The effects of economic news on commodity prices

The Quarterly Review of Economics and Finance 50 (2010) 377–385

Contents lists available at ScienceDirect

The Quarterly Review of Economics and Finance journal homepage: www.elsevier.com/locate/qref

The effects of economic news on commodity prices Shaun K. Roache ∗ , Marco Rossi International Monetary Fund, United States

a r t i c l e

i n f o

Article history: Received 13 July 2009 Received in revised form 7 February 2010 Accepted 27 February 2010 Available online 27 March 2010 JEL classification: G13 G14 G17

a b s t r a c t We assess how commodity prices respond to macroeconomic news and show that commodities have been relatively insensitive to such news over daily frequencies between 1997 and 2009 compared to other financial assets and major exchange rates. Where commodity prices are influenced by news, there is a procyclical bias and these sensitivities have risen as commodities have become increasingly financialized. However, models based on news still do a relatively poor job of forecasting commodity prices at daily frequencies. We also find some asymmetries in how commodity prices respond to news, most notably for gold, which alone among commodities acts as a safe-haven when “bad” economic news emerges. © 2010 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved.

Keywords: Commodities Macroeconomic announcements Futures pricing Financial forecasting

1. Introduction Commodity prices have not been immune to the recent protracted period of financial turmoil and economic recession. For some commodities, the pro-cyclical nature of demand has driven price moves, while for others, including gold, the crisis has underscored roles as safe-havens and stores of value. Recent commodity price volatility has overlapped with an increasing participation in commodity markets by financial investors rather than commercial traders, who are mainly seeking to enhance investment returns and/or achieve greater portfolio diversification. These two developments, elevated financial and economic volatility and the financialization of commodities, have triggered a vigorous debate about the determinants of commodity prices and the extent to which they may have changed (UNCTAD, 2009). Traditionally, the perception has been that the main drivers of prices are commodity-specific, with developments in the broader macroeconomy, typically serving as a proxy for demand, having less influence (Borensztein & Reinhart, 1994). However, as commodities have financialized, evidence has emerged that commodity prices have become increasingly exposed to market-wide

∗ Corresponding author at: International Monetary Fund, Research, 700 19th Street NW, Washington, DC 20431, United States. Tel.: +1 202 623 8207; fax: +1 202 589 8207. E-mail address: [email protected] (S.K. Roache).

shocks (Tang & Xiong, 2009), suggesting that announcements of macroeconomic indicators may have an impact on prices. In this paper we test the hypothesis that macroeconomic news is important for commodity prices, assess whether the sensitivity to such news has changed since 2002 (when commodity prices began their broad ascent), and identify the nature of these sensitivities. The results should prove useful to commodity producers, consumers, and financial investors keen to enhance their understanding of relatively high-frequency commodity price movements. Previous literature on the impact of macroeconomic announcements has mostly focused on bond and currency markets, with fairly clear evidence that macroeconomic news has significant price and volatility effects. Rossi (1998) finds that certain key economic announcements cause U.K government bond yield changes of between 2 and 6 basis points, including beyond the trading day. Fleming and Remolona (1999) find that the arrival of public information has a large effect on prices and subsequent trading activity, particularly during periods in which uncertainty (as measured by implied volatility) is high. Balduzzi, Edwin, Elton, and Green (2001) show that 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 these results are relevant given the relationship between commodity prices and some financial asset valuations. Frankel (2008) argues that inter-

1062-9769/$ – see front matter © 2010 The Board of Trustees of the University of Illinois. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.qref.2010.02.007

378

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

est rates can have a significant effect on commodity prices and Roache (2008) provides supporting empirical evidence. The strongest and most consistent relationship, however, is between the U.S. dollar and commodity prices, and there is a building consensus that macroeconomic news does affect exchange rates. Andersen, Bollerslev, Diebold, and Vega (2003) explore the relationship between macroeconomic news and the U.S. dollar exchange rate against six major currencies. They confirm macroeconomic news generally has a statistically significant correlation with intraday movements of the U.S. dollar, with “bad” news—for example, data indicating weaker-than-expected growth—having a larger impact than “good” news. Galati and Ho (2003) found similar results using daily data. Ehrmann and Fratzscher (2005) focused on the euro–dollar exchange rate and found that U.S. news tended to have more of an effect on the exchange rate than German news. Activity indicators such as GDP and labor market data had a particularly large and significant effect, with the news impact increasing during times of high market uncertainty. Rime, Sarno, and Sojli (2010) use a unique foreign exchange dealer dataset to link two previous findings in the empirical exchange rate literature: that both order flow and macroeconomic news are significant determinants of high-frequency exchange rate movements. They find that a significant amount of the variation in order flow variation can be explained using macroeconomic news, and that order flow appears to aggregate changes in expectations about fundamentals. Common themes have emerged from the literature focused on commodities and announcements. Macroeconomic announcements that have a significant impact on commodity prices are fewer than those that affect U.S. Treasury bonds, exchange rates, and equity markets, as noted by Cai, Cheung, and Wong (2001) and Christie-David, Chaudhry, and Koch (2000). The early literature focused on the impact on monetary variables and found that surprise interest rate and money supply declines led to higher commodity prices, as discussed in Frankel and Hardouvelis (1985) and Barnhart (1989). Some U.S. indicators, particularly real activity statistics, have shown the ability to move commodity prices, particularly at turning points in the business cycle; see Ghura (1990) and Hess, Huang, and Niessen (2008). In general, energy products are less sensitive to macroeconomic news (Kilian & Vega, 2008), while precious metals have tended to show relatively high sensitivities (Cai et al., 2001). This paper contributes to the literature in a number of ways. First, it provides an assessment of sensitivity to macroeconomic news during a period in which commodities have become increasingly financialized; this allows us to test the hypotheses that the reaction to news has changed as a result. Second, we use data for individual commodities rather than commodity indices which have been the focus of recent literature. Third, we identify whether the sensitivity to news is asymmetric along two dimensions: the type of surprise (“good” or “bad”) and the recent volatility behavior of the commodity price. This extends the literature for which conditioning information has tended to focus on business cycles. Fourth, we conduct out-of-sample tests to assess the forecasting performance of models based on news variables as regressors. Our results suggest that some commodity prices are influenced by the surprise element in macroeconomic news, with evidence of a pro-cyclical bias. However, sensitivities are heterogeneous across commodities and there is little evidence that commodities share similar sensitivities to market-wide shocks; i.e., there are very few news releases that affect a large number of commodity prices. Retail sales, non-farm payrolls, and inflation, generally regarded as among the more important news releases, tend to have the most significant and broadest effects. Commodities tend to be less sensitive

than financial assets, although the recent period of financialization has been associated with rising sensitivity, particularly to a small number of data releases. Asymmetries in the response of commodities are most apparent for those that have exhibited “safe haven” properties, such as gold, and recent volatility behavior appears to have little influence on price sensitivity to news, in sharp contrast to other financial assets and exchange rates. These results have two main implications. First, they add to the evidence supporting the hypothesis that the past few years of financialization have begun to change the high-frequency determinants of commodity prices, particularly to market-wide shocks. Second, they still caution against viewing commodity prices as responding reliably to news of macroeconomic developments over daily frequencies; even though sensitivities appear to be increasing, news variables still do a poor job of forecasting commodity prices at a daily frequency, a result which should inform the activities of commodity market participants. The paper is organized as follows. Section 2 reviews the data, and presents the methodology. Section 3 reports and discusses the results, while Section 4 concludes.

2. Methodology 2.1. Data We use daily price data for 12 commodity futures contracts available over the period from January 1, 1997 to December 31, 2009; our start date is determined by the availability of macroeconomic news variables from our source (Bloomberg). We have included precious metals, base metals, energy, and agricultural commodities (see Table A1 for details). Futures prices are taken from the nearest contract often used as the benchmark for that commodity and traded on exchanges in the United States. We focus on the futures market, rather than the spot market, for two reasons. First, the spot market for some commodities, including certain precious and base metals, is dominated by trading in London, which means that official fixing prices have less time to respond to daily developments in the United States because of the different time zone. Second, spot prices are often positively correlated with futures prices with a 1-day lag, which indicates that the impact of U.S. announcements on the futures price is likely to affect the spot price the following day. This is consistent with previous research indicating that commodities futures markets lead developments in spot markets (e.g., Antoniou & Foster, 1992; Yang, Balyeat, & Leatham, 2005). While many recent announcement studies use intraday data, we use daily data, finding the arguments of Ehrmann and Fratzscher (2005) in support of daily frequencies to be convincing. They note that Payne (2003) provides evidence of liquidity effects causing trades during the minutes following a news event that are not necessarily a response to the fundamental content of that news; e.g., trades based on participants covering a short position to reduce risk. Also, it may take longer than a few minutes for markets to absorb the significance of news events. For many commodities, which are often perceived to react to the response of other financial variables such as exchange rates (see below), this may be particularly relevant. The main objection to daily data—that it is noisy and polluted with many other market events—is a minor concern if we make the reasonable assumption, based on market efficiency, that non-announcement shocks on the release dates of specific reports are white noise and unbiased. Commodity prices, in common with financial assets, incorporate expectations regarding the future. As a result, the impact of news announcements should focus on the surprise component of

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

the news. A popular technique, which we use here, is to measure the surprise by the distance between the actual outturn for indicator i at period t, Xit and the publicly observable consensus estimate Et−1 (Xit ), scaled by the sample estimate of the variation in the surprises  X−E(X) . The surprises Zit may be interpreted as standard deviations from the consensus: Zit =

Xit − Et−1 (Xit ) ˆ X−E(X)

(1)

We use the analyst consensus estimates published by Bloomberg for each announcement and select a set of 15 monthly or quarterly macroeconomic announcements from those included by Ehrmann and Fratzscher (2005), with some substitutions, including the Employment Cost Index and Existing Home Sales (see Table A2). We focus mainly on announcements about U.S. macroeconomic developments since these have been shown to have the greatest influence on variables such as the U.S. dollar. However, we also include ECB and Bank of England interest rate decisions and the German IFO business climate survey; the IFO indicator was the only Euro area indicator shown to influence the U.S. dollar–euro exchange rate in Ehrmann and Fratzscher (2005). Commodities are traded globally and news from other emerging economies, particularly China given the growth in its demand across a wide range of products, may also influence prices. For now, the number of observations available to assess formally the impact of Chinese macroeconomic announcements is limited, which makes us cautious about their inclusion. However, this clearly remains a fertile area for future research. 2.2. Estimation strategy The simplest way to assess the significance of specific announcements is by estimating regressions in which the log change in the futures price p is the dependent variable, J surprise elements of the news announcements Zi including K − 1 lags, and L lags of the price return are the exogenous variables, and ε is the unexplained portion of the price return: pt = ˛ + εt

J K  

ˇjk Zjt−k +

j=1 k=0

∼(0,  2 )

L 

l pt−l + εt

l=1

(2)

This may be estimated using OLS, but the most obvious objection to this approach is that the price return variance of many commodity futures exhibit periods of high and low volatility, or heteroscedasticity. This violates the assumptions of OLS and leads to inefficient estimators. We find very strong evidence for commodity price volatility time-variation and clustering; this is a well-known result for financial assets and commodity prices and the results are surpressed. When asset return volatilities exhibit time-variation and clustering, a GARCH specification, which jointly models price returns and volatility, is often appropriate. In this model, the conditional variance of asset price returns ht is assumed to be a function of lagged values of the unexpected return εt−1 to εt−q and the conditional variance ht−1 to ht−s . The model can be written as: pt = ˛ + εt = t

ˇjk Zjt−k +

j=1 k=0



ht = 0 +

J K  

ht

l pt−l + εt

l=1

where t ∼N(0, 1)

Q  q=1

L 

1q ε2t−q +

S  s=1

2s ht−s

(3)

379

There are many variations on the GARCH theme but some studies have indicated that a simple GARCH(1,1) model—with one lag of the squared residual and one AR term—often outperforms other more complex specifications (Hansen & Lund, 2005) and we use this specification. However, our analysis of the volatility process for most commodities suggests that the conditional variance is sensitive to unexpected return shocks with lags of greater than one day. Also, formal tests on the residuals of GARCH(1,1) estimations still show the presence of heteroscedasticity for some commodities. To account for these features of the data, we present Bollerslev and Wooldridge (1992) standard errors, which are consistent in the presence of any remaining heteroscedasticity. We used likelihood ratio tests to identify the appropriate lag lengths and found that K = 2 and L = 2 in most cases.1 2.2.1. Controlling for the U.S. dollar effect The model shown in (3) is missing one important aspect of commodity prices—a high sensitivity to other financial variables. For example, macroeconomic news may exert an indirect influence through a commodity’s role as an effective hedge against lower interest rates or a depreciating U.S. dollar, together with other effects of exchange rate changes on commodity prices (e.g., IMF, 2008). In other words, might sensitivity to announcements merely reflect a relationship between the commodity and other financial assets, rather than the announcements themselves? To address this, we also include the U.S. dollar exchange rate in our analysis, as there is strong evidence that commodity prices have been sensitive to the U.S. dollar over a long period (Roache, 2008). We assume that all causality runs from the U.S. dollar—measured using the Federal Reserve’s trade-weighted index against major trading partners—to the commodity price. This assumption is not uncontroversial, as commodity prices may influence exchange rates, at least for economies for which commodities account for a large share of exports or through the emerging sovereign wealth fund (SWF) channel. However, recent evidence suggests that exchange rates play the dominant role as forcing variable—see Chen et al. (2008) and Clements and Fry (2008). We add the U.S. dollar index log change as an exogenous variable (e), including M lags, which would tend to introduce multicollinearity assuming the exchange rate is affected by economic announcements. The mean equation of the GARCH model (3) then becomes: pt = ˛+

J K   j=1 k=0

ˇjk Zjt−k +

L  l=1

l pt−l +

M 

m et−m + εt

(4)

m=0

2.2.2. “Good news”–“bad news” and volatility effects Up to now, our analysis assumes that commodity price sensitivity to announcements is symmetrical and constant over time. However, the asymmetrical nature of commodity markets suggests that it is reasonable to question these assumptions. We explore two possible factors that might condition the response of commodity prices to announcements. Do recent volatility patterns influence this sensitivity? Does it matter whether the news is “good” or “bad”? By conditioning the price response, we lose observations and increase the number of coefficients to be estimated and with a sample size of a little over 10 years, this may leave insufficient information to capture these effects. Consequently, following earlier studies, we use a composite indicator for these conditioning

1 Details on the GARCH and likelihood ratio tests are available from the authors by request.

380

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

models (see Ehrmann & Fratzscher, 2005; Galati & Ho, 2003). This composite aggregates the surprise element of the announcements into a single series, greatly simplifies the model, and increases the number of observations. The following analysis uses only U.S. announcements. The composite is the sum of the standardized scores for each announcement, excluding monetary policy shocks, and we do not impose any sign changes on these scores.2 In other words, the composite adds together the standardized surprises as calculated by Eq. (1) for each day. On a day with no announcement, the composite will have a value of zero (signifying no news). On a day with just one announcement, the composite’s value will be the standardized surprise of that announcement. On a day with more than one announcement, the surprise will be the summation of the individual standardized scores. We compare our results against a base model that estimates the regression of the log change in the commodity price p on to a constant ˛, the contemporaneous value and two lags of the composite indicator Z, and two autoregressive terms: pt = ˛ +

2 

ˇk Zt−k +

k=0

2 

l pt−l + εt

(5)

l=1

For the good news–bad news model, we define “good news” for the U.S. economy as an announcement surprise that should lead to an increase in the price of cyclically sensitive assets; this would include higher-than-expected GDP growth, industrial production, non-farm payrolls, consumer confidence, or inflation. Of course, unexpectedly higher inflation is not necessarily “good” news for the U.S. economy, but we have classified this as good news, since it should, a priori, lead to an increase in commodity prices. The mean equation of the GARCH model we estimate can then be written as: pt = ˛ +

K 

good good Zt−k

ˇk

+

K 

k=0

where Zgood

bad ˇkbad Zt−k +

k=0

(Zbad ) and ˇgood

L 

l pt−l + εt

(6)

l=1

K  k=0

high high Zt−k

ˇk

+

(ˇbad ) are the composite surprise vari-

K  k=0

low ˇklow Zt−k +

L 

l pt−l + εt

pt = ˛ +

J 2  

ˇjk Zjt−k +

j=1 k=0

able and coefficient for good (bad) news, respectively. To assess whether volatility—often used as a measure of investor uncertainty—affects commodity price sensitivities, we condition our analysis on the level of commodity price volatility over the preceding 30, 60, and 90 days. We classify an announcement as arriving in a high-volatility period if the daily standard deviation of the commodity price for this period is above its sample average and vice versa for low volatility. The mean equation for the GARCH model then becomes: pt = ˛ +

price indices began to trend higher (Table 2). This second sample is designed to assess whether the results change when considering only the most recent period of financialization. Our results are based on estimations that include the contemporaneous value and two lags of the explanatory news variables in the mean equation; in other words, the commodity price return for day t is a function of the news variables that are released at some specified time during the same day t and news released during the previous 2 days (t − 1 and t − 2). This specification assumes that between the open and close of the trading session, commodity prices will have the opportunity to react to the macroeconomic news released during the day. For some contracts—namely corn, soybeans and wheat—there is less time to absorb the effect of news released on the same day as the contracts stop trading earlier. However, all data releases included as exogenous variables are known before the daily close for these agricultural contracts; i.e., in all cases the news is released during the U.S. morning and the close for these contracts is 2.15 pm EST; the exception is the Federal Reserve interest rate announcement which is regularly scheduled for 2.30 pm EST, and which is included only with a lag for corn, soybeans, and wheat. The U.S. dollar is also included as a control variable. We show the results of estimations dropping the U.S. dollar as an exogenous variable in Table A3. In the variance equation, we implement a GARCH(1,1) specification with one lag each for the autoregressive and moving average terms (but no exogenous variables). From Eq. (4) for the mean equation K = L = M = 2 and from Eq. (3) for the variance equation S = Q = 1, a specification which is shown in Eq. (8). This specification was based on the results from likelihood ratio tests, combined with a check of t-ratios, which for almost all exogenous variables were insignificant at lags greater than two in the mean equation.

(7)

l=1

where Zhigh (Zlow ) and ˇhigh (ˇlow ) are the composite surprise variable and coefficient during high (low) volatility periods, respectively. 3. Results 3.1. Sensitivities to scheduled macroeconomic announcements We estimate the sensitivity of commodity prices to data releases for two sample periods: the full sample since 1997 (Table 1); and a sample that starts in January 2002, when the broad commodity

2 The results are robust to the inclusion of monetary policy shocks. We excluded them to allow comparisons with previous literature.

2  l=1

l pt−l +

2  m=0

m et−m + εt

(8)

ht = 0 + 11 ε2t−1 + 21 ht−1 One striking result is the degree of heterogeneity across commodities in terms of which macroeconomic data releases they show sensitivity to. A broad range of data releases is found to be statistically significant for at least one commodity, but for most of these releases the effect is found only for a small minority; e.g., retail sales is statistically significant only for aluminum in the full 1997–2009 sample. One emerging hypothesis related to the financialization of commodities is that investors are increasingly treating them as one asset class and as such, prices should be affected by market-wide shocks at the same time. These results argue against that hypothesis by suggesting that commodity prices have not broadly moved together in response to macroeconomic surprises over 1997–2009. A more consistent result across commodities, however, is that most of the variables have the anticipated positive sign and show most sensitivity to a smaller number of indicators. Data releases pointing to stronger-than-expected U.S. or German economic activity tend to increase commodity prices, underscoring their procyclical tendencies. Among the indicators that have the broadest and strongest effects either during the same day of the announcement or the following day are non-farm payrolls, advance GDP, the ISM survey, and the German IFO survey; these data releases were also relatively more important in explaining high-frequency exchange rate movements in Andersen et al. (2003), Ehrmann and Fratzscher (2005), and foreign exchange order flow in Rime et al. (2010) and as such may be regarded by market participants as particularly timely or informative in their content. Inflation surprises tend to have a negative effect on commodity prices, a finding that

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

381

Table 1 Commodity price sensitivity to macroeconomic announcements (with U.S. dollar control) January 1997 to December 2009. Gold

Silver

Platinum

Palladium

Crude oil

Heating oil

Natural gas

Wheat

Corn

Soybeans

Copper

Aluminium

Macroeconomic announcements Advance retail sales Change in non-farm payrolls Consumer confidence Consumer price index Employment cost index Existing home sales FOMC interest rate decision Advance GDP Housing starts Industrial production ISM manufacturing survey ECB interest rate decision German IFO survey

0.01 −0.07 −0.06 0.08 0.01 −0.02 0.05 0.00 −0.05 −0.17 0.06 0.13*** 0.01

0.08 0.07 0.15 0.05 0.14 −0.02 −0.05 0.20 −0.06 −0.16 0.12 0.02 0.00

0.06 −0.01 0.27 0.12 0.15 −0.11 −0.41** 0.10 0.06 −0.14 0.03 −0.40 0.08

0.18 0.12 −0.29 0.15 0.42 0.10 −0.03 0.57* 0.08 0.42** 0.21 0.20 −0.06

−0.28 0.27 0.20 0.01 0.18 0.00 −0.28 −0.14 0.18 0.36 −0.13 −0.10 −0.03

−0.25 0.30 0.22 −0.08 0.07 −0.24 −0.20 −0.02 0.15 0.28 −0.06 −0.33 −0.01

−0.09 0.19 0.78*** −0.21 0.19 −0.51* −0.38 0.20 −0.40** 0.01 −0.23 −0.30 0.85***

0.04 0.29* −0.04 0.09 −0.05 0.04 – −0.02 −0.17 0.02 0.23 0.05 0.07

0.15 0.25** −0.01 0.09 0.14 −0.18 – −0.27 −0.15 −0.02 0.06 0.20* 0.12

0.02 0.11 0.02 0.18 0.43** 0.00 – −0.30** −0.18 −0.02 −0.02 −0.09*** 0.21

0.14 0.20* 0.01 −0.26** −0.24 0.19* 0.13 0.09 0.12 −0.03 0.31* 0.00 0.15

0.17** 0.12 0.07 −0.22* −0.17 −0.06 0.02 0.16 0.22** 0.27 0.14 −0.01 0.16

Lagged by 1 day Change in non-farm payrolls Employment cost index Existing home sales FOMC interest rate decision Advance GDP Housing starts ISM manufacturing survey ECB interest rate decision German IFO survey Lagged price t − 1 Lagged price t − 2 U.S. dollar index change t U.S. dollar index change t − 1 U.S. dollar index change t − 2

−0.02 0.08 0.03 −0.03 0.02 −0.11** 0.05 −0.13* 0.17*** −0.04* −0.02 −1.04*** −0.11*** −0.04

−0.06 −0.10 0.10 −0.04 0.08 −0.03 0.19 −0.08 0.22*** −0.02 0.03 −1.19*** −0.06 0.04

−0.17* −0.17 −0.54*** −0.56 −0.06 0.08 0.02 −0.04 0.30*** −0.01 −0.07*** −0.83*** −0.19** −0.05

−0.21 0.70* −0.25* 0.04 0.36 −0.02 0.08 0.09 0.27* 0.11*** −0.02 −1.07*** 0.11 0.17*

0.09 0.08 0.32* 0.27 0.13 −0.19 0.07 0.03 −0.09 −0.02 −0.05*** −0.75*** 0.02 0.00

0.24 −0.10 0.39* 0.22* 0.21 0.00 0.22 0.17 −0.22 −0.03 −0.03 −0.78*** −0.05 −0.07

0.00 0.72* −0.21 0.81*** 0.90* −0.08 0.62*** 0.15 0.21 −0.04*** 0.00 −0.52*** 0.10 0.01

0.11 −0.12 0.09 −0.11 0.11 0.10 −0.04 0.24* 0.09 0.01 −0.02 −0.60*** −0.02 −0.12*

0.15 −0.04 0.15 −0.20* 0.23* 0.20 −0.05 0.01 0.13 0.03* 0.00 −0.49*** −0.08 −0.04

0.24* −0.15* 0.33 −0.14 0.26*** −0.10 −0.06 0.21** 0.07 −0.02 0.02 −0.47*** −0.12** −0.08

−0.03 0.23 −0.14 0.01 0.08 −0.19 0.25** −0.04 0.15 −0.08* −0.01 −0.67*** −0.06 −0.01

0.25*** −0.08 −0.08 −0.06 0.21 −0.15 0.11 −0.05 0.09 −0.10 −0.03 −0.43*** −0.25*** 0.00

0.04*** 0.94***

0.05*** 0.95***

0.27*** 0.67***

0.13*** 0.86***

0.07*** 0.91***

0.06*** 0.92***

0.08*** 0.91***

0.02*** 0.98***

0.03*** 0.96***

0.06*** 0.93***

0.03*** 0.96***

0.05*** 0.94***

GARCH equation coefficients Moving average Autoregressive

Coefficients shown are those that were statistically significant at the 10 percent level or more for at least one commodity. The coefficient on each announcement is multiplied by 100 and represents the percent change in the price of the nearest futures contract index for a one-standard deviation surprise. Bollerslev–Woolridge standard errors, which are robust to remaining heteroscedasticity. * Significance at the 90 percent level. ** Significance at the 95 percent level. *** Significance at the 99 percent level.

contrasts with research from the 1970s. In part, this may reflect a more pro-active U.S. monetary policy regime, in which real interest rate expectations increase in anticipation of monetary tightening in response to higher inflation. After controlling for the U.S. dollar, interest rate announcement surprises did not have a meaningful effect, with only two exceptions (palladium and natural gas). To a large extent, we would claim that it is not possible to draw strong conclusions about the effect of interest rates on commodity prices from this analysis since the number of interest rate surprises was very small (five for the FOMC, 18 for the ECB surprises, and 16 for the Bank of England over 1997–2009, excluding unscheduled inter-meeting changes). In most cases, it appears that central bank decisions had been correctly priced-in by financial markets before the announcement was made. We also find that a relatively large proportion of the effect of announcements on commodity prices comes with a 1-day lag; in other words, a news surprise released during the current trading session for a particular commodity sometimes has an effect during the following day’s trading session. This effect is not so apparent in studies of exchange rates (e.g., Andersen et al., 2003) and raises the issue of commodity market efficiency and the speed with which relevant information is incorporated into prices, which we do not address in this paper. To assess the extent to which financialization has increased the sensitivity to macroeconomic news surprises, we estimated the model described by (8) over a shorter sample period starting in

January 2002 and ending in December 2009.3 This period was chosen as its start date corresponds to the beginning of the sustained and broad rise in commodity prices that continued until peaking in July 2008 (as measured by the CRB commodity price index). Compared to the full sample, this sub-sample exhibits a larger number of statistically significant coefficients on news variables increases, at 57 compared to 47 for the 1997–2009 period. More importantly, a small number of news variables have a broader and larger effect on commodity prices, including non-farm payrolls and the ISM survey. Payrolls had a statistically significant impact on 10 out of 12 commodities, either during the day of the release or the following day with almost all of these exhibiting the expected positive sign. Only two precious metals—often regarded as safe havens during periods of increased uncertainty—had negative signs, platinum and palladium. Gold also exhibited these same safe haven tendencies, but much of the effect disappears when the U.S. dollar control is included in the estimation. The average size of the coefficients on the payrolls variable—when significant and positive—was 0.46, which indicates that for this group of commodities, a 1 percent positive surprise in the change in payrolls increased commodity prices by about 0.5 percent. Our finding that announcements can have

3 Chow tests based on breakpoints around January 2002 indicate that it was possible to reject the null hypothesis of model stability over the entire 1997–2009 sample at the 5 percent level for almost all commodities.

382

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

Table 2 Commodity price sensitivity to macroeconomic announcements (with U.S. dollar control) January 2002 to December 2009. Gold

Silver

Platinum

Palladium

Crude oil

Heating oil

Natural gas

Wheat

Corn

Soybeans

Copper

Aluminium

Macroeconomic announcements Advance retail sales Change in non-farm payrolls Consumer confidence Consumer price index Employment cost index Existing home sales FOMC interest rate decision Advance GDP Housing starts Industrial production ISM manufacturing survey PPI ex-food and energy German IFO survey

−0.03 −0.14 −0.05 0.08 −0.10 0.11 0.02** 0.01 −0.03 0.00 −0.02 −0.03 0.01

0.10 −0.06 0.13 0.37* 0.08 0.06 0.01 0.32 −0.16 −0.24 0.04 0.12 −0.03

0.02 0.21** −0.09 0.08 0.29 0.09 −0.02* 0.23 0.02 0.07 0.16* −0.05 0.17

0.20 0.16 −0.26* 0.17 0.64 0.08 0.05 0.70** 0.18 0.36* 0.15 0.06 −0.05

−0.07 0.50* 0.27 −0.15 0.16 0.12 −0.21*** 0.04 0.06 0.28 −0.13 0.14 0.26

0.09 0.59** 0.35 −0.15 0.36 −0.18 0.07 0.36 0.04 0.24 −0.10 −0.01 0.20

−0.14 0.55* 0.54 −0.32 0.42 −0.57 0.11 0.39 −0.54*** −0.06 0.00 −0.30 0.85**

0.01 0.54*** 0.03 0.14 −0.14 −0.08 – −0.20 −0.17 0.07 0.41 −0.17 0.11

0.13 0.51*** 0.04 0.11 0.11 −0.09 – −0.58 −0.15 −0.09 0.13 −0.08* 0.17

−0.20 0.28 0.11 0.15 0.35* 0.24* – −0.85 −0.18 −0.07 0.12 −0.07*** 0.22

0.37* 0.09 0.25* 0.02 −0.55 0.33** −0.11* −0.05 0.20 0.03 0.41* −0.04 0.13

0.12 0.19 0.05 −0.10 0.11 0.05 −0.13** 0.02 0.27** 0.23 0.18 0.02 0.23*

Lagged by 1 day Advance retail sales Change in non-farm payrolls Consumer confidence Consumer price index Existing home sales Advance GDP Housing starts ISM manufacturing survey PPI ex-food and energy German IFO survey Lagged price t − 1 Lagged price t − 2 U.S. dollar index change t U.S. dollar index change t − 1 U.S. dollar index change t − 2

−0.01 −0.10 −0.12 −0.13 −0.04 −0.14 −0.13** 0.14 −0.13 0.12 −0.09*** −0.02 −1.25*** −0.22*** −0.02

0.12 −0.09 −0.11 −0.09 0.11 0.16 0.03 0.28* −0.02 0.27* 0.00 0.03 −1.68*** −0.13 0.10

0.11 −0.31*** −0.06 0.16 −0.36*** −0.02 0.03 −0.05 0.00 0.26** −0.01 −0.02 −0.94*** −0.14** 0.07

0.31 −0.55*** 0.02 0.02 −0.19 0.36 −0.10 0.16 −0.25* 0.33* 0.09*** 0.01 −1.24*** 0.06 0.17*

0.19 −0.14 −0.08 −0.15 −0.03 0.31 −0.36* 0.14 −0.05 −0.21 −0.05** −0.02 −1.03*** −0.02 0.12

0.51* 0.05 −0.27 0.17 0.16 0.61 −0.13 0.44* 0.11 −0.16 −0.06*** 0.01 −1.07*** −0.08 0.06

0.28 0.13 0.73* 0.01 −0.29 0.79 −0.27 0.76*** −0.81 0.38 −0.03 0.02 −0.63*** 0.13 −0.03

−0.01 0.22 −0.41** −0.48** 0.63 0.26 −0.21 −0.06 −0.15 −0.28*** −0.01 0.01 −0.76*** 0.10 −0.10

0.17 0.33* −0.41* −0.38* 0.48 0.22 0.02 −0.01 −0.04 0.47*** 0.04* 0.01 −0.58*** −0.03 −0.02

−0.01 0.36** −0.14 −0.28* 0.56** 0.28*** −0.50* 0.02 −0.04 0.09 0.00 0.00 −0.62*** −0.05 −0.02

−0.03 0.07 −0.21 0.24 −0.36*** −0.35 −0.15 0.26 0.05 0.19 −0.08*** −0.01 −0.92*** −0.03 0.06

0.06 0.25* −0.09 0.16 −0.23** −0.01 −0.07 0.07 −0.10 0.19 −0.11*** −0.05* −0.54*** −0.24*** −0.02

0.04*** 0.95***

0.06** 0.93***

0.09*** 0.88***

0.08*** 0.91***

0.05 *** 0.93***

0.04*** 0.94***

0.08*** 0.91***

0.02*** 0.98***

0.03*** 0.96***

0.05*** 0.94***

0.03*** 0.96***

0.06*** 0.93***

GARCH equation coefficients Moving average MA(1) Autoregressive AR(1)

Coefficients shown are those that were statistically significant at the 10 percent level or more for at least one commodity. The coefficient on each announcement is multiplied by 100 and represents the percent change in the price of the nearest futures contract index for a one-standard deviation surprise. Bollerslev–Woolridge standard errors, which are robust to remaining heteroscedasticity. * Significance at the 90 percent level. ** Significance at the 95 percent level. *** Significance at the 99 percent level.

lagged effects—which perhaps signals market inefficiencies—did not change using this sub-sample. Around half of the significant coefficients on news variables for the 2002–2009 sample were at a 1-day lag, a similar proportion found for the full sample. The evidence that some macroeconomic news variables have a statistically significant effect on commodity prices does not tell us whether they might be useful in providing forecasts. To assess this, we evaluate out-of-sample performance by estimating three models from 1997 through the end of 2000: (i) the model largely based on (8), but excluding the contemporaneous change in the U.S. dollar as an explanatory variable since our intention is to assess how potential knowledge of news surprises, rather than knowledge regarding the direction of the dollar, may help to forecast commodity prices; (ii) a model based on (8) but including only the four most significant news releases which were retail sales, non-farm payrolls, advance GDP, and the ISM survey; and (iii) a model excluding news, but retaining autoregressive and lagged U.S. dollar variables. We first estimated these models for the 1997–2000 sample period (the smallest full year sample that ensured the model converged) and then forecast the values of commodity price levels over the next 12 months through December 31, 2001 using actual news surprises (i.e., assuming that we could perfectly predict macroeconomic news) and actual values for the autoregressive terms and the lagged values of the U.S. dollar (i.e., the parameters of the estimated model were kept constant during the 12-month forecast period, but the explanatory variables were updated each day). We then

proceed sequentially, adding a year to our in-sample estimation sample and forecasting for the subsequent 12 months to provide separate 12-month forecasts for each year from 2001 to 2009. The mean absolute percentage errors (MAPE) in the forecast of the commodity prices from this process over the entire outof-sample period for each commodity and the U.S. dollar for each specification are shown in Table 3. The results indicate that specifications including all news variables (denoted as the “full model” in the table) or just the most statistically significant news (the “partial model”) both underperform a specification that excludes news variables (the “no news model”), based on the smallest MAPE. News variables slightly improve the performance of the U.S. dollar model, but only when the most important news variables are included (the partial model). It appears that although increasing commodity financialization may have increased the importance of market-wide shocks in the form of macroeconomic news surprises, as yet they cannot contribute to better forecasting of daily frequency commodity price movements for any of the 12 commodities included in our analysis.4 The main implication of this

4 Similar results have been obtained for exchange rates, but recent work by Della Corte, Sarno, and Tsiakis (2009) suggest that assessing the predictive performance of empirical models incorporating alternative macroeconomic variables and using Bayesian methodology that explicitly accounts for parameter and model uncertainty can provide stronger evidence for predictability.

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

383

Table 3 Out-of-sample performance evaluation mean absolute percent error 2001–2009. Model

Gold

Silver

Platinum

Palladium

Crude oil

Heating oil

Full Partial No news

0.8662 0.8507 0.8428

1.3724 1.3629 1.3519

1.1619 1.1159 1.0708

1.5932 1.5649 1.5476

1.8848 1.8586 1.8418

1.8780 1.8509 1.8336

Model

Natural gas

Wheat

Corn

Soybeans

Copper

Alum.

U.S. dollar

Full Partial No news

2.7232 2 7064 2.6839

1.5509 1 5386 1.5231

1.3583 1 3464 1.3375

1.2567 1 2386 1.2364

1.3989 1 3847 1.3722

0.9962 0 9846 0.9756

0.3532 0 3514 0.3532

result is that financial investors taking active investment decisions in commodity markets should not attach too high a weight to macroeconomic data releases, at least over daily frequencies, when forming their views regarding the likely direction of commodity prices. 3.2. “Good news”, “bad news”, and volatility To assess the extent to which commodity prices are asymmetrically sensitive to macroeconomic news, we estimated a model in which the mean equation is given by (6), the news variables are aggregated together each day as described in Section 2.2, and the lag length of the explanatory news and exchange rate variables and autoregressive terms is K = L = 2. We estimate the models with and without the U.S. dollar as a control variable. The vari-

ance equation is a GARCH(1,1) specification with one lag each for the autoregressive and moving average terms (but no exogenous variables). Results indicate that for some commodities there are some statistically significant asymmetries in their sensitivity to news variables. Most prominently, bad news affects the gold price much more than good news—the coefficient on the bad news aggregate is statistically significant and much higher than that on good news—a result that is robust to the inclusion of the U.S. dollar control variable (Table 4). This contrasts sharply with the results from the symmetric specifications in which very few news releases were found to have an influence on the gold price, but also for other commodities using the asymmetric specification, indicating the unique safe haven properties of gold. This result provides evidence that market participants are attracted to gold, causing its price to rise,

Table 4 Commodity price sensitivity to macroeconomic announcements (composite models) January 1997 to December 2009. Gold

Platinum

Palladium

Crude oil

Heating oil

Natural gas

Wheat

Corn

Soybeans

Copper

0.01 −0.04 0.06

0.08 0.03 −0.01

0.03 −0.03 −0.02

0.00 0.08 −0.03

−0.10 0.06 0.12

0.02 −0.01 0.05

−0.01 −0.01 −0.02

−0.01 0.00 −0.01

0.04 −0.02 0.07*

0.05 0.00 0.05

0.05*** 0.01 0.01

Good/bad news with no U.S. dollar control Aggregate news regressor Good news t 0.00 0.04 0.00 Good news t − 1 −0.02 −0.04 −0.03 0.02 Bad news t −0.15*** −0.06 Bad news t − 1 −0.03 0.08 −0.04

0.17** 0.05 −0.02 0.02

−0.02 −0.07 0.09 0.01

0.01 0.07 −0.02 0.08

−0.06 −0.02 −0.14 0.17

0.10 −0.04 −0.08 0.03

0.12** 0.10* −0.09* 0.02 −0.18*** −0.13** 0.11 −0.02

0.11* 0.02 −0.05 −0.06

0.13*** −0.06 −0.04 0.08

0.05*** 0.01 0.05*** 0.01

Model 3—good/bad news with U.S. dollar control Aggregate news regressor Good news t 0.03 0.06 0.03 Good news t − 1 −0.03 −0.04 −0.03 0.05 Bad news t −0.10* −0.01 Bad news t − 1 −0.03 0.09* −0.04

0.20** 0.04 0.04 −0.01

0.01 −0.06 0.12 0.01

0.03 0.08 0.01 0.09

−0.04 −0.03 −0.13 0.17

0.11 −0.04 −0.06 0.03

0.12** 0.11** 0.02 −0.09* −0.17** −0.12* 0.11 −0.02

0.12** 0.02 −0.02 −0.06

0.14*** −0.06 −0.03 0.09*

– – – –

High/low volatility with no U.S. dollar control Regressors Hi vol – news t −0.06 0.02 0.09 Hi vol – news t − 1 −0.07 0.03 −0.12 Lo vol – news t −0.07*** −0.02 −0.01 Lo vol – news t − 1 −0.02 0.01 −0.02

0.08 0.12 0.08 0.01

0.00 −0.19 0.04 0.01

0.08 0.02 −0.04 0.11

−0.04 0.07 −0.12 0.07

−0.01 −0.09 0.04 0.01

−0.10 −0.19* 0.01 0.03

−0.13 −0.11 0.01 0.02

0.08 0.08 0.03 −0.04

0.04 −0.06 0.05 0.01

0.13 0.13 0.12** −0.01

0.03 −0.19 0.07 0.02

0.10 0.02 −0.01 0.12

−0.04 0.08 −0.10 0.07

−0.01 −0.08 0.05 0.02

−0.09 −0.18* 0.02 0.04

−0.12 −0.11 0.02 0.02

0.10 0.08 0.04 −0.04

0.05 −0.06 0.07* 0.02

Aggregate news t Aggregate news t − 1 Aggregate news t − 2

Silver

−0.07** −0.01 −0.02 0.01 0.02 0.05

Model 2—high/low volatility with U.S. dollar control Regressors Hi vol – news t −0.02 0.01 0.10 0.02 −0.10 Hi vol – news t − 1 −0.11* Lo vol – news t −0.03 0.03 0.03 Lo vol – news t − 1 −0.01 0.02 −0.02

Aluminium

U.S. dollar

0.12*** −0.03 0.03*** 0.01

– – – –

Coefficients shown are those that were statistically significant at the 10 percent level or more for at least one commodity. The coefficient on each announcement is multiplied by 100 and represents the percent change in the price of the nearest futures contract index for a one-standard deviation surprise. Bollerslev–Woolridge standard errors, which are robust to remaining heteroscedasticity. * Significance at the 90 percent level. ** Significance at the 95 percent level. *** Significance at the 99 percent level.

384

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

when the prospect of bad economic outcomes increases. This is not simply a reaction to U.S. dollar asymmetries; indeed, we also show the effect on the U.S. dollar which is, perhaps unsurprisingly, symmetric for both types of news. Other commodities that exhibit asymmetric sensitivity include corn, soybeans, copper, aluminum, and palladium. For the metals, there is a clear and positive sensitivity to good news, underscoring their typically pro-cyclical characteristics, but there is little sensitivity to bad news. It is also notable that the size of the good news coefficients in each case is larger than that for the U.S. dollar. For the agricultural commodities, there appears to be tendency for prices to rise whatever the type of news, a result that is more difficult to interpret. In terms of volatility, we confirm the standard result that the sensitivity of the U.S. dollar to macroeconomic news is higher following a period of elevated volatility, with a coefficient four times as large. For almost all commodities, except gold, however, recent price volatility behavior does not have a significant impact on how sensitive prices are to news. The impact of news on gold is stronger following periods of volatility, but only when we control for the U.S. dollar. 4. Conclusion Our objective in this paper is to measure the sensitivity of commodity prices to macroeconomic news, assess the extent to which this may have changed during the most recent period of financialization, and to identify potential asymmetries in this sensitivity. Our results suggest that commodities have been relatively insensitive to market-wide shocks in the form of macroeconomic news over daily frequencies between 1997 and 2009. Overall, the number of news variables that have an impact on prices is small and their effect relatively weak, at least compared to the results in the literature on other financial assets and major exchange rates. Where commodity prices are influenced by the surprise element in macroeconomic news, there tends to be evidence of a pro-cyclical bias, particularly when we control for the effect of the U.S. dollar. However, as commodity markets have become financialized in recent years, so their sensitivities appear to have risen somewhat to macroeconomic news, with some key data releases such as U.S. payrolls having a large and broad effect across many different commodities. However, our findings indicate that models based on macroeconomic news still do a poor job of forecasting commodity prices at daily frequencies compared to more naïve models. We find that there are some asymmetries in how commodity prices respond to news, most notably for gold, which alone among commodities acts as a safe-haven when “bad” economic news emerges. Finally, recent price volatility behavior appears not to influence the sensitivity of commodity prices to news, in sharp contrast to findings elsewhere in the literature on financial assets and exchange rates. We suspect that the effect of economic developments on commodity prices, particularly at high frequencies, will become a fertile area for research in the future, particularly as new emerging economies are becoming large consumers of primary commodities; for now, data limitations prevent a comprehensive study of their effects on these markets. Acknowledgements Excellent research assistance from Richard LaRock, formerly at the IMF, is greatly appreciated. We would like to thank two anonymous referees, Andy Berg, Kevin Cheng, Thomas Helbling, Ydahlia Metzgen, Roberto Perrelli, David Romer, and semi-

nar participants from the Finance and Research departments for helpful comments and suggestions. The usual disclaimer applies. Appendix A. See Tables A1–A3. Table A1 Commodity futures contracts specification. Contract

Exchangea

Specification summaryb

Closing price time

Gold Silver Platinum Palladium Oil

COM EX COM EX NYMEX NYMEX NYMEX

17:15 EST 17:15 EST 17:15 EST 17:15 EST 17:15 EST

Heating oil Natural gas

NYMEX NYMEX

Wheat Corn Soybeans Copper

CBOT CBOT CBOT COM EX

Aluminium

COM EX

100 troy ounces 5,000 troy ounces 50 troy ounces 100 troy ounces Light, sweet crude, 1,000 barrels 42,000 barrels 10,000 million British thermal units 5,000 bushels 5,000 bushels 5,000 bushels High grade, 25,000 pounds 44,000 pounds

17:15 EST 17:15 EST

13:15 CST 13:15 CST 13:15 CST 17:15 EST 17:15 EST

Source: New York Mercantile Exchange, Chicago Board of Trade, and the London Metals Exchange. a COMEX is a division of NYMEX, the New York Mercantile Exchange. CBOT is an abbreviation of the Chicago Board of Trade and LME is an abbreviation of the London Metals Exchange. b Refer to individual exchanges for full specifications.

Table A2 U.S. macroeconomic announcements: summary statistics (January 1997 to May 2008)a (monthly percent change, unless otherwise specified). Macroeconomic announcement

Advance retail sales Change in non-farm payrolls (thousands) Consumer confidence (index) Consumer price index Employment cost index (quarterly percent change) Existing home sales FOMC interest rate decision (basis point change) Advance GDP (annualized quarterly percent change) Housing starts (thousands) Industrial production ISM manufacturing survey PPI ex-food and energy (MoM) ECB interest rate decision (basis point change) German IFO survey U.K. interest rate decision (basis point change)

Average actual

Surprise Average

Standard deviation

0.3 55.9

0.0 −17.1

0.6 95.2

101.2

0.1

5.0

0.2 0.8

0.0 0.0

0.1 0.2

0.4 4

1.0 0

3.9 0.1

2.7

0.1

0.8

1460 0.1 51.9 0.1

10 0.0 0.1 0.0

86 0.4 2.0 0.3

3 95.8 5

0

0.1

0.0 0

1.2 0.1

Source: Bloomberg; Authors’ estimates. a Actuals denote the data as of the release date and do not reflect subsequent revisions.

S.K. Roache, M. Rossi / The Quarterly Review of Economics and Finance 50 (2010) 377–385

385

Table A3 Commodity price sensitivity to macroeconomic announcements (without U.S. dollar control). Gold

Silver

Platinum

Palladium

Crude oil

Heating oil

Natural gas

Wheat

Corn

Soybeans

Copper

Macroeconomic announcements Advance retail sales Change in non-farm payrolls Consumer confidence Consumer price index Employment cost index Existing home sales FOMC interest rate decision Advance GDP Housing starts Industrial production ISM manufacturing survey ECB interest rate decision German IFO survey

−0.02 −0.21** −0.13* 0.11 0.07 −0.01 −0.05 −0.10 −0.07 −0.28* −0.07 0.15** 0.12

0.05 −0.06 0.04 0.04 0.16 −0.05 −0.07 −0.02 −0.03 −0.24 −0.08 0.07 0.11

0.02 0.06 −0.04 0.13 0.27 −0.08 −0.41** 0.07 0.06 −0.08 −0.11 −0.48 0.26**

0.05 −0.04 −0.35* 0.20 0.50 0.09 −0.06 0.38 0.10 0.35 0.05 0.26 0.00

−0.34 0.13 0.14 0.02 0.24 −0.01 −0.29 −0.23 0.15 0.29 −0.21 −0.07 0.05

−0.34 0.18 0.15 −0.07 0.16 −0.22 −0.21 −0.12 0.13 0.21 −0.16 −0.32 0.08

−0.17 0.09 0.74*** −0.20 0.19 −0.49* −0.39 0.11 −0.39* −0.02 −0.29 −0.30 0.94***

0.02 0.21* −0.09 0.10 −0.04 0.05 – −0.08 −0.17 −0.01 0.14 0.06 0.12

0.13 0.19** −0.06 0.07 0.19 −0.17 – −0.32 −0.15 −0.05 −0.01 0.22* 0.17

−0.01 0.07 0.00 0.15 0.45** 0.01 – −0.34** −0.19 −0.05 −0.09 −0.08*** 0.24

0.09 0.15** 0.09 0.05 −0.03 0.02 −0.28*** −0.25* −0.21 −0.11 0.19* −0.04 0.12 0.01 0.03 0.12 0.12 0.22** −0.09 0.27 0.18 0.07 0.02 0.00 0.21** 0.20*

Lagged by 1 day Change in non-farm payrolls Employment cost index Existing home sales FOMC interest rate decision Advance GDP Housing starts ISM manufacturing survey ECB interest rate decision German IFO survey Lagged price t − 1 Lagged price t − 2

−0.02 0.04 0.04 −0.26*** 0.01 −0.08 0.06 −0.07 0.23*** 0.00 −0.01

−0.09 −0.14 0.10 −0.26*** 0.03 0.01 0.21 −0.02 0.24*** −0.02 0.02

−0.16 −0.14 −0.53*** −0.66** −0.10 0.16 0.02 0.01 0.42*** 0.02 −0.06***

−0.22 0.70* −0.30** −0.17 0.36 0.03 0.14 0.09 0.28** 0.11*** −0.04*

0.11 0.07 0.30* 0.16 0.16 −0.16 0.09 0.10 −0.08 −0.01 −0.04***

0.23 −0.11 0.39** 0.09 0.24 0.01 0.24 0.21 −0.21 −0.02 −0.02

0.02 0.72* −0.24 0.70*** 0.89* −0.06 0.66*** 0.20 0.21 −0.04** 0.01

0.08 −0.15 0.10 −0.12 −0.01 0.10 −0.02 0.23* 0.12 0.01 −0.01

0.13 −0.06 0.18 −0.20* 0.14* 0.21 −0.04 0.01 0.14 0.04** 0.00

0.22* −0.19* 0.34 −0.15 0.18*** −0.11 −0.05 0.22** 0.09 −0.01 0.03

−0.04 0.22 −0.14 −0.09 0.08 −0.18 0.26** −0.01 0.19** −0.07 −0.01

0.04*** 0.95***

0.04 0.96***

0.25*** 0.67***

0.12*** 0.86***

0.06*** 0.92***

GARCH equation coefficients Moving average Autoregressive

0.06 0.92***

0.08*** 0.91***

0.02*** 0.98***

0.03*** 0.96***

0.06*** 0.93***

0.03*** 0.96***

Aluminium

0.22*** −0.05 −0.09 −0.14 0.13 −0.16 0.11 −0.02 0.13 −0.08 −0.03

0.05*** 0.95***

Coefficients shown are those that were statistically significant at the 10 percent level or more for at least one commodity. The coefficient on each announcement is multiplied by 100 and represents the percent change in the price of the nearest futures contract index for a one-standard deviation surprise. Bollerslev–Woolridge standard errors, which are robust to remaining heteroscedasticity. * Significance at the 90 percent level. ** Significance at the 95 percent level. *** Significance at the 99 percent level.

References Andersen, T. G., Bollerslev, T., Diebold, F. X., & Vega, C. (2003). Micro Effects of Macro Announcements: Real-time price discovery in foreign exchange. American Economic Review, 93, 38–62. Antoniou, A., & Foster, A. J. (1992). The effects of futures trading on spot price volatility: Evidence for Brent crude oil using GARCH. Journal of Business Finance & Accounting, 19(4), 473–484. Balduzzi, P., Edwin, J., Elton, T., & Green, C. (2001). Economic news and bond prices: Evidence from the U.S. Treasury Market. The Journal of Financial and Quantitative Analysis, 36(4), 523–543. Barnhart, S. W. (1989). The effects of macroeconomic announcements on commodity prices. American Journal of Agricultural Economics, 71(2), 389–403. Bollerslev, T., & Wooldridge, J. M. (1992). Quasi-maximum likelihood estimation and inference in dynamic models with time varying covariances. Econometric Reviews, 11, 143–172. Borensztein, E., & Reinhart, C. M. (1994). The macroeconomic determinants of commodity prices. IMF Staff Papers, 41(2), 236–261. Cai, J., Cheung, Y.-L., & Wong, M. C. S. (2001). What moves the gold market? Journal of Futures Markets, 21(3), 257–278. Chen, Y.-C., Rogoff, K., & Rossi, B. (2008). Can Exchange Rates Forecast Commodity Prices? Economic Research Initiatives at Duke (ERID) Working Paper No. 1. Christie-David, R., Chaudhry, M., & Koch, T. W. (2000). Do macroeconomics news releases affect gold and silver prices? Journal of Economics and Business, 52, 405–421. Clements, K. W., & Fry, R. (2008). Commodity currencies and currency commodities. Resources Policy, 33(2), 55–73. Della Corte, P., Sarno, L., & Tsiakis, I. (2009). An economic evaluation of empirical exchange rate models. Review of Financial Studies, 22, 3491–3530. Ehrmann, M., & Fratzscher, M. (2005). Exchange rates and fundamentals: New evidence from real-time data. Journal of International Money and Finance, 24, 317–341. Fleming, M. J., & Remolona, E. M. (1999). Price formation and liquidity in the U.S. Treasury Market: The response to public information. Journal of Finance, 54, 1901–1915.

Frankel, J. A. (2008). The effect of monetary policy on real commodity prices. In J. Campbell (Ed.), Asset prices and monetary policy (pp. 291–327). University of Chicago Press. Frankel, J. A., & Hardouvelis, G. A. (1985). Commodity prices, money surprises and fed credibility. Journal of Money, Credit and Banking, 17(4, Part 1), 425–438. Galati, G., & Ho, C. (2003). Macroeconomic news and the euro/dollar exchange rate. Economic Notes, 32(3), 371–398. Ghura, D. (1990). How commodity prices respond to macroeconomic news. World Bank PRE Working Paper No. 354. Hansen, P., & Lund, A. (2005). A forecast comparison of volatility models: Does anything beat a GARCH(1,1)? Journal of Applied Econometrics, 20, 873–889. Hess, D., Huang, H., & Niessen, A. (2008). How do commodity futures respond to macroeconomic news? Journal of Financial Markets and Portfolio Management, 22(2). IMF. (2008). Dollar depreciation and commodity prices. Washington, DC: World Economic Outlook. (Chapter 1). Kilian, L., & Vega, C. (2008). Do energy prices respond to U.S. macroeconomic news? A test of the hypothesis of 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, 307–328. Rime, D., Sarno, L., & Sojli, E. (2010). Exchange rate forecasting, order flow and macroeconomic information. Journal of International Economics, 80, 72–88. Roache, S. (2008). Commodities and the market price of risk. IMF Working Paper No. 08/221. Rossi, M. (1998). Economic announcements and the timing of public debt auctions. IMF Working Paper No. 98/44. Tang, K., & Xiong, W. (2009). Index investing and the financialization of commodities. Unpublished manuscript. UNCTAD. (2009). The financialization of commodity markets. Trade and Development Report, New York (Chapter 2). Yang, J. R., Balyeat, B., & Leatham, D. J. (2005). Futures trading activity and commodity cash price volatility. Journal of Business Finance & Accounting, 32(1), 297–323.