South Korea Market-Implied Expected Policy Rate, Third Meeting

The market-implied expected policy rate is the path of the Bank of Korea base rate that bond market pricing implies for the next one to three Monetary Policy Committee meetings. It is read directly off the forward curve rather than from a survey or a time-series model.

The measure is term-premium-removed. Decomposing the forward rate fitted by an affine term structure model into a risk-neutral component and a term premium gives:

where $f^Q_t(\tau)$ is the risk-neutral forward rate obtained by setting the market price of risk to zero and $\text{TP}_t(\tau)$ is the term premium. The reported path discards the latter and integrates only the former over each meeting interval, the expectations-hypothesis component, which equals the expected average future short rate under the physical measure (Adrian, Crump, and Moench 2013). The only further adjustment is a convexity correction from compounding, which is not a risk premium.

A rise in the expected policy rate means the market is pricing in future rate hikes, while a fall signals expected cuts. This value is descriptive market pricing rather than the platform forecast, and each meeting is expressed as a full distribution over the 25 basis point decision grid. The next-meeting histogram is the most watched yet the least sharp of the three horizons, a property the backtest makes explicit.

(1) Term-premium removal. The daily zero-coupon curve is fitted by the Adrian, Crump, and Moench (2013) affine term structure model, and the risk-neutral forward curve is recomputed with the estimated market prices of risk set to zero, which yields the expected-policy-path forward absent any term premium.

(2) Inter-meeting implied rate. For each Monetary Policy Committee interval the instantaneous forward is integrated and exponentiated to a simple rate, which applies the Jensen convexity correction exactly rather than averaging the forward arithmetically (Piazzesi and Swanson 2008).

(3) Probability tree. Following the CME FedWatch construction, the expected change between consecutive meetings is split across the two bracketing 25 basis point levels by linear interpolation, the characteristic-and-mantissa step, and the per-meeting distributions are chained by Markov convolution so cumulative probabilities are sums over all paths reaching each level (Kuttner 2001; Krueger and Kuttner 1996).

(4) Backtest. Skill is measured against naive benchmarks because the Korean sample is roughly four-fifths holds, so accuracy alone is uninformative. Point forecasts are compared by root mean squared error and the Diebold and Mariano (1995) test with the Harvey, Leybourne, and Newbold (1997) small-sample correction against an always-hold random walk and an autoregressive model, the benchmark of Gürkaynak, Sack, and Swanson (2007). Probability forecasts are scored by the Brier (1950) score decomposed by Murphy (1973) into reliability, resolution, and uncertainty, with resolution the decisive term, and by the skill score against the unconditional base rate, the calibration-and-sharpness paradigm of Gneiting, Balabdaoui, and Raftery (2007).

(5) Disclosed substitution. Korea has no fed-funds-style futures, so the risk-neutral forward replaces the futures strip, a daily close-to-close window replaces intraday quotes, and the term-premium decomposition is model-based rather than read from a market quote (Bauer and Swanson 2023).

Expected-policy pricing matters for the economy because output and inflation respond to the expected path of policy and to its surprises, not to the level of the current rate alone (Woodford 2003; Bernanke and Kuttner 2005). The expectations channel is central to modern transmission, where guidance about the future path moves activity through long rates and expectations (Eusepi and Preston 2010; Campbell et al. 2012). Surprise tightening, the gap between the realized decision and this expectation, raises borrowing costs and lowers output and prices in the high-frequency identification of Gertler and Karadi (2015), consistent with the surprise-based rate effects of Kuttner (2001) and the narrative account of Romer and Romer (2004).

Recovering the expected path from market prices is an established tradition. Money-market and futures prices carry information about policy probabilities (Poole and Rasche 2000; Söderström 2001), and futures-based measures of expected policy are efficient in the sense that other variables add little to their forecast (Krueger and Kuttner 1996). Because raw futures and forwards are risk-adjusted forecasts that embed a small time-varying premium (Piazzesi and Swanson 2008), removing it yields the cleaner expectation that Gürkaynak (2005) and Bernanke and Kuttner (2005) treat as the object of interest.

The expectation is not a structural forecast. Announcements move private beliefs about the outlook as well as the stance, the information effect of Nakamura and Steinsson (2018), and Gürkaynak, Sack, and Swanson (2005) separate the immediate-rate surprise from revisions to the expected path, the path component that later work uses to measure the effects of forward guidance (Swanson 2021). Communication shapes how far ahead the market can read policy (Blinder et al. 2008), and Hanson and Stein (2015) show the expected path transmits even to distant real forward rates.

For Korea the honest use of the series is as a gauge of where the market sees policy heading and of how each decision lands against that expectation. The backtest shows the expectation carries genuine information at every horizon and beats naive time-series benchmarks at the longer horizons, and that its skill has improved as the Bank of Korea has become more predictable, the lengthening of anticipation documented by Lange, Sack, and Whitesell (2003). It is descriptive market pricing rather than a recommendation.

Rates and bond traders price the expected policy path continuously, and the market reaction to surprises relative to it is large and well documented. Surprise tightening raises bond yields and lowers equity prices (Kuttner 2001; Cook and Hahn 1989; Bernanke and Kuttner 2005), and the short end responds mainly to the current-rate surprise while longer maturities respond to revisions in the expected path (Gürkaynak, Sack, and Swanson 2005).

The expected path matters to duration managers because it shapes the pricing of maturities beyond the next decision. Policy news moves distant real forward rates through the term premium (Hanson and Stein 2015), and futures-implied expectations forecast the policy rate better than time-series models, the result this platform tests for Korea (Gürkaynak, Sack, and Swanson 2007). Around scheduled meetings asset prices carry a premium for bearing announcement risk, so realized returns are elevated (Savor and Wilson 2013; Lucca and Moench 2015; Ai and Bansal 2018), and the cross-section of equity responses varies with price stickiness (Gorodnichenko and Weber 2016).

The distribution served here is a descriptive input for positioning into meetings and for reading how much of a move is already priced, not a trading signal. The daily window blends some same-day non-monetary news into the measure (Cieslak and Schrimpf 2019), and the next-meeting histogram is overconfident on magnitude, so the series is best read alongside the platform term premium, cross-currency basis, and financial-conditions gauges rather than on its own.

KRMPR3 is the market-implied expected base rate at the third Monetary Policy Committee meeting ahead, an event-keyed series evaluated at meeting resolutions rather than in calendar time, so its evaluation object is the meeting-resolution forecast error and not the expected-rate level, on which a zero-mean test would reject trivially. The calendar unit-root battery of the augmented Dickey and Fuller (1979) regression, the Phillips and Perron (1988) test, and the KPSS test of Kwiatkowski et al. (1992), together with a calendar Bai and Perron (1998) break search and the seasonal tests of the Hylleberg et al. (1990) family, is not applied to an event index whose reference distributions assume a fixed calendar interval (Kuttner 2001; Gurkaynak, Sack, and Swanson 2005; Ghysels and Osborn 2001).

On the forecast-error object the martingale-difference t-test finds no systematic bias, with a mean error of −0.011, a plain t of −0.63 at p = 0.53, and a Newey and West (1987) event-order robust t of −0.35 at p = 0.73, and the exact sign test agrees with 136 of 261 errors positive at p = 0.536. The event-time Ljung and Box (1978) portmanteau in meeting lags rejects, at 164.82 for the first lag and 278.67 through five lags, each at p = 0.000, so the errors are unbiased in mean yet strongly serially dependent in event time as the longer overlap induces.

The meeting-level probability tree is scored against the realized decision by the Brier (1950) score, whose Murphy (1973) partition isolates the resolution term,

with skill against the base rate measured by $\text{BSS} = 1 - BS/BS_{\text{ref}}$. This is the strongest horizon, with resolution significantly positive at $\text{RES}=0.16$ on all 261 decisions and $\text{RES}=0.26$ on the post-2012 window of 135. It beats the base rate even on the full sample at $\text{BSS}=+0.19$ with a bootstrap interval from $0.07$ to $0.29$ that excludes zero, and rises to $\text{BSS}=+0.39$ post-2012 with an interval from $0.25$ to $0.50$, so the third-meeting horizon carries the high reliability badge.

Because four-fifths of the 261 decisions were holds, accuracy alone is uninformative, so the resolution term and the skill against the base rate carry the reading (Mincer and Zarnowitz 1969). Skill rises from the full sample to the post-2012 and post-2017 windows while benchmark difficulty stays roughly constant, following the Bank of Korea's growing transparency, and the exercise is a descriptive event-study evaluation and not a forecast (Conover 1999).