South Korea Monetary Policy Target Surprise

A monetary policy surprise is the component of a Bank of Korea policy rate decision that markets had not priced in before the announcement, recovered by comparing the realized policy rate with the market-implied rate expected just before the announcement (Kuttner 2001).

The realized change in the policy rate decomposes into an expected change and an unanticipated change:

where $\Delta i_t^e$ is the market-implied expected change just before the announcement and $\Delta i_t^u$ is the unanticipated change that the market did not foresee. This platform reports two series for the Bank of Korea, namely the expected change going into each meeting and the target surprise, the unexpected change in the current policy rate that Gürkaynak, Sack, and Swanson (2005) call the target factor.

Gürkaynak, Sack, and Swanson (2005) identify two factors on the United States sample, the target factor and a separately identified path factor for revisions to the expected future path. On the Bank of Korea sample, where the rate is held at most meetings, the path dimension is not robustly identified and so the path factor is not published here, leaving the expected component and the target surprise as the two published series, with the test described under methodology.

A large positive target surprise indicates a decision more contractionary than the market expected, and a large negative value indicates a more accommodative one. Both series are realized event-study identifications keyed to monetary policy meetings, roughly eight per year since 2017, and describe how each announcement moved the curve rather than serving as forecasts or tradable signals.

(1) Kuttner decomposition. For each meeting the expected change is the market-implied expected policy rate just before the announcement minus the prevailing rate, taken from the platform's MPC implied-rate model. The raw surprise is the realized change minus the expected change, so the two sum to the realized change by construction (Kuttner 2001; Gürkaynak 2005).

(2) Target factor. Following Gürkaynak, Sack, and Swanson (2005), announcement-day changes in the risk-neutral forward rate at five short horizons, namely the next-meeting horizon and 3, 6, 9, and 12 months, are stacked into a panel and the first two principal components are extracted (Stock and Watson 2002). The target factor is the component rotated to align with the next-meeting instrument, normalized so a one-unit change is a one-percentage-point surprise in the current rate. The number of factors the data support is tested with the Cragg and Donald (1997) minimum-distance rank test, which belongs to the high-frequency identification tradition of Cochrane and Piazzesi (2002) and Rigobon and Sack (2004).

(3) The unpublished path factor. On this sample the rank test does not support a second factor, namely the path factor that Gürkaynak, Sack, and Swanson (2005) identify on United States data and that the forward-guidance literature studies (Campbell et al. 2012; Swanson 2021). This weak identification cannot be separated into its economic source, a sample in which most meetings hold the rate, and its measurement source, the high collinearity of the risk-neutral-curve instruments, so the path factor is set aside for the current instrument and sample rather than permanently discarded. The event study of the curve's reaction supports this disposition, where the target factor loads most strongly at the next-meeting horizon and declines along the curve with a falling long-horizon fit. This pattern mostly reflects the absent slope dimension, though it may also include long-horizon instrument noise (Nakamura and Steinsson 2018).

(4) Disclosed substitutions. Korea has neither federal-funds-style futures nor a deep overnight-index-swap market, so three substitutions are made and disclosed. Risk-neutral forward changes replace the futures panel, the MPC implied-rate model replaces Kuttner's settlement-based scaling, and a daily close-to-close window replaces the 30-minute intraday window of Gürkaynak, Sack, and Swanson (2005). The daily window is Kuttner's own choice (Kuttner 2001) and admits same-day non-monetary news that an intraday window would exclude (Nakamura and Steinsson 2018).

Monetary policy surprises are a workhorse for measuring how policy affects the real economy, because the unanticipated component is plausibly closer to exogenous than the policy move itself (Bernanke and Kuttner 2005; Romer and Romer 2004). Early event studies that regressed rates on the unsplit change in the target found only a weak relationship (Cook and Hahn 1989), which Kuttner (2001) resolved by showing that rates respond strongly to the unanticipated component and barely to the anticipated one.

Surprise tightenings raise credit costs and reduce output and prices. The high-frequency-instrumented evidence of Gertler and Karadi (2015) shows this, consistent with the contractionary responses long estimated in the vector-autoregression tradition of Christiano, Eichenbaum, and Evans (1999) and with the surprise-driven rate effects in Kuttner (2001).

The interpretation of the surprise is complicated by the information effect. Because an announcement moves private beliefs about the outlook as well as the stance, a tightening surprise can be accompanied by upward revisions to expected growth (Nakamura and Steinsson 2018; Jarociński and Karadi 2020). Bauer and Swanson (2023), however, show that controlling for common news removes much of this pattern, which is why the surprises here are presented as realized identifications rather than as structural shocks ready for causal inference without further assumptions.

For Korea specifically, An, Lee, and Kim (2026) apply high-frequency identification to Bank of Korea communication and find that the path of policy carries more information for the economy than the current-rate decision. Kim, Kim, and Kwak (2026) show that United States and Korean yields synchronize in ways that shape domestic transmission in a small open economy. The honest use of the series here is as surveillance of how each decision landed against expectations, where a large target surprise flags a decision the market had mispriced rather than serving as a forecast of the macroeconomic response.

The financial-market response to monetary policy surprises is large and well mapped. A surprise tightening raises bond yields and lowers equity prices (Kuttner 2001; Cook and Hahn 1989; Bernanke and Kuttner 2005), and Gürkaynak, Sack, and Swanson (2005) show that the short end responds mainly to the current-rate surprise while longer maturities respond mainly to the expected path of policy.

The effects of a surprise extend beyond the level of the policy rate to asset prices broadly. Hanson and Stein (2015) find that surprises move distant forward real rates through term premia. Gorodnichenko and Weber (2016) show that the equity response varies with price stickiness, and Gürkaynak, Karasoy-Can, and Lee (2022) trace a cash-flow channel through firm liability structure. Cieslak and Schrimpf (2019) decompose the announcement reaction into monetary and non-monetary news, with the non-monetary part dominating around communication events.

The target surprise here is a descriptive input for that kind of analysis rather than a tradable signal, because the daily window admits same-day news (Rigobon and Sack 2004) and because the published series omits the path dimension that drives much of the long-end reaction. For Korea it is best read alongside An, Lee, and Kim (2026) on the high-frequency reaction of Korean markets and Kim, Kim, and Kwak (2026) on the synchronization of United States and Korean yields, and paired with the platform's term-premium, cross-currency-basis, and financial-conditions indicators rather than used on its own.

KRMPTGT is the Gurkaynak, Sack, and Swanson (2005) target-factor surprise, an event-keyed monetary-policy surprise indexed by Monetary Policy Committee meeting number rather than by calendar time, spanning 263 decisions from 2001-09 to 2026-05. It is published as the surprise itself and normalized to load one on the next-meeting risk-neutral forward at a near-unit fit of 0.9995, and its identification as the target factor rests on the Cragg and Donald (1997) rank test. The published target surprise correlates at 0.62 with the Kuttner (2001) anticipated-change surprise reported alongside it.

Because the index is meeting number and not calendar time, the standard unit-root battery of the augmented Dickey and Fuller (1979) regression with the augmentation of Said and Dickey (1984), the Phillips and Perron (1988) test, and the KPSS test of Kwiatkowski et al. (1992) is not applied to this surprise, and a calendar Bai and Perron (1998) break search together with the calendar seasonal tests of the Hylleberg et al. (1990) family was deliberately not run, because those reference distributions are all derived under a fixed-interval time-ordered process that an event index violates (Gurkaynak, Sack, and Swanson 2005; Ghysels and Osborn 2001; Perron 1989).

On the surprise object the martingale-difference zero-mean t-test finds no systematic ex-ante bias, since under a Newey and West (1987) event-order heteroskedasticity-robust overlay the published target surprise has mean and t-statistic zero to machine precision at p = 1, so it does not reject a zero mean. The exact sign test agrees, with 141 of 263 target surprises positive at p = 0.267, so the median is not distinguishable from zero (Conover 1999).

The event-time Ljung and Box (1978) portmanteau in meeting lags does reject, however, at meeting lag one giving 16.63 and at meeting lag five giving 38.86, each at p = 0.000, so the surprise carries event-time serial dependence read as departure from full anticipation.

The sample is short at 263 meetings and hold-heavy, with a point mass at a zero surprise, which is why the exact distribution-free sign test is reported beside the t-test and why the event-time portmanteau is read only at the first and fifth meeting lags. These are realized event-study identifications of the policy surprise at each decision and not forecasts, and the target-factor reading is reported so that a disagreement between the mean and the median is visible rather than averaged (Conover 1999).