The interbank market risk premium, central bank interventions, and measures of market liquidity

Annika Alexius, Helene Birenstam and Johanna Eklund September 17, 2012
y

Abstract The risk premium on the interbank market soared during the nancial crisis, thereby creating a wedge that prevented the interest rates actually paid by consumers and investors to follow policy interest downward. Central banks were attempting to ameliorate the situation by supplying liquidity to the interbank market. This paper studies the Swedish interbank market risk premium using a unique data set on traded volume between banks and between banks and the Riksbank. We nd that more trade both between Swedish banks and between the banks and the Riksbank results in a lower interbank market risk premium. The main determinants of the Swedish interbank premium are international variables such as US and EURO area risk premia. International exchange rate volatility and the EURO/USD deviations from CIP also matters, while standard mesures of domestic market liquidity and domestic credit risk are typically insignicant.

Keywords: JEL classications: F31, F41

Annika Alexius, Department of Economics, Stockholm University, 106 91 Stockholm, Sweden. E-mail: annika.alexius@ne.su.se. y Johanna Eklund, Financial Stability Department, Sveriges Riksbank. E-mail: johanna.eklund@riksbank.se.

Introduction

The interbank market received relatively little academic attention until the nancial crisis 2007-09, when risk premia in this particular market sky rocketed. The resulting wedge between central bank policy interest rates and the interest rates paid by consumers and investors turn into a major concern during the recession because it hampered the expansionary eects of monetary policy. For instance, the Federal Funds Rate was lowered by 5.25 percent between September 2007 and December 2008, but bank lending rates only fell by 1.1 percent during the same period. Instead, the interbank market risk premium increased from a stable level around x basis points to a modern time record of 3.60 percent. Policy makers and researchers started to devote much more attention to the interbank market risk premium, how it is determined and what central banks can do to aect it. The relative importance of illiquidity versus credit risk (or other risks) is important not only from an academic point of view but also because dierent policy measures are called for depending on which factors that cause x in the interbank market. Above all, central banks have a variety of tools at their disposal for improving market liquidity, but these interventions are less eective if the interbank market risk premium is mainly determined by other variables than illiquidity. Taylor and Williams (2008) and others conclude that movements in the interbank market risk premium are mostly due to credit risk, while their measure of liquidity does not have signicant eects.1 Using dierent measures of both liquidity and credit risk, Schwartz (2010) nds that liquidity risk is the more important factor. Fukuda (2011) studies interbank rate rates
1

REFs nding that credit risk is important

across currency denominations to identify the eects. For instance, both the Tokyo Interbank Oered Rate (TIBOR) and the London Interbank Oered Rate (LIBOR) are available in both Yen and US dollars. He nds that counter party credit risk in a specic country aects the TIBOR-LIBOR interest rate dierential across currency denominations, while liquidity conditions appear to be currency specic and aect dollar denominated interest rates (TIBOR and LIBOR) dierently from Yen denominated interest rates. Michaud and Upper () document a long run relationship between credit risk and the interbank market risk premium but no signicant eects on day to day movements in interbank spreads. They also investigate a cross section of LIBOR interbank rates for dierent banks and nd that credit default swap prices for the borrowing banks were statistically unrelated to the risk premia of these banks. According to Angelini et al. (2011), two thirds of the soaring interbank spread during the crisis was instead due to increased risk aversion rather than credit risk or liquidity. The empirical results concerning the relative importance of market liquidity for the interbank market risk premium appear to be systematically related to the proxies used to capture the dierent variables. Credit risk is typically measured using data on bank credit default spreads (Taylor and Williams (2008), Fukuda (2011), Michaud and Upper (xx)). The main exception is Schwartz (2010), who calculates the spread between interbank rates paid by banks with high versus low credit ratings. Since the probability of default can be calculculated using CDS prices, several studies calculate rather than estimate the eect of counterparty credit risk on the interbank market risk premium (REFs). Capturing liquidity in an OTC (over the counter) market turns out to be more di cult. Data on the measures of liquidity frequently used in studies of stock markets (bid-ask spreads, traded 3

volume) are typically unavailable in case of the interbank market since it is dominated by bilateral transaction, detailed information about which are rarely collected. Taylor and Williams (2008), Fukuda (2011) and others estimate the impact of liquidity provisions by central banks and conclude that liquidity does not aect the interbank market risk premium if liquidity provisions do not have signicant eects. However, liquidity provided by central banks is not necessarily a perfect substitute for market liqudity. Following xx, Schwartz (2010) uses the spread between bonds with (ideally) the same expected cash ows but dierent liquidity charachteristics to capture the liquidity premium. Similarily, Valenzuela (2010) measures liquidity risk as the spread between US sovereign bonds and AAA bonds of international organizations. The bond market liquidity premium is presumably correlated with the interbank market liquidity premium but does not necessarily capture liquidity problems that are specic to the interbank market. This paper utilizes a unique data set on the transaction volume in the Swedish interbank market to investigate how liquidity, credit risk, Riksbank interventions, and other factors aect the interbank market risk premium in Sweden. Transaction volume data from the Riksbank are compared to the measure of interbank market liquidity risk used by Schwartz (2010) and Valenzuela (2010), the interest rate spread between bonds with identical pay o structures but dierent liquidity charachteristics. These two measures of liquidity are barely correlated with eachother, but the bond spread displays high correlations with measures of credit risk and measures of nancial market risk (such as stock market volatility). The single most important variable behind movements in the Swedish interbank market risk premium turns out to be the US interbank market risk premium, followed by international exchange rate.volatiliy and the deviations from covered in4

terest rate parity between the EURO and the USD. Measures of domestic liquidity have signicant but quantitatively small eects on the interbank market risk premium. Somewhat surprisingly, we nd that domestic credit risk measured as average credit default spreads for major Swedish banks does not aect the interbank market risk premium signicantly once we control for international variables. The Riksbank interventions result in a signicant reduction of the risk premium in most specications, but this nding is not fully robust. Data on traded volume are weakly correlated to the interest spread between similar bonds with dierent liquidity charachteristics. The latter measure displays much higher correlations to credit risk, stock market volatility and other measures of market risk than the volume measure. More trade between the banks and the Riksbank results in a lower internbank market risk premium, although the eect is signicant only at the 10 percent level and the magnitude is much smaller than for trade between banks. Domestic credit risk as measured by CDS prices does not appear to have signicant eects on the interbank market risk premium. This implies that methods based on rst deducting the theoretical credit risk premium as calculated using CDS data and estimate the eects of liquidity and other variables on the remaining part of the risk premium (as in the papers by x, y) may be less relevant.

Measuring interbank market liquidity

As a general notion, liquidity is dened as the ease with which assets are traded. An illiquid asset is more di cult to sell than a liquid asset. Similarily, it is more di cult to buy and sell assets on an illiquid market than 5

on a liquid market. This is often modeled as an exogenous transaction cost that may vary over time and/or between dierent assets and markets. The most common empirical counterpart is bid-ask spreads, data on which is typically unavailable in the case of over the counter (OTC) markets such as the interbank market. Rather than just assuming an exogenous transaction cost to model illiquidity, this cost can be derived endogenously through several dierent mechanisms. In models with asymmetric information and adverse selection, the transaction cost depends on the probability of dealing with an informed trader and on how much additional information she has. The quantitative importance of this mechanism can be captured by measuring the impact of an additional net order ow on market prices. The more information revealed by an additional unit of trade, the more it aects the equilibrium price and the higher is the transaction cost due to asymmetric information. A common empirical proxy is the average change in market price per unit of trade (Korajdsyk and Sadka, 2008). There are also more complex measures such as the dierence in bid-ask spread between small orders and large orders divided by the corresponding dierence in order size (Goyenko et al.). A second strand of literature on liquidity as an endogenously derived transaction cost focuses the inventory risk of market makers. It is more risky and hence more costly to hold an asset before being able to re-sell it if returns are volatile, traded volume or turnover is low and/or volatile, the more of this asset the market maker already holds and the less diversiable these risks are. The theoretical concept of liquidity here centers around the value of "immediacy", i.e. the willingness to pay for immediate trade with a market maker rather than delaying the transaction while nding a buyer. As above, empirical measures capture how much the market price 6

is temporarily depressed by an unexpected sell order. A standard proxy is again the average change in market price per unit of trade. The third mechanism for creating endogenous transation costs is particularily relevant to OTC markets, such as the interbank market. Search frictions and bilateral bargaining can cause dierences in return also between assets with identical expected cash ows if they have dierent liquidity charachteristics. The theoretical measure of illiquidity in these model is the time it takes before the asset can be sold. Assets that are traded more frequently have lower search costs and hence a lower liquidity premium. Vayanos and Wang (2007) develop a search-based model of asset pricing with risk neutral investors. RISK AVERSION? Du e, Garleanu and Pedersen (2005) assume that there is a market maker, while Du e, Garleanu and Pedersen (2007) assume that investors meet randomly and bargain biletarally. Vayanos and Weill () show how bonds with identical cash ows but dierent liquidity characteristics can trade at dierent prices. Securities with larger oat (supply) or higher trading volume have less severe search problems and correspondingly lower liquidity premia. Observable related proxies are asset supply, traded volume, or the return dierential between assets with the same cash ow but dierent liqudity charachteristics (such as the yield spread between soverreign and covered bonds). Given that liquidity is modelled as a transaction cost (exogenous or endogenous), the next question is how it is priced in equilibrium. The theoretical literature contains several points about the pricing of liquidity that are relevant to the choice of empirical proxies for liquidity in the interbank market. For instance, asset specic liquidity and market liqudity are two different concepts. In standard models, deterministic dierences in transaction costs (liquidity) between assets only commands a negligable (second order) 7

premium in equilibrium because optimizing investors allocate their portfolios to trade their liquid assets while holding their illiquid assets (Constantinides 1986, Heaton and Lucas, 1996). In contrast, market wide liquidity shocks can result in a substantial liquidity premium, especially if the shocks are permanent rather than transitory. There is also a conceptual dierence between the amount of liquidity (as measured by e.g. the volume of trade on a specic market) and the equilbrium price of liquidity (as measured by yield dierences between two otherwise identical assets with dierent liquidity charachteristics). In Acharya and Pedersen (2005), the asset specic liquidity premium is higher for assets whose illiquidity covaries positively with market illiquidity and market return.2 Investors pay a premium for assets that are liquid in times of low market return and vice versa; the liquidity premium is higher for assets that are more di cult to sell during market downturns. The largest component of the estimated asset specic liquidity premium is due to the covariance between asset specic liquidity and market returns (REFERENCE!). This may be relevant when measuring market liquidity as the dierence between two asset specic premia. The most common empirical measure of liquidity is bid-ask spreads. For instance, when a recent study by Goyenko et al. () investigate how well 24 dierent proxies of liquidity actually capture the desired phenomenon, bidask spreads are used as the "true" benchmark measure of liquidity against which other proxies are evaluated. A disadvantage of using bid-ask spreads
2

In contrast to the standard result, asset specic liquidity risk is priced in Acharya and

Pedersen (2005) because they assume that all assets have to be sold each period. Also, because liquidity is stochastic, the liquid asset can be hit by a shock to the transaction cost.

is that they include various technical transactions costs that may vary for reasons that are unrelated to true liquidity. Other frequently used measures of liquidity focus on the volume of trade, such as turnover or xx. Data on bid-ask spreads or traded volume are rarely available in the case of the interbank market because information about bilateral transactions between banks is typically not collected by any agency. Instead, two measures of liquidity frequently encountered in empirical studies of the interbank market are the interest rate spread between bonds with similar expected cash ows but dierent liquidity charachteristics, and the estimated eects of central bank liquidity injections. The idea behind the latter proxy is that the more the interbank risk premium is reduced by central banks provisons of liquidity, the larger is the liquidity premium. Liquidity provided by central banks is however not necessarily a perfect substitute for genuine market liquidity and can also have signalling eects of the opposite sign. For instance, Brunetti et al. (2011) nd that unexpected liquidity provisions by ECB during the nancial crisis crowded out privately supplied liquidity and increased market volatility by revealing that the central bank had new private information. x y z measure interbank market liquidity as the interest dierential between twpo bonds with (ideally) the same cash ows but dierent liquidity. There are several theoretical objections to this proxy. First, it focuses on the bond market rather than the interbank market. The market liquidity shocks of these two markets could be highly correlated, but this is an empirical issue that remains to be studied. Second, the spread betwen two bonds captures asset specic liquidity rather than market liquidity. Some part of the movements in this spread is probably due to market liquidity shocks, but how we do not know how much. Third, the yield spread between two 9

bonds concerns dierences in the equilibrium price of liquidity rather than market lliquidity. The empirical ndings of Acharya and Pedersen (2005) imply that the bulk of an asset specic liquidity premium is due to the covariance between the asset specic liquidity shocks and market returns. The covariance between market returns and the asset specic liquidity shocks on covered bonds is a rather dierent animal than the interbank market liquidity, which is what we want to measure. Finally, the bundle of bonds covered by a sovereogn government often includes publicly owned companies with a non-zero probability of being sold within ve years, such as SBAB in the case of Sweden.3 Hence, the premium on covered bonds may include various factors specic to such companies. Turning to the availability of various proxies for market liquidity, a rst observation is that the interbank market is dominated by bilateral trade and details of these transaction are rarely collected by any agency. There is a European electronic platsform e-mid, covering 15-20 percent of the Eurozone interbank market trade, from which various information is recorded. Because the share of the European interbank trade covered by e-mid uctuates over time and has fallen considerably since the 2007-8 nancial crisis, volume data from this source is less reliable than e.g. price data. A specic bank can be assumed to pay approximately the same price on transactions using this platform as through other means, but a change in traded volume on e-mid can be due to a change in traded volume on the EMU interbank market or to a change in the bankschoice of trading mode. For instance, zz argues that banks were more reluctant to use e-mid during the nancial crisis because they did not want information about their trading activitites to be
3

Note on SBAB

10

revealed. E-mid data are used by Angelini et al. (2011) and Brunetti et al (2011), but neither paper studies liquidity or the volume of trade since these papers focus on the eects of bank specic factors. Vento and Ganga (2009) have e-mid transaction data but do not use it in their empirical investigation.Michaud and Upper (2008) calculate four dierent proxies for liquidity using e-mud data: the number of trades, trading volume, bid-ask spreads, and the average price impact of trades, but only report eyeball econometrics of the nding that market liquidity is important to the interbank market risk premium. Hejmans et al have a similar data set for Netherlands from a platform called Target2, but also do not use them to study the relationship between market liquidity and the interbank risk premium. Finally, Poskitt (2011) constructs two non-standard empirical proxies for internbank market liquidity: Average bid-ask spreads as quoted by dealers, and the number of dealers active in the market, using intraday quote data from the oshore market for three month US dollar funding from the Thomson Reuters Tick History database of SIRCA. He nds signcant eects of liquidity on the theoretically derived non-default (or credit) component of the LIBOR-OIS spread.4 The market transaction volume is a straight forward measure of market liquidity. We are not aware of previous papers studying the liquidity premium in the interbank market using transaction volume data.
4

Following Bank of England (2007), Poskitt calculated the default related component

of the spread from CDS data and estimate the eects of liqudiity etc on the remaning part of the spread.

11

Data

Most international studies of the interbank market risk premium dene it as the (average) interbank rate minus the expected average short money market rate as measured by the Overnight Index Swap Rate, OIS. For instance, U.S. OIS rates are calculated based on the daily federal funds rate. Unfortunately, such as nancial intrument does not exist in Sweden. The instrument closest to the desired measure of expected future daily policy rates is the STINA swap, where the oating leg is the average overnight intebank rate. Following the Riksbank, the Swedish interbank risk premium is calculated as the STINA swap rate plus the spread between the shortest interbank rate and the repo rate. These data are collected from Thomson Reuters. The credit risk of domestic banks is proxied by an equally weighted average of the credit default swap (CDS) spreads of the 4 largest Swedish banks. These data are also collected from Thomson Reuters, as are US and EURO zone interbank market risk premia dened as the average three month interbank rate minus the corresponding OIS rate. Data on the deviations from covered interest rate parity (CIP) between the EURO and the USD, and betwen the SEK and the EURO, are calculated using three months Tbill rates and matching three months currency futures.5 We also use data on the implied volatility of ve major international exchange rates (not including the SEK), the Swedish stock market and the U.S. stock market. As a robustness check, the Swedish interbank market risk premium is proxied by the TED spread (the three month interbank market rate minus the three month T-bill rate) in Section x.y. All data are daily and collected from
5

Formula for CIP.

12

Thompson Reuter. The sample period is January 2, 2007 to November 22, 2011, see Table 1 for details. Two dierent measures of interbank market liquidity are investigated. First, following Schwartz (2010) and Valenzuela (2010), we calculate the interest rate dierential between Swedish government bonds and bonds of the same maturity (two years and ve years) that are guarenteed or covered by the Swedish government. This measure of market liquidity is compared to the volume of transactions on the Swedish interbank market. The data on overnight interbank market trade are based on records of loan advances and repayments processed in the Swedish large value payment system, RIX, managed by the Swedish Riksbank. In the RIX-system, large value payments are recorded one by one by the participants themselves. In that process the banks indicate the type of transaction that the payment originates from, e.g. if the transaction is an overnight deposit with another participant bank. The data includes all transactions registered as an overnight deposit in the RIX-system, made between two banks and with a maturity from one day to the next. Hence, the transactions contain bilaterally unsecured overnight interbank deposits between the banks members in RIX. However, to the extent that the banks make overnight deposits in the Riksbank in addition to the transactions made to manage their own liquidity position, these are most likely not caught in the data. Indicatively, such transactions are rare. In addition, the banks also make overnight transactions in the form of repos and swaps which are not captured in the data either. Finally, not all banks are members in the

13

RIX-system.6 Specically, only a few foreign banks are members in RIX.7 These banks instead balance their liquidity by taking overnight loans with their correspondent banks, which are members in RIX. Such transactions are captured only indirectly, as they aect the correspondent banks net balances and thus the correspondent banks overnight loan requirements. The sample period for the data on interbank market transaction volume is July 16, 2007 to November 11, 2010.8 Several studies have shown that the deviations from covered interest parity (CIP) are strongly correlated to the interbank market risk premium (REFs). We include the deviations from CIP between the EURO and the USD as a control variable given the assumption that Sweden is a small country that does not aect international developments. The deviation from CIP between the SEK and the EURO is however only included in a robustness specication due to the unclear causal relationship between this variable and the interbank market risk premium. During the nancial crisis, a shortage
6

The following banks are members of RIX: Bankgirocentralen, Citibank, CLS Bank,

Crdit Agricole. Danske Bank. DnB NOR Bank, EMCF, Euroclear Sweden, Fortis Bank SA/NV, Kommuninvest, Landshypotek AB, Lnsfrskringar Bank, NASDAQ OMX, Nordea Bank, Nordnet Bank, Nykredit Bank A/S, Riksbanken, Riksglden, Royal Bank of Scotland, SBAB, SEB, Skandiabanken, Svensk Exportkredit, Svenska Handelsbanken, Swedbank AB, landsbanken 7 The following banks are not RIX members but have some activity on the Swedish interbank market: Deutsche Bank AG, JP Morgan Chase Bank, UBS AG. 8 The Riksbank has detailed raw data on Swedish interbank market volumes contained in one excel le for each day also after November 2010. Unfortunetaly, we have not been able to pursuade them to convert this information to time series data or grant us access to the original les to update the data set. For the sample period that we have, the data on transaction volumes had already been collected manually from each excel le by Riksbank sta for a dierent purpose.

14

of dollars developed in Europe.9 Swedish banks and rms were constrained and could not borrow at the US market interest rates. This resulted in sizable deviations from covered interest parity all over Europe. The Riksbank stepped in and started lending US dollar to Swedish banks and rms to alleviate their liquidity problems. Data on the amount of dollar loans to the Swedish economy allows us to evaluate the eects of this monetary policy action as well. It turns out that the dollar loans reduced the deviations from CIP signicantly but do not have a signicant direct eect on the risk premium. Table 1 shows descriptive statistics, sample periods and unit root tests for all included time series. According to the ADF unit root test, the main variables are clearly I(0): Swedish and US risk premia, traded volume in the interbank market, international stock volatility, international exchange rate volatility. This is an unusal nding, but our sample period is also unusually long and covers calm periods both before and after the nancial crisis. The null hypotheses of the presence of unit roots is marginally rejected for several other variables, including the EURO area risk premium and liquidity measured as the spread between sovereign and covered bonds. Given the well known low power of the ADF test to reject the null hypothesis of a unit root in small samples (as documented by Kwiatkowski et al. (1992), DeJong et al., (1992), and others), the variables are treated as I(0) in most of the empirical tests. The main specication is however estimated also using rst dierences of the data. The Riksbank nancial crisis interventions or extraordinary lending to domestic banks displays a much high p-value than the other variables. This is not surprising since the interventions are zero
9

The dollar shortage was not unique to Eiurope. For instance, xx studies the deviations

from CIP between the US dollar and the Australian c during the nancial crisis.

15

for most observations, with a single peak during the nancial crisis. The effect of the interventions is investigated using several dierent specications including dummy variables. Table 1: Descriptive statistics Interbank spread, SE Interbank spread, US Interbank spread, EURO CDS spread, SE CDS theoretical premium Liq: Bond Spread, 2 y Liq: Bond Spread, 5 y Liq: Trade Vol., SEK*109 RB interventions, SEK*109 Stock volatility, int. Stock volatility, SE XR volatility, int CIP dev, SEK CIP dev, EURO VIX Sample period 07-01-01 to 11-12-30 07-01-02 to 11-11-22 07-01-02 to 11-11-22 07-01-01 to 11-11-21 07-01-01 to 10-11-29 07-01-02 to 11-11-22 07-01-02 to 11-11-22 07-07-16 to 10-12-30 07-01-02 to 11-11-22 07-01-02 to 11-11-22 06-06-02 to 11-06-20 07-01-02 to 11-11-22 07-01-02 to 11-06-17 07-01-01 to 11-06-17 07-01-03 to 11-06-30 Mean 33.34 49.76 54.00 98.58 17.67 27.37 56.00 17.7 48.2 0.237 25.93 0.419 0.361 0.331 25.19 Max 146.30 363.88 195.80 226.74 61.18 95.53 111.35 79.2 290.1 2.198 77.92 2.250 1.412 1.483 80.86 Min 1.00 4.14 5.30 19.81 1.01 -6.55 -11.20 0.051 0 -0.224 11.52 -0.430 -0.778 -0.737 9.89 Std 22.43 54.01 34.44 46.68 12.93 23.00 36.79 18.4 60.9 0.406 10.36 0.449 0.498 0.544 11.52 ADF -3.038 -2.964 -2.184 -2.210 -1.826 -2.296 -2.234 -4.438 -1.352 -2.990 -2.381 -2.747 -2.027 -2.081 -3.592

Daily data. ADF is the Augmented Dickey Fuller unit root test, where the number of lags is chosen according to the AIC. The 5

The cross correlations between the variables are shown in Table 2. The three interbank riskpremia for Sweden, the US, and the EURO area are highly correlated. Volatility measures also display correlations above 0.7 with the risk premia. Figures 1 through 5 show the variables in questions. It is clear that the Swedish interbank market risk premium is closely related to the corresponding premia in both the US and the EURO zone. During the 2008 crisis, Swedish markets reacted stronger to the US development that the EURO interbank market did. The recent EURO debt crisis has however aected Sweden more than the US. Liquidity measures as the yield 16

spread between sovereign and covered bonds display a similar pattern as the interbank market risk premia and the volatility measures, while the volume of trade on the interbank market falls during the nancial crisis and does not recover within the sample period. The Riksbank interventions are large relative to the volume of trade between banks, but only occur during the nancial crisis.

Empirical results

In order to capture the eects of domestic credit risk, liquidity risk, and central bank interventions on the Swedish interbank market risk premium, we need to control for other factors in uencing the risk premium and use appropriate econometric techniques and/or instruments to avoid endogeneity. For instance, it is not appriori clear whether higher credit risk as measure by the CDS spread of Swedish banks result in a higher interbank market risk premium, whether a higher risk premium results in higher CDS spreads, or whether both variables simply react to international shocks. The interbank markets display a high degree of international integration. In the case of Sweden, international variables can be assumed to be exogenous. Sweden is a small, reasonably stable country and domestic events have had negligable eect on international nancial variables. Endogeneity is obviously a problem in the case of domestic variables and the central bank interventions in particular. The Riksbank s liquidity interventions may decrease the risk premium, but it is likely that the Riksbank decides to intervene when the market risk premium is perceived to be excessive. In addition, all these variables react to the same shocks. News, for instance information about the Greek haircut negotiations halting again, are likely to aect both credit risk, 17

market liquidity, the Riksbank credit provisions, and the interbank market risk premium. Hence instrumental variables are used to estimate these relationships. The choice of instruments is discussed in some detail below, but the main idea is to use lagged variables. Current shocks do not aect lagged variables, and a nancial market shock in t + 1 is by denition unpredictable given the information available at t.

4.1

Main ndings

Several papers have studied the determination of interbank market risk premia using factor models. However, since the factors are not identied, these studies do not answer questions such as whether credit risk or liquidity are more important, or whether central bank interventions are eective. Ideally, one would obviously like to have a pure measure of e.g. domestic credit risk which is unaected by other variables. Given data on the CDS spreads of domestic banks, controlling for other variables (international risk premia and international exchange rate risk), and estimating the relationship using instrumental variables, the estimated coe cient is interpreted as the eect of domestic credit risk on the interbank market risk premium: (1) Pk
i=1 i Xit

= rpSE t

+ "t ;

where rpSE is the Swedish interbank market risk premium and X is a t

set of variables includes the US and EURO area interbank market risk premia, international exchange rate risk (measured as the implicit volatility of z major exchange rates, and the deviations from covered interest parity, CIP, between the EURO and the US dollar), and domestic credit risk, liquidity, and the Riksbank interventions. The implicit volatility of international

18

variables can be assumed to be unaected by shocks to the small Swedish economy. The relationship nevertheless has to be estimated using instrumental variable techniques since international shocks aect both Swedish and international variables simultanously. A suitable set of instruments for estimating (1) consists of variables that are uncorrelated with the error term and su ciently correlated with the X -variables. From theory, the realization of nancial variables in period t should be impossible to predict using the information available at t 1,

lagged variables are valid instruments. Given that the residuals typically display rst order autocorrelation but not higher order autocorrelation, we use lags two and above. Table x report the p-value of the J -test for overidentifying restrictions and the Crabb-Donald test for instrument validity. The latter approximately a multivariate equivalent to the F-test of the rst stage regression in a 2SLS in the univariate case. The Crabb-Donald test falls radically when higher lags than lag three are included. Hence the standard set of instruments in Table x consists of the second and third lag of all variables in the specication in question. In case of the Riksbank interventions, one and two month lags are used, since the volume of extraordinay loans to the banks is not altered on a daily basis. The eects of this variable is studied more extensively in Section z. More details about the set of instruments used in each specication are provided in the footnotes of Table x. Table c shows the main results. Interbank markets are highly integrated across borders and two main determinants of the Swedish interbank market risk premium are the US and EURO zone interbank market risk premia. These two variables are highly correlated (0.89) and their relative in uence varies across the specications. In column two, only international variables are included. US or international stock volatility has been excluded since it 19

lacks signicant eects on the Swedish interbank market risk premium once we control for the US and EURO interbank market risk premia. The Swedish interbank market appears to react more than the US and EURO markets to international exchange rate factors since both international exchange rate volatility and the deviations from CIP between the EURO and the US are signicant. This is consistent with the exchange rate being more important to a small open economy than to the larger and more closed economies. In the third column of Table z, domestic variables are added to the regression. Credit risk of Swedish banks as measured by the CDS premium, domestic liquidity measured by the yield spread betweem sovereign and covered bonds, and the implicit volatility of Swedish stocks are estimated with negative, insignicant coe cients. Hence these domestic variables that are available for the full sample do not appear to have independent eects on the Swedish interbank market risk premium. Since several studies use CDS data to extract a theoretical credit Columns four and ve includes the traded volume on the interbank market. This is a separate regression because data on this variable is only available up to 2010-12-30 and we do not want to use this reduced sample period for the other specications. Traded volume in the interbank market is expected to be negatively related to the risk premium and this parameter is indeed signicantly negative. Section z shows that this nding is reasonably robust, but also that it is possible to nd specications where the volume of trade does not have a signicant negative eect on the interbank market risk premium. Columns x and z contains standardized coe cients, where all variables have been divided by their standard deviations. This results in coe cients of comparable sizes in the sense that the relative importance of the dierent 20

variables for movements in the Swedish interbank market risk premium is captured. The relative sizes of the unstandardized variables do not capture how important the variables are, since the variables are measured in dierent units and have dierent variances. For instance, judging from the non-standardized coe cients in e.g. column s one wiould conclude that the Eurozone risk premium is more important to the Swedish risk premium, because its estimated eect is larger. However, even though all riskpremia are measured in the same unit (basis points, annualized), the US risk premium has a much larger variance and therefore accounts for more of the variation in the Swedish risk premium. Traded volume and the amount of Riksbank lending to banks is measured in SEK, which renders an evaluation of the relative importance of the variables based on estimated coe cients that are not standardized even more hazardous. Columns x and z show that the US risk premium is the most important determinant of the Swedish interbank market risk premium, followed by the implicit volatility of international exchange rates, the EURO/USD deviations from CIP, the covered bonds spread (although the estimated eect is negative and insignicant), interbank market transactions volume, Swedish CDS premium and then the EURO interbank market risk premium. During the nancial crisis in 2009, many central banks attempted to alleviate the credit crunch and keep the nancial markets xx. The Riksbank initiated major interventions (lending to domestic banks) in October 2008 and phased these loans out during 2010. As discussed above, we have data set covering the daily volume of credit (loans) from the Riksbank to domestic banks. This amount peaked at 300 000 000 000 SEK or x percent of GDP in July 2009. The results in columns six and seven of Table x indicate that the Riksbank interventions in the interbank market lowered the risk premium 21

signicantly. The robustness of this nding is investigated in Section x. Since the unit of measurement is millions of SEK, the magnitude of the eect of the interventions on the interbank market risk premium is rather modest, around 60 basis points or 0.6 percent in July 2009, when the interventions peaked. Since several previous studies have treated these variables as integrated of order one and hence used dierenced data, the corresponding regression is performed here as well. The Cragg-Donald test statistic for instrument relevance falls below 2 if two lags of the dierenced variables are used as instruments, but is even lower for more lags or only one lag. The residuals do not display rst or higher order autocorrelation according to LM and LjungBox tests, but are heteroscedastic according to the Engel () test. Hence the White () heteroscedasticity consistent covariance matrix is used. Furthermore, the xx GMM estimator where both the covariance matrix and the coe cient vector are updated between iterations does not converge. Hence, the yy GMM estimator where only the weighting matrix is updated is employed. Several of the main results from Table z still stand when dierenced data are used. For instance, the US interbank market risk premium remains signicantly positive and neither domestic CDS spreads or domestic liquidity measured as the yield spread between sovereign and covered bonds are signicant. On the other hand, traded volume on the interbank market is now insignicant, as are the Riksbank interventions and all other variables.10 Instrument relevance is so low for the regressions using dierenced data that
10

The Riksbank interventions typically only move marginally between the days within

each week, so creating a daily dierenced series of this variable is actually not a very sensible operation. However, we still replicate the full Table z also for rst dierences for completeness.

22

they are not reliable - we only included them to enable comparison to other studies.

4.2

Robustness of the eectiveness of the Riksbank s interventions

According to the main ndings in Section c, the credit provisions by the Riksbank to domestic banks resulted in a signicant reduction of the interbank market risk premium. Table z shows two types of robustness tests of the eects of the interventions. First, the interventions variable is dened in dierent ways: as a continous variable simply measuring the amount lended to the banks each day, the logarithm of the same variable (to capture decreasing eects of the amount of credit), as a discrete dummy variable taking the value 1.0 when a positive amount is lent to the market, and as a discrete dummy variable that covers the days from the announcement of the rst loans to the announcement of the discountinuation of the loans. We also include the square of the amount. The fact that this type of interventions where the central bank provides liquidity to the interbank market has signicant eects on the risk premium is often taken as evidence of a market liquidity premium. For instance, Fukuda s conclusions about the eects of market liquidity are based on the size and signicance of the eects of central bank liquidity injections. Similarily, ... . The funding liquidity is dened as the ability to raise cash at short notice either via asset sales or new borrowing The potential shortage of funding liquidity is often called rollover risk, because it may be di cult or impossible for a bank to roll over its short-

23

term borrowings

4.3

Comparing the two liquidity measures

The two measures of market liquidity used in this paper are (i) the relative returns to a more liquid and a less liquid bonds with (ideally) the same expected payos and (ii) the turnover or traded volume on the interbank market. As shown in Table x, they have a correlation of -0.22. The negative sign is expected since a high premium on the illiquid bond signies a negative liquidity shock, as does low traded volume on the interbank market. This correlation is however rather low. The bond spread displays high correlations to C Table x summarizes the observable charachteristics on the two measures of liquidity. First, it is clear that the interest rate spread between covered bond and government bonds is much more correlated to measures of credit risk (CDS spreads) and to Swedish and international market risk than the transaction volume data. The correlation coe cients range between 0.67 and 0.8 in the former case and between 0.04 and 0.4 for the volume data. The volume data are much more variable as shown by the standard deviations of x and y in Table c. The correlation coe cient of the two measures is only -0.22 but nevertheless highly signicant with a p-value of 0.000. Interestingly, the volume variable is positively correlated with the interbank risk premium (0.22), while the theoretical relationship would imply that more trade reduced the risk premium. The regression coe cients from previous tables are also summarized in Table z. The trade volume consistently has the expected negative sign, but is statistically signicant only in about half of the cases. The covered bond spread has negative coe cients in 3 of x

24

cases and is positive and signicant in c. It is also highly colinear with the other variables in the regressions. The VIF measure of collinearity for this variable is sm , d, across the specications in Table z, compared to only s ,df, for the volume variable. Dierent measures of market liquidity could have dierent eects because that they capture dierent aspects of liquidity rather than because one measure captures liquidity better than the other measure. We have not seen theoretical models of the interaction or covariance between credit risk and liquidity. Most formal models of the pricing of liquidity focus on equity rather than xed income assets that are a- icted by credit risk. The results from repeating the exercise using the spread between covered and sovereign 2 year bonds as measure of market liquidity are reported in Table z. The expected sign of the estimated coe cient is positive here, since a high liquidity risk premium on the less liquid covered bond should indicate in a higher interbank market risk premium. The covered bond spread has a signicant positive eect in two out of six specications: when no control variables at all are included (column two) and when only the EURO interbank risk premium is added (columns three). In all other specications with more control variables, the estimated coe cient is either insignicantly positive or signicanlty negative. This can be interpreted in two ways: either market liquidity does not aect the Swedish interbank market risk premium or the interest rate spread between these two bonds with similar expected cash ows but dierent liquidity charachteristics is not a suitable measure of market liquidity. The nal column shows the results from replacing the original data on the yield spread between Swedish sovereign bonds and Swedish covered bonds by a puried or orthogonal measure, obtained as the residuals from an OLS 25

with the same independent variables as in the previous column. Ideally, other forms of risk than liquidity risk aecting the data have been removed by this procedure. The results are however not aected as this measure of liqudity still enters with a signicant negative sign rather than the expected positive coe cient. The nding that the interest rate spread between covered and uncovered bonds is highly correlated to market returns and credit risk but not to the volume of trade in the interbank market supports the view that it does not really measure interbank market liquidity.

4.4

Domestic credit risk

As evident from Table 3, the Swedish interbank market risk premium is mainly determined by international variables. The credit risk of domestic banks is one of the domestic variables that could be expected to in uence it, but the estimated coe cients in Table 3 are all negative and insignicant. This result is somewhat surprising, but not unique. Several previous studies have documented signicant eects of credit risk measured as domestic bank credit default spreads on the interbank risk premium (Taylor and Williams, 2009, Fukuda 2011, ). However, Michaud and Upper () nd a statistically and economically insignicant coe cient on CDS spreads on interbank risk premia in a very thorough study of a cross section of banks. In addition, Angelini et al (2009) document insigicant eects of observable measures of borrower bankscreditworthiness. When the robustness exercise above is repeated for this variable the CDS premium actually has a positive signicant eect on the interbank risk premium in several specication with only few control varialbes. These results

26

are shown in Table x. In columns two to four, the only included control variables are international interbank risk premia. This results in positive signicant estimates of the eect of domestic credit risk. Signicance is then lost as international exchange rate risk is included in columns x and y. Adding Swedish variables but excluding liquidity measured as the spread between covered and sovereign bonds still yields positive estimated coe cients. The nal column shows that including this spread yields a negative coe cient. Given that the CDS spread itself is theoretically expected to have a non-linear eect on the interbank market risk premium while the estimated model is linear, we also calculate the theoretical credit premium as in xxx. This transformed variable should have a linear relationship to the interbank market risk premium. As shown in column z, credit risk remains insignificant. A quick look at the results for the rst dierences of the data in Table x also c a negative estimated coe cient for the Swedish CDS premium. Hence there is little evidence that the credit risk of domestic banks aect the Swedish interbank market risk premium.

Conclusions

Analyzing the interbank market risk premium is important for several reasons. The recent surge in attention is mainly due to the fact that during the nancial crisis, the exploding risk premium on the interbank market drove a wedge between policy rates and the interest rates paid by consumers and investor, rendering the expansionary monetary policy measures pursued during the crisis much less eective. There is a sizable and rather inconclusive literature on the relative importance of credit risk versus liquidity risk in the 27

determination of interbank market risk premia. The most common measure of market liquidity is the interest rate spread between similar bonds with dierent liquidity charachteristics. This captures the dierential between two asset specic liquidity premia, which according to the theoretical model of Acharya and Pedersen (2005) is a function of the covariance between asset specic liquidity shocks and market returns. In this paper we use actual interbank market transaction volume data collected by the Riksbank to measure interbank market liquidity. We nd that the transaction volume has a reasonably robust negative eect on the interbank market risk premium. In contrast, the interest dierential between similar bonds with dierent liquidity charachteristics enters with the wrong sign in most full specications, implying that a higher liquidity premium on covered bonds relative to sovereign bonds is associated with a lower risk premium. Domestic credit risk as measured by the credit default spread of domestic bank does not have a signicant eect on the interbank market risk premium in the benchmark specications. This result is however not fully robust as several specications with fewer control variables result in the expected positive signicant eect. Instead, the main determinants of the Swedish interbank market risk premium are international variables. Using standardized coe cients, it turns out that the US interbank risk premium is the single most in uential factor. International exchange rate risk measured as the implicit volatility of major exchange rates has a major in uence over and above the eects through the interbank market risk premia in the major countries. Hence the small open Swedish economy appears to be more sensitive to exchange rate risk or the factors determining exchange rate risk than the larger and less open US and EURO zone economies. The sample period includes the international nancial crisis of 2008-2009 and the 28

European debt crisis in 2011, when the domestic shocks generated by the Swedish economy or Swedish nancial markets were relatively minor. Hence domestic factors could well have a larger in uence during periods with more domestic disturbances. Finally we study the e casy of the Riksbank s interventions in the interbank market. Using data on the amount lent to Swedish banks during the nancial crisis, we nd that the Riksbank was able to reduce the interbank market risk premium signicantly. Since the logarithm of the interventions is the statistically most appopriate transformation to use, the eect appears to be non linear and decreasing in the size of the interventions. A simple dummy variable is however enough to document that the interventions had signicant eects, as long as several control variables are included. The eect of liquidity provided by the Riksbank does not dier signicantly in magnitude from the eect of increased transactions between banks, although the point estimate is several times larger. Finding valid and relevant instruments for the interventions is obviously troublesome. We use very long lags to instrument for the level of the Riksbank interventions. Furthermore the endogeneity bias would generate a positive coe cient since the Riksbank intervenes when the interbank market risk premium is high.

References
Angelini et al. (2011),"The interbank market after August 2007: What has changed, and Why?", Journal of Money, Credit and Banking 43(5), 923-958. Acharya, V. and Pedersen, L., (2005), "Asset Pricing with Liquidity Risk", Journal of Financial Economics, 77, 375-410. Bank of England, 2007. "An indicative decomposition of Libor spreads", 29

in Markets and operations, Quarterly Bulletin, 490-510. Brunetti, C., Filippo, M., and Harris, J., (2011), "Eects of Central Bank intervention on the interbank market during the Subprime crisis", The Review of Financial Studies 24(6), 2053-2083. Constantinides, G., (1986), "Capital market equilibrium with transaction costs," Journal of Political Economy 94(4), 842-862. DeJong, D, Nankervis, J., Savin, N. E. and Whiteman, C, (1992), "Integration versus trend stationarity in time series", Econometrica 60(2), 423433 Du e, D., Garleanu, N. and Pedersen, L., (2005), "Over-the-Counter Markets", Econometrica 73(6), 1815-1847. Du e, D., Garleanu, N. and Pedersen, L.,(2007), "Valuation in Overthe-Countermarkets", The Review of Financial Studies 20(5), 1865-1900. Fukuda, S., (2011), "Market-specic and currency-specic risk during the global nancial crisis: Evidence from the interbank markets in Tokyo and London", NBER Working Paper 16962. Goyenko, R., Holden, C., and Trzinka, C., "Do liquidity measures measure liquidity?" Heaton J., and Lucas, D, (1996)"Evaluating the eects of incomplete markets on risk sharing and asset Pricing.", Journal of Political Economy 104(3), 443-487. Heijmans, R., Heuver, R., and Walraven, D., (2010), "Monitoring the unsecured interbank money market using TARGET2 data", DNB Working Paper 276. Korajczyk, R., and Sadka, R., (2008), "Pricing the commonality across alternative measures of liquidity", Journal of Financial Economics 87(1), 45 72. 30

Kwiatkowski, D, Phillips, P., Schmidt, P., and Shin, Y., (1992), "Testing the nullhypothesis of stationarity against the alternative of a unit root: How sure are we that economic time series Journal of Econometrics 54, 159-178. Michaud, F. and Upper, C., (2008), "What drives interbank rates? Evidence from the Libor panel", BIS Quarterly Review, 47-58. Poskitt, R., (2011), "Do liquidity or credit eects explain the behavior of the LIBOR-OIS spread?", Mimeo Schwartz, K., (2010), "Mind the Gap: Disentangling Credit and Liquidity in Risk Spreads", University of Pennsylvania Wharton School of Business, Working Paper. Taylor, J., and Williams, J., (2009) "A Black Swan in the Money Market", American Economic Journal: Macroeconomics, 1(1): 58 83. Valenzuela, P., (2010), "Rollover Risk and Corporate Bond Spreads", Mimeo. Vayanos, D., and Wang, T., (2007), "Search and endogenous concentration of liquidity in asset markets", Journal of economic theory 136(1), 66-104. Vento, G., and Ganga, P., (2009), "Interbank market and liquidity distribution during the Great Financial Crisis: The e-Mid Case", Journal of Money, Investment and Banking 18, 68-94. Weill (2002) have a unit root?",

31

Table 2: Main results International variables Constant Interbank US Interbank EURO FX Vol, int CIP dev E Domestic variables CDS SE Covered bond spread vol_ib RB Vol. Svol SE Crabb-Donald Prob(J-test) # Obs 627.34 0.724 1161 -0.845 (-3.426 71.34 0.729 (1) 7.573 3.978 -0.038 -0.941 0.479 5.920 6.518 2.724 5.502 2.291 (2) 29.284 (5.792) 0.214 (4.452) 0.232 (2.954) 18.011 (4.408) 6.994 2.603) -0.021 (-0.554) -0.066 (-0.750) (3) Standardized 0.476 (3.508) 0.229 (3.428) 0.015 (5.195) 0.306 (3.288) 0.189 (2.920) -0.017 (-0.269) -0.213 (-2.733) -0.138 (-2.386) (4) 12.432 (4.180 0.107 (3.872 0.346 (5.437 13.685 (3.161 8.179 (2.996 -0.027 (-0.959 -0.203 (-2.685 -2.01E-10 (-2.827 (5) 13.888 3.136 -0.0484 -0.568 0.494 3.339 30.684 3.284 -74.971 -0.115 0.068 0.728 -0.287 -1.476

-1.73E-10 -2.513

22.06 0.598

22.03 0.561

20.68 0.348 914

The Crabb-Donald statistics is the multivariate equivalent of the (mimimum) rst stage F-test. Prob(J) is the marginal probability of the J-test for overidentifying restrictions. 1 The instrument set includes the second and third lags of all variables in our data set that are available for the full sample period. 2 The instrument is the same as above plus the second and third lags of traded volume in the interbank market. 3 The instrument set is the same as in 1 plus the one and two months lag of the Riksbank interventions. 4 The instrument set includes the rst lags of all variables in our data set and the one month lag of the Riksbank interventions. (5) The instruments include the seond lag of the variables in the regression, except for the Riksbank intervention, where the one month lag is used. The results are similar for longer lags of the interventions, but the Crabb-Donald test for weak instruments falls drastically. .

32

Table 3: Regression results Constant Interbank, US Interbank, EURO SVol, int FX Vol, int CIP dev E -80.43951 -1.690667 -21038.10 -1.894211 22.20056 1.547859 -35.29447 -0.010865 -0.693611 -1.850002 -0.123351 -0.345403 15.72421 1.252796 -194.2340 -0.075332 -0.667564 -1.692632 -0.007919 -0.028440 9.58E-11 2.465128 4.510564 0.683718 2558.414 1.677352 -0.316889 -1.369474 -0.045563 -0.210873 2 lags all 0.398653 1.441222 1.044992 2.117426 -0.078707 -0.111136 dito 0.042062 0.331229 0.291515 2.330640 0.294811 0.842694 IV 0.224747 1.109924 0.270507 2.267999 0.286906 0.844618 IV -0.100125 -0.615190 0.336351 4.564589 0.200415 1.010814

CDS SE Covered bond spread vol_ib RB Vol. Dummy FX vol Relevance J # Obs
Daily data .

1.16E-12 0.452422

2.476954 0.811893 1117

1.199940 0.813417 1116

0.743994 0.958834

1.870262 0.784111 901

33

Table 4: The eects of the Riksbank interventions: Robustness c RB Dummy RB Volume ln(RB Volume) rp EURO US rp FX vol int CIP dev EURO Likv_bond CDS SE 0.445 3.751 -0.196 -2.585 8.743 1.294 5.401 1.850 -0.021 -0.243 0.120 2.546 0.502740 3.356793 -0.178674 -2.001058 18.13349 2.326924 -7.919193 -0.013889 -0.024635 -0.173932 0.129262 1.310266 0.299959 2.833974 0.026369 0.405414 35.56204 4.143847 0.524859 4.711453 -0.173835 -2.027897 20.24512 3.699920 0.296532 3.716212 0.102755 2.898220 6.292808 0.816361 IV1 10.306 3.688 -13.781 -3.893 IV2 10.59977 2.408672 IV3 20.49639 4.564749 IV4 9.664116 3.214585 12.61751 4.413215 lag 42 0.138 0.072 lag 42 15.38460 8.217984

-2.09E-10 -2.991703

-2.35E-10 -4.485964

-1.85E-10 -6.232320 0.755 7.643 0.274 8.098 -0.630481 -7.229575 0.512623 7.176816 -0.121012 -3.246655 9.943127 3.391388 10.349 5.858695

-0.061991 -0.410956 -0.011298 -0.192945

-0.184732 -2.372819 0.158138 3.198973

0.035004 1.225093

relevance J # Obs

36.41 0.505

4.562875 0.259707 901

52.49013 0.139079 909

37.19828 0.724558 754

2.232464

188.49 0.228 905

86.57151 0.836663

The Crabb-Donald statistics is the multivariate equivalent of the (mimimum) rst stage Ftest. Prob(J) is the marginal probability of the J-test for overidentifying restrictions. 1 The instrument set includes the second and third lags of all variables in our data set that are available for the full sample period (i.e. not the trade volume). 2 The instrument is the same as above plus lag 42 (the two months lag) of the Riksbank interventions. 3 The instrument set is the same as in 2 but lag 3 of all variables are excluded. 4 The instrument set is the same as in 1 plus the ve and seven months lags of the Riksbank interventions.

34

Table 5: The eects of the Riksbank interventions: Robustness c Traded volume rp EURO US rp FX vol int Likv_bond CDS SE cip_dev-euro 21.98372 14.50755 2.64E-10 3.760736 4.125181 4.019064 -9.45E-11 -2.408070 0.623046 37.39842 19.46125 19.21403 -1.42E-10 -3.345652 5.858159 3.640053 -1.17E-10 -2.407606 0.556418 11.44806 0.044202 1.636739 4.636057 2.492640 -3.25E-11 -0.529141 0.512080 8.264375 0.050336 1.705819 4.954219 1.425899 8.397781 4.305879 -1.24E-10 -2.038322 0.476949 7.615452 0.082306 2.988595 7.843929 1.839905 -0.167847 -2.236710 8.352245 2.660350 -1.22E-10 -1.957330 0.474111 7.502997 0.082699 2.974909 7.909892 1.858262 -0.165322 -1.994780 0.000299 0.010438

0.299439 25.03452

1 5 0 3 0 3 1 5

8 3

relevance J

103.7961 0.396356

103.7054 0.787234

101.5860 0.248266

68.54005 0.661734

23.54295 0.666541

22.24900 0.607099

22.26005 0.531941

1 0

Weekly data on returns to investments in U.S. and German ten-year zero coupon government benchmark bonds from Dahlquist et al. (2000).
f d Regression: st+ st = + rt rt + ut+ . + + t-statistics in brackets, p-values in parentheses. Columns four and ve contain Wald test statstics of the null hypotheses. The nal column contains the Sargan test of overidentifying restrictions.

35

Table 6: The eects of the Riksbank interventions: Robustness c Liq 2y spread rp EURO US rp FX vol int Likv_bond CDS SE CIP_Dev-E 7.006104 2.719712 0.484061 7.271180 9.749480 5.601126 0.117815 2.148282 0.292120 6.212194 18.57127 15.30395 0.014628 0.417016 3.939219 2.651042 0.022257 0.420672 0.598188 11.33996 0.009970 0.341094 4.542884 3.132220 -0.122589 -1.626472 0.513261 9.850130 0.041596 1.580327 11.22588 2.994119 2.436908 0.919998 -0.167102 -2.022280 0.515984 9.657757 0.034792 1.309935 10.94605 2.993773 0.031592 1.055717 ortho 5.208724 2.114610 -0.107596 -1.875115 0.542785 7.511017 -0.115628 -2.913214 8.521701 3.055751

0.228653 11.84335

-0.020697 -0.756734 838.7955 4.225953

relevance J

1291.641 0.244974

846.5931 0.263729

940.0514 0.485892

195.5917 0.551647

93.05970 0.606032

91.80553 0.565021

141.2430 0.754249

Weekly data on returns to investments in U.S. and German ten-year zero coupon government benchmark bonds from Dahlquist et al. (2000).
f d Regression: st+ st = + rt rt + ut+ . + + t-statistics in brackets, p-values in parentheses. Columns four and ve contain Wald test statstics of the null hypotheses. The nal column contains the Sargan test of overidentifying restrictions.

36

Table 7: The eects of the Riksbank interventions: Robustness c CDS SE credit rp EURO rp US FX vol int cip E likv 2y 0.553857 21.67773 0.256800 15.44961 0.466635 8.397692 0.082679 2.490179 0.407459 5.667398 0.093403 2.765957 5.363379 1.426068 147.3399 0.603268 -4.394894 -1.800772 0.112314 4.182207 4.274414 1.523503 0.154402 5.336724 -1.692548 -0.720320 0.091988 3.218963 0.778755 0.272908 0.059640 1.763875 9.036669 4.175232 5.205030 2.040579 -0.007424 -0.278494

-0.315225 -2.708260 0.321214 4.656136 0.139726 4.358464 14.07723 4.270341 355.6268 1.568778

7.668535 11.98834 -0.123069 -2.989171 11.98834 3.283479 761.5312 3.650274 -0.104258 -1.809445

relevance J

2974.331 0.616101

2895.365 0.437447

289.6359 0.737211

211.5479 0.574006

206.7498 0.289908

141.2430 0.765028

Weekly data on returns to investments in U.S. and German ten-year zero coupon government benchmark bonds from Dahlquist et al. (2000).
f d Regression: st+ st = + rt rt + ut+ . + + t-statistics in brackets, p-values in parentheses. Columns four and ve contain Wald test statstics of the null hypotheses. The nal column contains the Sargan test of overidentifying restrictions.

37

Table 8: Unconditional correlations

38
Daily data

Risk premium, SE Risk premium, US Risk premium, EURO CDS spread, SE theoretical CDS premium Liquidity: 2 year bonds Liquidity: 5 year bonds Liquidity: Volume Stock volatility, int. Stock volatility, SE XR volatility, int Riksbank interventions CIP deviation, EURO CIP deviation, SEK VIX

RP US 0.762 1 0.889 0.265 0.300 0.380 -0.117 0.188 0.741 0.838 0.507 0.169 0.568 0.235 0.785

RP E 0.864 0.889 1 0.265 0.332 0.414 -0.020 0.101 0.810 0.849 0.614 0.177 0.585 0.215 0.810

CDS SE 0.387 0.265 0.402 1 0.989 0.414 0.716 -0.435 0.605 0.529 0.720 0.456 0.152 -0.080 0.598

Liq 2y 0.426 0.380 0.414 0.414 0.786 1 0.738 -0.507 0.852 0.530 0.852 0.478 0.069 -0.075 0.676

Liq 5y 0.117 -0.117 -0.020 0.716 0.711 0.738 1 -0.695 0.286 0.082 0.581 0.377 -0.210 -0.134 0.266

VolIB -0.0166 0.188 0.101 -0.435 -0.406 -0.507 -0.695 1 -0.193 -0.033 -0.492 -0.683 0.385 0.420 -0.189

Svol, int 0.727 0.741 0.810 0.605 0.585 0.946 0.286 -0.193 1 0.946 0.834 0.363 0.314 -0.030 1.000

FXVol 0.622 0.507 0.614 0.720 0.286 0.852 0.581 -0.492 0.834 0.724 1 0.573 0.221 -0.034 0.841

VolRB 0.116 0.169 0.177 0.456 0.574 0.478 0.377 -0.683 0.363 0.377 0.573 1 0.385 0.420 0.424

CIP, E 0.590 0.568 0.585 0.152 0.104 0.069 -0.210 0.385 0.314 0.394 0.221 0.385 1 0.854 0.314