South Korea Debt Service Ratio, Private Non-Financial Sector

The debt service ratio is the share of disposable income that a borrowing sector devotes to interest and amortization payments on its outstanding stock of debt. The construction follows Drehmann and Juselius (2012, 2014), who define a sectoral DSR through an annuity formula that combines sector debt, augmented gross disposable income, an effective interest rate on the outstanding debt stock, and an average remaining maturity.

The DSR for each sector $j$ is computed from the following annuity formula:

where $i_{j,t}$ is the quarterly effective interest rate, $s_{j,t}$ is the average remaining maturity in quarters, $D_{j,t}$ is the sector debt stock, and $Y_{j,t}^{\text{aug}}$ is augmented gross disposable income. The leading term is the per-quarter annuity payment factor on a constant-maturity instalment loan, and multiplying it by the debt-to-income ratio gives the service burden relative to income. Dynan, Johnson, and Pence (2003) laid the foundation for this annuity framework for the United States.

A rising DSR means that the cash-flow constraint on households and non-financial corporations is tightening. A growing debt stock is sustainable as long as income grows in step and rates remain low, but it becomes unsustainable when either condition fails. KRDSR is not an independent national-accounts product but a faithful reproduction of the BIS Korea DSR adapted to the Korean institutional context (Drehmann et al. 2015), so it is read as a within-Korea signal of the credit-cycle phase relative to its own history rather than as a comparison against other countries' DSR levels or an absolute threshold.

(1) Sector debt. For each quarter $t$ and sector $j$, $D_{j,t}$ is the end-of-quarter outstanding stock of debt. For households, $D_{\text{HH},t}$ is the household credit total from the Bank of Korea household credit statistics, the broad definition that sums household loans and consumer credit and follows the definition the BIS applies to Korea per Drehmann et al. (2015). For non-financial corporations, $D_{\text{NFC},t}$ is computed as the residual between total credit to the private non-financial sector $r_t \times Y_t^{\text{nom}}$ and household credit, since non-financial corporate debt at the same coverage is not published separately. Here $r_t$ is the BIS-published credit-to-GDP ratio (the Federal Reserve QKRPAM770A, percent) and $Y_t^{\text{nom}}$ is the annual nominal GDP for the year containing quarter $t$. The PNFS aggregate is defined as

(2) Sector income. For each sector $j$ the augmented gross disposable income $Y_{j,t}^{\text{aug}}$ at quarterly frequency is

for households, where $\text{int}_{j,t}$ is interest paid approximated as

. For non-financial corporations,

where dividends are approximated as zero in the absence of directly observed data. The annual household GDI is Chow-Lin interpolated to quarterly with real GDP as the distribution indicator following Chow and Lin (1971), and the post-2010 quarterly observed series is used directly. The annual non-financial corporate GDI is approximated as annual nominal GDP times the enterprise share of national disposable income, Chow-Lin interpolated to quarterly. FISIM is allocated as

, where setting it at 5 percent of nominal GDP follows the BIS Korea practice documented in Drehmann et al. (2015). A four-quarter moving average is then applied to $Y_{j,t}^{\text{aug}}$ per the BIS adjustment.

(3) Interest rate. The quarterly rate is

, which follows the simple-conversion convention of Drehmann and Juselius (2014). The annual rate is the quarterly average of monthly outstanding-balance weighted-average bank loan rates for households and non-financial corporations. A compound-conversion variant $(1 + i^{\text{annual}})^{1/4} - 1$ is available behind a flag and is reported in the validation file.

(4) Maturity. The average remaining maturity $s_j$ in quarters is fixed at $s_{\text{HH}} = 72$ (18 years) for households and $s_{\text{NFC}} = 52$ (13 years) for non-financial corporations, following the Drehmann et al. (2015) BIS specification. The PNFS aggregate uses a debt-weighted average

(5) DSR annuity formula. Per Drehmann and Juselius (2012) and the Dynan, Johnson, and Pence (2003) US precedent,

where the bracket is the per-quarter annuity payment factor on a constant-maturity instalment loan. For the PNFS aggregate the interest rate is the debt-weighted average

and the income is the sectoral sum

(6) Low-confidence flag. Quarters before 2010Q1 use the Chow-Lin temporal disaggregation of annual income and are flagged with low-confidence, since the empirical correlation with the BIS reference is materially higher on the post-2010 sample where quarterly household GDI is observed directly.

(7) Validation. The full sample is compared against the BIS Korea DSR per sector. The correlation of mean-removed levels and of four-quarter differences is reported. The Chow-Lin period is also compared against a Denton-proportional variant as a robustness check.

The DSR is the single most powerful one-to-two-year-ahead predictor of banking crises among standalone indicators, outperforming the credit-to-GDP gap at the short horizon while the gap dominates at the three-to-five-year horizon (Drehmann and Juselius 2014). High credit and a rising debt service burden are the two limbs of the financial cycle, so the combined signal reaches its strongest forecast value when the two indicators move together, which Aldasoro, Borio, and Drehmann (2018) identify within the canonical early-warning hierarchy as the highest-conviction macroprudential alert. The Korean episode of late 2008 through 2009 is a clean instance of this combined signal, with KRCGAP rising into the global-financial-crisis peak while the DSR rose alongside it. The 2020–2022 episode is a second instance, where the COVID-era surge in household and corporate borrowing first lifted the gap and then, as policy rates rose sharply through 2022, transmitted to a rising DSR through the rate component.

A high debt service burden transmits into the real economy, compressing output and consumption. Mian, Sufi, and Verner (2017) show in a panel of 30 countries that a rise in the household debt-to-GDP ratio is followed by a fall in output growth two-to-three years later, mediated by the contraction in household consumption that a high debt service burden forces when income disappoints. Eggertsson and Krugman (2012) formalize this cash-flow channel in a macroeconomic model of debt deleveraging and demand contraction. Lombardi, Mohanty, and Shim (2017) reach a closely related conclusion on the household DSR in the BIS panel, with the short-run negative correlation between the DSR and consumption growth appearing in essentially every country with sufficient data coverage. Hofmann and Peersman (2017) close the loop by formalizing the debt-service channel of monetary policy transmission, where rate increases feed through the existing stock of variable-rate debt into a contraction in household and corporate spending. For Korea this channel is unusually sharp given the very high floating-rate share of household debt that the Bank of Korea Financial Stability Report has documented across successive issues, and KRDSR captures the resulting compression directly.

The DSR adds a cash-flow dimension that the credit-to-GDP gap does not capture. Schularick and Taylor (2012) and Jordà, Schularick, and Taylor (2016) show on long-run cross-country data that credit booms are the most reliable leading indicator of financial crises, with the housing-finance dimension contributing disproportionately to crisis depth. Borio and Lowe (2002) and Drehmann et al. (2011) embed this finding in the macroprudential framework that the credit-to-GDP gap operationalizes. A rising debt stock is sustainable when income grows in step and rates remain low, but it becomes unsustainable when either condition fails. Zabai (2017) reviews the household-debt dimension of this dynamic across advanced and emerging economies and identifies Korea as one of the cases where the household-debt level is high in international perspective but the DSR has remained within a containable range thanks to Bank of Korea macroprudential policy.

In Korea the household DSR has sat at the center of the macroprudential debate since 2017. The Bank of Korea Financial Stability Reports have kept it at the center of that debate, and the household DSR is a key input to the borrower-level loan-to-value, debt-to-income, and debt-service-ratio limits that the Financial Services Commission progressively tightened from 2017 to 2022. The KRDSRH series here approximates the BIS Korea household DSR rather than the Bank of Korea regulatory measure, which uses borrower-level micro data with a different income definition, but the cyclical signal is consistent across the two series. Juselius and Drehmann (2015, 2020) provide the deepest econometric analysis of this dynamic, showing in panel evidence that a rising DSR adds information beyond the credit-to-GDP gap and forecasts subsequent declines in output and consumption at a one-to-two-year lead. The leverage dynamics imply that when the level of credit is high, even at moderate rates, future income is mechanically committed to debt service, and the leverage-and-burden-of-debt mechanism transmits to lower investment and consumption in subsequent years.

DSR levels are not suited to cross-country comparison because the input data are not standardized. As the BIS itself emphasizes, the FISIM treatment, the sector-account definitions, and the average maturities all differ. Dembiermont, Drehmann, and Muksakunratana (2013) document the analogous comparability challenge for the BIS credit statistics that underlie the gap and the DSR, and Drehmann and Yetman (2018) extend the same caution to the gap measures derived from those statistics. The KRDSR series is therefore best read as a within-Korea, over-time signal of where the household and corporate sectors sit in the financial cycle rather than as a numerical comparison against US or European DSR levels. The interpretive frame is the credit-cycle phase, not the absolute level.

The decomposition into KRDSRH and KRDSRC matters because the two sectors respond differently to rate shocks and to macroprudential policy. Korean households hold predominantly floating-rate mortgages and secured consumer credit, so the household DSR responds quickly to Bank of Korea rate changes, with the 2022–2023 episode a recent instance of this fast pass-through. Non-financial corporations hold a mix of bank loans, bond debt, and trade-receivable credit at varied maturities, so the corporate DSR responds more slowly and more heterogeneously to the same rate shocks, and the corporate DSR remains elevated even after the household DSR has begun to fall as Bank of Korea rate cuts pass through to outstanding balances. Drehmann, Juselius, and Korinek (2017) document this asymmetry in the international panel and identify the corporate-DSR lag as a key vulnerability indicator.

Fixed-income desks operating in Korean credit markets use the DSR as a canonical short-horizon macro risk gauge. Its level relative to its own history, rather than against an absolute threshold, identifies the credit-cycle phase. A rising DSR signals a tightening of the cash-flow constraint on households and non-financial corporations, which transmits to wider credit spreads and to slower nominal-income growth that drags on the policy rate cycle (Schularick and Taylor 2012). The 2008–2009 and 2020–2022 episodes in the Korean data are the cleanest instances of this transmission, with KRDSR rising into the policy-tightening peak while credit spreads widened and growth slowed.

The sectoral DSRs give the most direct read on where each credit cycle sits. The household DSR (KRDSRH) is the most direct read on the consumer-credit-cycle position, and Lombardi, Mohanty, and Shim (2017) show that the household DSR predicts negative consumption growth at a one-to-two-year lead across the BIS country panel. For investors with positions in Korean consumer-finance debt, credit-card receivables, and household mortgage-backed securities, the KRDSRH level is the leading indicator of the credit-loss cycle that Mian, Sufi, and Verner (2017) document in the cross-country data. The corporate DSR (KRDSRC) is the analogous read on the corporate-credit-cycle position, and Drehmann, Juselius, and Korinek (2017) show that the corporate DSR predicts subsequent declines in investment and widening corporate-bond spreads in the international panel. For Korean corporate-bond portfolios, the KRDSRC level is the leading indicator of the spread cycle that the Bank for International Settlements documents in its Annual Economic Reports.

The DSR is a canonical input to macroprudential policy timing, and Drehmann et al. (2011) formalized the framework that the Basel III countercyclical capital buffer (CCyB) operationalizes. For investors anticipating Bank of Korea macroprudential policy, the combined KRDSR-and-KRCGAP signal is the best leading indicator of CCyB activation and of borrower-level LTV/DTI/DSR limit changes. The 2017–2022 macroprudential cycle in Korea, where the Financial Services Commission progressively tightened LTV, DTI, and DSR borrower limits in response to a rising household DSR, illustrates this transmission. For positions sensitive to macroprudential policy (Korean financial-sector equity, the KRW exchange rate, Bank of Korea reserve assets), the combined KRDSR-and-KRCGAP signal is the leading indicator of when the next macroprudential tightening cycle will begin.

The DSR is a canonical input to the credit-loss provisioning models of banks operating in Korea, and the panel evidence of Drehmann and Juselius (2014) shows that DSR rises predict credit-loss rises at the one-to-two-year horizon. For investors with positions in Korean bank debt and equity, the KRDSR level is the leading indicator of the credit-loss cycle that drives bank earnings and capital adequacy. In the 2020–2022 episode in the Korean data, the rise in household and corporate DSRs alongside the policy rate cycle foreshadowed the credit-loss recognition cycle that Korean banks reported in 2023–2024. Hofmann and Peersman (2017) provide the econometric foundation for this transmission, showing across the international panel that monetary policy rate increases transmit to credit losses with a one-to-two-year lag through the debt-service channel.

The DSR is a canonical input to forecasting the Bank of Korea policy rate cycle, and Hofmann and Peersman (2017) show that the debt-service channel of monetary policy transmission is the dominant transmission mechanism in countries with high household debt. In Korea, where the household debt-to-GDP ratio is among the highest in the OECD and the floating-rate share of household debt is substantial, the debt-service channel is unusually sharp, and KRDSR directly measures how far this channel has progressed. For Korean rate positions (KTBs, IRS, CRS), the KRDSR level is the leading indicator of how far the Bank of Korea can keep tightening before the debt-service-channel feedback begins to dominate, and conversely how rapidly it will need to ease once the cycle turns. The 2022–2024 cycle illustrates this dynamic, with Bank of Korea rate cuts beginning in earnest only after KRDSR had clearly passed its peak.

KRDSR is most useful when read alongside KRCGAP, KRFCI, KRSAHM, and KRRECP. KRCGAP captures the credit-stock dimension following Borio and Lowe (2002), KRDSR captures the cash-flow dimension following Drehmann et al. (2015), KRFCI captures the broad financial-conditions dimension, KRSAHM captures the labor-market transmission, and KRRECP synthesizes the near-term recession probability. The combination provides a layered read on credit-cycle stress that no single series can deliver alone. Aldasoro, Borio, and Drehmann (2018) document the cross-country evidence that the combined gap-and-DSR signal is the highest-conviction macroprudential alert.

The cross-country early-warning evidence applies to Korea as it stands, so KRDSR serves as a tradable short-horizon early-warning signal in the Korean macroprudential debate. Schularick and Taylor (2012) document on long-run cross-country data that credit booms are the strongest historical predictor of subsequent financial crises, and Jordà, Schularick, and Taylor (2016) refine this finding to identify mortgage-credit-driven booms as the most dangerous variety. Mian, Sufi, and Verner (2017) provide the correlation between household debt and output growth that anchors the cash-flow channel KRDSR captures, and Juselius and Drehmann (2015) and Juselius and Drehmann (2020) provide the leverage-dynamics evidence that ties together the credit stock, the debt service burden, and subsequent output and consumption. These cross-country findings apply to Korea given the institutional similarities documented by the Bank for International Settlements in its Annual Economic Reports and the analysis of the household-debt dimension by Zabai (2017). The KRDSR series therefore enters the Korean macroprudential debate as the canonical short-horizon early-warning channel that the Drehmann et al. (2011) framework specifies and that Dembiermont, Drehmann, and Muksakunratana (2013) embed in the broader credit-statistics framework.

KRDSR is a slow-moving quarterly debt-service ratio whose long-run behaviour is an open empirical question, since Korea's aggregate debt-service burden trended upward with credit deepening rather than reverting to a fixed mean. Its early-warning interpretation rests on deviations from a slowly varying norm as a matter of economics (Drehmann and Juselius 2012), not as a stationarity result, so the unit-root protocol is run on the level under a constant-plus-trend specification with the verdict reported honestly rather than forced. The ratio is tested over 94 quarterly observations from 2002-12-31 to 2026-03-31.

For this aggregate ratio the augmented Dickey and Fuller (1979) test with the lag augmentation of Said and Dickey (1984) gives p = 0.0428 and rejects a unit root, while Phillips and Perron (1988) give p = 0.3501 and fail to reject one and the KPSS test rejects level stationarity at p = 0.041 (Kwiatkowski et al. 1992), so the three procedures disagree and no clean order of integration is assigned. The DF-GLS escalation of Elliott, Rothenberg, and Stock (1996) is reserved for such ambiguous readings and enters only as confirmation of the ADF leg, never as a fourth vote, and the Ng and Perron (2001) M-tests are likewise reserved for flagged near-unit moving-average cases, so neither overturns the honest non-verdict. The limited power of unit-root and structural-break tests on roughly 94 quarterly observations is kept in view.

The descriptive first-order autoregressive coefficient on the level is 0.945, with a half-life of 12.24 quarters, reported as a persistence summary of the observed ratio rather than as an integration order.

Because the protocol assigns no clean stationary level, the serial-correlation and mean-break diagnostics are computed on the first difference, the object those procedures require, since a portmanteau or break search on a persistent level would spuriously read near-unit autocorrelations and segment the trend (Bai and Perron 1998; Perron 1989). On the differenced ratio the Ljung and Box (1978) portmanteau rejects white noise at both the lag-4 and lag-8 quarterly rungs, with lag-8 p = 0.000. The Bai and Perron (1998) multiple structural-change procedure, computed by the dynamic-programming algorithm of Bai and Perron (2003), finds no break in the mean of the differenced series. A constructed quarterly ratio of aggregated credit and income flows carries no posited seasonal quantity, so the seasonal-frequency battery is not applicable and was not run (Ghysels and Osborn 2001).