## 1. Variables in FCI by Country (wp17218)

## Source details

**Canonical URL:** [1. Variables in FCI by Country (wp17218)](https://www.imf.org/-/media/files/publications/wp/2017/wp17218.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17218.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17218.pdf.json)

---

### Introduction and scope
- Purpose: develop financial conditions indices (FCIs) for the six large and most financially-integrated Latin American economies (LA6) that are comparable across countries.
- Definition used: financial conditions = exogenous changes in financial markets as opposed to changes that reflect the business cycle or monetary policy decisions.
- LA6 countries: Argentina, Brazil, Chile, Colombia, Mexico, and Peru.
- Uses of estimated FCIs:
  - Characterize financial conditions and underlying common drivers in the region.
  - Investigate the impact of financial conditions on economic activity and time-variation in that relationship.
  - Characterize financial spillovers within the region.

### Methodology overview
- Estimation approach: Time-Varying Parameter Factor-Augmented Vector Autoregression (TVP-FAVAR), following Korobilis (2013) and Koop and Korobilis (2014).
- Distinctive methodological features:
  - Broad array of financial variables capturing credit conditions, leverage and quality of collateral, and intensity of perceived risk.
  - Loadings (weights) allowed to vary over time; variables may have varying or zero loadings in different periods.
  - Time-varying transmission of financial conditions to economic activity; method accommodates evolving macro-financial relationships.
- Technical implementation notes:
  - FCI estimated using Koop and Korobilis’ 2014 code; combines Primiceri’s (2005) time-varying parameter VAR with factor analysis for large datasets (FAVARs).
  - Approach relies on the Kalman filter and smoother and is described as simulation-free.
  - Rationale: TVP-FAVAR accommodates structural instabilities and changing relevance of financial variables over time.

### Variable selection and grouping
- Objective: measure exogenous financial shocks by purging FCI variables of developments originating outside the financial system (business cycle, monetary policy, exchange movements).
- Financial variables grouped into three categories:
  - Credit (quantity developments):
    - Examples: loans by segment, domestic bond and equity issuance, measures of credit quality such as NPLs and expected probabilities of default for borrowers.
    - Economic channel: measure ease of access to finance.
  - Collateral and leverage variables:
    - Examples: returns on stock indices, housing prices, price-to-earnings ratios, capital-to-assets ratios.
    - Economic channel: reflect balance sheet quality of lenders and borrowers.
  - Risk measures:
    - Examples: changes in CDS sovereign and EMBIG spreads, inter-bank, corporate and term spreads; volatility measures (equity returns and FX volatilities).
    - Economic channel: capture lenders’ tolerance of risk.
- Summary behavioral relationship: Declining credit volumes and collateral, and widening spreads, lead to tightening in financial conditions, and vice-versa.

### Data coverage and transformations
- Coverage: mostly monthly data from early 2000 until February 2017.
- Frequency conversion: a few quarterly variables converted to monthly using cubic spline interpolation.
- Missing data: replaced with zeros.
- Transformations applied to induce stationarity (legend from Table 1):
  - DL = first difference
  - DLN = first log-difference
  - DP12 = 12-month difference
  - LV = level
  - LVMA = ratio of level to 12-month moving average
- Example variables and transformations (verbatim list excerpts):
  - 1-Year EDF Banks (75th percentile) — LV — Credit
  - Domestic bond issuance — LVMA — Credit
  - Loans trade — DP12 — Credit
  - Nonperforming loans to bank credit ratio — LV — Credit
  - Total bank assets — LVMA — Credit
  - Capital to assets ratio — LV — Leverage
  - Housing prices — DLN — Leverage
  - Stock market general index — DLN — Leverage
  - CDS Sovereign 2YR — LV — Risk
  - EMBIG — LV — Risk
  - FX Volatility — LV — Risk
  - Net loan spread financial system, 30 to 89 days — DLV — Risk
- Data sources (verbatim listings): Banco Central de Brasil; Banco Central de Chile; Banco Central de Colombia; Banco Central de la República Argentina; Banco de México; Banco Central de Reserva del Perú; Bloomberg LP; Comisión Nacional Bancaria y de Valores México; FactSet; Moody's CreditEdge+; Superintendencia de Bancos e Instituciones Financieras de Chile; Superintendencia de Banca, Seguros y AFP del Perú; Superintendencia de Valores y Seguros de Chile; Thomson-Reuters Datastream.

### Empirical model and identification
- Model specification (text equation (1)):
  - X = vector of financial variables.
  - Y = vector of macroeconomic variables (inflation, a monthly proxy for growth, the real effective exchange rate, and the monetary policy rate).
  - f = unobservable (first) factor (the financial conditions index).
  - ε and u = uncorrelated (but heteroskedastic) error terms.
  - Coefficients λ_y and λ_f = factor loadings tracking effects of macroeconomic and financial conditions on each financial variable.
- Two-equation interpretation:
  - First equation: extraction of latent indicator f that, after removing contemporaneous macroeconomic effects, summarizes a large set of financial variables.
  - Second equation: models dynamic relationship between financial and macroeconomic conditions; with identifying assumptions, quantifies effects of financial conditions on macroeconomy and vice-versa.
- Identification for GDP effect: use of a Cholesky decomposition (ordering: inflation, GDP growth, REER, monetary policy rate, FCI).
- Normalization:
  - FCI normalized to mean zero and standard deviation one over observation period (2000–2016).
  - Zero = historical average financial conditions; values above (below) zero indicate “tight” (“loose”) conditions relative to historical average.

### Main findings — characterization of FCIs (2000–Feb 2017)
- Drivers of estimated FCIs:
  - A commodity cycle.
  - A global financial cycle.
  - Country-specific episodes of financial distress.
- Country-level status as of February 2017:
  - Loose: Argentina and Chile.
  - Broadly neutral: Colombia, Mexico, and Peru.
  - Somewhat tight: Brazil.
- Notable historical episodes:
  - 2007−9 financial crisis: sharpest spike across region (except Argentina).
  - European sovereign crisis (2011−12) and taper tantrum (early 2013): strongly felt in Brazil, Chile, and Peru.
  - Country-specific episodes: Argentina (collapse of currency board), Colombia (tight credit early 2000s), Peru (political uncertainty and slump of 2001).
- 2016 loosening drivers:
  - Argentina: easing credit and risk.
  - Chile: increasing credit and financial firms’ capital.
  - Peru: easing credit.
- 2016 tightening drivers:
  - Brazil: tighter credit and declining collateral.
  - Mexico: marginal credit deterioration.

### Common factors and variance decomposition
- Principal components analysis of the six country FCIs:
  - Two latent factors summarize up to 70 percent of the variance structure across region.
  - First common factor: accounts for about 43 percent of variance; correlation with world commodity prices = -0.73.
  - Second common factor: explains about 26 percent of variance; correlation with the VIX = 0.38; proxy for global financial cycle.
- Country correlations with principal components (as presented):
  - PC1 correlations: Argentina 0.73; Brazil 0.62; Chile -0.74; Colombia 0.62; Mexico 0.54; Peru -0.64.
  - PC2 correlations: Argentina 0.14; Brazil -0.37; Chile 0.60; Colombia 0.45; Mexico 0.60; Peru 0.71.
- Interpretation:
  - Commodity-price-driven factor closely correlated with financial conditions in all countries; highest absolute correlation with Chile and Argentina; lowest with Mexico.
  - Global financial factor (VIX-driven) strongly linked to Chile, Mexico, and Peru; less to Argentina, Brazil, and Colombia.
  - Chinn-Ito index values (2014): Peru 1, Chile 0.7, Mexico 0.7, Argentina 0, Brazil 0.4, Colombia 0.4 (reference for financial openness differences).

### Impact of financial conditions on activity (quantitative results)
- Identification and method:
  - Cholesky decomposition with ordering: inflation, GDP growth, REER, monetary policy rate, FCI.
  - For each month, time-varying VAR coefficients and factor loadings used to build impulse response functions.
- Key quantitative findings (using November 2016 loadings):
  - A one standard deviation tightening of financial conditions lowers half-year-ahead GDP growth by:
    - between 0.05−0.1 percentage points for Mexico, Argentina, Chile, and Peru.
    - between 0.25−0.3 percentage points for Brazil and Colombia.
  - Timing of trough in activity after shock:
    - Fifth month: Argentina and Chile.
    - Tenth month: Colombia.
  - Comparative interpretations:
    - A one standard deviation shock suffices to generate a ¼ percentage point fall in GDP growth in Brazil and Colombia.
    - Two standard deviations would be needed in Chile and Argentina to trigger a 0.25 percentage point fall.
    - Mexico and Peru would need around 4 ½ and 6 ½ standard deviation shocks respectively to trigger a 0.25 percentage point fall in GDP growth.
- Time variation:
  - Effect of financial tightening on activity is time-varying and can be between two and three times higher in periods of financial stress (example: Chile lost 2.4 percent of GDP after 24 months in October 2008 vs 1.5 percent of GDP in November 2016 from a one-standard deviation FCI shock).

### Model comparisons
- Models compared:
  - TVP-FAVAR (main model).
  - PCA-FAVAR (principal components-based).
  - KF-FAVAR (Kalman-filter based).
- Empirical observations:
  - PCA-FAVAR and KF-FAVAR closely track TVP-FAVAR for LA6.
  - Contributions of variables to the FCI fairly stable over time; profile driven primarily by changes in underlying financial variables rather than time-varying weights.
  - Differences: TVP-FAVAR points to somewhat looser financial conditions in Argentina and Brazil, and slightly tighter in Mexico at end of sample relative to alternatives.
  - TVP-FAVAR advantage: allows measured effect of financial conditions on economic activity to change over time, potentially improving forecast accuracy.

### Regional financial interconnectedness — approach
- Objective: assess importance of global, regional, and domestic shocks in shaping domestic financial conditions and quantify spillovers across LA6 (2014−16).
- Method:
  - Apply Diebold and Yilmaz’s (2014) spillovers framework.
  - Estimate VAR of monthly FCIs including world commodity prices and the VIX to remove common-factor co-movement.
  - Use generalized forecast-error variance decomposition (GVD) per Pesaran and Shin (1998) on 3-month ahead forecast errors.
- Focus period for stylized facts: 2014−16.

### Regional interconnectedness — stylized facts (2014−16)
- Domestic vs regional drivers:
  - Shocks originating from local financial sector explain largest fraction of FCI variance across LA6.
  - Brazil and Mexico: domestic shocks explain about 87 and 82 percent of the variance respectively.
  - Remaining LA6 countries: regional shocks explain between 33 percent (Colombia) and 43 percent (Peru) of FCI variability.
- Outward spillover potential:
  - Largest spillover potential: Brazil and Colombia, followed by Mexico, Argentina, and Peru.
  - Chile transmitted comparatively smaller FCI spillovers to the region.
- Inward spillovers:
  - Strongest receivers: Peru, Chile, and Argentina (fractions of variance decomposition of each country’s FCI explained by rest of region: about 44, 38, 37, and 33 percent respectively — note ordering in source text).
  - Brazil and Mexico: inward spillovers comparatively smaller (about 13 and 18 percent respectively).
- Notable bilateral links:
  - Argentina strongly affected by Mexico.
  - Chile significantly affected by Argentina.
  - Feedback loop between Colombia and Peru.

### Conclusions and policy-relevant research agenda
- Key summary findings:
  - Estimated FCIs are influenced by a commodity cycle and a global financial cycle and track country-specific episodes of financial distress.
  - By early 2017, financial conditions remained favorable in most LA6 countries relative to historical standards.
  - Quantified effect of financial shocks on output: a one standard deviation tightening of financial conditions lowers half-year ahead GDP growth by about 0.1 percentage points (0.2− ¼ percentage points in periods of financial stress).
  - Evidence of significant financial spillovers within the region.
- Going forward — research priorities and open questions:
  - Improve understanding of drivers of correlations between global factors (commodity prices, VIX) and each country’s FCI (candidate explanations include nature of reliance on commodity exports and degree of financial openness).
  - Investigate time-varying nature of GDP response to FCI shocks and which financial frictions drive changing transmission.
  - Further scrutiny of regional financial spillovers: do spillovers follow observed patterns of trade and cross-border investment flows? (Note: FCIs are purged of macroeconomic drivers, so real trade ties should not explain influence in this framework.)

### Appendix highlights — data tables and contribution mechanics
- Appendix A (data sources and variable coverage):
  - Transformations and groupings reiterated (DL, DLN, DP12, LV, LVMA; groups: Credit, Collateral/Leverage, Risk).
  - Institutional and acronym notes included verbatim (examples: Abrapp; BanRep; BCRA; BCRP; CDS; EDF; ELMI+LC; EMBIG; MMMF; MPR; SBIF; SBS; SVS).
- Appendix B (contributions to FCI):
  - Method: linear projection solving first equation for f yields expression where the k-th element of (1ˆˆ ˆ fff   ) gives gross contribution of kth financial variable in X to FCI f; netting of macro variables given by (1ˆˆ ˆˆ fffy y   ).
  - Interpretation: difference between actual change in f and estimated contributions from X attributed to changes in macroeconomic variables.
- Appendix C (time-varying loadings and contributions):
  - TVP-FAVAR allows loadings to change each period; time-varying weights can improve forecast accuracy and imply varying variable contributions to FCI.
  - Table A.2 (selected volatilities of variable contributions, standard deviations of scores by variable across countries, verbatim excerpts):
    - ELMI + LC: 0.0050 (ARG), 0.0190 (BRA), 0.0070 (CHL), 0.0060 (COL), 0.0080 (MEX), 0.000 (PER)
    - Total bank assets: 0.0140 (ARG), 0.0300 (BRA), 0.0030 (CHL), 0.0020 (COL), 0.0080 (MEX), 0.003 (PER)
    - EMBIG: 0.0060 (ARG), 0.0850 (BRA), 0.0110 (CHL), 0.0340 (COL), 0.0210 (MEX), 0.009 (PER)
    - Central bank rate - US T-Bill spread: 0.0000 (ARG), 0.0250 (BRA), 0.0030 (CHL), 0.0170 (COL), 0.0100 (MEX), 0.004 (PER)
    - Stock market general index: 0.0040 (ARG), 0.0240 (BRA), 0.0030 (CHL), 0.0070 (COL), 0.0080 (MEX), 0.003 (PER)
  - Table A.3 (stability of ranking of variable contributions — Top 3 Positive and Top 3 Negative Contributions by country, verbatim examples):
    - Argentina — Top 3 Positive: Total loans to private sector; Loans commercial; Loans trade. Top 3 Negative: ELMI + LC; Interest rates commercial loans - MPR spread; Volatility of stock market general index.
    - Brazil — Top 3 Positive: Central bank rate-US T-Bill spread; EMBIG; Total bank assets. Top 3 Negative: Total bank assets; Total loans to private sector; Total bank assets.
    - Chile — Top 3 Positive: Total loans to private sector; Total loans to private sector; Interest rates housing loans 3 years or more - MPR spread. Top 3 Negative: Capital to assets ratio; Loans housing / Bank assets; Loans housing / Bank assets.
    - Colombia — Top 3 Positive: Total loans to private sector; Total loans to private sector; Loans commercial / Bank assets. Top 3 Negative: FX Volatility; ELMI + LC; Central bank rate-US T-Bill spread.
    - Mexico — Top 3 Positive: Interest rates consumer loans - MPR spread; Interest rates commercial loans - MPR spread; Loans commercial. Top 3 Negative: ELMI + LC; Interest rates commercial loans - MPR spread; Central bank rate-US T-Bill spread.
    - Peru — Top 3 Positive: Total loans to private sector; Loans commercial; Total bank assets. Top 3 Negative: Interest rates consumer loans - MPR spread; Loans consumers / Bank assets; Capital to assets ratio.
  - Timeline markers in tables: "October 2008" and "November 2016" appear in original table context.

*Source: wp17218 - 1. Variables in FCI by Country and related sections; canonical URL: https://www.imf.org/-/media/files/publications/wp/2017/wp17218.pdf*

### 1. Variables in FCI by Country .........................................................................................

### 1. Variables in FCI by Country

### Introduction and scope
- Purpose: develop financial conditions indices (FCIs) for the six large and most financially-integrated Latin American economies (LA6) that are comparable across countries.
- Definition used: financial conditions = exogenous changes in financial markets as opposed to changes that reflect the business cycle or monetary policy decisions.
- LA6 countries: Argentina, Brazil, Chile, Colombia, Mexico, and Peru.
- Uses of estimated FCIs:
  - Characterize financial conditions and underlying common drivers in the region.
  - Investigate the impact of financial conditions on economic activity and time-variation in that relationship.
  - Characterize financial spillovers within the region.

### Main findings (as reported)
- Estimated FCIs are influenced by:
  - a commodity cycle,
  - a global financial cycle,
  - country-specific episodes of financial distress.
- As of February 2017, financial conditions are estimated to remain favorable in most LA6 countries.
- Impact on GDP:
  - A one standard deviation tightening of financial conditions lowers half-year ahead GDP growth by about 0.05−0.3 percentage points depending on the country.
- Evidence of significant financial spillovers within the region.

### Methodology overview
- Estimation approach: Time-Varying Parameter Factor-Augmented Vector Autoregression (TVP-FAVAR), following Korobilis (2013) and Koop and Korobilis (2014).
- Distinctive methodological features:
  1. Broad array of financial variables:
     - FCIs are constructed from a broad set of financial indicators capturing credit conditions, leverage and quality of collateral, and intensity of perceived risk.
  2. Changing weights:
     - Loadings (weights) on financial variables are allowed to vary over time; variables may have varying or zero loadings in different periods.
  3. Time-varying transmission:
     - The method allows for time-varying intensity in the transmission of financial conditions to economic activity, accommodating evolving macro-financial relationships.
- Technical implementation notes:
  - The FCI is estimated using Koop and Korobilis’ 2014 code.
  - The method combines Primiceri’s (2005) time-varying parameter VAR with factor analysis for large datasets (FAVARs).
  - The approach relies on the Kalman filter and smoother and is described as simulation-free.
  - Rationale: TVP-FAVAR accommodates structural instabilities and changing relevance of financial variables over time.

### Structure of the analysis (paper organization)
- Section II: data and methodology to construct FCIs for LA6 countries.
- Section III: characterization of financial conditions over the past 15 years, underlying common drivers, and impact on economic activity.
- Section IV: estimation and discussion of financial spillovers within the region.
- Section V: conclusions.

*Source: wp17218 - 1. Variables in FCI by Country; canonical URL: https://www.imf.org/-/media/files/publications/wp/2017/wp17218.pdf*

### 3.      Isolating financial shocks. The method aims to measure exogenous financial

### 3.      Isolating financial shocks. The method aims to measure exogenous financial

### A. Variable selection
- Objective: measure exogenous financial shocks by purging FCI variables of developments originating outside the financial system (business cycle, monetary policy, exchange movements).
- Rationale: FCIs should capture pure unanticipated financial shocks (investors’ shifts in preferences towards liquidity and risk) rather than endogenous responses to macroeconomic conditions or monetary policy.
- Financial variables grouped into three broad categories:
  - Quantity developments:
    - Examples: loans by segment, domestic bond and equity issuance, measures of credit quality such as NPLs and expected probabilities of default for borrowers.
    - Economic channel: measure ease of access to finance; when credit/equity financing is difficult, consumption and investment may be negatively affected.
  - Collateral and leverage variables:
    - Examples: returns on stock indices, housing prices, price-to-earnings ratios, capital-to-assets ratios.
    - Economic channel: reflect balance sheet quality of lenders and borrowers, affecting supply and demand of funds (Bernanke, Gertler, and Gilchrist 1999; Gilchrist and Zakrajsek 2012).
  - Risk measures:
    - Examples: changes in CDS sovereign and EMBIG spreads, inter-bank, corporate and term spreads; volatility measures (equity returns and FX volatilities).
    - Economic channel: capture lenders’ tolerance of risk, thereby affecting supply of funds.
- Summary behavioral relationship: Declining credit volumes and collateral, and widening spreads, lead to tightening in financial conditions, and vice-versa.

### B. Data
- Coverage: LA6 FCIs constructed using mostly monthly data from early 2000 until February 2017.
- Frequency conversion: A few quarterly variables converted to monthly using cubic spline interpolation.
- Coverage characteristics: Cross-country coverage best for variables describing credit conditions; length of coverage varies considerably by country (see Table 1).
- Transformations and treatment:
  - Each variable transformed to achieve stationarity (if needed) following Brave and Butters (2011).
  - Missing data are replaced with zeros.
- Example variable list and transformations (as reported in source Table 1 — selection shown verbatim):
  - 1-Year EDF Banks (75th percentile) — LV — Credit
  - 1-Year EDF Corporates (75th percentile) — LV — Credit
  - Domestic bond issuance — LVMA — Credit
  - Domestic equity issuance — LVMA — Credit
  - Loans trade — DP12 — Credit
  - Loans commercial DP12 — Credit
  - Loans commercial / Bank assets — DL12 — Credit
  - Loans consumers — DP12 — Credit
  - Loans consumers / Bank assets — DL12 — Credit
  - Loans housing — DP12 — Credit
  - Loans housing / Bank assets — DL12 — Credit
  - Nonperforming loans to bank credit ratio — LV — Credit
  - Total bank assets — LVMA — Credit
  - Total loans to private sector — DP12 — Credit
  - Capital to assets ratio — LV — Leverage
  - Financial firms Datastream return index — DLN — Leverage
  - Financials to Stock Total Market — LVMA — Leverage
  - Housing prices — DLN — Leverage
  - MMMF/ Bond fund assets — LVMA — Leverage
  - Pension fund assets (% GDP) — DLN — Leverage
  - Stock market general index — DLN — Leverage
  - Central bank rate-US T-Bill spread — LV — Risk
  - CDS Sovereign 2YR — LV — Risk
  - EMBIG — LV — Risk
  - FX Volatility — LV — Risk
  - Interbank - MPR spread — LV — Risk
  - Interest rates commercial loans - MPR spread — LV — Risk
  - Interest rates consumer loans - MPR spread — LV — Risk
  - Interest rates housing loans 3 years or more - MPR spread — LV — Risk
  - Interest rates trade loans - US 3-Month T-bill rate spread — LV — Risk
  - Net loan spread financial system, 30 to 89 days — DLV — Risk
  - Net loan spread financial system, 90 days to 1 year — DLV — Risk
  - Volatility of stock market general index — LV — Risk
- Data sources listed: Banco Central de Brasil; Banco Central de Chile; Banco Central de Colombia; Banco Central de la República Argentina; Banco de México; Banco Central de Reserva del Perú; Bloomberg LP; Comisión Nacional Bancaria y de Valores México; FactSet; Moody's CreditEdge+; Superintendencia de Bancos e Instituciones Financieras de Chile; Superintendencia de Banca, Seguros y AFP del Perú; Superintendencia de Valores y Seguros de Chile; Thomson-Reuters Datastream.
- Transformation legend (as per Table 1 notes):
  - CDS = Credit Default Spread; EDF = Expected Default Frequency; ELMI+LC = Emerging Local Markets Index, local-currency denominated; EMBIG = Emerging Market Bond Index Global; MMMF = Money Market Mutual Fund; MPR = Monetary Policy Rate.
  - Transformations: DL = first difference; DLN = first log-difference; DP12 = 12-month difference; LV = level; LVMA = ratio of level to 12-month moving average.
  - Groups: Credit — mainly quantity developments; Collateral/leverage — measures of balance-sheet health; Risk — spreads and volatilities.

### C. Empirical model
- Model specification (as described in text equation (1)):
  - X is a vector of financial variables.
  - Y is a vector of macroeconomic variables (inflation, a monthly proxy for growth, the real effective exchange rate, and the monetary policy rate).
  - f is an unobservable (first) factor (the financial conditions index).
  - ε and u are uncorrelated (but heteroskedastic) error terms.
  - Coefficients λ_y and λ_f are factor loadings tracking effects of macroeconomic and financial conditions on each financial variable.
- Two-equation interpretation:
  - First equation: extraction of latent indicator f that, after removing contemporaneous macroeconomic effects, summarizes a large set of financial variables.
  - Second equation: models dynamic relationship between financial and macroeconomic conditions; with identifying assumptions, quantifies effects of financial conditions on macroeconomy and vice-versa.
- Identification for GDP effect: use of a Cholesky decomposition (ordering implied in later sections).
- Normalization and stationarity:
  - The FCI is normalized to have a mean of zero and standard deviation of one over the observation period (2000–2016).
  - Zero indicates historical average financial conditions; values above (below) zero indicate “tight” (“loose”) conditions relative to historical average.
  - Variables are first transformed to induce stationarity as reported in Table 1.

### Key methodological notes
- Purpose of purging: unpurged FCIs cannot indicate whether tight/loose conditions are endogenous to macro conditions/monetary policy or reflect autonomous financial environment changes.
- Literature references: Brave and Butters (2011 and 2012) for variable grouping; Bernanke, Gertler, and Gilchrist 1999; Gilchrist and Zakrajsek 2012; Korobilis (2013); Koop and Korobilis (2014) (in other sections).

### Empirical implementation details
- Estimation sample: mostly monthly early 2000–February 2017.
- Missing data handling: replaced with zeros.
- Frequency conversion: cubic spline interpolation for quarterly variables.

---

### III. Financial conditions in LA6 over the past 15 years — Characterizing financial conditions
- Global events dominate FCI dynamics across LA6:
  - 2007-9 financial crisis: sharpest spike in financial conditions across region since 2000 (barring Argentina).
  - European sovereign crisis (2011−12) and the taper tantrum episode of early 2013: strongly felt in Brazil, Chile, and Peru.
  - U.S. elections impact: muted once FCIs purged of foreign exchange rate and monetary policy rate movements.
- Country-specific episodes captured by FCIs:
  - Argentina: collapse of currency board and ensuing crisis.
  - Colombia: tight credit conditions in early 2000s (post mortgage crisis of late 1990s).
  - Peru: period of political uncertainty and economic slump of 2001.
- As of February 2017 financial conditions:
  - Loose in Argentina and Chile.
  - Broadly neutral in Colombia, Mexico, and Peru.
  - Somewhat tight in Brazil.
- Annual/period changes (2016 example):
  - 2016 loosening drivers:
    - Argentina: easing credit and risk in about equal proportions (growing money market liquidity and commercial loans; reduced stock market and FX volatility).
    - Chile: increasing credit and financial firms’ capital (growing bank assets and total loans to private sector; higher financial firms’ valuation).
    - Peru: easing credit (growing bank assets and loans to private sector).
  - 2016 tightening drivers:
    - Brazil: tighter credit and declining collateral (declining total bank assets and loans to private sector; increasing NPLs; declining financial firms’ valuation).
    - Mexico: marginal credit deterioration (declining money market liquidity and loans to private sector, particularly consumer loans).
- Relationship to cyclical conditions (2013−17):
  - In most countries, FCI changes are close to what cyclical macroeconomic conditions would warrant.
  - Exceptions noted:
    - Deteriorated more than warranted: Brazil 2014, Colombia 2014, Argentina 2015.
    - Deteriorated less than warranted: Brazil 2015.
    - Improved beyond warranted: Argentina 2014, Brazil 2017.

### III. Financial conditions — Common factors and variance decomposition
- Principal components analysis of the six country FCIs:
  - Two latent factors summarize up to 70 percent of the variance structure across region.
  - First common factor: accounts for about 43 percent of variance; highly correlated with world commodity prices (correlation coefficient with commodity prices is -0.73).
  - Second common factor: explains about 26 percent of variance; closely linked to the VIX (correlation coefficient with the VIX is 0.38); regarded as proxy for a global financial cycle.
- Country correlations with principal components (as reported):
  - First principal component correlations (PC1): Argentina 0.73; Brazil 0.62; Chile -0.74; Colombia 0.62; Mexico 0.54; Peru -0.64 (note signs and magnitudes as presented).
  - Second principal component correlations (PC2): Argentina 0.14; Brazil -0.37; Chile 0.60; Colombia 0.45; Mexico 0.60; Peru 0.71.
- Interpretation:
  - Commodity-price-driven factor closely correlated with financial conditions in all countries; highest absolute correlation with Chile and Argentina; lowest with Mexico.
  - Global financial factor (VIX-driven) strongly linked to Chile, Mexico, and Peru (higher financial integration) and less to Argentina, Brazil, and Colombia (lower financial integration).
  - Financial openness reference: Peru, Chile, and Mexico had Chinn-Ito index values of 1, 0.7, and 0.7 in 2014, while Argentina, Brazil, and Colombia had 0, 0.4, and 0.4 respectively.

### III. B. Impact of financial conditions on activity
- Identification and method:
  - Use Cholesky decomposition with ordering: inflation, GDP growth, REER, monetary policy rate, FCI.
  - For each month, estimated time-varying VAR coefficients and factor loadings used to build impulse response functions for GDP growth and other variables.
- Key quantitative findings (using November 2016 loadings):
  - A one standard deviation tightening of financial conditions lowers half-year-ahead GDP growth by:
    - between 0.05−0.1 percentage points for Mexico, Argentina, Chile, and Peru.
    - between 0.25−0.3 percentage points for Brazil and Colombia.
  - Timing of trough in activity:
    - Fifth month: Argentina and Chile.
    - Tenth month: Colombia.
  - Comparative interpretation:
    - A one standard deviation shock suffices to generate a ¼ percentage point fall in GDP growth in Brazil and Colombia.
    - Two standard deviations would be needed in Chile and Argentina to trigger a 0.25 percentage point fall.
    - Mexico and Peru would need around 4 ½ and 6 ½ standard deviation shocks respectively to trigger a 0.25 percentage point fall in GDP growth.
- Time variation:
  - Impact of financial shocks on economic activity varies over time.
  - During periods of financial stress (e.g., 2007-9 financial crisis), the effect of financial tightening on economic activity is between two and three times as high as during more favorable conditions (example: Chile lost output after 24 months of 2.4 percent of GDP in October 2008 vs 1.5 percent of GDP in November 2016 from a one-standard deviation FCI shock).

### III. C. Model comparisons
- Model variants compared:
  - TVP-FAVAR (time-varying parameters) — main model.
  - PCA-FAVAR (principal components-based, Stock and Watson 2002; Bernanke, Boivin, and Eliasz 2005) — simpler alternative.
  - KF-FAVAR (Kalman-filter based, Doz, Giannone, and Reichlin 2011) — simpler alternative.
- Empirical observations:
  - For LA6, PCA-FAVAR and KF-FAVAR closely track TVP-FAVAR.
  - Contributions of each variable to the FCI fairly stable over time; profile driven primarily by changes in underlying financial variables rather than time-varying weights.
  - Differences:
    - TVP-FAVAR at end of sample points to somewhat looser financial conditions in Argentina and Brazil, and slightly tighter in Mexico, relative to alternatives — indicating potential importance of time variation.
    - Colombia during the GFC: PCA- and KF-FAVAR point to sharper deterioration than TVP-FAVAR.
  - TVP-FAVAR advantage: allows measured effect of financial conditions on economic activity to change over time, potentially improving forecast accuracy.

### IV. Regional financial interconnectedness — approach
- Objective: assess importance of global, regional, and domestic shocks in shaping domestic financial conditions and quantify spillovers across LA6 (2014−16).
- Method:
  - Apply Diebold and Yilmaz’s (2014) spillovers framework.
  - Estimate a VAR of monthly FCIs incorporating global control variables (world commodity prices and the VIX) to remove co-movement due to common factors.
  - Model specification (text equation (2)):
    - Y: vector of FCIs for LA6.
    - X: world commodity prices and VIX.
    - A(L): lag polynomial with order chosen by BIC.
    - Generalized forecast-error variance decomposition (GVD) using Pesaran and Shin (1998) to identify structural shocks to FCIs.
  - Focus: 3-month ahead forecast error decomposition.

### IV. Regional financial interconnectedness — stylized facts (over 2014−16)
- Domestic vs regional drivers:
  - Shocks originating from local financial sector explain largest fraction of FCI variance across LA6.
  - Brazil and Mexico: domestic shocks explain about 87 and 82 percent of the variance respectively.
  - Remaining LA6 countries: regional shocks explain between 33 percent (Colombia) and 43 percent (Peru) of FCI variability.
- Outward spillover potential:
  - Largest spillover potential: Brazil and Colombia, followed by Mexico, Argentina, and Peru.
  - Chile transmitted comparatively smaller FCI spillovers to the region.
- Inward spillovers:
  - Strongest receivers: Peru, Chile, and Argentina (fractions of variance decomposition of each country’s FCI explained by rest of region: about 44, 38, 37, and 33 percent respectively — note ordering in source text).
  - Brazil and Mexico: inward spillovers comparatively smaller (about 13 and 18 percent respectively).
- Notable bilateral links:
  - Argentina strongly affected by Mexico.
  - Chile significantly affected by Argentina.
  - Feedback loop between Colombia and Peru.

### V. Conclusion — summary of findings
- Methodology: development of cross-country comparable FCIs for LA6 following Korobilis (2013) and Koop and Korobilis (2014).
- Key findings:
  - Estimated FCIs influenced by a commodity cycle and a global financial cycle; they track country-specific episodes of financial distress.
  - By early 2017, financial conditions remained favorable in most LA6 countries relative to historical standards.
  - Quantified effect of financial shocks on output:
    - A one standard deviation tightening of financial conditions lowers half-year ahead GDP growth by about 0.1 percentage points (0.2− ¼ percentage points in periods of financial stress).
  - Evidence of significant financial spillovers within the region.

### Going forward — research priorities and open questions (policy-relevant research agenda)
- Improve understanding of drivers of correlations between global factors (commodity prices, VIX) and each country’s FCI; candidate explanations include:
  - Nature of reliance on commodity exports.
  - Degree of financial openness.
- Investigate time-varying nature of GDP response to FCI shocks:
  - Which financial frictions drive changing transmission of financial shocks to real economy?
- Further scrutiny of regional financial spillovers:
  - Do spillovers follow observed patterns of trade and cross-border investment flows?
  - Note: because FCIs are purged of macroeconomic drivers, real economic linkages (trade ties) should not explain influence of regional financial conditions in this framework.

*IMF staff (extracted from wp17218 — section 3 and subsequent sections of the source document).*

### Appendix A. Data Sources

### Appendix A. Data Sources

### Data sources and variable coverage
- Presents sources used to construct the FCI for each LA6 economy (Table A.1).
- Variables grouped into three categories: credit, collateral/leverage, and risk (see note 2/).
- Transformations (note 1/):
  - DL = first difference
  - DLN = first log-difference
  - DP12 = 12-month difference
  - LV = level
  - LVMA = ratio of level to 12-month moving average
- Institutional and acronym notes (verbatim extracts):
  - Abrapp = Associ ação Brasi l ei ra das Enti dades Fechadas de Previ dênci a Compl ementar
  - BanRep = Banco de l a Repúbl i ca
  - Banxi co = Banco de Méxi co
  - BCB = Banco Central do Brasi l
  - BCCH = Banco Central de Chi l e
  - BCRA = Banco Central de l a Repúbl i ca Argentina
  - BCRP = Banco Central de Reserva del Perú
  - CDS = Credit Default Swap
  - EDF = Expected Default Frequency
  - ELMI+LC = Emerging Local Markets Index, local-currency denominated
  - EMBIG = Emerging Market Bond Index Global
  - FSI = Financial Soundness Indicators
  - IFS = International Financial Statistics
  - MMMF = Money Market Mutual Fund
  - MPR = Monetary Policy Rate
  - SBIF = Superintendencia de Bancos e Instituciones Financieras
  - SBS = Superintendenc i a de Ba nc a, Seguros y AFP
  - SVS = Superintendencia de Valores y Seguros

### Example variable list structure (as presented)
- Variables include (examples from Table A.1):
  - 1-Year EDF Banks (75th percentile) — Transformation: LV — Group: Credit
  - 1-Year EDF Corporates (75th percentile) — Transformation: LV — Group: Credit
  - Domestic bond issuance — Transformation: LVMA — Group: Credit
  - Domestic equity issuance — Transformation: LVMA — Group: Credit
  - ELMI + LC — Transformation: DLN — Group: Credit
  - Loans trade, Loans commercial, Loans consumers, Loans housing — Transformations: DP12 or DL12 — Group: Credit
  - Nonperforming loans to bank credit ratio — Transformation: LV — Group: Credit
  - Total bank assets — Transformation: LVMA — Group: Credit
  - Capital to assets ratio — Transformation: LV — Group: Leverage
  - Financial firms Datastream return index — Transformation: DLN — Group: Leverage
  - Financials to Stock Total Market — Transformation: LVMA — Group: Leverage
  - Housing prices — Transformation: DLN — Group: Leverage
  - Pension fund assets (% GDP) — Transformation: DLN — Group: Leverage
  - Stock market general index — Transformation: DLN — Group: Leverage
  - Central bank rate - US T-Bill spread — Transformation: LV — Group: Risk
  - CDS Sovereign 2YR — Transformation: LV — Group: Risk
  - EMBIG — Transformation: LV — Group: Risk
  - FX Volatility — Transformation: LV — Group: Risk
  - Interbank - MPR spread; interest rate spreads for commercial, consumer, housing, trade loans — Transformation: LV — Group: Risk
  - Net loan spread financial system, 30 to 89 days and 90 days to 1 year — Transformation: DLV — Group: Risk
  - Volatility of stock market general index — Transformation: LV — Group: Risk

### Appendix B. Contributions to FCI

### Method to retrieve group contributions
- FCI summarizes information from variables grouped in three categories: risk, credit, and leverage.
- To retrieve contributions of each group to the change in the FCI over a period, a linear projection method is employed.
- Procedure:
  - Estimate the FCI using the system of equations (1).
  - Solve the first equation in (1) for the FCI (f) using a linear projection; this yields f as a function of X and y.
- Formal expression provided (verbatim from source):
  - 1ˆˆ ˆˆ , fffyfXy     
  - The k-th element of 1ˆˆ ˆ fff    yields the gross contribution of the kth financial variable in X to the FCI f and 1ˆˆ ˆˆ fffy y    the netting of the contribution of each macroeconomic variable in y to the financial variables X.
  - That is, 1ˆˆ ˆ fff    yields the change in f that can be attributed to changes in the variables in X when the variables in y do not change.
- The difference between the actual change in f and the estimated contributions from 1ˆˆ ˆ fff X     is given by 1ˆˆ ˆˆ fffy y    .
- Interpretation note: "This will typically happen when the macroeconomic conditions are above average."

### Appendix C. Time-Varying Loadings and Contributions

### Time-varying parameter FAVAR features and implications
- TVP-FAVAR methodology allows loadings to change each period.
- Rationale: previous literature shows time-variation in loadings and covariances of factor models using financial and macroeconomic data (Banerjee and others, 2008; Koop and Korobilis, 2014).
- Advantages:
  - Time-varying weights yield improved forecast accuracy of TVP-FAVARs relative to constant-parameter alternatives such as PC- and KF-FAVARs.
  - Time-varying weights imply that the contribution (or score) of each variable to the FCI may vary over time.
- For inspection of variability of variables’ weights throughout the sample, see Table A.2.

### Table A.2 — Factor loadings volatility by country (notes and examples)
- Sources: IMF staff.
- Note: Volatilities shown are the standard deviation of the score or variable contribution to the FCI by variable during the whole period (see Appendix B). Blanks indicate that the variable does not enter the computation of the FCI.
- Selected volatility entries (standard deviation of the score by variable across countries as shown in Table A.2):
  - ELMI + LC: 0.0050 (ARG), 0.0190 (BRA), 0.0070 (CHL), 0.0060 (COL), 0.0080 (MEX), 0.000 (PER)
  - Total bank assets: 0.0140 (ARG), 0.0300 (BRA), 0.0030 (CHL), 0.0020 (COL), 0.0080 (MEX), 0.003 (PER)
  - EMBIG: 0.0060 (ARG), 0.0850 (BRA), 0.0110 (CHL), 0.0340 (COL), 0.0210 (MEX), 0.009 (PER)
  - Central bank rate - US T-Bill spread: 0.0000 (ARG), 0.0250 (BRA), 0.0030 (CHL), 0.0170 (COL), 0.0100 (MEX), 0.004 (PER)
  - Stock market general index: 0.0040 (ARG), 0.0240 (BRA), 0.0030 (CHL), 0.0070 (COL), 0.0080 (MEX), 0.003 (PER)
  - Capital to assets ratio: 0.0010 (ARG), 0.0310 (BRA), 0.0040 (CHL), 0.0030 (COL), 0.001 (MEX)
  - Nonperforming loans to bank credit ratio: 0.0020 (ARG), 0.0250 (BRA), 0.0040 (CHL), 0.0060 (COL), 0.003 (MEX)
  - Loans consumers: 0.0030 (ARG), 0.0070 (BRA), 0.0080 (CHL), 0.0120 (COL), 0.0080 (MEX), 0.006 (PER)
  - Housing prices: 0.0030 (ARG), 0.0010 (BRA), 0.0140 (CHL), 0.0000 (COL), 0.000 (MEX)
  - FX Volatility: 0.0080 (ARG), 0.0090 (BRA), 0.0050 (CHL), 0.0030 (COL), 0.0020 (MEX), 0.000 (PER)
- Blanks in the original table indicate non-usage of that variable in a country's FCI computation.

### Table A.3 — Stability of ranking of variable contributions
- Finding: Rankings of variables by contribution for each country’s FCI are fairly stable over time.
- Table A.3 lists Top 3 Positive Contributions and Top 3 Negative Contributions by country (selected entries reproduced verbatim):
  - Argentina — Top 3 Positive Contributions:
    - Total loans to private sector
    - Loans commercial
    - Loans trade
  - Argentina — Top 3 Negative Contributions:
    - ELMI + LC
    - Interest rates commercial loans - MPR spread
    - Volatility of stock market general index
  - Brazil — Top 3 Positive Contributions:
    - Central bank rate-US T-Bill spread
    - EMBIG
    - Total bank assets
  - Brazil — Top 3 Negative Contributions:
    - Total bank assets
    - Total loans to private sector
    - Total bank assets
  - Chile — Top 3 Positive Contributions:
    - Total loans to private sector
    - Total loans to private sector
    - Interest rates housing loans 3 years or more - MPR spread
  - Chile — Top 3 Negative Contributions:
    - Capital to assets ratio
    - Loans housing / Bank assets
    - Loans housing / Bank assets
  - Colombia — Top 3 Positive Contributions:
    - Total loans to private sector
    - Total loans to private sector
    - Loans commercial / Bank assets
  - Colombia — Top 3 Negative Contributions:
    - FX Volatility
    - ELMI + LC
    - Central bank rate-US T-Bill spread
  - Mexico — Top 3 Positive Contributions:
    - Interest rates consumer loans - MPR spread
    - Interest rates commercial loans - MPR spread
    - Loans commercial
  - Mexico — Top 3 Negative Contributions:
    - ELMI + LC
    - Interest rates commercial loans - MPR spread
    - Central bank rate-US T-Bill spread
  - Peru — Top 3 Positive Contributions:
    - Total loans to private sector
    - Loans commercial
    - Total bank assets
  - Peru — Top 3 Negative Contributions:
    - Interest rates consumer loans - MPR spread
    - Loans consumers / Bank assets
    - Capital to assets ratio
- Timeline markers in the table: "October 2008" and "November 2016" appear in the original table context.

### Methodological references and supporting literature
- Appendix cites literature motivating TVP-FAVAR use and methods for factor and FCI construction, including (verbatim list from the source): Adrian T. and H. S. Shin, 2008; Banerjee, A., M., Marcellino, and I. Masten, 2008; Bernanke, B.S., J. Boivin, and P. Eliasz, 2005; Bernanke, B. S., M. Gertler, and S. Gilchrist, 1999; Brave, S. and A.R. Butters, 2011 and 2012; Chinn, Menzie D. and Hiro Ito, 2006; Cogley T. and T. J. Sargent 2001; Diebold, F. X. and K. Yilmaz, 2014; Doz, C., D. Giannone, and L. Reichlin, 2011; Galvao, A.B. and M.T. Owyang, 2014; Gilchrist, S. and E. Zakrajsek, 2012; Hatzius, J., P. Frederic, S. Mishkin, K. L. Schoenholtz, and M.W. Watson, 2010; International Monetary Fund, 2016; International Monetary Fund, 2017; Korobilis, Dimitris, 2013; Koop, G. and D. Korobilis, 2014; Matheson, T., 2011; Pesaran, M.H. and Y. Shin, 1998; Primiceri, G. E., 2005; Sims C. A., J. H. Stock, and M. W. Watson, 1993; Stock, J. H, and M. W. Watson, 2002.

*Source: wp17218 - Appendix A. Data Sources (IMF staff).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17218.pdf_