Macro-Prudential Early Warning System — Hong Kong

An automated Early Warning System (EWS) for monitoring systemic financial risk in Hong Kong. Built on the BIS Credit-to-GDP gap methodology and extended with property price and liquidity flow signals, the system produces a composite risk score and colour-coded regime classification (Amber / Orange / Red) on a quarterly basis.

---

Overview

Systemic financial crises rarely arrive without warning. Credit overexpansion, property price misalignment, and liquidity deterioration tend to build over multiple quarters before stress materialises. This EWS is designed to detect that buildup — providing a forward-looking signal suitable for macro-prudential policy decisions rather than real-time crisis detection.

The model draws on the same conceptual framework used by the BIS, HKMA, and IMF for countercyclical capital buffer guidance.

---

Signals

The composite EWS score is derived from the following indicators:

Signal	Source	Method
Credit-to-GDP gap (GAP)	BIS	Hodrick-Prescott (HP) filter
Credit gap momentum (GAP_DELTA)	BIS	Quarter-on-quarter gap change
Residential property price index (INDEX_YOY)	BIS	Year-on-year % change
Liquidity balance crossover (BALANCE_CROSSOVER)	HKMA	Net flow threshold breach
Fast liquidity balance (BALANCE_FAST)	HKMA	Short-window flow level

Each signal is scored on threshold breach, weighted, and summed into a composite score. The score maps to a regime:

Score	Regime	Interpretation
0 – 20	🟢 Green	No significant stress signals
20 – 45	🟡 Amber	Low-to-moderate risk, monitor
45 – 70	🟠 Orange	Elevated risk, consider tightening
70 – 100	🔴 Red	High systemic risk, action warranted

Thresholds, weights, regime bins, and persistence rules are fully configurable via config.yaml.

---

Configuration

All threshold and weighting parameters live in config.yaml — no code changes required to recalibrate the model.

thresholds:
  GAP:
    warn: 2.0
    critical: 10.0
  GAP_DELTA:
    warn: 1.5
    critical: 3.0
  INDEX_YOY:
    warn: 0.10
    critical: 0.15
  BALANCE_CROSSOVER:
    warn: -20.0
    critical: -50.0
  BALANCE_FAST:
    warn: 80.0
    critical: 60.0
weights:
  GAP: 3.0
  INDEX_YOY: 2.5
  GAP_DELTA: 2.0
  BALANCE_CROSSOVER: 1.5
  BALANCE_FAST: 1.0
regime_bins: [0, 20, 45, 70, 100]
persistence_quarters: 2

Thresholds

Each signal has a warn and critical level. Breaching warn contributes a partial score; breaching critical contributes the full weighted score for that signal.

GAP — Credit-to-GDP gap (HP filtered)

The deviation of the credit-to-GDP ratio from its long-run HP-filtered trend, expressed in percentage points. The BIS recommends this as the primary countercyclical capital buffer indicator.

Level	Value	Interpretation
warn	2.0 pp	Early-stage credit overexpansion; monitor
critical	10.0 pp	Severe credit excess; aligns with BIS Red zone guidance

GAP_DELTA — Quarter-on-quarter change in credit gap

The first difference of the GAP series. Captures acceleration in credit growth even when the absolute gap level is still moderate — useful for early detection of a building cycle.

Level	Value	Interpretation
warn	1.5 pp	Gap widening at a concerning pace
critical	3.0 pp	Rapid gap acceleration; strong momentum signal

INDEX_YOY — BIS residential property price index (year-on-year change)

Annual growth rate of the BIS property price index for Hong Kong. Property price inflation is a core systemic risk indicator for HK given the economy's structural exposure to real estate.

Level	Value	Interpretation
warn	10%	Above-trend price appreciation
critical	15%	Rapid price inflation consistent with bubble conditions

BALANCE_CROSSOVER — HKMA liquidity balance crossover

Captures the point at which aggregate HKMA liquidity flows turn negative, signalling net outflows from the banking system. A negative value means outflows are occurring; a more negative value indicates severity.

Level	Value	Interpretation
warn	−20.0	Mild net outflow; early liquidity deterioration
critical	−50.0	Severe net outflow; significant liquidity stress

BALANCE_FAST — Fast-moving liquidity balance

A shorter-window liquidity measure that reacts more quickly to sudden flow changes than the crossover signal. Note the inverted direction: a lower value indicates faster deterioration.

Level	Value	Interpretation
warn	80.0	Liquidity buffer thinning
critical	60.0	Liquidity buffer critically low

---

Weights

Signals are scored on breach and multiplied by their weight before summing into the composite EWS_SCORE. Higher weight = greater influence on the final score.

Signal	Weight	Rationale
GAP	3.0	Primary BIS indicator; highest predictive validity in literature
INDEX_YOY	2.5	HK property prices are the dominant systemic risk driver
GAP_DELTA	2.0	Adds momentum sensitivity to the level signal
BALANCE_CROSSOVER	1.5	Liquidity stress indicator; secondary to credit signals
BALANCE_FAST	1.0	Fast-reacting but noisier; lowest weight to reduce false positives

Total maximum weighted score normalises to 100.

---

Regime bins

regime_bins: [0, 20, 45, 70, 100]

Defines the score boundaries for regime classification:

Score range	Regime	Policy implication
0 – 20	🟢 Green	No significant stress signals
20 – 45	🟡 Amber	Low-to-moderate risk; maintain monitoring
45 – 70	🟠 Orange	Elevated risk; consider macro-prudential tightening
70 – 100	🔴 Red	High systemic risk; strong case for intervention

---

Persistence

persistence_quarters: 2

A regime upgrade (e.g. Amber → Orange) requires the score to breach the threshold for at least 2 consecutive quarters before the regime changes. This prevents single-quarter data noise from triggering false escalations. Regime downgrades apply immediately.

---

This separation of configuration from logic makes the system straightforward to back-test under alternative parameter assumptions without modifying source code.

---

Pipeline

BIS REST API  ──┐
                ├──▶  API-to-DataFrame pipeline  ──▶  Signal scoring  ──▶  EWS_SCORE / EWS_REGIME
HKMA REST API ──┘

• High-performance async data ingestion from BIS and HKMA public APIs
• HP filtering applied to credit and property series to extract cyclical components
• Composite scoring engine with YAML-driven threshold classification
• Output: quarterly EWS_SCORE (0–100) and EWS_REGIME label

---

Sample Output

TIME_PERIOD    EWS_SCORE    EWS_REGIME
2025-10-01         30.0         Amber
2024-01-01         30.0         Amber
2022-04-01         50.0        Orange
2018-04-01         75.0           Red
2015-01-01         90.0           Red   ← cycle peak

The model correctly identified sustained Red regimes during HK's two major property bubble peaks (2015, 2017–2018) and elevated Orange conditions through the post-GFC credit expansion cycle (2010–2014).

---

Methodology Notes

This model is designed as a buildup detector, not a real-time crisis monitor. Credit-to-GDP gap models are well-validated in the academic literature for signalling stress 2–6 quarters ahead of peak. The BIS reports approximately 70% sensitivity at a 10% false positive rate for this class of model globally; HK's property-driven cycle makes the property price index addition particularly informative.

Known limitations:

• Acute liquidity shocks (e.g. Q4 2008) may not be captured — the gap methodology goes quiet once credit contracts
• Regime transitions can lag on the downside; Orange may persist after conditions have already begun normalising
• With ~4 meaningful stress cycles in the 23-year sample, statistical validation is inherently limited — consistent with constraints faced by central bank EWS models globally

---

Requirements

python >= 3.9
pandas
requests
pyyaml

Install dependencies:

pip install -r requirements.txt

---

Usage

python src/main.py

Output is written to output/ews_scores.csv.

To adjust thresholds or signal weights, edit config.yaml before running.

---

References

• BIS Working Paper — Credit-to-GDP gaps and countercyclical capital buffers (Basel III)
• HKMA — Residential Mortgage Survey & Monetary Statistics
• Drehmann, M. & Tsatsaronis, K. (2014) — The credit-to-GDP gap and countercyclical capital buffers