The R-zone: when credit and asset prices boom together

The construction

The paper's Eq. (2a)/(2b): High-Debt-Growth = 1 if the three-year change in sector credit to GDP exceeds the pooled 80th percentile; High-Price-Growth = 1 if the three-year log change in the sector's real asset price exceeds the pooled 66.7th percentile; the R-zone is their product. The paper's published full-sample cutoffs are 8.99 pp for business credit, 7.60 pp for household credit, 26.56 for real equity growth and 12.67 for real home-price growth (100 x three-year log change). Those exact numbers are applied here to the JST panel: business credit is 100 x tbus/gdp and household credit 100 x thh/gdp (JST loans), the real equity index is cumulated from JST capital gains and deflated by JST CPI, and the real home price is hpnom/cpi. The same construction recomputed on this panel's own pooled 1950-2016 distribution gives cutoffs of 4.86, 6.10, 23.15 and 12.65: every one lower than the paper's, because the 18 JST advanced economies are calmer than the paper's 42-country panel. Both cutoff sets are run below. Full construction, sources and every labeled deviation: R-zone methodology.

Outcomes are counted two ways. The GHS-mirror runs use the Baron-Verner-Xiong revised crisis list, the chronology the paper itself uses (207 episodes here after excluding wartime episodes, 1870-2016), with forecasts stopping in 2013 so every three-year window is observable. The shelf runs use the Laeven-Valencia 2026 banking chronology (164 onsets, 1976-2023), the same onset list as the estate's credit-gap and DSR warning notes, with forecasts over 1973-2020.

The backtest: in-zone vs out-of-zone

Rows 1-5 and 6-7: JST annual country-years (observations are country-years). Final row: BIS quarterly country-quarters, household leg only, 4,404 observations across 41 economies. GHS report 45.3% business and 36.8% household on their 42-country panel; their 1-year unconditional crisis probability is 4.1%.

The interaction survives every variant run here: in-zone incidence is 3.6 times the out-of-zone rate in the headline row, and the joint business-and-household zone, rare as it is (12 country-years), was followed by a crisis 66.7% of the time, the pattern GHS call out for Japan 1988-89 and Spain 2005-07. The household zone weakens on the Laeven-Valencia chronology (23.4% in-zone against 5.3% out), and the reason is the chronology, not the indicator: over the common 1976-2016 window, BVX dates 36 episodes in these 18 economies where Laeven-Valencia dates 19, so episodes the zone catches under BVX, Switzerland 1990, the UK 1991 and Italy 1992 among them, have no LV counterpart and the same booms score as false alarms; Norway's boom is caught under BVX's 1987 dating but not under LV's 1991 dating of the same distress. Dropping ongoing-crisis years the way the credit-gap note does barely moves it (23.7%). The BIS quarterly household panel, fully out of the JST sample frame, lands at 23.3% in-zone against 4.5% out.

The signal is concentrated in the modern era. Before 1980 the zone almost never fires with the paper's cutoffs (6 business and 0 household in-zone country-years, none followed by a crisis); after 2000, 10 business in-zone country-years saw 60.0% crisis incidence and 42 household country-years 35.7%, against unconditional rates near 25.5% in an era dominated by 2007-08.

The cutoffs transfer as percentiles, not as numbers

At the crisis level (GHS's Table 10 rule: an onset counts as caught if the zone was on in any of the three prior years), the paper's numeric cutoffs catch 5 of 34 testable BVX crises (14.7%) with the business zone and 14 of 38 (36.8%) with the household zone, well short of the 40.0% and 47.7% the paper reports. Recompute the same 80th/66.7th percentiles on this panel's own pooled distribution and the crisis-level hit rates become 14 of 34 (41.2%) and 18 of 38 (47.4%), nearly exactly the paper's figures. The lesson for anyone operationalizing this indicator: the percentile construction is what replicates; the paper's numeric thresholds are statistics of its particular 42-country panel and travel poorly to a calmer advanced-economy panel.

Base rates, both ways

• Most in-zone observations are NOT followed by a crisis in most variants. On the shelf's own chronology, 76.6% of household in-zone country-years (100 minus the displayed 23.4%) saw no crisis onset within three years. Even in the headline GHS-mirror run the majority of in-zone country-years, 15 of 25 business and 45 of 73 household, passed without one.
• Counted as spells (consecutive in-zone runs) rather than country-years: 4 of 12 business spells and 13 of 27 household spells were followed by a crisis within three years of a spell year. The zone multiplies risk severalfold; it is not a prophecy.
• The unconditional three-year crisis probability in the JST-BVX panel is 12.0%. GHS's one-year unconditional is 4.1%. Any binary flag on a rare event will issue mostly false alarms at these base rates; the paper's own argument is about the cost trade-off of acting on them, not about certainty.

The zone today: 2025-Q4

The live reading applies the paper's household cutoffs (7.60 pp credit, 12.67 real home-price growth) to the latest BIS quarter where both three-year legs exist: household credit to GDP from the BIS total credit statistics (WS_TC) and real residential property prices (WS_SPP). 0 of the 41 rated countries are in the household R-zone today. 1 country clears the credit cutoff alone, 7 clear the price cutoff alone, and 33 clear neither. The business zone cannot be rated live: the estate holds no cross-country equity price index, and a zone that cannot see its price leg is unrated, not safe. The three-year change in business (NFC) credit to GDP is shown as the visible half of that pair.

Source: BIS total credit to the non-financial sector (WS_TC, households and NFCs, percentage of GDP) and BIS selected residential property prices (WS_SPP, real index, 2010 = 100), both from the BIS Data Portal bulk files. Unrated rows name the missing leg; 18 countries are unrated.

One reading needs a flag: Norway's +15.2 pp household credit change is partly a denominator effect. The BIS ratio divides by nominal GDP, which the 2022 energy-price windfall inflated; as that washed out of GDP the ratio recovered from 72.1 to 87.3 percent of GDP without a comparable credit boom. A credit-to-GDP change is not a credit change, in either direction.

Against the rest of the early-warning shelf

• The credit gap as crisis warning: the single-variable Borio-Lowe credit-to-GDP gap ranks pre-crisis country-years with an AUC of 0.66 on this estate's data, and that note's central finding is that the paper's low-noise thresholds did not reproduce; the R-zone is the literature's response to exactly that, conditioning credit growth on a simultaneous asset-price boom so the flag fires rarely (in-zone incidence 40.0% here, against 11.1% outside).
• The debt service ratio as crisis warning: the raw DSR manages an AUC of 0.59 within two years of onset on this estate's data; the R-zone interaction is the literature's answer to exactly this weakness of single-variable levels, trading the DSR's short-horizon framing for a three-year joint-boom condition.

The AUCs quoted are those notes' own computed values on this estate's panels. No AUC is computed for the R-zone itself: it is a binary indicator by construction, not a ranking, so threshold-free comparisons do not apply to it directly.