South Korea 5-Year Zero-Coupon Bond Term Premium

The term premium is the excess compensation that investors demand for bearing the risks associated with holding a long-term bond rather than sequentially rolling over short-term instruments. It arises because long-term bonds expose holders to interest rate risk, inflation risk, and liquidity risk that are absent from a rolling short-term strategy.

Under the no-arbitrage condition, the observed zero-coupon yield $y_t(n)$ approximately equals the model-fitted yield, and this fitted yield decomposes exactly into the sum of the risk-neutral rate $y^Q_t(n)$ and the term premium $\text{TP}_t(n)$:

The risk-neutral rate is the average of expected short-term rates over the bond's maturity under the physical measure, while the term premium is the risk compensation that investors require beyond that expected path. Since both components are latent variables that cannot be directly observed in market prices, they are estimated using affine term structure models (ATSMs).

The sign and magnitude of the term premium carry distinct economic meaning. A positive term premium indicates that investors require compensation beyond expected rate paths. A negative term premium suggests that demand for duration from price-insensitive buyers (central banks, pension funds, insurers) can more than offset the natural risk compensation, and has been observed in some advanced economies during periods of large-scale central bank asset purchases.

Estimated using the five-step Affine Term Structure Model (ACM) of Adrian, Crump, and Moench (2013).

(1) Input Construction. Par yields from 5 key rates (MSB 91d, KTB 1Y/3Y/5Y/10Y) are bootstrapped into zero-coupon yields under the Actual/Actual (ICMA) day count convention, using simple interest for $\tau < 1$yr and semi-annual compounding for $\tau \geq 1$yr. PCHIP interpolation generates a monthly maturity grid $Y \in \mathbb{R}^{T \times 120}$ (1–120 months). Monthly data are extracted via month-end resampling.

(2) PCA. $K=5$ principal components are extracted from the demeaned yield matrix:

, where PC 1 = Level, PC 2 = Slope, PC 3 = Curvature, PC 4–5 = higher-order variation. Variance and sign normalization are applied.

(3) VAR(1). State dynamics follow a driftless VAR(1):

with $\mu = 0$ imposed post-estimation. The OLS estimate $\hat{\Phi}$ is then bias-corrected using the Bauer-Rudebusch (2012) iterated bootstrap procedure, which corrects for finite-sample downward bias in the persistence of yield curve factors and prevents systematic overestimation of term premiums.

(4) Risk Price Estimation. Excess holding-period returns are defined as:

These are regressed on

. After Jensen's inequality correction:

and orthogonal projection to remove VAR residual components, the market prices of risk $(\lambda_0, \Lambda_1)$ are extracted via cross-sectional regression.

(5) Affine Recursion. Starting from $A_1 = -\delta_0$, $B_1 = -\delta_1$ (short rate equation), coefficients are recursed for $n = 2, \ldots, 120$:

Yields are decomposed as:

under the physical measure, and $y^Q_t(n)$ is the corresponding yield under the risk-neutral measure ($\lambda = 0$). The term premium is:

(6) Pricing Errors. Over the 6307 daily observations from 2000-12-18 to 2026-06-09 the mean absolute difference between the model-fitted yield and the observed zero-coupon yield is 4.99 basis points at twelve months, 1.05 at thirty-six months, 2.11 at sixty months and 1.04 at one hundred and twenty months. Those figures are a property of the input panel the estimation runs on rather than of the bias correction, and the same measurement on the same panel returns the same values to two decimal places when that correction is switched off.

(7) Second Configuration. A second KRED configuration of this estimator, which omits the small-sample persistence correction, is published beside this family at the same four maturities. It is a different estimand rather than a second measurement of this one, and neither family validates, corroborates, or outranks the other in either direction (Adrian, Crump, and Moench 2013; Bauer, Rudebusch, and Wu 2012).

Fluctuations in the term premium reflect a confluence of macroeconomic forces, including interest rate uncertainty, inflation risk, fiscal deficits and sovereign debt supply, global risk appetite, and the portfolio balance effects of central bank asset holdings (Rudebusch, Sack, and Swanson 2007).

The recognition that the term premium is an economically significant variable rests on the empirical failure of the expectations hypothesis (EH). The EH holds that long-term yields should equal the average of expected future short-term rates, which implies a zero term premium (Fisher 1896; Lutz 1940). However, empirical violations of the EH are extensively documented and show that term premiums are time-varying, economically significant, and predictable (Fama and Bliss 1987; Campbell and Shiller 1991; Cochrane and Piazzesi 2005). Because the risk-neutral rate and the term premium are latent, they are identified by the Gaussian ATSM, which makes yields affine functions of the state vector and links the physical-measure and risk-neutral-measure dynamics through no-arbitrage restrictions (Vasicek 1977; Duffie and Kan 1996).

A rising term premium signals that investors perceive elevated uncertainty about the future path of interest rates, often associated with heightened macroeconomic volatility (Wright 2011). Kim and Wright (2005) show that much of the 'conundrum' observed in U.S. Treasury yields during 2004–2005, namely the failure of long rates to rise despite Fed tightening, could be attributed to a declining term premium rather than shifts in rate expectations.

The sign and magnitude of the term premium are explained by preferred habitat theory. When investor demand segmented by maturity is not fully arbitraged away, the supply and demand for duration govern risk compensation, and a positive term premium is consistent with investor risk aversion (Modigliani and Sutch 1966; Vayanos and Vila 2021). Conversely, when demand for duration from price-insensitive buyers exceeds the natural risk compensation, the term premium can fall to negative values (Greenwood and Vayanos 2014).

Central bank quantitative easing (QE) compresses the term premium by absorbing duration supply from private portfolios. Gagnon et al. (2011) estimate that the Federal Reserve's Large-Scale Asset Purchases (LSAPs) reduced the 10-year term premium by 30–100 basis points. Li and Wei (2013) formalize this through a preferred-habitat model where the central bank's demand for long-term bonds crowds out private duration holders, reducing the equilibrium risk compensation. Bernanke (2015) extends this argument to the global context, noting that QE in one jurisdiction can spill over to compress term premiums in other sovereign bond markets through portfolio rebalancing channels.

For emerging market economies like South Korea, the term premium also reflects external factors such as global risk-on/risk-off dynamics, U.S. Treasury term premium spillovers (Obstfeld 2015), and foreign investor positioning in the domestic government bond market. The Korean term premium is therefore jointly determined by domestic monetary policy expectations and global financial conditions.

From a policy evaluation perspective, the term premium decomposition is essential for central bank communication. When long-term yields rise, policymakers need to distinguish whether the increase reflects higher expected policy rates or a rising term premium. The former suggests the market anticipates tightening, while the latter suggests increased uncertainty or shifting risk appetite. Misattribution can lead to policy errors, as emphasized by Adrian, Crump, and Moench (2013) and Bauer and Rudebusch (2014).

Decomposing yields into risk-neutral rates and term premiums is fundamental for fixed income portfolio management, as it enables investors to distinguish whether movements in long-term rates are driven by changes in expected policy paths or shifts in risk compensation.

A high term premium relative to historical norms suggests attractive risk compensation for long-term bonds, potentially signaling a favorable entry point for duration extension. Conversely, a low or negative term premium may indicate insufficient compensation for bearing duration risk, warranting caution in long-dated positions. Ilmanen (2011) provides a comprehensive framework for incorporating term premium signals into systematic bond allocation strategies.

The term premium also serves as a key input for relative value analysis across the yield curve. Cochrane and Piazzesi (2005) demonstrate that a single tent-shaped linear combination of forward rates predicts excess bond returns with an $R^2$ of approximately 44%, and this predictability is closely linked to term premium dynamics. Strategies that exploit term premium mean-reversion (buying duration when the premium is high and selling when it is compressed) have historically delivered positive risk-adjusted returns (Ang and Piazzesi 2003).

In the context of Korean Treasury Bonds (KTBs), the term premium decomposition is particularly valuable for foreign investors who must jointly manage duration risk and currency risk. When the KTB term premium is elevated relative to the U.S. counterpart, estimated by Kim and Wright (2005) or the ACM model at the New York Fed, it may present cross-market value opportunities after accounting for FX hedging costs.

For liability-driven investors such as insurers and pension funds, the term premium provides a measure of the compensation received for matching long-duration liabilities. A structurally low term premium environment challenges the ability of these institutions to generate sufficient returns on their fixed-income portfolios, a concern highlighted by the Bank for International Settlements (BIS 2018) in the context of prolonged low-rate regimes.

Over 6307 observations from 2000-12-18 to 2026-06-09, the five-year term premium carries a unit root, with the Dickey and Fuller (1979) test not rejecting at p = 0.1182, the Phillips and Perron (1988) test concurring at p = 0.0901, and the Kwiatkowski et al. (1992) test rejecting stationarity, an agreed I(1). On the level the first-order autocorrelation is 0.998 with an implied half-life of about 381.8 observations, a descriptive persistence summary and not an integration claim, the Ljung and Box (1978) portmanteau on the level rejects white noise at lags 10 and 20, Q = 61730.60 and Q = 121092.00 at p = 0.000 and p = 0.000, and the Bai and Perron (1998) procedure, by the Bai and Perron (2003) algorithm, finds no break in the mean (Andrews 1993; Perron 1989).

This signal, the five-year term premium, has no daily seasonal period, so the seasonal-unit-root and seasonal-stationarity machinery of Hylleberg et al. (1990) and Canova and Hansen (1995) is deliberately not run (Beaulieu and Miron 1992; Ghysels and Osborn 2001).