---
ticker: "KRKTBNOISE"
title: "South Korea Treasury-Bond Market Noise Measure"
unit: "bp"
frequency: "Daily"
source: "Hu, Pan, and Wang (2013); Nelson and Siegel (1987); Diebold and Li (2006)"
release: "Updated Daily"
category: "Financial Indicators"
country: "KR"
language: "en"
canonical: "https://kred.dev/en/series/KRKTBNOISE"
license: "https://creativecommons.org/licenses/by-nc-nd/4.0/"
latest_value: 5.36
latest_date: "2026-10-23"
first_date: "2000-12-18"
observations_total: 6399
observations_shown: 120
---

# South Korea Treasury-Bond Market Noise Measure

## Overview

Gauges how far KTB yields stray from a smooth curve, tracking bond-market liquidity strain and arbitrage gaps.

## Key Figures

|  | Value | Date |
|---|---|---|
| Latest | 5.36 | 2026-10-23 |
| Change from previous | -0.05 | 2026-10-22 |
| Change over one year | +2.37 | 2025-10-23 |
| Highest on record | 18.78 | 2008-12-02 |
| Lowest on record | 0.03 | 2024-05-07 |
| Period covered | 2000-12-18 – 2026-10-23 |  |
| Observations | 6399 |  |

## Recent observations

| Date | Value | Change |
|---|---|---|
| 2026-04-27 | 0.91 | +0.18 |
| 2026-04-28 | 0.84 | -0.08 |
| 2026-04-29 | 0.78 | -0.06 |
| 2026-04-30 | 0.75 | -0.03 |
| 2026-05-04 | 0.76 | +0.01 |
| 2026-05-06 | 0.84 | +0.08 |
| 2026-05-07 | 0.91 | +0.07 |
| 2026-05-08 | 1.23 | +0.33 |
| 2026-05-11 | 1.38 | +0.15 |
| 2026-05-12 | 1.75 | +0.37 |
| 2026-05-13 | 2.13 | +0.38 |
| 2026-05-14 | 2.24 | +0.11 |
| 2026-05-15 | 2.13 | -0.11 |
| 2026-05-18 | 3.21 | +1.07 |
| 2026-05-19 | 3.40 | +0.19 |
| 2026-05-20 | 3.50 | +0.10 |
| 2026-05-21 | 3.19 | -0.31 |
| 2026-05-22 | 3.26 | +0.07 |
| 2026-05-26 | 3.71 | +0.45 |
| 2026-05-27 | 3.42 | -0.29 |
| 2026-05-28 | 3.21 | -0.22 |
| 2026-05-29 | 3.47 | +0.27 |
| 2026-06-01 | 3.79 | +0.32 |
| 2026-06-02 | 3.76 | -0.03 |
| 2026-06-04 | 3.48 | -0.29 |
| 2026-06-05 | 3.52 | +0.05 |
| 2026-06-08 | 3.40 | -0.13 |
| 2026-06-09 | 3.98 | +0.58 |
| 2026-06-10 | 1.50 | -2.48 |
| 2026-06-11 | 2.07 | +0.56 |
| 2026-06-12 | 2.00 | -0.06 |
| 2026-06-15 | 1.49 | -0.52 |
| 2026-06-16 | 2.30 | +0.81 |
| 2026-06-17 | 2.40 | +0.11 |
| 2026-06-18 | 1.95 | -0.45 |
| 2026-06-19 | 1.99 | +0.03 |
| 2026-06-22 | 2.04 | +0.05 |
| 2026-06-23 | 2.57 | +0.53 |
| 2026-06-24 | 2.55 | -0.02 |
| 2026-06-25 | 2.50 | -0.04 |
| 2026-06-26 | 2.60 | +0.10 |
| 2026-06-29 | 2.62 | +0.02 |
| 2026-06-30 | 2.56 | -0.06 |
| 2026-07-01 | 2.75 | +0.19 |
| 2026-07-02 | 3.44 | +0.68 |
| 2026-07-03 | 3.64 | +0.20 |
| 2026-07-06 | 3.41 | -0.23 |
| 2026-07-07 | 3.41 | 0.00 |
| 2026-07-08 | 3.94 | +0.53 |
| 2026-07-09 | 3.92 | -0.03 |
| 2026-07-10 | 3.95 | +0.03 |
| 2026-07-13 | 3.68 | -0.27 |
| 2026-07-14 | 2.82 | -0.86 |
| 2026-07-15 | 3.05 | +0.23 |
| 2026-07-16 | 3.10 | +0.05 |
| 2026-07-20 | 3.00 | -0.10 |
| 2026-07-21 | 3.42 | +0.42 |
| 2026-07-22 | 3.07 | -0.34 |
| 2026-07-23 | 3.17 | +0.09 |
| 2026-07-24 | 2.66 | -0.51 |
| 2026-07-27 | 3.20 | +0.55 |
| 2026-07-28 | 3.29 | +0.09 |
| 2026-07-29 | 3.07 | -0.22 |
| 2026-07-30 | 2.80 | -0.28 |
| 2026-07-31 | 3.02 | +0.23 |
| 2026-08-03 | 2.94 | -0.09 |
| 2026-08-04 | 2.96 | +0.02 |
| 2026-08-05 | 3.29 | +0.32 |
| 2026-08-06 | 3.13 | -0.16 |
| 2026-08-07 | 3.06 | -0.07 |
| 2026-08-10 | 3.16 | +0.10 |
| 2026-08-11 | 3.48 | +0.32 |
| 2026-08-12 | 3.96 | +0.48 |
| 2026-08-13 | 4.39 | +0.43 |
| 2026-08-14 | 4.35 | -0.04 |
| 2026-08-18 | 4.41 | +0.06 |
| 2026-08-19 | 4.10 | -0.31 |
| 2026-08-20 | 4.09 | 0.00 |
| 2026-08-21 | 3.93 | -0.17 |
| 2026-08-24 | 3.87 | -0.06 |
| 2026-08-25 | 4.19 | +0.32 |
| 2026-08-26 | 3.95 | -0.24 |
| 2026-08-27 | 4.13 | +0.18 |
| 2026-08-28 | 4.56 | +0.43 |
| 2026-08-31 | 4.29 | -0.27 |
| 2026-09-01 | 4.01 | -0.28 |
| 2026-09-02 | 3.71 | -0.30 |
| 2026-09-03 | 3.57 | -0.14 |
| 2026-09-04 | 4.08 | +0.51 |
| 2026-09-07 | 4.00 | -0.08 |
| 2026-09-08 | 3.88 | -0.12 |
| 2026-09-09 | 3.44 | -0.44 |
| 2026-09-10 | 0.30 | -3.14 |
| 2026-09-11 | 0.34 | +0.04 |
| 2026-09-14 | 0.42 | +0.08 |
| 2026-09-15 | 0.35 | -0.07 |
| 2026-09-16 | 0.72 | +0.37 |
| 2026-09-17 | 1.82 | +1.10 |
| 2026-09-18 | 1.93 | +0.11 |
| 2026-09-21 | 1.16 | -0.77 |
| 2026-09-22 | 1.32 | +0.17 |
| 2026-09-23 | 1.29 | -0.04 |
| 2026-09-28 | 1.01 | -0.28 |
| 2026-09-29 | 0.84 | -0.16 |
| 2026-09-30 | 2.11 | +1.27 |
| 2026-10-01 | 2.95 | +0.83 |
| 2026-10-02 | 3.83 | +0.88 |
| 2026-10-06 | 4.03 | +0.20 |
| 2026-10-07 | 3.50 | -0.53 |
| 2026-10-08 | 3.21 | -0.30 |
| 2026-10-12 | 3.26 | +0.06 |
| 2026-10-13 | 3.71 | +0.45 |
| 2026-10-14 | 3.80 | +0.09 |
| 2026-10-15 | 4.10 | +0.29 |
| 2026-10-16 | 4.21 | +0.12 |
| 2026-10-19 | 3.98 | -0.23 |
| 2026-10-20 | 4.75 | +0.77 |
| 2026-10-21 | 5.71 | +0.96 |
| 2026-10-22 | 5.40 | -0.31 |
| 2026-10-23 | 5.36 | -0.05 |

## Definition

The treasury-market noise measure is the root-mean-squared deviation, in basis points, of observed KTB benchmark yields from the yields implied by a smoothed zero-coupon yield curve. Following Hu, Pan, and Wang (2013), the construction fits each day a three-factor Nelson and Siegel (1987) zero-coupon curve to the yields of five benchmark tenors (0.5, 1, 3, 5, and 10 years) by OLS, using the Diebold and Li (2006) fixed decay parameter $\bar{\lambda} = 0.0609$ (monthly maturities), and aggregates the residuals of the observed yields about the fitted curve as the root-mean-squared error (RMSE) over the 1-to-10-year window.

$$\text{KRKTBNOISE}_t = \text{RMSE}_{m}\bigl(y_t(m) - \hat{y}_t(m)\bigr)$$

where $y_t(m)$ is the observed benchmark yield at maturity $m$ and $\hat{y}_t(m)$ is the yield implied by the fitted curve. The measure stays low when relative-value arbitrage capital is abundant and price discrepancies across bonds are corrected quickly, and rises when arbitrage capital is scarce or benchmark tenors fragment.

Because KRKTBNOISE takes already-smoothed benchmark yields rather than individual bond prices as its input, it measures cross-tenor curve-fitting tension rather than the security-level mispricing of the original measure. Its level is structurally lower than the U.S. measure, which uses individual securities, so a numerical comparison is meaningless. It remains valid as a relative, within-Korea, time-series signal that rises when benchmark tenors fragment owing to demand concentration in specific maturities, flight to safety, benchmark-bond squeezes, or funding stress. The model is structured so that, once issue-level KTB pricing becomes available, the yield panel alone can be replaced while the same Nelson-Siegel RMSE computation is retained unchanged.

## Methodology

**(1) Yield panel.** Daily benchmark KTB yields $y_t(m)$ in percent are loaded for maturities $m \in M_{\text{fit}} = \{6, 12, 36, 60, 120\}$ months. Following the homogeneous-issuer requirement of Hu, Pan, and Wang (2013), only KTBs are included and MSB and corporate paper are excluded.

**(2) Loading matrix.** With the Diebold and Li (2006) fixed decay $\bar{\lambda} = 0.0609$, the Nelson and Siegel (1987) zero-coupon yield at maturity $m$ months is

$$z(m;\beta_0,\beta_1,\beta_2) = \beta_0 + \beta_1\,\frac{1-e^{-\bar{\lambda}m}}{\bar{\lambda}m} + \beta_2\!\left[\frac{1-e^{-\bar{\lambda}m}}{\bar{\lambda}m}-e^{-\bar{\lambda}m}\right]$$

With $\bar{\lambda}$ fixed, this is linear in $(\beta_0,\beta_1,\beta_2)$, giving an $|M_{\text{fit}}| \times 3$ loading matrix $\Lambda$.

**(3) OLS estimation.** For each day $t$ with $n_{\text{fit}} \geq 4$ observed tenors, ordinary least squares yields the factor estimates

$$\hat{\beta}_t = (\Lambda_t^\top\Lambda_t)^{-1}\Lambda_t^\top y_t$$

Because the solution is closed-form, the estimation is deterministic and no numerical optimizer is involved. Here $\Lambda_t$ and $y_t$ are restricted to the observed tenors. This follows Gürkaynak, Sack, and Wright (2007) and Andreasen, Christensen, and Rudebusch (2019) in adopting a fixed-shape NS family with OLS as the workhorse estimator.

**(4) Noise RMSE.** The fitted yields

$$\hat{y}_t(m) = z(m;\hat{\beta}_t)$$

are computed. Following HPW's convention of restricting the noise statistic to the 1-year-and-longer window, the noise is evaluated on $M_{\text{noise}} = \{12, 36, 60, 120\}$ months as

$$\text{KRKTBNOISE}_t = \sqrt{\frac{1}{|M_{\text{noise}}|}\sum_{m\in M_{\text{noise}}}\bigl(y_t(m)-\hat{y}_t(m)\bigr)^2} \times 100\;[\text{bp}]$$

The factor of 100 converts from percent to basis points.

**(5) Low-confidence flag.** Days with $n_{\text{fit}} = 4$ (one residual degree of freedom) are flagged low-confidence. Days with $n_{\text{fit}} < 4$ are dropped.

**(6) Robustness diagnostics.** To confirm that the primary series is not an artifact of the fixed-$\lambda$ choice, validation diagnostics are computed but not persisted to the database. A Nelson-Siegel variant that estimates $\lambda$ daily by nonlinear least squares (NLS) and, on days with $n_{\text{fit}} = 5$, a six-parameter Svensson (1994) fit are additionally computed and written to the validation CSV.

## Applications in Economics

The core economic content of the noise measure is that it gauges the aggregate capital available for relative-value arbitrage in the sovereign bond market. This interpretation originates in the limits-of-arbitrage tradition, where well-functioning markets are kept smooth by relative-value arbitrageurs who correct price discrepancies across bonds quickly, while the noise measure rises when arbitrage capital is scarce (Shleifer and Vishny 1997; Gromb and Vayanos 2002).

That the noise measure captures funding-liquidity constraints has been confirmed empirically on U.S. data. Hu, Pan, and Wang (2013) fit a Svensson (1994) forward-rate model to the daily cross-section of U.S. Treasury securities and define noise as the RMSE of observed minus fitted yields, documenting that the measure is negatively correlated with broker-dealer leverage, positively correlated with CDS spreads, and spikes from a normal-times level of roughly 3 basis points to 14.5 basis points during funding crises. This is consistent with the proposition of Brunnermeier and Pedersen (2009) that market liquidity deteriorates when funding liquidity is constrained.

The slow movement of arbitrage capital provides the micro-foundation for why noise reflects persistent price discrepancies. Under the slow-capital mechanism analyzed by Duffie (2010) and Mitchell, Pedersen, and Pulvino (2007), once a large shock drains arbitrage capital, price discrepancies across bonds can persist for weeks before sufficient capital returns to exploit them.

In the Korean KTB market, noise captures the supply-demand imbalances that arise from preferred-habitat segmentation (Vayanos and Vila 2021; Greenwood and Vayanos 2014). Insurance companies and pension funds concentrate demand in specific maturity buckets, and arbitrageurs offset this only partially. Domestic banks' collateral demand for the 3-year KTB benchmark in RP transactions with the BOK, and life insurers' and pension funds' liability-duration-matching demand for the 10-year benchmark, form habitats prone to dislocation in stress episodes. KRKTBNOISE captures this dislocation as a rising RMSE.

The two largest KRKTBNOISE spikes in the Korean sample coincide with episodes of broad financial stress. The 2008 global financial crisis produced a maximum reading of approximately 18 basis points as KTB yields fragmented across tenors amid deleveraging and foreign outflows. The October–November 2022 Legoland/Heungkuk credit stress episode generated a substantial rise in noise as the CP-market disruption following the revocation of the Legoland debt guarantee spilled into the KTB market through collateral channels. These episodes are consistent with the intermediary-asset-pricing framework, in which the balance-sheet capacity of financial intermediaries is the common factor behind market-wide asset-price dislocations (He, Kelly, and Manela 2017; Adrian, Etula, and Muir 2014).

The flight-to-liquidity channel is also closely tied to Korean data (Longstaff 2004). When investors concentrate in the most liquid KTB benchmarks (typically the 3-year and 5-year), demand becomes skewed toward a subset of tenors, the yield curve departs from a smooth NS shape, and the noise measure rises. This is genuine cross-tenor segmentation, distinct from the parallel shift of the curve that the NS $\beta_0$ level factor absorbs.

The safe-asset convenience yield on government bonds generates demand that is inelastic with respect to supply, which amplifies noise (Krishnamurthy and Vissing-Jorgensen 2012). The same mechanism applies to Korean KTBs as the sovereign safe asset, so during the Legoland episode the flight to safety simultaneously raised KTB prices and widened credit spreads, which appeared in KRKTBNOISE as a distortion in the cross-section of KTB benchmark yields relative to the smooth NS curve.

The term structure of bond liquidity itself creates the baseline heterogeneity that amplifies noise (Goyenko, Subrahmanyam, and Ukhov 2011). Because shorter maturities are generally more liquid than longer ones, when the short-end benchmarks (6-month, 1-year) in a stressed Korean market sometimes trade more liquidly than the long end (10-year), yield-curve curvature arises that the fixed-$\lambda$ NS cannot absorb, and this shows up as elevated noise. Fontaine and Garcia (2012) identify a bond-specific liquidity premium in Canadian government bonds that is independent of the term premium, and although KRKTBNOISE captures only the cross-sectional dimension, it is correlated with the same aggregate liquidity cycle.

Applying a similar noise framework to corporate bonds reveals comovement with sovereign noise, which confirms the presence of a common funding-liquidity factor (Bao, Pan, and Wang 2011). In the Korean context, the BOK's open market operations through RP and monetary stabilization bonds directly affect the funding capacity of the arbitrage capital available for KTB cross-sectional smoothing, so KRKTBNOISE also carries information about the BOK's liquidity stance beyond the signal in policy-rate expectations.

The tendency for subsequent KTB expected excess returns to be higher when the noise measure is elevated is the bond-market analogue of liquidity-risk-premium theory (Amihud 2002; Pástor and Stambaugh 2003). Amihud (2002) showed that expected illiquidity predicts returns and Pástor and Stambaugh (2003) built a factor model on that foundation, and such premia are consistent with the liquidity-adjusted CAPM of Acharya and Pedersen (2005), in which assets whose liquidity dries up when market liquidity is low command a positive premium. Where the ACM term premium series (KRTP) on this site captures the forward-looking compensation for bearing KTB duration risk, KRKTBNOISE provides a real-time gauge of the current market-liquidity environment in which that risk premium is being earned.

## Applications in Financial Markets

The noise measure functions as a real-time stress thermometer for the sovereign bond market, complementing the term premium (KRTP), the expected compensation for bearing duration risk, and the financial conditions index (KRFCI), which aggregates five financial-stress variables. When KRKTBNOISE is elevated, the KTB market is less liquid, bid-ask spreads in benchmark bonds are wider, and the ability of relative-value desks to exploit cross-tenor pricing errors is impaired.

For a bond portfolio manager, a rising KRKTBNOISE has two direct implications. First, transaction costs for rebalancing a KTB portfolio increase as bid-ask spreads widen and price impact deepens. Second, the Nelson-Siegel curve that KRKTBNOISE fits is also the basis for pricing non-benchmark KTBs relative to benchmark KTBs, so a high noise level signals that this pricing relationship is less stable than usual and that hedges constructed from benchmark bonds may be less effective at offsetting non-benchmark exposures.

Intermediary capital is a leading global predictor of noise-type liquidity measures (He, Kelly, and Manela 2017). Primary dealers and bank arbitrage desks supply the smoothing capacity that keeps the yield curve clean, so when their equity capital is impaired they scale back arbitrage activity and noise rises. For Korean desks this manifests through the BOK repo market and the KRW/USD cross-currency swap basis (KRXCCB on this site), and when global dollar funding is tight, domestic banks reduce their KTB arbitrage positions to manage balance-sheet exposure while KRKTBNOISE rises in tandem with KRXCCB.

Fitting the Nelson-Siegel family to individual bond prices rather than to benchmark yields reduces estimation error and improves measurement of the term premium (Andreasen, Christensen, and Rudebusch 2019), and this is the improvement that upgrading KRKTBNOISE from benchmark to issue-level data would capture. For practical portfolio purposes, the current benchmark version already suffices to identify the episodes where curve smoothness breaks down and cross-tenor trades become difficult to execute.

In relative-value strategies, KRKTBNOISE provides a timing signal for entering cross-tenor butterfly trades. Because the Nelson-Siegel curvature factor absorbs the expected hump shape of the yield curve, a high KRKTBNOISE reading indicates that the curvature of the actual market curve has departed from the smooth model shape, creating potential entry opportunities for butterfly trades (long 3-year, short 1-year and 5-year) or slope trades (steepeners and flatteners). Gürkaynak, Sack, and Wright (2007) documented similar RMSE dynamics in U.S. Treasuries and showed that the residuals are correlated with on-the-run/off-the-run premia, which in Korean markets maps to the benchmark/non-benchmark spread.

From a risk-management perspective, KRKTBNOISE functions as a market-microstructure stress indicator that leads some traditional metrics. Because it is computed from actual benchmark quote data rather than from a model or survey, it responds to market conditions in real time. Just as Bao, Pan, and Wang (2011) showed that the corporate bond illiquidity measure leads sharp credit-spread widening by 2–3 weeks in stress episodes, KRKTBNOISE may similarly lead the KRFCI, and a sustained rise above its historical 75th percentile (approximately 4.4 basis points in the Korean sample) warrants reducing position sizes in KTB relative-value strategies and increased caution across any strategy that assumes the yield curve maintains its normal smooth shape.

Fontaine and Garcia (2012) estimated a systematic bond-specific liquidity premium reaching several basis points during funding stress in Canadian government bonds, and the same premium exists in KTBs. KRKTBNOISE provides a daily gauge of the current level of that premium, so it can be used to adjust duration positions by reducing long positions when liquidity stress is high and to assess the adequacy of the term premium as compensation for the full bundle of risks, including liquidity risk.

Aggregate liquidity predicts returns cross-sectionally through the liquidity risk premium on low-liquidity assets (Pástor and Stambaugh 2003). The bond-market analogue is that when KRKTBNOISE is at a cycle high, the forward-looking KTB term premium should also be elevated, because the market demands additional compensation for bearing duration risk and liquidity risk together. Reading the KRTP series on this site together with KRKTBNOISE makes it possible to distinguish whether a high term premium stems from rate uncertainty or from sovereign-bond illiquidity.

For duration and funding-cost management, the KRKTBNOISE level indicates whether the cost of intermediating KTB risk has risen. This is consistent with the finding of Brunnermeier and Pedersen (2009) that market and funding liquidity co-move and that intermediary constraints amplify stress propagation across asset classes. When preferred-habitat demand concentrates in a specific benchmark tenor, holdings in adjacent tenors carry a segmentation premium in excess of the curve risk-neutral rate, and a high KRKTBNOISE reading is the observable footprint of that premium (Vayanos and Vila 2021). The time-varying KRKTBNOISE maps directly into a time-varying liquidity risk premium on KTBs under the liquidity-adjusted CAPM, in which expected returns depend on both systematic market risk and systematic liquidity risk (Acharya and Pedersen 2005). The result that expected illiquidity predicts returns and that illiquidity is priced in the cross-section is confirmed in bond markets as well, which shows that the same logic applies to fixed-income instruments across the maturity spectrum (Amihud 2002; Goyenko, Subrahmanyam, and Ukhov 2011).

## Statistical Tests

Over 6323 observations from 2000-12-18 to 2026-07-01, the unit-root reading of the Treasury-market noise measure is ambiguous, since the Dickey and Fuller (1979) and Phillips and Perron (1988) tests reject a unit root at p = 0.0 and p = 0.0 while the Kwiatkowski et al. (1992) test also rejects stationarity at p < 0.01, the near-unit-root disagreement of a persistent bounded signal, so no clean order is assigned. On the level the first-order autocorrelation is 0.974 with an implied half-life of about 26.4 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 = 48983.00 and Q = 83000.60 at p = 0.000 and p = 0.000, and the Bai and Perron (1998) procedure, by the Bai and Perron (2003) algorithm, finds 5 breaks in the mean at 2004-12-20, 2008-10-14, 2012-07-20, 2016-12-07, 2022-08-18 (Andrews 1993; Perron 1989).

This signal, the Treasury-market noise measure, 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).

## Frequently Asked Questions

### What does the government bond noise measure quantify?

The root mean squared fitting error between observed government bond yields and a smooth zero-coupon curve fitted through them, expressed in basis points. A high reading means individual issues trade away from what the curve implies.

### Why does curve fitting error carry information about liquidity?

When arbitrage capital is plentiful, deviations from the curve are closed quickly. Persistent deviations signal that the capital and balance sheet capacity supporting relative-value trades has contracted, which makes this a proxy for liquidity in the government bond market.

### Is the government bond noise measure the same as a credit spread?

No. This concerns the capacity to arbitrage within the government curve itself, whereas a credit spread compensates for default risk. They often rise together in a crisis, but they measure different things.