Macroeconomic announcements : empirical test of the covered interest parity

(1)

Academic Year 2015/2016

Macroeconomic Announcements:

Empirical Test of the Covered Interest

Parity

Faculty of Economics and Business

Supervisor: Gabriele Ciminelli

Phil Etwi 10457771

This paper examines the effect of the monthly US non-farm payrolls release on the individual variables of the covered interest parity, consequently measuring the effect on the prolonged deviations of the CIP during the financial crisis. The study considers the change in each individual CIP variable separately and finds that there is indeed a period of time during the crisis for which CIP deviations were significantly large as a result of a breakdown of arbitrage. Moreover, the results imply negative CIP deviations during the period of crisis, relating to a potential arbitrage opportunity of borrowing funds from the US Government, whilst simultaneously lending to the UK Government resulting in positive earnings. This paper goes on to explain the fundamental underlying reasons behind the arbitrage market failure.

(2)

Statement

This bachelor’s thesis is written by Student Phil Etwi who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion

of the work, not for the contents. This paper was not submitted to any other examination committees.

Amsterdam

(3)

2 | P a g e

The Covered Interest Parity (thereafter CIP) is a theory established from the efficient market hypothesis. This latter notion explains that an efficient market is a market in which asset prices always fully reflect and incorporate all available information. Consequently, efficient markets leave no room for arbitrage opportunities. In the most simplistic form, arbitrage can be explained as borrowing money in one currency and lending it in another in order to take advantage of the interest rate differentials. Whilst simultaneously hedging foreign exchange rate risk (Griffoli & Ranaldo, 2011, p. 6). The CIP engages in a similar manner, when in equilibrium, the return of a domestic investment should be equal to that of a foreign investment considering both the spot and forward exchange rates. The reason for this being once the parity is found to be in state of disequilibrium, the ‘corrective’ arbitrage market, consisting of investors and speculators will profit from this arbitrage opportunity and hereby adjusting the market interest rates and/or exchange rates, leading to a restoration of the equilibrium. For instance, as soon as the CIP is hit by a shock causing disequilibrium, arbitrageurs immediately start investing within the country with the relatively higher return. Since this investment is country-specific, demand for the investment currency increases, given the fact that the arbitrageurs will initially need to change their currency into that of the investment currency. This excess demand may potentially lead to an appreciation of the investment currency and in turn the return to equilibrium of the CIP. In addition to this one, there are many more arbitrage adjustment mechanisms which, in theory, automatically restore the CIP. The general CIP can be illustrated as:

₍

₎

where and refer to the nominal interest rate of the home and foreign country on similar assets of a certain maturity, respectively, is the foreign exchange (thereafter FX) spot rate between the foreign currency and home currency at time t and is the FX forward rate contracted at time t for exchange at time t + s. Both the spot and forward FX rates are defined as the price of home currency in terms of foreign currency, throughout this paper. is the part of the equation which illustrates the distinction between the CIP and the Uncovered Interest Parity (thereafter UIP). The forward rate is included in the CIP, whereas the UIP does not cover

(5)

4 | P a g e

for arbitrage opportunities. Like any other theoretical notion, the CIP has several underlying assumptions. In order for the CIP to hold, there must be free capital mobility. There should not be any limitation for arbitrageurs to make investments across borders, meaning that the flow of money across borders should not be hindered. Secondly, the domestic and foreign assets must be perfect substitutes. This second assumption is critical to the CIP because, among other factors, the default risk determines the path of return of an investment. Once all aspects of the two assets are identical, a fluctuation in the return of the domestic asset should automatically also be reflected in the return of the foreign asset, maintaining the CIP. During times of crisis, it becomes more likely for disturbances of the CIP to arise from unanticipated shocks to the economy, such as an unexpected inflation rate or employment rate. Moreover, I believe that during volatile periods, financial markets and institutions may be less efficient possibly leading to short term deviations of the CIP. The sum of these shocks and a volatile period could potentially be the base of a risk-free investment opportunity for smart arbitrageurs.

The purpose of this paper is to empirically investigate whether or not the CIP may have held during the financial crisis of 2008. More specifically, I will examine the degree to which the deviations from short-term CIP observed in the monthly US Dollar/British Pound (thereafter USD/GBP) foreign exchange (thereafter FX) market are associated with factors initiating turbulence in the US financial markets. Daily observations will be made to during monthly 24-hour windows of the US employment report to spot any potential effects such an macroeconomic release has on the individual variables of the CIP. Whilst controlling for other important and relevant factors, particular emphasis will be put on how the surprise component of the monthly Bureau of Labor Statistics (thereafter BLS) Employment Report affects short term change in the variables of the CIP in the period of 2001-2015. Any change already anticipated by the market cannot serve as a shock to the economy and will therefore not affect the USD, the FX market or the bond market. To my knowledge, the CIP has not previously been tested before during a financial crisis in combination with exogenous shocks in the form of monthly unemployment reports. The remainder of the paper is organised as follows. Section 2 entails the critical review of existing literature on similar topical scientific papers and a brief review of the general knowledge and data on CIP gathered up until now by our scientific predecessor. Section 3 proposes the two main hypotheses of the research. Section 4 describes the data and research

(6)

method. Section 5 provides the framework for the results of the empirical analysis, followed by a discussion in Section 6. Finally, Section 7 concludes the paper.

2. Literature Review

According to Griffoli and Ranaldo, there is a distinction between secured arbitrage, which involves collateral against risk of counterparty default and unsecured arbitrage. For the remainder of this research paper, the focus will be on unsecured arbitrage. The reason for this being that the empirical analysis of the CIP will be based on short term (1 year until maturity) government bonds of the US and the UK, which in theory have a counterparty default risk which can be regarded as trivial.

Arbitrage is the solidifying component of financial markets. It links financial securities through pricing relationships and allows for the efficient functioning of markets. Since an efficient market is defined as one in which all asset prices fully reflect and incorporate all available information, it follows that agents should not be able to earn arbitrage profits by exploiting current information. Taylor (1989) conducted one of the first detailed analyses of covered interest arbitrage using high-frequency data surrounding various turbulent historical periods, including also a ‘calm’ period to act as control. Although this study was conducted a long time ago, he managed to obtain high-quality data from the BoE consisting of daily observations of the spot, forward and corresponding Eurosterling/Eurodollar deposit rates. Taylor’s main finding was that over time efficiency in the FX markets, measured by the reductions in the size, frequency and persistence of arbitrage opportunity increased. However, deviations from CIP still tend to rise during periods of uncertainty and turbulence and persist for some time before they are arbitraged away (Taylor 1989, p. 386). The reason behind this being that uncertainty and crisis bring along with them shocks, in turn causing inefficiencies.

A similar, but more recent study conducted by Batten and Szilagyi (2010) supports Taylor’s findings. They investigated the long-term CIP relationship between the USD and the Japanese Yen, as opposed to this empirical study which considers the relationship between the USD and the GBP. Their study uses a sample consisting of 5603 daily observations and is thereby the longest sample period yet testing the CIP. Unlike Taylor, Batten and Szilagyi (2010)

(7)

6 | P a g e

used London bank policy rates for the FX spot and forward, yet the results were nearly identical. Over the 25-year sample period, they found deviations from CIP equilibrium to peak and vary considerably during periods of economic instability. Nevertheless, Batten and Szilagyi too made the finding of CIP arbitrage opportunities becoming less frequent, if not nearly non-existent, in recent years (2010, p. 285). Consequently, attributing this improved efficiency of FX markets to the advancement of electronic trading (i.e. trading platform like Reuters D2000 and connected product-pricing systems).

As can be gathered from the preceding two scientific papers, the general view on short term CIP deviations is that when it comes to G10 currency markets, they have diminished significantly. This leads to us having to consider the underlying principle of the CIP, that arbitrage, not only in theory but also in practice, ensures that the parity holds and that such profits are minimized. For instance, Griffoli and Ranaldo (2011, p. 13) find that over the sample period, profits from arbitrage are small or even negative during the time surrounding the crisis. Even as we lead up to the first signs of crisis at the end of 2007, they observe that levels remain relatively small. However, as the financial situation worsens, they noticed that the Lehman Brothers bankruptcy coincides with what they call a breakdown of arbitrage. Griffoli and Ranaldo show that at their peak, arbitrage profits grew substantially, nearly reaching 400 basis points (thereafter bps) on an annualized basis. Moreover, profits remained high for several months. According to their research, the main responsible factor leading to such a significant breakdown of arbitrage was the lack of USD funding liquidity. They argue that it became increasingly problematic to raise capital for, amongst other things, arbitrage related purposes. Policies aimed at avoiding/containing the effects of the financial downturn led to a failure to re-balance the CIP condition.

Contradictory to the former statement, Coffey et al. (2009, p. 3) argue a slightly opposing view on the resolution of the arbitrage collapse surrounding the failure of Lehman Brothers. Although they both believe that the collapse of the arbitrage market was a result of two fundamental reasons. Firstly, their results imply that opportunities to exploit arbitrage were limited during the crisis due to funding constraints. Borrowing USD in the unsecured capital market became increasingly more difficult as a result of the simultaneous decrease of supply, and increase of global demand of USD. Secondly, they argue that the CIP deviations can also partly be explain by the increase in counterparty risk during the crisis. However, where the two views

(8)

begin to differ is that Coffey et al. do state that the fact that the Federal Reserve (thereafter FED) agreed to supply USD to foreign central banks, eased short-term USD-funding constraints and consequently lowered basis (deviation in CIP) by an average of 5 bps. Whilst Griffoli and Ranaldo (2011) believed that policy intervention actually further harmed the arbitrage market.

Besides the effects of an economic downturn, other factors may also cause disequilibrium to the CIP. Since the late 1980s, it has been known to economists that asset price movements resulting from regularly planned macroeconomic announcements provide information on the development of private sector expectations and how the economy, in turn, reacts towards these expectations (Faust et al. 2007, p. 1052). For instance, macroeconomic announcements tend to adjust financial variables and asset prices such as interest rates, exchange rates, stocks and bonds. Faust et al. (2007) studied the joint reactions of numerous financial markets in the US and Europe to US macroeconomic announcements, from a UIP perspective. In particular, they observed the high-frequency response of exchange rates and interest rates to 10 various macroeconomic data announcements, including CPI, GDP (advance release) and the non-farm payroll (thereafter NFP) releases. Faust et al. observed 14 years of high-frequency data, and found that unanticipated strong releases regarding real activity in the US lead to an appreciation of the USD in the short-run. Furthermore, the simultaneous movements of interest and exchange rates imply that this news either leads to a fall in the risk premium required for holding foreign assets or an expected net depreciation over the following decade. In some cases, the surprise component of the release may even lead to both scenarios (2007, p. 1066).

Anderson et al. (2007) tries to determine to what extent, if at all, macroeconomic releases of variables are incorporated into the pricing of stocks, bonds and the FX. They distinguish between two main fundamental views of financial markets. Firstly, the traditional view which is built upon the notion of efficient markets, suggests that asset prices constantly and instantaneously reflect movements in available information and underlying fundamentals. Conversely, others believe that asset prices and underlying fundamentals may be cyclically disconnected, implying a failure of arbitrage (Anderson et al. 2007 p. 252). They find that the exogenous shocks from the announcements produce conditional mean jumps. Consequently, explaining that high-frequency stock, bond and exchange rate dynamics are linked to underlying pricing fundamentals.

(9)

8 | P a g e

The literature review extensively explains the two different stands of existing literature. The initial method of workings simply looked at how the individual variables of the CIP change across times of tranquillity and times of economic turbulence causing the CIP equilibrium as a whole to break down. As time progressed, economists started to look at the issue from a different perspective. The more recent studies introduced an additional factor of macroeconomic announcements to observe how the individual variables of the CIP adjust as a result of these macroeconomic official announcements. This paper is structured in such a way to combine the two categories of existing literature with the objective of broadening the current knowledge on the reactions of the CIP variables, and in turn the deviations of the CIP entirely. It will test the individual variables of the CIP by means of, in addition to allowing for change across time, observing the change in the variables as a result of these macroeconomic official announcements, particularly the unanticipated segment of these releases.

3. Hypotheses

The central hypothesis that will be tested regards what we have named the error. The CIP-error is a simple rearranged form of the original CIP by taking all the variables over to the left hand side:

(

₎

₍

₎

Since the CIP-error is supposed to measure the variable change, it can be manipulated even further into equation (3) by considering the natural logarithm of the variables. The small letters in equation (3) denote the variables in natural logarithm. If the CIP holds, this error term must be zero or at least significantly close to this value. This rearranged form measures the impact of the disequilibrium caused by the macroeconomic shock on potential CIP arbitrage profits. We expect that profits from arbitrage will initially be negligible or even negative, as you would expect, up until the first signs of crisis in august 2007. From this point in time, we expect to see profits from arbitrage grow significantly as a response to different factors caused by the

(3)

(2)

(10)

financial crisis, which in turn exemplify the disturbances from the US macroeconomic announcements. Finally, we believe that the results will indicate a return to equilibrium from the end of 2009 onwards.

The secondary hypothesis which will be put to the test regards the sign (in terms of positive or negative order) of the CIP-error. We expect the estimated CIP-error to be significantly negative during the second period throughout. The reason behind this idea being that the macroeconomic shocks that the US economy will receive from the strong announcements will lead to lower policy interest rates. This simply means that the rate/price of borrowing from the US Government is cheaper than the identical rate/price which can be obtained from selling UK Government bonds. An arbitrage opportunity associated with such a negative CIP-error would be to borrow money from the US Government, change this amount of USD into GBP and finally lend this amount to the UK Government whilst hedging against the exchange rate. Clearly, an identical but reversed arbitrage strategy would be required for a positive CIP-error.

4 Dataset and Methodology

Following in the steps of Andersen et al. (2007), Faust et al. (2009), Fratzscher (2009) and Ciminelli (2015) I will conduct an event study estimating the reaction of the four CIP variables to real macroeconomic announcements. Consequently considering only the change of the variable of interest over a specific time-window around the US Employment Report release, and exclude all observations in which no release takes place. I will employ a 24-hour window surrounding the 8.30am New York Eastern Standard Time (thereafter EST), BLS announcement observing the change in the particular variables between 12.00pm before and 12.00pm after the announcement. Besides data on NFP, the BLS Employment Report also releases additional information on variables (i.e. unemployment rate and average hourly and weekly earnings growth, etc.). The fact that the report releases information on multiple variables at the same time imposes the risk of an empirical trade-off. Since these variables tend to be correlated to one another, including additional explanatory variables could possibly lead to high standard errors and thus inaccurate estimators. Whilst on the other hand, the use of merely one explanatory

(11)

10 | P a g e

variable may initially be seen as an invitation to omitted variable bias. However, since we are only interested in the surprise component (exogenous shock) of the announcement, there is no correlation, hence no empirical issues regarding the methodology.

4.1 Data

The sample of data gathered consists of one observation per month, covering a time range from January 2001 until January 2015, making a total of 169 month-observations. The dataset will be split up into three parts to represent the period before, during and ‘after’ (recovery period) the financial crisis. More specifically, the first period ranges from January 2001 until July 2007. The reason for this being that I follow (Baba & Packer, 2009) in their decision to use BNP Paribas’ announcement on the 9th of August 2007 as the starting point of the 2007-08 turmoil. On this date, BNP Paribas announced their decision to freeze redemptions for three of its hedge funds as a result of their inability to value them. This lead to a chain of events, one of which being the European Central Bank (thereafter ECB) injected overnight liquidity totalling 95 billion euros into the interbank market on the 9th and 10th of August 2007. The second dataset covers the most volatile period, ranging from August 2007 until September 2009. Here Baba & Packer (2009) are followed again in their decision to end their sample period at the month in which the Lehman Brothers filed for chapter 11 protection. Finally, the last period consists of the dataset covering the range October 2009 until January 2015.

The single explanatory variable in the regression is made up of the surprise component of the BLS Employment Report, and in particular the NFP release. NFP data are detailed estimates of the change of employment levels, released monthly, usually on the first Friday of every month at 8.30am New York EST. As mentioned above, the importance of the data is contained in the surprise component (shock), which is the part that remains after correcting for market expectations. Following precedent from existing literature, the proxy that will be used for market expectations is the economists’ median response to a survey asking economists their forecast.

Any change already anticipated by the market cannot serve as a shock to the economy and should therefore not affect the USD, the FX market, the bond market.

The dependent variables of the regression include data on the short-term government bonds yields of both the US and the UK. The government bond yields, which are assumed to be

(12)

risk-free, are used for the purpose of testing the CIP in order to avoid risk of default and the need for risk premiums. In addition to the interest rates, the spot and forward FX rates are used as dependent variables for the main regressions. For the purpose of this research, the FX rate, from a domestic perspective, is defined as USD/GBP, meaning that an increase in the exchange rate portrays a depreciation of the USD. The spot exchange rate refers to the exchange rate at that particular moment, t, whilst the forward exchange rate refers to a predetermined and contracted spot exchange rate at time t for a moment in the future, t+s. In the case of this research, all investments have a period of maturity of 1 year, unless stated otherwise. Therefore, the CIP for the purpose of this research looks as follows:

₍

₎

4.2 Methodology

As mentioned above, initially, I will estimate the reaction of the four CIP variables to the NFP surprise component using the ordinary least squares (thereafter OLS) method. This is done by composing four individual linear regressions for each of the three sample period all with a constant and the same single explanatory variable, NFP surprise ( ):

All these linear regressions are constructed robust standard errors in order to correct for heteroskedasticity. Again the small letters s and f denote the natural logarithm form of the corresponding variables in order to measure the actual percentage change. As can be seen from the last two regressions, the bid and ask quotes are taken into consideration in order to

(6)

(8)

(7)

(4)

(13)

12 | P a g e

incorporate transaction costs related to arbitrage. Once the results from the twelve independent regressions are in, we have an estimate of the coefficients of the respective

.

Each

represents to which extent the news component of the NFP employment announcement causes the dependent CIP variable to change as a result of the unanticipated deviation of the announcement from its expected value. The use of the 24-hour window decreases the risk that the regressions are distorted by any other potential variables which may also affect the four dependent variables. Therefore, it may be argued that for the purpose of the linear regressions, the change in the four dependent variables is solely measured by the shock, excluding any other potentially disturbing factors.

As mentioned above in the hypotheses section, the CIP-error is a simple variation to the original CIP. This rearranged form simplifies matters as it now measures the direct, natural logarithm change in each CIP variable. By merely filling in the respective

found from the linear regressions, the size of a potential CIP deviation during a time period will become immediately clear. Consequently, the estimation results obtained from the twelve individual linear regressions will be substituted into the CIP-error (3), providing an empirical answer to the central research question.

5. Results

The monthly mean and standard deviation (SD) of the CIP deviations are illustrated in

Figure 1 and Figure 2, respectively. The horizontal axis shows the years, whilst the deviation from CIP equilibrium is captured on the x-axis. Note that there is evidence supporting Taylor’s (1989) view of bidirectional arbitrage. Yet, during the financial crisis the negative deviations appear to dominate. The main observation that can be withdrawn from Figure 2 is that the greatest disruption takes place in 2008 to 2009 (period 2), as a result of the global financial crisis. Furthermore, it is also evident that the standard deviation of the CIP-error is substantially low and stable (approximately < 0.004) in the period before and during the recovery period of the crisis, namely period 1 and 3 of the sample.

(2)

(14)

Figure 1 Monthly mean deviation from CIP (2001-2015)

Figure 2 Monthly SD of deviation from CIP (2001-2015)

-0,01 -0,008 -0,006 -0,004 -0,002 0 0,002 0,004 0,006 0,008

Monthly Mean CIP-Error

0 0,002 0,004 0,006 0,008 0,01 0,012

SD

(15)

14 | P a g e

The estimation results are to a certain extent what was predicted by the hypothesis constructed beforehand. First of all, since the sample has been separated into three periods, all the coefficients tell a very specific story about that particular period in itself. As illustrated by Table 1, in both the first and third period, the βs associated to the change in US and UK government bond rates are positive, whilst those of the change in spot and forward FX rates are negative. What can be taken from this observation is that in essence, in the periods of economic tranquillity, it is estimated that shocks from unexpected macroeconomic employment announcements positively affect the policy rates associated with the government bonds with 12 month maturity. On the other hand, both the spot and forward FX rates have a negative reaction to the NFP surprise component. This result can be explained by the definition given to the exchange rate. As mentioned above, throughout this paper, all exchange rates have been assigned the definition as to be the price of USD in terms of GBP. An unanticipated, positive shock (increase in employment rate is greater than previously expected by the public) therefore decreases the exchange rate, based on our definition, hence an appreciation of the home USD currency. So to finalize, an expansionary BLS Employment Report has a strengthening effect on the USD relatively to the GBP.

Table 1 Regression Analysis: effect of macroeconomic shock (β)

Dependent Variable

Period 1 January 2001 until July 2007 (79 obs.) 3.683*** (0.568) ** 1.515*** (0.375) ** -0.203*** (0.064) ** -0.213*** (0.057)***

Period 2 August 2007 until September 2009 (26 obs.) 3.698*** (1.657) ** -0.031 (3.544) -0.130 (0.166) -0.037 (0.156)

Period 3 November 2009 until January 2015 (64 obs.) 0.542*** (0.168) ** 0.105 (0.563) -0.189*** (0.074) ** -0.240*** (0.093) **

(16)

From the coefficients of the first and third period together, all but one is found to be not significantly different from zero at a 1% statistical level, whereas for the second period, we solely find one of the four coefficients to actually be significant at the 1% statistical level. The second period from the data set is therefore seemingly the most chaotic period. This is evident from Table 1, which shows the substantially large standard errors for this period. Additionally, the change in the US government bond yield variable is the only one out of the four with a positive coefficient for β.

The following and final step in our analysis process is to substitute the estimation results from the linear regressions into the CIP-error in order to estimate the deviation from the CIP during the three sample periods. The outcome of this process is presented in Table 2. The first observation that can be made regards the nature of the estimates. The estimated CIP-error from the first (5.78E-05) and third period (0.000506) are both of positive order (>0), as opposed to that of the second period. This result implies the potential opportunity for a negatively directional arbitrage strategy. Such a strategy is in line with our second hypothesis and entails borrowing in the US market in order to lend in the UK market. Secondly, the estimate associated with period one is relatively negligible considering its small size. This finding supports our central hypothesis and implies that arbitrage opportunities during the period of January 2001 until July 2007 were miniscule, if even present at all. Most importantly, the absolute value of the estimated CIP-error in period two (0.00098) is greater than that of all three periods, however only a fraction larger than that of period three. Even though this estimate is closely followed by that of period three, this small difference may be explained by the fact that during the period from November 2009 until January 2015 the world as a whole was still partly recovering from the global financial crisis. The implications were that the ingredients for instability in the arbitrage market were still present.

(17)

16 | P a g e

Table 2 Estimated CIP-error

Period 1 January 2001 until July 2007 (79 obs.)

CIP-error = 5.78E-05 > 0 Period 2 August 2007 until September 2009 (26 obs.)

CIP-error = -0.00098 < 0 Period 3 November 2009 until January 2015 (64 obs.)

CIP-error = 0.000506 > 0

*, **, *** denotes significance at 10%, 5%, 1% statistical level, respectively. Constant included, using robust standard errors. Calculated according to formula (3) with the respective βs.

6. Robustness Checks

In this section, the robustness of the results is checked to the change of the start of the crisis. Previously, we used August 2007 as the start date and September 2009 as the end date of the global financial crisis. Critics may argue that the financial crisis was still on-going in September 2009. Therefore, for the purpose of the robustness check, we alter the end date of the crisis to December 2010, in order to avoid criticism, and run another series of identical linear regressions in order to verify the robustness of the results. At the end of September 2010, the Federal Reserve Bank of New York announced the closing of the recapitalization of American International Group, Inc. (thereafter AIG) and the full repayment of its loans to AIG. This termination eventually took place at the start of January 2011 and may, for the purpose of the robustness check, be considered the end of the global financial crisis. The results of the linear regressions are shown in Table 3 and seem to be robust to the alteration of the sample periods since the coefficients do not change severely. Moreover, the signs and the magnitudes of the estimated coefficients also seem to be plausible.

(18)

Table 3 Regression Analysis: effect of macroeconomic shock (β)

Dependent Variable

Period 1 January 2001 until July 2007 (79 obs.) 3.683*** (0.568) ** 1.515*** (0.375) ** -0.203*** (0.064) ** -0.213*** (0.057)***

Period 2 August 2007 until September 2009 (26 obs.) 2.801*** (1.161) ** 0.016 (2.323) -0.141 (0.117) -0.145 (0.126)

Period 3 November 2009 until January 2015 (64 obs.) 0.281*** (0.151) ** 0.345 (0.587) -0.197*** (0.081) ** -0.197*** (0.099) **

*, **, *** denotes significance at 10%, 5%, 1% statistical level, respectively. Constant included, using robust standard errors

7. Discussion

Our empirical results steer us towards the direction of two fundamental implications. First of all, as was hypothesized in the initial stages of the research, the estimated CIP-error has been found to be significantly larger in the second period, relative to the other two periods. The underlying explanation being, that unanticipated macroeconomic shocks create a greater disturbance of the CIP equilibrium during times of financial instability as compared to financially stable periods. This result provides sufficient evidence to reject the null hypothesis assigned to our first proposition. Even though the results indicate only a small difference between the CIP deviation of period two and three, the substantial difference in absolute value of the deviation between that of the first and second period suggests the following progressive cycle. Initially in period one, when there are no signs whatsoever of a potential financial downturn, the global arbitrage market performs according to theoretical expectations. The presence of infinite capital mobility and constant perfect substitutability, in terms of similarity in riskiness and liquidity of the

(19)

18 | P a g e

investments, provides a stable setting for corrective arbitrage. Next, leading into the second period of the sample, labelled the turbulent period, we find the underlying assumptions of arbitrage to breakdown. Finally, as we see the economy recover from the recession, we see that the CIP deviations begin to diminish, again somewhat in line with the primary hypothesis. There is however an overlooked novelty pointed out by the result which was not taken into consideration during the construction of the central hypothesis. This proposition wrongly expected the breakdown in arbitrage leading to prolonged CIP deviations to have been recovered somewhere in the third period. As expressed by the results, the CIP-error did somewhat diminish, yet the deviations were still of considerable size relatively to the first period. On the other hand, the secondary hypothesis regarding the negative CIP-error expectation during the second period did however fully coincide with our estimation results. As initially expected, the arbitrage strategy associated with the CIP deviation consisted of borrowing funds from the US Government, whilst simultaneously lending to the UK Government resulting in positive earnings.

Existing literature in this field of research have silently agreed upon two main fundamental explanations behind the breakdown of arbitrage and consequently, significant and persistent CIP deviations. The primary reason mentioned by the likes of Baba & Packer (2009), Coffey et al. (2009) and (Griffoli & Ranaldo, 2011) refers to the liquidity constraint. At this point it is clear that the stream of capital is essential to arbitrage in order for it to operate properly. As soon as arbitrageurs spot an opportunity to exploit the interest rate differentials and thereby make a risk-free profit, they must instantly have a line of credit in order to actually invest in the risk-free arbitrage strategy. Leading up to the first signs of crisis, at the end of 2007 going onwards, banks and other financial institutions started becoming aware of the severity of the economic situation we were about to enter. As a result of the slowing economy, Central Banks began to inject money into the economy, by purchasing eligible public and private sector assets, with the aim of providing an additional stimulus to nominal spending in order to meet the inflation target (Benford et al. 2009, p. 90). This procedure is known as ‘quantitative easing’ (thereafter QE). However, in hindsight, it can evidently be disputed that the required effect was not truly achieved. Banks issued fewer loans in spite of the increase in non-borrowed reserves provided by central banks through QE policies. Banks became more risk averse during this period and decided that it might be safer to deposit these funds back at the CB’s deposit facility.

(20)

As soon as such a constraint on liquidity presents itself, the risk of arbitrage breakdown, and therefore prolonged CIP deviations, comes into existence.

The second possible factor explaining the collapse of CIP arbitrage refers to risk. Although Griffoli & Ranaldo (2011) mention three types of risk, namely contract, rollover and counterparty default risk, as the cause of the failure in the arbitrage market, this study merely recognizes contract risk since it deals with government bonds. Contract risk refers to the default of the trader's FX forward counterparty during the term of arbitrage (Griffoli & Ranaldo, 2011, p. 15). Yet, since risk tends to affect both currencies equally, it is unlikely that its conrtibution to the market failure was significant, relatively to the liquidity constraint.

8. Conclusions

This paper has empirically investigated the effect of macroeconomic announcements on the CIP during the financial crisis. The data set chosen for this study ranges from January 2001 to January 2015, consisting of 169 daily observations. The results from the empirical analysis have two fundamental implications. Namely, the CIP -error tends to significantly increase during times of financial distress, allowing for potentially prolonged and risk-free arbitrage opportunities. The reason for the presence of such arbitrage opportunities, results from the fact that the NFP surprise component of the macroeconomic release, creates unforeseen shocks which disrupt the arbitrage market causing significant deviations in the CIP. Furthermore, the results conclude that during the peak of the global financial crisis, the deviation in CIP was negative involving an arbitrage strategy of borrowing in the US market whilst lending in the UK market. Finally, the central factor for the length of the CIP deviations is the liquidity constraint which significantly contributed to the failure of the corrective arbitrage market.

(21)

20 | P a g e

Bibliography

Andersen, T. G., Bollerslev, T., Diebold, F. X., & Vega, C. (2007). Real-Time Price Discovery in Global Stock, Bond and Foreign Exchange Markets. Journal of International Economics, 73, 251–277. Baba, N., & Packer, F. (2009). Interpreting Deviations from Covered Interest Parity During the Financial

Turmoil of 2007-08. Journal of Banking & Finance, 33, 1953-1962.

Batten, J. A., & Szilagyi, P. G. (2010). Is Covered Interest Parity Arbitrage Extinct? Evidence from the Spot USD/Yen. Applied Economics Letters, 17(3), 283-287.

Benford , J., Stuart, B., Nikotov, K., Young , C., & Robson, M. (2009). Quantitative Easing. Bank of England

Quarterly Bulletin, 49(2), 90-100.

Ciminelli, G. (2015). On the Signaling Channel of Monetary Policy and the Sensitivity of. MPhil Thesis

Graduate Program in Economics of the Tinbergen Institute, 1-76.

Coffey, N., Hrung, W., & Sarkar, A. (2009). Capital Constraints, Counterparty Risk and Deviations from Covered Interest Rate Parity. SSRN Working Paper, 1-47.

Faust, J., Rogers, J. H., Wang, S.-Y. B., & Wright, J. H. (2007). The High-Frequency Response of Exchange Rates and Interest Rates to Macroeconomic Announcements. Journal of Monetary Economics,

54, 1051-1068.

Fratzscher, M. (2009). What Explains Global Exchange Rate Movements. European Central Bank, 1-33. Griffoli, T. M., & Ranaldo, A. (2011). Limits to Arbitrage During the Crisis; Funding Liqudity Constraints

and Covered Interest Parity. SSRN Working Paper, 1-31.

Taylor, M. P. (1989). Covered Interest Arbitrage and Market Turbulence. The Economic Journal, 99(396), 376-391.

Macroeconomic announcements : empirical test of the covered interest parity