Persistent deviations from the CIP condition

(1)

Persistent deviations from the CIP

condition

by Erik Gunnink

Supervisor: Artem Tsvetkov Student No: 2206005 University of Groningen

The Netherlands June 8, 2017

(2)

Abstract

This paper takes a closer look at the deviations from the covered interest rate parity (CIP) for the five major currencies using daily time series data from March 7 2007 until December 30 2016. The results show persistent deviations from the CIP condition for the EUR/USD, GBP/USD, AUD/USD and JPY/USD. The empirical evidence reveals that these CIP deviations are related to liquidity risk, credit risk and counterparty risk. The results show that money market risk was especially important during the global financial crisis, whereas counterparty risk played an important role during the European debt crisis.

(3)

1 Introduction

“I don’t throw darts at a board. I bet on sure things. Read Sun-tzu, The Art of War. Every battle is won before it is ever fought” (Gordon Gekko). Arbitrage is an example of a sure thing when it comes to trading. Arbitrage normally emerges as a result of inefficiencies in the market. An example of a no-arbitrage condition is the Covered Interest Rate Parity (CIP), which states that the exchange rate forward premiums (discounts) offset interest rate differentials between two countries. This means that investors are neutral toward funding in different currencies. To interpret it differently, a Japanese bank with assets in euros can choose to fund the bank through euros. But it can also use its own deposit currency, the Japanese yen, to buy euros on the spot market and a forward contract to hedge the exchange rate. The CIP condition promises that both options give the same cost for the bank. If this CIP does not hold, so when there are deviations in the parity, this will mean that there are arbitrage opportunities.

To take advantage of CIP deviations you will need to trade in the currency market. The foreign exchange markets (Forex) is the largest and most liquid market in the world with a global average daily turnover of $5.1 trillion (Bank of International Settlements, 2016). On this market all available currencies can be traded 24 hours a day, where the US dollar is by far the most traded. Forward exchange rates are based on interest rate differentials between the two currencies traded, where the currency with the lower interest rate will trade on a forward premium and the currency with the higher interest rate will trade on a forward discount. Meaning that for a premium (discount) the forward exchange rate will be higher (lower) than the spot rate.

Baba and Packer [2009b] record disturbances in the FX markets across some of the major currencies since the beginning of the financial turmoil. They report problems in dollar funding markets and increase in counterparty risk as some of the major drivers causing these problems. This is due to deterioration in funding liquidity of banks and the ability to liquidate certain asset positions.

We will show that the CIP condition has been violated among some of the most used currencies since the global financial crisis. In the absence of transaction costs, the following equation hold for the CIP condition to be true.

Ft,t+s St

= (1 + r U SD₎

(1 + r∗₎ (1)

(5)

USD return, and 1 + r∗ is the foreign currency return.

This paper will be using the overnight index swap (OIS) as proxy for rU SD _{and r}∗_. The OIS rate is an interest swap rate where the overnight rate is exchanged for a fixed rate. OIS rates will be used instead of the London Interbank Offered Rate (LIBOR). The LIBOR is a reference rate produced by the British Bankers Association, which represents the average short-term interest rates that banks charge each other. The LIBOR will not be taken as proxy for interest rate, since with the use of OIS rate there is no principal ex-changed so the counterparty risk is smaller for this rate. The exchanging parties calculate the difference in interest to be paid on the initial principal of the swap, however there is no exchange of the principal. The OIS is also being used in the papers of Mancini-Griffoli and Ranaldo [2011] and Fukuda [2016] to measure the CIP condition. If equation 1 does not hold there will exist an arbitrage opportunity to gain a riskless profit by borrowing money in domestic currency and selling it on the spot price in the foreign currency, which will be lent in the base currency. After a particular timespan the domestic currency will be bought back via a forward contract. These trades will be made until the arbitrage opportunity has been exploited to equilibrium. However in textbook arbitrage there is no requirement of capital and involves no risk, but in reality most arbitrage opportunities need capital and can be risky [Shleifer and Vishny, 1997].

The simplest arbitrages are more complex than the textbook shows us. Suppose that the 1-year interest rates in the United States and Japan are 4% and 7%, and the spot rate is 112.3 JPY per USD. From equation 1 we should arrive at a forward rate of 112.3 ∗ (1+0.07)_(1+0.04) = 115.53. But what if the forward exchange rate is smaller than 115.53. If the forward exchange rate is for example 114, then a trader could do the following:

• Borrow 1000 USD at a 4% per year for one year and convert it to 112.3 ∗ 1000 = 112.300 JPY and invest that in Japan at an interest rate of 7%

• Get into a Forward contract and buy 1054.04 USD for 112.300 ∗ 1, 07 = 120.161 JPY, since 120.161/114 = 1054.04 USD

The amount you needed to payoff for the borrowings is 1040 USD, so the arbitrageur acquired a profit of 14.04 USD. Of course such big deviations will not happen in reality, however this is the simplistic way of how it works. Despite that it looks like free money, arbitrage can still be risky and costly. To take advantage of the arbitrage opportunities one needs to deal with credit risk where one borrows and invests their money.

(6)

Tepper, and Verdelhan [2017]. It shows that studies in recent years all show deviations in the CIP conditoin, which indicate that the CIP condition did not hold. Akram, Rime, and Sarno [2008] show that before the crisis the deviations of the CIP were only short lived.

This paper will show the repeated failure of the CIP condition since the start of the financial crisis and identify what factors contributed to deviations. This study will be looking at the CIP condition with the USD as the quoted currency and the EUR, GBP, AUD and JPY as base currencies. This paper will reveal evidence on why the CIP con-dition does not hold. Before the appropriate regression analysis is done, some empirical explanations on why the condition fails will be explained. Firstly, taking advantage of this arbitrage opportunity is not completely riskless. These risks increased especially during the crisis, such as counterparty risks. Secondly, the ability to obtain funding to take advantage of this arbitrage opportunity has become a problem since the financial crisis.

This study differs and improves upon previous empirical work in a few aspects. First of all the dataset that is used will cover a larger time-span then other papers, namely march 2007 until December 2017. This study covers the financial crisis, the European debt crisis and post crisis periods, whereas other studies focus on one of the three periods. Furthermore it will cover multiple maturities for CIP deviations. I will be looking at maturities of 1 month, 3 months, 6 months and 1 year periods for CIP. The rest of the paper will answer the following question:

Research question: What are the main factors that cause CIP deviation for the five major currencies?

The remaining of the paper is organized as follows. Section 2 gives background infor-mation on the CIP condition and its components. Section 3 will give a literature review of previous studies about the CIP. Section 4 gives a basic overview of the methodology and the variables used. Section 5 describes the data that has been used to do this research. Section 6 shows the results of my empirical analysis, and section 7 reports a robustness check. In section 8 a discussion, and section 9 concludes the paper.

2 Background information

(7)

multiple periods will give: Ft,t+s St = (1 + r U SD T (t) ) (1 + r_{T (t)}∗ ) (2)

On the left side of equation 2 we have St, which is the currency spot exchange rate in units of the domestic currency, Ft,t+s is the corresponding FX forward contracted at time t and exchanged at time t+s. In this paper deviations will are calculated for the 1-month, 3-month, 6-month and 1-year period. The right side of the equation shows r∗_{T (t)}, which is the foreign interest rate calculated by annualized value, and rU SD

T (t) is the dollar interest rate that this papers uses as the quoted currency in the rest of the paper. Furthermore r∗_{T (t)} and rU SD

T (t) are annualized rates, which are calculated for the different periods used. If we take a look at equation 2, it is possible to break it up into two components.

(1 + rU SD_{T (t)} ) = Ft,t+s St

∗ (1 + r∗_{T (t)}) (3)

The right hand side of equation 3 is widely known as the FX swap implied dollar rate, whereas the left handside, 1 + rU SD, is known to be the dollar cash rate. An FX swap is a contract to have a currency exchange between two foreign participants. The agreement in the contracts contains a purchase of a certain amount in one currency and the sale of an equal amount for another currency. This exchange is clearly shown in figure 1.

Figure 1: Flow of funds in FX swap

Source: Baba and Packer [2009a]

Figure 1 shows that at the beginning of the contract bank A borrows an amount of X ∗ S USD and lends X EUR. And after a period of 3 months the contract closes and bank A needs to payback X ∗ F and receives X. Where S is the spot exchange rate and F the forward exchange rate.

(8)

Z_t∗ = Ft,t+s St

∗ (1 + r∗_{T (t)}) − (1 + r_{T (t)}U SD) (4) With Z_t∗ as notation for CIP deviations between the US dollar and one of the four other currencies at different maturities. These deviations are graphed in figure 2, where the daily deviations from the CIP condition between the USD and the EUR, GBP, AUD and GBP. It shows the deviations of the 3-month maturity for the period from march 7 2007 until December 30 2016 for the 3-month maturity, and in figure 9 (Appendix) the 1-month, 6-month and 1-year maturities are depicted. Figure 2 shows that there were big deviations for all currencies starting from the beginning of the financial crisis. The biggest deviations can be seen in 2008, where it reached up to 100 basis points for the 3-month maturity. The increased demand for USD made the its interest lower than these other currencies on the forward market. These upward deviations mean that the US dollar had lower interest rate on the forward market.

The fall of Lehman Brothers on 15 September 2008 can be regarded as the point where the deviation for all the major currencies started to take off. Du et al. [2017] show that before the financial crisis the CIP was close to zero for these currencies. However, these deviations did not go away after the crisis. Shortly after the crisis there is another important peak, namely in October 2011 when the European Union decided to agree on a debt relief for Greece. After the Euro crisis the CIP starts to deviate again and according to Claudio Borio and Sushko [2016] this is because the evolving demand for FX hedges and constraints on arbitrage activity.

Furthermore, figure 2 shows that the Australian dollar is the only currency experienc-ing negative CIP deviations. Unlike the other currencies, the USD had higher interest rates than the implied dollar rate on forward contract. Fukuda and Tanaka [2017] shows that Australia did not adopt unconventional monetary policy to aid recovery from a de-flationary economy, while the US, Euro zone and Japan did adopt these policies. These currencies ended up in a liquidity trap. It is shown in figure 8 (Appendix) how all the OIS rates drop except of the AUD.

3 Literature Review

This section will give an outline of the existing literature written on the topic of covered interest rate parity.

3.1 Previous work

(9)

Figure 2: Deviations from the CIP condition between the US dollar and the EUR, GBP, AUD and JPY for 3-month maturity. These daily deviations in the CIP condition are calculated in basis points. The covered interest rate parity implies that the deviations should be zero, where an upward movement implies that the US dollar had lower interest rate on the forward market. The CIP deviation for the 1-month, 6-month and 12-month maturity are shown in figure 9.

However, he argues that if violation of the CIP would occur it could be due to two reasons. Firstly, arbitrage opportunities could arise, however those opportunities are traded away so quickly that they do not come forward in the data sample. Secondly arbitrage opportunities can come forward if one or more good prices are offered in the market. He criticizes earlier empirical work on CIP that uses data from published sources, which he improves by collecting the data himself. Two years later Taylor [1989] again tests for CIP violations, however this time using multiple historic ”calm” and ”turbulent” periods. Again his results support the CIP theorem. Batten and Szilagyi [2011] research concluded the same as Taylor, but with different timespan. They also argue that due to the recent electronic trading platforms, the extent of possibilities for CIP arbitrage has declined. Poitras [1988] looks at the CIP formulation from the perspective of the t-bill market. In case of CIP deviations investors will sell the lower yielding t-bill and buy the higher yielding t-bill with the appropriate covering of exposure to the underlying currencies.

(10)

efficiency is not interpreted as a statement about prices being correct at each point in time but the notion that in efficiently-functioning financial markets very short-term arbitrage opportunities can arise and invite traders to exploit them, which makes it worthwhile to watch the relevant markets”. Fletcher and Taylor [1994] results show evidence of arbi-trage opportunities in the Eurobond and foreign bond market with long-term maturities. However on average they find that the condition holds for long-term markets on which they tested.

Skinner and Mason [2011] point out covered interest parity holds, but only for coun-tries that are rated triple A, such as UK and Norway. They also state that the size of their economy and transaction costs is no likely source of CIP deviations. However, they do find evidence for CIP violations for emerging markets on the long-term. Skinner and Mason contribute these deviations to factors related to credit risk.

According to the efficient markets hypothesis, all current information is reflected in its prices, so that means that agents in the market cannot make huge profits by utilizing current information [Fama, 1970]. Crowder [1995] looks at the efficiencies of markets, because according to the efficient market hypothesis arbitrage possibilities cannot persist longer than one day. Crowders results show that one of the four markets he analyzed could be marked as efficient, the other three did not produce the satisfactory result he was hoping for.

Still, some papers found evidence for CIP deviations prior the financial crisis, Kui-Yin and Po-ming [1994] find arbitrage opportunities in the Hong Kong foreign exchange market. The profit opportunities from CIP deviation is especially large for the one-way arbitrage. Additionally, they found larger profit opportunities for long-term contracts, since those contracts had lower swap costs. Furthermore, Bhar et al. [2004] study CIP deviations as a consequence of higher volatility in exchange market. They look at a 90-day period, and they find evidence of more daily deviations in CIP when there is higher volatility in exchange market.

(11)

about counterparty risk became an issue after the bankruptcy of Lehman Brothers. This is also argued by Melvin and Taylor [2009], banks were hesitant to lend each other not knowing the details of other balance sheet. Not only was the market highly volatile, but as a financial institution you needed to consider counterparty risk.

Furthermore, Fukuda (2016) looks at how CIP deviations vary across different markets during the financial crisis, for example the difference between the New York and Tokyo markets. One of their main findings was that larger dollar-specific risk and smaller yen-specific risk explained why CIP deviations were larger in the Tokyo market. On the other hand Bottazzi, Luque, Pascoa, and Sundaresan [2012] look at cross currency bias between the euro and the dollar. They show that it was extremely costly to carry out arbitrage after the fall of Lehmann and during the European debt crisis. The banks were cautious to lend each other scarcer currency, which enforced the continuation of the cross currency basis. They find that there was a significant “convenience yield” in the scarcer currency, the dollar. This has also been discussed by Coffey et al. [2009] with the model of Brunnermeier and Pedersen [2009], to explain CIP deviations due to the difficulties in borrowing US dollar during the crisis. It was difficult to perform arbitrage trades, since increased global demand in US dollars and decrease in supply.

Du et al. [2017] examine CIP deviations post crisis and find that these deviations are especially strong for forward contracts that appear on the banks balance sheet at the end of the quarter. Pinnington and Shamloo (2016) did CIP research on the first half of 2015 and argue that deviations arose from reduced liquidity in foreign exchange markets, instead of imbalances in international funding markets which was argued to be big factor during the crisis by some researchers (see Coffey et al., 2009; Brunnermeier and Pedersen, 2009).

Various studies use different measure for interest rate, for instance this study uses OIS rate whereas other papers use LIBOR as measure (see Coffey et al., 2009; Skinner and Mason, 2011). Mancini-Griffoli and Ranaldo [2011] sketch the difference in the textbook CIP and real world CIP. The big distinction in the real world is that arbitrage can be undertaken by borrowing and lending funds on secured or unsecured terms. The two strategies are almost identical, except that lending in secured terms involves pledging collateral. However they find that arbitrage can be undertaken by means of both secured and unsecured funds, and come to the same results.

(12)

3.2 Liquidty risk

Previous literature will be used to consider liquidity risk as a possible factor for CIP deviations, before it is used in the model. During the financial crisis, liquidity in the in-terbank market shrank quickly and banks preferred to hoard their cash instead of lending it out [Heider, Hoerova, and Holthausen, 2009]. Since they were afraid of their own fu-ture funding constraints, banks started to hoard liquidity. McGuire and Von Peter [2009] document in their paper how difficult US dollar funding became after the fall of Lehmann Brothers. European banks held large US dollar assets beyond their dollar deposits, and those assets were difficult to sell in an illiquid market. These US dollar assets were fi-nanced mostly at short maturity and rollovers to support these assets became difficult for European banks since funds were pulled back. Furthermore, since it was difficult for European banks to borrow on the interbank market they needed to rely on the secured funding markets and foreign exchange swaps. However, these banks tried to swap euros for dollars, but there was less demand in euros by US banks. So there was a large imbal-ance in currencies on a global level, and the banking system suffered from it [McCauley and McGuire, 2009]. The holdings of the European banks in assets denominated in US dollar were larger than US banks in euro assets, so there was an imbalance in the currency hedging. These potential sources of liquidity hoarding and funding are also documented by Mancini-Griffoli and Ranaldo [2011], but they also note how financial institution felt pressure to deleverage. With the deleveraging of their balance sheet, the banks tried to rebalance the portfolio to exchange rate exposure and gain liquidity. As many firms started doing this deleveraging at the same time, it drove down asset prices and caused panic with investors about the wealth of the firm. Shleifer and Vishny [2010] document in their paper about so called “fire sales”, where firms are forced to liquidate their assets because they owe cash to someone else.

3.3 Counterparty risk

(13)

an issue since 2008. Heider et al. [2009] discuss the important role of counterparty risk in the breakdown of the interbank market. In their paper they argue that for higher levels of counterparty risk there will be more adverse selection in the interbank market. The banks that are considered safe will leave the unsecured market, but this will give rise to interest rates. The interbank may break down if this risk becomes to high, since banks will rather keep their liquidity than lend it too borrowers with asymmetric information.

4 Methodology

The objective of this paper is to analyze the data for arbitrage possibilities and to be able to explain what components cause these deviations in CIP. I will be looking at CIP for the following currencies: USD, EUR, GBP, JPY and AUD.

My paper builds on works such as Baba and Packer [2009a], Fong, Valente, and Fung [2010] and Fukuda [2016] who study the CIP arbitrage profits and their relationship with credit risk and market liquidity. They focus on the CIP relationship particularly during the crisis. Their results show large number of deviations in CIP. This study will be using time series analysis to show what factors contributed to these deviations. The CIP deviations for each currency will be calculated in the following way:

Z_i∗ = Ft,t+s St

(1 + r_{T (t)}∗ ) − (1 + rU SD_{T (t)} ) (5)

(14)

4.1 The model

My main model will be a regression with the CIP deviation as the dependent variable: Z_i∗ = β0+β1Liquidity+β3LOISU SD+β4LOIS∗+β5(CDSU SD−CDS∗)+β6V IX + (6)

,where • Z∗

i: CIP deviations for particular maturity i and currency ∗.

• Liquidity: denotes liquidity risk measure, which is calculated using the FX forward bid-ask spread.

• LOIS: LIBOR-OIS spread, for both US dollar and the corresponding currency ∗. • CDSU SD _{− CDS}∗_{: denotes the spread in CDS premium of the USD and}

corre-sponding currency ∗. This factor measure sovereign specific credit risk. • V IX: The VIX index, which is a proxy for global market risk.

4.2 Liquidity Risk

It is quite difficult to test for liqudity risk, since there is no empirical measure that captures it fully. What is meant in this paper by liquidity is the ability to easily buy and sell with low transaction costs. If markets are more illiquid, then this will increase the volatility. However to test for liquidity risk I will be taking the bid-ask spread of the FX forwards. This liquidity measure is also used in previous empirical work (see Fong et al., 2010; Fukuda, 2016). This liquidity risk is an important factor according to previous empirical work, because shortages of the USD during the financial crisis caused distortion in the money markets. Counterparty risk became more extreme after the bankruptcy of Lehmann Brothers, which caused dollar liquidity problems for European firms. In the summer of 2007 it was very hard to price forward contracts due to the worsening of the liquidity of the USD, GBP and EUR [Baba and Packer, 2009b]. The market conditions in the equity sector were worsening and carry traders witnessed an increase in volatility. Pinnington, Shamloo, et al. [2016] point out that large spreads in the bid-ask can reflect trouble for future exchange rate movements.

4.3 LIBOR-OIS

(15)

almost no credit risk (OIS). With all the LIBOR and OIS maturities for every currency, this spread can be calculated and it can be seen that the spread has increased significantly since the crisis. As former Chairman of the Federal Reserve, Alan Greenspan stated “the LOIS remains a barometer of fears for the banks insolvency”. Meaning that a smaller spread shows the confidence in banks to be able to repay their loans. The use of LOIS as measure for credit risk has been discussed in multiple papers (see Baba and Packer, 2009b; Coffey et al., 2009; Mancini-Griffoli and Ranaldo, 2011; Fong et al., 2010). Gefang, Koop, and Potter [2011] debate in their paper, that for short-term (1-month and 3-month maturities) LOIS spreads the role of liquidity risk is more important than that of credit risk. However, for the longer-term (1 year) LOIS spreads both the liquidity risk and credit risk have important contribution. Filipovi´c and Trolle [2013] did a study on interbank risk and the spread of LOIS. They get a similar conclusion as Gefang et al. [2011], that increase in the spread gives rise to liquidity risk on the short-term and default risk on the longer-terms. Furthermore, it is necessary to distinguish credit risk and market risk, since credit risk arises from the counter parties default probability and market risk arises from market variables such as interest rate and exchange rates. Market risk is more easily hedged by entering in offsetting contracts, whereas for credit risks it is more difficult.

4.4 Sovereign credit risk

(16)

4.5 VIX

To measure volatility, the model will use the CBOE Volatility Index (VIX), which is the measure of market’s expectation of 3-day volatility. It measures the implied volatility of the S&P 500 index options that is considered to represent the market as a whole. This index is widely chosen as a reflection for investors judgement of the global market. Expected is that increase in global market risk will have an impact on CIP deviations, since there will be a stronger move towards US dollars when there was this increase in market risk in the financial crisis. This increase in global market risk has also happened during the EU debt crisis, where there was shift in demand from the euro to the British pound. Filipozzi and Staehr [2013] show that the VIX had an impact on the deviations of two Eastern European countries during the crisis. Furthermore, previous work has found the volatility index to be a significant source (see Brunnermeier, Nagel, and Pedersen, 2008; Coffey et al., 2009; Mancini-Griffoli and Ranaldo, 2011).

5 Data

In this section the data will be described that is used to calculate the CIP condition and the variables used in the model. The data covers a period from march 2007 until December 2016. This period captures the financial crisis and the aftermath to study CIP deviations. The CIP condition will be measured at maturities of 1-month, 3-month, 6-month and 1-year with the use of daily data. As mentioned earlier, this paper will be using the OIS rates as a proxy for the interest rates. For the OIS rates this paper has taken the daily middle rate quotes from Thompson Reuters electronic brokerage system, which are retrieved from datastream. There are some small remarks regarding the OIS data for some currencies. First of all the OIS dataset for GBP at maturities of 1-month, 3-month and 1-year starts at August 27 2007. Secondly the data set of the 1-year JPY ends on November 13 2012. This is not a big problem, since the dataset is still big enough to capture deviations. In table 2 the variables used in this study are explained.

In table 11 panel A (Appendix) the descriptive statistics of all OIS rates are shown. From the table it is clear that Australia always had quite high rates, whereas Japan always experienced very low rates. Before the crisis this was a very popular carry trade strategy, where investors borrow in a country with low interest rate (Japan) and invest in a country with high interest (Australia). From panel A in table 11 it is visible that there are some big differences in the maximum and minimum OIS rates. This is due to the fact that OIS rates have gone down in recent years and nowadays are on historical low levels.

(17)

AUD/USD and JPY/USD have been retrieved. These exchange rate quotes are from Reuters electronic brokerage system, which are retrieved from datastream.

The data from OIS rates as well as spot FX and forward FX rates in table 11 are used for the calculation of CIP deviations. The descriptive statistics of the calculated deviations are reported in table 1 and graphically shown in figure 2 for the 3-month maturity. The statistics clearly show that the JPY CIP deviation mean and median are the highest, whereas the AUD show the lowest mean and even an average negative deviation.

Table 1: This table shows descriptive statistics of CIP deviations for each currency and maturities of 1-month, 3-month, 6-month and 1-year. These deviations have been calculated with the use of equation 5. The deviations are shown in basis points with the USD as quoted currency. The statists are for the period March 7 2007 - December 30 2016

Variable Mean Median SD Max. Min. Skew. Kurt. N

EUR 1M 2.720 1.976 3.232 53.086 -0.535 6.060 60.521 2563 3M 7.901 6.070 8.050 101.991 -1.462 4.359 33.144 2563 6M 15.680 12.569 14.019 143.578 -8.122 3.023 18.391 2563 1Y 32.626 28.153 24.908 186.355 -7.116 1.625 7.388 2563 GBP 1M 1.468 0.747 2.889 52.389 -1.744 7.319 82.443 2180 3M 3.996 2.043 6.677 103.556 -4.349 6.469 61.666 2563 6M 7.152 3.545 11.008 133.049 -5.981 4.387 30.517 2180 1Y 12.360 6.627 17.916 147.190 -13.709 2.767 13.616 2180 AUD 1M 0.008 -0.396 3.289 34.528 -7.361 4.729 38.287 2563 3M 0.708 -1.046 8.425 78.155 -9.760 4.276 28.035 2563 6M 0.352 -2.155 15.120 130.931 -18.267 3.667 21.696 2563 1Y -3.262 -7.788 24.959 228.999 -40.865 2.973 16.209 2563 JPY 1M 3.443 2.327 3.651 52.655 -3.359 4.731 40.352 2563 3M 10.407 7.906 9.545 106.614 -9.227 4.131 30.432 2563 6M 21.690 17.586 16.486 149.402 -18.208 2.500 14.009 2563 1Y 44.398 39.035 30.720 192.929 -37.388 0.944 4.375 2563

Equation 5 is used to explain the consistent deviations in the CIP condition. For these explanatory variables the necessary datasets have been obtained. Table 3 panel A gives the descriptive statistics of the forward FX bid-ask spread, which will be used as a measure of liquidity risk. This dataset includes all daily bid and ask FX forward exchange rates for the five currencies. The spread is calculated as follows:

%Spread = 100 ∗ (Ask price − Bid price)

Ask price (7)

(18)

forwards are the most liquid. However the spreads do not differ that much compared to the GBP, but it does show that the AUD is the least liquid.It is also clearly visible in figure 3 that reports the 3-month bid-ask spreads, showing how these spreads only moved significantly during the financial crisis. This indicates again that during the crisis there was an liquidity problem.

Table 2: Variable definition

OIS Overnight Index swap, which is used as proxy for interest rate for the calculation of CIP. The OIS is an interest swap rate where the overnight rate is exchanged for a fixed rate

LIBOR London interbank offered rate, which is a reference rate produced by the British Bankers Association. It represents the average short-term interest rate that banks charge each other.

CDS Credit default swaps are financial contracts that transfer the risk of a corporate/sovereign bond between two parties. It is a form of insurance when you buy a bond, since it gives you protection when the underlying bond defaults. LOIS LIBOR-OIS, which is the difference between LIBOR and

OIS rates. This spread incorporates liquidity and credit risk. FX forward bid-ask Spread in the bid and ask prices of the forward exchange spread rates. If the spread increases, it means hat the market is

getting more illiquid.

VIX Volatility Index (VIX), which is the measure of market’s expectation of 3-day volatility. It measures the implied volatility of the S&P 500 index options that is considered to represent the market as a whole.

(19)

all spreads are increasing with maturity and all have experienced some high maximums, which occurred during the crisis. The USD experienced the highest LOIS spread on the 3-month maturity of 364 basis points. Figure 4 graphs the daily LOIS spread of the five currencies for the 3-month period. During 2008-09 there are clear maximums for all currencies, and during the Euro crisis there is a clear peak for the EUR. From the figure it is also visible that the LOIS for the JPY is quite stable after the crisis.

This paper will be using CDS rates for the sovereign specific credit risk, with the descriptive statistics reported in panel C of table 4. I took the five-year CDS premiums, since at that maturity they are the most liquid. They are considered the most liquid because they are the most transacted and are regarded to be the fundamental of the credit derivatives market. CDS data of both Germany and France will be used as a proxy for the Euro zone, since there are obviously no specific CDS quotes available for the whole of Europe. The CDS quotes are taken from CMA data vision, which are downloaded from datastream. It is clear from table 4 panel C that France has the highest mean, median and standard deviation of all six countries. One of the reasons is the European debt crisis, which had an huge impact on the CDS on European countries as is clearly graphed in figure 5. Germany was also affected by the European debt crisis, but less since they are a more stable country and have less sovereign risk. That is also why there are two different European countries included; maybe they will give different results in the regressions. Furthermore the statistics show that Japan has a relatively high CDS mean and median, however the reason is that Japan’s CDS market is not as developed as its counterparts. For the regression the CDS difference between the USA and the corresponding country will be taken, and from table 4 panel C and figure 5 it is clear that the USA has the lowest CDS premium. So the data of this CDS difference will be negative in all samples that are used.

In panel D of table 4 we can find the descriptive statistics of the VIX index, which will be used as a proxy for market volatility. The VIX index measures the volatility of US equities from the S&P500 and the index quotes are taken from the Chicago Board Options Exchange, which are retrieved from datastream. An high value of the index corresponds to a more volatile market. The statistics report a mean and median of around 20, which can also clearly be seen in figure 6. During the financial it reached a maximum of 65 points, after which there are some small peaks in may 2010 and October 2011 that both can be accounted to the Euro crisis.

For all these explanatory variables the correlations can be found in table 12 (Ap-pendix). From the correlation it is clear that LOIS USD is mostly correlated with the dependent variable, CIP deviations. And that the two LOIS variables are highly cor-related to each other, which is expected since the same instruments are used for the calculation of this variable.

(20)

Table 3: Descriptive statistics of the explanatory variables. Panel A of this table summa-rizes the forward exchange rate bid-ask spread. Panel B reports the descriptive statistics LOIS spread for 1-month, 3-month, 6-month and 1-year maturities, which have been cal-culated by taking the LIBOR rates minus the OIS rates. The statistics are given in basis points in both panels for the period March 7 2007 - December 30 2016

Panel A: FX Forward bid-ask spread

(21)

Table 4: Descriptive statistics of the explanatory Variables. Panel C of this table summarizes the CDS statistics that are based on the level of its premium. Panel D reports the VIX index, which are reported based on its price index.

Panel C: CDS

United States 23.724 22.230 13.194 90.000 6.000 1.820 8.364 2345 Australia 50.468 44.765 23.808 185.000 19.770 2.104 9.019 2126 Japan 60.724 55.354 27.865 152.640 15.750 0.720 3.190 2306 United Kingdom 49.410 45.213 28.732 165.000 10.500 0.917 3.807 2267 Germany 34.178 25.860 24.196 118.380 5.200 1.315 4.047 2356 France 68.688 51.660 49.654 245.270 11.000 1.555 4.688 2205 Panel D: VIX VIX 21.868 19.750 8.059 65.860 12.600 1.889 7.377 2475

Figure 3: 3-month forward FX bid-ask spread in basis points for the USD and the four currencies. The graph shows the period March 7 2007 - December 30 2016.

(22)

Figure 5: CDS premiums for every country, with France and Germany as proxy for the Euro zone. The initial period is from March 7 2007 - December 30 2016, but some datasets start some months later.

Figure 6: Daily price level of the VIX index for the period March 7 2007 - December 30 2016.

6 Results

This section reports the empirical results regarding the factors that have an influence on CIP deviations of the EUR, GBP, AUD and JPY. For each time-series regression this paper uses daily data for the period of 7 March 2007 until 30 December 2016. The depen-dent variable, CIP deviations, will be reported in basis points. The time-series regression that will be used has been discussed in section 4 equation 6. First the EUR/USD results will be discussed, after which the other currencies will be compared to it.

6.1 Results EUR

(23)

results indicate that the EUR/USD is very sensitive to liquidity risk, which in this case means that an increase in the bid-ask spread will positively increase deviations. For example the 3-month period of table 5, an increase of 1 basis point in the FX spread will increase the deviations with 1.334 basis point. This is in line with what has been discussed before, that global dollar shortage during the crisis impacted these FX forward prices. Baba and Packer [2009b] showed that these dollar liquidity problems caused turmoil in the FX swap market, since during the crisis European institutions wanted to convert euro’s into US dollars, resulting into more demand than supply of FX swaps. Only for the 6-month maturity in table 5 it is visible that this liquidity risk factor is not significant. Also Fong et al. [2010] results show that liquidity risk is an important component of the repeated deviations. Furthermore Pinnington et al. [2016] report that the large bid-ask spreads caused large persistent deviations in the CIP that cannot be arbitraged away. Table 5: Time-series regression of daily EUR/USD CIP deviations using equation (6). This table shows the estimation results for a 1-month, 3-month, 6-month and 1-year maturity of EUR/USD CIP deviations. The dependent variable, CIP deviations, is given in basis points. These results are with use of the CDS of France as proxy for the Euro zone. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. The regression is estimated for the period March 7 2007 - December 30 2016. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

1 Month 3 Month 6 Month 12 Month

FX spread 0.353 1.334 0.757 2.450 (0.207)* (0.384)*** (0.518) (0.599)*** LOIS USD 0.084 0.236 (0.368) 0.632 (0.017)*** (0.026)*** (0.037)*** (0.051)*** LOIS EUR -0.015 -0.125 -0.091 -0.070 (0.015) (0.026)*** (0.041)** (0.051) CDS (US-FRA) -0.004 -0.038 -0.081 -0.121 (0.003)* (0.006)*** (0.012)*** (0.025)*** VIX -0.021 0.004 -0.333 -0.784 (0.022) (0.067) (0.180)* (0.219)*** Constant 1,026 -1.034 3.653 -5.670 (0.676) (1.224) (2.920) (3.341)* Observations 2089 2089 2089 2089 Adjusted R-squared 0.542 0.704 0.682 0.666

(24)

short-term and credit risk on the long-short-term. The table reports that LOIS spread of the USD is positively significant at a 1%-level for every maturity. Indicating that if the LOIS spread of the USD will increase with one basis point in the 3-month maturity, the CIP deviations will increase with 0.236 basis points. Furthermore it is also clearly visible from figure 4 that these spreads have risen quite extremely and pushing these deviations up. There is almost no difference in reported coefficients when comparing the two tables. The OIS market has very small default risk compared to the LIBOR market, however the market’s perception of risk increased causing an increase in the LOIS spread. And this in turn caused a turmoil in the CIP condition, creating some fraction of these big deviations. Figure 12 (Appendix) shows a correlation table of the variables, which also indicated that LOIS mostly correlated with the CIP deviations. Also indicating that these variables move together Baba and Packer [2009a] results report that the dollar LOIS spread had a significant positive impact, but only since the financial crisis.

As for the LOIS spread of the EUR, table 5 reports that the coefficients have a significantly negative effect on the CIP deviations. This currency specific risk measure the pushes the deviations down, instead as for the LOIS spread of the dollar that is pushing it up. However this heightening of counterparty risk during the financial crisis of 2008-09, but also the European debt crisis in 2011-12 played a crucial role in the widening of these spreads. For the 1-month maturity the LOIS spreads of the EUR are not significant and in table 13 only at a 10%-level, that may reflect that the problems in the short-term money markets was less problematic in the whole sample than the long-term market. A possible explanation could also be that money market risk was only significant during the financial crisis, since during the Euro crisis there was an increase in sovereign risk. Their is only a small difference in reported coefficients between sample with France or Germany as CDS, since the results with Germany shows greater negative coefficients. A possible explanation could be the difference in riskiness between the two countries, France and Germany.

Regarding the sovereign credit risk, which is reported as the CDS premium difference. Table 5 shows that all the coefficients have significant negative effect on the deviations. The financial crisis increased the CDS premiums of both the USA and Europe, which represented by France and Germany in this regression model. Looking at the 3-month maturity in table 5, it shows that an widening of the CDS by 1 will decrease the CIP deviations by 0.038 basis points. This is negative because the EUR is seen as a substitute for the USD when credit risk of the United States increases. So during the Euro crisis there was a flight to USD, since the sovereign risk of the EUR was rising. Fukuda [2016] finds the same effect, where the CDS started to have an impact on the CIP deviations after the financial crisis.

(25)

that for the 1-year maturity an increase in the volatility index is related to an increase in demand for USD and lower interest rate. As Baba and Packer [2009a] discuss in their paper, since the financial crisis banks were shortening their debt structure to be able to payoff their loans. The VIX volatility index is just not statistically and economically significant enough to be able to explain CIP deviations. Furthermore, Coffey et al. [2009] find the VIX index to be significant for the period 2008-09.

6.2 Results GBP, AUD and JPY

The regression results of the GBP, AUD and JPY will be discussed and compared to the results of the EUR. In table 6 the results of the GBP are reported. Firstly, sign of FX spreads is positive and significant at a 1%-level for every maturity, which indicates that higher bid-ask spreads also give rise to CIP deviations of the GBP. Furthermore the LOIS spreads give almost identical significant results as for the EUR, where the USD LOIS is positively significant and GBP LOIS is negatively significant at 1%-levels. Fukuda [2016] finds similar results for the GBP, whrere the currency-specific money market risk is a important factor in explaining these deviations. For the CDS variable there is a difference with the EUR, since it is only statistically significant at a 1%-level for the 12-month maturity and 10%-level for the 3-month maturity. A possible reason is that the sovereign credit risk played a smaller role for the United Kingdom, since they were less involved in the European debt crisis. This is also visible in the CIP deviations itself, since after the crisis the deviations for the GBP stayed more stable than for the EUR. Furthermore the coefficient for the 12-month period shows a positive significance, whereas the EUR gives negative significance. This was because financial institutions switched from the EUR to the GBP, since it was considered more safe during the Euro crisis. The reason would be that the GBP is more exposed to shocks in the United States than the EUR, since the EUR is more a potential substitute when credit risk of the United States increases. However the GBP was much less volatile during that period as can be seen in the descriptive statistics in table 15 of the appendix. For the VIX the same accounts for the EUR, where it is also not statistically significant.

(26)

Table 6: Estimation results time-series regression of daily GBP/USD CIP deviations using equation (6). It shows regression results for the 1-month, 3-month, 6-month and 1-year maturity where the dependent variable, CIP deviations, is reported in basis points. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. The regression is estimated for the period March 7 2007 - December 30 2016. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

FX spread 1.201 1.600 1.701 1.571 (0.073)*** (0.095)*** (0.544)*** (0.411)*** LOIS USD 0.077 0.232 0.329 0.446 (0.002)*** (0.005)*** (0.038)*** (0.040)*** LOIS GBP -0.029 -0.124 -0.168 -0.138 (0.004)*** (0.005)*** (0.034)*** (0.028)*** CDS (US-UK) 0.002 -0.009 0.026 0.267 (0.003) (0.005)* (0.032) (0.045)*** VIX -0.011 0.063 -0.009 0.191 (0.009) (0.044) (0.112) (0.156) Constant -2.474 -4.660 -6.504 -17.700 (0.261)*** (0.390)*** (1.677)*** (2.606)*** Observations 2115 2148 2115 2148 R-squared 0.639 0.797 0.751 0.757 Adjusted R-squared 0.638 0.796 0.750 0.756

in line with the EUR. However, for the AUD LOIS it shows only a negative significance at a 1%-level for the 6-month period. This means that this change in risk premiums for the Australian dollar does not affect the CIP deviations in the shorter term periods (1-month and 3-month). It is also visible in figure 4 that the LOIS spread was not as wide during the crisis compared to the other currencies. Another possible explanation can be that these longer term periods are more affected, because they were less liquid. The CDS variables coefficients are all statistically positive at a 1% significance level, and the same reasoning of the GBP applies to the AUD of why it is positive. Table 7 does report statistical significance for the volatility VIX index for the 1-month and 6-month period, whereas the EUR did not report significance. When looking at the R-squared, the 1-month maturity only shows a R-squared of 0.365, which is significantly lower than the other samples.

(27)

Table 7: Estimation results time-series regression of daily AUD/USD CIP deviations using equation (6). It shows regression results for the 1-month, 3-month, 6-month and 1-year maturity where the dependent variable, CIP deviations, is reported in basis points. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. The regression is estimated for the period March 7 2007 - December 30 2016. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

1 Month 3 Month 6 Month

FX spread 0.270 0.296 0.620 (0.058)*** (0.122)*** (0.222)*** LOIS USD 0.080 0.171 0.195 (0.010)*** (0.023)*** (0.034)*** LOIS AUD -0.0175 -0.033 -0.074 (0.015) (0.021) (0.024)*** CDS (USD - AUS) 0.017 0.126 0.181 (0.006)*** (0.023)*** (0.051)*** VIX -0.028 0.038 0.269 (0.013)** (0.033) (0.071)*** Constant -1.856 -2.397 -5.261 (0.316)*** (0.725)*** (1.554)*** Observations 1910 1910 1910 R-squared 0.365 0.676 0.680 Adjusted R-squared 0.364 0.675 0.679

(28)

Table 8: Estimation results time-series regression of daily JPY/USD CIP deviations using equation (6). It shows regression results for the 1-month, 3-month, 6-month and 1-year maturity where the dependent variable, CIP deviations, is reported in basis points. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. The regression is estimated for the period March 7 2007 - December 30 2016. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

(29)

7 Robustness

This section reports a robustness check to assess my results in section 6 Results.

7.1 Seperation sample size

This robustness check will be done for the 3-month maturity by reducing the sample size. The original sample will be cut into three different periods, containing the financial crisis, European debt crisis and post-crisis periods. The purpose of splitting my data into periods is to capture period specific factors. Maybe for the financial crisis period different variables will be significant compared to post-crisis. The results of my robustness test can be found in table 9, table 18, table 19 and table 20. The results show for every period the corresponding currency, for which period 1 is from 07-03-2007 until 30-12-2009, period 2 is from 01-01-2010 until 30-12-2013, period 3 is from 01-01-2014 until 30-12-2016 and the sample period contains the whole data sample from 07-03-2007 until 30-12-2016. Table 21, table 22 and table 23 (Appendix) show the same results, but these are reported per period instead of individual currency.

First regarding the results of the the variables of the EUR in table 9. For these results only CDS of France is taken as proxy for Europe. The FX spread is positively significant in every period, as well as the initial results. This is consistent with my view that liquidity risk plays a part in the explanation of CIP deviations. Furthermore the USD LOIS also shows positive significance through every period, indicating what was expected during this paper is true. Regarding the LOIS of the EUR, the results of period 3 are the only ones that are true to the original results. In period 1 there is a significance of 5% for the second subsample and in period 2 there is significance of 10% for the first subsample, whereas the original results indicate a positive significance of 1%. This may suggest that credit risk of the EUR only has significant effect in period 3, after the European debt crisis. For the CDS spread between the US and France or Germany, it is consistent with my view that it had a significant effect during the European debt crisis. In period 1 it is only significant at a 10%-level, which is less significant that for period 2 and 3. Finally the VIX index is almost identical in the period to the sample, except for period 2 were it is significant. This can suggest that the increase of the global market risk had an big impact on the EUR/USD deviations during the Euro crisis. Probably because during the Euro crisis there was an increase in demand of the USD.

(30)

Table 9: Estimation results robustness test EUR/USD for each period, where CDS of France is taken as proxy for Europe. It shows regression results for the 1-month, 3-month, 6-month and 1-year maturity where the dependent variable, CIP deviations, is reported in basis points. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

EUR/USD

Period 1 Period 2 Period 3 Sample

FX spread 1,517 0.901 1.157 1.334 (0.714)** (0.163)*** (0.224)*** (0.384)*** LOIS USD 0.251 0.156 0.343 0.236 (0.051)*** (0.027)*** (0.031)*** (0.026)*** LOIS EUR -0.144 -0.024 -0.473 -0.125 (0.089) (0.014)* (0.063)*** (0.026)*** CDS -0.204 -0.031 0.236 -0.038 (0.104)* (0.006)*** (0.030)*** (0.006)*** VIX -0.166 -0.031 0.076 0.004 (0.121) (0.006)*** (0.051) (0.067) Constant 3.091 -6.632 6.359 -1.034 (3.221) (0.727)*** (1.882)*** (1.224) Observations 360 984 743 2089 Adjusted R2 0.823 0.803 0.891 0.704

period 3. These country-specific risks had no significant impact during the financial crisis and the European debt crisis, however in recent years it has become a significant factor. Finally the VIX index shows the same results, which is insignificant. A last remark re-mark is the R-squared, which is is quite high except for period 2 where it gives a value of 0.405. This indicates that there were some other factors influencing CIP deviation in period 2.

(31)

than the the results from the whole sample.

Table 20 reports the JPY/USD per period CIP deviation, where the coefficient of the FX spread was insignificant in period 3. However the other period are consistent with the results from the whole sample. The LOIS of both the USD and JPY are significantly the same and support the view that credit risk and liquidity risk were factors for CIP deviations. The country specific risk, CDS variable, is again only in period 3 insignificant. And for the VIX index only period 1 is insignificant, but overall most variables are consistent with the results from the whole sample.

7.2 CIP with LIBOR

A second robustness check is done on the computation of the interest rate in the CIP condition. This paper used OIS rates as proxy for the interest rate, while it is also common to use LIBOR. This section will show if there are any differences when using the LIBOR rate in the CIP condition. Table 10 shows the descriptive test statistics of the daily CIP deviations. For the EUR the results show similar results, where the OIS rate reports slightly higher statistics. The GBP also shows similar results, but with the LIBOR showing slightly higher mean. For the JPY and AUD the differences are bigger, where the AUD reports a positive mean for OIS rate and negative mean for LIBOR. As a whole the OIS rates give higher statistics then the LIBOR does, but the trend of CIP deviation is almost the same as can be seen from the figure 7, which graphs the 3-month EUR/USD CIP of both LIBOR and OIS.

Table 10: Descriptive statistics of CIP deviation using LIBOR or OIS. These deviations are given in basis points for the period 07-03-2007 until 30-12-2016. The first table show the CIP results with the use of LIBOR and the second table with the use of OIS.

CIP with LIBOR

EUR 7.087 5.821 6.743 72.872 -6.060 2.231 11.619 2481 GBP 4.678 3.132 6.278 84.681 -6.068 4.410 33.119 2481 AUD -2.449 -2.485 3.586 29.106 -29.209 1.078 14.671 2431 JPY 6.292 5.264 5.060 61.148 -12.455 2.136 15.708 2481

CIP with OIS

(32)

Figure 7: Deviations from the CIP condition for EUR/USD with the use of either LIBOR or OIS rates. These daily deviations in the CIP condition are shown in basis points. The covered interest rate parity implies that the deviations should be zero, where an upward movement implies that the USD had lower interest rate on the forward market.

8 Discussion

In this discussion section the econometric results will be interpreted to answer the research question and the study done in this paper will be critically evaluated. In the introduction the research question was introduced, which stated:

Research question: What are the main factors that cause CIP deviation for the five major currencies?

The results section reported regressions that have been done to test for the statistical significance of these factors. The results show that liquidity risk, currency specific risk and sovereign risk are significant factors across these currencies deviations. It is difficult to conclude how big the contribution each factor is for these deviations. However, from previous studies and this empirical work it becomes evident that liquidity risk, currency specific credit risk and sovereign credit risk are the main factors that cause deviations in the CIP condition.

Regarding the FX bid-ask spread, it gives significance for every currency and also every period in the results. However, when looking at the daily spread graphed in figure 3, it shows only big deviation during 2008-09 and the rest of the sample it stays quite constant. This result I find conspicuous and should be investigated further. From these results I do believe that this liquidity risk factor had a big impact during the crisis, but not significant in the the periods afterwards.

(33)

FX market and many assets of financial institutions are denominated in dollars. Coffey et al. [2009], Fukuda [2016], Du et al. [2017] and Pinnington et al. [2016] all report the importance of this factor for CIP deviations. Since the LIBOR-OIS spread incorporates both liquidity risk and credit risk, it is hard to point out which part has had more influence on the deviations over the whole sample

As for the CIP deviations of the Australian dollar, which was the only currency that experienced negative deviations. This is explained by the fact that Australia did not adopt unconventional monetary policy [Fukuda and Tanaka, 2017]. This paper did not focus on the monetary policies of the particular currencies, but used the results of other studies to explain the unexplained.

This paper showed the main factors contributing to deviations in the CIP condition. The factors are derived from existing literature and tested for their statistical significance. As with many empirical papers there are always factors which are not looked at. This is quite a broad study and did not focus on specific country effects or banking regulation.

9 Conclusion

This study shows the persistent failure of the covered interest rate parity in the the US dollar, euro, British pound, Australian dollar and the Japanese yen since the start of the global financial crisis. These CIP deviations were calculated using OIS rates as proxy for interest rates. With the help of the necessary empirical work I found that these deviations could be related to liquidity risk, credit risk or sovereign risk. The model in this paper uses four variables as possible characteristics of CIP deviation. Firstly, CIP deviations increase when there is a widening of the FX forward bid-ask spread. Secondly, there will be an increase in deviations if the LOIS widens more. Thirdly, increase of CDS premiums of the four currencies relative to the USD is an important factor for deviations. Lastly, the influence of the VIX index measuring global market volatility.

My empirical results show that a change in the FX forward bid-ask spreads has a positive significant effect for every currency. This liquidity risk was particularly profound during the crisis, where USD shortages caused turmoil in the FX market. The same accounts for the USD LOIS, which is a significant factor explaining the currency specific risk. The OIS market has a very small credit risk compared to the LIBOR market, however the market’s perception of risk increased causing an increase in the LOIS spread. The sovereign specific risk, CDS, played an important role for the EUR, AUD and JPY, as reported in the results. This was due to the fact that during the Euro crisis there was a switch from the EUR to the GBP, since it was a more safe currency..

(34)

Splitting the sample in smaller periods does give a better indication of the importance of different variables in particular periods.

(35)

Appendix

(36)

Figure 9: Deviations from the CIP condition between the US dollar and the EUR, GBP, AUD and JPY for 1-month, 6-month and 1-year maturity. These daily deviations in the CIP condition are calculated in basis points. The covered interest rate parity implies that the deviations should be zero, where an upward movement implies that the US dollar had lower interest rate on the forward market.

(a) CIP 1-month maturity

(b) CIP 6-month maturity

(37)

Table 11: Descriptive Statistics of variables used in the CIP condition for the USD, EUR, GBP, AUD, and JPY. Panel A in this table shows the statistics of OIS rates, where these rates are shown in %. Panel B reports the statistics of spot and FX forward rates with USD as domestic currency in basis points.

Panel A: OIS Rates

USD 1M 0.745 0.158 1.405 5.269 0.068 2.371 7.234 2563 3M 0.745 0.166 1.387 5.305 0.069 2.399 7.419 2563 6M 0.757 0.187 1.363 5.329 0.065 2.412 7.537 2563 1Y 0.819 0.315 1.325 5.408 0.063 2.370 7.450 2563 EUR 1M 0.924 0.350 1.477 4.313 -0.351 1.444 3.465 2563 3M 0.923 0.356 1.487 4.351 -0.379 1.458 3.526 2563 6M 0.934 0.351 1.499 4.454 -0.413 1.452 3.547 2563 1Y 0.974 0.361 1.510 4.670 -0.463 1.407 3.520 2563 GBP 1M 1.086 0.460 1.607 5.938 0.195 2.204 6.013 2439 3M 1.293 0.460 1.842 6.003 0.176 1.767 4.225 2563 6M 1.062 0.475 1.564 5.953 0.138 2.242 6.187 2439 1Y 1.099 0.540 1.517 6.017 0.059 2.242 6.261 2439 AUD 1M 3.744 3.240 1.653 7.318 1.465 0.719 2.477 2563 3M 3.699 3.150 1.681 7.395 1.453 0.771 2.541 2563 6M 3.664 3.013 1.728 7.550 1.405 0.813 2.573 2563 1Y 3.684 2.920 1.805 7.850 1.340 0.791 2.488 2563 JPY 1M 0.143 0.078 0.173 0.630 -0.109 1.461 3.731 2557 3M 0.142 0.075 0.183 0.708 -0.151 1.450 3.920 2557 6M 0.141 0.073 0.197 0.738 -0.209 1.441 4.206 2557 1Y 0.190 0.098 0.252 0.858 -0.273 0.814 2.743 1676

Panel B: Exchange Rates

(38)

Table 12: Correlation table of EUR/USD, GBP/USD, AUD/USD and JPY/USD for the variables of the regression model in equation 6. These correlations are for the 3-month maturity.

EUR/USD Correlation

EUR CIP FX spread LOIS USD LOIS EUR CDS VIX

EUR CIP 1.000 FX spread 0.626 1.000 LOIS USD 0.791 0.687 1.000 LOIS EUR 0.574 0.637 0.799 1.000 CDS (US-FR) -0.006 -0.072 0.139 -0.109 1.000 VIX 0.419 0.529 0.676 0.812 0.018 1.000 GBP/USD Correlation

GBP CIP FX spread LOIS USD LOIS GBP CDS VIX

GBP CIP 1.000 FX spread 0.710 1.000 LOIS USD 0.795 0.658 1.000 LOIS GBP 0.643 0.574 0.864 1.000 CDS(US-UK) -0.139 -0.103 -0.276 -0.414 1.000 VIX 0.457 0.444 0.673 0.783 -0.481 1.000 AUD/USD Correlation

AUD CIP FX spread LOIS USD LOIS AU CDS VIX

AUD CIP 1.000 FX spread 0.666 1.000 LOIS USD 0.767 0.784 1.000 LOIS AUD 0.382 0.411 0.537 1.000 CDS (US-AU) -0.238 -0.428 -0.598 -0.272 1.000 VIX 0.631 0.788 0.793 0.540 -0.555 1.000 JPY/USD Correlation

JPY CIP FX spread LOIS USD LOIS JPY CDS VIX

(39)

Table 13: Time-series regression of daily EUR/USD CIP deviations using equation (6). This table shows the estimation results for a month, 3-month, 6-month and 1-year maturity of EUR/USD CIP deviations. The dependent variable, CIP deviations, is given in basis points. These results are with use of the CDS of Germany as proxy for the Euro zone. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. The regression is estimated for the period March 7 2007 - December 30 2016. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

(40)

Table 14: Descriptive statistics for period 1 (07-03-2007 until 30-12-2009) of the variables used in the regression model. The statistics are shown in basis points.

Panel A: Forward bid-ask spread

Variable Mean Median SD Maximum Minimum N

(41)

(42)

(43)

Table 17: Descriptive statistics of CIP deviation for each individual period. This data is used for the robustness test, where the statistics are shown in basis points

CIP Period 1

EUR/USD 3M 8.963 5.884 12.484 101.991 -1,461 736 GBP/USD 3M 6.783 3.722 11.003 102.854 -2,661 736 AUD/USD 3M 8.026 5.047 12.500 78.155 -6,381 736 JPY/USD 3M 12.060 8.374 14.683 106.614 -9,226 736 CIP Period 2 EUR/USD 3M 6.964 5.801 4.122 25.880 0,415 1042 GBP/USD 3M 2.103 1.829 1.724 11.264 -1,877 1042 AUD/USD 3M -1,528 -1,288 2,207 4,560 -9,759 1042 JPY/USD 3M 7,292 6,483 4,013 26,463 2,266 1042 CIP Period 3 EUR/USD 3M 8.164 6.488 6.391 28.124 -0,269 783 GBP/USD 3M 3.941 1.855 3.864 14.842 0.508 783 AUD/USD 3M -3,187 -3,710 2,627 4,909 -9,736 783 JPY/USD 3M 13,0137 10,641 7,212 31,675 3,534 783

(44)

Table 18: Estimation results robustness test GBP/USD for each period, where CDS of France is taken as proxy for Europe. It shows regression results for the 1-month, 3-month, 6-month and 1-year maturity where the dependent variable, CIP deviations, is reported in basis points. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

GBP/USD

(45)

Table 19: Estimation results robustness test AUD/USD for each period, where CDS of France is taken as proxy for Europe. It shows regression results for the 1-month, 3-month, 6-month and 1-year maturity where the dependent variable, CIP deviations, is reported in basis points. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

AUD/USD

(46)

Table 20: Estimation results robustness test JPY/USD for each period, where CDS of France is taken as proxy for Europe. It shows regression results for the 1-month, 3-month, 6-month and 1-year maturity where the dependent variable, CIP deviations, is reported in basis points. The standard errors are adjusted for heteroskedasticity using the Newey-West procedure. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

JPY/USD

(47)

Table 21: Estimation results Robustness test period 1 (07-03-2007 until 30-12-2009) for each currency. The dependent variable, CIP deviations, is reported in basis points. he standard errors are adjusted for heteroskedasticity using the Newey-West procedure. Numbers in parentheses are robust standard errors. ***, **, and * denote the 1%, 5%, and 10% significance level.

Period 1: 07-03-2007 until 30-12-2009

EUR/USD GBP/USD AUD/USD JPY/USD

(48)

Period 2: 01-01-2010 until 30-12-2013

(49)

Period 3: 01-01-2014 until 30-12-2016

(50)

Figure 11: Deviations from the CIP condition for GBP, AUD and JPY with the use of either LIBOR or OIS rates. These daily deviations in the CIP condition are shown in basis points. The covered interest rate parity implies that the deviations should be zero, where an upward movement implies that the US dollar had lower interest rate on the forward market.

(a) LIBOR CIP deviations GBP

(b) LIBOR CIP deviations AUD

(51)

References

Q Farooq Akram, Dagfinn Rime, and Lucio Sarno. Arbitrage in the foreign exchange market: Turning on the microscope. Journal of International Economics, 76(2):237– 253, 2008.

Naohiko Baba and Frank Packer. Interpreting deviations from covered interest parity during the financial market turmoil of 2007–08. Journal of Banking & Finance, 33(11): 1953–1962, 2009a.

Naohiko Baba and Frank Packer. From turmoil to crisis: dislocations in the fx swap market before and after the failure of lehman brothers. Journal of International Money and Finance, 28(8):1350–1374, 2009b.

Jonathan A Batten and Peter G Szilagyi. The recent internationalization of japanese banks. Japanese Economy, 38(1):81–120, 2011.

Ramprasad Bhar, Suk-Joong Kim, and Toan M Pham. Exchange rate volatility and its impact on the transaction costs of covered interest rate parity. Japan and the world economy, 16(4):503–525, 2004.

Jean-Marc Bottazzi, Jaime Luque, Mario Pascoa, and Suresh M Sundaresan. Dollar shortage, central bank actions, and the cross currency basis. 2012.

Markus K Brunnermeier and Lasse Heje Pedersen. Market liquidity and funding liquidity. Review of Financial studies, 22(6):2201–2238, 2009.

Markus K Brunnermeier, Stefan Nagel, and Lasse H Pedersen. Carry trades and currency crashes. NBER macroeconomics annual, 23(1):313–348, 2008.

PAtrick McGuire Claudio Borio, Robert McCauley and Vladyslav Sushko. Covered inter-est parity lost: understanding the cross-currency bias. BIS Quarterly Review, Septem-ber 2016.

Niall Coffey, Warren B Hrung, and Asani Sarkar. Capital constraints, counterparty risk, and deviations from covered interest rate parity. 2009.

William J Crowder. Covered interest parity and international capital market efficiency. International Review of Economics & Finance, 4(2):115–132, 1995.

Wenxin Du, Alexander Tepper, and Adrien Verdelhan. Deviations from covered interest rate parity. 2016.

Persistent deviations from the CIP condition