
13
Column (4) presents the regression where the control variables unexpected CPI
inflation rate in the UK and Eurozone are added. This UK control variable is significant at the 5% level and the Eurozone control variable is insignificant. The coefficient on intervention size changes from 0.575 to 0.583 and the p-value remains at 0.004. The log pseudo-likelihood changes from 976.24 to 977.94, which indicates a small improvement of the model.
Finally, in column (5), the control variable first difference of the log of the FTSE 100 index is added. The coefficient on this variable has a p-value of 0.200, which means it is not a strong variable in explaining the movement of the Pound-euro exchange rate. Nevertheless, the log pseudo-likelihood changes from 977.94 to 979.22. In addition, coefficient on
intervention size changes from 0.583 to 0.604 and its p-value improves slightly from 0.004 to 0.003. Thus, adding this control variable again improves the model.
4.2. Conditional variance equation
The p-value of the intervention size coefficient improves from 0.695 in column (1) to 0.009 in column (5) when all control variables are added. The sign of this coefficient is negative, which indicates that QE decreases the volatility in the sample.
The ARCH specifications show how well previous period’s volatility explains current period’s volatility. In columns (2) – (5), which include the control variables, the estimated ARCH coefficient is around 0.13 and statistically significant at the 5% level. The estimated GARCH coefficients in the regressions are significant at the 1% level and approximately take on the value of 0.84. The sum (0.97) of the ARCH and GARCH coefficients is close to unity, which indicates that changes in the conditional variance are persistent (Stock & Watson, 2013, p. 704). It confirms the earlier ARCH-effect tests of volatility clustering performed in Section 3.
4.3. Tests with dummy variable
Table 5 and Table 6 on the next two pages show the MLE estimators of the model with a dummy variable and interaction term to investigate whether QE had a different effect in different parts of the sample. In the discussion, the coefficients of the conditional mean equation are multiplied by 1,000 for convenience again.
Table 5 shows the output of the model where the dummy variable takes on the value of 1 after 06 Oct 2011 and 0 otherwise. Column (1) in Table 5 shows the regression without the dummy and interaction term. In column (2), the dummy variable and interaction term are added in both the conditional mean and variance equation. The coefficients on these variables
14
are highly insignificant for the conditional mean equation. Thus, no evidence is found that QE had a different effect on the exchange rate before 06 Oct 2011 than after that date.
Table 5: GARCH model with dummy where timedummy = 1 after 06 Oct 2011
Dependent variable: Δlog(pound-euro). Period: 01/2008 - 12/2013, n = 305. Timedummy = 1 after 06 Oct 2011.
[1] [2] [3]
Conditional mean equation
intervention size 6.04E-04 6.47E-04 6.73E-04
(0.003)*** (0.041)** (0.003)***
timedummy -6.05E-05 1.65E-04
(0.955) (0.878)
(int.size * timedummy) -5.78E-05 -1.19E-04
(0.897) (0.783) Δlog(dollar-euro) 0.3407 0.3405 0.3367 (0.000)*** (0.000)*** (0.000)*** unexp. inflation UK -0.0166 -0.0165 -0.0161 (0.002)*** (0.002)*** (0.003)*** unexp. inflation EZ -0.0088 -0.0090 -0.0065 (0.262) (0.323) (0.425) unexp. GDP growth UK 0.0073 0.0071 0.0074 (0.016)** (0.021)** (0.014)** unexp. GDP growth EZ -0.0075 -0.0102 -0.0105 (0.560) (0.404) (0.393) Δlog(FTSE 100) -0.0370 -0.0377 -0.0365 (0.200) (0.180) (0.197) Intercept -0.0001 -0.0001 -0.0005 (0.257) (0.464) (0.451) AR L1. -0.1490 -0.1453 -0.1480 (0.029)** (0.035)** (0.031)** Conditional variance equation
intervention size -0.3561 -0.0515 (0.009)*** (0.713) timedummy -0.8837 -1.2260 (0.020)** (0.005)*** (int.size * timedummy) -0.2829 (0.267) ARCH L1. 0.1352 0.1177 0.1373 (0.015)** (0.032)** (0.011)** GARCH L1. 0.8445 0.8168 0.7913 (0.000)*** (0.000)*** (0.000)*** Intercept -12.60 -11.46 -11.40 (0.000) (0.000) (0.000) Summary statistics and joint tests
Wald Chi2 139.84 139.73 134.95
(0.000)*** (0.000)*** (0.000)***
Log pseudo-likelihood 979.22 981.65 980.77
The p-values are given in parentheses. The coefficients are statistically significant at the *10%, **5%, or ***1% level.
In the conditional variance equation, the coefficients on intervention size and the interaction term are insignificant. The insignificant coefficient on the interaction term means that no evidence is found that QE had a different effect on the volatility before and after 06 Oct 2011. The dummy is statistically significant at the 5% level. In column (3), the
15
insignificant variables are removed. The dummy is now significant at the 1% level and its sign is negative. This indicates that the volatility was lower after 06 Oct 2011 than before that date.
Table 6: GARCH model with dummy where timedummy = 1 after 28 Jan 2010
Dependent variable: Δlog(pound-euro). Period: 01/2008 - 12/2013, n = 305. Timedummy = 1 after 28 Jan 2010.
[1] [2] [3]
Conditional mean equation
intervention size 6.04E-04 9.44E-04 9.41E-04
(0.003)*** (0.010)*** (0.007)***
timedummy 0.0018 0.0018
(0.187) (0.178)
(int.size * timedummy) -4.09E-04 -4.30E-04
(0.349) (0.311) Δlog(dollar-euro) 0.3407 0.3421 0.3351 (0.000)*** (0.000)*** (0.000)*** unexp. inflation UK -0.0166 -0.0160 -0.0156 (0.002)*** (0.002)*** (0.004)*** unexp. inflation EZ -0.0088 -0.0069 -0.0056 (0.262) (0.390) (0.457) unexp. GDP growth UK 0.0073 0.0066 0.0070 (0.016)** (0.041)** (0.032)** unexp. GDP growth EZ -0.0075 -0.0085 -0.0087 (0.560) (0.506) (0.489) Δlog(FTSE 100) -0.0370 -0.0368 -0.0372 (0.200) (0.174) (0.179) Intercept -0.0001 -0.0021 -0.0020 (0.257) (0.104) (0.108) AR L1. -0.1490 -0.1609 -0.1576 (0.029)** (0.023)** (0.022)** Conditional variance equation
intervention size -0.3561 -0.1988 (0.009)*** (0.353) timedummy -1.2086 -1.1253 (0.037)** (0.042)** (int.size * timedummy) -0.1488 (0.538) ARCH L1. 0.1352 0.0931 0.1375 (0.015)** (0.019)** (0.010)*** GARCH L1. 0.8445 0.8466 0.8210 (0.000)*** (0.000)*** (0.000)*** Intercept -12.60 -11.04 -11.55 (0.000) (0.000) (0.000) Summary statistics and joint tests
Wald Chi2 139.84 139.93 130.75
(0.000)*** (0.000)*** (0.000)***
Log pseudo-likelihood 979.22 982.48 980.49
The p-values are given in parentheses. The coefficients are statistically significant at the *10%, **5%, or ***1% level.
Table 6 shows the output of the model where the dummy variable takes on the value of 1 after 28 Jan 2010 and 0 before that date. Column (1) in Table 6 shows the regression
16
interaction term are added. The interaction term coefficient in the conditional mean equation (-0.409) is insignificant with a p-value of 0.349. The coefficient on intervention size increases from 0.604 to 0.944. The sum of the intervention size and interaction term coefficient is 0.535, which is close to 0.604. Although the coefficient on the interaction term is
insignificant, its negative sign combined with the increased coefficient on intervention size is a weak indication that QE only appreciated the pound before 28 Jan 2010 and that it actually depreciated the pound in the period thereafter.
In the conditional variance equation, the coefficient on the dummy variable is significant at the 5% level. As in the previous test, the insignificant intervention size and interaction term are removed from the regression in column (3). The coefficient on the
dummy is again negative and takes on the value of -1.23. This indicates that the volatility was lower in the period after 28 Jan 2010 than before that date. The value is also close to the -1.13 of the previous dummy test, which indicates that the level of volatility did not change much after 28 Jan 2010.
In sum, there is strong evidence that QE appreciated the pound relative to the euro in the whole sample. Furthermore, the negative sign on the interaction term (although not significant) indicates that QE first appreciated the pound just after the collapse of Lehman Brothers and depreciated it thereafter. There is also strong evidence that QE decreased the volatility over the whole sample period. In addition, the volatility is found to be lower after 28 Jan 2010 than before that date, although no evidence is found that the impact of QE on the volatility was different between those two periods.
5. Discussion
This paper uses weekly intervention size data to test the effects of QE on the Pound-euro exchange rate and its volatility. For the analyses, an AR(1)-GARCH(1,1) model is applied. This thesis also tests whether QE had a different effect in different parts of the sample. In fact, such findings are reported on the price of the Pound-euro exchange rate.
Macroeconomic theory predicts that an increase in money supply will depreciate the currency. Previous literature supports this theory and also reports that central bank
interventions tend to increase exchange rate volatility. The findings of this paper challenge this by providing strong evidence that the APP appreciated the pound relative to the euro and decreased the Pound-euro volatility over the whole sample.
On the other hand, this paper also finds weak evidence that the pound appreciated only at the onset of the QE program, in the period just after the collapse of Lehman Brothers and
17
actually depreciated it thereafter. The finding of an appreciating pound can be explained because of the calming effect QE can have on the financial markets and on the overall economy through the confidence channel by boosting consumer confidence, as discussed in Section 2. The weak indication that QE depreciated the pound after 28 Jan 2010 is in line with the prediction of macroeconomic theory of an increasing money supply and previous findings in the literature.
The fact that some papers find that QE affect exchange rates and others do not, suggests that it is hard to reach consensus. However, studies that do report effects of QE on exchange rates point in the same direction. Namely, that in some cases QE depreciates the currency and increases its volatility. This thesis finds some evidence that effects of QE on the exchange rate can be ambiguous and that it depends on the economic environment.
The results are also important for the current discussion about ‘currency wars’, where some authors claim that QE gives an unfair competitive advantage by weakening the currency (for the discussion, see e.g. Bergsten (2013) and Sibert (2010)). The message for
policymakers and investors is that there probably are some effects of QE, but that these are not always clear-cut.
This thesis has made several assumptions. One assumption made in the data is that expectations about GDP growth or CPI inflation only change upon their announcements. Nevertheless, the analysis in Section 4 shows that the figures used are helpful in obtaining more reliable estimates of the intervention size coefficient.
Another assumption is that the effects of the asset purchases are incorporated in the price of the exchange rate within the same week. This assumption of efficient markets is supported by the EMH. It is also one of the drawbacks of this thesis, because it remains possible that it takes time for the asset purchases to be incorporated in the price. This research is limited in the sense that it could not incorporate lags of the intervention size, because in many weeks the size of interventions is much the same. Incorporating those lags thus gives some econometrical issues. A challenge for future research is to work a way around this problem.
One other drawback of this thesis is the usage of weekly intervention size data. It remains possible that the outcome is affected by other variables during the week, which are not accounted for in the regressions. If possible, future research should also investigate the effects on a higher frequency (e.g. days/hours/minutes). In this way one can obtain a more direct measure of QE on the exchange rate.
18
References
Aguilar, J., Nydahl, S., (2000). Central bank interventions and exchange rates: the case of Sweden. Journal of International Financial Markets, Intitutions and Money, volume 10, pp. 303-322.
Benford, J., Berry, S., Nikolov, K., Robson, M., Young, C., (2009). Quantitative Easing.
The Bank of England Quarterly Bulletin, volume 2, pp. 90-100.
Bergsten, C. F., (2013). Currency wars, the economy of the United States and reform of the international financial system. Peterson Institute for International Economics, pp. 1-48.
Bodie, Z., Kane, A., Marcus, A. J., (2011). Investments and Portfolio Management (pp. 371-401). New York: McGraw-Hill/Irwin.
Bonser-Neal, C., Tanner, G., (1996). Central bank intervention and the volatility of foreign exchange rates: evidence from the options market. Journal of International Money and
Finance, volume 15, pp. 853-878.
Breedon, F., Chadha, J. S., Waters, A., (2012). The financial market impact of UK quantitative easing. Oxford Review of Economic Policy, volume 28, pp. 702–728. Butt, N., Churm, R., McMahon, M., Morotz, A., Schanz, J., (2014). QE and the bank lending
channel in the United Kingdom. Bank of England Working Paper No. 511. Retrieved from www.bankofengland.co.uk/research/Pages/workingpapers/default.aspx.
Christensen, J. H. E, Rudebusch, G. D., (2012). The response of interest rate to US and UK quantitative easing. The Economic Journal, volume 122, pp. 385-414.
Dominguez, K., (1998). Central bank intervention and exchange rate volatility. Journal of
International Money and Finance, volume 17, pp. 161-190.
Dropsy, V., (1996). Macroeconomic determinants of exchange rates: a frequency-specific analysis. Applied economics, volume 28, pp. 55-63.
Engle, R., (2001). GARCH 101: The Use of ARCH/GARCH Model in Applied Econometrics. Journal of Economic Perspectives, volume 15, pp. 157-168. Fatum, R., Scholnick, B., (2008). Monetary policy news and exchange rates: Do only
surprises matter? Journal of Banking & Finance, volume 32, pp. 1076-1086.
Fratzscher, M., (2006). On the long-term effectiveness of exchange rate communication and interventions. Journal of International Money and Finance, volume 25, pp. 146-167. Frenkel, M., Pierdzioch, C., Stadtmann, G., (2005). The effects of Japanese foreign exchange
market interventions on the Yen/US dollar exchange rate volatility. International
19
Glick, R., Leduc, S., (2012). Central bank announcements of asset purchases and the impact on global financial and commodity markets. Journal of International Money and
Finance, volume 31, pp. 2078-2101.
Joyce, M. A. S., Lasaosa, A., Stevens, I., Tong, M., (2010). The financial market impact of quantitative easing in the United Kingdom. The Bank of England Working Paper No.
393, volume 7, pp. 113-161. Retrieved from http://ssrn.com/abstract=1638986. Joyce, M. A. S., Tong, M., Woods, R., (2011). The United Kingdom’s quantitative easing
policy: design, operation and impact. The Bank of England Quarterly Bulletin 2011, volume 3, pp. 200-212. Retrieved from
www.bankofengland.co.uk/research/Pages/workingpapers/default.aspx. Joyce, M. A. S., Miles, D., Scott, A., Vayanos, D., (2012a). Quantitative easing and
unconventional monetary policy – An introduction. The Economic Journal, volume 122, pp. 271-288.
Joyce, M. A. S., Tong, M., (2012b). QE and the gilt market: a disaggregated analysis. The
Economic Journal, volume 122, pp. 348-384.
Kapetanios, G., Mumtaz, H., Stevens, I., Theodoridis, K., (2012). Assessing the economy-wide effects of quantitative easing. The Economic Journal, volume 122, pp. 316-347. Kenourgious, D., Papadamou, S., Dimitriou, D., (in press). Intraday exchange rate volatility
transmissions across QE announcements. Finance Research Letters.
http://dx.doi.org/10.1016/j.frl.2015.05.007.
Kim, S., Sheen, J., (2006). Interventions in the Yen-dollar spot market: A story of price, volatility and volume. Journal of Banking & Finance, volume 30, pp. 3191-3214. Mankiw, N., G., (2013). Macroeconomics (pp. 131-230). Houndmills: Palgrave Macmillan. Nelson, E., (2013). Key aspects of longer-term asset purchase programs in UK and US
monetary policy. Oxford Economic Papers, volume 65, pp. i92-i114.
Pilbeam, K., (2013). International Finance (pp. 152-156). Hampshire: Palgrave Macmillan. Sibert, A., (2010). Quantitative easing and currency wars. European Parliament, pp. 1-13.
Retrieved from http://eprints.bbk.ac.uk/9647/.
Stock, J., M., Watson, M. M., (2012). Introduction to Econometrics (pp. 558-710). Harlow: Pearson.
20
Appendix
Table 3: Akaike information criterion (AIC) for lag-selection
Dependent variable: Δln(PDEUR) n = 299
Lag Log likelihood p-value AIC
0 889.92 - -5.966 1 1122.34 0.000 -7.026* 2 1160.00 0.157 -6.850 3 1212.45 0.001 -6.772 4 1255.45 0.035 -6.632 5 1296.61 0.061 -6.479 6 1352.35 0.000 -6.424
The lowest AIC value indicates the best fit to the model and is denoted by *.
Figure 2: Autocorrelation test for MA(q) terms
21
Figure 3: Pound-euro exchange rate in the period 2008-2013
