Excerpt

## List of Contents

List of Tables

List of Figures

List of Abbreviations

Introduction

1. Literature Review

2. Data

2.1 Data Description

2.2 Volatility Clustering

2.3 Multiple Breakpoint - Bai - Perron test

2.4 Test for stationarity - Unit Root

2.5 Correlogram

2.6 Descriptive Statistics

3. Methodology

3.1 ARCH

3.2 GARCH(1,1)

3.3 GJR GARCH(1,1)

3.4 Exponential GARCH(1,1)

3.5 Integrated GARCH(1,1)

3.6 Volatility Forecasting

4. Empirical Results

4.1 Results of the ARCH Effects testing

4.2 Results of the GARCH(1,1) model

4.3 Results of the GJR GARCH(1,1) model

4.4 Results of the Exponential GARCH(1,1) model

4.5 Results of the Integrated GARCH(1,1) model

4.6 Results of Volatility Forecasting

Conclusion

References

Appendix

## List of Tables

1. Table: Bai - Perron - Multiple Breakpoint Test - Global L breaks vs. none

2. Table: Test for heteroscedasticity - Augmented- Dickey Fuller test

3. Table: Correlogram - Test for autocorrelation

4. Table: Descriptive Statistics

5. Table: Results of the ARCH test

6. Table: Results of the GARCH(1,1) model

7. Table: Results of the GJR GARCH(1,1) model

8. Table: Results of the EGARCH(1,1) model

9. Table: Results of the IGARCH(1,1) model

10. Table: Results of the ARCH - LM test

11. Table: Forecast error statistics

## List of Figures

1. Figure: Volatility Clustering

2. Figure: Conditional Variance - GJR GARCH(1,1)

3. Figure: Dynamic forecast (6 months ahead) - USDEUR - GJR GARCH(1,1)

4. Figure: Dynamic forecast (6 months ahead) - GBPEUR - GJR GARCH(1,1)

5. Figure: Dynamic forecast (6 months ahead) - JPYEUR - GJR GARCH(1,1)

6. Figure: Dynamic forecast (6 months ahead) - INREUR - GJR GARCH(1,1)

## List of Abbreviations

Abbildung in dieser Leseprobe nicht enthalten

## Abstract

This research explores the impact of foreign exchange rateinterventions on the behaviour of exchange rate returns and theirvolatility. In 2017, monetary interventions are actual as they have neverbeen before. However, they have been criticised for not being effectiveand existing empirical evidence is mixed. The present research appliesmodels from the *GARCH framework,* while using monthly data from2001 to 2017 on the *USDEUR, GBPREUR, JPYEUR* and *INREUR* rate.The results indicate that monetary interventions have a higher impacton developed country currencies than on emerging markets currencies.In addition, evidence is found that the volatility increased after thefinancial crisis.

## Introduction

The Federal Reserve Bank. The European Central Bank. The Bank of England. The Bank ofJapan and recently the Reserve Bank of India, as well: In recent years, central banks became acentre of attention in the media. In June 2017, Bloomberg titles ‘ *Central Banks Won't End theParty Soon ’*, implying that central banks will continue foreign asset purchases in order tostabilise the global monetary system. The Reserve Bank of India intervenes in order to keep therupee stable and exports competitive to guarantee constant economic growth as the EconomicTimes (2017) reports. However, in recent years central banks have been publicly criticized asbeing less powerful. In addition, the Financial Times (2016) published that the ‘ *power of central banks is waning* ’ and their monetary interventions have a weaker effect on foreign exchangerates and their volatility than they had 30 years ago.

The IMF (2003) defines monetary interventions as the selling or purchasing of foreigncurrency assets by central banks in order to influence the exchange rate. In accordance toDominguez (1998) central bank interventions have always been controversial and criticised asnot having an impact on the foreign exchange rate. However, a great part of existing literatureand economic decisions have been made based on Milton Friedman’s (1953) argument thatmonetary interventions increase the volatility of currency rates. How much of that is true?

The increasing amount of FOREX rate intervention in order to stabilise the globalmonetary system after the financial crisis in 2008 provides the motivation to absorb newinsights on how monetary interventions impacts currency rates. This research analyses theeffect of monetary interventions by the respective central bank in the medium-term and usestime series data for the *USDEUR*, *GBPEUR*, *JPYEUR* and *INREUR* rate. By applying the *GARCH(1,1)*, *GJR GARCH(1,1)*, *EGARCH(1,1)* and *IGARCH(1,1)*, this research providesevidence on how sales and purchases in foreign currency assets influence the exchange ratereturn under consideration of the financial crisis. Furthermore, this study researches thebehaviour of the FOREX volatility after the financial crisis by applying a dummy variable inthe variance equation.

It is expected that FOREX intervention influence the exchange rate of developed countries more effectively, since *EMs* are dependent on many more macroeconomic factors. Furthermore, the plummeting financial markets during the crisis in 2008 caused overall high volatility, therefore it is predicted that the dummy variable will have a positive significant coefficient implying that the volatility of each currency increased.

The first chapter provides further definitions on FOREX intervention and introduces existing empirical evidence. The second section gives information on the data and tests whether the requirements for applying GARCH models are fulfilled. Further, the Methodology part proposes the mean and variance equation for each GARCH model. Moreover, the results section analyses and discusses the test results, defines the most efficient GARCH model and performs a six month ahead forecast the FOREX returns and its volatility. The conclusion summarises the most significant results and provides motivation for further research.

## 1. Literature Review

The following part reviews previous research on the correlation between monetary interventions and FOREX rates. Additionally, various approaches are discussed critically and it is outlined how the present research fits within existing literature.

After the breakdown of the Bretton Wood^{1} system in 1973, the major *DMs* stopped pegging their currency rate to the *USD*, thereby fallowing a period of floating currency exchange rates in accordance to Doroodian and Caporale (2001). Schwartz (1996) published a paper analysing the post- Bretton Wood period and the increasing amount of monetary interventions by G3 nations in order to prevent the depreciation of the dollar.^{2}

Hueffner (1983) defines monetary interventions as ‘purchases and sales by central banksof foreign exchange assets. As stated by Reinert, et al. (2009) monetary interventions have thepurpose to keep or target a predetermined foreign exchange spot price. They add thatinterventions are also used to ‘directly alter macroeconomic conditions’. Additionally, asoutlined by Burda and Wyplosz (2017), interventions are distinguished between sterilized andnon-sterilized interventions. Sterilized interventions involve changes of foreign reserve assetswithout changing the foreign monetary base. On the opposite, non-sterilized interventionschange the foreign monetary base by selling and purchasing foreign currencies. Furthermore,Hueffner (1983) adds that interventions differentiate ‘between active and passiveinterventions’. In case of an active intervention a central bank buys or sells foreign assets toinfluence the currency rate, whereas during a public intervention the initial initiator are clientsand the central bank only executes the transaction (Arnon, Weinblatt, & Young, 2010).^{3} Furthermore, it is important to distinguish between interventions that are communicatedpublicly and interventions that are executed secretly (Arnon, Weinblatt, & Young, 2010). Mostempirical evidence in literature considers both kinds of interventions, since they are notdifferentiated in the official data of central banks (Dominguez, Central bank intervention andexchange rate volatility, 1998).^{4} However, according to Edison (1993), researchers had alwaysdifficulties to receive adequate data of interventions, since central banks have been reluctant to publish intervening actives. Neely (2000) introduces the most successful strategy that researches have been using in literature. They use the official data on foreign exchange reserves, which include securities and bank deposits of foreign currencies. Dependent on the central bank, the reserves are published in various time intervals.

Starting from the post- Bretton Wood period, a first series of studies was published covering the impact of monetary intervention on currency exchange rate levels. Obstfeld (1983), Gartner (1987) and Kearney & MacDonald (1986) were one of the first researchers that analysed the effect of sterilized interventions on FOREX and provided empirical evidence by using the changes of foreign exchange reserves as a proxy for interventions.^{5}

The evidence has been the basis for further analysis on central bank’s impact onvolatility of currency rates. Baillie & Humpage (1990) were one of the first researchersanalysing the effect of internventions on the volatility of exchange rates by applying the *GARCH * model developed by Bollerslev (1986). They observed that G3 countries intervenedmainly to maintain current exchange rate levels during the post- Louvre^{6} period. However, theywere not able to prove that ‘the intervention successfully influenced currency rates’. Likewise,Almekinders & Eijffinger (1994) did not find a significant impact on the exchange ratevolatility caused by central bank interventions while analysing the *USDDM* rate. Nevertheless,they found out than an appreciation of the daily conditional variance lead to interventions bythe Deutsche Bundesbank and the *FED*. Tanner and Bonser-Neal (1996) followed a differentapproach, by estimating ‘ex ante volatility using the implied volatilities of currency optionprices’ instead of the *GARCH* modelling. While analysing the *USDDM* and *JPYUSD* exchangerate, they were also only able to find little proof of correlation between central bankinterventions and exchange rate volatility. Similarly, by using the same approach as Tanner andBonser-Neal, Aguilar & Nydahl (2000) found only a weak confirmation of the impact forinterventions on exchange rate volatility. Likewise, Frenkel, et al. (2001) analysed the monetaryinterventions by the ECB and could only determine weak short-term effects on the FOREXrate.

Dominguez and Frankel (1993a) were the first releasing a study that found evidence ofsignificant effects on exchange rate volatility through monetary interventions by analysingweekly intervention data by the Deutsche Bundesbank and Federal Reserve and its impact on the US Dollar/ Deutsche Mark rate for the period from 1987 to 1989. Additionally, Almekinders (1995) found out that monetary interventions not always lead to an increase in return byanalysing daily data of the *DMUSD* rate from 1985 to 1988. Likewise, Dominguez (1998) foundfurther evidences and confirms that usually none-reported interventions lead to an increase involatility, whereas reported interventions have little impact on the volatility of currency rates.Moreover, Doroodian and Caporale (2001) analysed exchange rate interventions on theJPYUSD rate by using daily data from 1985 to 1997 and also applied the *GARCH* model. Incontrast to previous researches, Doroodian and Caporale (2001) concluded that monetaryinterventions are related to higher uncertainty of currency exchange rates.

Over the years the *GARCH(1,1)* model established itself as a well-serving framework for analysing volatility as pointed out by Bénassy-Quéré, et al. (2002). However, according to Andersen, et. at. (2009), in recent years the model has been criticized for the assumption of stationarity. Moreover, Bénassy-Quéré (2002) state that ‘the effect of a volatility shock vanishes over time at an exponetial rate.’

As can be withdrwan from previous literature, most studies used high-frequencay ordaily data to analyse foreign exchange rate volatility in the short-term. Only fewer studiesanalysed the impact of interventions on exchange rate volatility in the medium term. Behera, etal. (2008) used monthly data within the *GARCH)* model and found out that interventions by theRBI reduced the volailtiy from April 1995 to December 2006. On the contrary, Olowe (2009)studied the volatility of the *Nigerian Naira/ USD* during the period January 1970-December2007 by using various *GARCH* models. He concluded that the exchange rate volatility ispersistent against monetary interventions. Likewise, Khan, et al. (2012) applied symmetric andasymmetric *GARCH* models for the period January 2001 to December 2009. They foundvolatility clustering in the results of the *GARCH-M*, whereas the *EGARCH* found asymmetriceffects in the exchange rate return. In support of Khan, et al. (2012) study, Elsherif (2016)discovered evidence of volatility clustering while applying the GARCH model on interventionsof Egypt’s central bank and the *EGPUSD*. However, he found none persistent volatility shockswithin the model.

In conclusion, it is obvious that there has been done extensive research about the impactof CB interventions on FOREX returns and its volatility. Various *GARCH* models, time framesand geographical regions have been analysed during the last three decades. Some studiesdelivered significant evidence that prove an impact in the currency exchange rate volatility afterMI. Since the results are not consistent and very mixed, the effectiveness of interventions onFOREX rates needs to be re-examined, which provides the motivation for the present study.

## 2. Data

This section examines the financial data used in this research and tests, whether it fulfils the requirements to proceed with *GARCH* model tests.

### 2.1 Data Description

The present empirical study includes monthly data sets from the United Kingdom, the United States, Japan, Eurozone and India in order to analyse the volatility on *GBPEUR, USDEUR, JPYEUR and INREUR*. The time series period is based on availability, starting from January 2001 to May 2017 and includes 197 observations. It is sourced from the central bank responsible for the respective currency.^{7} In alignment to previous studies^{8}, the macroeconomic variables of each country used in this research are:

Exchange rate return

Foreign Exchange Reserves

The present research uses returns instead of the absolute currency spot price. That is similar toCampell, Lo and MacKinlay (1997) who suggest two reasons for using returns: (1) Returnssummarise the investment opportunity of an asset and (2) returns have less attractive statisticalproperties. The exchange rate return is estimated through the monthly spot price using thefollowing formula:

Abbildung in dieser Leseprobe nicht enthalten

Where *S* determines the monthly spot price of either *GBPEUR, USDEUR, JPYEUR or INREUR* and *t* is equal to the period of the respective spot price.

Based on Neely (2000) and Behera, et al. (2008) the intervention variables of the presentmodel are derived from the foreign exchange reserves *(FER)* by estimating the absolutemonthly change. Therefore, a positive change in net intervention is equal to a purchase of FER,whereas a negative change in net intervention equals a sale. The concept is expressed in thefollowing:

Abbildung in dieser Leseprobe nicht enthalten

The present model uses the variables of purchase [Abbildung in dieser Leseprobe nicht enthalten] and sales [Abbildung in dieser Leseprobe nicht enthalten] in million euros for each country. Central bank variables of the base currency are expressed as [Abbildung in dieser Leseprobe nicht enthalten] and[Abbildung in dieser Leseprobe nicht enthalten]^{9} Since the quote currency is always *EUR*, interventions by the *ECB* are expressed as [Abbildung in dieser Leseprobe nicht enthalten]and [Abbildung in dieser Leseprobe nicht enthalten] .therefore include securities and bank deposits of the *ECB,* the *BOE,* the *BOJ,* the *FED* and the *RBI* in foreign currencies.

### 2.2 Volatility Clustering

Brooks (2008) states that volatility clustering is a fundamental characteristic of financial data.Asteriou and Hall (2015) summarise volatility clustering by the size of prize changes. Theyexplain further that large price changes are followed by price changes of similar size andtherefore obtain higher volatility. In contrast, small price changes are followed by price changesof similar size while the volatility is low.^{10} Figure 1 illustrates the deviation of all currency ratereturns and proves that the time series data used in this study exhibits volatility clustering.

*Figure 1: Volatility Clustering*

Abbildung in dieser Leseprobe nicht enthalten

*Source: Estimated by author trough Eviews 9.5*

All currency returns indicated a relative high level of deviation until 2003, which was causedby the early 2000s recession and under consideration of the dot-com bubble burst. As the chartof *INREUR* illustrates, the returns deviated much more than the currencies from *DMs*.^{11} *USDEUR, GBPEUR and JPYEUR* experienced a period of low volatility and stable marketsuntil 2007, when the financial crisis started to impact. During the period from 2007 to 2010, allcharts indicated high volatility, which decreased in 2011 to a moderate level. However, thatwas not the case for returns on *INREUR*, since the slowdown of economic growth in Indiacaused higher volatility as pointed out by Riley & Yan (2013). Until the estimation of thepresent research, *INREUR* exhibited a high level of volatility, except for a short period in 2014. *DMs* recovered moderately from the volatility during the financial crisis and indicated lowlevels of volatility during 2011 until 2015. In 2015, *DMs* and *EMs* experienced periods of highreturn deviation, which was caused by lower commodity prices and the discontinuation ofquantitative easing by the fed (De Bock & De Carvalho Filho, 2015).

### 2.3 Multiple Breakpoint - Bai - Perron test

Like Vadivel (2009) and Behera, et al. (2008), the present model includes a dummy variable based on the existence of volatility breakpoints (k). This study uses the multiple break point test by Bai - Perron (1998) to investigate for volatility breaks in the spot price return.

*Table 1: Bai - Perron - Multiple Breakpoint Test - Global L breaks vs. none*

USDEUR GBPEUR JPYEUR INREUR

Abbildung in dieser Leseprobe nicht enthalten

*Source: Estimated by author trough Eviews 9.5.*

The multiple breakpoint test verified that the *DMs* had their most significant volatility break around the financial crisis from 2007 to 2009. However, the Indian Rupee also had a breakpoint during the financial crisis in July 2008, but the most significant breakpoint was in September 2013. Riley and Yan (2013) point out that the *INR* depreciated, which was caused by a high oil price, high interest rates and a slowdown in the Indian economic growth. The dummy variable used in this research is defined as:

In the same manner as Behera, et al. (2008), the dummy variable ()\ ) is equal to zero during the pre-crisis period and one during the post-crisis period.^{12}

### 2.4 Test for stationarity

The present study uses the Augmented-Dickey Fuller *(ADF)* test to verify the stationarity of thetime series, based on the DF- test developed by Fuller (1976) and Dickey-Fuller (1979). TheADF test proves the presence of a unit root in a data stream *(y)* under the null hypothesis (,Hkand uses a number of lags (p) for the dependent variable. It follows the t- distribution andincludes the critical values of *-3.464101*, *-2.876277* and *-2.574704* on a *1%*, *5%*, and *10%* level,respectively. As Brooks (2008) states, the Akaike Info Criterion *(AIC)* suggests the bestnumber of lags *(p) * in order to maximise the efficiency of GARCH model tests. Table 2 providesthe results of the ADF test.

*Table 2: Test for stationarity - Augmented- Dickey Fuller test*

Abbildung in dieser Leseprobe nicht enthalten

*****Automatic- based AIC, maxlag = 5*

** Significant at 1% level, ** at 5%t level, *** at 10% level. Note: Values are rounded. Source: Estimated by author trough Eviews 9.5*

As can be withdrawn from the table above, the test statistic of each test variable is more negativethan the critical value hence each return series of the currency pair are stationary at a 1% level.Therefore, the present *ADF* test rejects the null hypothesis (,Hk and shows that the data stream

differs from having a unit root. Furthermore, based on the Akaike Info Criterion *(AIC)* only the return of *GBPEUR* showed evidence of lags.

### 2.5 Correlogram

In accordance with further literature and previous studies, Agung (2011) suggests using acorrelogram in order to analyse the time series data for autocorrelation (AC). Testing the nullhypothesis j,Hk of no autocorrelation and the alternative j,Ik of present autocorrelation. *Table *

*3* provides evidence of the test results.

*Table 3: Correlogram - Test for autocorrelation*

USDEUR GBPEUR INREUR JPYEUR

Abbildung in dieser Leseprobe nicht enthalten

**Significant at 1% level, ** at 5% level, *** at 10% level. Note: All values are rounded. Source: Estimated by author trough Eviews 9.5*

The autocorrelation values are close to zero throughout the model and the probability is insignificant for all currency pairs. Therefore, the null hypothesis fails j,Hk to be rejected, implying that the data for returns of all currency pairs has no autocorrelation.

### 2.6 Descriptive Statistics

In addition, Asteriou & Hall (2007), Vogelvang (2005), Brooks (2008) and Tsay (2010) pointout to test the financial data series for skewness and kurtosis matching a normal distribution.The table below shows the descriptive statistics of the data sets including the Jarque - Bera testfor normality.^{13}

*Table 4: Descriptive Statistics*

USDEUR GBPEUR INREUR JPYEUR

Abbildung in dieser Leseprobe nicht enthalten

**Significant at 1% level, ** at 5% level, *** at 10% level. Note: All values are rounded. Source: Estimated by author trough Eviews 9.5.*

*Table 4* shows that the returns on *USDEUR* and on *JPYEUR* are skewed right, whereas the returns on *GBPEUR* and *INREUR* are skewed left. Furthermore, based on Westfall (2014) returns on *GBPEUR* and on *JPYEUR* are highly skewed, while the returns on *USDEUR* and *INREUR* are approximately symmetric. In addition, grounding on the information provided by Westfall (2014), and on the table above, the returns on currency pairs are leptokurtic. Further, according to Jarque and Bera (1980), all returns are non-normal distributed, since they reject the null hypothesis j, k for normality.

Concluding, the time series data shows evidence of (1) non-autocorrelated returns, (2) non-normal distributed returns and (3) positive auto correlated return squares. Therefore, the financial data used in this research fulfils the requirements to perform ARCH model tests based on Brooks (2008) and Tsay (2010).

**[...]**

^{1} Based on Steil (2013) the Bretton Wood System was international monetary agreement from 1944 in order to support the post-war economy and international cooperation.

^{2} Schwartz (1996) identified three main periods of central bank interventions: (1) From October 1974 to March 1975 the FED cooperated with the BB and the SNB. (2) From Septmeber 1977 to Decemeber 1979 the FED intervened togerther with the BB and the BOJ. (3) In 1985 the G5 nations started to intervene against the overvalued dollar in order to fulfill the Placa Accord.

^{3} Dominguez and Frankel (1993a) name the domestic governement or foreign governements as potential initiators in the case of passive interventions.

^{4} Klein (1993) provides evidence that 72 percent of the total monetary interventions were published in the Wall Street Journal during 1985 - 89.

^{5} Further publications about the impact of monetary interventions on the foreign exchange rate were published by Henderson & Sampson (1983), Jurgenson (1983), Henderson (1984), Loopesko (1984), Obstfeld (1990), Dominguez (1990, 1992), Frankel (1992), (Eijffinger & Gruijters, 1992), Dominguez & Frankel (1993a,b,c), Osterberg & Baillie (1997) and Edison H. (1993).

^{6} The Louvre Accord was a monetary agreement aiming to stabilize the foreign currencies and prevent the depreciation of the US-Dollar (Bakker, et al. 2013).

^{7} Data from the ECB, the BOE, the BOJ, the FED and the RBI has been used.

^{8} Behera, et al. (2008), Vadivel (2009) and Elsherif (2016) use the same variables.

^{9} In this mode, central banks of the quote currencies are the FED, the BOJ, the RBI and the BOE.

^{11} In accordance to Jochum and Kodres (1998), EMs usually show higher volatility than developed countries, which is the result of political, economic and social instability.

^{12} In order to estimate the dummy variable for the variance equation, the Bai - Perron test was performed with squared spot price returns of each currency pair.

^{13} Based on Jarque and Bera (1980), the test for normality examines a data set under the null hypothesis (,Hk that it is normally distributed.

- Quote paper
- Marvin Arras (Author), 2017, Intervention and Foreign Exchange Volatility. New evidence from Developed and Emerging Countries, Munich, GRIN Verlag, https://www.grin.com/document/387544

Publish now - it's free

Comments