South Korea 3-Year Expected Inflation

The 3-year expected inflation is the average annual rate of price increase anticipated by market participants and economic agents over the next 3 years from the current date. It is the time-integrated average of the instantaneous (forward) expected inflation path over the 3-year segment, capturing both the near-term inflation outlook and the expected convergence to the long-run anchor.

In general, the $n$-year average expected inflation is defined as the integral average of the short-term component $\pi^{short}_t$ decaying and converging to the long-run anchor $\pi^{long}_t$:

where $\kappa$ is the decay parameter that governs the speed of convergence.

Because expected inflation is a latent variable not directly observable in market prices or surveys, it is estimated with a model that combines the UCSV trend, the household inflation survey, and the Nelson-Siegel term structure (Nelson and Siegel 1987).

A rise in the 3-year expected inflation reflects the expectation of markets and households that the average rate of price increase over the next 3 years will be higher, while a fall indicates disinflation expectations. The gap between the short-term component and the long-run anchor $\pi^{long}_t$ summarizes whether the market views the current inflation shock as transitory or persistent.

Estimated in three stages, comprising UCSV trend extraction, the survey-anchored short-term expectation, and the Nelson-Siegel term-structure projection.

(1) Trend inflation extraction. The UCSV model of Stock and Watson (2007) extracts trend inflation $\tau_t$ from headline CPI year-over-year. The state-space representation is

where $h_t$ and $g_t$ are the log-volatility processes for the observation error and trend innovation, respectively. The precision-based Gibbs sampler of Chan and Jeliazkov (2009) cycles through the following five blocks. These sample (i) $\tau_{1:T}$ via tridiagonal precision sampling in $O(T)$, (ii) $h_{1:T}$ via KSC mixture approximation, (iii) $g_{1:T}$ via KSC mixture approximation, (iv) the mixture indicators via multinomial, and (v) $\sigma^2_h, \sigma^2_g$ via inverse-gamma conjugate, respectively. The MCMC settings are 2,000 burn-in draws, 5,000 posterior draws, an IG prior $\nu_0 = 10$, $E[\sigma^2] = 0.02$, and a diffuse initial state variance $V_0 = 10.0$, and the sampler is initialized from a 12-month moving average of inflation, which damps monthly seasonal variation in its starting values.

The trend that enters expected inflation is a one-sided estimate, not a two-sided posterior path over the full sample. For each month the sampler is run afresh on the expanding window of observations up to that month, and the trend value is the posterior mean of the window's final month, so no later observation enters it. The series begins only once the window reaches its minimum length of 60 monthly observations, so it starts in December 1995, and the months before that are not estimated. Every window uses the same random seed, so consecutive months share common random numbers and their month-to-month moves are not independent Monte Carlo noise. The Nelson-Siegel decay, the long-run anchor weights and the volatility prior are fixed calibrations chosen on the full sample, the target leg of the long-run anchor takes a later-dated value in the months from December 1995 through December 1999, as described below, and the input is the latest vintage of the price index, so the series is a quasi-real-time estimate, not a real-time one (Orphanides and van Norden 2002). It is a model estimate, neither survey data nor a central bank's estimate.

(2) Survey-anchored short-term expectation. The 1-year expected inflation $\pi^{short}_t$ combines the UCSV trend with the Bank of Korea Consumer Survey of Inflation Expectations (1-year ahead, monthly, from January 2002). Expanding-window OLS estimates $(\alpha, \beta)$ from the regression

where only observations with confirmed 12-month-ahead realized inflation at time $t$ are used to prevent look-ahead bias. The resulting forecast is

where $\beta \in [0,1]$ and the intercept $\alpha$ absorbs the systematic upward bias of the household survey (Ang, Bekaert, and Wei 2007). A minimum training sample of 84 months (7 years) is required, and below this threshold or when the survey is unavailable, $\pi^{short}_t = \tau_t$. The long-run anchor $\pi^{long}_t$ reflects strong evidence that the trend converges to the target after the adoption of explicit inflation targeting (Garnier, Mertens, and Nelson 2015) and the limited long-run information content of the 1-year household survey (Chan, Clark, and Koop 2018). When the survey is available, the weights are 60% inflation target, 30% UCSV trend, and 10% survey, and when it is unavailable, they are 70% inflation target and 30% UCSV trend. The BOK inflation target history is 2.5% (2000–2003), 3.0% (2004–2015), and 2.0% (2016–present). The model's target schedule starts in January 2000, so for the months from December 1995 through December 1999 the target leg of the long-run anchor takes the schedule's latest value, the current 2.0% target, rather than a target in force at the time. In those months the anchor therefore rests on a later-dated value, and so does every maturity of the curve through the anchor's weight, which rises with maturity.

(3) Nelson-Siegel term structure. Instantaneous (forward) expected inflation follows

and the $n$-year average expected inflation is the integral average

The decay parameter is fixed at $\kappa = 1.0$. At the 3-year maturity, $\pi^{short}$ carries a 32% weight, capturing variation during inflation shock periods.

Expected inflation is a central variable spanning modern macroeconomics and finance. It links nominal and real interest rates in the Fisher (1930) equation, serves as the primary forward-looking determinant of current inflation in the New Keynesian Phillips Curve (Galí and Gertler 1999), and, in asset pricing, governs the real return on nominal assets and the breakeven inflation rate embedded in the term structure.

Expected inflation is a latent variable not directly observable in market prices or survey responses, so each measurement approach carries its own bias. Survey-based measures such as the BOK Consumer Survey, the University of Michigan Survey, and the Survey of Professional Forecasters capture stated expectations but may differ from the expectations actually embedded in economic decisions, while market-based breakeven measures such as TIPS spreads confound pure inflation expectations with inflation risk premiums and liquidity premiums. The model-based approach used here combines the UCSV trend, the household survey, and the Nelson-Siegel term structure to draw on the strengths of each source, in line with the methodology of the Federal Reserve Bank of Cleveland, which combines financial market data, surveys, and time-series models to construct an expected inflation term structure (Haubrich, Pennacchi, and Ritchken 2012).

Expected inflation is a linchpin of monetary policy analysis. In the Taylor (1993) rule, the optimal policy rate is a function of the inflation gap ($\pi - \pi^*$), and the forward-looking version of the rule computes the gap using expected inflation rather than current inflation. This distinction matters because monetary policy operates with long and variable lags (Friedman 1961), so a central bank that responds only to current inflation is systematically behind the curve.

The anchoring of inflation expectations, namely the degree to which long-term expectations remain stable in the face of short-term inflation shocks, is a key measure of central bank credibility. Bernanke (2007) argues that well-anchored expectations are the single most important asset a central bank possesses. When agents expect inflation to return to target, their pricing and wage-setting behavior acts as a self-stabilizing mechanism that helps bring about that return, whereas when expectations become unanchored, inflation shocks can become self-reinforcing through wage-price spirals.

The term structure of expected inflation, namely how expectations vary across the 1-year, 3-year, 5-year, and 10-year horizons, provides richer information than any single-horizon measure. A steep term structure, in which short-term expectations lie substantially above long-term, suggests that the market views the current inflation shock as transitory and expects a return to the long-run anchor. A flat or inverted term structure, in which long-term expectations rise to meet elevated short-term expectations, signals a more persistent inflation regime shift, a pattern observed in many countries during 2022–2023.

For Korea specifically, comparing model-based expected inflation with the BOK inflation target (currently 2%) reveals the degree of target credibility. A persistent deviation of the 10-year expected inflation from 2% would suggest that economic agents do not fully believe the central bank will achieve its target, which has implications for the effective conduct of monetary policy (Gürkaynak, Levin, and Swanson 2010).

In open economies, cross-country expected inflation differentials drive expected real exchange rate movements through the relative purchasing power parity (PPP) channel. The Korea-U.S. expected inflation differential informs expectations about the long-run KRW/USD exchange rate trajectory and is used in trade competitiveness analysis and FX risk management.

In bond markets, expected inflation is a key input for bond valuation, where the nominal yield decomposes into the real yield, expected inflation, and an inflation risk premium:

Changes in expected inflation directly affect nominal bond prices, and an unexpected rise in expected inflation reduces the real value of fixed nominal coupons and inflicts capital losses on bondholders. Ang, Bekaert, and Wei (2008) show that inflation risk, driven by both expected inflation dynamics and inflation uncertainty, is a priced factor in the cross-section of bond returns.

For countries with inflation-linked bond markets (e.g., U.S. TIPS, UK Linkers), the breakeven inflation rate, defined as the spread between nominal and inflation-linked yields of the same maturity, provides a market-based proxy for expected inflation. However, breakeven rates confound pure expectations with inflation risk premiums and liquidity differentials. In Korea, where no inflation-linked sovereign bond exists, model-based expected inflation estimates like those in KRED fill this gap by providing a clean measure of inflation expectations without the distortions present in breakeven rates.

For equity investors, expected inflation affects valuations through two channels. In the discount rate channel, higher expected inflation raises nominal discount rates, while in the cash flow channel, inflation affects revenues, costs, and profit margins differentially across sectors. Stocks of companies with strong pricing power tend to outperform during periods of rising expected inflation, while those with fixed-price contracts or high input cost sensitivity underperform (Weber 2015).

In portfolio construction, the expected inflation term structure informs the optimal mix of nominal bonds, inflation-protected securities, commodities, and real assets. Multi-asset frameworks like the one proposed by Ilmanen (2011) explicitly condition asset allocation on the inflation regime (distinguishing between rising/falling and high/low inflation environments), with the expected inflation term structure serving as the regime identification variable.

For pension funds and life insurers with long-duration real liabilities, namely benefit payments indexed to wages or prices, the expected inflation term structure directly affects liability valuation and hedge ratios. An unexpected increase in long-term expected inflation raises the present value of real liabilities, necessitating portfolio rebalancing toward inflation-hedging assets.

KREXPINF3 is the three-year point of KRED's expected-inflation term structure, a Nelson-Siegel average of a short-run expectation and a long-run anchor that are both built on the one-sided trend of the unobserved-components stochastic-volatility model (Stock and Watson 2007), and never on a full-sample posterior path of that trend. The short-run leg combines the household survey with the trend through an expanding-window regression that uses only outcomes already known at each date, and the long-run leg puts fixed weights on the inflation target, the trend and the survey, so the curve inherits the trend's filter-dominated persistence (Cogley 2002) and the statistical object is a persistence summary of the one-sided curve 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 path, because a verdict from them would test the random-walk trend the curve inherits rather than a property of the data, and the random-walk-versus-constant character of such a trend is not pinned down by the likelihood (Stock and Watson 1998). 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 an estimated trend reflects the estimator'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 path it reads the estimator's weighting rather than evidence about the data-generating process.

What the matrix mandates in their place is a descriptive persistence summary and a statement about revision. Over the 366 months from 1995-12-01 to 2026-05-01 the first-order autoregressive coefficient is 0.968 and the implied half-life is 21.50 months, 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). The revision cell records that the curve inherits the one-sided trend and the expanding-window survey regression, that a stored trend month is recomputed only when the inflation data up to that month are revised or rebased or when the sampler itself changes, and that each load replaces the whole history, so a fixed past month can move across updates although no smoothing enters it, 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 curve is one-sided in the data but it is not a real-time estimate, since the Nelson-Siegel decay, the long-run anchor weights and the volatility prior are fixed calibrations chosen on the full sample and the input is the latest vintage of the price index, which makes it quasi-real-time evidence in the sense of Orphanides and van Norden (2002), and it is a model estimate, neither survey data nor a central bank's estimate. And every trend window uses the same random seed, so consecutive months share common random numbers, their month-to-month moves are not independent Monte Carlo noise, and the extent to which shared simulation error enters the persistence measured above is not quantified here, while the half-life is measured in months.