War and emerging market default risk: the case of India and the Iraqi invasion of Kuwait.

Clark, Ephraim ; Lakshmi, Geeta

ABSTRACT

We use the performance of Indian Eurobonds over the period 1990-1992 to examine the sensitivity of India's creditworthiness to the Iraqi invasion of Kuwait on August 2, 1990. We also explore the related question of whether the changes in creditworthiness, measured as the effect of changes in default probabilities on bond prices, were accurately assessed by the market in a timely manner. We find that the markets systematically mis-estimated these effects. They anticipated no effects on India's default probabilities in the invasion quarter. All the change in Indian bond prices in the quarter that the invasion took place was due to changes in the risk free term structure of interest rates. In the quarter following the invasion, effects of changes in default probabilities were significant and caused a fall of nearly 3 points in Indian Eurobond prices. In the quarter when the Gulf War took place changes in default probabilities caused a further fall of 1.34 points in Indian bond prices. We find evidence of market over-reaction to country specific invasion effects.

JEL: 0530, 0160, G150, P330, F340

Keywords: Term structure of interest rates; Duration; Spline

I. INTRODUCTION

The purpose of this paper is to use the performance of Indian Eurobonds over the period 1990-1992 to examine the sensitivity of India's financial creditworthiness to the Iraqi invasion of Kuwait on August 2, 1990. To this end, we divide bond price changes into those induced by general market conditions and those induced by changes in default probabilities. We also explore the related question of whether the changes in creditworthiness, measured as the effect of changes in default probabilities on bond prices, were accurately assessed by the market in a timely manner.

Extreme political events such as terrorism and war have become a major concern of the international capital markets since September 11, 2001. Events like these have worldwide as well as country specific economic, financial, political and social consequences and their effect on markets can be dramatic. They also threaten to become more frequent. Besides the recent war in Afghanistan and the coalition invasion of Iraq, India and Pakistan are on the warpath, Turkey is threatening the Iraqi Kurds, the US is threatening the Axis of Evil, Al Queda is threatening the West and North Korea is threatening any country within range of its missals. Given that many emerging economies are saddled with structural imbalances, social and political fragility, and financial dependence, they may be particularly vulnerable to such events, whose effects can highlight and exacerbate certain weaknesses above and beyond what is warranted by the fundamentals. (1) This would be the case, for example, if information assymetries, such as those discussed by Calvo (1998) and Calvo and Mendoza (2000) in the context of contagion, hindered timely and accurate analysis. The first contribution of this paper, then, is that we develop a methodology that makes it possible to distinguish between general market effects and country specific changes in default probabilities. The second contribution is that we apply this methodology to determine the particular case of country specific effects that the Iraqi invasion of Kuwait had on the prices of Indian Eurobonds.

India and the invasion of Kuwait is an interesting case study. First of all, India was not directly involved in the conflict and was far enough from the war theater that it would not be threatened directly by the fighting. It did, however, have close links to the region through its large contingent of emigrants working in the Middle-East and their sizeable contribution to Indian foreign exchange earnings. (2) Thus, its situation was likely to be directly related to the invasion's effects but limited to the type of events outside the realm of physical and human destruction that lend themselves to evaluation. Secondly, India had a relatively large amount of Eurobonds outstanding in a wide range of currencies, coupons and maturities. This makes econometric testing feasible and guarantees that the case of India will reflect the market in general and not a specific feature of a particular bond, currency coupon or maturity. Third, at the beginning of the period under consideration, India's structural and political difficulties were longstanding and well known. Its credit rating was still a longstanding and respectable A2. Furthermore, over the period in question India was undertaking structural reforms urged by the IMF (International Monetary Fund) that were designed to reduce default probabilities. Thus, it was in a well-known, sensitive but solid and improving position, vulnerable to the effects of war but far from desperate. Thus, if invasion effects above and beyond those generated by overall market conditions are present, they will be reflected in the data without being contaminated by bias due to conditions generated by extreme situations. In fact, although India experienced financial distress and its official credit rating was downgraded three times, it did not default or reschedule.

We proceed in three steps:

1. We present a simple default risk model that separates changes in bond prices into two categories: those caused by changes in the risk free term structure of interest rates and those caused by changes in country specific default risk.

2. We compute the riskless term structure of interest rates and use it to calculate the price of the theoretical bond presented in the model.

3. We use regression analysis to estimate the relationship between the risky and theoretical bonds to compute the price changes in the risky bond due to changes in the riskless term structure (market risk) along with dummy variables timed to measure the effect of changes in country specific default risk.

When this methodology is applied to Indian Eurobonds 1990-1992, we find that the markets anticipated no country specific effects of the invasion on India's default risk in the quarter that the invasion took place. All the changes in Indian bond prices in this quarter were due to changes in the risk free term structure of interest rates. In subsequent quarters this assessment was revised and default risk is found to account for a fall of over 4 points in Indian bond prices. Interestingly, we find that the markets systematically mis-estimated the effect that the events set off by the invasion would eventually have on India's default risk.

The rest of the paper is organized as follows. In section II we develop the relationship between the risky and theoretical riskless bonds. Section III presents the data and methodology. Section IV presents the results and section V concludes.

II. BOND PRICES, THE TERM STRUCTURE OF INTEREST RATES AND DEFAULT RISK

The price of a risky bond can be represented as the difference between the expected loss from default on the risky bond and the price of a theoretical bond identical in every way to the risky bond except that the theoretical bond has no default risk and its price is determined by the riskless zero coupon term structure of interest rates. (3)

Consider the following notations:

[P.sub.0]= observed price of the risky bond at time 0

[T.sub.0] = the price of theoretical bond at time 0.

t =1,2 ... n = payment dates where n = maturity date of the bond.

[r.sub.t] = 1 + the riskless zero coupon rate for period t.

[C.sub.t] = the cash flow for time t.

[R.sub.t] = the given (constant) recovery rate at time t in the case of default as a percent of the expected value of the theoretical bond at time t.

[F.sub.0, t] = the forward price of the theoretical bond for delivery at time t.

[K.sub.t] = [t.summation over (i=0)][C.sub.i][r.sup.-i.sub.i] = present value of the coupons paid out up to the delivery date at time t.

[[lambda].sub.t] = the risk neutral probability of default at time t.

To simplify the exposition, we assume that default can only occur immediately before each payment date. Thus, the difference between the theoretical riskless bond and the risky bond can be written as

[T.sub.0] - [P.sub.0] = [n.summation over (t=1)][[lambda].sub.t] [[F.sub.0,t] - [R.sub.t][F.sub.0,t]][r.sub.t.sup.-t] (1)

where [F.sub.0,t] represents the expected value of the riskless asset. The value of [F.sub.0,t] is equal to [[T.sub.0] - [K.sub.t]][r.sup.t.sub.t]. Substituting this value into (1) and rearranging gives

[P.sub.0] = [n.summation over (t=1)][[lambda].sub.t][K.sub.t](1 - [R.sub.t]) + [T.sub.0] [1 - [n.summation over (t=1)] [[lambda].sub.t] (1 - [R.sub.t])] (2)

Equation (2) breaks the riskless cash flows into two parts. The first term on the right hand side (RHS) of equation (2) represents the present value of the coupons that will be paid out before default occurs at each default date. The second term represents the certainty equivalent of the uncertain cash flows. In the absence of default risk, the two bonds are equivalent. This can be seen if we set the [[lambda].sub.t]'s equal to zero. It is also clear that in the absence of changes in the default probabilities ([[lambda].sub.t]), any change in the price of the risky bond is due to changes in the term structure of interest rates. To see this, we can write the value of [T.sub.0] as

[T.sub.0] = [C.sub.1][r.sup.-1.sub.1] + [C.sub.2][r.sup.-2.sub.2]+ ... + [C.sub.n] [r.sup.-n.sub.n] (3)

Then take the total differential of T

[ILLUSTRATION OMITTED] (4)

The differential of P in equation (2) is equal to

d[P.sub.0] = [1 - [n.summation over (t=1)] [[lambda].sub.t] (1 - [R.sub.t])] dT + [DELTA][LAMBDA] (5)

where

[DELTA][LAMBDA] = [n.summation over (i=1)] [partial derivative]P / [partial derivative][[lambda].sub.i] d[[lambda].sub.i] (6)

refers to price changes due to changes in default probabilities.

Substituting from (4) for dT in (5) shows that in the absence of changes in the bond's default risk probabilities, changes in the risky bond's price are due to changes in the term structure of interest rates, which we can call market risk.

Changes in the default probabilities change both the present value of the coupons that will be paid out before default occurs at each default date and the present value of the certainty equivalent. It is important to note that changes in the term structure are continuous changes driven by the market whereas changes in default probabilities are discrete and relatively rare as evidenced in credit ratings and migrations from one category to another. In the regressions that follow, we will use this fact to test for changes in default probabilities.

III. DATA AND METHODOLOGY

A. Data Set

The market prices of Indian bonds and data for modelling the term structure were obtained from the Handbooks published by the International Securities Market Association (ISMA), formerly known as the Association of International Bond Dealers (AIBD).

Our data is quarterly and the observation period runs from June 29, 1990 to September 30, 1992, a total of 10 observations for each bond. The quarterly window was chosen, based on the timing of the invasion, as the smallest window wide enough to encompass price variations due to changes in the term structure as well as changes in default probabilities. Our sample is the subset of eight Indian bonds with varying amounts and maturities issued by public sector and quasi public sector borrowers--3 in USD, 4 in DEM and 1 in JPY--that remained outstanding over the entire observation period. We also considered the Fung and Rudd (1986) argument that the time period should be not be too close to the issue date of any bond, since these prices often mirror issue costs along with interest-rate driven price movements. There were no direct sovereign issues made but all the above issuers were under the control, management and ownership of the Government of India and were guaranteed by it. Apart from ONGC, they are all financial institutions. The details of these bonds are given in Table 1.

To estimate the riskless term structure in the unregulated, tax-free Eurobond market of 1990-1992, we constructed sample sets for each observation date of not less than 50 bonds (4) issued by officially backed supranationals, for each of the three currencies constituting India's external debt--US dollars, German marks, and Japanese yen. (5) We use the supranationals to estimate the international riskless term structure rather than the corresponding treasuries in order to avoid biases that can creep into national credit markets through taxes, regulations, government intervention and the like. The supranationals included in our sample are guaranteed at least de facto by their member governments and borrow at terms equivalent to, and at times better than, the treasuries of the currencies in question. Thus, they are effectively riskless and give the best picture of the international riskless term structure of interest rates. The large number of bonds in each sample was necessary to ensure the desirable asymptotic qualities of consistency and sufficiency. To ensure a balanced sample over the whole term structure, bonds in equal numbers were chosen with term left to maturity of less than three years, between three and six years, and over six years.

B. The Methodology

We proceed in three steps.

In step 1, we estimate the riskless term structure for each time period, developed from McCulloch's cubic spline methodology, on the cross section of supranational bonds in the USD market, the JPY market and the DEM market. (6) This gives us three time series for the riskless term structure, one in USD, one in DEM, and one in JPY.

In the estimation of the riskless yield curve, we used two spline (7) knot points of three and six years. The choice of these two points was based on the observation that the Eurobond market typically deals in shorter maturities than their respective domestic bond markets. Thus, we reasoned that the break points for investor perceptions of uncertainty, liquidity and risk in the Eurobond market could reasonably be represented as relatively short term: up to three years, relatively medium term: between three and six years, and relatively long term: above six years.

Bond prices are quoted clean in the Eurobond market, i.e. they are quoted free from any accrued coupon in order to facilitate yield comparisons but the actual sale is on the basis of the dirty price, i.e. the clean price cum accrued interest. Thus, we computed the dirty prices based on the number of days the bond was not held by the buyer. (8) The ask prices were used to compute the dirty prices. (9)

We used this information in ordinary least square (OLS) regressions to estimate the parameters of the cubic spline model using observed values of prices, coupons and times to maturity. These parameters were then used to compute the discount curves and the risk-free spot rate curves. The discount curve was computed for twelve years to allow comparability between data sets.

In step 2, we apply the corresponding riskless term structure for each time period to each of the Indian Eurobonds in our sample to estimate their "theoretical riskless price". This gives us eight time series, one for each bond, of the theoretical riskless price of the Indian Eurobonds in the sample. (10)

Finally in step 3, we use the "theoretical riskless prices" in the relationship with the observed risky prices developed in section 2 along with dummy variables timed to the invasion and its aftermath to test the effects of the invasion of Kuwait on Indian bond prices.

IV. RESULTS

A. Estimates of the Term Structure

Five parameters were estimated for the cubic spline model. The results, not reported here, of the 30 regression coefficients, i.e., the three currency markets over ten time periods, are very good. The linear coefficient is always significant and always negative for all three currencies. The results are best for the dollar. The quadratic and cubic coefficients are usually significant at the 5% level. Otherwise, except in one case, they are significant at the 10% level. The curvature coefficients are also often significant at the 5% level and usually significant at the 10% level, more so for the first knot than for the second, thereby indicating more curvature effect at the short end of the structure than the long end. For the yen, the quadratic and cubic coefficients are usually significant at the 10% level. However, the curvature coefficients are clearly significant together in only three periods: June 1990, March 1991, and December 1991. In March 1992 short term curvature is significant and in June 1992 long term curvature is significant. For the mark no parameters except the linear coefficient are significant at the beginning of the observation period. However, starting in March 1991 the quadratic coefficient becomes significant at the 5% level and the cubic at the 10% level. Except for June 1991, they stay significant until the last period, September 1992. The curvature coefficients are only significant in December 1991 and June 1992.

We used the parameters estimated above to compute the yield curves and discount functions and compared the results with those estimated with the Cox, Ingersoll and Ross (1985) model. (11) They are almost indistinguishable, which is strong evidence for their accuracy. The shape of the yield curve showed a consistently upward sloping curve for the dollar market The yen long-term rate fell below the short-term rate from September 1991 to March 1992, down to almost 0, in the last two-mentioned time periods. In two cases it was almost the same as the short-term rate i.e. in September 1990 and March 1991. During this period i.e. from June 1990 to March 1991, it was only marginally above the short rate. During the remaining last period however, the difference widened. The DEM yield curves showed either the interest rate at the long end to be roughly of the same magnitude as the short, or inverted. During this period, Germany was also undergoing the event of reunification and the yield curve was inverted. In summary, the shape of the two ends of the yield curves seems satisfactory.

B. Regression Results on Indian Bonds

We then applied the foregoing riskless term structures to each of the Indian bonds in our sample for each observation period in order to estimate their theoretical prices. In all, 80 observations were collated (10 quarters x 8 bonds) from the thirty yield curves (3 currency markets x 10 quarters).

We test for stationarity in the panel data series for P and T using the Im, Pesaran, and Shin (1995) T-bar test as applied in Wu and Chen (1999). The Z scores are -0.5297 and -0.3548, respectively. (12) The corresponding 95% critical values are [+ or -] 3.09177 generated by the Monte Carlo simulations. Thus we cannot reject non-stationarity. Differencing both series once and applying the T-bar test gives Z scores of -27.58 and -21.44 for the first differences of P and T, respectively. Thus, based on the 95% critical values, we reject non-stationarity for the differenced series.

Using first differenced data, we start testing with the benchmark case where there are no changes in default probabilities and price changes are due exclusively to changes in the risk free term structure, that is, equation 5 with [DELTA][LAMBDA] = 0

[DELTA][P.sub.it] = [a.sub.1] + [a.sub.2][DELTA][T.sub.t] + [a.sub.3][e.sub.t-1] + [u.sub.t] (7)

where [DELTA] denotes the first difference [e.sub.t-1] is the error correction term, (13) [u.sub.t] is the error term, and [a.sub.1], [a.sub.2] and [a.sub.3] are estimated coefficients. We include the error correction term because from equation (5) we can see that in its absence, the coefficient [a.sub.2] is likely to be time varying, thereby causing the regression to suffer from omitted variable bias.

To overcome problems of heteroskedasticity and cross sectional correlation arising from the data pooled over 8 bonds and 10 time periods, we used the Kmenta (1990) full cross-sectionally correlated and time-wise autogressive model. (14) We expect al to be equal to zero and [a.sub.2] to be positive. (15) The results are reported in table 2. They show that [a.sub.1] is very small and not statistically different from zero with a p-value of 0.564. (16) On the other hand, [a.sub.2] is highly significant with a p-value of 0.00 and, as expected, it is positive. (17) The coefficient [a.sub.3] of the error correction term is also highly significant. Furthermore, the overall equation is very good with an adjusted [R.sup.2] of over 55% and no evidence of autocorrelation in the residuals.

To test for changes in default probabilities, we add three dummy variables: D1 for the third quarter of 1990 when the invasion took place, D2 for the fourth quarter of 1990, the period in between the invasion of Kuwait and the Gulf War, and D3, the first quarter of 1991 when the Gulf War was fought. Each dummy takes the value of 1 for the quarter in question and zeros everywhere else and is designed to capture the effect of country specific changes in default probabilities on bond prices. We then test equation 5 where the dummy variables capture the term [DELTA][LAMBDA]:

[DELTA]P = [a.sub.1] + [a.sub.2][DELTA]T + [a.sub.3][e.sub.t-1] + [b.sub.1]D1 + [b.sub.2]D2 + [b.sub.3]D3 + [u.sub.t] (8)

With the inclusion of the dummy variables, we have no expectations about [a.sub.1], which will capture any constant effects associated with effects that were not anticipated by the market. We expect that [a.sub.2] will be similar to [a.sub.2] in Table 2. If the markets anticipate or perceive changes in default probabilities, the coefficients [b.sub.1], [b.sub.2], and [b.sub.3] will be statistically significant. If, on the other hand, no changes were anticipated or perceived, they will not be statistically significant.

Table 3 shows the results. (18) The overall results are much improved with respect to those of the benchmark in table 2. The adjusted [R.sup.2] has increased to 0.9710. As expected, [a.sub.2] is similar to [a.sub.2] in table 2, changing by only 0.0306 but the t-statistic is much higher. The coefficient [b.sub.1] is not significant, which is evidence that no change in default probabilities was perceived or anticipated in the quarter that the invasion took place. However, coefficients [b.sub.2] and [b.sub.3] are highly significant. This is evidence that the market perceived changes in default probabilities in these two quarters. It is interesting that changes are significant in both quarters. If the full extent of the change in default probabilities had been accurately assessed in the period following the invasion, [b.sub.3] would not be significant. The significance of [b.sub.3] suggests that the market either underestimated the full extent of the invasion's effects on bond prices in the preceding period or over-reacted and overestimated them in the following period. We attribute the increase in default probabilities to invasion effects and not to a weakening of India's intrinsic position because India's structural problems and economic inefficiencies were well known for many years and basically unchanged over the 7 months following the invasion. Thus, the consequences of the invasion, the buildup to war and the war itself seem to have fragilized India's financial position beyond all expectations. The over-reaction that this implies is reflected in the enormous loss of four grades in its credit rating in the space of less than 6 months.

V. CONCLUSIONS

In the current international climate of conflict and confrontation, political events such as invasions and war have become important factors in the performance of international capital markets. Given the structural imbalances, social and political fragility, and financial dependence of many emerging economies, these economies may be particularly vulnerable to such events. Furthermore, whilst the events themselves are often anticipated far in advance, it is unclear how accurately the effects of these events can be assessed by the markets. We use the case of Indian Eurobonds to examine these questions with respect to Iraq's invasion of Kuwait on August 2nd 1990. India was chosen because it was far enough away from the immediate destruction of the war itself but its close links with Indian emigrants in the Middle East and its large dependence on oil imports made it vulnerable to events in the Middle East. India also had a wide enough range of Eurobonds outstanding to make testing feasible. Furthermore, India's pre-invasion credit rating of A2 indicated a healthy financial position such that its financial troubles subsequent to the invasion could be attributed in large part to the invasion and its aftermath.

We test a simple default risk model that uses the prices of theoretical riskless bonds in USD, DEM and JPY calculated from cubic spline estimates of the international term structure of interest rates over the period 1990-1992. We find that in the quarter that the invasion took place, the markets anticipated no country specific effects of the invasion on India's default risk. All the changes in Indian bond prices in that quarter were due to changes in the risk free term structure of interest rates. However, in the quarter following the invasion, increased default risk, reflected in a two-notch downgrade of India's credit rating, caused a fall of nearly 3 points in Indian Eurobond prices. A further increase in perceived default probabilities, reflected in a further two notch downgrade of India's credit rating, caused a further fall of 1.34 points in Indian bond prices in the succeeding period. This suggests that effects were either underestimated in the preceding period or over-estimated in the succeeding period. Over the entire seven-month period following the invasion, India's credit rating fell by four grades, a huge amount. This is strong evidence of India's extreme vulnerability to the invasion's effects and suggests that over-estimation was present. The lagged reaction of the market to invasion effects on India's default probabilities is strong evidence that the markets were unable to effectively assess this vulnerability in a timely manner.

BIBLIOGRAPHY

Bradley, F., 1991, "Neglected Factors in the Market Pricing of Eurodollar Bonds," Journal of Portfolio Management, Vol. 17, No. 2, 62-73.

Brown, S.J. and Dybvig, P.H., 1986, "The Empirical Implications of the Cox, Ingersoll and Ross Theory of the Term Structure of Interest Rates," The Journal of Finance, Vol. 41,3-42

Carleton, W.T. and Cooper, I.A., 1976, "Estimation And Uses Of Term Structure Of Interest Rates," The Journal of Finance, Vol. 31, No. 4, 1067-83.

Calvo, G. A., 1998, "Varieties of Capital Market Crises," in G. A. Calvo and M. King, eds., The Debt Burden and its Consequences for Monetary Policy, (New York: Macmillan Press).

Calvo, G. A. and E. Mendoza, 2000, "Rational Contagion and the Globalization of Securities Markets," Journal of International Economics, 51, pp. 79-113.

Carleton, W.T. and Cooper, I.A., 1976, "Estimation and Uses of Term Structure of Interest Rates," The Journal of Finance, Vol. 31, No. 4, 1067-83.

Chambers, D.R, Carleton, W.T. and Waldman, D.W., 1984, "A New Approach to Estimation of the Term Structure of Interest Rates," Journal of Financial and Quantitative Analysis, Vol. 19, No. 3, 233-52.

Cox, J.C., Ingersoll, J.C. and Ross, S.A., 1985, "A Theory Of Term Structure Of Interest Rates," Econometrica, Vol. 53, No. 2, 385-407.

Deacon, M. and Derry, A., 1994a, "Deriving Estimates of Inflation Expectations From the Prices of UK Government Bond," Bank of England, Working Paper Series No. 23.

Deacon, M. and Derry, A., 1994b, "Estimating Market Interest Rate and Inflation Expectations from the Prices of UK Government Bonds," Bank of England Quarterly Bulletin, 232-240.

Doukas, J., 1989, "Contagion Effect on Sovereign Interest Rate Spreads," Economic Letters, 29, pp. 237-241.

Duffie, D. and Singleton, K, 1999, "Modeling Term Structures of Defaultable Bonds," Review ofFinancial Studies, Vol. 12, 687-720.

Eichengreen, B., A. Rose and C. Wyplosz, 1996, "Contagious Currency Crises," NBER Working Paper no. 5681, (Cambridge MA, National Bureau of Economic Research).

Fung, W.K.H. and Rudd, A., 1986, "Pricing New Corporate Bond Issues: An Analysis of Issue Cost and Seasoning Effects," The Journal of Finance, Vol.41, No.3, 633-43.

Govt. of India, Economic Survey, New Delhi- various issues.

Im, KS, Persaran, H.M and Shin, Y., 1995, "Testing for Unit Roots in Heterogeneous Panels," Working Paper, Cambridge University

"International Securities Market Association (ISMA)," Weekly Eurobond Guide, Various issues.

Jarrow, R A., Turnbull, S. M., 1995, "Pricing Derivatives on Financial Securities Subject to Credit Risk," Journal of Finance, 50, 1, 53-85.

Kmenta, J., 1990, Elements Of Econometrics, 2nd Ed., Maxwell Macmillan International Editions, New York.

Litzenberger, RH. and Rolfo, J., 1984, "An International Study of Tax Effects on Government Bonds," Journal of Finance, Vol. 39, No. 1, 1-22.

Luther, RG. and Matatko, J., 1992, "Tax Effects In Gilt Edged Security Valuation," in Cooke T.E., Matatko J. and Stafford D.C., Risk Portfolio Management And Capital Markets, 81-89, Macmillan.

Madan, D. and. Unal H, 1998, "Pricing the Risks of Default," Review of Derivatives Research, Vol.2, 79-105.

Mastronikola, K, 1991, "Yield Curves for Gilt Edged Stocks: A New Model," Bank of England Discussion Paper (Technical Series), No 49.

McCulloch, H.J., 1971, "Measuring the Term Structure of Interest Rates," Journal of Business, Vol. 44, 19-31.

McCulloch, H.J., 1975a, "An Estimate Of The Liquidity Premium," Journal of Political Economy, Vol. 83, No. 1, 95-119.

McCulloch, H.J., 1975b, "The Tax Adjusted Yield Curve," Journal of Finance, Vol. 30, 811-29.

Schaefer, S.M., 1981, "Measuring A Tax-Specific Structure Of Interest Rates In The Markets For British Government Securities," The Economic Journal, 91, 415-38.

Shea, G.S., 1985, "Term Structure Estimations with Exponential Splines," The Journal of Finance, Vol. 40, No. 1, 319-25.

Valdes, R O., 1997, "Emerging Markets Contagion: Evidence and Theory," Working Paper No. 7, (Santiago: Banco Central de Chile).

Vasicek, O.A. and Fong, H.G., 1982, "Term Structure Modeling Using Exponential Splines," The Journal of Finance, Vol. 37, No 2, 339-48.

Wu, J.L. and Chen, S.L., 1999, "Are Real Exchange Rates Stationary Based on Panel Unit Root Tests?" International Journal of Finance and Economics, Vol. 4, 243-252.

NOTES

(1.) Questions such as these resemble those in the literature on contagion that looks at things such as the transmission of a crisis from one country to another that is unwarranted by the fundamentals (Eichengreen et al., 1996) and excess co-movement of credit spreads (Doukas, 1989) or returns across countries (Valdes, 1997).

(2.) See, various annual issues of Economic Survey published by Govt. of India during this period.

(3.) See, for example, Jarrow and Turnbull (1995), Madan and Unal (1998), and Duffle and Singleton (1999).

(4.) Most studies like Brown and Dybvig (1986) use the same dataset. However, even when the currency market was the same, in this study, the data set was varied to enable the use of very short bonds. This was to prevent the underestimation of the very short end of the yield curve to the extent of the time between June 1990 and September 1992. The June 1990 observations would have had to otherwise include observations at least 27 months away from maturity.

(5.) These issuers include the World Bank, Eurofima, the European Investment Bank, the African Development Bank, the Asian Development Bank etc.,

(6.) Although several models such as Carleton and Cooper (1976), Schaefer (1981), Vasichek and Fong (1982), Chambers, Carleton and Waldman (1984), Mastronikola (1991) exist to estimate the term structure, Shea (1985) compares them and finds McCulloch's (1971, 1975a 1975b) cubic spline model empirically tractable, easily computable by OLS and parsimonious. Furthermore, Litzenberger and Rolfo (1984), Luther and Matatko (1992), Deacon and Derry (1994 a and b) and Bradley (1991) have successfully applied this model in their empirical studies. In this study we use the McCulloch cubic spline

(7.) A spline is a model which incorporates switching coefficients of regression in two or more periods of time. To make this a smooth transition and to estimate it, it is essential that two regression lines meet at a switching point (knot) in a manner that in the example of a cubic spline satisfies the following:

Yt = [alpha]1 + [beta]1Xt + [gamma]1Xt2 + d1Xt3 + et (t=1,2, ..., [t.sup.*])

Yt = [alpha]2 + [beta]2Xt + [gamma]2Xt2 + d2Xt3 + et (t=[t.sup.*]+1, [t.sup.*]+2, ..., n)

The requirement is that at point t=[t.sup.*] the first and second derivative of these curves be the same.

(8.) Unlike more imprecise methods like Chambers et al (1984) who computed interest accrued to the nearest quarter, dirty prices used in this study were precise to the day.

(9.) There were two other choices in this matter:

1)Bid prices could have been used on the argument that they are prices the market makers are ready to buy at, or

2)The midpoint of the two, i.e. the mean of the bid and ask price could have been used. However it was found that the bid-ask spread was very narrow in the market. In view of this we found it reasonable to calculate the prices on the basis of ask quotations along with the accrued coupon.

(10.) We also estimate the yield curve using the Cox, Ingersoll and Ross (1985) model, which gives similar results. These results are not reported here but are available on request.

(11.) Results available on request.

(12.) Details of the simulations are available on request.

(13.) The error correction term is the error term [e.sub.t] in the regression [P.sub.t] = [c.sub.1] + [c.sub.2] [T.sub.t] + [e.sub.t].

(14.) We found that the Kmenta model worked best with no correction for autocorrelation.

(15.) See Equation 5.

(16.) Tests for bond specific fixed effects were also negative at conventional levels of significance

(17.) Tests for bond specific slope effects confirm that the slopes for the individual bonds are all positive at conventional levels of significance.

(18.) We also controlled the dummy variables for currency specific and maturity specific effects and found that none were present.

Ephraim Clark (a) and Geeta Lakshmi (b)

(a) Professor of Accounting and Finance, Middlesex University Business School The Burroughs, Hendon, London NW4 4BT E.Clark@mdx.ac.uk

(b) Visiting Lecturer, Nottingham University, Nottingham, UK Geeta.Lakshmi@nottingham.ac.uk

Table 1
Bonds issued in the Euromarket by India (1980-92)

Name Date of Issue Currency Amount

Industrial 6/1989 dollar 100 million
Development
Bank of India

Oil and Natural 12/1988 dollar 125 million
Gas Commission

Oil and Natural 3/1990 dollar 125 million
Gas Commission

State Bank of 6/1988 yen 15 billion
India

Industrial 3/1987 DM 200 million
Development
Bank of India

Industrial 9/1988 DM 250 million
Development
Bank of India

Industrial 2/1986 DM 100 million
Development
Bank of India

Oil and Natural 2/1987 DM 150 million
Gas Commission

Name Maturity Coupon

Industrial 6/6/96 10%
Development
Bank of India

Oil and Natural 16/11/1993 9.75%
Gas Commission

Oil and Natural 16/03/1997 10%
Gas Commission

State Bank of 21/06/1993 5.25%
India

Industrial 21/12/1994 6.38%
Development
Bank of India

Industrial 1/9/95 6.625%
Development
Bank of India

Industrial 1/2/93 7%
Development
Bank of India

Oil and Natural 25/02/1994 6.375%
Gas Commission

Table 2

Regression results of Equation 6 on first differences with the error
correction term

 Coefficient Value t-statistic p-value

[a.sub.1] -0.19622 -0.5644 0.564

[a.sub.2] 0.70642 8.29 0.000

[a.sub.3] -0.2974 -3.337 0.001

Adjusted R square = 0.5513

Table 3

Regression results of Equation 6 on first differences with the error
correction term

 Coefficient Value t-statistic p-value

[a.sub.1] 0.31532 2.274 0.026

[a.sub.2] 0.73702 36.72 0.000

[a.sub.3] -1.3681 -16.96 0.000

[b.sub.1] 0.46582 1.288 0.202

[b.sub.2] -2.7901 -6.986 0.000

[b.sub.3] -1.3363 -3.436 0.001

Adjusted R square = 0.9710