Measuring Oil-price Shocks Using Market-based Information

Transcription

1 Measuring Oil-price Shocks Using Market-based Information Tao Wu Michele Cavallo This version: January 29 First draft: September 26 Abstract This paper takes on a narrative and quantitative approach to examine the dynamic effects of oil-price shocks to the U.S. economy. Based on market information collected from various oil-industry trade journals, we separate different kinds of oilprice shocks, and construct measures of exogenous oil shocks that are free of endogeneity and anticipatory problems. Estimation results indicate that oil shocks have had substantial and statistically significant impacts on the U.S. economy during the past two and a half decades. By contrast, traditional VAR identification strategies lead to a much weaker and insignificant real effect for the same period. Further investigation suggests that this discrepancy is possibly due to a lack of identification on the VAR approach, originating from mixing the exogenous oil-supply shocks with endogenous oil-price movements driven by changes in oil demand. jel classification codes: C32, C82, E31, E32, Q43 For helpful comments, we thank James Hamilton, Òscar Jordà, Lutz Kilian, Monika Piazzesi, Christina Romer, David Romer, Eric Swanson, Mark Bergen, John Fernald, and seminar and conference participants at the NBER summer institute, Oxford University, Bank of England, Bank of Italy, the SCE conference, and the SED conference. Tiffany Smith, Thien Nguyen, Michael Simmons and Nina Ozdemir provided excellent research assistance. The views expressed in this paper do not necessarily reflect those of the Federal Reserve Bank of Dallas or the Federal Reserve System. Corresponding author. Federal Reserve Bank of Dallas, Tao.Wu@dal.frb.org. Federal Reserve Board, Michele.Cavallo@frb.gov.

2 1. Introduction The relationship between oil-price shocks and the macroeconomy has attracted extensive scrutiny by economists over the past three decades. However, literature has not reached a consensus on how oil price increases have affected the real economy, or by how much. A large body of the literature relies on various Vector Autoregression (VAR) identification strategies to isolate exogenous oil-price shocks and estimate the effect of such shocks. Nevertheless, estimation results are generally inconsistent with conventional wisdom, and the estimated relationship is often unstable over time. This is why, after a careful examination of various methodologies, Bernanke, Gertler, and Watson (1997) conclude that finding a measure of oil price shocks that works in a VAR context is not straightforward. It is also true that the estimated impacts of these measures on output and prices can be quite unstable over different samples. Conventional VAR-based measures of oil-price shocks have two obvious flaws: endogeneity and predictability. First, VAR identification strategies often cannot separate the effects of an exogenous oil-price shock from those of an endogenous response of oil prices to other kinds of structural shock. For instance, the oil price increases since 22 are viewed by many as the result of an expanding world economy driven by gains in productivity (The Wall Street Journal, August 11, 26). Kilian (26, 27) has indeed found that standard measures of oil-price shocks based on changes in spot oil prices do not represent the exogenous component of oil prices, but are likely to reflect movements driven by endogenous factors. Such endogenous movements will undoubtedly lead to biased estimates of the effects of oil shocks. On the other hand, part of the observed price changes might have already been anticipated by private agents well in advance, therefore, they are hardly shocks. Most measures of oil-price shocks in the literature are constructed using only the spot oil price. However, when the market senses any substantial supply-demand imbalances in the future, changes on the spot prices maynotfullyreflect such imbalances. A number of authors (for instance, Wu and McCallum, 25; Chinn, LeBlanc, and Coibon, 25) have found that oil futures prices are quite powerful in predicting the spot price movement, which indicates that at least some of the oil-price movement may have been anticipated several months in advance. Both of these concerns underscore the necessity of taking different approaches to obtain better measures of exogenous oil-price shocks. In this paper, we take a narrative and quantitative approach and construct alternative measures of exogenous oil-price shocks that avoid these problems. We begin by identifying the 1

3 events that have driven the daily oil-price movement for the entire period from 1984 to 27. For this purpose, we construct a database identifying the major oil-related events that have occurred each day since January 1984, based on the information we collected from daily oilmarket commentaries published in several oil-industry trade journals, including Oil Daily, Oil & Gas Journal, andmonthly Energy Chronology. We then categorize these daily oil-related events into 22 different groups according to their identity, such as weather changes in the U.S., military actions in the Middle East, OPEC proposals on oil production, and U.S. oil inventory announcements, etc. (Table 1). Next we construct our measures of oil shocks for each kind of event by running oil-price forecasting equations on a daily basis. Finally, the series of shocks from exogenous events are selected and aggregated into a single series of exogenous oil-price shocks. By construction, such shock series should be free of both endogeneity and anticipatory problems. For robustness, we provide several alternative definitions of exogeneity and construct shock measures for each of them. We employ our new, market-based measures to analyze the responses of U.S. output, inflation, and monetary policy to exogenous oil-price shocks, and compare them with the responses one may obtain following two conventional VAR-based identification strategies that are very popular in the literature. We obtain substantial and statistically significant output and price responses to exogenous oil-price shocks as identified by market information. In contrast, responses obtained based on VAR identification strategies are much weaker, statistically insignificant, and unstable over time. Moreover, we find that after a demand-driven oil-price shock, real GDP rises and price level declines. This finding is consistent with the scenarios in which changes in oil prices are endogenous responses to economic expansion rather than exogenous oil shocks. We argue that the VAR identification strategies cannot separate the effects of these two kinds of shocks, and consequently the estimated output responses are much weaker, insignificant, and unstable over time. The approach we take in this study is similar in spirit to the narrative approach as advocated by Romer and Romer (24, 26) in their analyses of monetary policy and tax shocks. Several previous studies also have tried to isolate some exogenous events associated with abrupt oil-price increases to examine their effects on the U.S. economy. For instance, Hamilton (1983) identifies eight oil price episodes before 1981, mainly the geopolitical tensions that occurred in the Middle East, and concludes that such oil shocks had effectively contributed to the post-war U.S. 2

4 recessions. Dotsey and Reid (1992) use the same set of dummies and reach a similar conclusion. Hoover and Perez (1994) revise Hamilton s (1983) quarterly dummies into a monthly dummy series and find that oil shocks lead to declines in the U.S. industrial production. Bernanke, Gertler, and Watson (1997) further construct a quantitative measure, weighing Hoover and Perez s dummy variables by log-change in oil prices, yet they cannot find statistically significant macroeconomic responses to oil shocks in a VAR setting. Finally, Hamilton (23) identifies five military conflicts during the postwar period and re-examines the effect of associated oil shocks to the postwar U.S. GDP growth. Our study, by constructing a chronology of all major oil-related events on a daily basis, is much more comprehensive than previous relevant studies. Extracting surprise components of the oil-price movement using an oil futures price-based forecasting model is another novelty of our methodology. A recent study by Kilian (27) also uses information from the oil market to disentagle various kinds of oil-price shocks. In particular, he constructs an index of global real economic activity and includes it in a three-variatevar, along with data on world oil production and real oil prices, and separates the oil-supply shocks, aggregate demand-driven shocks, and oil market-specific demand shocks assuming a recursive VAR structure. Although his approach is completely different from ours, the estimated dynamic effects of all three kinds of structural shocks on the U.S. economy are quite close to our empirical estimates. This in turn confirms the appropriateness of our approach and we will discuss the relevant findings in details in later sections. Our study also sheds light on the ongoing debate about how the real effects of oil shocks on the economy have changed over time. For instance, VAR studies such as Hooker (1996) and Blanchard and Gali (27) usually find a much weaker and statistically insignificant relationship between the identified oil-price shocks and real GDP growth in the U.S. and other developed economies in the past two or three decades, and these results are often used as evidence suggesting that the U.S. economy has become much less volatile and more robust to external shocks, a result of either better economic policy, or a lack of large adverse shocks, or less energy dependency (more energy efficient and a larger share of services sector), all contributing to a great moderation of the U.S. economy starting the early 198s. Although we do not challenge this general description of great moderation, the estimation results presented below still indicate a substantial and significant adverse effect on the U.S. economy from external oil shocks. 3

5 Evidences from VAR studies may suffer from a poor identification strategy. The rest of the paper is organized as follows. Section 2 describes our methodology in identifying the oil-related events and constructing the oil-shock measures. Section 3 illustrates the procedures that we follow to estimate the macroeconomic effects of oil-price shocks. Section 4 presents the empirical results. Section 5 offers conclusions. 2. Measures of Exogenous Oil-Price Shocks Based on Market Information The derivation of our market information-based measures of oil-price shocks consists of three key steps. We first conduct a thorough and careful examination of the oil-related events that have driven oil-price movement over the past two and a half decades on a day-to-day basis, and categorize them into 22 event types. Then we perform a daily oil-price forecasting exercise, and construct shock measures that capture the unpredictable component of oil price movements, one for each event type. Finally, we combine shock series of all types of events that are exogenous and construct a single series of exogenous oil-price shocks. To be robust, we provide several alternative definitions of exogeneity, and for each definition we aggregate daily shock series of various event types into a single series of exogenous shocks, which by construction is free of endogeneity and anticipatory problems A Comprehensive Study of Daily Oil-Related Events The first step of our methodology is to identify the events behind the observed oil-price movements. 1 For this purpose we collect information from various oil-industry trade journals, such as Oil Daily, andoil & Gas Journal, etc. We then cross-examine the accuracy of this information with various government publications, such as Monthly Energy Chronology, published by the Energy Information Administration, a statistical agency of the U.S. Department of Energy. For each trading day of our sample period, which runs from January 3, 1984 to October 31, 27, we collect information on major oil-related events that occurred on that day from the above sources, in particular the market commentaries or reviews published in the industry trade 1 To be consistent with the literature and oil industry terminology, throughout the paper we refer spot oil price to price quoted on one-month futures contracts of West Texas Intermediate light sweet crude oil traded at the New York Mercantile Exchange (NYMEX). This is also the spot price most financial presses, such as The Wall Street Journal, quote every day. 4

6 journals. We define major oil-related events as those that significantly affected oil prices and received extensive coverage in the market analysis for the day. After a careful reading of the materials about all 5,971 trading days, we categorize oil-related events into 22 different types (Table 1), such as weather changes in the U.S., military actions in the Middle East, OPEC development on oil production, and U.S. oil inventory announcement, etc., and assign a code to each trading day based on the event analysis, or more than one code if more than one oil-related event occurred on the same day. We conduct the event study on a daily frequency, because the crude oil market, like other well-developed financial markets, is highly volatile and very sensitive, and responds immediately to economic, political, or industry-specific news. Choosing a lower frequency, such as monthly or even weekly, will result in a situation in which several events might have happened during thesameperiod,anditwillthenbedifficult to quantify how much a shock each of these events has brought to the oil market, or distinguish the exogenous shocks from endogenous ones. Daily is the highest frequency for which we can find relevant market information. To minimize the possibility that the information interpretation and event categorization may be biased by the analyst s subjective predispositions, a thorough content analysis, whichisa practice widely used in marketing literature, is conducted (Kassarjian 1977, Levy, Dutta, and Bergen 22). Specifically, three independent groups of analysts were engaged in reading the documents and categorizing the events, and the results were compared to make sure that they agree in most cases. Figure 1 shows the observed frequencies of oil-related events from 1984 to 27. Excluding thedayswithnoparticularreasonobservedorwhenthepricemovementwasprimarilydrivenby speculations, the most frequent event is OPEC development on oil production (742 trading days, or 22% of the sample), followed by U.S. oil inventory announcement (73 days, or 22%), and political development in the Middle East (476 days, or 15%). Oil production or transportation disruptions in the U.S. and around the world (types 3 and 4) had influenced oil-price movement on about 486 trading days, or 14% of the sample. Other types of events include changes in the market s expectations of U.S. oil inventories, U.S. weather changes, and changes in oil demand in the U.S. and around the world, etc. 5

7 2.2. Two Measures of Oil-Price Shocks The next step is to quantify the shocks implied by each oil-related event on a daily basis. Two approaches are adopted. The first one is based on a modified version of an oil-price forecasting model in Wu and McCallum (25). In particular, for each day, we run the following regression of oil-price changes on spreads between oil spot and futures prices at different horizons quoted by the end of the previous trading day, with a rolling sample consisting of the previous 2 trading days: 2 log P S t+1 log P S t = α + 6X j=2 β j (log P F j,t log P S t )+ε t+1, (2.1) where Pt S and Pt+1 S are the spot price at t and t +1, respectively, P j,t F the j-month oil futures price at time t, andα and β j s the estimation coefficients. We then calculate the unexpected change in spot price as realized on the following day, and define the predicting error as our shock measure for the day. Equation (2.1) incorporates term structure information on futures-spot spreads in forecasting future oil-price movement. This equation is in the same spirit as the bond-yield forecasting model in Cochrane and Piazzesi (22, 25), who also use information embodied in term spreads of interest rates at all available horizons to forecast future bond-yield movement, without imposing Expectations Hypothesis. Wu and McCallum (25) compare the out-of-sample forecasting performance of such a futures-spot spread model with that of several other models and conclude that the futures-spot spread model performs the best, particularly when the forecasting horizons are within the next few months. On the other hand, we exclude price quotes on futures contracts beyond six months from the equation, as the futures market becomes much less liquid for those horizons, so the quoted futures prices become a much less accurate measure of oil price expectations. Wu and McCallum (25) also find that the out-of-sample performance of the futures-spot spread model is much worse when the forecasting horizon goes beyond one year. Alternatively, we define the shock as simply the change in the logarithm of spot oil price on the day. This definition is consistent with the belief that oil futures prices do not have any predictability on future oil prices (Alquist and Kilian, 27), and that log oil price follows a random walk process. However, one needs to keep in mind that both our log-price change 2 Changing the length of the rolling sample has negligible effects on the forecasting results. 6

8 measure and predicting error measure are formulated on a daily basis and, with the identification of the oil-related events, will be constructed around exogenous event days so both will be legitimate measures of exogenous oil shocks for later econometric analysis What Does Exogeneity Mean? Having categorized the event days and constructed daily shock measures for each event type, the next step is to combine shock series of all types of events that are exogenous and construct a single series of exogenous oil-price shocks. To this end, one needs to first explicitly define what exogeneity he is referring to, and then combine the relevant series into one single series. A genuine exogenous shock, ideally, should be defined as exogenous to the U.S. economy in the most rigorous sense. Therefore, any events that could possibly correlate with the U.S. economy cannot be a genuine exogenous shock. For instance, weather changes in New England (category 1 in Table 1) do not qualify as exogenous oil-demand shocks, as such changes may affect not only oil demand but also utility output, construction activities, and retail sales. For the same reason, oil production and transportation disruptions in the U.S. (category 3) do not qualify as exogenous oil-supply shocks, as the energy industry is a substantial component of the national economy. Moreover, military conflicts (categories 1-12) do not qualify, as they will affect U.S. defense spending even when the U.S. is not directly involved, one may still suspect that the military buildup in response to such conflicts would make them correlate with real GDP in addition to their impacts on oil supply and demand (Ramey and Shapiro 1998, Ramey 28). For the same reason, political developments (categories 7-9) may not qualify as genuine exogenous shocks. These concerns basically rule out most of the oil-related events that researchers usually think are exogenous, as during the past 24 years, there were only six trading days in which non-u.s. weather changes (category 2) had significantly affected the oil market, and there was no new oil field discovered anywhere (categories 5 and 6) that had significantly affected the oil market. Therefore, we need to take a step back and allow for some ambiguity in the following econometric analysis. Again, different assumptions may lead to quite different interpretations of exogeneity. So instead of providing one single series and calling it the exogenous oil shock measure, we provide several alternative definitions of exogenous oil events and accordingly con- 7

9 struct several series of exogenous oil shock, 3 and will examine their dynamic implications in later sections: 1). Our baseline definition consists of event categories 1-12, including the U.S. and non-u.s. weather changes, oil production or transportation disruptions, and political and military actions. These are typically the kinds of events that many think are exogenous to the U.S. economy. It is interesting to note that political developments and military tensions in non-oil producing countries (categories 9 and 12) were not mentioned even once in the oil market analysis during the past 24 years, which indicate that they had essentially no effect whatsoever on the world oil supply and demand. 2). Alternatively, we make a narrow definition, consisting of only event categories 2-9, that is, the non-u.s. weather changes, oil production or transportation disruptions, and political developments. Compared with the baseline specification, this narrower definition excludes the U.S. weather changes and military actions because they are more likely correlated with real GDP in the U.S. 3). We also adopt a broad definition of exogeneity, consisting of event categories 1-12 and This definition includes not only the categories in our baseline definition, but also events such as oil and gas inventory announcements (for instance, the Energy Information Administration s weekly inventories reports) or changes in market expectations of oil inventories. These oil events are described in Kilian (27) as precautionary demand shocks, as they are likely associated with concerns about the availability of future oil supplies. We do not include in our definition events such as OPEC or non-opec oil exporters changes of their oil production plan or proposals (categories 13 and 14), because these events are more likely endogenous responses to development in other sectors of the world economy. For the same reason, changes in oil demand, such as global economic development, and improvement in oil usage efficiency (categories 18 and 19) are not included in any of the exogeneity definitions. However, these events are a very important part of the world oil market movement, and their implications will be separately examined in the econometric analysis later. 4 3 We thank Christina and David Romer for making this suggestion. 4 In constructing the database, we have preserved as much primitive information on the oil-market movement as possible to facilitate following researchers. Interested readers can choose their own definitions of exogeneity and construct alternative measures of their own. 8

10 2.4. Constructing Monthly Shock Series To facilitate the monthly VAR analysis, we sum up the daily shock series to obtain monthly series. Specifically, for each trading day, the daily shocks computed in equation (2.1) are assigned to an event category based on the code assigned to that particular trading day (or equally divided if there are multiple codes on that day). We then sum up the 22 daily shock series into 22 monthly series. Finally, a monthly oil-shock measure is constructed for the VAR analysis that correspondes to each exogeneity definition or any other interesting combination of oil-related events. Figures 2a and 2b display the market information-based measures of oil shocks, with the shocks defined as the predicting error in equation (2.1) and as the log-price change, respectively. The three market information-based measures are quite similar. Take the predicting error shocks for example: the correlations between the baseline measure and the broad measure are 78 percent, and the correlations are 72 percent between the baseline and the narrow measures and 55 percent between the narrow and the broad measures. On the other hand, the two methods of constructing the shocks, (i.e., the predicting error and the log-price change ), yield very similar shock series: The correlations between the baseline measures constructed in these two different ways are 9 percent, and the correlations are 87 percent for the broad measure and 88 percent for the narrow measure. Figures 3a and 3b compare our market information-based measures with two VAR-based measures that are very popular in the literature. The first one is the Net Oil Price Increase (NOPI) measure, or asymmetric VAR-based measure proposed by Mork (1989) and Hamilton (1996), and the second one is a symmetric VAR-based measure as in Bernanke, Gertler, and Watson (1997) and Blanchard and Gali (27). Both VAR-based measures are defined as residuals from a recursive VAR that includes macroeconomic variables and an oil-price variable on the bottom. The only difference is whether it is the net oil price increase (the maximum of zero and real oil price change over the preceding year) or the real oil price change that enters the VAR. Details of variable construction will be explained in Section 3. As shown in Figures 3a and 3b, there is substantial similarity as well as significant differences between the traditional VAR-based measures and our market information-based measures. Both kinds of oil-price shock measures capture major oil-price spikes reasonably well, for instance, in March-April 1986, August-September 199, December-February 1991, April 1999, 9

11 and September-October 24. However, the magnitudes of the shocks are somewhat different, as the symmetric VAR-based measures are more volatile than the market information-based measures, and the asymmetric VAR-based measures are the least volatile. Correlations between our baseline oil-shock measure and the asymmetric VAR-based measure are 24 percent, and 23 percent between the baseline and the symmetric VAR-based measures. Correlations between the two VAR-based measures are 53 percent. Such differences are not surprising. First, the two VAR-based measures are residuals from a macroeconomic VAR, whereas the market information-based measures are residuals from the oil price forecasting equation that does not incorporate macroeconomic variables, or simply the oil price change. Second and more important, their identification strategies are completely different, as we adopt an event study approach and rely on the market information to identify the exogenous oil shocks, whereas the traditional VAR-based measures rely on the recursive VAR identification strategy. For instance, if the oil price rises sharply because of an expansion of the world economy, the traditional VAR will interpret the price increase as an oil shock, yet our methodology will identify it as an increase in oil demand due to economic development (categories 18 or 19), and will exclude it in any of our exogenous oil-shock measures. Another example is the spike in April 1986, during which the crude oil price rose rapidly from $1.42 on March 31, 1986, to $13.3 on April 3, 1986, a 28 percent increase in a single month. Our investigation points to the following events that occurred during that month: (1) rumors and official confirmation of George Bush s plan to visit Saudi Arabia to discuss recent oil price increases during the first week of April (type 7: political development in the Middle East); (2) a strike in the Norway oil industry on April 7 (type 4: oil production disruptions outside the U.S.); (3) Iranian rockets hit a Saudi Arabian oil tanker on April 7 (type 1: military actions in the Middle East); (4) U.S. bombing of Libya on April 14 and 15 (type 1: military actions in the Middle East); (5) on April 21 and 22, reports on the likely change of OPEC s production quota; and (6) in the last two weeks of April, reports suggest that stronger-than-expected gasoline demand had tightened gasoline inventories and put upward pressure on crude oil prices (type 15: changes in U.S. gas inventory). Accordingly, events 1-4 are included in constructing our baseline shocks measure for this month, which implies a 26% price increase for the month. However, the asymmetric VAR-based shock measure is almost zero, as the higher oil price in the previous year led to a zero reading of the Net Oil Price Increase. 1

12 3. Estimating the Effects of Oil-Price Shocks Next, we examine the effects of oil-price shocks on the U.S. economy in a vector autoregressive model. In particular, the following equations are estimated X t = A + A 1 t + A 2 (L) X t 1 + B (L) O t + ε t, (3.1) where X t is a vector that contains (1) the logarithm of real GDP; (2) the logarithm of Consumer Price Index (CPI); (3) the level of the federal funds rate; and (4) the logarithm of real oil price, which is defined as the difference between the logarithm of Producer Price Index (PPI) for crude oil and the logarithm of the CPI. O t is an oil-price shock measure, t the time trend, and ε t is a vector of disturbance terms. A and A 1 are vectors of coefficients, while A 2 (L) and B (L) are polynomials in nonnegative power of the lag operator L. The estimated dynamic responses of the endogenous variables in X t to an oil-price shock in k periods are given by the estimates of the coefficients on L k in the expansion of [I A 2 (L) L] 1 B (L). This strategy is also adopted by Christiano, Eichenbaum, and Evans (1999) and Burnside, Eichenbaum, and Fisher (24) in estimating the effects of identified monetary policy shocks and fiscal policy shocks, respectively. Data are monthly, the VAR is estimated using six lags, and the sample period runs from January 1984 to October 27. Our choice of the VAR variables is very similar to Bernanke, Gertler, and Watson (1997), except that they also include a commodity price index in their VAR to examine the effects of monetary policy shock. As the primary focus of our study is oil-price shocks, we choose not to include the commodity price index in X t. The real GDP is only available on quarterly frequency, and we follow Bernanke, Gertler, and Watson (1997) and interpolate the quarterly real GDP into a monthly series using Chow and Lin (1971) algorithm. In an earlier version of the paper, we have also estimated a uni-variate autoregressive model to examine the effect of oil-price shocks, the strategy adopted by Ramey and Shapiro (1998) and Kilian (26), and the estimated dynamic effects are similar. We substitute O t by the alternative market information-based shock measures in Section 2 (baseline, broad, or narrow definition of exogeneity, and shock size calculated by log-price change or prediction error methods), one at each time, to examine the effects of oil-price shocks. As mentioned in Section 2, for comparison we also estimate the impulse responses of two traditional VAR-based oil shock measures. The first is the asymmetric VAR-based shock measure as in Hamilton (1996). In particular, the Net Oil Price Increase (NOPI), i.e., the 11

13 maximum of zero and the difference between the logarithm of current oil price and the maximum value of the logged oil price during the previous year, is constructed and enters the bottom of a recursive four-variate VAR consisting of the logarithm of real GDP, the logarithm of CPI, the level of the federal funds rate, and the NOPI. The fitted residual of the NOPI is thus defined as the asymmetric VAR-based oil-price shocks. The second measure is the symmetric VAR-based measure as in Bernanke, Gertler, and Watson (1997) and Blanchard and Gali (27), and is constructed in a similar fashion as the asymmetric VAR-based measure, except that it is the logarithm of the real oil price (deflated by the CPI), rather than the NOPI, that enters the bottom of the four-variate VAR. 4. Empirical Results 4.1. Impulse Responses of Market Information-Based Exogenous Oil-Price Shocks Figures 4a-4f display the estimated impulse response functions for real GDP, the CPI, the federal funds rate, and real oil price to an oil-price shock constructed using market information. As the impulse responses of the real oil price will be different when different oil-price shock measures are substituted into equation (3.1), to facilitate the comparison of their macroeconomic effects, we normalize these oil-shock measures so that the peak response of oil price is 1 percent. 5 This is roughly 1.75 times the estimated standard deviations of oil-price shocks constructed according to the baseline market information-based measure. Inference is based on a bootstrap Monte Carlo procedure with 1, replications, and 95 percent confidence intervals for the point-estimates are denoted by the shaded area. The estimated impulse responses fit very well with the conventional wisdom about the effects of an exogenous oil-price shock. Take the case of baseline definition, log-price change for example (Figure 4a). In response to the shock, real GDP gradually declines, and the maximal response arrives 18 months after the shock. The response becomes significant three months after the shock and remains significant throughout the 24-month horizon. The point estimate 5 Blanchard and Gali (27) choose to normalize the size of oil shocks so that they raise the oil price by 1 percent on impact. As can be seen in Figures 4a-4f, our estimation implies that the responses of the real oil price usually are hump-shaped, with the peak arriving in the first or second month after the shock. Therefore, we choose to normalize the size of the shocks according to their maximal responses rather than responses on impact. In most cases the normalized shock sizes are very similar. 12

14 suggests a substantial impact of the oil shock to the real economy: The output loss accumulates to 6.8 percent of a month s real GDP over the 24 months after the shock, or about.6 percent of annual real GDP in two years. The CPI shifts up immediately on impact, and the peak arrives three months after the shock. The price increase is both substantial and persistent, on average 14 basis points higher during the 24-month period, and the price level remains.1 percent higher than the original level even 24 months after the shock. In responding to the initial rise in the consumer price level, the federal funds rate rises a few basis points in the first three months after the oil shock. However, with the real GDP continuing to decline and inflation gradually decelerating, monetary policy becomes more accommodative, and the 24-month cumulative decline in the federal funds rate reaches 2.6 percentage points, or about 11 basis points lower than the pre-shock level each month, on average. The response of the real oil price to the oil shock is hump-shaped, with the peak arriving one month after the shock, and the oil price increase remains significant even eight months after the shock. The impulse responses to broadly defined exogenous oil shocks and narrowly defined shocks are similar. As shown in Figure 4b, in response to a broadly defined exogenous oilprice shock, real GDP declines immediately, and the output loss is persistent and even stronger than in the baseline case, with the 24-month cumulative output loss reaching 8.4 percent, or.7 percent of a year s real GDP in two years. The CPI rises on the shock, and price increase is significant for the first six months. Again, the federal funds rate initially rises in response to the CPI inflation but then declines as the output loss persists. A narrowly defined exogenous oil-price shock also leads to a persistent output decline and consumer price increase, and the monetary policy is again restrictive at first but switches to accommodative as the output loss continues. Figures 4d-4f display the impulse responses to exogenous oil-price shocks, defined according to the baseline definition, the broad definition, and the narrow definition, respectively, with the shock sizes calculated according to the oil-price forecasting equation (2.1) using information from oil futures prices. Again, in response to the exogenous oil-price shocks, real GDP declines and consumer price increases, and the federal funds rate becomes first restrictive and then accommodative. Compared with Figures 4a-4c, the point estimate of output loss is slightly smaller. For instance, the 24-month cumulative output loss is 5.6 percent in response to the baseline oil-price shock in Figure 4d, whereas it is 6.8 percent when shock size is calculated 13

15 according to the log-price change method in Figure 4a. The responses of real GDP and the CPI, however, are still statistically significant and persistent Impulse Responses of VAR-Based Measures of Exogenous Oil-Price Shocks The impulse responses displayed in Figures 4a-4f are substantially stronger and much more significant than previous estimates of the effects of oil shocks in the literature, in particular the output responses. For instance, Bernanke, Gertler, and Watson (1997) estimate the macroeconomic effects of oil-price shocks in a VAR setting, using four alternative indicators of oil shocks: (1) changes of log of nominal PPI for crude oil, (2) Hoover-Perez dummies for political and military events in the Middle East, and scaled by log-change of the nominal oil price, (3) Mork measure (positive monthly changes of logarithm of real oil price), and (4) Hamilton s NOPI measure. The VAR was estimated over the period , and none of the specifications generate a statistically significant output response. Moreover, as the federal funds rate persistently rises in responding to the higher price level after the oil shock, Bernanke, Gertler, and Watson (1997) argue that it is impossible to determine how much of the output decline is the direct result of the oil shock, as opposed to the tightening monetary policy, an identification problem in the VAR. Therefore, they conclude that finding a measure of oil price shocks that works in a VAR context is not straightforward. Hooker (1996) employs a similar VAR and examines two specifications, with oil price entering the VAR in nominal log-differences and in real log-levels, respectively. He finds that oil prices no longer Granger-cause the U.S. GDP or unemployment from 1973 to Indeed, GDP growth exhibits a large positive response to an oil price increase for about four quarters and quickly returns to zero, contradicting the conventional wisdom of the macroeconomic effects of oil shocks. Bernanke, Gertler, and Watson (1997) also report a similar output response when using log-change of nominal oil price to measure the state of the oil market. However, by using his Net Oil Price Increase measure in a uni-variate time-series model, Hamilton (1996) still finds a negative output response to an oil-price shock after 1973, although the response is much weaker than pre-1973 and coefficient estimates are insignificant. To illustrate the difference between the macroeconomic implications of our market informationbased measures and those of the traditional measures, we construct two VAR-based measures (defined in Section 3) and estimate the same VAR equations (3.1). Figure 5a displays the im- 14

16 pulse responses to an oil-price shock, constructed according to Hamilton s (1996) measure, i.e., an asymmetric VAR-based measure. Again, the shocks are normalized so that the peak response of oil price is 1 percent. After such a shock, real GDP declines and the CPI rises. However, the output response is no longer statistically significant, and much weaker than what is implied by the market information-based measure. For instance, the 24-month cumulative output loss implied by the Hamilton measure is only 1.7 percent, only a quarter of the 6.8 percent output loss implied by the baseline market information-based measure in Figure 4a. Responses of the CPI and the federal funds rate are also substantially weaker and become statistically insignificant. The output response to an oil-price shock constructed by the symmetric VAR-based measure (i.e., with log real oil price directly entering the VAR), is slightly stronger than what is implied by the Hamilton s (1996) measure, as shown in Figure 5b. In particular, the accumulated output loss in the 24 months after the shock is now 2.9 percent, although still less than half of the 6.8 percent as implied by our baseline market information-based measure. More important, real GDP response is still statistically insignificant during most of the 24-month period after the shock. It is also worth noting that the point estimates and the statistical inference of the output response as reported in Figure 5b are very close to Blanchard and Gali (27) s estimates. They also estimate a similar VAR using the log oil price level as an indicator of the state of the oil market. In particular, in their second subsample period (1984:Q1 to 25:Q4), the real GDP loss accumulates to about 1.6 percent of quarterly GDP over three years (their Figure 6a), or about.13 percent of annual GDP each year on average, almost identical to the reported estimates of 2.9 percent of monthly GDP loss in two years in our Figure 5a, i.e.,.12 percent of annual GDP loss each year on average. This is not surprising, as our sample period (1984:M1 to 27:M1) is very close to theirs, the VARs are constructed in a similar fashion, and the shock size is normalized to a similar magnitude. They also find that in the 196s and the 197s, the U.S. GDP response is much stronger and statistically significant, and they draw the conclusion that oil shocks are having much less of an effect on the U.S. economy since However, our estimation results suggest that even for the past two decades, exogenous oil shocks still have a substantial and significant effect on the U.S. economy, and the associated output loss may be much larger than their estimates. We illustrate this point more clearly in Figures 6a and 6b, by plotting the point estimates 15

17 of the output responses implied by the two VAR-based measures and the baseline market information-based measure, along with the 95 percent confidence intervals of the latter. Apparently, the real GDP response implied by market information-based measure is significantly stronger than that implied by the traditional VAR-based measures, as the point estimates of the latter stay outside the 95 percent confidence intervals for several months, in particular the response from the Hamilton s (1996) measure. The responses of the CPI are more similar, which is not surprising as part of the increase in oil prices is reflected in the oil component of the CPI by index construction. Why are the output responses implied by the VAR-based measures so different from the responses implied by market information-based measures? One possible explanation is that the VAR identification strategies fail to separate exogenous oil-price shocks from endogenous oilprice movements responding to other kinds of structural shocks. For instance, a productivity shock may lead to an economic expansion and higher oil prices, and consequently a positive correlation between real GDP growth and oil-price movement. Therefore, although the pure exogenous oil shock would lead to a substantial output decline, a VAR estimation may mix up these two kinds of shocks and generate a much weaker and insignificant output response. Our narrative approach provides an opportunity to directly examine this conjecture. In particular, if the above guess is correct, then we should be able to obtain a much less negative, or even positive output response, from an oil-price increase that is induced by gains in productivity or other endogenous oil-price movement. To this end, we construct another series of oil-price shocks, by combining the shock series of categories 18 and 19 (that is, events related to changes in oil demand due to economic development, improvement in oil usage efficiency, technology, etc). We substitute this series into the four-variate VAR equations (3.1), and plot the impulse responses in Figures 7a and 7b, with the shock sizes calculated using the log-price change or predicting error methods, respectively. Impulse responses in both figures indicate that the narrative approach has correctly identified this kind of shock. After an oil-demand shock as described above, the real GDP increases, and the CPI level declines despite the fact that the oil-price component of the CPI rises. The federal funds rate moves little as the output expansion is partly offset by the declines in the consumer price level. This is exactly the pattern one would expect from a positive productivity shock. The insignificance of the responses may simply reflect the lack of enough observations of such shocks 16

18 in our sample period. Such a results also confirms Kilian s (27) finding that an aggregate demand expansion would lead to an increase in the U.S. real GDP growth (and like us, he also finds that the responses are not significant at the 5 percent level) Other Kinds of Oil-Price Shocks Based on Market Information Next, we examine macroeconomic implications of other kinds of oil-price shocks, as identified by our narrative methodology. These shocks are an indispensable part of the world oil-market movement in the past two and a half decades, and are also mixed up by the traditional VAR identification strategies with the exogenous oil shocks. Thus they deserve a separate examination. OPEC and Non-OPEC Oil-Price Shocks Another example of endogenous oil-price movement, which is often discussed by the literature, is the OPEC and non-opec oil exporters decisions to change their oil production plan, which many argue may well be endogenous responses to world economic development. Such events are included in our event categories 13 and 14 (Table 1). Again, we construct a shock series and examine its impact on the U.S. economy. However, we have not found strong evidence suggesting that such events are purely related to gains in productivity. As displayed in Figures 8a and 8b, in responding to such a shock, real GDP declines and the CPI rises. This pattern is different from a typical productivity shock as displayed in Figures 7a and 7b. However, one can still see that the output loss here is much weaker than the output loss induced by an exogenous oil-price shock (as in Figures 4a-4f), and the output response becomes statistically insignificant. 6 Again, this is consistent with literature s finding that oil prices no longer Granger-cause real GDP and other macroeconomic variables since 1973, when the OPEC actions began to significantly affect the world oil market (Hooker 1996). On the other hand, consistent with the weak output declines, monetary policy becomes tighter in order to fighttheriseintheconsumer price level. Oil Market-Specific Demand Shocks Event categories reflect demand shocks that are specific to the oil market, i.e., changes in oil demand that are not related to general economic growth as examined above. For instance, 6 Note that the OPEC development is the single largest event type in our sample, with an observation sample of 742 trading days. Therefore, this statistical insignificance is not likely from a lack of observations. 17

19 changes in the U.S. Strategic Petroleum Reserve (SPR) in the past two decades have always been associated with substantial oil price changes. 7 Variations in the commercial oil and gas inventories, including both actual changes and market expectations of changes, have also affected the oil market significantly. These oil market-specific demand changes often reflect changes in the precautionary oil demand, as discussed in Kilian (27). An oil market-specific demand shock drives up the real oil price by definition, yet the price increase is not as persistent as the exogenous oil-price shocks (Figures 9a and 9b), and the real oil price goes back to its pre-shock level in about five months. In response to the shock, real GDP declines, although the decline is much less persistent and much weaker than in responding to an exogenous oil shock: Real GDP goes back to the pre-shock level in about six months, and the 24-month cumulative output loss is 2.2 percent of monthly GDP, compared with a 6.8 percent output decline implied by the exogenous oil shock. Output decline is statistically significant at the 5 percent level for the first three months after the shock. The CPI increases on the shock but goes back to its pre-shock level in four months. Because of the short lives of output and CPI responses, the accommodation of monetary policy is quite limited. The estimated output decline is close to Kilian s (27) estimates of real GDP response to an oil market-specific demandshock using world oil-production data and a uni-variate regression model, yet his estimate of the CPI response is more persistent than ours, showing an increasing statistical significance throughout a three-year horizon. Oil-Price Shocks Related to Military Actions The last kinds of shocks that we examine are the shocks related to military actions. Because there was only one military conflict outside the Middle East that was found to have significantly affected the oil market over our sample period (the U.S. invasion of Panama in 1989), we decide to examine the shocks related to all military actions. Not surprisingly, military actions tend to drive up the real oil prices and lead to a substantial output loss (Figures 1a and 1b). In particular, real GDP gradually declines, with the maximal response arriving 11 months after the shock. The decline is statistically significant at the 5 percent level for most horizons, and 7 For example, on January 23, 27, the U.S. Department of Energy announced that the U.S. would double the size of its Strategic Petroleum Reserves (SPR) over the next two decades, and the crude oil price responded by a 7.6 percent rally on the same day; on September 13, 2, after the White House mentioned it was considering a release of SPR in response to oil price increases, the oil price dipped 1.5 percent on that day and dropped another 3 percent three weeks later on October 5, 2, the day the SPR release was officially announced. 18