## _wp11170

## Source details

**Canonical URL:** [_wp11170](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11170.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11170.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11170.pdf.json)

---

### Introduction
- Objective: provide a quantitative assessment of overall financial conditions in 13 Asian economies by developing a new Financial Condition Index (FCI).
- Motivation: the 2008 global financial crisis highlighted linkages between financial conditions and the real economy; sudden tightening at end-2008 contributed to a sharp decline in output in many Asian economies, followed by an unprecedented monetary policy response and rebound in equity markets that supported a fast recovery since mid 2009.
- Sample timeframe: develop FCIs for 13 Asian economies between 2001 up to the beginning of 2011.
- Sample economies: Australia, China, Hong Kong SAR, India, Indonesia, Japan, Korea, Malaysia, New Zealand, Philippines, Singapore, Thailand, and Taiwan Province of China.
- Conceptual point: an ideal FCI should measure exogenous changes in financial conditions; the endogenous component of financial variables that reflects the business cycle should be removed.

### Methodological overview
- Two complementary approaches used to construct FCIs:
  - Weighted-sum approach (weights reflect estimated impact of each financial variable on real GDP).
  - Principal-components approach (estimate an unobserved common factor from several financial variables).
- Novelty:
  - Employ both weighted-sum and principal-component approaches and take the average of the two indices to construct a combined financial indicator.
  - Control for endogenous responses of financial variables to economic activity, so the index presents exogenous changes in financial conditions.
- Financial channels captured by FCIs: investment (quantity and cost of capital), consumption (wealth, confidence, and income effects), and trade (price competitiveness of tradable goods).

### Weighted-sum approach (VAR-based)
- Endogenous variables included in the VAR: GDP, equity prices, prices, the effective exchange rate, the spread of lending rates over policy rates, and bank credit to the private sector.
- Weights are derived from generalized impulse responses to ensure independence from variable ordering in the VAR.
- FCI calculation uses cumulative responses of GDP growth to one-unit shocks to each financial variable, averaged over a specified horizon (between 4–6 quarters), and scales by an average of the financial variable over the whole sample period (1990–2010).

### Principal-component approach (GDFM-based)
- Two-step construction:
  1. Estimate the unobserved common component using the Generalized Dynamic Factor Model (GDFM) of Forni and Lippi (2005).
  2. Regress the estimated common factor on current and lagged business cycle variables (real GDP growth and inflation) and take the residuals to purge endogenous predictive content.
- Data set for each economy is a balanced panel comprised of: interest rate spreads of government bonds of different maturity/lending rates over policy rates, the nominal effective exchange rate, nominal credit to the private sector, and stock market indices.
- In the GDFM, each observation is decomposed into a common component driven by a small number of common shocks and an idiosyncratic component; the common factor is estimated through dynamic principal component analysis.
- The common factor dimension ݍ
 ௝ is set to 1 for each economy, and the resulting factor represents the FCI.
- Purging endogeneity: regress the common factor on current and lagged Zj (real GDP growth and inflation rates); the residual is the FCI reflecting only exogenous shifts in financial conditions.

### Combined index and interpretation
- Combined FCI: simple average of the VAR-based FCI and the GDFM-based FCI for each economy.
- Rationale: VAR approach is simple and intuitive but assumes a static relationship (fixed weights); GDFM approach captures dynamism but requires reestimation as new data arrive.
- Interpretation: an upward movement of the index suggests more accommodative overall financial conditions; a decline suggests tighter financial conditions that should lead to deceleration in economic growth.
- Empirical performance: the final index performs well in tracking economic activity; the contemporaneous correlation of the FCI and GDP growth is about 0.6 for emerging Asia.

### Drivers and relative contributions to the FCI
- FCI components: interest rate spread/real interest rate, real exchange rate (REER), credit growth, and stock market indices.
- Cross-country patterns:
  - China and the Philippines: contribution of credit growth to the FCI is relatively larger.
  - Hong Kong SAR and Taiwan Province of China: contribution of exchange rates is relatively greater.
  - Economies with larger contributions from exchange rate and equity movements (Hong Kong SAR, Taiwan Province of China, Singapore) tend to have greater GDP growth volatility.
  - Economies with larger contributions from interest rates and credit (India, China) have less volatile GDP growth.
- House prices excluded from the FCI due to data limitations in several countries (authorities in Australia noted housing wealth should ideally be included).

### Predictive power — in-sample methodology and results
- Regression framework used (equation 5): growtht = autoregressive terms + γi * FCIt−i + ...; significance of γi indicates FCI explanatory power for year-on-year real GDP growth.
- Partial R²: measures marginal contribution of FCI when autoregressive growth terms are held constant.
- In-sample findings (h = 2 forecast horizon):
  - FCIs are powerful leading indicators for GDP growth; statistical tests confirm significance over a two-quarter horizon.
  - Partial R² values are in most cases above 0.7, implying errors in predicting GDP growth based on autoregressive parts could be reduced by 70 percent if FCI is included.
  - Selected in-sample partial R² and t-stat highlights (h = 2):
    - Australia (VAR 0.62 t-stat 2.03; GDFM 0.71 t-stat 2.47; Comb 0.73 t-stat 2.82)
    - China (VAR 0.71 t-stat 2.15; GDFM 0.8 t-stat 3.79; Comb 0.86 t-stat 2.14)
    - India (VAR 0.82 t-stat 4.94; GDFM 0.81 t-stat 1.97; Comb 0.69 t-stat 1.58)
    - Korea (VAR 0.72 t-stat 1.94; GDFM 0.8 t-stat 5.61; Comb 0.78 t-stat 4.96)
    - Malaysia (VAR 0.76 t-stat 9.08; GDFM 0.57 t-stat 2.37; Comb 0.71 t-stat 4.22)
  - VAR-based and GDFM-based FCIs have comparable performances; combined series generally outperform both in in-sample tests.

### Out-of-sample predictive performance (RMSE) and real-time forecasting
- Out-of-sample tests compute RMSEs recursively for h = 2 forecast horizon; results reported both excluding and including the global crisis.
- Combined-series FCIs generally produce lower RMSEs than models with NO FCI, VAR-only, or GDFM-only for most economies.
  - Examples (RMSE excluding crisis / including crisis):
    - Australia: NO FCI 0.81 / 0.91; VAR 0.74 / 0.86; GDFM 0.77 / 0.85; Comb 0.76 / 0.84
    - China: NO FCI 1.1 / 1.18; VAR 1.03 / 1.05; GDFM 0.87 / 1.07; Comb 0.82 / 0.79
    - Japan: NO FCI 1.59 / 2.12; VAR 1.37 / 1.69; GDFM 1.92 / 2.01; Comb 1.23 / 1.88
    - Malaysia: NO FCI 1.41 / 2.28; VAR 1.44 / 1.84; GDFM 1.15 / 2.1; Comb 0.93 / 1.71
    - Singapore: NO FCI 2.8 / 4.1; VAR 3.14 / 3.74; GDFM 3.14 / 3.94; Comb 2.72 / 3.59
- Real-time forecasting experiments:
  - Real-time forecasts use only information available at the time.
  - Correlation of real-time GDP forecasts with actual outturns increases substantially when FCIs are used: average correlation rises from 0.68 to 0.87.
  - Country-level correlations without and with FCIs (selected):
    - Australia: 0.68 → 0.86
    - China: 0.58 → 0.97
    - Hong Kong SAR: 0.71 → 0.92
    - India: 0.77 → 0.88
    - Japan: 0.74 → 0.84
    - Philippines: 0.56 → 0.72
    - Singapore: 0.55 → 0.81
    - Taiwan Province of China: 0.76 → 0.85

### Developments through the global financial crisis (periodization and impacts)
- Periods defined: Precrisis (2006–08:Q2), Financial Crisis (2008:Q3–2009:Q1), Recovery (2009:Q2–2010:Q1), Normalization (2010:Q2 onwards).
- Precrisis (2006–08:Q2):
  - Financial conditions in Asia relatively accommodative; driven by favorable equity markets and credit.
  - Weaker REERs contributed positively in China, Hong Kong SAR, Japan, Taiwan Province of China.
  - Monetary policy was tightened in many economies to curb inflation.
  - Local FCI peaks typically in late 2007; FCI peaks led GDP growth peaks by one to two quarters in many cases.
- Financial Crisis (2008:Q3–2009:Q1):
  - Market confidence collapsed end-2008; sharp pullout of funds from Asia.
  - Declines in stock valuations and tighter credit conditions caused sharp tightening of FCIs in late 2008/early 2009.
  - Monetary easing could not fully offset equity and credit shocks; FCI trough in 2008:Q4/2009:Q1; GDP trough typically one to two quarters later.
- Recovery (2009:Q2–2010:Q1):
  - Strong policy reactions and rapid equity market recovery eased financial conditions significantly late 2009.
  - In India, Indonesia, Korea, the Philippines, and Thailand, the FCI surpassed precrisis levels by 2009:Q4.
  - Exchange rate depreciation in H2 2009 contributed positively to FCIs in Australia, Korea, Taiwan Province of China.
  - FCIs local peaks in 2009:Q4/2010:Q1; GDP peaks generally one quarter later.
- Normalization (2010:Q2 onwards):
  - Financial conditions began to tighten in 2010 despite improved credit growth since mid-2010.
  - Equity gains stabilized; real exchange rates appreciated in much of the region; policy rates slowly normalized.
  - Tightening from equity stabilization, stronger exchange rates, and higher policy rates were partly offset by continued credit recovery.

### Scenario simulation (second half of 2010)
- Two scenarios simulated for 2010: second half:
  - Scenario A: FCI remains constant at 2010:Q2 level.
  - Scenario B: FCI eases to the level of 2010:Q1.
- Result:
  - Under Scenario B, GDP growth accelerates by about one-half of a percentage point in the next four quarters.

### Conclusions and policy implications
- Rationale:
  - Policy interest rate captures only one dimension of financial conditions; a composite FCI summarizes multiple financial channels that influence the economy and provides timely real-time information that macro variables (e.g., GDP) do not.
- Methodology:
  - FCIs constructed using two methodologies: VAR-based weighted sum and GDFM-based principal component. Combination of both methods yields superior predictive performance.
- Main findings:
  - FCIs display leading indicator properties: turning points in FCIs precede turning points in economic activity.
  - Swings in equity markets were a major factor driving FCI movements during the crisis and subsequent recovery.
  - Combined FCIs improve in-sample and out-of-sample GDP forecasting and increase real-time forecast correlations substantially (average from 0.68 to 0.87).
- Implication for policymakers:
  - Monitoring FCIs and the relative contributions of component financial variables can help identify which financial variables are causing excessively accommodative or tight conditions and guide the appropriate policy reaction.

*Source: _wp11170 - References.*

### References .............................................................................................................

### _wp11170 - References

### Introduction
- Objective: provide a quantitative assessment of overall financial conditions in 13 Asian economies by developing a new Financial Condition Index (FCI).
- Motivation: the 2008 global financial crisis highlighted linkages between financial conditions and the real economy; sudden tightening at end-2008 contributed to a sharp decline in output in many Asian economies, followed by an unprecedented monetary policy response and rebound in equity markets that supported a fast recovery since mid 2009.
- Sample timeframe: develop FCIs for 13 Asian economies between 2001 up to the beginning of 2011.
- Sample economies: Australia, China, Hong Kong SAR, India, Indonesia, Japan, Korea, Malaysia, New Zealand, Philippines, Singapore, Thailand, and Taiwan Province of China.
- Conceptual point: an ideal FCI should measure exogenous changes in financial conditions; the endogenous component of financial variables that reflects the business cycle should be removed.

### Methodological overview
- Two complementary approaches used to construct FCIs:
  - Weighted-sum approach (weights reflect estimated impact of each financial variable on real GDP).
  - Principal-components approach (estimate an unobserved common factor from several financial variables).
- Novelty of this paper:
  - Employ both weighted-sum and principal-component approaches and take the average of the two indices to construct a combined financial indicator.
  - Control for endogenous responses of financial variables to economic activity, so the index presents exogenous changes in financial conditions.
- Financial channels captured by FCIs: investment (quantity and cost of capital), consumption (wealth, confidence, and income effects), and trade (price competitiveness of tradable goods).

### Weighted-sum approach (VAR-based)
- Endogenous variables included in the VAR: GDP, equity prices, prices, the effective exchange rate, the spread of lending rates over policy rates, and bank credit to the private sector.
- Weights are derived from generalized impulse responses to ensure independence from variable ordering in the VAR.
- FCI calculation uses cumulative responses of GDP growth to one-unit shocks to each financial variable, averaged over a specified horizon (between 4–6 quarters), and scales by an average of the financial variable over the whole sample period (1990–2010).

### Principal-component approach (GDFM-based)
- Two-step construction:
  1. Estimate the unobserved common component using the Generalized Dynamic Factor Model (GDFM) of Forni and Lippi (2005).
  2. Regress the estimated common factor on current and lagged business cycle variables (real GDP growth and inflation) and take the residuals to purge endogenous predictive content.
- Data set for each economy is a balanced panel comprised of: interest rate spreads of government bonds of different maturity/lending rates over policy rates, the nominal effective exchange rate, nominal credit to the private sector, and stock market indices.
- In the GDFM, each observation is decomposed into a common component driven by a small number of common shocks and an idiosyncratic component; the common factor is estimated through dynamic principal component analysis.
- In the analysis, the common factor dimension ݍ
 ௝ is set to 1 for each economy, and the resulting factor represents the FCI.
- Purging endogeneity: regress the common factor on current and lagged Zj (real GDP growth and inflation rates); the residual is the FCI reflecting only exogenous shifts in financial conditions.

### Combined index and interpretation
- Combined FCI: simple average of the VAR-based FCI and the GDFM-based FCI for each economy.
- Rationale: VAR approach is simple and intuitive but assumes a static relationship (fixed weights); GDFM approach captures dynamism but requires reestimation as new data arrive.
- Interpretation: an upward movement of the index suggests more accommodative overall financial conditions; a decline suggests tighter financial conditions that should lead to deceleration in economic growth.
- Empirical performance: the final index performs well in tracking economic activity; the contemporaneous correlation of the FCI and GDP growth is about 0.6 for emerging Asia.

### Key methodological and data notes
- Weighted-sum weights are based on generalized impulse responses and use a GDP response horizon of between 4–6 quarters.
- The average used to scale financial variables in the weighted-sum approach is computed over the whole sample period 1990–2010.
- Data availability considerations influenced variable selection and the country sample.
- Footnote data usage: in practice, the paper uses data from 2000 up to the end of 2011 for consistency with the VAR-based index construction and the GDFM estimation; see the data appendix for details.

*Source: _wp11170 - References.*

### 0.9 for advanced Asia (Figure 1). However, assessing the predictive power of the FCI

### _wp11170 - 0.9 for advanced Asia (Figure 1). However, assessing the predictive power of the FCI

### Drivers and relative contributions to the Financial Conditions Index (FCI)
- The FCI is constructed from interest rate spread/real interest rate, real exchange rate (REER), credit growth, and stock market indices.  
- Relative contributions vary across economies and reflect financial/economic structure:
  - China and the Philippines: contribution of credit growth to the FCI is relatively larger, reflecting a greater role for banking intermediation.
  - Hong Kong SAR and Taiwan Province of China: contribution of exchange rates is relatively greater, reflecting export dependence.
  - Economies with larger contributions from exchange rate and equity movements (Hong Kong SAR, Taiwan Province of China, Singapore) tend to have greater GDP growth volatility.
  - Economies with larger contributions from interest rates and credit (India, China) have less volatile GDP growth.
- House prices were excluded from the FCI due to data limitations in several countries (authorities in Australia noted housing wealth should ideally be included).

### Predictive power: methodology and in-sample results
- Regression framework used (equation 5): growtht = autoregressive terms + γi * FCIt−i + ...; significance of γi indicates FCI explanatory power for year-on-year real GDP growth.
- Partial R²: measures marginal contribution of FCI when autoregressive growth terms are held constant. Partial R² is obtained by regressing residuals of growth on residuals of FCI (both after removing growth t−i effects).
- In-sample findings (h = 2 forecast horizon):
  - FCIs are powerful leading indicators for GDP growth; statistical tests (t-statistics and F-statistics) confirm significance over two-quarter horizon.
  - Partial R² values are in most cases above 0.7, implying errors in predicting GDP growth based on autoregressive parts could be reduced by 70 percent if FCI is included.
  - Selected in-sample partial R² and t-stat highlights (h = 2):
    - Australia (VAR 0.62 t-stat 2.03; GDFM 0.71 t-stat 2.47; Comb 0.73 t-stat 2.82)
    - China (VAR 0.71 t-stat 2.15; GDFM 0.8 t-stat 3.79; Comb 0.86 t-stat 2.14)
    - India (VAR 0.82 t-stat 4.94; GDFM 0.81 t-stat 1.97; Comb 0.69 t-stat 1.58)
    - Korea (VAR 0.72 t-stat 1.94; GDFM 0.8 t-stat 5.61; Comb 0.78 t-stat 4.96)
    - Malaysia (VAR 0.76 t-stat 9.08; GDFM 0.57 t-stat 2.37; Comb 0.71 t-stat 4.22)
  - VAR-based and GDFM-based FCIs have comparable performances; combined series generally outperform both in in-sample tests.

### Out-of-sample predictive performance (RMSE) and real-time forecasting
- Out-of-sample tests compute RMSEs recursively for h = 2 forecast horizon; results reported both excluding and including the global crisis.
- Combined-series FCIs generally produce lower RMSEs than models with NO FCI, VAR-only, or GDFM-only for most economies.
  - Examples (RMSE excluding crisis / including crisis):
    - Australia: NO FCI 0.81 / 0.91; VAR 0.74 / 0.86; GDFM 0.77 / 0.85; Comb 0.76 / 0.84
    - China: NO FCI 1.1 / 1.18; VAR 1.03 / 1.05; GDFM 0.87 / 1.07; Comb 0.82 / 0.79
    - Japan: NO FCI 1.59 / 2.12; VAR 1.37 / 1.69; GDFM 1.92 / 2.01; Comb 1.23 / 1.88
    - Malaysia: NO FCI 1.41 / 2.28; VAR 1.44 / 1.84; GDFM 1.15 / 2.1; Comb 0.93 / 1.71
    - Singapore: NO FCI 2.8 / 4.1; VAR 3.14 / 3.74; GDFM 3.14 / 3.94; Comb 2.72 / 3.59
- Real-time forecasting experiments:
  - Real-time forecasts use only information available at the time.
  - Correlation of real-time GDP forecasts with actual outturns increases substantially when FCIs are used: average correlation rises from 0.68 to 0.87.
  - Country-level correlations without and with FCIs (selected):
    - Australia: 0.68 → 0.86
    - China: 0.58 → 0.97
    - Hong Kong SAR: 0.71 → 0.92
    - India: 0.77 → 0.88
    - Japan: 0.74 → 0.84
    - Philippines: 0.56 → 0.72
    - Singapore: 0.55 → 0.81
    - Taiwan Province of China: 0.76 → 0.85

### Developments through the global financial crisis (periodization and impacts)
- Periods defined: Precrisis (2006–08:Q2), Financial Crisis (2008:Q3–2009:Q1), Recovery (2009:Q2–2010:Q1), Normalization (2010:Q2 onwards).
- Precrisis (2006–08:Q2):
  - Financial conditions in Asia relatively accommodative; driven by favorable equity markets and credit. Weaker REERs contributed positively in China, Hong Kong SAR, Japan, Taiwan Province of China.
  - Monetary policy was tightened in many economies to curb inflation.
  - Local FCI peaks typically in late 2007; FCI peaks led GDP growth peaks by one to two quarters in many cases.
- Financial Crisis (2008:Q3–2009:Q1):
  - Market confidence collapsed end-2008; sharp pullout of funds from Asia.
  - Declines in stock valuations and tighter credit conditions caused sharp tightening of FCIs in late 2008/early 2009.
  - Monetary easing could not fully offset equity and credit shocks; FCI trough in 2008:Q4/2009:Q1; GDP trough typically one to two quarters later.
- Recovery (2009:Q2–2010:Q1):
  - Strong policy reactions and rapid equity market recovery eased financial conditions significantly late 2009.
  - In India, Indonesia, Korea, the Philippines, and Thailand, the FCI surpassed precrisis levels by 2009:Q4.
  - Exchange rate depreciation in H2 2009 contributed positively to FCIs in Australia, Korea, Taiwan Province of China.
  - FCIs local peaks in 2009:Q4/2010:Q1; GDP peaks generally one quarter later.
- Normalization (2010:Q2 onwards):
  - Financial conditions began to tighten in 2010 despite improved credit growth since mid-2010.
  - Equity gains stabilized; real exchange rates appreciated in much of the region; policy rates slowly normalized.
  - Tightening from equity stabilization, stronger exchange rates, and higher policy rates were partly offset by continued credit recovery.

### Scenario simulation (second half of 2010)
- Two scenarios simulated for 2010: second half:
  - Scenario A: FCI remains constant at 2010:Q2 level.
  - Scenario B: FCI eases to the level of 2010:Q1.
- Result:
  - Under Scenario B, GDP growth accelerates by about one-half of a percentage point in the next four quarters.

### Conclusions and policy implications
- Rationale:
  - Policy interest rate captures only one dimension of financial conditions; a composite FCI summarizes multiple financial channels that influence the economy and provides timely real-time information that macro variables (e.g., GDP) do not.
- Methodology:
  - FCIs constructed using two methodologies: VAR-based weighted sum and GDFM-based principal component. Combination of both methods yields superior predictive performance.
- Main findings:
  - FCIs display leading indicator properties: turning points in FCIs precede turning points in economic activity.
  - Swings in equity markets were a major factor driving FCI movements during the crisis and subsequent recovery.
  - Combined FCIs improve in-sample and out-of-sample GDP forecasting and increase real-time forecast correlations substantially (average from 0.68 to 0.87).
- Implication for policymakers:
  - Monitoring FCIs and the relative contributions of component financial variables can help identify which financial variables are causing excessively accommodative or tight conditions and guide the appropriate policy reaction.

*Source: Authors' estimates.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11170.pdf_