1. Introduction
Traditional fixed-income allocations based on market-value weights often fail to reflect meaningful differences in how portfolios respond to macroeconomic conditions. Two portfolios with identical value-based allocations may behave differently when interest rates change because the durations of their underlying bonds vary. To address this limitation, practitioners rely on duration contributions, which combine each segment’s market weight with its duration to produce a more accurate measure of interest rate sensitivity.
Credit portfolio management adopts a related metric—spread-duration contribution—to capture the portfolio’s exposure to movements in credit spreads. Allocating spread-duration exposure across sectors or issuers is a central task in credit portfolio construction. However, spread duration alone does not fully capture credit risk. A portfolio may match a benchmark’s spread-duration profile yet assume greater credit risk if it holds bonds with wider spreads within each segment. Such bonds trade at wider spreadsbecause the market assigns them higher credit risk and, as a result, their spreads tend to react more sharply to systematic shocks. Although this high-beta behavior is well recognized, it is often treated as secondary rather than a fundamental component of portfolio allocation.
The broader credit risk literature underscores the importance of incorporating spread risk into portfolio construction and valuation. Foundational studies show that credit spreads embed both issuer-specific and systematic risk factors (Collin-Dufresne et al., 2001; Jones & Mason, 1980), while other work highlights the need for more refined measures that capture differences in credit quality and sector-specific exposures (Elton et al., 2001; Huang & Huang, 2012). To address these issues, Dor, Dynkin, Hyman, Houweling, van Leeuwen, and Penninga (2007) introduce Duration Times Spread (DTS), which extends spread duration by incorporating the spread level itself. By linking volatility to relative rather than absolute spread changes, DTS offers a more informativerepresentation of credit risk and has been shown to improve the forecasting of excess return volatility.
Furthermore, existing studies on the Korean corporate bond market provide extensive evidence on the behavior and determinants of credit spreads. Prior research documents that Korean corporate bond spreads reflect default risk, liquidity risk, and macroeconomic conditions, and that spreads widen and become more persistent during periods of economic stress (Byun & Jang, 2004; Kim, 2009; Kim, Jang, & Kim, 2023; Kim & Lee, 2011; Nam, 2014; Ryu & Lee, 2015). More recent studies further highlight the role of policy uncertainty, information, and sentiment in shaping spread dynamics (Hwang & Rhee, 2022; Koo, Lee, & Kang, 2024; Yoon, Son, & Lim, 2025).
Despite this rich literature, prior studies seldom examines how spread volatility scales with the level of spreads in the Korean corporate bond market, nor evaluates whether relative spread changes provide a stable and economically meaningful measure of credit risk in the same market. In particular, the DTS framework—widely used by practitioners and validated in global markets—has not been formally tested in the Korean context. This study addresses this gap by empirically evaluating the DTS framework and its implications for excess return volatility in the Korean corporate bond market. To guide the empirical analysis, we develop a set of testable hypotheses that capture the expected relationships between spreads, volatility, and DTS. These hypotheses are presented in Section III and provide the analytical foundation for the subsequent empirical tests.
This study makes several contributions to the literature. First, it provides the comprehensive empirical evaluation of DTS in the Korean market—an environment that remains understudied despite its distinctive structural characteristics, including concentrated issuance, relatively low secondary-market liquidity, and a rating distribution skewed toward high-grade issuers. Second, we document a strong positive relationship between spread levels and both systematic and idiosyncratic spread volatility, offering empirical support for the theoretical foundation of DTS in a non-U.S. setting. Third, we show that excess return volatility increases proportionally with DTS and that DTS-based volatility forecasts are more stable and informative than those derived from absolute spread changes. Collectively, these findings extend the applicability of DTS beyond the global markets in which it was originally developed and demonstrate its value as a risk-sensitive metric for portfolio construction, volatility forecasting, and relative-value assessment within the Korean fixed-income market.
The remainder of the manuscript is structured as follows. Section II reviews the existing literature and explains how this study builds upon it. Section III describes the data and methodology. Section IV presents the empirical analysis. Section V discusses how DTS can be used to measure excess return volatility. Sections VI and VII discuss implications and conclude.
2. Literature Review
Research on corporate bond credit spreads emphasizes the role of credit risk, liquidity, and macroeconomic forces in determining bond pricing. Foundational work by Jones and Mason (1980), Elton et al. (2001), and Collin-Dufresne et al. (2001) establishes that credit spreads incorporate both firm-specific default risk and systematic components related to market-wide conditions. Huang and Huang (2012) further show that only a portion of the corporate-treasury spread is attributable to pure credit risk, with the remainder linked to liquidity and risk premia. These insights motivate the development of improved credit risk sensitivity measures.
Dor et al. (2007) introduce DTS as a framework for capturing a bond’s sensitivity to relative spread changes rather than absolute spread changes. DTS incorporates both spread duration and the level of the spread, making it particularly effective when credit spread volatility is proportional to prevailing spread levels. Dor, Polbennikov, and Rosten (2007) extend the DTS concept to credit derivatives, and de Jong (2023) demonstrates its usefulness in modeling and predicting volatility patterns. Despite its conceptual importance, DTS has rarely been empirically evaluated in Asian markets.
2.1 Evidence from the Korean Corporate Bond Market
Empirical research on the Korean corporate bond market examines credit spreads from two complementary perspectives: the behavior and dynamics of credit spreads themselves, and the economic factors that determine their level and variation.
Early work by Byun and Jang (2004) analyzes credit risk in Korean corporate bonds using a structural model framework and shows that default risk is an important component of observed corporate bond spreads. Their study provides one of the earliest systematic analyses of credit risk pricing in the Korean bond market and establishes a foundation for subsequent research on spread behavior. Focusing on the informational content of credit spreads, Kim and Lee (2011) document that spreads constructed from BBB-rated corporate bonds contain significant predictive power for future real economic activity in Korea, highlighting the macroeconomic relevance of spread movements. Kim (2009) further shows that credit spreads in emerging bond markets, including Korea, display sharp increases in both level and volatility during periods of financial stress, with spread volatility rising disproportionately during downturns.
Several studies investigate the dynamic properties of Korean credit spreads across credit quality and market conditions. Ryu and Lee (2015) find that spreads associated with lower-rated bonds exhibit greater persistence and slower mean reversion, implying that higher-risk bonds experience more prolonged spread adjustments. Kang and Lee (2015) show that credit spreads provide valuable information for forecasting movements in the Korean government bond yield curve, particularly when structural breaks are taken into account. Together, these studies indicate that Korean credit spreads vary substantially across ratings and economic regimes and that spread behavior becomes more pronounced during periods of stress.
A second strand of literature focuses on identifying the determinants of credit spread variation. Nam (2014) documents the countercyclical nature of Korean corporate bond spreads, showing that spreads widen during economic downturns and narrow during expansions. Hwang and Rhee (2022) demonstrate that economic policy uncertainty— especially uncertainty originating from the United States—significantly affects Korean corporate bond spreads, with heterogeneous effects across bond types and credit ratings. Thesefindings suggest that Korean credit spreads reflect not only issuer-specific default risk but also broader macroeconomic and global risk factors.
More recent studies employ advanced empirical techniques to analyze and forecast credit spreads. Koo, Lee, and Kang (2024) show that firm-specific textual sentiment extracted from news and online sources is significantly related to subsequent changes in Korean corporate bond spreads, particularly during crisis periods. Yoon, Son, and Lim (2025) find that deep learning models such as GRU and LSTM outperform traditional econometric approaches in forecasting Korean corporate bond credit spreads, with predictive performance varying across credit ratings and forecast horizons. These studies underscore the growing importance of information and expectation channels in shaping spread dynamics.
Finally, Kim, Jang, and Kim (2023) provide direct evidence that liquidity risk is systematically priced in the Korean corporate bond market. Using a comprehensive dataset, they show that bonds with greater exposure to liquidity risk command higher yieldseven after controlling for credit ratings and other bond characteristics. Their results reinforce the view that liquidity conditions play a central role in determining both the level and variability of credit spreads in Korea.
In summary, existing Korean studies consistently indicate that (i) credit spreads in the Korean corporate bond market reflect default risk, liquidity risk, and macroeconomic conditions; (ii) spread levels and persistence increase for lower-rated and riskier bonds; and (iii) spread behavior becomes more pronounced during periods of heightened market stress. However, despite this extensive literature on the level, dynamics, and determinants of Korean credit spreads, no prior study explicitly examines how spread volatility scales with spread levels, nor evaluates whether relative spread changes provide a stable and economically meaningful measure of credit risk. This study addresses that gap by empirically testing the DTS framework in the Korean corporate bondmarket and linking spread volatility directly to excess return volatility.
3. Methodology
3.1 Duration-Times-Spread (DTS)
DTS is introduced by Dor et al. (2007) as a measure of credit risk sensitivity that captures a bond’s exposure to relative changes in credit spreads rather than absolute changes. The distinction is important because empirical evidence, including the results in this study, shows that the volatility of spread changes tends to scale with the level of the spread. DTS formalizes this intuition within a standard risk-sensitivity framework.
Let D denote the spread duration of bond i, which measures the sensitivity of the bond’s price to a 1-basis-point absolute change in its credit spread, and s denote the credit spread of bond i, expressed in basis points.
3.1.1 Absolute spread change framework
Consider a monthly change in credit spread, Δsi. The return attributable solely to this spread movement is:
which states that a bond’s excess return is approximately proportional to its spread duration and the magnitude of the absolute spread change. This formulation is commonly used in fixed-income risk management.
Rewriting Eq. (1), we obtain:
Equation (2) expresses return sensitivity in terms of the relative change in spreads, Δ s i s i
which is the Duration-Times-Spread measure, which represents the sensitivity of a bond’s price to a 1-percent relative change in its spread. More precisely, spread duration captures sensitivity to absolute spread changes while DTS captures sensitivity to relative spread changes. When wider-spread bonds experience larger relative changes in spreads—an empirical regularity shown later in the paper—DTS becomes a more stable indicator of risk across securities. Thus, DTS embeds the economic idea that higher-spread bonds are inherently riskier, not merely because they have higher spreads, but because their spread changes are proportionally larger.
3.1.2 Volatility implications
Under Eq. (1), the volatility of excess returns is approximated by:
Under Eq. (2), the corresponding volatility expression becomes:
Equations (3) and (4) highlight the distinction between the two frameworks. Eq. (3) links excess-return volatility to the volatility of absolute spread changes while Eq. (4) links excess-return volatility to the volatility of relative spread changes.
Although Equations (3) and (4) are algebraically related, they differ in their statistical behavior when spread volatility depends on spread levels, one question remains: Why prefer Eq. (4) over Eq. (3)? We provide evidence that (i) the volatility of bothsystematic and idiosyncratic spread changes is proportional to the spread level; (ii) the volatility of relative spread changes is considerably more stable across time, sectors, ratings, and duration groups; and (iii) DTS explains excess return volatility more accurately than spread duration.
Given these empirical properties, Eq. (4) provides a better-behaved and more stable risk estimator than Eq. (3), because the latter is mechanically sensitive to the scale of the spread. This justifies the use of DTS as the primary measure of spread-risk exposure, particularly in markets such as Korea, where dispersion in spread levels is substantial, and volatility is strongly level-dependent.
3.1.3 Hypotheses
Empirical studies of the Korean corporate bond market consistently show that credit spreads are sensitive to default risk, liquidity conditions, and macroeconomic fluctuations, and that spread behavior becomes more pronounced during periods of market stress (Byun & Jang, 2004; Kim, 2009; Kim, Jang, & Kim, 2023; Kim & Lee, 2011; Nam, 2014).. Evidence also indicates that lower-rated bonds exhibit greater persistence and slower adjustment in spreads, implying higher volatility and risk exposure (Ryu & Lee, 2015). More recent studies document that policy uncertainty and information-related factors significantly affect Korean corporate bond spreads, further amplifying spread dynamics in turbulent periods (Hwang & Rhee, 2022; Koo, Lee, & Kang, 2024).
While these studies provide important insights into the level, dynamics, and determinants of credit spreads in Korea, they do not directly address how spread volatility scales with spread levels, nor whether relative spread changes offer a stable basis for measuring credit risk. Therefore, building on the DTS framework proposed by Dor et al. (2007) and subsequent evidence on proportional spread volatility (de Jong, 2023), we formulate the following testable hypotheses.
First, credit spreads in global markets have been shown to exhibit volatility that scales with spread levels (Dor et al., 2007; de Jong, 2023). If this proportionality holds in the Korean corporate bond market, both systematic and idiosyncratic spread volatilities should increase with the underlying spread. Therefore, we propose:
H1. Spread volatility is positively related to the level of credit spreads in the Korean bond market.
Under the DTS framework, volatility of excess returns should be approximately proportional to the product of duration and spread (Dor et al., 2007). If the framework applies in Korea, portfolios (or buckets) with higher DTS should exhibit correspondingly higher excess return volatility. Hence, we propose:
H2: Excess return volatility increases with DT) in the Korean bond market.
If relative spread changes are more stable than absolute spread changes—as documented by Dor et al. (2007) and de Jong (2023)—then DTS, which embeds relative spread volatility, should produce volatility forecasts that more closely match realized excess return volatility than traditional spread-duration measures. This implies smaller forecast errors and more stable residual distributions. Therefore, we propose:
H3: Volatility forecasts based on DTS provide more stable and informative predictions than those based on absolute spread changes in the Korean bond market.
3.2 Data
Although a fully integrated dataset would be ideal, no single source provides all information required for Korean corporate bond analysis. We therefore construct a unified dataset by combining several complementary sources and applying consistent preprocessing steps.
The first dataset (the price_year dataset) is obtained from Korea Asset Pricing and contains bond-level yields, prices, modified durations, and basic bond characteristics. The sample spans January 2010 to August 2020, covering 51,229 corporate bonds and producing 3,830,850 bond-month observations. This dataset serves as the primary input for computing excess returns, yields, and duration measures.
The second dataset (the sector_data dataset) is obtained from FnGuide and provides issuer- and sector-level classifications that are not available in the price_year dataset. This dataset covers the same period and includes 512 bonds with 711,718 observations. Sector identifiers are merged into the main dataset for constructing sector-duration buckets.
The third dataset (the bootstrap_ZCB_ytm dataset) consists of zero-coupon Treasury yields bootstrapped by the Korea Financial Investment Association (KOFIA). These benchmark yields are used to compute credit spreads by subtracting the matched-maturity Treasury yield from each corporate bond’s yield-to-maturity.
Finally, Credit ratings are taken from the Korea Asset Pricing dataset, and bonds rated CCC, CC, or C are excluded.
After merging all sources by bond code and date, we remove observations with missing yields, benchmark Treasury yields, sector identifiers, or price information. This cleaned and unified dataset forms the basis for all empirical analyses.
4. Results
4.1 Analysis of Spread Behavior of Corporate Bonds
How should the risk associated with a particular market sector be measured? Typically, the historical volatility of a particular sector over some previous time period is used to forecast its volatility for the coming period. For this approach to be reliable, we must find that these volatilities are fairly stable. Unfortunately, this is not always possible.
Figure 1 shows the 36-month trailing volatility of spread changes for different credit ratings, such as AA+, AA-, and BBB+, in the Korean fixed income securities data from January 2010 to August 2020.
<Fig. 1>
Trailing 36-month volatility of monthly credit spread changes by credit rating (January 2010-August 2020).
Figure 1 shows that the spread volatility fluctuates over the sample period. It declines until the year 2016 and increases overall until 2020 while fluctuating to varying degrees depending on the credit rating.
Instead of dividing by ratings, we also partition the investment-grade corporate bond universe by spread levels. The results are reported in Figure 2. The spread level is the lowest for the 1st and the highest for the 6th.
<Fig. 2>
Trailing 36-month volatility of monthly credit spread changes by spread-level bucket (January 2010-August 2020).
Figure 2 shows that, compared to the patterns in Figure 1, spread volatilities are considerably more stable when corporate bonds are grouped by spread level rather than by credit rating. This suggests that spread level provides a more reliable basis for categorizing volatility in the Korean investment-grade market. The figure also reveals a positive relationship between spread level and volatility: buckets with higher spreads exhibit higher spread volatility, measured in basis points. It is important to note that this relationship reflects an absolute, not a relative, sensitivity. Even if the percentage (relative) volatility of spread changes were similar across buckets, bonds with higher spreads would mechanically produce larger absolute spread movements.
This observation motivates the use of relative spread volatility as a more stable descriptor of risk. One way to address the instability of absolute volatility is to approximate it by multiplying the historically observed relative spread volatility (expressed as a percentage change per month) by the current spread level (in basis points). If relative spread volatility is indeed more stable than absolute volatility, this approach would yield a more consistent volatility estimate across time and across spread buckets. The patterns documented in Figure 2 support this interpretation by showing that absolute spread volatility increases proportionally with spread levels, consistent with the proportionality assumption later tested in our analysis.
Figure 3 plots the comparison of the spread volatilities using either absolute or relative spread changes of the investment grade and high yield corporate bonds in the Korean corporate bond market. Relative spread changes are calculated as the ratio of spread change to the beginning-of-month spread level. The sample period is from January 2010 to August 2020.
<Fig. 3>
Time-series volatility of absolute and relative credit spread changes for investment-grade and high-yield corporate bonds (January 2010-August 2020).
The comparison in Figure 3 shows that while the spread volatility measured by relative spread changes is more stable, the difference is not significant. In fact, both spread volatilities are unstable over time.
Although relative spread volatility is more stable than absolute volatility, the absolute level of volatility still increases with spread. This motivates further analysis of both systematic and idiosyncratic components in Sections 4.2 and 4.3.
4.2 Systematic Spread Volatility
We examine the relationship between systematic spread volatility and spread levels. To do this, we first partition our dataset by spread level, measure the volatility of each spread level bucket, and examine the relationship between spread levels and spread volatility.
However, several challenges arise due to the nature of the dataset. First, it is heterogeneous, containing bonds from various industries, credit ratings, and maturities. Second, corporate bond spreads exhibit significant fluctuations over the sample period, resulting in considerable variation in the composition of each spread level group over time. We aim to create a partition that ensures bonds within each group share similar risk characteristics while maintaining a well-balanced distribution, such that each group contains a diverse set of corporate bonds.
The credit index is initially partitioned by sector into three rather broad categories: Financials, Industrials, and Utilities. The sector classifications—Financials, Industrials, and Utilities—follow the issuer-sector categories provided by FnGuide, which are widely used in Korean fixed-income research and reflect firms’primary business activities. Using these predefined classifications ensures consistency and transparency in segmenting the corporate bond universe.
Each sector is then further divided by duration (i.e., short, medium, and long). To ensure that each sector-duration bucket is adequately populated each month, we divide each sector into three equally populated duration groups. The purpose of dividing each sector into equally populated duration groups is to create homogeneous buckets for estimating the relationship between spread level and spread volatility. This grouping is performed only for the empirical analysis of volatility behavior; it is not required ex ante when forecasting volatility or constructing portfolios. In practical forecasting applications, investors can classify bonds using observable sector and duration information at the beginning of each period without needing to reproduce the exact historicalbucket sizes used in this study. Our grouping approach therefore serves to obtain reliable estimates of systematic and idiosyncratic volatility patterns, while remaining compatible with real-world forecasting practices where ex ante sector and duration data are readily available.
In the final step, bonds within each sector-duration bucket are assigned to one of several spread-level buckets. To facilitate a detailed partitioning of the entire spread range while minimizing instances of sparsely populated buckets, the spread breakpoints are set differently for each sector. Additionally, the Financials and Industrials sectors are divided into six spread buckets. The highest spread buckets across all sectors are excluded from the analysis due to the presence of extreme values.
The systematic spread change in bucket J in month t can be represented as the average spread change across all bonds in that bucket in month t. Therefore, for each bucket in the partition, we compute every month the median spread, the average spread change, and the cross-sectional standard deviation of spread change. This procedure produces 45 distinct time-series datasets; each consists of a fairly homogeneous set of bonds for which we have monthly spreads and spread changes. We then calculate the time series volatility of these systematic spread changes. Similarly, the spread level for bucket J is calculated as the time series average of the monthly median spread (rather than the average spread).
Figure 4 plots the relationship between systematic spread volatility and the corresponding spread level for each of the 45 sector-duration buckets.
<Fig. 4>
Time-series volatility of systematic spread changes versus spread level by sector-duration bucket (January 2010-August 2020).
The figure illustrates a pronounced positive association: buckets with higher median spreads consistently exhibit higher systematic spread volatility. This pattern holds across all three sectors and across short-, medium-, and long-duration groups. Differences between sectors—particularly between Industrials and the other categories—are modest, and variation attributable to duration appears limited. Overall, the figure indicates that systematic spread volatility scales closely with the level of spreads, with sector and duration playing only secondary roles.
4.3 Idiosyncratic Spread Volatility
To examine how idiosyncratic spread volatility varies with spread levels, we rely on the same sector-duration-spread partition used in the analysis of systematic spread volatility. Rather than focusing on the average spread change within each bucket, we evaluate the cross-sectional dispersion of spread changes among the bonds that comprise the bucket. For each bond i in market bucket J at time t, the idiosyncratic spread change is defined as the deviation of its individual spread change from the bucket’s average spread change in that month:
Equation (5) defines the idiosyncratic component of spread changes and is distinct from the volatility expressions in Equations (3)-(4). The volatility of idiosyncratic spread changes is equal to the cross-sectional standard deviation of total spread changes.
Figure 5 shows a scatterplot of the cross-sectional volatility for all months and spread buckets for investment-grade bonds.
<Fig. 5>
Cross-sectional volatility of idiosyncratic spread changes versus spread level for investment-grade bonds.
The figure shows the general pattern of volatilities increasing with spread, as well as the relative paucity of data at the higher-spread levels.
To derive a single measure of idiosyncratic spread volatility for each bucket, we aggregate all idiosyncratic spread changes for market bucket J across all investment-grade bonds and months, and compute their pooled standard deviation. This pooled measure of idiosyncratic spread volatility per market bucket is plotted in Figure 6 against the median spread of the bucket.
Although dispersed, Figure 6 reveals a positive relationship between spread levels and spread volatility. Buckets composed primarily of lower-grade bonds display greater dispersion than those dominated by higher-grade bonds; however, they still conform tothe same proportional pattern documented for systematic volatility.
5. A Novel Measure of Excess Return Volatility
What are the implications of spread proportionality? Which measure—duration times spread, or spread duration—is more appropriate for representing the risk of credit securities? In this section, we demonstrate that excess return volatility increases proportionately with DTS. Furthermore, portfolios with significantly different spreads and spread durations but with similar DTS exhibit the same excess return volatility. For example, a portfolio with a weighted spread of 200 bps and spread duration of two years is as risky as a portfolio with a spread of 100 bps and spread duration of four years. We also show that DTS generates better estimates of future excess return volatility than those calculated by spread duration.
5.1 DTS, Spread Duration, and Excess Returns
If the volatility of both systematic and idiosyncratic spread changes is proportional to the level of spread, the volatility of excess returns should be proportionately related to DTS, with the proportionality factor equal to the volatility of relative spread changes over the corresponding period.
To examine this prediction, each month bonds are assigned to quintiles according to their DTS value. Each quintile is further divided into six buckets based on spread. Every month the average excess returns and median DTS are calculated, and then the timeseries volatility of excess returns and average DTS are calculated separately for each bucket.
This formulation yields two empirical predictions. First, excess return volatility should increase proportionately with DTS, where the ratio of the two (or slope) represents the volatility of relative spread changes previously estimated. Second, the levelof excess return volatility should be approximately equal across spread buckets with a similar DTS value.
Figure 7 shows that excess return volatility increases with the level of DTS. In addition, consistent with the second prediction, observations representing the same DTS quintile but with differing spread levels exhibit very similar excess return volatilities, although the buckets with the highest spread for all quintiles are apart from the rest of the buckets. The one exception to this is in the highest DTS quintile, where the subdivision by spread causes wide variations in DTS as well. As a result, the points no longer form a tight cluster, although they do continue to exhibit the same general relation between DTS and volatility. Overall, the results presented in Figure 7 strongly support both predictions.
<Fig. 7>
Excess return volatility versus Duration Times Spread (DTS) for investment-grade corporate bonds (January 2010-August 2020).
To highlight the importance of this second finding, Table 1 presents the average spread and spread duration for all 30 buckets. The table shows how substantially spreads and spread durations can differ across buckets that nonetheless exhibit nearly identical DTS values.
<Table 1>
Average spread levels and spread durations across DTS and spread buckets.
Panel A reports average credit spreads, and Panel B reports average spread durations for each DTS-spread bucket. The table illustrates that portfolios with similar DTS values can exhibit substantially different combinations of spread levels and spread durations.
Panel A. Spread level versus DTS
Table 1 illustrates this point clearly. For example, in the second DTS quintile, the lowest and third spread buckets have nearly identical DTS values but markedly different underlying characteristics. The lowest-spread bucket contains bonds with an average spread duration of 6.40 and a spread of about 12 basis points, whereas the third bucket comprises bonds with a spread duration of 3.56 and a spread of 95 basis points. This comparison demonstrates that a portfolio composed of higher-spread, shorter-duration bonds can exhibit risk levels comparable to those of a portfolio composed of lower-spread, longer-duration bonds, provided that their DTS values are similar.
5.2 A Comparison of Excess Return Volatility Forecasts
We compare DTS-based forecasts directly against forecasts based on the traditional spread-duration framework, using the volatility of absolute spread changes as the benchmark. To evaluate the forecasting performance of DTS relative to traditional spread-duration-based measures, we compare two volatility estimators for excess returns:
Absolute spread-based estimator:
which relies on the historical volatility of absolute spread changes; and
DTS-based estimator:
which uses the historical volatility of relative spread changes.
These expressions follow directly from Equations (3) and (4), which link excess return volatility to absolute and relative spread changes, respectively. The key difference is whether volatility is scaled by spread duration alone or by DTS, which incorporates both duration and the level of the spread.
To directly compare their forecasting accuracy, we compute, for each month, the realized excess return of every DTS-spread bucket and construct normalized residuals under each volatility estimator. We then summarize the mean and standard deviation of these residuals across all 30 buckets. A volatility estimator is considered more accurate if its normalized residuals are closer to zero on average and exhibit a standard deviation closer to one, consistent with the interpretation of a well-calibrated volatility forecast.
Figure 8 presents these summary statistics.
<Fig. 8>
Mean and standard deviation of normalized excess return realizations under alternative volatility estimators.
In Figure 8, the DTS-based estimator produces residuals that are more tightly centered around zero and generally closer to unit variance than those generated using absolute spread-based volatility. In contrast, forecasts derived from absolute spread changes are more sensitive to the choice of estimation window; short histories tend to overreact to recent conditions, whereas long histories may understate changes in volatility. This instability complicates the selection of an optimal window for absolute spread volatility.
Figure 9 reinforces these findings by comparing the distribution of standardized excess return realizations under the two estimators.
<Fig 9>
Distribution of standardized excess return realizations under DTS-based and absolute spread-based volatility forecasts.
In Figure 9, the DTS-based forecasts yield a distribution with a sharper peak around zero and thinner tails, indicating fewer extreme forecast errors. In contrast, the distribution based on absolute spread changes exhibits a higher prevalence of outliers,reflecting noisier forecasts. Taken together, these results indicate that DTS produces volatility forecasts that more closely approximate the standardized normal benchmark implied by the modeling framework.
Based on these diagnostics, DTS demonstrates superior forecasting performance relative to the traditional spread-duration approach. This superiority is reflected in (i) lower dispersion of normalized residuals, (ii) greater stability of volatility estimates across time, and (iii) a standardized residual distribution that more closely matches the expected reference distribution. The statistical behavior of the normalized residuals provides a consistent basis for assessing and comparing forecasting accuracy.
6. Discussion
While our study does not directly test portfolio strategies, the empirical patterns we document imply several practical considerations for portfolio managers, particularly within the context of the Korean corporate bond market. Korea’s fixed-income marketis characterized by concentrated issuance among large conglomerates and financial institutions, relatively low secondary-market liquidity, and a ratings distribution skewed toward high-grade issuers. These structural features influence spread behavior inways that differ from major developed markets, underscoring the importance of examining whether tools such as DTS—originally developed and tested elsewhere—remain effective in this environment.
First, the results demonstrate that incorporating DTS into bond valuation frameworks can improve the assessment of credit risk. By establishing that both systematic and idiosyncratic spread volatilities are proportional to the level of spreads, our findings support the use of DTS as a measure that captures the relative sensitivity of bond prices to spread changes. This proportional structure enhances the accuracy of risk assessments, especially in markets such as Korea where spread levels and liquidity conditions vary substantially across issuers and sectors.
Second, DTS provides a more reliable basis for forecasting excess return volatility than traditional spread-duration-based measures. Because DTS leverages the stability of relative spread volatility, the resulting forecasts better align with realized excess return behavior. For practitioners, this implies that DTS-based volatility estimates may support more informed decisions regarding trade timing, hedging, and overall risk management, particularly during periods of heightened uncertainty.
Third, the evidence that spread volatility differs across sectors—yet consistently scales with spread level—highlights the importance of sector-aware risk evaluation. DTS provides a unified metric that enables comparisons of credit risk exposure across sectors with distinct liquidity conditions and spread dynamics. This can guide more effective sector allocation and diversification strategies, especially in a market where issuance concentration heightens sector-level risk.
Fourth, the observed link between high spreads and larger spread changes has meaningful implications for asset allocation. Investors may adjust their exposure to higher-spread securities based on their risk tolerance, as bonds with wider spreads are more prone to pronounced volatility. In the Korean bond market, where liquidity is uneven across issuers, DTS can help distinguish between sectors and securities that contribute disproportionately to portfolio-level spread risk.
Finally, the relationship between spread levels, volatility, and liquidity underscores a broader implication: DTS may serve as an indicator of liquidity-sensitive risk. In a market with relatively thin secondary trading activity, such as Korea, identifying securities that are more vulnerable to liquidity-driven volatility is particularly valuable for both portfolio managers and regulators. DTS provides a systematic way to flag these securities by linking volatility expectations to observable characteristics.
Overall, the empirical validation of DTS in the Korean market broadens the applicability of this measure beyond the markets in which it was originally developed. By showing that DTS reliably captures the behavior of spread volatility and excess return volatility in an underexplored and structurally distinct market, this study helps establish DTS as a robust framework for credit risk assessment and portfolio management in a wider range of fixed-income environments.
7. Conclusion
This study provides a comprehensive examination of spread behavior in the Korean corporate bond market and evaluates the effectiveness of DTS as a measure of credit risk sensitivity. Using an extensive 11-year dataset containing more than 3.8 million bond-month observations, we document a strong positive relationship between spread levels and both systematic and idiosyncratic spread volatilities. This proportionality is consistent with the core empirical assumption underlying the DTS framework, originallyarticulated by Dor et al. (2007) and later reaffirmed by de Jong (2023). By validating this relationship in an underexplored and structurally distinct market, our study extends the generalizability of DTS beyond the global markets in which it was first developed.
Across sector, duration, and spread partitions, we show that excess return volatility increases proportionally with DTS and that volatility forecasts based on relative spread changes are substantially more stable than those derived from absolute spread changes. These findings highlight the advantages of DTS as a risk-sensitive measure for credit portfolios, particularly in the Korean bond market, where concentrated issuance, limited secondary-market liquidity, and a rating distribution skewed toward high-grade issuers produce volatility patterns that differ from those in larger developed markets. In this setting, DTS offers a coherent framework for characterizing credit spread risk, supporting more accurate volatility forecasts, and facilitating consistent comparisons of risk exposures across sectors.
Relative to the original work by Dor et al., which demonstrated the usefulness of DTS primarily in U.S. and global investment-grade markets, our study provides an important extension by confirming the empirical validity of DTS in a market with different institutional characteristics and liquidity conditions. Whereas prior studies focus on demonstrating DTS behavior or its forecasting advantages globally, our analysis shows that the proportionality assumption and resulting volatility relationships also hold in the Korean corporate bond market, thereby broadening the environments in which DTS can be considered a reliable tool for portfolio management.
Several avenues for future research emerge from our findings. First, evaluating whether DTS improves risk-adjusted performance in active portfolio strategies through formal backtesting would provide a deeper understanding of its practical value. Second, examining the behavior of DTS across different market regimes—such as periods of severe liquidity stress or credit tightening—could help assess its robustness under extreme conditions. Third, applying DTS to individual traded instruments such as credit default swaps (CDS) or exchange-traded bond funds could offer insights into how spread dynamics propagate across related credit markets. Finally, extending the analysis to other Asian markets with similar structural features may shed light on whether the DTS framework can serve as a unifying measure of credit risk across diverse fixed-income environments.
Overall, this study confirms that DTS is an effective and interpretable measure of credit spread risk in the Korean corporate bond market and contributes to the growing body of literature demonstrating its utility for credit risk measurement, volatility forecasting, and portfolio management.
PDF Links
PubReader
ePub Link
Full text via DOI
Download Citation
