South Korea Taylor-Rule Policy Stance Gap

The Taylor rule is a benchmark monetary-policy reaction function that maps the natural real rate, current inflation, an inflation target, and the output gap into a prescribed nominal policy rate. The canonical form of Taylor (1993) is

with $\phi_\pi = \phi_y = 0.5$.

KRTAYLOR computes this rule and four further variants on Korean quarterly inputs, taking the natural real rate $r^*$ from the Holston, Laubach, and Williams (2023) natural-rate estimates, and KRTAYLORGAP reports the policy stance as the gap between the actual Bank of Korea base rate and the prescribed rate. Within the same group, the natural real rate $r^*$ itself is provided as KRNR, and the real-rate gap, the distance of the real policy rate from $r^*$, is provided as KRRRGAP.

A negative KRTAYLORGAP means actual policy is easier than the rule prescribes, while a positive value means it is tighter. KRRRGAP measures only a one-dimensional distance in real-rate space, whereas KRTAYLORGAP measures the gap from a rule that responds to the inflation gap and the output gap with explicit coefficients, so the two diverge when inflation deviates materially from target or when the output gap is large.

The Taylor (1993) rule, its four variants, and a constant-$r^*$ benchmark are computed on Korean quarterly inputs in the steps below, and the policy-stance gap between each rule and the actual base rate is produced.

(1) Inputs. The quarterly natural real rate $r^*_t$ comes from the HLW DB results and the quarterly COVID-adjusted output gap $y_t$ from the HLW model, per Holston, Laubach, and Williams (2023). Both are the model's one-sided Kalman-filtered states, so each quarter's value uses the data only up to that quarter, but the model's parameters and the filter's start value are estimated on the full sample and re-estimated at every update. They are therefore quasi-final inputs, filtered at parameters estimated on the full sample, not real-time ones (Orphanides and van Norden 2002), and the prescriptions begin well after the filter's first quarter, past the earliest quarters that lean most on that start value. The actual base rate $i_t$ is the BOK daily series resampled to quarter-end under the period-end convention, taking the rate in effect on the last business day of the quarter. Headline CPI year-on-year $\pi_t$ is computed from the monthly CPI level as a four-quarter percent change on the quarter-end series. One-year expected inflation $\pi^{1y}_t$ for the forecast-based rule is KREXPINF1, KRED's model estimate of one-year expected inflation, which combines the household survey with the one-sided CPI trend and a target-weighted long-run anchor and is therefore not survey data. The inflation target $\pi^*_t$ comes from the Bank of Korea schedule, with the 2016Q1 step from 3.0% to 2.0% preserved as a discontinuity per the spec lock.

(2) Taylor (1993). Per Taylor (1993),

Response coefficients $\phi_\pi = 0.5$ on the inflation gap and $\phi_y = 0.5$ on the output gap give a long-run inflation response of

, satisfying the Taylor principle.

(3) Balanced approach (Taylor 1999). Same form with $\phi_y = 1.0$,

which is the more appropriate specification in low-inflation environments where the central bank should respond more aggressively to slack.

(4) Inertial smoothing. The realistic single benchmark, with the Taylor (1993) target inside the recursion,

seeded at the actual base rate in the first usable quarter. Coibion and Gorodnichenko (2012) document central-bank interest-rate smoothing in OECD economies and confirm $\rho$ values in this range as the empirical norm.

(5) First-difference rule (Orphanides and Williams 2002). Uses no level of $r^*$ or the gap and is robust to the Orphanides (2001) real-time critique,

anchored at the actual base rate at the first usable quarter.

(6) Forecast-based rule (Clarida, Galí, and Gertler 1998). Replace $\pi_t$ by one-year expected inflation $\pi^{1y}_t$ in the Taylor (1993) form,

(7) Constant-$r^*$ benchmark. One additional Taylor (1993) line with $r^*$ replaced by the sample mean of HLW $r^*$,

The constant equals the empirical sample mean of HLW $r^*$ rather than the literature-style 2% used in Hofmann and Bogdanova (2012), because Korea's mean is materially different. The benchmark isolates how much of the prescribed-rate decline is due to the falling neutral rate.

(8) Rule-range band. Cross-specification span over the four anchor-free level rules,

The first-difference path is excluded from the band because it depends on the starting anchor, and is drawn as a standalone line. Taylor and Williams (2010) describe the broader robustness-rule framework in which this band sits.

(9) Stance gaps. For each rule $x$,

so a negative gap means policy is easier than the rule prescribes and a positive gap means it is tighter. The primary stance line is $\text{gap}_{93}$ (Taylor 1993 reference); the inertial-rule gap is the more realistic single read and is reported alongside.

The Taylor rule was introduced not as a binding prescription but as a description of how a well-functioning central bank's policy rate should respond to inflation and the output gap (Taylor 1993). Asso, Kahn, and Leeson (2010) document the rule's adoption in central-bank practice and the way it has been used as a starting point for policy deliberation rather than a strict commitment. The Bank of Korea's actual policy responses also weigh the exchange rate, financial stability, household debt, and global conditions that the rule does not capture, so the rule-range band and the cross-specification stance gaps presented here read as an analytical benchmark against which actual policy can be measured rather than a verdict on past policy.

The estimated long-run inflation response of the Bank of Korea reaction function falls below the strict Taylor principle threshold, a finding that reflects a gradual adjustment style rather than a defect. Clarida, Galí, and Gertler (1998) document Taylor-rule estimates for the G-3 plus several emerging economies and find substantial cross-country variation in the long-run inflation response, and Clarida, Galí, and Gertler (2000) extend this to a within-country regime classification and argue that satisfying the Taylor principle (a long-run inflation response above one) is the key feature that distinguishes proactive from reactive central banks. The Korean 2001–2026 sample produces an estimated long-run inflation response of approximately 0.83 in a simple OLS reaction function on the actual base rate, which Coibion and Gorodnichenko (2012) attribute to the heavy interest-rate smoothing that empirical central-bank specifications uniformly require.

A central concern in the Taylor-rule literature is that the inputs a central bank uses in real time differ materially from the ex-post-revised series researchers use. Orphanides (2001) shows that this gap can reverse the policy-stance reading on individual episodes. The HLW $r^*$ and the output gap are one-sided Kalman-filtered states, so each quarter's value uses the data only up to that quarter, but their parameters and start value are estimated on the full sample and the whole history is recomputed at every estimation. They are quasi-final rather than real-time inputs in the sense of Orphanides and van Norden (2002), filtered at parameters estimated on the full sample, so the prescribed rates computed here are benchmarks built from them rather than what the BOK could have computed at the time (Orphanides 2001, 2003). The first-difference rule of Orphanides and Williams (2002) is included for precisely this reason since it needs no level of $r^*$ or the gap and is robust to the real-time problem, providing a cross-check that does not depend on how those levels are revised.

The Korean sample produces several episodes worth reading carefully. The 2008–2009 global financial crisis brought a deeply negative output gap and falling inflation, but in the committed one-sided record inflation was still above target at the turn of 2009, so every rule prescribed a rate above the BOK's policy rate in 2008Q4 and 2009Q1, and only in 2009Q2 did the Taylor (1993) and balanced rules fall below the 2.0% policy rate. On this record the rule logic points to more accommodation only briefly, while exchange-rate and financial-stability considerations weighed on the BOK as the Korean won depreciated sharply. The 2020 COVID trough produced a negative output gap of 2.41 percent of potential at 2020Q2, against 3.66 at 2009Q1, with filtered standard errors of 1.23 and 1.18 percentage points, and every rule prescribed a rate below zero in 2020Q2, the Taylor (1993) prescription having been below zero in every quarter since 2019Q1, while the BOK reached zero only nominally against a 0.5% effective lower bound. In that episode the inertial rule reads as a more realistic guide because it incorporates the BOK's actual gradualist style. The 2021Q4–2022 inflation surge produced a sharp positive deviation, with the Taylor (1993) and balanced prescriptions reaching 9.45% and 9.87% in 2022Q2 against a BOK policy rate that peaked at 3.5%, indicating that BOK policy was substantially easier than a strict mechanical rule reading would prescribe.

Prolonged deviation from Taylor-rule prescriptions carries macroeconomic costs. Prolonged below-rule episodes, namely periods where policy is easier than the rule prescribes, are followed by higher inflation volatility and more financial-cycle activity, which provides the macroeconomic rationale for the policy-stance interpretation (Nikolsko-Rzhevskyy, Papell, and Prodan 2014). In the committed record the Korean Taylor (1993) stance gap is negative in every quarter from 2021Q1 to 2023Q1, while in every quarter of 2020 all the rules prescribed rates below the policy rate, so the below-rule period that overlaps the household-debt and housing-price boom documented in the Bank of Korea Financial Stability Reports begins in 2021 rather than 2020.

The Korean stance gap can be read alongside cross-country evidence to position Korea within the global pattern. Hofmann and Bogdanova (2012) document a cross-country Great Deviation where global policy rates ran systematically below Taylor-rule prescriptions in 2003–2007 and again in 2010–2014, and Kahn (2012) reviews the cross-country estimated reaction functions that complement the Hofmann-Bogdanova comparison. In the committed record the Korean Taylor (1993) stance gap is negative in every quarter from 2003Q1 to 2005Q4, inside the first of those windows.

The real-rate gap KRRRGAP measures only the distance of the real policy rate from $r^*$, while KRTAYLORGAP measures the gap from a rule that also responds to the inflation gap and the output gap with explicit coefficients. The two diverge when inflation deviates materially from target or when the output gap is large, with the Taylor stance gap moving by more in those episodes because it adds two response channels to the single real-rate channel that KRRRGAP captures.

A Taylor-type rule functions as one benchmark in optimal-policy analysis. Woodford (2001) embeds the Taylor rule in the optimal-policy literature and shows that a Taylor-type rule approximates an optimal commitment policy under standard New-Keynesian assumptions, and Svensson (2003) argues that targeting rules, where the central bank commits to forecast inflation and output, dominate instrument rules of the Taylor type in optimal-policy analysis. Taylor and Williams (2010) survey the robustness-rule literature in which both approaches sit. The Korean stance gap can be read as a Taylor-type benchmark with the understanding that the BOK's actual targeting framework is closer to a Svensson-style targeting rule than to a strict Taylor-instrument rule.

In the committed one-sided record the Korean natural real rate is 3.86% at 2001Q4 and 0.11% at 2026Q1, with filtered standard errors of 1.69 and 2.18 percentage points at those quarters, conditional on the estimated parameters. Holston, Laubach, and Williams (2017, 2023) document a cross-country decline in $r^*$ that is roughly synchronous across advanced economies, and Laubach and Williams (2003) provide the original methodology that the HLW papers extend. The constant-$r^*$ benchmark line in KRTAYLOR isolates how much of the prescribed-rate decline is due to this falling neutral rate.

KRTAYLOR provides a model-based benchmark against which the Bank of Korea base rate can be assessed at each quarter, and the inertial-rule gap is the most realistic single read of the stance because it captures the gradualism documented in Coibion and Gorodnichenko (2012). A positive gap (actual rate above prescribed) means that the rule prescribes a lower rate than the one in force, so policy is tighter than the rule prescribes. A negative gap means the opposite, namely that the rule prescribes a higher rate than the one in force and policy is easier than the rule prescribes. The rule-range band is the cross-specification uncertainty envelope and should be read as a quantification of how much disagreement exists across reasonable rule specifications.

The degree to which the Taylor principle is satisfied maps into how steeply the term structure responds to short-rate moves. Clarida, Galí, and Gertler (2000) document that countries whose central banks satisfy the strict Taylor principle (a long-run inflation response above one) tend to anchor inflation expectations more effectively, and Woodford (2001) embeds this finding in the optimal-policy literature. A more proactive central bank produces a flatter term-structure response, while a more reactive central bank produces a steeper one. The KRTP series on this site captures the term premium that compensates for this risk channel and can be read alongside KRTAYLORGAP.

For positioning in Korean rates (KTBs, IRS, CRS), KRTAYLORGAP shows in each quarter how far the base rate stands above or below the rule's prescription, which describes where policy stands relative to the rule rather than when the policy rate will move. For desks running outright duration positions, KRTAYLORGAP is a useful complement to the implied rate path series on this site, which captures market pricing of the next three MPC meetings.

Exit from an accommodative monetary-policy phase can be guided by Taylor-rule benchmarks. Rudebusch (2010) argues that the Federal Reserve's exit strategy from accommodative policy after the 2008 financial crisis should be informed by Taylor-rule benchmarks rather than by ad-hoc judgment alone, and similar arguments apply to the BOK's transitions out of accommodative phases. The 2021–2026 Korean cycle, where in the committed record the BOK held policy below the Taylor (1993) prescription in every quarter from 2021Q1 to 2023Q1 and above it in every quarter from 2023Q4 to 2026Q1, illustrates the trade-off Nikolsko-Rzhevskyy, Papell, and Prodan (2014) identify between rule-based and discretionary policy.

KRTAYLOR and KRTAYLORGAP are best read alongside KRNR (the HLW natural rate that drives the rules), KRRRGAP (the one-dimensional real-rate gap that the Taylor stance generalizes), KRGDPGAP (the output gap that enters the rules), KREXPINF1 (the expected-inflation input to the forecast-based rule), KRIMPR (the market-implied rate path that captures forward-looking BOK expectations), and KRTP (the term premium that compensates for the term-structure risk the policy stance creates). The combination provides a multi-dimensional read of monetary policy that no single series can deliver alone. The forward-looking read of where BOK policy moves next comes from the implied rate path, which records market pricing, while the Taylor stance gap shows where policy stands relative to the rule.

The forecast-based rule of Clarida, Galí, and Gertler (1998) is the most theoretically attractive Taylor variant because it incorporates inflation expectations rather than backward-looking inflation, and Woodford (2001) embeds it in the optimal-policy framework. For Korean applications, the forecast-based rule replaces realized headline inflation with KREXPINF1, the one-year point of the expected-inflation curve, which draws on the BOK Consumer Survey. Desks tracking the BOK rate cycle can read the forecast-based stance gap alongside the headline-inflation-based Taylor (1993) stance gap.

The constant-$r^*$ Taylor (1993) line in KRTAYLOR isolates how much of the prescribed-rate decline over the past two decades is due to the falling natural rate documented by Holston, Laubach, and Williams (2017, 2023). The gap between the constant-$r^*$ line and the time-varying-$r^*$ line at each quarter reads as the contribution of the natural-rate decline to the prescribed rate. In the committed record the time-varying line lies above the constant line at the start of the sample and below it at the end, with the natural rate at 3.86% at 2001Q4 and 0.11% at 2026Q1 (filtered standard errors of 1.69 and 2.18 percentage points), while Holston, Laubach, and Williams (2017) document a roughly synchronous natural-rate decline across advanced economies.

KRTAYLORGAP is the Taylor (1993) stance gap, the observed base rate minus the prescribed rate, and so a deterministic transform of two one-sided Kalman-filtered states of KRED's quarterly Holston-Laubach-Williams model, the natural rate and the COVID-adjusted output gap, combined with the observed policy rate, headline inflation and the inflation target, and never a function of smoothed states. Because two of its legs are filtered latent states, its measured serial correlation carries the filter's gain and not only the persistence of the data, so the statistical object is a persistence summary of the stance gap and not an integration order of the data.

The augmented Dickey and Fuller (1979) regression with the lag augmentation of Said and Dickey (1984), the nonparametric Phillips and Perron (1988) test and the KPSS stationarity test of Kwiatkowski et al. (1992) were deliberately not run on this series, because a verdict from them would read the random-walk dynamics the model imposes on its filtered inputs, whose signal-to-noise ratios were fixed by the median-unbiased procedure of Stock and Watson (1998) built on the exponential Wald statistic of Andrews and Ploberger (1994). The GLS-detrended test of Elliott, Rothenberg, and Stock (1996) was not run for the same reason, and neither was the modified-statistic battery of Ng and Perron (2001), since both confirm an integration reading this archetype does not take. The level break search of Bai and Perron (1998) was likewise not run, because a break in the mean of a transform of filtered states reflects the filter's response to its inputs rather than a change in the economy (Perron 1989). A level portmanteau such as Ljung and Box (1978) is not applied either, since on this series it reads the filter gain passed through the inputs rather than evidence about the data-generating process.

What the matrix mandates in their place is a descriptive persistence summary, the record of how the prescription's variance splits across its legs, and a statement about revision. Over the 98 quarters from 2001-12-31 to 2026-03-31 the first-order autoregressive coefficient of the stance gap is 0.861 and the implied half-life is 4.63 quarters, reported as a property of this series and not as an order of integration, and read with the small-sample downward bias of near-unit autoregressive estimates in mind (Andrews 1993). In the Taylor (1993) prescription the natural-rate, output-gap and observed-inflation legs carry variance shares of 0.208, 0.114 and 0.188 of the prescription's own variance, so the natural-rate leg alone outweighs the observed-inflation leg, the observed leg does not dominate, and the filter's persistence and revision pass into the stance gap through the two filtered legs. The revision cell records that the stance gap inherits the one-sided natural rate and output gap, whose parameters are re-estimated at every update by a seeded basin-hopping search, and that each load replaces the whole history, so the value at a fixed past quarter moves across updates and no dated vintage panel exists from which the size of that movement could be measured (Orphanides and van Norden 2002).

Three limits are stated rather than implied. No order of integration is assigned, by ruling and not by an inconclusive test (Hamilton 2018). The gap is measured against a benchmark built from quasi-final inputs, filtered at parameters estimated on the full sample in the sense of Orphanides and van Norden (2002), and it is not a real-time stance reading, which is the problem Orphanides (2001) raises for rules evaluated on revised inputs. And the frequency is quarterly, so the half-life is measured in quarters and is not comparable with a half-life measured at another frequency (Holston, Laubach, and Williams 2023; Laubach and Williams 2003).