The January effect in the corporate bond market: a systematic examination.
A January effect has also been ascertained in corporate bond returns and yields (see Chang and Pinegar, 1986; Chang and Huang, 1990; Fama and French, 1993; and Barnhill, Joutz, and Maxwell, 1997) and municipal bond returns (see Kihn, 1996). The prevalence and scope of the January effect is apparent in the findings that international stock markets have similar seasonal variations (see Gultekin and Gultekin, 1983).
The purpose of this paper is to examine the January effect in the corporate bond market, systematically examining first its strength, and then its possible causes. The paper extends the previous research by examining a larger sample of bonds, testing previously proposed causes using new data and methodologies, and proposing and testing new theories.
The paper is organized as follows. In Section I, I examine previous research on the January effect in the corporate bond market. Section II investigates the strength of the anomaly. In Section III, I examine the relation between the January effect in the bond market and firm size. Section IV discusses different supply and demand theories that have been suggested as causes of the January effect. Section V presents conclusions about the causes of the January effect in the corporate bond market.
I. Seasonal Variation in Corporate Bond Returns and Yields
In this section, I review previous work regarding the January effect in the corporate bond market, examine possible theories for the anomaly, and explore the relation between bond credit quality and the January effect.
A. The January Effect in the Corporate Bond Market
The January effect in the corporate bond market is well-documented. Chang and Pinegar (1986) analyze the monthly holding returns of treasuries and bonds rated, Aaa, Aa, A, Baa, Ba, and B. They find an excess return in January for the noninvestment-grade segments. Specifically, Chang and Pinegar find a positive excess return at the 92% confidence level for a sample of Ba-rated bonds and at the 99% confidence level for a sample of B-rated bonds in January, but they find no statistically significant excess returns for either treasury bonds or investment-grade bonds.
Chang and Huang (1990) use a different methodology but find similar results. The lower the quality of the bonds the more pronounced the January effect. In contrast to Chang and Pinegar, Chang and Huang find a statistically significant excess return for the lowest level of investment-grade bonds (Baa).
Fama and French (1993) analyze common risk factors in stock and bond returns. Their study finds a statistically significant excess return in January for portfolios of A, Baa, and noninvestment-grade bonds. The excess return increases monotonically as bond rating decreases.
Barnhill, Joutz, and Maxwell (1997) use cointegration techniques and error correction models to examine factors that affect the yields on noninvestment-grade securities. The authors find statistically significant negative changes in yield on CS First Boston's BB, B, and aggregate noninvestment-grade indices in January. The lowest credit quality B index had a larger January coefficient and level of statistical significance than the BB index.
B. Theories that Explain the January Effect
There have been numerous attempts to explain the January effect in the financial markets. One of the most frequently suggested reasons is tax-loss selling. However, numerous studies challenge the validity of this hypothesis (see Brown, Keim, Kleindon, and Marsh, 1983; Gultekin and Gultekin, 1983; Jones, Pearce, and Wilson, 1987; and Tinic, Barone-Adesi, and West, 1987). At best, the previous research suggests that tax-related selling only partly explains the January effect.
DeRosa-Farag (1996) suggests that the January effect in the corporate bond market is the result of supply and demand considerations. Bond coupon payments are not evenly distributed throughout the year, but instead reach their highest level in December and their lowest level in January. DeRosa-Farag therefore suggests that the January price increase is due to fund flow increases in December and decreases in January. However, Fridson and Garman (1995) and Barnhill, Maxwell, and Joutz (1997) find no supporting evidence for a coupon-based payment flow theory.
Several other explanatory hypotheses that apply to the stock market can be ruled out, since they do not apply to the corporate bond market. For example, Bhardwaj and Brooks (1992) find that the January effect is closely related to low share price; however, no systematic pricing differences are found across bonds or categories of bonds. Seyhun (1988) finds support for the theory that insider trading by small-firm management is a cause for both the January and small-firm anomalies. However, since management rarely owns a company's bonds, this theory cannot explain the seasonality of returns in the bond market.
C. The January Effect and Bond Ratings
As noted, previous research demonstrates an anomaly in the corporate bond market and finds the January effect more pronounced in the lowest quality bonds (low-quality-investment grade and noninvestment-grade bonds). This provides the focus for this paper. First, I examine the seasonality of the lowest-quality-investment grade categories (Split-BBB and BBB) and noninvestment-grade bonds.(1) Second, the discrepancy between the strength of the January effect for investment-grade and noninvestment-grade bonds suggests a systematic difference between the two types of bonds. A systematic difference could explain the reason for the anomaly in the corporate bond market.
I note the danger inherent in viewing the corporate bond market as a homogeneous market. In fact, the market is highly segmented. Describing the causes of the segmentation could clarify the reasons why the January effect is more prevalent among the lowest-quality bonds.
The segmentation between investment-grade and noninvestment-grade bonds starts with market participants. Investment-grade and noninvestment-grade bond investors are different: noninvestment-grade bonds are held almost exclusively by institutional investors [ILLUSTRATION FOR FIGURE 1 OMITTED]. Investment-grade bonds, in contrast, are commonly held by individual investors and regulated financial institutions which are restricted from owning non-investment-grade bonds.
Another difference is the unique nature of noninvestment-grade bonds. Noninvestment-grade bonds are considered part debt and part equity instruments. This view is consistent with contingent-claims theory (see Black and Scholes, 1973; and Merton, 1974). The contingent claims theory implies that as the default probability of the bond increases (bond rating decreases), the bond becomes less sensitive to changes in interest rates but more sensitive to changes in stock returns. The observed comovements of noninvestment-grade bonds to both debt and equity instruments is thus consistent with contingent claims theory (see Shane, 1994, and Fridson, 1994).
Portfolio management strategies also differ between investment-grade and noninvestment-grade bonds. Given the large yield spread between treasuries and noninvestment-grade bonds, high-yield portfolio managers stay close to fully invested at all times. High-yield portfolio managers time the market by shifting the portfolio's weighting across credit qualities, but they rarely shift the portfolio's assets out of the noninvestment-grade market.
The liquidity of investment-grade and noninvestment-grade bonds can also vary significantly. As credit quality decreases so does the liquidity of the bonds. This is a function of two phenomena. The average size of bond issues decreases with the credit rating, as does the number of bond issues per company. This effectively limits both the secondary market and the number of market makers. Also, institutional investors can be restricted from owning noninvestment-grade bonds or from holding more than a specified percentage of their portfolio in noninvestment-grade bonds. Even noninvestment-grade mutual funds are limited in the percentage of their portfolio they may hold in the lowest rated categories (B, CCC and NR or not rated). Smaller issue size and restrictions on ownership effectively limit the demand for and liquidity of noninvestment-grade bonds.
II. Strength of the January Effect
Because the majority of bonds are not traded on exchanges, reliable yield and return data must be obtained from one of the investment banks that is an active market maker. Therefore, to analyze the strength of the anomaly, I collected monthly market return data for noninvestment-grade bonds and the lowest investment-grade category (Split-BBB) from CS First Boston for the period January 1986 to April 1997. CS First Boston calculates return data, based on accrued interest and change in the bond's price, for all bonds in its indices. Because there are new issues and redemptions, the number of bonds tracked in the indices changes monthly, but the indices currently include over 1,200 bonds. Table 1 shows the mean and standard deviation of monthly returns by rating category. As expected, the mean and standard deviation of return increase as credit quality declines.
Table 2 shows the monthly return patterns. Previous research has found that the 1987 stock market crash, the Iraq invasion of Kuwait, and the subsequent liberation of Kuwait all had one-month impacts on noninvestment-grade bonds (see Barnhill, Joutz, and Maxwell, 1997). To account for these exogenous shocks to the market and to reduce "noise" in the model, I included one-month dummy variables representing these events.(2) Thus, the constant represents the average monthly returns for February through November and excludes January and December. The January and December coefficients represent any additional return above or below the average return. I included the December variable to examine any fluctuations in returns, since several of the theories examined later have implications concerning return patterns in December. The coefficients for the exogenous shocks represent any change in the average return over the one-month time period the shock took place.
The January coefficient is positive for all the indices and statistically significant, at over the 95% confidence level, in three of the five indices. While the coefficient is positive for the lowest investment-grade category, it is not statistically significant. The January coefficient ranges from a premium over the average monthly return of 17 basis points for the Split-BBB index to 248 basis points for the Split-CCC/CCC category. I find no support for a statistically significant negative return in December consistent with year-end-selling theories, which are examined in more detail in Section V.
The strength of the January effect is readily apparent if the excess January return is compared to the average monthly return. The January premium is 17% for Split-BBB, [TABULAR DATA FOR TABLE 1 OMITTED] [TABULAR DATA FOR TABLE 2 OMITTED] 76% for BB, 63% for Split-BB, 101% for B, and a 248% for Split-CCC/CCC. Overall, the results clearly indicate the strength of the anomaly and the increase in the January premium as the bond rating decreases. These result are consistent with previous findings.
III. The January Effect in the Corporate Bond Market and Firm Size
A number of researchers have found that the turn-of-the-year effect in stock returns is related to a small-firm effect (see Keim, 1983; Reinganum, 1983; Roll, 1983; and Cho and Taylor, 1987). In the corporate bond market, Chang and Pinegar (1986) conclude that seasonal returns are consistent with the small-firm effect.
To empirically examine the relation between the small-firm effect and the January anomaly in the corporate bond market, I use Compustat data to compare firm market value to minor bond rating categories (see Table 3). On average, as minor ratings decrease, there is a monotonic decrease in firm value, except for the change between the AA+ and AA categories and between the AA- and A+ categories.
Because of the small sample size in some rating classes and heterogeneous variances, I use the Wilcoxon rank sum test, a nonparametric test, to compare the significance of the differences between adjacent minor rating categories. Higher rating categories show statistically significant larger market values in nine of the 15 categories. Though not included, I also compare market value and major rating categories, again using the Wilcoxon rank sum test. All higher major rating categories have [TABULAR DATA FOR TABLE 3 OMITTED] statistically significant (at the 99% confidence level) larger market values.
Given the relation between firm size, as measured by market value, and bond rating, the results suggest that the January effect in the bond market is indeed related to the small-firm effect, as also found in the stock market.
IV. Supply and Demand Theories
This section investigates various supply and demand theories that have been suggested as causes of the January effect observed in the equity market.
A. Seasonal Buying and Selling Patterns
One explanation of financial markets seasonality is the tendency for investors to overreact to information around the end of the year. For example, DeBondt and Thaler (1985, 1987) and Dyl and Maberly (1992) find evidence to suggest that the seasonal returns found in stock prices are related to investor psychology and subsequent seasonal buying and selling patterns. In the same context, Lakonishok, Shleifer, Thaler, and Vishny (1991) find evidence that equity pension fund managers tend to sell "losers" at the end of quarters, especially at the end of the fourth quarter. In the corporate bond market, Cooper and Shulman (1994) find a similar practice of selling "winners" and "losers" at year-end.
While there is support for year-end selling, it does not adequately explain the January effect. Investors may sell their holdings at year-end but if they reinvest the proceeds the net impact on the market should be neutral unless there is a shifting of investments across categories. Therefore, for year-end selling to explain the January effect adequately, there should be a delay between the selling and subsequent reinvestment around the turn of the year. But if there is a delay in reinvestment, the delay would provide the shift in the demand necessary to explain the January effect.
To examine the hypothesis that the January effect in the bond market is related to a shift in demand around the turn of the year, I collected data from the Investment Company Institute. Mutual funds report month end information on the percentage of their portfolios held in liquid assets to the Investment Company Institute. A shift in demand from December to January, as assets are sold but not reinvested until January, suggests that the percentage of liquid assets held by mutual funds should increase at the end of December. Moreover, to adequately explain the discrepancy in the strength of 'the January effect in investment-grade as compared to noninvestment-grade bonds, there should be an increase in the liquidity position of noninvestment-grade-bond mutual funds but not in the investment-grade-bond mutual funds.
To examine this theory, I perform a seasonal analysis on the liquidity position of investment-grade and noninvestment-grade bond funds on month end data from January 1986 to March 1997. The results appear in Table 4. The December coefficients represent changes in the liquidity position of the funds. Both the investment-grade and noninvestment-grade funds have negative, though not statistically significant, coefficients. Clearly, these results indicate that mutual funds do not actively sell assets in December and retain liquid assets until January. A puzzling result is the statistically significant increase in the liquidity position of noninvestment-grade bond funds in January. I suggest an explanation for this phenomenon later in the paper.
Even though previous studies have demonstrated the tendency for investors to sell winners and losers around the turn of the year, I find no evidence that this selling would lead to the shift in demand necessary to explain the January effect. Nor do I find support for any downward pressure on prices in December, as suggested by year-end-selling theories. Lastly, none of the year-end-selling theories would have a rationale for the strength and prevalence of the January effect for noninvestment-grade bonds but not for investment-grade bonds.
B. Noninvestment-Grade Bond Supply
A commonly heard hypothesis on Wall Street for the January effect is an undersupply of noninvestment-grade bonds in January. It has been suggested that there is seasonal variation in new issues of noninvestment-grade bonds or, more specifically, a lack of new issues in January, which is believed to drive up the prices of existing bonds. A decrease in the new-issue bond supply of noninvestment-grade bonds in January could certainly account for the January effect in the corporate bond market and at the same time explain the different behavior of investment-grade and noninvestment-grade bonds.
To examine this hypothesis, I obtain monthly supply information for noninvestment-grade bonds from CS First Boston, which collects monthly information on the total supply of noninvestment-grade bonds. Their calculation of the total supply of noninvestment-grade bonds includes new issues (public and 144a), debt retirements, and net rating downgrades/upgrades. New-issue supply is the dominant factor in the change in the total supply of high-yield bonds, but to account for other concurrent changes, I use the total-market supply.(3) Unfortunately, the data are only available on a monthly basis starting in January 1992, which limits the conclusions that can be drawn.
While the sample period is limited, the data reported in Table 5 at least suggest that there is no systematic reduction in the total supply of noninvestment-grade bonds in January. In fact, over the limited sample period, there is an average increase in the net new-issue supply in January. This finding calls into question the validity of the hypothesis that the January effect in the bond market results from a lack of supply in January.
C. Individual Investor Seasonal Demand
The impact of investment flow into mutual funds on market returns is clearly demonstrated in the stock and bond markets (see Warther, 1995) and on the yield of noninvestment-grade bonds (see Fridson and Jonsson, 1995). These results suggest that the January effect could be a function of a seasonal shift in demand by individual investors in favor of high-yield bonds, which would be consistent with the findings of DeBondt and Thaler (1985, 1987) and Dyl and Maberly (1992).
To examine this hypothesis, I model the monthly percentage change in the total assets invested in high-yield mutual funds using Investment Company Institute data (see Table 6). Since total assets under management have increased over time, the constant is positive for both the investment-grade and noninvestment-grade funds. To account for an autoregressive process, I included a single lag of the dependent variable as well as the exogenous shocks discussed in Section II. The January and December coefficients reflect any positive or negative change above or below the constant.
The results in Table 6 show that the amount of noninvestment-grade mutual fund assets has increased historically by 0.63% per month. In January, noninvestment-grade mutual fund assets increase an additional, statistically significant, 1.89%. This finding of a systematic increase in the investment flow in January into the noninvestment-grade-bond mutual funds would put upward pressure on prices, which is consistent with the January effect.
For a seasonal increase in demand to be a convincing theory, it must explain why the January effect is limited to the noninvestment-grade-bond [TABULAR DATA FOR TABLE 4 OMITTED] market and also explain why the anomaly is stronger for lower-rated bonds. To examine why there is little, if any, January effect for investment-grade bonds, I obtain the percent monthly change in total assets for investment-grade-bond-mutual funds from the Investment Company Institute and examine any seasonal shifts in demand. Table 6 reports the results. Unlike the noninvestment-grade mutual funds, there is no statistically significant increase in supply in January for the investment-grade bond funds. This result helps answer the question of why there is little or no corresponding January effect found in the investment-grade bond market.
Table 5. Change in Monthly Supply of High-Yield Bonds (January 1992 to April 1997) This table presents the results of a regression on the average change in the monthly supply of high-yield bonds. The January coefficient measures an additional change that can be ascribed to that particular month. The monthly supply of high-yield bonds is nonstationary, so the process is modeled using the stationary first difference. Variable Coefficient Standard Error t-Value t-Probability Constant -60.46 308.85 -0.196 0.845 January 1,601.50 1,096.30 1.461 0.149
An increase in mutual fund flow into high-yield funds could also provide an answer for why the January effect increases as credit quality decreases. While high-yield-bond mutual funds make up 21% [ILLUSTRATION FOR FIGURE 1 OMITTED] of the total high-yield market, they are believed to be the dominant holder of B-rated bonds, which represent 43% of the market.(4) This could explain why there is a stronger January effect among the bonds with the lowest ratings.
A significant increase in investor demand for noninvestment-grade bonds in January also provides a possible explanation of the anomalous finding regarding the liquid assets of bond mutual funds. The increase in the percentage of liquid assets held by noninvestment-grade-bond funds at the end of January is consistent with the increased flow of funds into noninvestment-grade-bond funds in January.
Finally, note the effect of the exogenous shocks on the demand for investment-grade and noninvestment-grade bonds in Table 6. While the signs of the changes are consistent for both the investment-grade and noninvestment-grade mutual funds during all the [TABULAR DATA FOR TABLE 6 OMITTED] shocks, the magnitude and statistical significance are very different.
D. Window Dressing
Window dressing offers another possible explanation for the anomaly. Window dressing is the year-end practice of financial institutions to "clean up" their financial reports. The term can refer to a number of different practices. For example, Lakonishok, Shleifer, Thaler, and Vishny (1991) and Cooper and Shulman (1994) describe the practice as selling "losers" at year-end. Fridson (1994) describes window dressing as the year-end practice of adjusting portfolio weighting to meet compliance restrictions on industry weighting and bond credit quality. In practice, window dressing often results in bond portfolio managers, insurance companies, and pension funds selling lower-quality issues at year-end to raise the average quality of their portfolios. This practice is then reversed at the beginning of the year as funds are reinvested in the lower-credit-quality issues. This shift in demand from lower to higher-credit quality and then back again to lower-credit quality could explain the difference in the strength of the anomaly as bond ratings decrease.
The shifting of funds out of the lower-credit qualities by mutual funds is not done in a vacuum. If there are significant supply and demand pressures in the market that impact the BB-rated and B-rated bonds, these same forces should also apply in the opposite direction to an offsetting asset class.
One possible alternative asset class is liquid investments. Another alternative asset class is Split-BBB bonds. The Split-BBB category is an attractive alternative asset class to investors who shift funds around the turn of the year. Split-BBB bonds are rated investment grade by one of the two major rating agencies and noninvestment-grade by the other. Regulatory agencies generally use the highest bond rating in classifying an institution's portfolio. Hence, the Split-BBB category of bonds provides an attractive yield while allowing institutional investors to report a higher percentage of investment-grade bonds in their portfolios at the fiscal year-end. To examine the hypothesis that the anomaly is in part a shifting of assets to different classes due to window dressing, I model the seasonal component of the yield on the CS First Boston Split-BBB index. For comparison purposes, I also model the two adjacent rating indices, the BBB and BB rating categories.
1. Modeling the Seasonal Component of the BBB, Split-BBB, and BB Indices
To examine the impact of seasonal fluctuation on yields accurately and to reduce "noise" in the model, I control for other factors that may affect relative yields. Barnhill, Joutz, and Maxwell (1997) analyzed factors that affect the yield on noninvestment-grade securities. They found that the market drivers for the noninvestment-grade-bond market include a long-run mean reverting relation between the yield on noninvestment-grade securities and treasury bond yields and default rates. Using error correction models (ECMs), the authors found that exogenous shocks impact the short-run dynamics of the different noninvestment-grade indices. Thus, to analyze the January effect, I developed error correction models for the Split-BBB index and, for comparison, the BBB and BB indices.
2. Cointegration Analysis
I test corporate bond yields, treasury yields, and default rates(5) for stationarity utilizing the augmented Dickey-Fuller test statistic. I find that all the variables are nonstationary, but the first differences are stationary. To avoid a potentially spurious regression, the traditional approach in modeling nonstationary variables is to difference the variables to induce stationarity.(6) However, this can lead to a loss of information on the long-run relationship among variables. Therefore, I tested to determine if there are any cointegrating vectors. The implication of a cointegrating vector is that, although the individual variables may be nonstationary, a linear combination of the variables is stationary (see Enders, 1995). Thus, a cointegrating vector implies a long-run stationary relationship among the variables.
Two methodologies are available to test for a cointegrating vector. The Engle-Granger methodology (see Engle and Granger, 1987) is appropriate for a bivariate analysis. However, the Johansen maximum likelihood procedure for finite-order vector autoregressions (see Johansen, 1988, 1991) is more appropriate for a multivariate analysis. Given the multivariate nature of the present study, I used the Johansen methodology.
The first step in testing for a cointegrating vector is to determine the appropriate lag structure in the vector autoregression (VAR), which is done by testing the residuals as the lag structure is reduced. The null hypothesis is that there is no significant difference in the residuals as the model lag structure is reduced. The log-likelihood, Schwartz criterion, Hannan-Quinn, and the F-statistics are then used to test the significance of the change in the model's residuals. All the test statistics suggested that a lag structure of two periods is appropriate.
Table 7 reports the results of the test of a long-term relation among treasury yields, default rates, and the yields on the different indices. The maximal and trace eigenvalue statistics test the stationarity of the cointegrating vector and can be considered a multivariate Dickey-Fuller test. For example, the null hypothesis of r(rank) = 0 is that there is no stationary cointegrating vector. The Johansen maximal and trace eigenvalues are the test statistics. All the maximal and trace eigenvalue statistics reject (at the 99% confidence level) the null hypotheses that there are no cointegrating vectors for both the Split-BBB and BB indices.
The implication is that there is at least one cointegrating (stationary) vector or long-run solution for these indices. The null hypothesis of r I is that there is one or more cointegrating vector(s). All the maximal and trace eigenvalue statistics suggest little evidence of more than one cointegrating vector, and I conclude that there is a single cointegrating vector for both the Split-BBB and BB indices.
By restricting the beta coefficients of the treasury bond yield and then the default rate and then reestimating the cointegrating vectors, the significance of the variables in the cointegrating vectors can be tested by using chi-square statistics. The chi-square [Mathematical Expression Omitted] statistics suggest that both the treasury bond yield and the default rate individually add explanatory power to the long-run solution for both the Split-BBB and BB indices.
No statistically significant conclusion can be drawn about a cointegrating vector for the BBB index. However, given the cointegration results and the chi-square statistics, the yield on the BBB index seems to be insensitive to the default rate or at least to small changes in the default rate.
The standardized [[Beta].sup./] eigenvectors in Table 7 are the estimated cointegrating vectors for the indices' long-term market equilibrium or the long-run solution. The cointegrating vectors for the Split-BBB and BB indices are written as follows:
[Epsilon] = Split-BBB Yield - 0.980 T-Bond Yield - 0.065 Moody's Default Rate (1)
[Epsilon] = BB Yield - 0.843 T-Bond Yield - 0.149 Moody's Default Rate (2)
or algebraically manipulated with [Epsilon] having an expectation of zero as:
Split-BBB Yield = 0.980 T-Bond Yield + 0.065 Moody's Default Rate (3)
([Alpha] coefficient = -0.134)
BB Yield = 0.843 T-Bond Yield + 0.149 Moody's Default Rate (4)
([Alpha] coefficient = -0.169)
[TABULAR DATA FOR TABLE 7 OMITTED]
The standardized eigenvector found in Equations (3) and (4) indicates that in the long run, the Split-BBB and BB indices' yields are essentially the risk-free rate plus a premium that reflects default risk. As expected, the default premium increases as the bond rating declines. The [Alpha] coefficients in Table 7 (also found under Equations (3) and (4)) represent the speed of adjustment to disequilibrium comparable to a mean reversion rate. The signs of the a coefficients are as expected: a negative sign indicates that as the variables move away from equilibrium, there is an adjustment back towards the equilibrium relation. The smaller the [Alpha] coefficient, the quicker the adjustment. Thus, given the smaller [Alpha] coefficient, the BB index reverts faster to the long-run relation.
3. Estimating an Error Correction Model for the Indices
After finding a cointegrating vector in a system of equations, I estimate the indices as single-equation error correction models (ECM) (for further discussion of ECMs, see Enders, 1995, and Hendry, 1995). An error correction model is simply a reparameterized single-equation autoregressive distributed lag (ADL) model. For ease of comparison, I also use an ECM to estimate the yield on the BBB index even though no statistically significant cointegrating vector was found.(7) The resulting ECMs combine a short-run dynamic analysis with the long-run relation found in the cointegration analysis. For example, Equation (5) represents a two-period ADL model of the Split-BBB index algebraically manipulated into a one-period ECM which directly incorporates the long run solution:
[Delta][Split-BBB.sub.t] = [a.sub.0] + [b.sub.1][Delta][Split-BBB.sub.t-1] + [summation of] [b.sub.2i][Delta][T-Bond.sub.t-i] where i = 0 to 1 + [summation of] [b.sub.3i][Delta][DefaultRate.sub.t-i] where i = 0 to 1 + [c.sub.1] [(Split-BBB + [Epsilon] T-Bond + [Delta] DefaultRate).sub.t-1] + [a.sub.4] 1987[Crash.sub.t] + [a.sub.5][IraqInvasion.sub.t] + [a.sub.6][KuwaitLiberation.sub.t] + [a.sub.7][January.sub.t] + [a.sub.8][December.sub.t] (5)
I use the reparameterized general ECM formulations as models to discuss the single equation results. [c.sub.1] is the feedback coefficient that reflects a long-run adjustment to disequilibrium, and is comparable to the a coefficients from the cointegration analysis found in Table 7. Therefore, the error correction variable [(Split-BBB + [Epsilon] T-Bond + [Delta] DefaultRate).sub.t-1] can be represented from the cointegration results found in Equation (1) as:
[ECM.sub.Split-BBB] = [[Split-BBB - 0.98(T-Bond) - 0.07(Default Rate)].sub.t-1] (6)
Table 8 contains the results of the estimated ECMs on the yield on the different indices. based upon global F tests, all the models are statistically significant. The ECMs decrease in accuracy, based on [R.sup.2] and standard deviation, as the credit quality declines.
In comparing the January effect, the BBB index had little variation in return associated with January or December, and the BB index had a statistically significant decrease in yield in January. The decrease in the BB yield in January is consistent with the findings in previous studies and with a January effect.
The important result in Table 8 is the finding that there is a statistically significant (at the 95% confidence level) increase in yield on the Split-BBB index in January, and that, though not statistically significant, the coefficient for December is negative. The signs of the January and December coefficients for the Split-BBB index are the opposite of what would be expected in the presence of a January effect. Overall, a contra-January effect is evident in the behavior of the yield on the Split-BBB index, which supports the hypothesis that part of the January anomaly is the result of window dressing by portfolio managers. That is, there is a substitution between the lower-quality noninvestment-grade bonds and the lowest investment-grade bonds.
I find a statistically significant January effect in the return data for noninvestment-grade bond indexes. While I also find a positive excess return in January in the lowest investment-grade category, the results are not statistically significant. Similar to previous studies, I find that excess returns in January increase as credit quality decreases. Given the demonstrated relationship between a firm's market value and its bond rating, these results are consistent with the small-firm effect in stocks.
My results reject two hypotheses about the cause of the January anomaly in the bond markets. First, no support is found for a systematic decrease in January in the supply of noninvestment-grade bonds. Second, I find no seasonal anomaly in the selling and buying patterns of high-yield and investment-grade bond mutual funds around the turn of the year.
However, I do find support for the January anomaly in the bond market resulting in part from an increase in the demand by individual investors during January for high-yield bonds but not for investment-grade bonds. This difference between the shift in demand for investment-grade and noninvestment-grade bonds is consistent with previous findings of a January effect in the market for noninvestment-grade bonds but not in the market for investment-grade bonds. My finding is also consistent with the suggestion made by some market commentators that individual investor psychology partly accounts for the January effect.
I also find evidence that the January effect in the corporate bond market is at least partly the result of window dressing. I find a statistically significant increase in yield in the lowest investment-grade bond category (Split-BBB) in January. The increase in yield in the Split-BBB category is the opposite of what would occur with the January effect, which I found results in a decrease in the yield of noninvestment-grade bonds in January. Thus, an increase in the Split-BBB yield, an alternative asset class, is consistent with a window-dressing explanation for the January effect: portfolio managers shift their portfolio weighting around the [TABULAR DATA FOR TABLE 8 OMITTED] turn-of-the year to improve the appearance of their year-end reports to investors. Overall, my results demonstrate that the January effect in the corporate bond market is not a function of a single causal factor, but is instead a function of several factors coinciding around the turn of the year.
This work was completed while working as a research fellow at the Financial Markets Research Institute at George Washington University. The author gratefully acknowledges the financial support of Shenkman Capital Management, Inc. and data provided by CS First Boston, the Investment Company Institute, and Moody's Investors Service. I would also like to thank Theodore Barnhill, Michael Ferri, Mark Klock, and Fred Joutz for comments and suggestions.
1 A split rating occurs when Moody's and Standard & Poor's assign different ratings to a bond.
2 To ensure the robustness of the findings, 1 also estimated the models without the exogenous variables. The results were similar but the models that excluded the shocks had larger standard errors.
3 The results are insensitive to the use of new issue supply or total market supply.
4 I reached this conclusion after conversations with market participants. However, I can find no data to document this implied market segmentation.
5 Citibase is used to determine the investment grade and treasury yield information. Citibase provides Moody's yield information. For convenience, the Citibase Baa category is labeled BBB to coincide with the rating method used by CS First Boston, which provided the yield information for the Split-BBB, BB, and B indices utilized in this study. Moody's Investors Service furnished the default information.
6 To examine the robustness of the results, I also estimated the models using the traditional approach of differencing the variables to induce stationarity. The results from the differenced models were consistent with those from the cointegration models. However, the overall fits of the differenced models were at least 20% worse based on a comparison of [R.sup.2].
7 The BBB index is also estimated using a differenced model. The results of the differenced model are similar to the results reported for the ECM.
Barnhill, T., F. Joutz, and W. Maxwell, 1997, "Factors Affecting the Yield of Noninvestment Grade Bond Indices," George Washington University Working Paper 97-46 (August).
Bhardwaj, R. and L. Brooks, 1992, "The January Anomaly: Effects of Low Share Price, Transaction Costs, and Bid-Ask Bias," Journal of Finance (June), 553-575.
Black, F. and M. Scholes, 1973, "The Pricing of Options and Corporate Liabilities," Journal of Political Economy (May/June), 637-654.
Brown, P., D. Keim, A. Kleidon, and T. Marsh, 1983, "Stock Return Seasonality and the Tax-Loss Selling Hypothesis: Analysis of the Arguments and Australian Evidence," Journal of Financial Economics (June), 105-128.
Chang, E. and R. Huang, 1990, "Time-Varying Return and Risk in the Corporate Bond Market." Journal of Financial and Quantitative Analysis (September), 323340.
Chang, E. and M. Pinegar, 1986, "Return Seasonality and Tax-Loss Selling in the Market for Long-Term Government and Corporate Bonds," Journal of Financial Economics (December), 391-415.
Cho, C. and W. Taylor, 1987, "The Seasonal Stability of the Factor Structure of Stock Returns," Journal of Finance (December), 1195-1211.
Cooper, R. and J. Shulman, 1994, "The Year-End Effect in Junk Bond Prices," Financial Analysts Journal (September/October), 61-65.
DeBondt, W. and R. Thaler, 1985, "Does the Stock Market Overreact?" Journal of Finance (July), 793-805.
DeBondt, W. and R. Thaler, 1987, "Further Evidence on Investor Overreaction and Stock Market Seasonality," Journal of Finance (July), 557-581.
DeRosa-Farag, S., 1996, 1995 High Yield Market Review, New York, NY, Chase Securities, Inc.
Dyl, E and E. Maberly, 1992. "Odd-Lot Transactions Around the Turn of the Year and the January Effect," Journal of Financial and Quantitative Analysis (December), 591-604.
Enders, W., 1995, Applied Econometric Time Series, New York, NY. John Wiley & Sons, Inc.
Engle, R. and C. Granger, 1987, "Cointegration and Error-Correction: Representation, Estimation, and Testing," Econometrica (March), 251-276.
Fama, F. and K. French, 1993. "Common Risk Factors in the Returns on Stocks and Bonds," Journal of Financial Economics (February), 3-56.
Fridson, Martin, 1994, "Do High-Yield Bonds Have an Equity Component'?" Financial Management (Summer), 8284.
Fridson, M. and C. Garman, 1995, "January Effect: Probably Not a Function of Coupon Flows," This Week in High Yield (December), 1-3.
Fridson, M. and C. Garman, 1995, "January Effect: The Longer-Term Evidence," This Week in High Yield (December), 4-6.
Fridson, M. and J. Jonsson, 1995, "Spread Versus Treasuries and the Riskiness of High-Yield Bonds," Journal of Fixed Income (December), 79-88.
Gultekin, M. and B. Gultekin, 1983, "Stock Market Seasonality: International Evidence," Journal of Financial Economics (December), 469-481.
Gultekin, M. and B. Gultekin, 1987, "Stock Return Anomalies and Tests of the APT," Journal of Finance (December), 1213-1224.
Hendry, D., 1995, Dynamic Econometrics, Oxford, England, Oxford University Press.
Johansen, S., 1988, "Statistical Analysis of Cointegration Vectors," Journal of Economic Dynamics and Control (June/September), 231-254.
Johansen, S., 1991, "Estimation and Hypothesis Testing of Cointegrating Vectors in Gaussian Vector Autoregressive Models," Econometrica (November), 1551-1580.
Jones, C., D. Pearce, and J. Wilson, 1987, "Can Tax-Loss Selling Explain the January Effect?" Journal of Finance (June), 453-461.
Keim, D., 1983, "Size Related Anomalies and Stock Return Seasonality: Further Empirical Evidence," Journal of Financial Economics (June), 13-32.
Kihn, J., 1996, "The Financial Performance of Low-Grade Municipal Bond Funds," Financial Management (Summer), 52-73.
Lakonishok, J., A. Shleifer, R. Thaler, and R. Vishny, 1991, "Window Dressing by Pension Fund Managers," American Economic Review (May), 227-231.
Merton, R., 1974, "On the Pricing of Corporate Debt: The Risk Structure of Interest Rates," Journal of Finance (May), 449-470.
Reinganum, M., 1983, "The Anomalous Stock Market Behavior of Small Firms in January: Empirical Tests for Tax-Loss Selling Effects," Journal of Financial Economics (December). 89-104.
Roll, R., 1983, "Vas Ist Das? The Turn-of-the-Year Effect and the Return Premia of Small Firms," Journal of Portfolio Management (Winter), 18-28.
Seyhun, N., 1988, "The January Effect and Aggregate Insider Trading," Journal of Finance (March), 129-141.
Shane, H., 1994, "Comovements of Low-Grade Debt and Equity Returns of Highly Leveraged Firms," Journal of Fixed Income (March), 79-89.
Tinic, S., G. Barone-Adesi, and R. West, 1987, "Seasonality in Canadian Stock Prices: A Test of the "Tax-Loss-Selling" Hypothesis." Journal of Financial and Quantitative Analysis (March), 51-63.
Warther, V., 1995, "Aggregate Mutual Fund Flow and Security Returns," Journal of Financial Economics (October/November), 209-235.
William F. Maxwell is a Visiting Assistant Professor of Finance at Georgetown University.
|Printer friendly Cite/link Email Feedback|
|Author:||Maxwell, William F.|
|Date:||Jun 22, 1998|
|Previous Article:||Market efficiency and the returns to technical analysis.|
|Next Article:||The information content of corporate offerings of seasoned securities: an empirical analysis.|