Commodity Price Responses to Monetary Policy Surprises Dean Scrimgeour∗ Colgate University 15 April, 2014

Abstract Exploiting information in high-frequency financial market data I find that a 10 basis point surprise increase in interest rates causes commodity prices to fall immediately by about 0.6%. This is similar to the estimated responses of both the S&P500 and a U.S. trade weighted exchange rate index, and about five times larger than the response in a standard VAR even twelve months after the shock. Metals prices tend to respond more than agricultural commodities. The point estimate for oil prices is similar to other commodities, but is estimated less precisely.

∗

Contact: Dean Scrimgeour, Economics Department, Colgate University, 13 Oak Drive, Hamilton, NY 13346. Email: [email protected]. This work has benefited from comments from Emily Conover, Ed Gamber, Chad Jones, Yuriy Gorodnichenko, Lutz Kilian, Roisin O’Sullivan, Ann Owen, Brian Roe, Christina Romer, David Romer, Nicole Simpson, Julie Smith, Chad Sparber, and Marc Tomljanovich, and to seminar participants at various locations. Thanks also to RC Research for providing futures data.

1.

Introduction

I explore the influence of monetary policy on commodity prices. Since commodity prices help determine a wide range of consumer and producer prices, the response of commodity prices to monetary policy is an important aspect of the monetary transmission mechanism. The relationship between commodity prices and monetary policy has been given a lot of attention over the years. Some blame the inflation of the 1970s on rising commodity prices (see for example, Blinder, 1982). By contrast Barsky and Kilian (2001) argue that commodity prices tended to rise in the 1970s in response to anticipated inflation brought on by loose monetary policy. During the 2005-2008 period, elevated commodity prices brought renewed attention to commodity markets. Would high oil prices lead to inflation? Would they cause a recession? Would they generate both, returning the economy to stagflation as in the 1970s? Explanations for why commodity prices have been high include growing demand in China and speculative behavior in financial markets (Hamilton, 2009). In addition, revisiting Barsky and Kilian (2001), Taylor (2009) has argued that loose monetary policy may have been behind the surge in commodity prices. In addition, many commodity-producing countries link the value of their currency to the U.S. dollar. Monetary policy changes in the United States could affect economic outcomes in these countries through the effect on commodity prices as well as the conventional interest rate channels. This article attempts to measure the size of monetary policy surprises’ effects on commodity prices. Like other financial market prices, commodity prices are relatively flexible, adjusting quickly in response to shocks. Any effects of monetary policy announcements on commodity prices likely occur within a short period of the announcement, by contrast with retail prices, which are stickier. It would be interesting to know how the long-run response of commodity prices differs from the short-run response. Long-run responses are harder to estimate in general. Efficient Market Hypothesis reasoning suggests that commodity prices should respond relatively quickly to news about interest rates, so that the short-run and long-run response of commodity prices are similar. More sophisticated theories of commodity markets loosen the link between short-run and long-run effects, but the sign of the difference is not clear. Overshooting theories suggest short-run responses could be larger, while models with storage constraints might predict smaller short-run 2

responses. I study the relationship between commodity prices and interest rates around particular newsrelated events. The event days are when the Federal Reserve’s Open Markets Committee meets and financial markets acquire new information about the course of monetary policy. The article builds on the literature starting with Cook and Hahn (1989) who focused on meeting dates to measure the response of bond markets to changes in the Federal Reserve’s target federal funds rate. However, Cook and Hahn’s study ignores the fact that financial market participants learn about more than just monetary policy on these event days. Rigobon and Sack (2004), whose method I implement, account for the fact that other forces may move both interest rates and asset prices on event days, confounding the event study estimator. They show how to estimate the effect of monetary policy on asset prices using instrumental variables techniques that rely on non-event days to provide information about the typical relationship between interest rates and commodity prices. Deviation from this normal relationship on event days is interpreted as being due to monetary policy surprises. The results in this article indicate a rapid, statistically significant, and economically significant response of commodity prices to monetary policy. Point estimates suggest that a 10 basis point increase in interest rates would cause commodity prices to fall by approximately 0.6% immediately. This finding poses a challenge for vector autoregression (VAR) studies that impose the restriction that commodity prices can only respond to monetary policy surprises from previous months or quarters (Christiano et al., 1999). In a standard recursively-identified VAR, a 10 basis point surprise increase in interest rates would lower commodity prices by approximately 0.1% after a year, with smaller responses closer to the timing of the surprise.1 The response of oil prices is estimated less precisely, and in many specifications it is not statistically significantly different from zero. This is consistent with Kilian and Vega (2011) who find that energy prices are predetermined with respect to macroeconomic news. It is also consistent with intertemporal arbitrage being more difficult for oil, which is bulky compared with metals. 1 Kim (1999) uses a non-recursive identification scheme in a VAR that allows for a contemporaneous response of commodity prices to monetary supply shocks. His estimates imply smaller and more gradual responses of commodity prices to monetary policy than I estimate.

3

2.

Related Work

Since commodities and bonds are substitutes as assets that can store value, when the Federal Reserve sells bonds to raise interest rates, demand for commodities falls. Therefore commodity prices in the spot market should fall when interest rates rise due to monetary intervention. Other shocks, such as news about bond risk, might move bond prices and commodity prices in opposite directions. Taylor (2009) presents a monetary explanation for the rise in commodity prices during the early stages of the recent financial crisis. He argues that oil prices increased in 2007 and 2008 because the Federal Open Market Committee (FOMC) reduced in interest rates.2 Similarly Frankel (2008) has argued that reductions in interest rates could increase commodity prices. In his work he emphasizes an overshooting mechanism in the response of commodity prices to monetary policy.3 The overshooting story aims to explain the magnitude of the rise in commodity prices in response to a monetary policy surprise.4 Several other studies use data at monthly or lower frequency to examine the interaction of commodity prices with other macroeconomic variables. For example, Hua (1998) estimates a long-run (cointegrating) relationship between non-oil commodity prices and macroeconomic indicators. Kwon and Koo (2009) estimate a vector autoregression that includes primary commodity prices.5 In each case, the estimated response of commodity prices is potentially confounded by the fact that commodity prices are forward looking. Moreover, the interpretation of shocks in the VAR setting is not always clear (Rudebusch, 1998). This article estimates effects of the monetary policy surprise by looking at movements in com2 Glick and Leduc (2011) explore whether the Federal Reserve’s quantitative easing policies drove up commodity prices, but find they did not. In fact, Glick and Leduc (2012) argue that large-scale asset purchases by the Federal Reserve, though intended as stimulus, tended to cause commodity prices to decline, possibly due to the policy action signalling Federal Reserve information about the state of the economy. 3 See also Frankel (1984); Frankel and Hardouvelis (1985); Frankel (1986). 4 Caballero et al. (2008a,b) also link interest rates and commodity prices, noting that commodity prices have been high at the same time real interest rates have been low in the 2000s. They emphasize a global savings glut (Bernanke, 2005) to explain levels of interest rates and commodity prices. During the early stages of the financial crisis, when debt began looking riskier, there was a sell-off in bonds and investors substituted some commodities for bonds in their portfolios. 5 Kwon and Koo’s VAR is identified using techniques from machine learning. See Heckman and Pinto (2013) for a critique of these approaches to identification in economics.

4

modity prices that occur within a day of the surprise. It is possible that this short-run response of commodity prices differs from the long-run response. One possible reason for this would be a gradual incorporation of the monetary news into commodity prices. A problem with this hypothesis is that it relies on predictable movements in asset prices, which present opportunities for arbitrage. Dornbusch (1976) provides another reason: a large movement of the asset price happens exactly at the moment of the shock, but is followed by a gradual reversion, so the short-run response is larger than the long-run response. Time series properties of the commodity price series I study suggest any short-run impact of monetary policy surprises is likely to be persistent, so that the short-run response is informative about longer-horizon responses. Specifically, the commodity prices are relatively well-characterized as unit root processes, without additional autoregressive or moving average components that would indicate significant differences between short-run and long-run responses to a shock, or the hump-shaped responses that some macroeconomic variables exhibit. One interpretation of monetary policy surprises is that they occur due to changes in the central bank’s preferences.6 An unexpected increase in the central bank’s inflation target causes a surprise fall in interest rates. The (sticky) consumer price level p does not move when the shock occurs but the (flexible) commodity price does. The lower interest rate causes the commodity price to overshoot, rising more in the short run than the long run. With an unexpected, permanent change in the central bank’s nominal target, the effect on the commodity price level is largest when the shock occurs so that the model has implications for commodity futures prices. Spot prices should move more than futures prices. This is the analog of Dornbusch’s overshooting result, and I test for it using an index of commodity price futures at various horizons. Monetary policy surprises may affect commodity prices through other channels too. For example, monetary contractions induce falls in aggregate demand that depress commodity prices. Alternatively, a portfolio rebalancing might occur where a higher safe interest rate causes investors to substitute away from risky assets such as commodities causing a softening of commodity prices. Another common story linking commodities and monetary policy is the idea that commodities are 6

¨ See Gurkaynak et al. (2005a) for more on this perspective. An alternative is that the monetary policy surprise reflects information differences, for which Romer and Romer (2000) find evidence in an earlier period.

5

an inflation hedge (Bodie, 1983). For this theory to rationalize the negative effect of interest rates on commodity prices, surprise falls in interest rates would have to increase the uncertainty about future inflation, while surprise increases in interest rates decrease uncertainty. In measuring the effect of monetary policy surprises on commodity prices I follow Rigobon and Sack (2004) who modify the event study procedure to account for other factors that influence both interest rates and asset prices on event days. They apply these methods to estimate the effect of monetary policy surprises on stock and bond yields. Other studies using these methods have estimated the effects of monetary policy on stock prices, interest rates, and exchange rates.7 To my knowledge, mine is the only study to apply these methods to analyze the effect of monetary policy on commodity prices using high-frequency data in this way.8 Others to use daily data to examine the influence of monetary policy on commodity prices have used standard event study methods (for example, Glick and Leduc, 2011), or use survey measures of expectations rather than market-based measures of expectations (for example, Barnhart (1989); Roache and Rossi (2010)), or focus only on energy (Kilian and Vega (2011)). In another recent paper, Anzuini et al. (2012) estimate the effect of monetary policy shocks on commodity prices using monthly data in a VAR. Following Rigobon and Sack, I measure monetary policy surprises using three-month Eurodollar futures rate. Since these are not spot rates, they are less affected by surprises due to the Fed adjusting its target rate a month earlier or later than expected (a timing surprise).9 I also use the Kuttner (2001) method as a robustness check. Kuttner’s work shows how to use federal funds futures contracts to measure monetary policy surprises.10 Conditional on accurate 7

Craine and Martin (2008) apply the same approach to estimate the effect of U.S. policy surprises on Australian stocks and bonds and the USD/AUD exchange rate (as well as effects of Australian monetary policy surprises). 8 Frankel and Hardouvelis (1985) look at the commodity price response to weekly money supply announcements in the early 1980s, but give a different interpretation to the results. If commodity prices respond to the announcements, it is interpreted as evidence that the Federal Reserve’s previously announced monetary target is disregarded by market participants. If there is no response, this is interpreted to mean that markets regard the monetary surprise as an error in hitting the target rather than a change in the target. The methods in this article, because they rely on forward-looking measures of policy, are not confounded by a targeting error. 9 Timing surprises relate to the level and slope of the yield curve. Significant monetary policy surprises affect the level of the yield curve (longer-term interest rates), whereas timing surprises affect the slope. For example, a change in the federal funds target rate that affects the anticipated rate in two weeks from the present but does not affect the anticipated rate three months into the future contains a timing surprise. The Kuttner method, since it uses primarily current month futures is more susceptible to timing surprises. 10 ¨ Gurkaynak et al. (2005) and Bernanke and Kuttner (2005), among others, follow this approach.

6

measurement of the surprise, assessing its effects on financial markets is relatively straightforward. The federal funds futures shock may capture important timing surprises related to when the Federal Reserve adjusts policy, so the federal funds futures shock mingles timing and level information. As in Kuttner, I use the spot month federal funds futures contract to measure the shock, switching over to using the next month federal funds futures contract if an event day is within three days of the end of the month.

3.

Empirical Setup

Rigobon and Sack (2004) modify the standard event study method to account for the endogeneity of the change in the interest rate.11 Consider estimating βj in

∆pjt = βj ∆it + jt

(1)

∆it = mpt + νt

(2)

where ∆pj is the change in the (log) price of commodity j and ∆i is the change in an interest rate. In particular, mp is the monetary surprise, which is zero on non-event days but may be non-zero on events days. Endogeneity problems arise because interest rates move for many reasons, even on event days, and these reasons may be related to other factors that affect commodity prices (jt ).12 The shock ν may be correlated with  causing ∆i to be endogenous in equation 1, even on event days. On event days the variance-covariance matrix of [∆pjt , ∆it ]0 changes in a way that allows us to identify βj . The event study estimator applies ordinary least squares to equation 1 using data from event days only. Rigobon and Sack show how to identify β using an instrumental variables approach.

11

See also Rigobon (2003). As an application, the “policy surprise” measured from Federal Funds Futures contracts shows apparent surprises on non-event days as well as event days. It is likely that some of the movements in the Federal Funds Futures on non-event days (and possibly also on event days) represent the anticipated response of the Federal Reserve to news, rather than a monetary policy surprise. 12

7

The instrumental variable is defined as

zt =

     

∆it

     −∆it

on event days on pre-event days

and the equation is estimated using all the event days and the days immediately prior to an event day (so the event days are half of the sample). Asymptotic analysis shows that βˆj,IV

→ = =

cov(zt , ∆pjt ) cov(zt , ∆it ) 1 1 2 2 2 2 (βj (σm + σν ) + σ,ν ) − 2 (βj σν + σ,ν ) 1 1 2 2 2 2 (σm + σν ) − 2 σν βj

where σx2 is the variance of x, σx,y is the covariance of x with y, and we assume that monetary policy shocks are uncorrelated with either  or ν. The IV estimator remains consistent even if we move away from a triangular system to one in which interest rates explicitly respond to commodity prices. Intuitively, the comovement between interest rates and commodity prices on event days is due to the effect of interest rates on commodity prices and the background correlation between interest rates and commodity prices. The instrumental variables estimator uses the non-event days the measure the background correlation between p and i, then subtracts that from the correlation on event days. Defining the instrument as −∆it on non-event days rather than ∆it ensures that the background comovement in commodity prices and interest rates is subtracted from the comovement on event days. Apart from the expected assumptions about the noncorrelation of monetary policy shocks with other shocks affecting the financial markets, we also assume that the variation in these other shocks is the same on event days and non-event days. If this is not the case, perhaps because everyone gives the Federal Reserve space to make its announcement, then the estimator above will not be consistent. In addition to estimating the response of commodity prices to interest rates separately for each

8

commodity, I estimate values for β that are restricted to be the same across groups of commodities, pooling data on different commodities and estimating a single regression, either for all the commodities I study or for some subset. I also estimate coefficients using a system estimator (3SLS) that accounts for covariances of jt across commodities. Table 1 shows some summary statistics for changes in spot commodity prices and the threemonth Eurodollar futures rate on event days (when the FOMC meets) and nonevent days.13 Figure 1 presents this information graphically. The variances of commodity price movements tend to be higher on event days, and the covariances with the interest rate tend to be more negative. In my base sample, I use all regularly scheduled Federal Open Market Committee meeting dates from 1994 to March 2008 as event days, so my sample ends in the early stages of the recent financial crisis, before interest rates got near zero and the Federal Reserve began using unconventional monetary policy operations. The sample starts in 1994 since this was when the Federal Reserve began announcing its decisions publicly. This gives 114 event days. As a robustness check, I also look at FOMC chairs’ testimonies presented under the Humphrey-Hawkins regulation as additional event days, include unscheduled announcements of target interest rate changes as event days, and extend the sample of events to include FOMC meetings back to 1989. I use daily data on spot prices for seventeen commodities over this period. There are nine metals (gold, silver, copper, aluminum, tin, zinc, platinum, lead, and nickel)14 , seven agricultural commodities (cocoa, coffee, cotton, wheat, hogs, live cattle, and livestock), and oil. The market for these commodities usually closes before the Federal Reserve releases a post-meeting statement, typically around 2:15pm Eastern Standard Time (Kuttner, 2001) during the period I study. Therefore, the daily change in the commodity price is taken to be the difference in closing prices between the day after a Federal Reserve announcement and the closing price on the day of the Federal Reserve’s announcement.15 For each commodity I fail to reject the null hypothesis of a unit root in the log of the price level with a Phillips-Perron test. This suggests the effects I estimate 13

Variables in my data are recorded before the FOMC announcements come out, so the event day movements are for the difference between prices the day after the announcement and the day of the announcement. 14 Data for lead prices start in 1995. 15 If the measurement of price changes is amended so the price change is the closing price on the day of the announcement less the price the day before, the empirical strategy I use detects no effect of monetary policy on commodity prices.

9

using daily data are likely to be very persistent, and will eventually affect output and consumer prices.

4.

Results

The following section presents the empirical results. It first shows the response of commodity prices to monetary surprises, then presents a range of robustness checks. It then studies commodity futures and exchange rates before exploring the role of carrying cost in explaining the responses of different commodities. Finally, there is a discussion of how the results in this article could be used in identifying a VAR.

4.1.

Response of Commodity Prices to Monetary Policy Surprises

Tables 2 and 3 show estimates of the effect of monetary policy surprises on commodity prices. Table 2 shows estimates when commodity prices are grouped together. The commodity prices are not aggregated, but a single effect is estimated using data on prices of a group of commodities. The groups are All Commodities, All Metals, All Agricultural commodities (of which there are Vegetable commodities and Animal commodities), and Oil. Since there is correlation across commodities in daily price changes, the standard errors are clustered at the day level. Table 3 shows the estimated effects for each commodity. The tables show both the event study estimates (from the OLS regression of 100 times the change in the log commodity price on the change in the interest rate, in percentage points, on event days) and the Rigobon-Sack estimate.16 Table 3 also includes Three-Stage Least Squares estimates of the full system of equations, with one equation for each commodity price. The results show that monetary policy surprises have substantial effects on commodity prices. Consider a particular meeting of the Federal Open Markets Committee at which market participants expect the Federal Reserve to either keep interest rates at some level (with 40% probability) or raise them 25 basis points (with probability 60%), with these subjective probabilities based on the state of the economy and perceived stance of the Federal Reserve. On average, interest rates 16

The Rigobon-Sack estimate corresponds to α ˆ ihet in Rigobon and Sack’s paper.

10

are expected to rise by 15 basis points. If the FOMC actually raises interest rates 25 basis points, there is a 10 basis point surprise. According to my estimates, a 10 basis point surprise increase in interest rates reduces commodity prices overall by about 0.6%, with a standard error of 0.2%. Rigobon and Sack estimate that the S&P500 responds to a shock of this size by about 0.7%. (The corresponding coefficient is around -6.81 rather than -6.37.) The response of metals is somewhat larger, while the response of agricultural commodities is smaller. For a 10 basis point surprise increase in interest rates, metals prices tend to fall by 0.75%, while agricultural commodities fall by 0.49%. The price of oil falls by 0.67% in response to such a monetary surprise. The estimated effect on the price of oil is economically significant, but the standard error is relatively large. These results are somewhat consistent with Kilian and Vega (2011) who find that macroeconomic news tends not to affect energy prices immediately, since the estimated response of oil prices is statistically insignificant. The results are not consistent with the findings of Roache and Rossi (2010), who show that gold is “unique” in its response to news, though their study examines a broader range of sources of news. While my article studies only one source of news, the response of gold is similar to many other commodities, especially the large range of metals I include. The variation across commodities in responsiveness to monetary policy surprises, and a reason why gold and other metals seem to respond more to news, could be due to physical characteristics of the commodities, as I discuss below in section 4.6. Compared with Anzuini et al. (2012), I estimate slightly larger responses for commodities overall, and much larger responses of metals prices (around 7% compared with 1% in response to a 100 basis point surprise), though the size of the monetary policy shock in the VAR is not directly comparable to the surprises I use. When we consider the effect of monetary policy surprises on individual commodities some prices appear to respond much more than others. Tin, lead, cotton, wheat and coffee each appear to respond less to monetary policy surprises. By contrast prices of gold, silver, platinum, and nickel respond more strongly to monetary policy surprises. Some of the standard errors are large, but the data reveal a general pattern of commodity prices falling in response to monetary tightening. The results estimated on the full sample are sensitive to the inclusion of the 18 March, 2008,

11

FOMC announcement. Tables 2 and 3 include a column of estimates based on the data excluding March 2008. The estimates are substantially smaller when March 2008 is excluded. For three commodities (coffee, cotton, and wheat) the point estimate becomes positive. Overall, the estimates still suggest a substantial response of commodity prices in general to monetary policy surprises. The estimated coefficient for all commodities is -4.52 instead of -6.37. On 18 March, 2008, days after the collapse of Bear Stearns, the FOMC reduced the funds rate 75 basis points. In its statement, the FOMC noted that some members had wanted a smaller change. The three-month Eurodollar futures rate increased around 20 basis points, and commodity prices fell around 4% on average. This is four times greater than the response implied by estimates using only the data in the rest of my dataset. The discrepancy could be due to market participants updating their views on how bad the economic outlook was at that time, or perhaps because of the apparent lack of concern with the growing financial and economic turmoil among some FOMC members. Regardless of the reasons, this observation is an outlier, and I have presented results both with and without it. The main purpose of the empirical exercise so far is to produce an estimate of the effect of a monetary policy surprise of a given size. The results in this article can also form the basis of a kind of variance decomposition for commodity prices. Consider the volatility of gold price movements. If we assume that monetary policy surprises only occur on event days, and these are about one-thirtieth of all trading days, then the variance of the change in gold prices is about five percent due to monetary policy surprises. The variance of gold price changes on event days alone is around 2/3 due to monetary policy surprises. Since event days are relatively rare, about 5% of the overall variance of gold price movements is accounted for by monetary policy surprises. This is comparable to findings in other studies regarding the effects of U.S. monetary policy surprises on stock prices, foreign interest rates, and exchange rates.17 Note that these calculations have no 17 Using the numbers in table 1, the variance of the change in gold prices is higher by about 0.86 on event days, relative to a baseline of 0.49. Since event days are about 3% of all days, monetary policy surprises account for 0.03 ∗ 0.86/0.49, or about 5% of the variability of gold price movements. For comparison, Craine and Martin (2008) find similar shares of the variability of Australian short-term interest rates due to U.S. monetary policy surprises on U.S. monetary policy event days. Stock returns have a smaller share of their variability accounted for by monetary policy surprises, and the USD/AUD exchange rate has an even smaller share of volatility due to either U.S. or Australian monetary policy surprises. Other commodities generally have a smaller fraction of volatility accounted for by monetary policy surprises.

12

implications for the role of systematic monetary policy in affecting the variability of commodity prices.

4.2.

Robustness

This section reports on several robustness checks. The previous section noted the importance of the 18 March, 2008, event. Aside from this event, the findings reported so far are generally robust to variations in the sample period, the definition of events, and the method used to estimate the response of commodity prices. First, consider extending the sample period. Table 4 shows how the estimated responses of commodity prices to monetary policy are affected by using date from 1989 to 1993 in addition to the 1994 to 2008 period. The magnitudes of the estimates are similar but are less precise even though more data is used. When even earlier data are added, the estimated effect becomes smaller. The reason for the larger standard errors is that the first-stage regression is much weaker when the earlier period is included. The strength of the first stage is determined by the difference between the variance of the Eurodollar futures rate on event days and pre-event days. The Federal Reserve’s policy was much less transparent before 1994, so the news generated on the event day was less clear and did not move markets as much. This is especially true prior to 1989. From 1994, the Federal Reserve released more information immediately upon making a decision. Enhanced transparency does not mean there are no surprises. There have continued to be surprises in this era, and the Federal Reserve’s enhanced transparency means these surprises and their effects are more easily measured. Adding the Federal Reserve chairman’s Humphrey-Hawkins testimony dates as event days to the original sample (as in Rigobon and Sack) reduces the estimated response of commodity prices somewhat. For example, the estimated response for all commodity prices is -5.25 instead of -6.37. The estimated responses of the other categories reported are also attenuated relative to the original results. (And in contrast to adding the earlier years to the sample, adding the chairman’s testimonies lowers the standard errors.) Demiralp and Jorda (2004) argue that the Federal Reserve’s actions have a larger effect when 13

it reverses the previous direction of policy. That is, when it begins raising interest rates or begins lowering them, markets respond more. A simple test of whether this is true of commodity prices by excluding dates when the Federal Reserve reverses the path of the federal funds target from my regressions.18 I do not find a systematically smaller response of commodity prices to interest rates. Metals prices do seem to respond less, but agricultural commodities seem to respond more when the policy reversals are excluded. I implement the Kuttner (2001) method using federal funds futures contracts to estimate the effect of monetary policy surprises. These estimates suggest a smaller, more precisely estimated, effect of monetary policy on commodity prices. The magnitude of the responses is broadly similar, with the overall response of commodity prices to a 10 basis point monetary policy shock being 0.43% according to the Kuttner estimates, instead of -0.64% according to the Rigobon and Sack estimates, when the base sample is used.

4.3.

Current and Expected Changes in the Interest Rate

The Federal Reserve, when announcing a decision, may change the target interest rate (“target”), ¨ and also may hint at changes to the likely future trajectory of interest rates (“path”). Gurkaynak et al. (2005b) argue that signalling about future interest rate changes is the more important channel through which Federal Reserve actions affect bond prices. The three-month Eurodollar contract used in this article contains elements of both these channels, and so the estimated effects of monetary policy on commodity prices include a response to current changes in the short-term interest rates and a response to expected future changes. Nonetheless it may be interesting to examine the two effects separately. ¨ I rescale the target and path factors provided by Gurkaynak et al. (2005b) so that a one unit change in each corresponds to a one unit change in the three-month Eurodollar futures rate. Table

18 There are eight such dates in my main sample. An alternative approach that yields the same conclusion is to include an indicator for reversals interacted with the interest rate change. Since there are so few actual reversals, the estimated coefficient on this interaction has a large standard error.

14

4 reports the estimated response of commodity prices to each factor, based on the regression ∆pjt = βgt targett + βgp patht + jt

(3)

where g indicates the group of commodities to which commodity j belongs. Estimation is by ordinary least squares applied to only the event days. Several findings stand out. First, the magnitude of responses is broadly similar to the responses estimated by other means. Second, the estimated responses to the path factor are generally larger than to the target factor for metals, but the target factor has a larger effect than the path factor on agricultural commodities. In general these differences in point estimates are not statistically significant. Third, the estimated responses for oil are large, but also have large standard errors. (Different estimators, such as seemingly unrelated regression, suggest much bigger standard errors for oil.)

4.4.

The Response of Commodity Futures

In addition to the estimated response of commodity prices in the spot market, I estimate the response of commodity futures. In principle, since commodity futures reveal the expected future level of the commodity index, we can examine whether shorter horizon futures respond more, as implied by the theory of overshooting. In practice, markets in longer-dated futures contracts are typically thin, so detecting any response may be difficult. Table 5 presents the estimated responses of the Goldman Sachs Commodity Index (GSCI) futures contracts for one month through five months. The GSCI is heavily weighted toward oil and energy products, so the estimated responses are similar to the response of the price of oil. The table presents separate estimates for each commodity, first from single equation methods, then from the seemingly unrelated regression, estimated by GLS. I fail to reject the hypothesis that all contracts respond the same amount to a given shock. (The null is equivalent to no overshooting.) However it is interesting to note that the pattern of responses, with smaller responses for contracts with more distant settlement dates, is consistent 15

with the overshooting model. Vector autoregression studies often include commodity prices to ameliorate the price puzzle – a temporary increase of consumer prices in response to a monetary tightening (Sims, 1992). My findings indicate that a common identification assumption used in these models – that monetary policy responds to contemporaneous commodity price movements but not vice versa – is invalid. I estimate that a surprise monetary policy tightening of 10 basis points would cause about a half percent reduction in commodity prices on the same day the tightening occurs. However, according to a baseline VAR (Christiano et al., 1999), a monetary policy tightening of approximately 10 basis points of the federal funds rate reduces commodity prices about 0.1% after one year and 0.2% after two years. The gradual response estimated in the VAR is in stark contrast to the immediate response I find. Figure 2 compares these estimates in graphical form.19 The confidence interval for the GSCI futures responses are wide and encompass the VAR impulse response function. According to my results, the near zero response of commodity prices estimated in the VAR is just as likely to be the truth as a 4% response. The recursively identified VAR cannot separately estimate immediate effects of commodity prices on monetary policy and of monetary policy on commodity prices. The results in this article suggest the effect of monetary policy on commodity prices, often assumed to be zero, may be substantial.

4.5.

Is the Commodity Price Response Driven by the Exchange Rate?

One interpretation of the commodity price movements generated by monetary policy surprises is that they are due to simultaneous changes in the foreign exchange value of the U.S. dollar. If the foreign price of gold is unaffected by a Federal Reserve tightening, but the exchange rate appreciates, then the U.S. dollar price of gold should fall so that U.S. dollar prices and foreign prices remain close together. I examine this idea by estimating the effect of a monetary policy surprise on a trade-weighted exchange rate index in a system of equations that also includes the 19 The data for the VAR come from Christiano et al. (1999). The commodity price variable is the percentage change in a smoothed index of commodity prices. The impulse response function I report is for the cumulative effect of the shock on the level of commodity prices.

16

commodity prices.

∆pjt = βg ∆it + jt

var(jt ) = σj2

(4)

∆et = βe ∆it + et

var(et ) = σe2

(5)

The system is displayed here, with j identifying the commodity, which belongs to a particular commodity group g. Within this system, I estimate one effect for metals, one for softs and grains, one for livestock, one for oil, and one for the exchange rate. Then I test whether the exchange rate response βe is equal to the commodity price response βg for each commodity group. I estimate that the U.S. trade weighted index exchange rate moves about 0.23% in response to a ten basis point monetary policy surprise. This is smaller than the response Craine and Martin (2008) find for the USD/AUD exchange rate. It is similar to what Faust et al. (2003) and Faust et al. (2007) find for the USD/DEM (or USD/EUR) exchange rate and larger than the response of the British Pound in the same paper. None of the effects is significantly different from the exchange rate response at the 5% level, though the p-value is 0.08 for metals. On balance, the evidence indicates that the response of commodity prices is broadly similar to the response of exchange rates.

4.6.

Does Carrying Cost Explain the Commodity Price Response?

The intertemporal arbitrage in the Hotelling model may be more difficult for some products than for others. Commodities such as hogs are more difficult to store and sell later than metals. This suggests the intertemporal arbitrage of the Hotelling model, and hence the responsiveness to changes in interest rates, may be more limited for hogs than for gold. If the gold spot price is below the expected future spot price, a higher interest rate should reduce the current price of gold, and storage should enforce this outcome. By contrast, the spot and expected hog prices may not be linked so closely, and a higher interest rate may not cause the current spot price of hogs to fall if storage is sufficiently costly to eliminate the possibility of arbitrage profits. As a first-pass attempt to test this idea, I construct a proxy measure of carrying costs then 17

interact this with interest rate changes. The carrying cost proxy is generated as the natural log of the value of a cubic meter of the commodity. For some commodities, this is very high (platinum, gold) while for others it is lower (cocoa, coffee). (Details are in a supplemental appendix.) When this measure is included in the Rigobon-Sack regression, interacted with the interest rate and instrumented for with the interaction of the carrying cost proxy and the usual Rigobon-Sack instrument, the coefficient on the interaction of carrying costs and the interest rate is significantly negative (p-value 0.045). This suggests that some of the variation in commodity price response to interest rate changes could be due to carrying cost variations. The proxy I use for carrying costs could also be related to transportation costs. An alternative interpretation for the finding that the carrying cost proxy is correlated with the response of commodity prices to interest rates is that commodities that are bulky are hard to transport, so arbitrage across space is difficult. A change in interest rates moves the exchange rate, so should move commodity prices more in internationally integrated commodity markets, which are more likely to be metals than agricultural commodities.

4.7.

Identifying the VAR

Recursively identified VARs commonly assume that changes in monetary policy have no immediate effect on commodity prices. The exclusion restriction allows monetary policy to respond to commodity prices. Commodity price movements may forecast inflation, so the central bank’s forward-looking behavior can be modelled through a response to commodities. It is possible to estimate the VAR under a nonzero restriction and consider the dynamics of the system.20 Using the data from Christiano, Eichenbaum, and Evans, I estimate the VAR imposing a nonzero instantaneous response of commodity prices to monetary policy. Christiano et al. impose a zero impact, while I impose a contemporaneous response that is consistent with estimates in table 3, but otherwise retain the recursive identification scheme. Christiano, Eichenbaum, and Evans use a smoothed index of commodity prices in their VAR. This series is very persistent and its growth rate (which is the form used in the VAR) is also very persistent. Commodity prices I 20

Anzuini et al. (2012) study the impact of monetary policy shocks on commodity prices in a VAR using sign restrictions for identification. My approach is closer to the recursive identification scheme.

18

study tend to be persistent in levels but not in first differences. (Unit root tests typically lead to the conclusion that the prices are I(1).) Smoothing commodity prices creates a variable that is much more persistent in growth rates than commodity prices really are. Consider the commodity price equation in the VAR. It has the form

(6)

∆pcom,t = α∆pcom,t−1 + β∆it + δzt + t

where z is a vector of other variables (including lags). The impact of a monetary policy surprise on the level of commodity prices is β in the short run and approximately

β 1−α

in the long run.

Christiano et al. impose β = 0. Earlier in this article, I estimated values of β that vary across commodities but that are close to -5 (the percent response of commodity prices to a 100 basis point monetary surprise), and I impose this value in estimating the VAR. In the CEE VAR α is close to one because of the persistence in the smoothed commodity price series. As a result, a one-time 100 basis point monetary policy surprise would cause commodity prices to fall by about five percent each month, month after month. That is, the impulse response function is unrealistic.

5.

Conclusion

On 16 November 1999, the Federal Open Markets Committee raised the federal funds rate 25 basis points. This move was mostly anticipated by financial markets, but the three-month Eurodollar futures rate did increase by five basis points. On the same day, commodity prices fell by about a quarter of one percent, broadly consistent with the effect of a monetary policy surprise on commodity prices as I estimate it. My results are in line with estimates of the effect of monetary policy on other asset prices. I estimate that monetary policy surprises have a smaller impact on commodities than on stock prices, but the effect is the same order of magnitude. In addition, I show that the movements in commodity prices following a monetary policy surprise are similar to the change in the foreign exchange value of the U.S. dollar, consistent with global commodity market integration in which changes in U.S. monetary policy have minor effects on the world prices of commodities, but induce large changes in U.S. dollar prices. 19

There is some heterogeneity in the responsiveness of commodity prices, with the response of individual commodities being correlated with the value of a given volume of the product, a proxy for carrying costs. A range of metals prices, not just gold, tend to respond to monetary policy. This adds to our knowledge of the impact of monetary policy on asset markets and the economy in general. Prices of many commodities doubled between 2000 and 2007. Some analysts have blamed this fact on the Federal Reserve’s loose monetary policy. According to the estimates in this article, each percentage point reduction in interest rates instigated by the Fed increases commodity prices around 5%. Monetary policy surprises over this period of time simply cannot have been large enough to generate this sustained increase in commodity prices if the estimates in this article are taken at face value.

20

References Anzuini, Alessio, Marco J Lombardi, and Patrizio Pagano, “The impact of monetary policy shocks on commodity prices,” 2012. Bank of Italy Temi di Discussione Working Paper 851. Barnhart, Scott W, “The effects of macroeconomic announcements on commodity prices,” American Journal of Agricultural Economics, 1989, 71 (2), 389–403. Barsky, Robert and Lutz Kilian, “Do We Really Know That Oil Caused the Great Stagflation? A Monetary Alternative,” NBER Macroeconomics Annual, 2001, 16, 137–183. Bernanke, Ben S., “The Global Saving Glut and the US Current Account Deficit,” Sandridge Lecture, Virginia Association of Economics, Richmond, Virginia, Federal Reserve Board March, 2005. and Kenneth N. Kuttner, “What Explains the Stock Market’s Reaction to Federal Reserve Policy?,” Journal of Finance, 2005, 60 (3), 1221–1257. Blinder, Alan S., “The Anatomy of Double Digit Inflation in the 1970s,” in Robert Hall, ed., Inflation: Causes and Effects, Chicago: University of Chicago Press, 1982, pp. 261–282. Bodie, Zvi, “Commodity futures as a hedge against inflation,” The Journal of Portfolio Management, 1983, 9 (3), 12–17. Caballero, R.J., E. Farhi, and P.O. Gourinchas, “An Equilibrium Model of “Global Imbalance” and Low Interest Rates,” American Economic Review, 2008, 98 (1), 358–393. ,

, and

, “Financial Crash, Commodity Prices, and Global Imbalances,” Brookings Papers on

Economic Activity, 2008, 2, 1–55. Christiano, Lawrence J., Martin Eichenbaum, and Charles Evans, “Monetary Policy Shocks: What Have We Learned and to What End?,” in John B. Taylor and Michael Woodford, eds., Handbook of Macroeconomics, Vol. 1A, Elsevier, 1999, chapter 2, pp. 65–148. Commodity Research Bureau, The CRB Commodity Yearbook 2007, Wiley, 2007. 21

Cook, Timothy and Thomas Hahn, “The Effect of Changes in the Federal Funds Rate Target on Market Interest Rates in the 1970s,” Journal of Monetary Economics, November 1989, 24, 331–351. Craine, R. and V.L. Martin, “International Monetary Policy Surprise Spillovers,” Journal of International Economics, 2008, 75 (1), 180–196. Demiralp, S. and O. Jorda, “The Response of Term Rates to Fed Announcements,” Journal of Money, Credit & Banking, 2004, 36 (3), 387–406. Dornbusch, Rudiger, “Expectations and Exchange Rate Dynamics,” Journal of Political Economy, December 1976, 84 (6), 1161–76. Faust, J., J.H. Rogers, E. Swanson, and J.H. Wright, “Identifying the Effects of Monetary Policy Shocks on Exchange Rates Using High Frequency Data,” Journal of the European Economic Association, 2003, 1 (5), 1031–1057. Faust, Jon, John H Rogers, Shing-Yi B Wang, and Jonathan H Wright, “The high-frequency response of exchange rates and interest rates to macroeconomic announcements,” Journal of Monetary Economics, 2007, 54 (4), 1051–1068. Frankel, J.A., “Commodity Prices and Money: Lessons from International Finance,” American Journal of Agricultural Economics, 1984, 66 (5), 560–566. , “Expectations and Commodity Price Dynamics: The Overshooting Model,” American Journal of Agricultural Economics, 1986, 68 (2), 344–348. , “The Effect of Monetary Policy on Real Commodity Prices,” in John Campbell, ed., Asset Prices and Monetary Policy, University Of Chicago Press, 2008, pp. 291–327. and G.A. Hardouvelis, “Commodity Prices, Money Surprises and Fed Credibility,” Journal of Money, Credit and Banking, 1985, 17 (4), 425–438. Glick, R. and S. Leduc, “Are large-scale asset purchases fueling the rise in commodity prices?,” FRBSF Economic Letter, 2011, 10 (4). 22

Glick, Reuven and Sylvain Leduc, “Central bank announcements of asset purchases and the impact on global financial and commodity markets,” Journal of International Money and Finance, 2012, 31 (8), 2078–2101. ¨ Gurkaynak, Refet S., Brian Sack, and Eric Swanson, “The Sensitivity of Long-Term Interest Rates to Economic News: Evidence and Implications for Macroeconomic Models,” American Economic Review, March 2005, 95 (1), 425–436. ¨ Gurkaynak, Refet S, Brian Sack, and Eric T Swanson, “Do Actions Speak Louder Than Words? The Response of Asset Prices to Monetary Policy Actions and Statements,” International Journal of Central Banking, 2005, 1 (1), 55–93. Hamilton, J. D., “Causes and Consequences of the Oil Shock of 2007-08,” Brookings Papers on Economic Activity, 2009, pp. 215–261. Heckman, James J. and Rodrigo Pinto, “Causal Analysis after Haavelmo,” Working Paper 19453, National Bureau of Economic Research September 2013. Hua, Ping, “On primary commodity prices: the impact of macroeconomic/monetary shocks,” Journal of Policy Modeling, 1998, 20 (6), 767–790. Jarvis, HFT, “The Thermal Variation of the Density of Beef and the Determination of its Coefficient of Cubical Expansion,” International Journal of Food Science & Technology, 1971, 6 (4), 383–391. Kilian, L. and C. Vega, “Do energy prices respond to US macroeconomic news? A test of the hypothesis of predetermined energy prices,” Review of Economics and Statistics, 2011, 93 (2), 660– 671. Kim, Soyoung, “Do Monetary Policy Shocks Matter in the G-7 countries? Using Common Identifying Assumptions about Monetary Policy Across Countries,” Journal of International Economics, August 1999, 48 (2), 387–412. Kuttner, Kenneth N., “Monetary Policy Surprises and Interest Rates: Evidence from the Fed Funds Futures Market,” Journal of Monetary Economics, June 2001, 47 (3), 523–544. 23

Kwon, Dae-Heum and Won W Koo, “Interdependence of Macro and Agricultural Economics: How Sensitive is the Relationship?,” American Journal of Agricultural Economics, 2009, 91 (5), 1194–1200. Obstfeld, Maurice, “Pricing-to-Market, the Interest-Rate Rule, and the Exchange Rate,” in Carmen Reinhart, Carlos V´egh, and Andr´es Velasco, eds., Money, Crises, and Transition: Essays in Honor of Guillermo A. Calvo, MIT Press, 2008, pp. 3–20. Rigobon, R., “Identification Through Heteroskedasticity,” Review of Economics and Statistics, 2003, 85 (4), 777–792. Rigobon, Roberto and Brian Sack, “The Impact of Monetary Policy on Asset Prices,” Journal of Monetary Economics, November 2004, 51 (8), 1553–1575. Roache, Shaun K and Marco Rossi, “The effects of economic news on commodity prices,” The Quarterly Review of Economics and Finance, 2010, 50 (3), 377–385. Romer, Christina D. and David H. Romer, “Federal Reserve Information and the Behavior of Interest Rates,” American Economic Review, June 2000, 90 (3), 429–457. Rudebusch, Glenn D, “Do Measures of Monetary Policy in a VAR Make Sense?,” International Economic Review, November 1998, 39 (4), 907–31. Sims, C.A., “Interpreting the Macroeconomic Time Series Facts,” European Economic Review, 1992, 36 (5), 975–1011. Taylor, J.B., “The Financial Crisis and the Policy Responses: An Empirical Analysis of What Went Wrong,” 2009. NBER Working Paper 14631.

24

Figure 1: Commodity prices and interest rates on event days and pre-event days Note: scatter plots of the percentage change in commodity prices against the change the three-month Eurodollar futures rate, on both pre-event days (left panel) and event days (right panel). The change in the interest rate is measured in basis points, while the change in the commodity price is measured in percentage terms. The graph includes data from each commodity series used in the article.

25

Figure 2: Dynamic response of commodity prices to monetary policy surprises Note: this graph shows the estimated response over time of commodity prices to a monetary policy surprise using the Christiano, Eichenbaum and Evans (1999) VAR and GSCI commodity futures. The responses are shown with 95% confidence intervals. In the VAR, these are constructed using the 2.5 percentile and the 97.5 percentile of the distribution of responses in a Monte Carlo simulation where the parameter vector is drawn from a normal distribution centered on the VAR estimates, with the estimated parameter covariance matrix.

26

Table 1: Variation of Commodity Prices and Covariation with Interest Rate

Commodity Gold Aluminum Copper Lead Nickel Platinum Silver Tin Zinc Cocoa Coffee Cotton Wheat Hogs Livestock Live Cattle Oil

Std. dev. of Commodity Price Event Day Pre-event 1.16 0.70 1.25 1.02 1.48 1.30 1.75 1.59 1.98 2.01 1.34 1.14 1.86 1.27 1.27 1.26 1.76 1.52 1.94 1.98 2.58 2.77 1.52 1.47 1.82 1.30 1.53 1.50 1.01 0.93 0.93 0.97 2.76 2.39

Covariance with Policy Rate Event day Pre-event -2.60 -0.10 -1.80 -0.04 -2.32 -0.36 -1.77 -0.26 -2.98 -0.68 -2.14 0.28 -3.98 0.39 -0.74 -0.19 -2.37 -0.32 -2.57 0.85 -1.97 -1.08 -0.62 0.12 -1.32 -0.52 -1.63 0.06 -0.97 0.05 -0.92 0.16 -2.18 -0.26

Note: statistics from regularly-scheduled FOMC meeting dates from January 1994 to March 2008. (Lead prices only available from 1995.)

27

Table 2: Estimated Effects of Interest Rates on Commodity Prices Commodity All Commodities Metals All Agricultural Softs Livestock Oil

Event Study -5.55 (1.75) -6.63 (2.24) -4.09 (1.87) -4.64 (2.69) -3.37 (1.69) -6.25 (3.67)

Rigobon-Sack -6.37 (2.09) -7.47 (2.70) -4.92 (2.27) -5.05 (3.33) -4.74 (2.15) -6.72 (4.73)

Note: standard errors, clustered at the day level, in parentheses. Standard errors for the effect on oil prices are heteroskedasticity-robust standard errors, since clustering is redundant with a single commodity. The softs category includes wheat, cotton, coffee, and cocoa. The livestock category includes livestock, hogs, and live cattle.

28

29

Event Study -7.43 (2.86) -5.15 (2.04) -6.64 (2.42) -5.30 (3.12) -8.54 (4.60) -6.13 (3.29) -11.40 (3.75) -2.11 (1.90) -6.78 (3.53 )

Rigobon Sack -8.54 (3.45) -6.03 (2.50) -6.68 (3.02) -5.04 (3.90) -8.18 (5.73) -8.51 (3.95) -15.13 (4.79) -1.67 (2.54) -7.20 (4.30)

Excl. March 08 -4.74 (1.65) -5.20 (2.50) -5.96 (3.08) -2.83 (3.47) -7.90 (6.21) -5.05 (2.90) -9.42 (2.67) -2.06 (2.78) -5.13 (4.11) 3SLS -4.52 (1.91) -6.28 (2.53) -7.38 (3.06) -2.83 (3.67) -10.85 (4.37) -5.22 (2.67) -8.87 (3.26) -3.21 (2.76) -6.76 (3.63) Oil

Live Cattle

Livestock

Hogs

Wheat

Cotton

Coffee

Commodity Cocoa

Event Study -7.35 (2.82) -5.65 (4.05) -1.77 (3.73) -3.78 (5.22) -4.68 (2.40) -2.77 (1.63) -2.64 (1.59) -6.25 (3.67)

Rigobon Sack -11.93 (5.00) -2.96 (5.51) -2.72 (4.56) -2.61 (6.55) -6.54 (3.12) -3.88 (2.03) -3.81 (2.06) -6.72 (4.73)

Excl. March 08 -8.32 (4.09) -1.04 (5.70) 0.12 (4.22) 1.70 (3.91) -7.23 (3.39) -4.09 (2.18) -3.73 (2.08) -5.82 (4.45)

3SLS -7.81 (4.27) 2.11 (5.76) 2.26 (3.31) 2.83 (3.35) -7.98 (3.42) -5.17 (2.15) -5.19 (2.07) -4.32 (5.78)

system of equations. 3SLS estimates use data excluding March 2008 and also excluding 1994 since lead prices are unavailable for that year.

Note: heteroskedasticity-robust standard errors in parentheses, except for 3SLS estimates, which are derived from a generalized least squares estimate of the

Zinc

Tin

Silver

Platinum

Nickel

Lead

Copper

Aluminum

Commodity Gold

Table 3: Estimated Effects of Interest Rates on Commodity Prices

Table 4: Robustness Checks

Commodity All Commodities All Metals All Agricultural Softs Livestock Oil

With ’89-’93 -4.84 (1.86) -5.68 (2.38) -4.13 (1.88) -4.64 (2.68) -3.45 (1.71) -3.03 (4.20)

With Testimonies -8.62 (2.95) -9.88 (3.80) -7.04 (3.24) -6.32 (4.50) -8.01 (3.64) -8.59 (7.77)

Excluding Reversals -6.22 (2.31) -6.74 (3.05) -5.55 (2.47) -5.87 (3.62) -5.12 (2.43) -6.38 (5.24)

Kuttner Method -4.25 (1.63) -4.60 (2.14) -3.87 (1.72) -4.52 (2.31) -3.01 (1.70) -3.83 (3.55)

With Unscheduled -5.63 (2.22) -7.67 (2.94) -3.44 (2.07) -3.59 (2.87) -3.24 (1.88) -2.68 (4.09)

Target -4.76 (1.08) -3.01 (2.15) -6.62 (2.68) -6.54 (3.56) -6.73 (3.16) -7.11 (7.57)

Path -6.78 (2.86) -7.40 (4.21) -4.11 (3.44) -2.91 (5.14) -5.72 (5.33) -20.13 (9.86)

Note: standard errors, in parentheses, are clustered at the day level (or robust standard errors for oil, as in table 2).

30

Table 5: Response of Commodity Price Futures

Contract GSCI 1Month GSCI 2Month GSCI 3Month GSCI 4Month GSCI 5Month

Single Equation -4.03 (4.11) -3.69 (4.04) -3.38 (4.03) -3.38 (3.98) -3.34 (3.95)

3SLS -4.03 (2.74) -3.69 (2.56) -3.38 (2.48) -3.38 (2.39) -3.34 (2.31)

Note: this table presents estimates of the effect of monetary policy surprises on commodity futures contracts, using the Goldman-Sachs Commodity Index futures for one through five months. Robust standard errors for the first column, 3SLS standard errors for the second column.

31

A

Supplement: Data

The following gives information on data sources.

A1.

Data Sources Table A.1: Data Sources Commodity Gold Aluminum Copper Lead Nickel Platinum Silver Tin Zinc Cocoa Coffee Cotton Wheat Livestock Live Hogs Live Cattle Oil GSCI 1-month GSCI 2-month GSCI 3-month GSCI 4-month GSCI 5-month Three-Month Eurodollar Rate Federal Funds Futures Rate Trade-weighted Index

A2.

Source Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Datastream Energy Information Administration Datastream Datastream Datastream Datastream Datastream Global Financial Database Wikiposit FRED

Code GSGCSPT GSIASPT GSICSPT GSILSPT GSIKSPT PLATFRE GSSISPT LTICASH GSIZSPT GSCCSPT GSKCSPT GSCTSPT GSWHSPT GSLVSPT GSLHSPT GSLCSPT RCLC1 GS1MSPT GS2MSPT GS3MSPT GS4MSPT GS5MSPT IBUSA3D FFF0, FFF1 DTWIM

Carrying Cost Proxy

The carrying cost proxy is the log of the dollar value of a cubic meter of the commodity. I value a cubic meter of the commodity using the average price over the sample period, converted into dollars per kilogram, combined with an estimate of the density of the commodity. The density 32

information is taken from Wolfram Alpha, si metric.co.uk, and engineeringtoolbox.com, sources that report a density of each commodity. The livestock is the least standard of these, and I use an estimate of the density of beef from Jarvis (1971). I use Commodity Research Bureau (2007) for various conversions and some price information where necessary to corroborate other sources.

33