## 1. External Shocks and Responses of Domestic Variables

## Source details

**Canonical URL:** [1. External Shocks and Responses of Domestic Variables](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15269.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15269.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15269.pdf.json)

---

### I. Introduction — scope and motivation
- Emerging market (EM) vulnerabilities to external headwinds and intrinsic domestic fundamentals are central to policy discussions after the global financial crisis (GFC).
- Spillovers from advanced economies’ policies, combined with uncertainty and weak domestic fundamentals, affect EMs with large trade links and high exposure to the global financial cycle.
- The paper proposes a methodology to quantify the impact of external and domestic macro-financial impulses on the dynamics of a home economy, with both forward-looking and historical analyses.

### II. Methodology overview
- Model class:
  - Mixed-Frequency Bayesian VAR (MFBVAR) with two blocks: home and foreign.
  - Mixed-frequency feature combines monthly and quarterly information to compute conditional forecasts (nowcasts).
  - Model parsimony: limited to key explanatory variables to focus on main channels of propagation and spillover.
- Block composition:
  - Home block: consists of five variables.
  - Foreign block: consists of three variables.
  - Foreign-block variables explicitly mentioned: global oil prices, US corporate risk spread, and US industrial production (IP).
- Application target: GDP nowcasting (note: for other targets, e.g., inflation forecasting, different indicators would be required).

### III. Application case: Turkey (home economy) and practitioner guidance
- Rationale for selection:
  - Turkey is a very open economy exporting to major advanced economies, especially Europe.
  - Completely dependent on the importation of primary commodities.
  - Major trade links with many countries in its region.
  - Financial factors such as risk taking channel and capital flow are important.
  - These characteristics make Turkey an ideal test case; toolbox can be used for other small open economies.
- Practitioner guidance:
  - Identify relevant external variables by assessing main trading and financial partners, degree of trade and financial openness, and reliance on oil and other commodities.
  - Choose leading indicators in the home economy that convey information on investment, trade, and financial activities.
  - Application dictates variable choice.
- Footnote/example (verbatim): "For example in 2013 Iraq was the Turkey’s 2nd biggest export market."

### IV. Model specification and estimation (exact design choices)
- Model: mixed-frequency structural Bayesian VAR with two blocks (home and foreign).
- Lag length: one lag only, to avoid over-parameterization.
- Sample period: 2003Q2 to 2014Q1.
- Data treatment:
  - All variables demeaned and detrended using the Hodrick-Prescott (HP) filter.
  - Series converted to cyclical components by applying the HP filter to 100 times the natural log of data.
  - HP filter applied to level of domestic economy interest rates and US financial risk spread data.
- Stationarity: all variables are stationary and have zero mean after detrending.
- Structural form (closed-form when demeaned): Y(t) = A(1) Y(t-1) + ε(t), with ε(t) ~ i.i.d. (0, Σ).
- Identification assumption: block exogeneity — no spillover from domestic economy to external block; external block affects domestic economy.
- Estimation:
  - Standard OLS estimates imposing block exogeneity.
  - Cholesky decomposition used to estimate an orthogonal covariance matrix of residuals.
- Forecast evaluation: multi-step forecasting error approach (recursively computing h-step ahead forecasts) following Marcellino, Stock, and Watson (2006).
- Missing observations / mixed frequency: Kalman filter used in conditional forecasting to account for missing observations.

### V. Data and variables (exact variable lists and definitions)
- Domestic block variables (monthly except GDP):
  - Real GDP (quarterly)
  - Industrial production
  - Three-month weighted average interest rates (short-term)
  - Imports (foreign trade: total merchandise imports, c.i.f.)
  - Loans to the private sector (Deposit Money Banks Loans: Private Sector)
- Foreign block variables:
  - Price of oil (World Crude Oil {Petro}: Brent, WTI, Dubai Fateh Spot Price Avg. (US$/Bbl))
  - US industrial production excluding construction (US IP)
  - US risk spread (defined as the spread between Moody’s BAA and AAA corporate bond rates)
- Seasonal adjustment: variables seasonally adjusted by official sources and authors’ computations (X12 ARIMA).
- Rationale for chosen foreign variables:
  - Global oil price and US IP reflect commodity cycle and demand pressure.
  - US risk spread signals foreign investors’ risk-taking appetite.
  - For Turkey, US risk spread suited nowcasting application better than alternative indicators (EMBI, VIX, FSI).

### VI. Identification, ordering, and timing
- In the domestic block, slow-moving variables (e.g., GDP) are ordered first so they can contemporaneously affect fast-moving variables.
- GDP is presumed uncorrelated with other (non-GDP) domestic shock processes.
- By construction, FEVDs of external variables are explained solely by shocks originating in the foreign block (no domestic → foreign spillovers).

### VII. Impulse response analysis (key quantitative result)
- Comparing impulse responses of GDP growth to external shocks:
  - Upon the impact of a one-standard-deviation oil shock (spread shock and US IP), output drops 1.3 (0.7 and 1) percentage points relative to its steady state.
- Comovements and policy reaction:
  - IP, credit, and imports comove with output growth.
  - Interest rate increases in response to adverse supply shocks (reflecting output-inflation stabilization tradeoff).
  - Interest rate decreases in response to demand shocks.

### VIII. Forecast Error Variance Decompositions (FEVD) — aggregated findings (exact statements)
- FEVD over 10 quarters: aggregate contributions reported as “home” (domestic shocks aggregated) vs “foreign” (external shocks aggregated).
- Key observations:
  - The impact of domestic shocks is strongest in the first quarter.
  - More than 60% of output FEVD is driven by external factors.
  - Loans to the private sector: largely determined by domestic factors in the short term; medium-to-long-term share of external shocks increases (attributed mainly to capital flows converted to credit).
  - Short-term interest rates: explained mainly by domestic factors (monetary policy response to business cycle shocks).
- Interpretation: large share of volatility in domestic indicators is explained by external factors — consistent with Turkey’s increasing openness after 2001 and capital flow dynamics.

### IX. Historical Variance Decompositions (HVD) — selected findings (periods and directional roles)
- GDP dynamic: large persistence between 2011 and 2013.
- External channels (real and financial) were very important during 2012-2013 turmoil.
  - Drivers cited for turmoil: Euro Area double-dip recession, US fiscal cliff uncertainty, Turkey’s political unrest, and other factors.
- Role of individual foreign variables between 2012 and mid-2013:
  - US IP: playing a negative but small role.
  - US risk spread: playing a large and positive role in explaining the domestic business cycle.
  - The US IP and US risk spread show opposite contributions, highlighting the countercyclical nature of the risk spread.
- Domestic factors:
  - Minor role in explaining the dynamics of short-term rates.
  - 2012 saw a positive and significant contribution from risk spread which diminished later.
  - Credit comoves with GDP but domestic explanatory role decreases toward end of sample.

### X. Counterfactual simulations (three scenarios and main result)
- Simulated economies computed by zeroing out selected residuals:
  - (i) Baseline: all shocks included.
  - (ii) No domestic shocks: exclude domestic shocks, external shocks still operate.
  - (iii) No external shocks: exclude external shocks, domestic shocks still operate.
- Key result:
  - During the 2009 recession and 2010 rebound, external shocks help explain the downturn and rebound better.
  - A model excluding external shocks predicts a shorter-duration 2009 downturn and a too-speedy recovery.
  - Interpretation: external shocks are important to capture joint dynamics of downturns and rebounds.

### XI. Conditional forecast / Nowcasting methodology and implementation details
- Nowcasting objective: compute current-quarter forecast of GDP using mixed-frequency information flow; revise forecasts as monthly data arrive.
- Conditional forecast notation: nowcast of quarterly GDP at period t conditional on information up to month M+1 denoted E_t|M+1[Δ_Q ...]. Forecasts updated monthly as new high-frequency data are released.
- Evaluation metric: one-quarter ahead root-mean-square errors (RMSEs) computed over different horizons and vintages.
- RMSE computation: RMSE formula employed as specified in the text (equation (9)); vintages include unconditional and conditional on month 1, month 2, and month 3 data availability within the quarter. Kalman filter used for missing observations.
- Nowcasting improves as new monthly information becomes available throughout the quarter.
- Conditional forecasting implementation specifics:
  - Conditional on M1 includes information of first month of quarter for all monthly variables.
  - Conditional on M2 includes information of second month of quarter only for short-term interest rate and external variables in the model.

### XII. Forecast performance — out-of-sample comparisons (exact figures)
- Out-of-sample forecast errors (difference between forecast and actual), four-steps-ahead, comparison with WEO:
  - Forecast Turkey’s output Growth — Forecast Errors:
    - WEO: 2009 = 4.0; 2010 = 4.2; 2011 = -0.2; 2012 = 0.6
    - Our Model: 2009 = -2.3; 2010 = 2.3; 2011 = 0.4; 2012 = -0.1
  - Note example: WEO’s forecast of 2009 GDP growth was 4% above actual; the VAR-based model predicted the 2009 recession with a somewhat bigger trough.
- Out-of-sample RMSEs (exact figures):
  - Forecast Turkey’s output Growth:
    - WEO: 2009-2012 = 2.9; 2010-2012 = 2.4
    - Our Model: 2009-2012 = 1.6; 2010-2012 = 1.3
- RMSE vintage summary: Figure 5 (described) shows unconditional forecasts (blue bars) and nowcasts conditional on month 1, month 2, month 3 information (red, green, yellow as described in text).

### XIII. Cross-correlations and stylized facts (selected exact correlation sequences from Table 4)
- Cross-correlation of Output GAP with leads/lags (t-3 t-2 t-1 t t+1 t+2 t+3):
  - Output GAP: 0.33 0.56 0.82 1 0.82 0.56 0.33
  - Domestic IP: 0.3 0.55 0.79 0.94 0.76 0.51 0.22
  - Short-term Rate: 0.5 0.4 0.2 1 -0.12 -0.29 -0.38 -0.38
  - Import: 0.39 0.64 0.83 0.89 0.62 0.23 -0.03
  - Credits: 0.69 0.83 0.87 0.75 0.52 0.31 0.15
  - Nominal POIL: 0.26 0.51 0.7 0.79 0.56 0.13 -0.13
  - US IP: 0.62 0.8 0.9 0.85 0.67 0.44 0.19
  - US Financial Riskness: 0.38 0.05 -0.35 -0.68 -0.77 -0.61 -0.52

### XV. Conclusion — key messages and practical guidance
- The paper proposes a nowcasting toolbox for output growth that explicitly incorporates global factors.
- Impulse responses, variance decompositions, and counterfactual simulations show external shocks, together with domestic shocks, can explain domestic variable dynamics.
- The empirical setup parsimoniously combines domestic and external leading indicators, and is suitable for small open economies with large trade and financial links to advanced economies.
- Conditional forecast results indicate improved forecasting power by exploiting mixed-frequency data and measuring impulses from external factors.
- Practical guidance: while the VAR framework with domestic leading indicators and global variables provides a parsimonious forecasting framework with good accuracy, practitioners should not rely on a single model; combinations of forecasting techniques are recommended.

*Source: IMF Working Paper _wp15269 (excerpt provided).*

### 1.  External Shocks and Responses of Domestic Variables .................................................... 10

### 1.  External Shocks and Responses of Domestic Variables

### I. Introduction — scope and motivation
- Emerging market (EM) vulnerabilities to external headwinds and intrinsic domestic fundamentals are central to policy discussions after the global financial crisis (GFC).
- Spillovers from advanced economies’ policies, combined with uncertainty and weak domestic fundamentals, affect EMs with large trade links and high exposure to the global financial cycle.
- The paper proposes a methodology to quantify the impact of external and domestic macro-financial impulses on the dynamics of a home economy, with both forward-looking and historical analyses.

### II. Methodology overview
- Model class: Mixed-Frequency Bayesian VAR (MFBVAR) with two blocks: home and foreign.
- Mixed-frequency feature: combines monthly and quarterly information to compute conditional forecasts (nowcasts).
- Model parsimony: limited to key explanatory variables to focus on main channels of propagation and spillover.
- Home block: consists of five variables.
- Foreign block: consists of three variables.
- Foreign-block variables explicitly mentioned: global oil prices, US corporate risk spread, and US industrial production (IP).
- Model application target: GDP nowcasting (note: for other targets, e.g., inflation forecasting, different indicators would be required).

### III. Application case: Turkey (home economy)
- Rationale for selection:
  - Turkey is a very open economy exporting to major advanced economies, especially Europe.
  - Completely dependent on the importation of primary commodities.
  - Major trade links with many countries in its region.
  - Financial factors such as risk taking channel and capital flow are important.
  - These characteristics make Turkey an ideal test case; toolbox can be used for other small open economies.
- Practitioner guidance for applying the toolbox to other economies:
  - Identify relevant external variables by assessing main trading and financial partners, degree of trade and financial openness, and reliance on oil and other commodities.
  - Choose leading indicators in the home economy that convey information on investment, trade, and financial activities.
  - Application dictates variable choice.

- Footnote/example from source:
  - "For example in 2013 Iraq was the Turkey’s 2nd biggest export market."

### IV. Key findings and empirical results (as reported)
- External shocks can account for more than 50% of the GDP growth dynamic in Turkey.
- Historical variance decompositions indicate:
  - A decrease in the risk-taking appetite in the US provides positive impulses for the emerging market economy’s output growth.
  - Strong US industrial production (IP) provides positive impulses for the emerging market economy’s output growth.
- Counterfactual simulations show:
  - External shocks play a key role in explaining the recession and subsequent rebound in Turkey.
- Nowcasting insights:
  - Exploiting the mixed-frequency nature of the model, nowcasting can quantify the direction of macroeconomic developments as high-frequency data become available.
  - The proposed model improves upon the benchmark naïve model.

### V. Structure of the paper (as organized in source)
- Section 2: model specifications and database.
- Sections 3–6: results — impulse response functions, variance decomposition analysis, counterfactual simulations, and conditional forecast results.
- Section 7: conclusions.
- Sections 8–11: References and Appendices.

*Source: IMF working paper chapter "1.  External Shocks and Responses of Domestic Variables"*

### 1. We use a mixed-frequency Bayesian Vector Autoregression (VAR) model for our

### _wp15269 - 1. We use a mixed-frequency Bayesian Vector Autoregression (VAR) model for our

### Model specification and estimation
- Model: mixed-frequency structural Bayesian VAR with two blocks (home and foreign).
- Lag length: one lag only, to avoid over-parameterization.
- Sample period: 2003Q2 to 2014Q1.
- Data treatment: all variables demeaned and detrended using the Hodrick-Prescott (HP) filter; series converted to cyclical components by applying the HP filter to 100 times the natural log of data. HP filter applied to level of domestic economy interest rates and US financial risk spread data.
- Stationarity: all variables are stationary and have zero mean after detrending.
- Structural form (closed-form when demeaned): Y(t) = A(1) Y(t-1) + ε(t), with ε(t) ~ i.i.d. (0, Σ).
- Identification assumption: block exogeneity — no spillover from domestic economy to external block; external block affects domestic economy.
- Estimation: standard OLS estimates imposing block exogeneity; Cholesky decomposition used to estimate an orthogonal covariance matrix of residuals.
- Forecast evaluation: multi-step forecasting error approach (recursively computing h-step ahead forecasts) following Marcellino, Stock, and Watson (2006).
- Missing observations / mixed frequency: Kalman filter used in conditional forecasting to account for missing observations.

### Data and variables
- Domestic block variables (monthly except GDP):
  - Real GDP (quarterly)
  - Industrial production
  - Three-month weighted average interest rates (short-term)
  - Imports (foreign trade: total merchandise imports, c.i.f.)
  - Loans to the private sector (Deposit Money Banks Loans: Private Sector)
- Foreign block variables:
  - Price of oil (World Crude Oil {Petro}: Brent, WTI, Dubai Fateh Spot Price Avg. (US$/Bbl))
  - US industrial production excluding construction (US IP)
  - US risk spread (defined as the spread between Moody’s BAA and AAA corporate bond rates)
- Seasonal adjustment: variables seasonally adjusted by official sources and authors’ computations (X12 ARIMA).
- Rationale for foreign variables: global oil price and US IP reflect commodity cycle and demand pressure; US risk spread signals foreign investors’ risk-taking appetite. For Turkey, US risk spread suited nowcasting application better than alternative indicators (EMBI, VIX, FSI).

### Identification, ordering, and timing
- In the domestic block, slow-moving variables (e.g., GDP) are ordered first so they can contemporaneously affect fast-moving variables.
- GDP is presumed uncorrelated with other (non-GDP) domestic shock processes.
- By construction, FEVDs of external variables are explained solely by shocks originating in the foreign block (no domestic → foreign spillovers).

### Impulse response analysis (key quantitative result)
- Comparing impulse responses of GDP growth to external shocks:
  - Upon the impact of a one-standard-deviation oil shock (spread shock and US IP), output drops 1.3 (0.7 and 1) percentage points relative to its steady state.
- Comovements and policy reaction:
  - IP, credit, and imports comove with output growth.
  - Interest rate increases in response to adverse supply shocks (reflecting output-inflation stabilization tradeoff).
  - Interest rate decreases in response to demand shocks.

### Forecast Error Variance Decompositions (FEVD) — aggregated findings
- FEVD over 10 quarters: aggregate contributions reported as “home” (domestic shocks aggregated) vs “foreign” (external shocks aggregated).
- Key observations:
  - The impact of domestic shocks is strongest in the first quarter.
  - More than 60% of output FEVD is driven by external factors.
  - Loans to the private sector: largely determined by domestic factors in the short term; medium-to-long-term share of external shocks increases (attributed mainly to capital flows converted to credit).
  - Short-term interest rates: explained mainly by domestic factors (monetary policy response to business cycle shocks).
- Interpretation: large share of volatility in domestic indicators is explained by external factors — consistent with Turkey’s increasing openness after 2001 and capital flow dynamics.

### Historical Variance Decompositions (HVD) — selected findings
- GDP dynamic: large persistence between 2011 and 2013.
- External channels (real and financial) were very important during 2012-2013 turmoil.
  - Drivers cited for turmoil: Euro Area double-dip recession, US fiscal cliff uncertainty, Turkey’s political unrest, and other factors.
- Role of individual foreign variables between 2012 and mid-2013:
  - US IP: playing a negative but small role.
  - US risk spread: playing a large and positive role in explaining the domestic business cycle.
  - The US IP and US risk spread show opposite contributions, highlighting the countercyclical nature of the risk spread.
- Domestic factors: minor role in explaining the dynamics of short-term rates; 2012 saw a positive and significant contribution from risk spread which diminished later; credit comoves with GDP but domestic explanatory role decreases toward end of sample.

### Counterfactual simulations (three scenarios)
- Simulated economies computed by zeroing out selected residuals:
  - (i) Baseline: all shocks included.
  - (ii) No domestic shocks: exclude domestic shocks, external shocks still operate.
  - (iii) No external shocks: exclude external shocks, domestic shocks still operate.
- Key result from simulations:
  - During the 2009 recession and 2010 rebound, external shocks help explain the downturn and rebound better. A model excluding external shocks predicts a shorter-duration 2009 downturn and a too-speedy recovery.
  - Interpretation: external shocks are important to capture joint dynamics of downturns and rebounds.

### Conditional forecast / Nowcasting methodology
- Nowcasting objective: compute current-quarter forecast of GDP using mixed-frequency information flow; revise forecasts as monthly data arrive.
- Conditional forecast notation: nowcast of quarterly GDP at period t conditional on information up to month M+1 denoted E_t|M+1[Δ_Q ...]. Forecasts updated monthly as new high-frequency data are released.
- Evaluation metric: one-quarter ahead root-mean-square errors (RMSEs) computed over different horizons and vintages.
- RMSE computation: RMSE formula employed as specified in the text (equation (9)); vintages include unconditional and conditional on month 1, month 2, and month 3 data availability within the quarter. Kalman filter used for missing observations.
- Nowcasting improves as new monthly information becomes available throughout the quarter.

### Forecast performance — out-of-sample comparisons (exact figures)
- Out-of-sample forecast errors (difference between forecast and actual), four-steps-ahead, comparison with WEO:
  - Forecast Turkey’s output Growth — Forecast Errors:
    - WEO: 2009 = 4.0; 2010 = 4.2; 2011 = -0.2; 2012 = 0.6
    - Our Model: 2009 = -2.3; 2010 = 2.3; 2011 = 0.4; 2012 = -0.1
  - Note example: WEO’s forecast of 2009 GDP growth was 4% above actual; the VAR-based model predicted the 2009 recession with a somewhat bigger trough.
- Out-of-sample RMSEs (exact figures):
  - Forecast Turkey’s output Growth:
    - WEO: 2009-2012 = 2.9; 2010-2012 = 2.4
    - Our Model: 2009-2012 = 1.6; 2010-2012 = 1.3
- RMSE vintage summary: Figure 5 (described) shows unconditional forecasts (blue bars) and nowcasts conditional on month 1, month 2, month 3 information (red, green, yellow as described in text).

### Cross-correlations and stylized facts (selected exact correlation sequences from Table 4)
- Cross-correlation of Output GAP with leads/lags (t-3 t-2 t-1 t t+1 t+2 t+3):
  - Output GAP: 0.33 0.56 0.82 1 0.82 0.56 0.33
  - Domestic IP: 0.3 0.55 0.79 0.94 0.76 0.51 0.22
  - Short-term Rate: 0.5 0.4 0.2 1 -0.12 -0.29 -0.38 -0.38
  - Import: 0.39 0.64 0.83 0.89 0.62 0.23 -0.03
  - Credits: 0.69 0.83 0.87 0.75 0.52 0.31 0.15
  - Nominal POIL: 0.26 0.51 0.7 0.79 0.56 0.13 -0.13
  - US IP: 0.62 0.8 0.9 0.85 0.67 0.44 0.19
  - US Financial Riskness: 0.38 0.05 -0.35 -0.68 -0.77 -0.61 -0.52

### Nowcasting in practice (illustrative exercise and calendar)
- Practical example: assume Turkish GDP available up to 2014Q2; aim to point-forecast 2014Q3.
  - Domestic variables (except short-term rates) available up to 2014M7; external variables available up to 2014M8.
  - Update variance-covariance matrix and nowcasts as new monthly information is published.
  - Findings: IP and imports are important leading indicators for Turkish aggregate output; loans to private sector also contain useful leading information.
- Calendar of Turkey’s GDP releases (Table 5 exact wording):
  - March/April: Last week of March or 1st week of April — Q4
  - June: After first week — Q1
  - September: 2nd week — Q2
  - December: 2nd week — Q3
- Conditional forecasting implementation:
  - Conditional on M1 includes information of first month of quarter for all monthly variables.
  - Conditional on M2 includes information of second month of quarter only for short-term interest rate and external variables in the model.

### Conclusion (paper’s key messages)
- The paper proposes a nowcasting toolbox for output growth that explicitly incorporates global factors.
- Impulse responses, variance decompositions, and counterfactual simulations show external shocks, together with domestic shocks, can explain domestic variable dynamics.
- The empirical setup parsimoniously combines domestic and external leading indicators, and is suitable for small open economies with large trade and financial links to advanced economies.
- Conditional forecast results indicate improved forecasting power by exploiting mixed-frequency data and measuring impulses from external factors.
- Practical guidance: while the VAR framework with domestic leading indicators and global variables provides a parsimonious forecasting framework with good accuracy, practitioners should not rely on a single model; combinations of forecasting techniques are recommended.

*Italicized source attribution: IMF Working Paper _wp15269 (excerpt provided).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15269.pdf_