An International Study of Causality-In-Variance: Interest Rate and Financial Sector Returns

Transcription

1 An International Study of Causality-In-Variance: Interest Rate and Financial Sector Returns V. T. Alaganar State Street Global Advisers Level 38, Aurora Place 88 Phillips Street, Sydney 2000, AUSTRALIA Ramaprasad Bhar School of Banking and Finance The University of New South Wales Sydney 2052, AUSTRALIA Abstract: We demonstrate that causality-in-variance test could be employed to model the direction and lags in information flow between two variables and to avoid misspecifications. We apply this methodology to test the causality between the financial sector returns and interest rates of the G7 countries and show that the direction and the lead/lag structure of causality in the mean and the variance are more complex and dynamic than that have previously been reported. In most cases, we found two-way information flow both at the mean and the volatility level. Causality results give us insights into (i) how and when information is impacted on different market segments, and (ii) design more objective bi-variate models with the appropriate lag structure. Corresponding author

2 An International Study of Causality-In-Variance: Interest Rate and Financial Sector Returns Abstract: We demonstrate that causality-in-variance test could be employed to model the direction and lags in information flow between two variables and to avoid misspecifications. We apply this methodology to test the causality between the financial sector returns and interest rates of the G7 countries and show that the direction and the lead/lag structure of causality in the mean and the variance are more complex and dynamic than that have previously been reported. In most cases, we found two-way information flow both at the mean and the volatility level. Causality results give us insights into (i) how and when information is impacted on different market segments, and (ii) design more objective bi-variate models with the appropriate lag structure.. 2

3 I. Introduction A change in interest rate affects the value of financial sector firms for many reasons. One source is the risk created by mismatches in maturities of assets and liabilities held by the firm. 1 The operations and the asset-liability structure of financial sector firms differ depending upon the country, mainly due to differences in the degree of national and international financial market integration and government regulations. Other factors that affect the relationship between financial sector returns and interest rates are default risk on bank loans and composition of holdings such as real estate and marketable securities. There is a large body of research literature in finance that examines the relationship between interest rates and banking sector returns (For example, Sweeney and Warga (1986), Elyasiani and Mansur (1998)). Some studies such as by Flannery and James (1984) and Bae (1990) focus on the decomposition of interest rate changes into expected and unexpected components and associating these changes with banking stock returns. The methodologies used for defining and constructing the expected and unexpected interest rate change series vary and so are the conclusions about the relationship between interest rate changes and banking sector returns. Another approach involves a two-factor model with the market index and interest rate as factors (Madura and Zarruk ). Researchers such as Kane and Unal (1988) and Kwan (1991) employ models that reflect the time varying nature of the sensitivity of stock returns to interest rates. Elyasiani and Mansur (1998) assume that interest rate is an exogenous process and estimate its volatility using an ARCH (1) model. The estimated conditional volatility of the interest rate is used in the second stage regression in a GARCH-M framework. But, they do not incorporate the market index as a source of systematic risk. Tai (2000), on the other hand, adopts a GARCH framework for interest rate change and carries out the joint estimation along with the equity market return. Tai claims that in two-stage estimation approach, as in Elyasiani and Mansur, the disadvantage is that cross-asset correlations are ignored and this may sacrifice efficiency of estimation. Joint estimation process, however, involves a large number of parameters to be estimated which has to be constrained to avoid computational difficulties. 3

4 In this paper we examine the causality between the interest rate and aggregate financial sectors (banking, insurance and financial services) of the seven OECD countries using a recent causality-in-variance test following the procedure proposed by Cheung and Ng (1996). Traditional Granger causality test focuses on the mean changes, while the causality-in-variance examines the conditional volatility dependence between two variables. The knowledge of how volatility between two variables (or markets) are related is important in that the second-moment or variance is directly linked to information flow 2. Causality-in-variance tests we employ can detect the direction of causality as well as the number of leads/lags involved. The main advantage of this test is the flexible specification of the innovation process and the nondependence on normality. Our results are useful for practitioners and academics in understanding how interest rate and financial sector return volatility are related and how this relationship differs across different industries in the financial sector within the G7 countries. This is in contrast to the most reported studies, which mainly use US data for empirical analysis. However, it is not the intent of this article to analyze global impact and/or transmission of interest rate impacts across countries. The findings here will help understand earlier reported results in view of the more robust causality tests adopted. This knowledge will further assist in the design of better models of interest rates and equity returns, hedging and risk management tools in banking and portfolio management. Our paper is organized as follows: Section II contains a selected literature review. Section III discusses the data followed by the methodology in section IV. Section V presents the results. We conclude the paper in section VI. II. Literature Review Sweeney and Warga (1986) regress stock returns of 21 industry portfolios against the market and a series of simple changes in the long-term interest rates. Their results indicate that only stocks of electric utilities, and the banking, financial and real estate sectors are consistently sensitive to interest rates over the period Madura and Zarruk (1995) show that interest rate risk is greater for non-us banks than for US banks. This variation is due to the differences in risk-based 4

5 capital requirements among banks in different countries. They also claim that non-us banks are sensitive to domestic as well as international interest rates. Akella and Chen (1990) demonstrate that bank share return tends to exhibit more sensitivity to changes in long-term rates than shortterm rates. Faff and Howard (1999) find that Australian bank stock returns are more sensitive to interest rates and this sensitivity is not stable over different periods. There are differences in the banking and financial system and regulations that could explain the non-uniform behaviour of financial sector returns across countries. La Porta, Lopez-de-Silanes, Shleifer, and Vishny (1997) suggest that the legal systems of different countries offer different degrees of protection to outside investors. If the country risk is so dominant, any differences in returns of banks between countries may be common to all firms in the country and not specific to the banks. Allan and Gale (1995) distinguish between transactional and relationship banks. Relationship banks such as German Hausbanks and Japanese main banks provide both debt and equity financing to their clients. These banks have in-depth involvement with their clients in terms of ties, directorship and managerial positions etc. Transactional banks such the banks in the UK, USA, Australia, and Canada provide short-term bank loans to corporations. These banks do not have extensive involvement with corporate management. Lee (1989) uses a GMM procedure with time-varying risk premia, and conditional heteroskedasticity in the disturbance for models of interest rates. He claims that allowing for conditional heteroskedasticity is more important than incorporating the time varying risk premia. Elyasiani and Mansur (1998) employ a GARCH-M technique to investigate the effect of interest rate and its volatility on bank stock returns. They found that changes in interest rate and its volatility to have a direct impact on the first and the second moment of bank stock returns, respectively. One of the shortcomings of this research is that it does not use the market index as an explanatory variable in the bank stock return equation. Besides, this model pre-specifies the direction of possible causality. Tai (2000) examines the role of interest rate and exchange rate on bank stocks using a number of techniques. His NLSUR (Non-Linear Seemingly Unrelated Regression) via the GMM approach indicates that the interest rate is the only significant factor. In his MGARCH-M approach, 5

6 however, he finds that time varying interest rate, exchange rate and the market risk premium to be related to the stock returns. In order to minimise the computational difficulties, Tai imposes constraints on the parameters of the system. However, he does not provide all the parameter estimates of the volatility equation and hence it is not clear how many of the estimated parameters are significant. Tai claims that in the two-stage estimation approach, the disadvantage is that the cross-asset correlations are ignored and this may sacrifice efficiency of estimation. On the other hand, his joint estimation process involves a large system to be estimated which has to be constrained to be computationally tractable, and thus defeating the very gain it was expected to deliver. III. Data We obtain the daily values of the total market index, total financial sector and the two subsectors in the financial sector, banking and insurance, for the seven OECD countries from Datastream International. Datastream claims that its industry classification system is more detailed and consistent across countries, which minimises the risk of misclassification of firms. Datastream classifies indices into one of six levels. Level 1 is the market index for a country and levels 2 and 3 are general industry groupings. At each additional level there are more disaggregated industry definitions until the most disaggregated industry classification, level 6. We use the terms megasector and industry to represent level 2 and 6 classifications respectively. The term sector is used in a general sense. Griffin and Karolyi (1997) argue that most disaggregated industry indices should be used to examine industry effects. We use the aggregate financial sector index at level 2 and retail banking and insurance sector at level 6 in our study. At level 6, we limit the study to banking and financial sectors due to lack of complete data over the test period for other industries such as life assurance, investment companies and real estate. Appendix I contains the Datastream codes used to identify the sectors in each country in our sample. We calculate the weekly returns of sectors using closing prices of consecutive Wednesdays, which minimise the impact of weekends and holidays, which mostly fall on Mondays 3. We also calculate weekly holding period return for the ten-year bond and the threemonth bill return for each of the countries. Period covered in the study is from January 1990 to December The sample size is, therefore, 574 for all data sets. 6

7 IV. Methodology The method developed by Cheung and Ng (1996) relies on analysing cross-correlation function of the standardised residuals from the univariate models of a pair of series under consideration. It does not depend on simultaneous modelling of the inter- and intra- variable dynamics. Adequacy of each univariate model may, however, be established with the help of portmanteau statistics computed from the standardised residual as well as from the squared standardised residual. Since simultaneous modelling is not required in order to study causality, this method is suitable for dealing with relatively large number of series. Besides the test statistic has a well-defined asymptotic distribution. Following the approach in Elyasiani and Mansur (1998), and Tai (2000) we model the long-term interest rate as a GARCH (1,1) process. Also, Gray (1996) reports that the interest rate model may be augmented to follow the square root GARCH (1,1) process. This process maintains the essential characteristic of the GARCH effect including the important persistence effect. Additionally, it incorporates the interest rate level effect as in the square root process of Cox, Ingersoll and Ross (1985). Incorporating these characteristics as in Gray (1996), the model for the long-term interest rate is given by the following equations: r = r r =φ + φ r +ε, ε ~ N(0, σ ) (1) 2 t t t t 1 t t t σ = a + a ε + a σ + a r. (2) t 0 1 t 1 2 t 1 3 t 1 We adopt the standard market model for the financial sector returns with GARCH (1,1) errors. Representing the sector excess return by x t and the market excess return by m t, the following equations describe the return generating process for each of the banking, financial services, and the insurance sectors (excess returns are with respect to short-term interest rates 4 ): x =α+β m +η, η ~ N(0, ν ) (3) 2 t t t t t 7

8 ν = b + b η + b ν. (4) t 0 1 t 1 2 t 1 εt The standardized residuals for the long-term rates are generated from and that for the sector σ ηt return are generated from. These residual series are used to establish the appropriateness of ν t the models with the help of Ljung-Box statistics. In addition, these form the basis of Cheung and Ng (1996) measure of pair-wise causality tests. They show that under the no-causality hypothesis, the cross correlation at different lags are not significantly different from zero and the statistic T ϒ 1,2 (k) is asymptotically normally distributed, where ϒ (k) 1,2 is the cross correlation t between the series 1 and 2 at lag k. Cheung and Ng 5 show that this result is robust to distributional assumptions. The causality in mean is tested using standardised residuals and the causality in variance is tested using squared standardised residuals. In the next section we discuss the results of these causality tests for each of the seven member countries and for each of the three sectors. V. Results Table 1 contains the results of the diagnostics tests of the univariate models given in equations (1) and (3) for interest rates and sector returns respectively. Ljung-Box test with 36 lags is conducted on standardised residuals from the respective models to detect serial correlation and the resulting test statistic is labelled as LB(36). Similarly, LB 2 (36) is used to determine whether there is any serial correlation in the squared standardised residual series. Long-term yield, which is assumed to follow an AR-GARCH(1,1) process, appears to have a good fit as seen by both the LB(36) and LB 2 (36) statistics with just one exception for LB 2 (36) statistic for Italy. In short, there is no significant serial correlation or heteroskedasticity in the residuals and therefore the AR-GARCH(1,1) process selected describes the long-term yield formation process for all countries adequately. We employ a GARCH(1,1) model to represent sector returns. Out of 21 country-sector combinations, model adequacy is supported by 17 cases. The cases involving 8

9 non-significant LB(36) statistic, which suggests model inadequacy at a 5% level of significance, is limited to Canada s banking and insurance industries, France s and USA s financial megasectors. However, the LB 2 (36) statistic for all 21 cases of sector returns support the chosen GARCH(1,1) model specification at 5% level of significance. We present the causality in mean and variance test results in Table 2 through Table 8 for Canada, France, Germany, Italy, Japan, UK, and USA respectively. Lags are measured in weeks, which range from -12 to +12. Significance at positive (negative) lags implies that the directions of the causation are from (towards) the long-term yield to (from) the sector returns at the mean or variance level. Significance at zero lag/lead implies cotemporaneous two-way causality or current innovations in one variable causing current innovations in the other variable. In Table 9, significant leads, lags and cotemporaneous causality for the seven countries are summarized. The cross-correlations of standardized residuals reveal that feedback from/to the interest rate on sector returns is not uniform across the G7 countries. In terms of the direction and the lead/lag structure of causality, the interaction between interest rate and sector returns appears to be more complex and dynamic than the standard ARCH and GARCH representations could reveal. The shortest lag of 1 week in causality at the mean level from interest rate to sector returns is seen with all US groupings. In general, causation is more prevalent at the mean level. It also appears some form of mean level causation exists for all cases showing second order causation. In the case of Canada, France, Germany, and the USA, financial megasector as well as banking and insurance industries show two-way causality at the mean level. Similarly, banking and insurance industries exhibit causality at the mean level across all countries with the exception of Japan. All sectors in Japan display the lowest levels of causation, suggesting that interest rate innovations and sector returns are decoupled. This is an interesting result given the recent history where interest rate changes have failed to vitalize the Japanese economy. Another insightful aspect of our result is that the megasector/industry returns cause innovations in interest rates. This causation is most pronounced for France, Germany, and the USA at the mean level. For USA, both mean and volatility causation from sector returns to interest rates spans the maximum lead allowed in our model, which is 12 weeks 6. In empirical research 9

10 relating to interest rates and equity returns, it is common to assume one-way causality from interest rates to equity returns. This traditional assumption fails to recognize the forward-looking nature of the equity markets where current prices reflect expected future interest rate and the state of the economy. We can make a strong argument for the value of banking firms to be more forward looking given the relationship between the banking profits and the levels of interest rates. For example, higher interest rates and slowing economy could mean lower profits and higher defaults. In short, interest rate research that attempts to model the innovations (unexpected changes) in interest rates should incorporate lagged equity returns as additional explanatory variables. Cotemporaneous causality at the mean level is evident for all sectors in Canada, France, Germany and the USA. However, second order cotemporaneous causality is observed only in 3 cases out of a total of 21 cases. This observation has important implications towards models involving lagged mean and volatility variables. The absence of significant causality in volatility in Japan does not justify the use of GARCH type of models for financial sector returns in Japan. However, we observe significant causality in volatility at different leads and lags in all other countries, which provides a basis for applying a GARCH framework as modeled by Elyasiani and Mansur (1998). Volatility feedback that occurs at long lags (example, banking firms in USA, Italy, France and Canada) reflects volatility persistence. Causality in variance from sector returns to interest rates is most striking for Germany followed by the US, France, UK and Canada. VI. Conclusion In this article we extend the debate over causal relationship between interest rate and financial sector returns and demonstrate that the interaction between these two variables is more complex than that has so far been reported. We adopt a recent and robust test of causality to explore any feedback effect between these two important variables. The approach does not depend upon joint modelling but the dependence information generated may be used to develop better bi-variate models. 10

11 We document the general direction and the lead/lag structure of causality in the mean and the variance between financial sector returns and interest rates for the G7 countries in the OECD. In most cases, we find two-way information flow both at the mean and the volatility level 7. Most researchers in this area tend to employ models that pre-specify the causal link, and the estimation results are used simply to establish their statistical significance or otherwise. Causality results obtained in this paper using Cheung and Ng (1996) procedure give us insights to design more objective bi-variate specification with the appropriate lag structure. Furthermore, our procedure provides the basis for including volatility as a factor in the financial sector asset pricing models. The analysis also reveals feedback effect between the financial sector equity return and interest rate, which has not been reported in earlier studies. 11

12 References: Akella, S.R., and Chen, S, 1990, Interest rate sensitivity of bank stock returns: specification effects and structural changes, Journal of Financial Research 13, Allen, F., and Gale, D., 1995, A welfare comparison of intermediaries and financial markets in Germany and the US, European Economic Review 39, Bae, S.C., 1990, Interest rate changes and common stock returns of financial institutions: Revisited, Journal of Financial Research 13, Cheung, Y, and Ng, L.K., 1996, A causality-in-variance test and its application to financial market prices, Journal of Econometrics 72, Cox, J. C., Ingersoll, J. E. and Ross, S. A. (1985), A theory of term structure of interest rates, Econometrica, 53, Elyasiani, E., and Mansur, I. 1998, Sensitivity of the bank stock returns distribution to changes in the level and volatility of interest rate: A GARCH-M model, Journal of Banking and Finance 22, Engle, R.F., Ito, T., and Lin, K.L., 1990, Meteor showers or heat waves? Heteroskedastic intradaily volatility in the foreign exchange market, Econometrica 58, Faff, R.W., and P.F. Howard, 1999, Interest rate risk of Australian financial sector companies in a period of regulatory change, Pacific-Basin Finance Journal 7, Flannery, M.J., and C.M. James, 1984, The effect of Interest rate changes on the common stock returns of financial institutions, Journal of Finance 39,

13 Gray, S. F., 1996, Modelling the conditional distribution of interest rates as a regime switching process, Journal of Financial Economics, 42, Griffin, J. G. and G. A. Karolyi, 1997, Another look at the role of industrial structure of markets for international diversification strategies, Arizona State and Western Ontario Working Paper. Kane, E.J., and H. Unal, 1988, Change in market assessment of deposit institutions riskiness, Journal of Financial Services Research 1, Kwan, S.H., 1991, Re-examination of interest rate sensitivity of commercial bank stock returns using a random coefficient model, Journal of Financial Services Research 5, La Porta, R., Lopez-de-S., Shleifer A. and Vishny, R.W., 1997 Legal determinants of external finance, Journal of Finance 52, Lee, B.S., 1989, A non-linear expectations model of the term structure of interest rates with time varying risk premia, Journal of Money, Credit and Banking, 21, Madura, J., and Zarruk, E.R, 1995, Bank exposure to interest rate risk: a global perspective, Journal of Financial Research 18, Ross, S.A., 1989, Information and volatility: the no-arbitrage Martingale approach to timing and resolution irrelevancy, Journal of Finance 44, Sweeney, R.J., and Warga, A.D., 1986, The pricing of interest rate risk: Evidence from the stock market, Journal of Finance 14, Tai, C. S., 2000, Time varying market, interest rate, and exchange rate risk premia in the US commercial bank stock returns, Journal of Multinational Financial Management, 10,

14 Appendix I: Datastream Codes Country Banking Insurance Financial Total Market Short Rate Long Rate Canada BANKSCN INSURCN TOTLFCN TOTMKCN CNTBL3M CN13895 France BANKSFR INSURFR TOTLFFR TOTMKFR FRVGTHB FRBRYLD Germany BANKSBD INSURBD TOTLFBD TOTMKBD FIBOR3M BDBRYLD Italy BANKSIT INSURIT TOTLFIT TOTMKIT ITIBKSN ITLGLTB Japan BANKSJP INSURJP TOTLFJP TOTMKJP JPIBK3M JAPGLTB U.K. BANKSUK INSURUK TOTLFUK TOTMKUK LDNTB3M UKMGLTB U.S.A. BANKSUS INSURUS TOTLFUS TOTMKUS USFTB3M USAGLTB 14

15 Table 1 Univariate Model Diagnostics for Member Countries Long Yield Banking Sector Insurance Sector Financial Sector Canada LB (36) LB 2 (36) France LB (36) LB 2 (36) Germany LB (36) LB 2 (36) Italy LB (36) LB 2 (36) Japan LB (36) LB 2 (36) U.K. LB (36) LB 2 (36) U.S.A. LB (36) LB 2 (36) Diagnostic tests are carried using standardized residuals from the respective models. LB (36) is Ljung-Box statistic at lag 36 to detect serial correlation in the standardized residual series. Similarly, LB 2 (36) is Ljung-Box statistic using squared standardized residual series to detect remaining heteroscedasticity. The entries are p- values and if it is larger than 0.05 (0.01), the null hypothesis of no serial correlation is supported at 5% (1%) level. Long yield is modelled as an AR-GARCH (1,1) series with the additional specification in the variance equation as discussed in the text. Sector returns are all market models with GARCH (1,1) error specifications. 15

16 Table 2 Causality Between Long-Term Yield and Sector Returns Cross Correlation Based Analysis for Canada Causality in Mean Causality in Variance Lag Bank Insurance Financial Bank Insurance Financial * * * * * * * * * * * * * * * * * The entries in the table are the Cheung and Ng (1996) statistics based on cross-correlation function between long-term yield and each of the other three sector returns. Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. 16

17 Table 3 Causality Between Long-Term Yield and Sector Returns Cross Correlation Based Analysis for France Causality in Mean Causality in Variance Lag Bank Insurance Financial Bank Insurance Financial * * * * * 2.05 * * * * * 2.37 * * * * * * * * * * * * The entries in the table are the Cheung and Ng (1996) statistics based on cross-correlation function between long-term yield and each of the other three sector returns. Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. 17

18 Table 4 Causality Between Long-Term Yield and Sector Returns Cross Correlation Based Analysis for Germany Causality in Mean Causality in Variance Lag Bank Insurance Financial Bank Insurance Financial * * * * * * * * * * * * * * * * * 1.96 * * * The entries in the table are the Cheung and Ng (1996) statistics based on cross-correlation function between long-term yield and each of the other three sector returns. Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. 18

19 Table 5 Causality Between Long-Term Yield and Sector Returns Cross Correlation Based Analysis for Italy Causality in Mean Causality in Variance Lag Bank Insurance Financial Bank Insurance Financial * 2.08 * * * * * * * * * * The entries in the table are the Cheung and Ng (1996) statistics based on cross-correlation function between long-term yield and each of the other three sector returns. Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. 19

20 Table 6 Causality Between Long-Term Yield and Sector Returns Cross Correlation Based Analysis for Japan Causality in Mean Causality in Variance Lag Bank Insurance Financial Bank Insurance Financial * * * * * * * The entries in the table are the Cheung and Ng (1996) statistics based on cross-correlation function between long-term yield and each of the other three sector returns. Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. 20

21 Table 7 Causality Between Long-Term Yield and Sector Returns Cross Correlation Based Analysis for U.K. Causality in Mean Causality in Variance Lag Bank Insurance Financial Bank Insurance Financial * * * * * * * * * * * * * The entries in the table are the Cheung and Ng (1996) statistics based on cross-correlation function between long-term yield and each of the other three sector returns. Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. 21

22 Table 8 Causality Between Long-Term Yield and Sector Returns Cross Correlation Based Analysis for U.S.A. Causality in Mean Causality in Variance Lag Bank Insurance Financial Bank Insurance Financial * 2.67 * 2.12 * 1.90 * * * * * * 2.09 * 2.50 * 1.86 * * * * * * * * * * * * * * The entries in the table are the Cheung and Ng (1996) statistics based on cross-correlation function between long-term yield and each of the other three sector returns. Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. 22

23 Table 9 Summary of the Causality Between Long-Term Yield and Sector Returns Leads and Lags with Significant Causality Causality in Mean Causality in Variance Country Bank Insurance Financial Bank Insurance Financial Canada -Lag -6-8,-6, ,-1-9 -Contemp Lead France-Lag ,-3-3 -Contemp Lead 7,9,11 6,9 6,7,9, Germany-Lag -12, Contemp Lead 4,12 6,9,10, ,10 2,10 Italy -Lag ,-1-5,-2 - -Contemp Lead 5,9 7, Japan -Lag Contemp Lead 5 1,2,9 2, UK -Lag ,-5,-4, Contemp Lead 4 10, USA -Lag -1-7,-1-12,-1-12,-4,-1-12,-5-12,-4 -Contemp Lead 4,6 1,4,7 1,4, Lags are measured in weeks and negative lag implies lead. Under the null hypothesis of no causality the statistic has standard normal distribution. Significance at 5% is indicated by *. Significance at positive lag length indicates long-term yield is causing either mean or variance in the other sector return. Similarly significance at negative lag implies the opposite. '0' in the respective cell indicates the cotemporanoues two-way causation that is significant at 5%. 23

24 Endnotes: 1 Most studies in this area focus on banking stocks only, however, we define financial sector firms in a broad sense. These include firms involved in banking and insurance related activities as identified by Datastream International. 2 For example, (Ross (1989), Engle, Ito, and Lin (1990) attribute information assimilation and processing time to the variance of equity returns. 3 Weekly observations avoid irregular spacing that often results due to variable number of trading days involved in a week when working with daily observations. Daily data is reported to be more prone to noise as well. Besides, Tai (2000), which is a closely related paper, also employs weekly observations. 4 The rationale for using short-term bill rate to generate excess return is that it is close to the weekly sampling period of our data and the high liquidity of short-term bills. 5 The efficacy of this method to study causality is that it does not require joint modelling of the two series of interest to us. It, thus, does not require imposition of any additional constraint e.g. on the correlation structure of the residuals. Such a specification would be required if we were to model this using other econometric technique such as bivariate GARCH process where such specifications may contaminate the natural interaction that exists. 6 We would like to interpret the result as the average behavior of the causation between the two series over the sample period. It is not suggesting that every 12 weeks or so there is some causation. This quarterly interaction may be related to important macro economic information releases. In most countries studied in this paper important macro-economic data such as GDP and industrial production are released quarterly. 7 We believe that such information on causality would be of use to the academia, the practitioners and the regulators. For academia and the practitioners the application of the Cheung and Ng (1996) methodology in this paper illustrates how to infer lead-lag relationship between the variables of interest without any prior specification of the covariance structure. Policy-makers, on the other hand, would like to know the impact and the duration of the flow through process of changes in monetary policy. 24