## _wp09140 - Two-step estimation

## Source details

**Canonical URL:** [_wp09140 - Two-step estimation](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09140.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09140.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09140.pdf.json)

---

### Introduction
- Commodity prices affected by recent protracted financial turmoil: pro-cyclical demand drove price moves for some commodities; gold underscored its role as a safe-haven asset and store of value.
- Event study methodology applied to investigate how scheduled and periodic macroeconomic announcements affect commodity futures prices.
- Timing of scheduled announcements is known in advance, making them potentially key factors for traders and long-term market participants.
- Gold behaves differently from other commodities:
  - Gold prices react to specific scheduled announcements in the United States and the Euro area (indicators of activity or interest rate decisions) consistent with safe-haven and store-of-value roles.
  - Other commodity prices, where macroeconomic news is significant, exhibit pro-cyclical sensitivities, albeit much less than financial assets.

### Methodology and literature review (asset prices and macro announcements)
- Prior findings relevant to approach:
  - Rossi (1998): certain key economic announcements cause U.K. government bond yield changes of between 2–6 basis points.
  - Fleming and Remelona (1999): arrival of public information has a large effect on prices and subsequent trading activity, especially when implied volatility is high.
  - Balduzzi, Elton, and Green (2001): labor market, inflation, and durable goods orders data have large impacts on U.S. Treasury bond prices.
  - Strongest and most consistent relationship for commodities is with the U.S. dollar.
  - Andersen et al (2002) and Galati and Ho (2003): macroeconomic news has statistically significant correlation with intra-day movements of the U.S. dollar; “bad” news has larger impact.
- Commodities-specific regularities from literature:
  - Compared with bonds, exchange rates, equities, fewer and weaker macro announcement effects for commodities.
  - Key U.S. indicators (inflation, GDP, employment) repeatedly move some commodity prices.
  - Energy products tend to be less sensitive; gold most sensitive.
  - Gold often rises with unexpected U.S. inflation and output increases, or tighter labor market (Ghura (1990); Christie-David, Chaudry and Koch (2000)).
  - Sensitivity of gold varies through time; increases during recessions (Hess, Huang, and Niessen (2008)).

### Data and empirical setup
- Daily futures price data for 12 commodity futures contracts, January 1997 to June 2009, covering precious metals, base metals, energy, and agricultural commodities.
- Futures contracts (closing price times shown):
  - Gold: COMEX, 100 troy ounces, 17:15 EST
  - Silver: COMEX, 5,000 troy ounces, 17:15 EST
  - Platinum: NYMEX, 50 troy ounces, 17:15 EST
  - Palladium: NYMEX, 100 troy ounces, 17:15 EST
  - Oil: NYMEX, Light, sweet crude, 1,000 barrels, 17:15 EST
  - Heating oil: NYMEX, 42,000 barrels, 17:15 EST
  - Natural gas: NYMEX, 10,000 million British thermal units, 17:15 EST
  - Wheat: CBOT, 5,000 bushels, 13:15 CST
  - Corn: CBOT, 5,000 bushels, 13:15 CST
  - Soybeans: CBOT, 5,000 bushels, 13:15 CST
  - Copper: COMEX, High grade, 25,000 pounds, 17:15 EST
  - Aluminium: COMEX, 44,000 pounds, 17:15 EST
- Surprise measure:
  - Surprise Z_it = (X_it - E_{t-1}(X_it)) / σ_X, using Bloomberg analyst consensus estimates; set of 13 monthly or quarterly U.S. macro announcements plus ECB and Bank of England interest rate decisions and German IFO survey.
- Frequency choice: daily data preferred over intraday to address liquidity effects and allow markets time to absorb news.
- Summary statistics (selected, January 1997–April 2009):
  - Advance retail sales: Average actual 0.2, Average surprise 0.0, Standard deviation surprise 0.6
  - Change in non-farm payrolls (thousands): Average actual 43.6, Average surprise -24.0, Standard deviation surprise 77.5
  - Consumer confidence (index): Average actual 89.8, Average surprise -0.3, Standard deviation surprise 5.0
  - Consumer price index: Average actual 0.2, Average surprise 0.0, Standard deviation surprise 0.2
  - Employment cost index (quarterly percent change): Average actual 0.7, Average surprise 0.0, Standard deviation surprise 0.2
  - FOMC interest rate decision (absolute basis point change): Average actual 18, Average surprise -16
  - Advance GDP (annualized quarterly percent change): Average actual 2.5, Average surprise 0.0, Standard deviation surprise 0.0
  - Housing starts (thousands): Average actual 1,648, Average surprise 9.4, Standard deviation surprise 101
  - Industrial production: Average actual 0.1, Average surprise 0.0, Standard deviation surprise 0.0
  - ISM manufacturing survey: Average actual 52.9, Average surprise 0.1, Standard deviation surprise 2.1
  - ECB interest rate decision (absolute basis point change): Average actual 30, Average surprise 38
  - German IFO survey: Average actual 96.7, Average surprise 0.1, Standard deviation surprise 1.2

### Estimation strategy
- Benchmark mean equation (OLS) for log change in futures price Δp_t:
  - Δp_t = α + Σ_{j=0}^{J} β_j Z_{j,t-k} + Σ_{k=1}^{K} λ_k Δp_{t-k} + ε_t, with ε_t ~ i.i.d.(0, σε^2).
- Volatility modeling and corrections:
  - Evidence of heteroscedasticity and volatility clustering; GARCH approach used to jointly model returns and volatility.
  - Conditional variance h_t modeled as function of lagged squared residuals and past h; innovations ν_t ~ N(0,1).
  - Preference for GARCH(1,1) but likelihood ratio tests identified K =2 and L =2 in most cases.
  - Bollerslev and Wooldridge (1992) robust standard errors used.
- U.S. dollar controls:
  - Include Federal Reserve trade-weighted U.S. dollar index log change Δe_t with M lags as exogenous regressor.
  - Augmented mean equation: Δp_t = α + Σ_{m=0}^{M} θ_m Δe_{t-m} + Σ_{l=1}^{L} λ_l Δp_{t-l} + Σ_{k=0}^{K} β_k Z_{t-k} + ε_t.
  - Assumed causality from U.S. dollar to commodity price.
- Conditioning on volatility and good/bad news:
  - Composite aggregate news indicator constructed (sum of standardized surprises, excluding monetary policy shocks).
  - Volatility classification: high-volatility if prior 30-, 60-, or 90-day daily standard deviation > sample average.
  - Good news defined as surprises that should raise prices of cyclically-sensitive assets (e.g., higher-than-expected GDP, industrial production, non-farm payrolls, consumer confidence, inflation).
  - Models include separate regressors for good and bad news and for high/low volatility states.

### Results — Scheduled macroeconomic announcements (selected findings and coefficients)
- General patterns:
  - Energy products tend to exhibit little sensitivity; agricultural products and base metals show pro-cyclical sensitivity; gold tends to be counter-cyclical.
  - Most influential indicators for gold: U.S. retail sales, non-farm payrolls, housing starts, ISM survey.
  - German IFO survey strongly influences base metals, even when controlling for the U.S. dollar.
  - Controlling for the U.S. dollar often heightens measured pro-cyclical sensitivity for non-gold commodities.
  - Commodities tend to be inversely related to Federal Reserve interest rate surprises where significant.
- Selected coefficients (from Table 3; coefficients multiplied by 100 represent percent change in nearest futures price for a 1-standard deviation surprise; only those significant at the 10 percent level or more are shown):
  - Gold:
    - Change in non-farm payrolls: -0.18 *
    - Consumer confidence: -0.15 **
    - Industrial production: -0.31 *
    - Consumer price index: 0.10
    - Advance GDP: -0.13
    - ISM manufacturing survey: -0.09
    - ECB interest rate decision: 0.15 **
    - German IFO survey: 0.13
  - Crude Oil:
    - Change in non-farm payrolls: 0.12
    - Industrial production: 0.27
    - Existing home sales (lagged): 0.34 *
    - Advance retail sales (lagged): -0.47
  - U.S. dollar (selected entries):
    - Advance retail sales: 0.05
    - Change in non-farm payrolls: 0.13 ***
    - Consumer confidence: 0.08 ***
    - Advance GDP: 0.17 **
    - German IFO survey: -0.11

- Subperiod robustness:
  - Sample split 1997–November 2001 and December 2001–March 2009: number of indicators affecting prices and pro-cyclical sensitivity among non-gold commodities tended to rise since 2001; Chow tests around November 2001 indicated inability to reject model stability at the 5 percent level for almost all commodities.

### Results — Including U.S. dollar controls (selected findings)
- Including U.S. dollar index as regressor changes significance patterns and magnitudes; pro-cyclical sensitivities for many commodities are heightened when controlling for the U.S. dollar.
- Selected coefficients with U.S. dollar control (examples):
  - Gold:
    - Change in non-farm payrolls: -0.05 (Table 4)
    - Consumer price index: 0.08
    - German IFO survey: 0.00 (Table 4 and Table 5)
  - Crude Oil:
    - Change in non-farm payrolls: 0.24
    - Industrial production: 0.33
    - Advance retail sales: -0.38
- U.S. dollar index change contemporaneous coefficients (examples, Table A4/A6):
  - U.S. dollar index change t: Crude -1.04 ***; Silver -1.13 ***; Gold -1.02 ***
  - U.S. dollar index change t-1: Crude -0.10 **; Gold 0.18 **

### Results — Good/bad news and volatility conditioning (selected findings)
- Good vs. bad news:
  - Few commodities show differences between good and bad news; gold is a prominent exception.
  - Gold: bad news aggregate coefficient is statistically significant and much larger than good news.
    - Good news t-0 for gold: -0.02
    - Bad news t-0 for gold: -0.17 ***
  - U.S. dollar reactions are more symmetric across good and bad news.
  - Interpretation: gold behaves as a safe-haven; negative surprises increase gold sensitivity and volatility.
- Volatility conditioning:
  - For the U.S. dollar, sensitivity is higher following periods of elevated volatility.
  - For almost all commodities except gold, good/bad news distinction does not significantly alter responses.
  - Gold impact of news is stronger following periods of elevated volatility, mainly when controlling for the U.S. dollar.
- Selected entries from Table 6 (coefficients multiplied by 100, percent change per 1-standard deviation surprise; significance markers preserved):
  - Aggregate news t: Gold -0.08 ***
  - Aggregate news t-1: Gold -0.03
  - High/Low volatility regressors:
    - Hi vol - news t-0: Gold -0.07
    - Lo vol - news t-0: Gold -0.09 ***

### Two-step estimation diagnostics and additional empirical outputs
- Evidence of GARCH effects in intraday prices; longer lag structures sometimes required beyond GARCH(1,1).
- Six of the largest 25 absolute returns associated with central banks’ selling of gold reserves.
- Table A3–A6 key empirical output format:
  - Coefficients multiplied by 100 represent percent change in price of nearest futures contract and U.S. dollar index for a 1-standard deviation surprise.
  - Significance notation: ***, **, * denote 99 percent, 95 percent, and 90 percent significance respectively.
  - Examples reported across tables:
    - Employment cost index (same day): Palladium 0.80 ***
    - Advance GDP (same day): Silver 0.35
    - Lagged price t-1 effects: Palladium 0.11 ***
    - ECB interest rate decision (same day): Gold 0.15 **; Natural gas -0.49
    - Change in non-farm payrolls (same day, Dec 2001–May 2009): Crude -0.32 **; Heating oil -0.41 **
    - Advance retail sales (same day, Dec 2001–May 2009): Corn -0.34 ***
    - ECB interest rate decision (same day, Dec 2001–May 2009): Corn -0.50 **; Soybeans -0.55 ***; Aluminium -0.19 **

### Conclusion and policy/market implications
- Commodities are not just financial assets; gold is not just another commodity.
- Some commodity prices respond to standardized macroeconomic news surprises; non-gold commodities show pro-cyclical bias, especially after controlling for the U.S. dollar.
- Crude oil shows little responsiveness to most scheduled announcements at daily horizons.
- Financialization has likely increased commodity sensitivity to macroeconomic news and surprise interest rate changes in recent years.
- Gold specifics:
  - Sensitive to scheduled U.S. and Euro area announcements including retail sales, non-farm payrolls, and inflation.
  - High sensitivity to real interest rates and unique safe-haven role lead to counter-cyclical short-term reactions and heightened response to negative surprises.
- Market and policy implications:
  - Traders may time order flow to avoid releases known to affect gold returns to reduce return uncertainty.
  - Long-term participants: confirmation of pro-cyclical bias in many commodities and gold’s safe-haven role during economic uncertainty.
  - Key future issue: extent to which increasing financialization further heightens commodity sensitivity to macroeconomic developments.

*Source: _wp09140 - 97. Two-step estimation (excerpted content).*

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

### _wp09140 - References..............................................................................................................

### Introduction
- Commodity prices have been affected by the recent protracted period of financial turmoil; for some commodities pro-cyclical demand drove price moves, while for gold the crisis underscored its role as a safe-haven asset and store of value.
- The paper uses an event study methodology (applied previously to asset prices) to investigate how scheduled and periodic macroeconomic announcements affect commodity prices.
- Timing of scheduled announcements is known in advance, making them potentially key factors for traders and long-term market participants.
- Gold behaves very differently from other commodities:
  - Gold prices react to specific scheduled announcements in the United States and the Euro area (such as indicators of activity or interest rate decisions) consistent with its role as a safe-haven and store of value.
  - Other commodity prices, where macroeconomic news is significant, exhibit pro-cyclical sensitivities, albeit much less than financial assets.
- Paper organization: Section II reviews literature, data, and methodology; Section III reports and discusses results; Section IV concludes.

### Methodology — Literature Review (asset prices and macroeconomic announcements)
- Prior literature finds macroeconomic news has significant price and volatility effects in bond and currency markets:
  - Rossi (1998): certain key economic announcements cause U.K. government bond yield changes of between 2–6 basis points, including beyond the trading day.
  - Fleming and Remelona (1999): arrival of public information has a large effect on prices and subsequent trading activity, particularly when implied volatility is high.
  - Balduzzi, Elton, and Green (2001): a wide variety of economic announcements affect U.S. Treasury bond prices, with labor market, inflation, and durable goods orders data having the largest impact.
- Commodities are not financial assets, but related empirical results are relevant because of relationships between commodity prices and some financial asset valuations.
  - Frankel (2008) argues interest rates can have a significant effect on commodity prices; Roache (2008) provides supporting empirical evidence.
  - The strongest and most consistent relationship is between the U.S. dollar and commodity prices.
- Macroeconomic news affects exchange rates:
  - Andersen et al (2002): macroeconomic news generally has a statistically significant correlation with intra-day movements of the U.S. dollar; “bad” news (weaker-than-expected growth) has a larger impact than “good” news.
  - Galati and Ho (2003): similar results using daily data.
  - Ehrmann and Fratzscher (2005): U.S. news tended to have more effect on the euro-dollar exchange rate than German news; activity indicators such as GDP and labor market data had particularly large and significant effects, with news impact increasing during times of high market uncertainty.

### Focus on commodities and gold — literature themes and empirical regularities
- Compared with U.S. Treasury bonds, exchange rates, and equity markets, the number and significance of macroeconomic announcements affecting commodity prices is lower.
- Key U.S. indicators (inflation, GDP, employment statistics) repeatedly show the ability to move some commodity prices.
- Energy products have tended to be less sensitive; gold has been most sensitive.
- Earlier studies (1980s and 1990s) generally confirm conventional wisdom that gold is a hedge against higher inflation and economic uncertainty:
  - Gold prices tend to rise if U.S. inflation and output unexpectedly increase, or if the labor market tightens by more than the market projects (Ghura (1990); Christie-David, Chaudry and Koch (2000)).
- Gold also appears sensitive to supply and demand news:
  - Some studies indicate central bank announcements regarding sales of gold reserves have tended to cause price declines (Cai, Cheung, and Wong (2001)).
  - Sensitivity of gold to news varies through time; Hess, Huang, and Niessen (2008) present evidence that sensitivity increases during recessions.

### Key empirical findings from Table 1 (selected study results — preserved exactly)
- Frankel and Hardouvelis (1985), daily reactions of nine commodities to U.S. money supply announcements from 1980-1982:
  - Gold and other commodities negatively related.
  - "1 percentage point positive shock in the money supply leads to a 0.7 percent decline in gold."
  - Authors contend that following a positive money supply surprise, the market anticipates quick Fed action that would lead to a tightening in policy.
- Barnhart (1989), sensitivity of 15 commodity futures prices to the surprise component of announcements for 12 U.S. economic variables (OLS single equation and SUR system estimations, daily data 1980-1984):
  - Rejects hypothesis that all parameters are equal to zero for just four commodities, including gold.
  - "Just two announcements were significant for the gold price: the M1 money aggregate with a negative coefficient; and the Federal Reserve surcharge rate indicating that a surprise 100 basis point increase in the rate would lead to a fall in the gold price of nearly 1 percent."
  - Similar results for the metals sub-group—including gold, silver, and copper—with Fed Discount Rate announcements also significant.
- Ghura (1990), regression of daily commodities futures price on 14 U.S. macroeconomic announcements from 1985-89:
  - "Gold sensitive only to employment reports, with positive surprise leading to higher price; no significant effect from inflation or activity."
  - Notes results may be biased lower by inclusion of other financial variables (e.g. exchange rates) as regressors.
- Christie-David, Chaudry, and Koch (2000), sensitivity of gold and silver futures prices over 15 minute intervals to 23 U.S. macroeconomic news announcements from 1992-1995:
  - Formal variance tests show gold and silver price volatility is higher during days in which there are announcements.
  - Metals prices are sensitive to a fewer number of announcements than bond futures.
  - "GDP, inflation, and capacity utilization are all significant, with the expected positive sign."
- Cai, Cheung, and Wong (2001), regression of 5 minute gold futures prices on 23 U.S. macroeconomic announcements over 1994- (table truncated in source).

*Source: _wp09140 - References (extracted content).*

### 97. Two-step estimation

### 97. Two-step estimation

### Study data and literature findings
- Two-step estimation using GARCH and a flexible Fourier form to capture smooth intraday patterns.
  - Clear evidence of GARCH effects in intraday prices.
  - Six of the largest 25 absolute returns associated with central banks’ selling of gold reserves.
  - Number and significance of announcements lower for gold than for bonds or currencies.
  - Coefficients on most announcements had the correct sign, with three statistically significant: employment reports, inflation, and GDP.
- Hess, Huang, and Niessen (2008): OLS regression of two commodity indices—the CRB and Goldman Sachs index—on 17 U.S. macroeconomic announcements using daily data from 1989 to 2005.
  - Commodities sensitive to far fewer announcements than either bonds or equities.
  - Impact of news on prices is dependent upon the state of the economy.
  - Unconditionally, only inflation surprises are statistically significant, although their economic impact is very small.
  - Conditioning on the NBER-defined business cycle shows increased effect during recessions; surprise news on GDP, payrolls, and retail sales are significant (small), with the expected positive sign.
- Kilian and Vega (2008): Regressions of WTI crude oil and U.S. gasoline prices on 30 U.S. macroeconomic announcements using daily data from 1983 to 2008.
  - No evidence of statistically significant responses of either oil or gasoline to U.S. macroeconomic news at daily horizons.
  - Some evidence that a broad set of selected forward-looking indicators were statistically significant over a horizon of one month.
  - Economic significance of leading indicators was low, with minimal explanatory power.

### Data (authors’ dataset and construction)
- Daily price data for 12 commodity futures contracts from January 1997 to June 2009, including precious metals, base metals, energy, and agricultural commodities.
- Futures prices taken from the nearest contract traded on exchanges in the United States.
- Rationale for focusing on futures rather than spot:
  - London-dominated spot markets (e.g., fixing prices) have time-zone delays relative to U.S. announcements.
  - Spot prices often positively correlated with futures with a one-day lag; futures tend to lead spot (examples and prior literature cited).
- Gold futures and spot correlation matrix (same day / previous day) presented; correlation coefficients in bold are significant at the 5 percent level.
  - Selected entries from Table 2 (formatted as correlations in source): Gold future–Gold spot same day 0.26; Gold future–U.S. dollar same day -0.44; Gold spot–U.S. dollar same day -0.19; previous-day correlations include Gold future–Gold spot 0.72 (previous day).
- Frequency choice justification:
  - Daily data preferred over intraday due to liquidity effects and possibility that markets take longer than minutes to absorb significance of news; assumption that non-announcement shocks on release dates are white noise and unbiased.
- Macroeconomic announcement surprise measure:
  - Surprise Z_it = (X_it - E_{t-1}(X_it)) / σ_X, i.e., standardized surprise interpreted as standard deviations from the consensus.
  - Analyst consensus estimates from Bloomberg used.
  - Set of 13 monthly or quarterly U.S. macroeconomic announcements selected (with some substitutions, including Employment Cost Index and Existing Home Sales).
  - Also include ECB and Bank of England interest rate decisions and the German IFO business climate survey.
  - Chinese macroeconomic announcements excluded due to limited observations.

### Estimation strategy
- Benchmark mean equation (OLS formulation) for log change in futures price Δp_t:
  - Δp_t = α + Σ_{j=0}^{J} β_j Z_{j,t-k} + Σ_{k=1}^{K} λ_k Δp_{t-k} + ε_t, with ε_t ~ i.i.d.(0, σε^2). (Refer to equation (2) in source.)
- Heteroscedasticity and volatility clustering observed in commodity returns; OLS is inefficient.
- GARCH approach:
  - Joint modeling of returns and volatility; conditional variance h_t modeled as function of lagged squared residuals and past h.
  - Model structure referenced as equation (3) in source; innovations ν_t ~ N(0,1).
  - Preference for GARCH(1,1) noted but evidence suggests longer lag structure in many commodities.
  - Bollerslev and Wooldridge (1992) standard errors used to account for remaining heteroscedasticity.
  - Likelihood ratio tests identified K =2 and L =2 in most cases.
- Controlling for U.S. dollar:
  - Include the Federal Reserve’s trade-weighted U.S. dollar index log change Δe_t with M lags as exogenous variable to control for dollar effects.
  - Mean equation augmented to: Δp_t = α + Σ_{m=0}^{M} θ_m Δe_{t-m} + Σ_{l=1}^{L} λ_l Δp_{t-l} + Σ_{k=0}^{K} β_k Z_{t-k} + ε_t. (Refer to equation (4) in source.)
  - Assumption: causality runs from U.S. dollar to commodity price; literature cited acknowledging controversy but supporting evidence.
- Conditioning on volatility and good/bad news:
  - Composite indicator: sum of standardized surprises for each announcement (excluding monetary policy shocks), used to aggregate news and preserve degrees of freedom.
  - Base reduced model for gold: Δp_t = α + Σ_{l=1}^{2} γ_l Z_{t-l} + Σ_{k=1}^{2} φ_k Δp_{t-k} + ε_t. (Refer to equation (5) in source.)
  - Volatility conditioning: classify days as high-volatility if prior 30-, 60-, or 90-day daily standard deviation > sample average; mean equation extended to include low-vol and high-vol news regressors (refer to equation (6)).
  - Good vs. bad news classification: “Good news” defined as surprises that should raise prices of cyclically-sensitive assets (e.g., higher-than-expected GDP, industrial production, non-farm payrolls, consumer confidence, inflation). Mean equation with separate good and bad news regressors presented (refer to equation (7)).

### Results — Scheduled macroeconomic announcements (selected findings and coefficients)
- General findings:
  - A number of U.S. and Euro area announcements impact commodity prices; energy products tend to exhibit little sensitivity, agricultural products and base metals show pro-cyclical sensitivity, gold tends to be counter-cyclical.
  - U.S. retail sales, non-farm payrolls, housing starts, and the ISM survey are among the most influential indicators for gold.
  - German IFO survey is a strong influence, particularly for base metals, even when controlling for the U.S. dollar.
  - Controlling for the U.S. dollar often heightens measured pro-cyclical sensitivity for non-gold commodities.
  - Commodities tend to be inversely related to Federal Reserve interest rate surprises where significant.
  - Fewer surprises in scheduled FOMC rates; ECB surprises have more datapoints and, when controlling for the U.S. dollar, precious and base metals prices are inversely related to ECB interest rate shocks.
- Selected coefficients from Table 3 (each coefficient multiplied by 100 and represents percent change in nearest futures price for a 1-standard deviation surprise; only those significant at the 10 percent level or more are shown):
  - Gold:
    - Change in non-farm payrolls: -0.18 *
    - Consumer confidence: -0.15 **
    - Industrial production: -0.31 *
    - Consumer price index: 0.10
    - Advance GDP: -0.13
    - ISM manufacturing survey: -0.09
    - ECB interest rate decision: 0.15 **
    - German IFO survey: 0.13
  - Crude Oil:
    - Change in non-farm payrolls: 0.12
    - Industrial production: 0.27
    - Existing home sales (lagged): 0.34 *
    - Advance retail sales (lagged): -0.47
  - Wheat, Corn, Copper, Aluminium: selected significant coefficients reported in Table 3 (see source for complete list).
  - U.S. dollar responses also reported in Table 3 (selected entries): Advance retail sales 0.05, Change in non-farm payrolls 0.13 ***, Consumer confidence 0.08 ***, Advance GDP 0.17 **, German IFO survey -0.11.
- Results robust across subperiods:
  - Sample split into 1997–November 2001 and December 2001–March 2009; number of indicators affecting prices and pro-cyclical sensitivity among non-gold commodities tended to rise since 2001, but main qualitative conclusions remain.
  - Chow tests around November 2001 indicated inability to reject model stability at the 5 percent level for almost all commodities.

### Results — Including U.S. dollar controls (selected findings from Table 4 and Table 5)
- Including the U.S. dollar index as regressor changes some significance patterns and magnitudes:
  - Gold:
    - Change in non-farm payrolls: -0.05 (Table 4)
    - Consumer price index: 0.08
    - German IFO survey: 0.00 (Table 4) but 0.00 in Table 5 as well (selected rows retained from source).
  - Crude Oil:
    - Change in non-farm payrolls: 0.24
    - Industrial production: 0.33
    - Advance retail sales: -0.38
  - U.S. dollar coefficients and impact retained in tables; Bollerslev-Woolridge standard errors reported, significance denoted by ***, **, * for 99 percent, 95 percent, and 90 percent levels respectively.
- Results indicate pro-cyclical sensitivities for many commodities are heightened once U.S. dollar is controlled for.

### Results — Good/bad news and volatility conditioning (selected findings)
- Good vs. bad news:
  - Few commodities show differences between good and bad news; gold is a prominent exception.
  - Gold: bad news aggregate coefficient is statistically significant and much higher than good news (Table 6).
    - Good news t-0 for gold: -0.02
    - Bad news t-0 for gold: -0.17 ***
  - U.S. dollar reactions are more symmetric across good and bad news.
  - Interpretation: gold behaves as safe-haven; negative surprises increase gold sensitivity and volatility.
- Volatility conditioning:
  - For the U.S. dollar, sensitivity is higher following periods of elevated volatility.
  - For almost all commodities except gold, good/bad news distinction does not significantly alter responses.
  - Gold impact of news is stronger following periods of elevated volatility, but this is evident mainly when controlling for the U.S. dollar.
- Selected entries from Table 6 (coefficients multiplied by 100, percent change per 1-standard deviation surprise; significance markers preserved):
  - Aggregate news t: Gold -0.08 ***
  - Aggregate news t-1: Gold -0.03
  - High/low volatility regressors (Hi vol - news t-0): Gold -0.07; Lo vol - news t-0: Gold -0.09 ***

### Conclusion and policy/market implications
- Commodities are not just financial assets; gold is not just another commodity.
- Some commodity prices are influenced by standardized surprises in macroeconomic news, with evidence of pro-cyclical bias across many non-gold commodities, especially after controlling for the U.S. dollar.
- Crude oil shows little responsiveness to most scheduled announcements at daily horizons.
- Financialization has likely increased commodity sensitivity to macroeconomic news and surprise interest rate changes in recent years.
- Gold:
  - Sensitive to scheduled U.S. and Euro area announcements including retail sales, non-farm payrolls, and inflation.
  - High sensitivity to real interest rates and unique role as safe-haven lead to counter-cyclical short-term reactions and heightened response to negative surprises.
- Implications:
  - Traders may time order flow to avoid releases known to affect gold returns to reduce return uncertainty.
  - Long-term participants: confirmation of pro-cyclical bias in many commodities and gold’s safe-haven role during economic uncertainty.
  - Key future issue: extent to which increasing financialization further heightens commodity sensitivity to macroeconomic developments.

*Source: Authors (excerpt from _wp09140 - 97. Two-step estimation).*

### REFERENCES

### _wp09140 - REFERENCES

### References (selected entries from source)
- Andersen, Torben G., Tim Bollerslev, Francis X. Diebold, and Clara Vega, 2003, “Micro Effects of Macro Announcements: Real-Time Price Discovery in Foreign Exchange,” American Economic Review, Vol. 93, pp. 38-62.
- Antoniou, Antonios, and Andrew J. Foster, 1992, “The Effects of Futures Trading on Spot Price Volatility: Evidence for Brent Crude Oil Using GARCH,” Journal of Business Finance & Accounting, Vol. 19 No. 4, pp. 473-484.
- Attié, Alexander P., and Shaun K. Roache, 2009, “Inflation Hedging for Long-Term Investors,” IMF Working Paper, No. 09/90.
- Balduzzi, Pierluigi, Edwin J. Elton, and T. Clifton Green, 2001, “Economic News and Bond Prices: Evidence from the U.S. Treasury Market,” The Journal of Financial and Quantitative Analysis, Vol. 36, No. 4, pp. 523-543.
- Barnhart, Scott W., 1989, “The Effects of Macroeconomic Announcements on Commodity Prices,” American Journal of Agricultural Economics, Vol. 71, No. 2, pp. 389-403.
- Bollerslev, Tim and Jeffrey M. Wooldridge, 1992, “Quasi-Maximum Likelihood Estimation and Inference in Dynamic Models with Time Varying Covariances,” Econometric Reviews, 11, pp. 143-172.
- Cai, Jun, Yan-Leung Cheung, Michael C. S. Wong, 2001, “What moves the gold market?” Journal of Futures Markets, Volume 21 Issue 3, pp. 257-278.
- Chen, Yu-Chin, Kenneth Rogoff, and Barbara Rossi, 2008, “Can Exchange Rates Forecast Commodity Prices?,” Economic Research Initiatives at Duke (ERID) Working Paper, No. 1.
- Christie–David, Rohan, Mukesh Chaudhry, and Timothy W. Koch, 2000, “Do Macroeconomics News Releases Affect Gold and Silver Prices?,” Journal of Economics and Business, 52, pp. 405–421.
- Clements, Kenneth W., and Renee Fry, 2008, “Commodity Currencies and Currency Commodities,” Resources Policy, Volume 33, Issue 2, pp. 55-73.
- Cornell, Bradford, and Kenneth R. French, 1986, “Commodity Own Rates, Real Interest Rates, and Money Supply Announcements,” Journal of Monetary Economics, 18, pp. 3-20.
- Ehrmann, M., and M. Fratzscher, 2005, “Exchange Rates and Fundamentals: New Evidence from Real-Time Data,” Journal of International Money and Finance, 24, pp. 317-341.
- Engle, Robert F. and Kenneth F. Kroner, 1995, “Multivariate Simultaneous Generalized ARCH,” Econometric Theory, Vol. 11, No.1, pp. 122-150.
- Fleming, M. J., and E. M. Remelona, 1999, “Price Formation and Liquidity in the U.S. Treasury Market: The Response to Public Information,” Journal of Finance, 54, pp. 1901-1915.
- Frankel, Jeffrey A., 2008, “The Effect of Monetary Policy on Real Commodity Prices,” in Asset Prices and Monetary Policy, John Campbell, ed., U. of Chicago Press, pp. 291-327.
- Frankel, Jeffrey A., and Gikas A. Hardouvelis, 1985, “Commodity Prices, Money Surprises and Fed Credibility,” Journal of Money, Credit and Banking, Vol. 17, No. 4, Part 1, pp. 425-438.
- Galati, Gabriele, and Corrine Ho, 2003, “Macroeconomic News and the Euro/Dollar Exchange Rate” Economic Notes, vol. 32, no. 3, pp. 371–398.
- Ghura, Dhaneshwar, 1990, “How Commodity Prices Respond to Macroeconomic News,” World Bank PRE Working Paper, No. 354.
- Hansen, Peter, and Asger Lund, 2005, “A Forecast Comparison of Volatility Models: Does Anything Beat a GARCH (1,1)?,” Journal of Applied Econometrics, 20, pp. 873–889.
- Hess, Dieter, He Huang, and Alexandra Niessen, 2008, “How Do Commodity Futures Respond to Macroeconomic News?,” Journal of Financial Markets and Portfolio Management, Volume 22, Number 2.
- Kaul, Gautam, 1987, “Stock Returns and Inflation—The Role of the Monetary Sector,” Journal of Financial Economics, Vol. 18, No. 2, pp. 253–276.
- Kilian, Lutz, and Clara Vega, 2008, “Do Energy Prices Respond to U.S. Macroeconomic News? A Test of the Hypothesis od Predetermined Energy Prices,” Federal Reserve International Discussion Paper, No. 957.
- Payne, R., 2003, “Informed trade in Spot Foreign Exchange Markets: An Empirical Investigation,” Journal of International Economics, 61, pp. 307-328.
- Roache, Shaun, 2008, “Commodities and the Market Price of Risk,” IMF Working Paper, No. 08/221.
- Rossi, Marco, 1998, “Economic Announcements and the Timing of Public Debt Auctions,” IMF Working Paper, No. 98/44.
- Yang, Jian, R. Brian Balyeat, and David J. Leatham, 2005, “Futures Trading Activity and Commodity Cash Price Volatility,” Journal of Business Finance & Accounting, 32(1), pp. 297-323.

### Appendix — Contract specifications and summary statistics
- Table A1. Commodity Futures Contracts Specification (closing price time shown):
  - Gold: COMEX, 100 troy ounces, 17:15 EST
  - Silver: COMEX, 5,000 troy ounces, 17:15 EST
  - Platinum: NYMEX, 50 troy ounces, 17:15 EST
  - Palladium: NYMEX, 100 troy ounces, 17:15 EST
  - Oil: NYMEX, Light, sweet crude, 1,000 barrels, 17:15 EST
  - Heating oil: NYMEX, 42,000 barrels, 17:15 EST
  - Natural gas: NYMEX, 10,000 million British thermal units, 17:15 EST
  - Wheat: CBOT, 5,000 bushels, 13:15 CST
  - Corn: CBOT, 5,000 bushels, 13:15 CST
  - Soybeans: CBOT, 5,000 bushels, 13:15 CST
  - Copper: COMEX, High grade, 25,000 pounds, 17:15 EST
  - Aluminium: COMEX, 44,000 pounds, 17:15 EST
- Source note: COMEX division of NYMEX, NYMEX, and CBOT. COMEX is a division of NYMEX. CBOT is an abbreviation of the Chicago Board of Trade.

- Table A2. U.S. and Euro area Macroeconomic Announcements: Summary Statistics, January 1997–April 2009 (Monthly percent change, unless otherwise specified) — selected entries:
  - Advance retail sales: Average actual 0.2, Average surprise 0.0, Standard deviation surprise 0.6
  - Change in non-farm payrolls (thousands): Average actual 43.6, Average surprise -24.0, Standard deviation surprise 77.5
  - Consumer confidence (index): Average actual 89.8, Average surprise -0.3, Standard deviation surprise 5.0
  - Consumer price index: Average actual 0.2, Average surprise 0.0, Standard deviation surprise 0.2
  - Employment cost index (quarterly percent change): Average actual 0.7, Average surprise 0.0, Standard deviation surprise 0.2
  - Existing home sales: Average actual -1.1, Average surprise 0.1, Standard deviation surprise 3.3
  - FOMC interest rate decision (absolute basis point change): Average actual 18, Average surprise -16
  - Advance GDP (annualized quarterly percent change): Average actual 2.5, Average surprise 0.0, Standard deviation surprise 0.0
  - Housing starts (thousands): Average actual 1,648, Average surprise 9.4, Standard deviation surprise 101
  - Industrial production: Average actual 0.1, Average surprise 0.0, Standard deviation surprise 0.0
  - ISM manufacturing survey: Average actual 52.9, Average surprise 0.1, Standard deviation surprise 2.1
  - PPI ex-food and energy (MoM): Average actual 0.2, Average surprise 0.0, Standard deviation surprise 0.5
  - ECB interest rate decision (absolute basis point change): Average actual 30, Average surprise 38
  - German IFO survey: Average actual 96.7, Average surprise 0.1, Standard deviation surprise 1.2
  - U.K. interest rate decision (absolute basis point change): Average actual 10, Average surprise 49
- Note: “Actuals denote the data as of the release date and do not reflect subsequent revisions.” Source: Bloomberg; Authors’ estimates.

### Appendix — Key empirical outputs from tables A3–A6 (coefficients and interpretation format)
- Table A3. Commodity Price Sensitivity to Economic Announcements, January 1997–May 2009:
  - Coefficients are multiplied by 100 and represent the percent change in the price of the nearest futures contract and U.S. dollar index for a 1-standard deviation surprise.
  - Significance notation: ***, **, * denote 99 percent, 95 percent, and 90 percent significance respectively.
  - Examples of reported coefficients (percent change for a 1-standard deviation surprise):
    - Change in non-farm payrolls (same day): Gold -0.18 *
    - Consumer confidence (same day): Gold -0.15 **
    - Employment cost index (same day): Palladium 0.80 ***
    - Advance GDP (same day): Silver 0.35
    - Lagged price t-1 effects: Gold 0.00; Silver -0.02; Platinum 0.02; Palladium 0.11 ***
    - ECB interest rate decision (same day): Gold 0.15 **; Natural gas -0.49
  - Source: Authors' estimates. Standard errors: Bollerslev-Wooldridge robust to heteroscedasticity.

- Table A4. Commodity Price Sensitivity to Economic Announcements (with U.S. dollar control), January 1997–May 2001:
  - U.S. dollar index change contemporaneous coefficients (examples):
    - U.S. dollar index change t: Crude -1.04 ***; Silver -1.13 ***; Gold -1.02 ***
    - U.S. dollar index change t-1: Crude -0.10 **; Gold 0.18 **
  - Example announcement effects (percent change):
    - ECB interest rate decision (same day): Gold 0.13 ***; Wheat 0.20 **; Soybeans -0.09 ***
    - Change in non-farm payrolls (same day): Soybeans 0.29 *; Wheat 0.29 *
  - Lagged announcement and lagged price coefficients reported similarly.

- Table A5. Commodity Price Sensitivity to Announcements, December 2001–May 2009:
  - Examples (percent change for 1-standard deviation surprise):
    - Change in non-farm payrolls (same day): Crude -0.32 **; Heating oil -0.41 **
    - Advance retail sales (same day): Corn -0.34 ***
    - PPI ex-food and energy (same day): Crude 0.09; Heating oil 0.25 **
    - ECB interest rate decision (same day): Corn -0.50 **; Soybeans -0.55 ***; Aluminium -0.19 **
  - Lagged effects and lagged price t-1, t-2 coefficients reported for each commodity.

- Table A6. Commodity Price Sensitivity to Announcements (with U.S. dollar control), December 2001–May 2009:
  - U.S. dollar index change contemporaneous coefficients (examples):
    - U.S. dollar index change t: Crude -1.23 ***; Heating oil -1.52 ***; Gold -1.15 ***
    - U.S. dollar index change t-1: Crude -0.20 ***; Gold 0.16
  - Examples of announcement coefficients (percent change):
    - Change in non-farm payrolls (same day): Oil 0.45 **; Oil (other column) 0.42 ***
    - ECB interest rate decision (same day): Gold -0.28; Wheat -0.08 *; Corn -0.45 **
  - Lagged announcement and lagged price terms included; significance indicated as in prior tables.

- Table A6 (continued). Models distinguishing High/Low volatility and Good/Bad news (Jan 1997–May 2009):
  - Aggregate news regressors (examples):
    - Aggregate news t: Crude -0.08 ***; Gold 0.06
    - Good news t (no US$ control): Copper 0.10 *
    - Bad news t (no US$ control): Crude -0.17 ***; Oil (other) -0.20 ***
  - High/Low volatility model examples:
    - Lo vol - news t (no US$ control): Crude -0.09 ***; Gold 0.04
    - Hi vol - news t-1 (with US$ control): Crude -0.11 *

### Figures — Diagnostics
- Figure A1. Sample Autocorrelation Functions for the Squared Residuals from an AR(1) equation of log returns, 1997-2009:
  - Panels for Gold, Silver, Platinum, Palladium, Crude oil, Heating oil, Natural gas, Wheat, Corn, Soybeans, Copper, Aluminium.
  - Dashed lines represent the 95 percent confidence intervals.
- Figure A2. Sample Partial Autocorrelation Functions for the Squared Residuals from an AR(1) equation of log returns, 1997-2009:
  - Panels for the same commodities as in Figure A1.
  - Dashed lines represent the 95 percent confidence intervals.

*Source: _wp09140 - REFERENCES (authors’ estimates, tables and figures as provided in the source PDF).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2009/_wp09140.pdf_