## _wp11169

## Source details

**Canonical URL:** [_wp11169](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11169.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11169.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11169.pdf.json)

---

### Introduction: motivation and scope
- Emerging market economies face vulnerability to exogenous shocks such as sudden stops, global liquidity crunches, and terms of trade movements as they integrate into the global economy.
- Main domestic tools: liability composition (short-term vs long-term; local vs foreign currency) and accumulation of international reserves.
- Underexplored tool: hedging through financial instruments (derivatives tied to global financing indicators, commodity options).
- Examples of potential hedging instruments discussed:
  - VIX (advanced-economy market volatility index)
  - S&P 500 (SPX) put options or index movements
  - U.S. interest rates (10-year government bond spread, term spread)
  - US high-yield corporate bond spread (HYS)
  - Cross-country spreads (EMBICDX)
  - Commodity derivatives (oil, gold, copper; put options, collars)

### Data and hedging strategies: series and measurement
- Country spreads:
  - EMBI and emerging markets’ CDS spreads used; spliced EMBI-CDS series where CDS starts later.
  - EMBI Global Index Stripped Spreads (GISS) available for 21 countries: Argentina, Brazil, Bulgaria, Chile, China, Colombia, Ecuador, Korea, Hungary, Lebanon, Malaysia, Mexico, Panama, Peru, Philippines, Poland, Russia, South Africa, Thailand, Turkey, Venezuela.
- Hedging instruments and indices:
  - VIX index: monthly return proxy via rolling futures; correlation with rolling-future returns = 0.69; average bid-ask spread for VIX futures = 27 basis points.
  - US High Yield (HY) Bond Spreads and HY CDX returns: correlation between monthly changes in HY spread and returns on HY CDX index = 0.67; bid-ask spread for CDX index = 17 basis points.
  - US term spread: spread between 10 year government bond rate and 3 month T-bill rate; replicable via standard swap; tracked by FTSE interest rate swap index.
  - EMBICDX: index constructed as average change in CDS (or EMBI if CDS not available) of all sample countries excluding the country itself.
  - Commodity hedgers: S&P GSCI total return indices for oil, gold, copper using 1-month futures rolling.
- Assumptions and limitations:
  - Hedging positions modeled as actuarially fair (expected return equal to risk-free rate) to construct an upper-bound on hedging demand.
  - Transaction costs and bid-ask spreads abstracted from analysis (results represent upper-bound performance).

### Preliminary tests: principal component analysis (PCA)
- Data transformations:
  - Monthly differences for spreads; log differences for level variables (VIX, SPX, oil).
- PCA results:
  - First principal component on monthly changes of country spreads for 26 EMs explains about the 48 percent of the total variation.
  - First three principal components account for about the 64 percent of the total variation.
- Regression of principal components on hedging instruments:
  - Only the first factor correlates with hedgers in univariate regressions for first four factors.
  - Sign relations:
    - VIX and HYS moves positively correlated with spreads (liquidity/risk appetite channels).
    - SPX returns and U.S. government bond yield negatively correlated with spreads.
    - Oil returns have negative coefficients (higher oil prices acting as negative shocks for many EM oil importers).
  - Multivariate regression: stock returns (SPX), HY spread (HYS), and U.S. yield remain significant; oil effect appears captured by these three.

### Preliminary tests: panel regression analysis
- Panel regression setup and sample:
  - Monthly data covering the sample period from 1993M1 to 2009M6.
  - Total Observations 128.
- Key results:
  - Univariate regressions (Columns 1–5) show expected signs for all variables.
  - EMBICDX remains highly significant in multivariate regressions, indicating other country spreads are informative after controlling for global factors.
  - Interaction with country risk:
    - Countries classified as “high” or “low” risk based on average spread above or below 500bp.
    - Riskier countries' spreads are more sensitive to global risk conditions.
- Reported statistics and notes:
  - Adjusted R-squared = 0.7169 (reported under PCA table block).
  - Significance notation: *, **, and *** denote significance at the 10%, 5% and 1% levels, respectively, in two-tail tests.
  - Variable definitions: VIXreturn = monthly return on VIX index; SPXreturn = monthly return on SP500 index; HYSreturn = monthly change of US HY bond spread; FED10Yreturn = monthly change of US 10 year bond spread; OILreturn = monthly return of oil.

### Hedging approach (methodology)
- Notation and setup:
  - tspreadΔ denotes the change in country spreads.
  - Country takes position θ on a hedging instrument with return tr, so hedged change = tspreadΔ + r t θ.
  - Assumption: E[r_t-1] = 0 (actuarially fair).
- Two objective functions:
  - Variance minimization:
    - var arg min[] tt VAR spreadrθ θ Δ.
    - Optimal hedge obtained from OLS ttt spreadrαβϵΔ  ; optimal θ = −β.
  - Lower partial moment (LPM) attenuation:
    - Objective: arg min, f Lcnm where m spread r c Δ; threshold c set at country-specific mean plus two standard deviations.
    - Parameter n controls tail weighting; alternatives: n=0 (minimize tail probability), n=1 (minimize mean of the tail), n=2 (minimize variance of the tail).
    - First-order condition: { } max 0, * tLPMt nE spread r c θ  Δ  .
    - Second-order condition satisfied when 2 n .
    - Nonparametric estimation: distribution F estimated non-parametrically; kernel density estimator with standard normal kernel g(z) = (2π)^(-1/2) exp(-z^2/2); LPM estimated via equations using optimal bandwidth and grid search to find θ that minimizes estimated LPM.
- Multi-hedger extension:
  - Portfolio return = Σ i r i θ i.

### Results — instrument selection and in-sample findings
- Instrument selection:
  - PCA and panel regressions identify EMBICDX, VIX, and HYS as the three most promising hedging instruments (reasonable liquidity and market depth).
  - Commodity prices not relevant for explaining EMBI spread variation and not pursued further.
- Static (in-sample) hedging — key patterns:
  - Positive correlation between hedge ratios and EMBI spreads: higher-risk countries take larger hedging positions.
  - Tail risk minimization (LPM) requires more hedging effort, particularly for high-risk countries.
- In-sample hedging performance (variance minimization) — average effects:
  - EMBICDX reduces standard deviation by 26 percentage points on average.
  - HYS reduces standard deviation by 20 percentage points on average.
  - Two-hedger strategy (EMBICDX + HYS) reduces standard deviation by 31 percentage points on average.
  - Reduction in standard deviation tends to be larger for lower-risk countries (average spread < 500bp).
- In-sample LPM (n=0, tail probability minimization, threshold = two standard deviations):
  - EMBICDX and HYS reduce tail probability, on average, by about 4 percentage points (5 percentage points when both are used).
  - This corresponds to about 40 percent reduction in tail probability (47 percent when both instruments are used).
  - Hedging performance comparable across high and low risk groups, although still better for low-risk countries.
- In-sample LPM (n=1, tail mean minimization):
  - EMBICDX and HYS remain the best hedgers; HYS performs slightly better on this metric.
  - Combined they reduce the tail mean by, on average, 56 percent of its unhedged value.
- For n=2 (extreme tail variance), EMBICDX produces the most significant reduction in probability of extreme tail events.

### Out-of-sample and dynamic hedging results
- Out-of-sample exercise design:
  - Initial estimation sample: from first available observation up to 2002Q4 to estimate hedge ratio for 2003Q1; then one-quarter-ahead rolling until 2009Q3. Monthly-frequency robustness check also performed.
  - Full sample referenced: sample starts from 1993 Q1 to 2009 Q2 (starting date varies across countries).
- Out-of-sample performance (variance minimization):
  - Substantial deterioration in performance relative to in-sample.
  - Using HYS as hedger often increases variance of spreads.
  - EMBICDX continues to reduce variance but much less than in-sample: average reduction out-of-sample is just by 9 percent.
  - Large cross-country variation; in some cases hedging adds significant volatility.
- Out-of-sample performance (tail probability and tail mean):
  - Tail probability: out-of-sample performance worse than in-sample; instances where hedging increases tail probability (quarterly and monthly frequencies).
  - Tail mean: out-of-sample performance shows improvement concentrated on higher-risk countries with more extreme tail events.
- Interpretation:
  - Hedge ratios estimated in volatile/crisis-rich samples perform poorly when extrapolated to tranquil periods, but may become relevant again during crises.

### Tranquil vs crisis times — conditional effectiveness
- Experiment:
  - Hedge ratios estimated using data up to 2004Q4 (include major EM crisis episodes).
  - Evaluate performance in tranquil period 2005Q1–2007Q1 and crisis period 2007Q2–2009Q2.
- Findings:
  - In tranquil period (2005Q1–2007Q1): volatility largely unchanged; EMBICDX slightly increases volatility for low-risk countries and reduces for high-risk countries; HYS slightly reduces volatility for both groups.
  - In crisis period (2007Q2–2009Q2): dramatic reductions in volatility and tail risk, especially for high-spread countries.
    - EMBICDX and HYS hedging would have reduced standard deviation by roughly 80 and 75 percent on average, respectively.
    - Reductions in tail mean are reduced by about 90 percent for both hedgers.
    - Either EMBICDX or HYS hedging would have virtually eliminated tail risk in 2007Q2–2009Q2 for most high-spread countries in the sample.
- Practical implication:
  - Hedging pays off substantially in crises but is of limited use (and may be counterproductive once transaction costs are considered) in tranquil times.
  - Whether hedging proves worthwhile depends on frequency of global crises and expected costs during crises.

### Practical considerations and omitted strategies
- Market liquidity and tradability:
  - VIX futures trading started in 2005; rolling-future strategies and total return indices invoked to proxy implementable instruments.
  - HY CDX and emerging market CDX indices are tradable in OTC markets via TRS, CDS, credit options, credit-linked notes.
- Costs and realism:
  - Analysis abstracts from transaction costs and bid-ask spreads, so results represent an upper-bound on hedging uptake and effectiveness.
  - Option-based strategies were not analyzed due to explicit costs and large bid-ask spreads (at-the-money VIX options bid-ask spreads on order of 5-10 percent).
  - Shorting SPX not considered due to prohibitive expected cost and poor hedging of asymmetric risks.
- Policy and political constraints:
  - Limited practical uptake may reflect governments' reluctance to spend budgetary resources on insurance against events they will not be blamed for, given protection primarily valuable in tail events.
  - Behavioral and political economy note: post-crisis complacency in tranquil times reduces political appetite to buy protection against tail events despite potential high payoffs during crises.

### Conclusions (policy-relevant takeaways)
- Promising instruments:
  - Strategies based on U.S. high-yield spreads (HYS) and positions on other emerging market spreads (EMBICDX) are somewhat promising as hedgers.
- Performance summary:
  - These instruments explain much of in-sample movements in spreads but perform poorer out-of-sample, except when large crises are looming.
  - Hedging based on correlations derived during crisis sample periods can be very helpful for hedging against future crises (can reduce volatility of tail risk up to 90 percent).
  - The same hedging is ineffective or counter-productive during tranquil times.
- Distribution of benefits:
  - Hedging is particularly useful for high-risk countries (those with high average spreads).
  - High-risk countries often have higher hedge ratios and therefore higher potential payoffs during crises.
- Policy stance:
  - Results support cautious “experimentation” with the considered hedging strategies as a complement to the policy toolkit for insuring against global shocks.
  - Uneven payoff over the cycle, transaction costs, innovation and implementation costs, and political obstacles help explain why countries do not widely adopt such hedging strategies.

### Appendix Table I — regression framework and selected coefficient highlights
- Dep Var: Chg(log(Spread))
- Number of observations in each model: 3944
- Key coefficient estimates (selected models and values preserved exactly):
  - Chg(logVIX):
    - Model 1: 0.5808*** (0.0339)
    - Model 2: 0.1177** (0.0280)
    - Model 3: 0.1164** (0.0274)
    - Model 4: 0.0942** (0.0452)
    - Model 5: 0.1314*** (0.0355)
    - Model 6 (High Risk (Spd>500bp)): 0.2478*** (0.0391)
    - Model 7 (Low Risk (Spd<500bp)): 0.2350*** (0.0387)
  - Chg(logSPX):
    - Model 1: -2.1219*** (0.1287)
    - Model 2: -0.1467 (0.0981)
    - Model 3: -0.1464 (0.0972)
    - Model 4: -0.3357** (0.1258)
    - Model 5: -0.0430 (0.1285)
    - Model 6 (High Risk): -0.9576*** (0.0707)
    - Model 7 (Low Risk): -0.4611*** (0.1434)
  - Chg(log(HYS)):
    - Model 1: 1.1269*** (0.0576)
    - Model 2: 0.4120*** (0.0605)
    - Model 3: 0.4098*** (0.0606)
    - Model 4: 0.4806*** (0.0867)
    - Model 5: 0.2994*** (0.0753)
    - Model 6 (High Risk): 0.8630*** (0.1065)
    - Model 7 (Low Risk): 0.8535*** (0.0681)
  - Chg(log(EMBICDX)):
    - Model 1: 0.7996*** (0.0486)
    - Model 3: 0.5750*** (0.0510)
    - Model 4: 0.6149*** (0.0513)
    - Model 6 (High Risk): 0.7409*** (0.0935)
    - Model 7 (Low Risk): 0.4796*** (0.0432)
  - Chg(FED10Y):
    - Model 2: -0.5101*** (0.0484)
    - Model 3: 0.1476** (0.0508)
    - Model 4: 0.1399** (0.0507)
    - Model 5: 0.0952 (0.0974)
    - Model 6 (High Risk): 0.1696*** (0.0570)
    - Model 7 (Low Risk): 0.2597*** (0.0531)
    - Model 8: 0.2697*** (0.0925)
  - Chg(logOIL) and interactions:
    - Chg(logOIL): -0.7670 (0.4972) in one specification; -0.0523 (0.1034); -0.0934 (0.1691) in others
    - EXP*Chg(logOIL): 0.0404 (0.0911); -0.0114 (0.0775); 0.0479 (0.1542)
    - EXP*Chg(logGOLD): -0.1847* (0.0609); -0.2284* (0.1026); -0.2392* (0.1114)
    - EXP*Chg(logCOPPER): -0.0105 (0.0468); -0.0353 (0.0924); 0.0039 (0.0857)
- Goodness-of-fit (R-Squared by model):
  - Model 1: 0.1333
  - Model 2: 0.1327
  - Model 3: 0.1645
  - Model 4: 0.2537
  - Model 5: 0.0872
  - Model 6: 0.2739
  - Model 7: 0.2793
  - Model 8: 0.2996
  - Model 9: 0.1824
- P values comparing coefficients between high- and low-risk groups (selected):
  - EMBICDX: P-value=0.0177
  - SPX: Pvalue=0.0042
  - VIX: P-value=0.5483 and P-value=0.8202 (reported)
  - HYS: P-value=0.1267 and P-value=0.9404 (reported)
  - FED10y: P-value=0.9255
- Notes:
  - Series of EMBICDX are calculated as the average EMBI spreads of all the countries excluding the own country.
  - FED10Y is the spread between US 10 year government bond and US 3 month T-bill.
  - Countries with average spread higher than 500bp are considered high risk group, while the rest are in the low risk group.
  - Oil exporters: Russia, Venezuela, Mexico and Nigeria.
  - South Africa is the gold exporter and Chile the copper exporter.
  - Significance notation: *, **, and *** denote significance at the 10%, 5% and 1% levels, respectively, in two-tail tests.
  - Standard deviations of the coefficients are in the parentheses.

*Source: Appendix Table I (from document _wp11169 - Appendix Table I).*

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

### _wp11169 - References .............................................................................................................

### Introduction: motivation and scope
- Emerging market economies face vulnerability to exogenous shocks (sudden stops, global liquidity crunches, terms of trade movements) as they integrate into the global economy.
- Main domestic tools: liability composition (short-term vs long-term; local vs foreign currency) and accumulation of international reserves.
- Underexplored tool: hedging through financial instruments (derivatives tied to global financing indicators, commodity options).
- Examples of potential hedging instruments discussed:
  - VIX (advanced-economy market volatility index)
  - S&P 500 (SPX) put options or index movements
  - U.S. interest rates (10-year government bond spread, term spread)
  - US high-yield corporate bond spread (HYS)
  - Cross-country spreads (EMBICDX)
  - Commodity derivatives (oil, gold, copper; put options, collars)

### Data and hedging strategies: series and measurement
- Country spreads:
  - EMBI and emerging markets’ CDS spreads used; spliced EMBI-CDS series where CDS starts later.
  - EMBI Global Index Stripped Spreads (GISS) available for 21 countries (Argentina, Brazil, Bulgaria, Chile, China, Colombia, Ecuador, Korea, Hungary, Lebanon, Malaysia, Mexico, Panama, Peru, Philippines, Poland, Russia, South Africa, Thailand, Turkey, Venezuela).
- Hedging instruments and indices:
  - VIX index (monthly return proxy via rolling futures; correlation with rolling-future returns = 0.69; average bid-ask spread for VIX futures = 27 basis points).
  - US High Yield (HY) Bond Spreads and HY CDX returns (correlation between monthly changes in HY spread and returns on HY CDX index = 0.67; bid-ask spread for CDX index = 17 basis points).
  - US term spread: spread between 10 year government bond rate and 3 month T-bill rate (replicable via standard swap; tracked by FTSE interest rate swap index).
  - EMBICDX: index constructed as average change in CDS (or EMBI if CDS not available) of all sample countries excluding the country itself.
  - Commodity hedgers: S&P GSCI total return indices for oil, gold, copper using 1-month futures rolling.
- Assumptions and limitations:
  - Hedging positions modeled as actuarially fair (expected return equal to risk-free rate) to construct an upper-bound on hedging demand.
  - Transaction costs and bid-ask spreads abstracted from in analysis (results represent upper-bound performance).

### Preliminary tests: principal component analysis (PCA)
- Data transformations:
  - Monthly differences for spreads; log differences for level variables (VIX, SPX, oil).
- PCA results:
  - First principal component on monthly changes of country spreads for 26 EMs explains about the 48 percent of the total variation.
  - First three principal components account for about the 64 percent of the total variation.
- Regression of principal components on hedging instruments (Table 1):
  - Univariate regressions for first four factors show only the first factor correlates with hedgers.
  - Signs and relations:
    - VIX and HYS moves positively correlated with spreads (liquidity/risk appetite channels).
    - SPX returns and U.S. government bond yield negatively correlated with spreads.
    - Oil returns have negative coefficients (higher oil prices acting as negative shocks for many EM oil importers).
  - Multivariate regression: stock returns (SPX), HY spread (HYS), and U.S. yield remain significant; oil effect appears captured by these three.

### Preliminary tests: panel regression analysis
- Panel regression set-up and sample:
  - Monthly data covering the sample period from 1993M1 to 2009M6.
  - Total Observations 128.
- Key results (Table 2 summary):
  - Univariate regressions (Columns 1–5): expected signs for all variables.
  - EMBICDX (average EM spreads excluding own country) remains highly significant in multivariate regressions (Column 6), indicating other country spreads are informative after controlling for global factors.
  - Interaction with country risk (Columns 8–9):
    - Countries classified as “high” or “low” risk based on average spread above or below 500bp.
    - Riskier countries' spreads are more sensitive to global risk conditions.
- Reported statistics and notes:
  - Adjusted R-squared = 0.7169 (reported under PCA table block).
  - Significance notation: *, **, and *** denote significance at the 10%, 5% and 1% levels, respectively, in two-tail tests.
  - Variable definitions: VIXreturn = monthly return on VIX index; SPXreturn = monthly return on SP500 index; HYSreturn = monthly change of US HY bond spread; FED10Yreturn = monthly change of US 10 year bond spread; OILreturn = monthly return of oil.

### Hedging strategy evaluation: in-sample vs out-of-sample and tail protection
- Hedging objectives:
  - Minimize variance and different Lower Partial Moments (LPM) of hedged spreads, including:
    - Probability of a tail realization (2 standard deviations from mean).
    - Mean and variance of hedged spreads conditional on tail occurrence.
- Performance summary:
  - Strong in-sample performance consistent with existing literature.
  - Out-of-sample performance is much poorer.
  - During tranquil periods, strategies yield limited gains or may add risk to spreads.
  - Strategies tend to pay off during spikes in global risk aversion (example: late 2008/early 2009).
  - Implication: strategies may be more useful for protection against tail events than for smoothing ongoing volatility.

### Practical considerations and omitted strategies
- Market liquidity and tradability:
  - VIX futures trading started in 2005; rolling-future strategies and total return indices invoked to proxy implementable instruments.
  - HY CDX and emerging market CDX indices are tradable in OTC markets via TRS, CDS, credit options, credit-linked notes.
- Costs and realism:
  - Analysis abstracts from transaction costs and bid-ask spreads, so results represent an upper-bound on hedging uptake and effectiveness.
  - Option-based strategies were not analyzed due to explicit costs and large bid-ask spreads (at-the-money VIX options bid-ask spreads on order of 5-10 percent).
  - Shorting SPX not considered due to prohibitive expected cost and poor hedging of asymmetric risks.
- Policy and political constraints:
  - Limited practical uptake may reflect governments' reluctance to spend budgetary resources on insurance against events they will not be blamed for, given protection primarily valuable in tail events.

* _wp11169 - References .............................................................................................................*

### Appendix Table I is analogous to Table II but uses the log change in country spreads as

### _wp11169 - Appendix Table I is analogous to Table II but uses the log change in country spreads as

### IV. Hedging approach (methodology)
- Dependent variable notation:
  - tspreadΔ denotes the change in country spreads.
  - Country takes position θ on a hedging instrument with return tr, so hedged change = tspreadΔ + r t θ.
  - Assumption: 1 0 tt Er  .
- Two objective functions considered:
  - Variance minimization:
    - Problem: var arg min[] tt VAR spreadrθ θ Δ (equation (1)).
    - Optimal hedge can be obtained from OLS regression ttt spreadrαβϵΔ   (equation (2)), with var θ β .
  - Lower partial moment (LPM) attenuation:
    - Objective: minimize expected value of an increasing function of tail realizations for hedged spread:
      - arg min, f Lcnm where m spread r c Δ (equation (3)); f d is the pdf of the return on the hedged spread.
      - Threshold c is set at the country-specific mean plus two standard deviations.
      - Parameter n controls tail weighting; alternative values considered: n=0, 1, or 2.
        - n=0: minimize tail probability (probability that change in hedged spreads is beyond threshold c).
        - n=1: minimize mean of the tail.
        - n=2: minimize variance of the tail.
      - First-order condition for optimal hedge ratio θLPM: { } max 0, * tLPMt nE spread r c θ  Δ   [equation (4)].
      - Second-order condition satisfied when 2 n .
    - Nonparametric estimation:
      - Distribution F estimated non-parametrically over whole sample.
      - Kernel density estimator for portfolio return fm: y f (.) estimated by (5) with kernel g(.) taken as standard normal g(z) = (2π)^(-1/2) exp(-z^2/2) (equation (6)).
      - LPM estimated via equations (7)-(8) using optimal bandwidth; grid search used to find θ that minimizes estimated LPM.
- Multi-hedger extension: portfolio return = Σ i r i θ i.

### V. Results — instrument selection and in-sample findings
- PCA and panel regressions (Tables I and II) identify three most promising hedging instruments: EMBICDX, VIX, and HYS (reasonable liquidity and market depth).
- Commodity prices found not relevant for explaining EMBI spread variation; thus not pursued further.
- Static (in-sample) hedging — key patterns:
  - Positive correlation between hedge ratios and EMBI spreads: higher-risk countries take larger hedging positions.
  - Tail risk minimization (LPM) requires more hedging effort, particularly for high-risk countries.
- In-sample hedging performance (variance minimization):
  - On average:
    - EMBICDX reduces standard deviation by 26 percentage points on average.
    - HYS reduces standard deviation by 20 percentage points on average.
    - Two-hedger strategy (EMBICDX + HYS) reduces standard deviation by 31 percentage points on average.
  - Reduction in standard deviation tends to be larger for lower-risk countries (average spread < 500bp).
- In-sample LPM (n=0, tail probability minimization, threshold = two standard deviations):
  - EMBICDX and HYS reduce tail probability, on average, by about 4 percentage points (5 percentage points when both are used).
  - This corresponds to about 40 percent reduction in tail probability (47 percent when both instruments are used).
  - Hedging performance comparable across high and low risk groups, although still better for low-risk countries.
- In-sample LPM (n=1, tail mean minimization):
  - EMBICDX and HYS remain the best hedgers; HYS performs slightly better in this metric.
  - Combined they reduce the tail mean by, on average, 56 percent of its unhedged value.
- For n=2 (extreme tail variance), EMBICDX produces the most significant reduction in probability of extreme tail events.

### V.C Out-of-sample and dynamic hedging results
- Out-of-sample exercise design:
  - Initial estimation sample: from first available observation up to 2002Q4 to estimate hedge ratio for 2003Q1; then one-quarter-ahead rolling until 2009Q3. Monthly-frequency robustness check also performed.
  - Full sample referenced elsewhere: sample starts from 1993 Q1 to 2009 Q2 (starting date varies across countries).
- Out-of-sample performance (variance minimization):
  - Substantial deterioration in performance relative to in-sample.
  - Using HYS as hedger often increases variance of spreads (i.e., counterproductive).
  - EMBICDX continues to reduce variance but much less than in-sample: average reduction out-of-sample is just by 9 percent.
  - Large cross-country variation; in some cases hedging adds significant volatility.
- Out-of-sample performance (tail probability and tail mean):
  - Tail probability: out-of-sample performance worse than in-sample; instances where hedging increases tail probability (quarterly and monthly frequencies).
  - Tail mean: out-of-sample performance shows improvement concentrated on higher-risk countries with more extreme tail events.
- Interpretation:
  - Hedge ratios estimated in volatile/crisis-rich samples perform poorly when extrapolated to tranquil periods, but may become relevant again during crises.

### V.D Tranquil vs crisis times — conditional effectiveness
- Experiment:
  - Hedge ratios estimated using data up to 2004Q4 (include major EM crisis episodes).
  - Evaluate performance in tranquil period 2005Q1–2007Q1 and crisis period 2007Q2–2009Q2.
- Findings:
  - Hedging based on 1993–2004 data produces:
    - In tranquil period (2005Q1–2007Q1): volatility largely unchanged; EMBICDX slightly increases volatility for low-risk countries and reduces for high-risk countries; HYS slightly reduces volatility for both groups.
    - In crisis period (2007Q2–2009Q2): dramatic reductions in volatility and tail risk, especially for high-spread countries.
      - EMBICDX and HYS hedging would have reduced standard deviation by roughly 80 and 75 percent on average, respectively.
      - Reductions in tail mean are reduced by about 90 percent for both hedgers.
      - Either EMBICDX or HYS hedging would have virtually eliminated tail risk in 2007Q2–2009Q2 for most high-spread countries in the sample.
- Practical implication:
  - Hedging pays off substantially in crises but is of limited use (and may be counterproductive once transaction costs are considered) in tranquil times.
  - On average, whether hedging proves worthwhile depends on frequency of global crises and expected costs during crises.

### VI. Conclusions (policy-relevant takeaways)
- Promising instruments:
  - Strategies based on U.S. high-yield spreads (HYS) and positions on other emerging market spreads (EMBICDX) are somewhat promising as hedgers.
- Performance summary:
  - These instruments explain much of in-sample movements in spreads but perform poorer out-of-sample, except when large crises are looming.
  - Hedging based on correlations derived during crisis sample periods can be very helpful for hedging against future crises (can reduce volatility of tail risk up to 90 percent).
  - The same hedging is ineffective or counter-productive during tranquil times.
- Distribution of benefits:
  - Hedging is particularly useful for high-risk countries (those with high average spreads).
  - High-risk countries often have higher hedge ratios and therefore higher potential payoffs during crises.
- Policy stance:
  - Results support cautious “experimentation” with the considered hedging strategies as a complement to the policy toolkit for insuring against global shocks.
  - Uneven payoff over the cycle, transaction costs, innovation and implementation costs, and political obstacles help explain why countries do not widely adopt such hedging strategies.
- Behavioral and political economy note:
  - Post-crisis complacency in tranquil times reduces political appetite to buy protection against tail events despite potential high payoffs during crises.

*Source: _wp11169 - Appendix Table I is analogous to Table II but uses the log change in country spreads as (PDF). *

### Appendix Table I

### Appendix Table I

### Regression framework and dependent variable
- Dep Var: Chg(log(Spread))
- Expected Sign: indicated per variable in table (e.g., Chg(logVIX) (+), Chg(logSPX) (-), Chg(log(HYS)) (+), Chg(log(EMBICDX)) (+), Chg(FED10Y) (-), Chg(logOIL) (+), EXP* interactions (signs indicated))
- Number of observations in each model: 3944

### Key coefficient estimates across models (preserve model-specific values)
- Chg(logVIX)
  - Model 1: 0.5808*** (0.0339)
  - Model 2: 0.1177** (0.0280)
  - Model 3: 0.1164** (0.0274)
  - Model 4: 0.0942** (0.0452)
  - Model 5: 0.1314*** (0.0355)
  - Model 6 (High Risk (Spd>500bp)): 0.2478*** (0.0391)
  - Model 7 (Low Risk (Spd<500bp)): 0.2350*** (0.0387)

- Chg(logSPX)
  - Model 1: -2.1219*** (0.1287)
  - Model 2: -0.1467 (0.0981)
  - Model 3: -0.1464 (0.0972)
  - Model 4: -0.3357** (0.1258)
  - Model 5: -0.0430 (0.1285)
  - Model 6 (High Risk): -0.9576*** (0.0707)
  - Model 7 (Low Risk): -0.4611*** (0.1434)

- Chg(log(HYS))
  - Model 1: 1.1269*** (0.0576)
  - Model 2: 0.4120*** (0.0605)
  - Model 3: 0.4098*** (0.0606)
  - Model 4: 0.4806*** (0.0867)
  - Model 5: 0.2994*** (0.0753)
  - Model 6 (High Risk): 0.8630*** (0.1065)
  - Model 7 (Low Risk): 0.8535*** (0.0681)

- Chg(log(EMBICDX))
  - Model 1: 0.7996*** (0.0486)
  - Model 3: 0.5750*** (0.0510)
  - Model 4: 0.6149*** (0.0513)
  - Model 6 (High Risk): 0.7409*** (0.0935)
  - Model 7 (Low Risk): 0.4796*** (0.0432)

- Chg(FED10Y)
  - Model 2: -0.5101*** (0.0484)
  - Model 3: 0.1476** (0.0508)
  - Model 4: 0.1399** (0.0507)
  - Model 5: 0.0952 (0.0974)
  - Model 6 (High Risk): 0.1696*** (0.0570)
  - Model 7 (Low Risk): 0.2597*** (0.0531)
  - Model 8: 0.2697*** (0.0925)

- Chg(logOIL) and interactions
  - Chg(logOIL): -0.7670 (0.4972) in one specification; -0.0523 (0.1034); -0.0934 (0.1691) in others
  - EXP*Chg(logOIL): 0.0404 (0.0911); -0.0114 (0.0775); 0.0479 (0.1542)
  - EXP*Chg(logGOLD): -0.1847* (0.0609); -0.2284* (0.1026); -0.2392* (0.1114)
  - EXP*Chg(logCOPPER): -0.0105 (0.0468); -0.0353 (0.0924); 0.0039 (0.0857)

### Goodness-of-fit and sample
- R-Squared by model:
  - Model 1: 0.1333
  - Model 2: 0.1327
  - Model 3: 0.1645
  - Model 4: 0.2537
  - Model 5: 0.0872
  - Model 6: 0.2739
  - Model 7: 0.2793
  - Model 8: 0.2996
  - Model 9: 0.1824
- # of Obs: 3944 (for all reported models)

### P values of tests comparing coefficients between high- and low-risk groups
- VIX: P-value=0.5483
- VIX: P-value=0.8202
- SPX: Pvalue=0.1346
- SPX: Pvalue=0.0042
- HYS: P-value=0.1267
- HYS: P-value=0.9404
- EMBICDX: P-value=0.0177
- FED10y: P-value=0.9255

### Notes on variables, group definitions, and interpretation
- Models 1-5 examine pairwise correlations between the monthly PERCENTAGE changes of EMBI spreads and monthly returns of various financial instruments.
- Series of EMBICDX are calculated as the average EMBI spreads of all the countries excluding the own country.
- FED10Y is the spread between US 10 year government bond and US 3 month T-bill.
- VIX, SPX, HYS, OIL, GOLD and COPPER track movements of the VIX index, SP500 index, US high yield spread, Oil, gold and copper price movements respectively.
- Countries with average spread higher than 500bp are considered high risk group, while the rest are in the low risk group.
- Model 8 introduces dummies for both high risk and low risk groups; coefficients are group specific. Example (from Model 8): "1 percentage change of EMBICDX corresponds to 0.7410 percentage change of country spread for the high risk group, and 0.4794 percentage change for the low risk group (after controlling for the impact from other variables)."
- T tests (P values) indicate whether coefficients are significantly different between groups.
- Oil exporters: Russia, Venezuela, Mexico and Nigeria.
- South Africa is the gold exporter and Chile the copper exporter.
- Significance notation: *, **, and *** denote significance at the 10%, 5% and 1% levels, respectively, in two-tail tests.
- Standard deviations of the coefficients are in the parentheses.

*Source: Appendix Table I (from document _wp11169 - Appendix Table I).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp11169.pdf_