This item is only available as the following downloads:
1 THE EFFECTS OF OIL SHOCKS ON THE US ECONOMY BY DAVID JALIMAN A THESIS Submitted to the Division of Social Sciences New College of Florida in partial fulfillment of the requirements for the degree Bachelor of Arts in Economics Under the sponsorship o f Dr. Tarron Khemraj Sarasota, Florida May, 2010
2 Abstract This thesis sets out to examine the impact of oil shocks and oil volatility on the US business cycle. Through the use of both a GARCH (1,1) combined with an AR (2) model and a VAR model. The res ults from these models indicate that oil shocks cause a small, but statistically significant impact on GDP. Oil shocks can adversely affect U.S. economic growth. Additionally, the VAR model also showed oil shocks causing a slight increase in both the unemp loyment level and the consumer price index. While the models demonstrate the effects oil shocks can have on the economy, they are perhaps too small to have a sizeable impact on the US business cycle.
3 Acknowledgements Perched on my balcony I look out onto a quiet and peaceful campus. I know that these will be the last few lines I will write here at New College. While there are many individuals that I must thank, first I have to acknowledge the institution that gave me the freedom to explore a nd the framework to find my own way. I want to thank my advisor and Sponsor, professor Tarron Khemraj for his advice, support and patience in this daunting endeavor. This process would not have been possible without all the amazing teachers I've had th e privilege of learning from. Especially my committee members Prof. Catherine Elliott and Prof. Zhang Jing who have pushed me beyond my comfort level and profoundly shaped my experience. My parents have made all of this possible. I don't even what to thin k about where I would be without their support. Thank you. When I needed them my friends have been there to; advise me, sustain me, distract me and excite me, you can't be replaced. Don't worry, we'll meet again on the journey. There is one person in pa rticular that has been there in my times of need. I owe you a million.
4 Table of Contents Abstract 2 Acknowledgments .. 3 Table of Contents ... 4 Introduction .. 5 Chap ter 1: Literature Review ... 9 Chapter 2: The US Macroeconomy: An Overview 28 Chapter 3: Methodology .. 35 Chapter 4: Econometric Model .. 41 Chapter 6: Conclusion 47 Bibliography 51
5 Introduction Oil is vital to the US economy. Oil provides 37 percent of US energy usage, 71 percent of which is used for transportation (US Department of Energy 2008). Oil is also a significant input into chemicals production, pharmaceut icals manufacturing, electricity and heating. Oil is the largest global industry with sales of $3 trillion dollars followed by food with sales of $1.7 trillion. (Kammen, 2005) Although sales of food excludes a significant portion of agricultural in the th ird world, which is produced for personal consumption and not recorded as part of the cash economy. In the early 1920s the US replaced Russia as the world's largest oil producer. By 1924 US production exceeded 2 million barrels a day. For the first hal f of the 20 th century the US was the dominant global oil producer as well as consumer. The US continues to be the leading consumer, although the rate of growth of Chinese oil consumption far outstrips the US. Currently, China consumes approximately 8 mill ion barrels per day with an annual growth rate of 16 percent as measured by the first quarter of 2010 over first quarter 2009. The US is no longer the leading oil producer and its thirst for oil far exceeds domestic production. The US first began import ing oil in 1948 and US oil production continued to increase for the next two and a half decades with production in excess of 9 million barrels a day from 1968 to 1974. After 1974 US production declined, by 2008 production had fallen to approximately 5 mill ion barrels a day. The US imports approximately
6 11 million barrels of oil a day, which is 57 percent of total domestic oil consumption according to the US government statistics. In January 2010 the US trade deficit was $37.3 billion, with $27.5 billion or 73 percent of the US trade deficit coming from imported oil.( US Department of Energy 2008) When the price of oil is high, it leads to a significant wealth transfer from the US to other oil producing nations. In 2008 Americans driving across country acute ly experienced the oil price increases. Each stop at the pump cost more than the previous time. Americans were scrambling to buy hybrids and other fuel efficient vehicles, some of which were selling at a premium to their sticker price. Many dealerships we re frantically trying to unload the gas guzzlers, with large cash back rebates. In the summer of 2008 the long held axiom of the relatively inelastic demand for gasoline was fully tested. Americans finally ran into a price increase, which they could no lon ger afford to ignore. The year 2008 was a unique time for oil prices. From January to July of 2008, the price of crude oil shot from $92 a barrel to $147, then in December plummeting to less than $40(Khan2009). Had the fundamentals changed so drasticall y in those six months to account for the price swings? What was responsible for run up in oil prices? To some economists the price of oil no longer appeared to be driven by real factors of supply and demand. The amount of oil traded by Wall Street specula tors far exceeded the trade between oil
7 producers and consumers. How would this increased volatility impact the US business cycle? What is the historical relationship of oil price volatility to the US macro economy in the post World War II period? The ec onomic impact of changes in oil prices on the US business cycles is a matter of dispute by economists. Hamilton has built a reputation as an economist advancing the notion that oil price shocks are a strong negative force on the US economy. In opposition t o Hamilton's assertions about oil shocks, Hooker has analyzed the data from 1948 to 1994 period and used regression analysis to show that there is no longer an important impact from oil price increases. There is continuing debate within the economics pro fession on the actual impact of oil prices on the US economy. The objective of this thesis is to contribute to the economics literature the results of some measures of statistical analysis on the impact of oil price volatility on the US economy in the last decades. The research I conducted uses different econometric techniques than those used by Hamilton and Hooker to look for correlations between oil price volatility and US business cycles. The results of this topic of could have potentially broad implica tions for US economic and foreign policy makers. In this thesis I will examine the impact of oil shocks and oil volatility on the US business cycle. In order to examine this debated question I synthesized the available literature written by the experts in this field in Chapter One. The
8 literature I examined was published in highly respected journals and utilized econometric analysis to examine oil shocks and their effect on the US GDP. In Chapter Two, I outline the methodology of my model, used to trace out the impact of oil shocks. In Chapter Three the model is used to analyze the data on US GDP, industrial production, consumer price index, unemployment level, and crude oil prices. This analysis was performed with Eviews software. The conclusion examine s the implications of the results and possible explanations for recent trends of increased volatility in crude oil prices.
9 Chapter 1: Literature Review Introduction Over the last thirty years economists have studied the effects of oil prices on world ec onomies. Their findings are valuable for policy makers, governments, businesses and investors. There is a range of theories and supporting research to explain the relationship between oil prices and the business cycle, which will be measured through GDP gr owth, unemployment, industrial production and inflation. This chapter will examine the different theories put forth by economists and the results of the research they conducted. Evidence of Link between Oil Price Shocks and US Recessions Hamilton's (1983 ) contribution to understanding the causes of the underperforming US economy in the eight years following the OPEC embargo was his analysis of the correlation of oil shocks to economic performance. Hamilton began examining the effects of oil shocks to the economy for his doctoral thesis. He continued with this line of research in subsequent years establishing himself as a key economic expert in the field of oil prices and their impact on the macroeconomy. Hamilton's work is a major background source for th is thesis. Lagging Economic Performance follows OPEC Embargo
10 James D. Hamilton begins his article: Oil and the Macroeconomy since World War II, with a discussion of the dramatically poor US economic performance from 1973 to 1981. This eight year period be gan with the OPEC oil embargo, which led to the pronounced oil price shock of 1973. Prices increased 300 percent between 1973 and1974. This period was characterized as a time of economic malaise with a significantly lower growth rate, higher inflation and higher unemployment than previous decades. The average unemployment rate of 6.7 percent during the period of 1973 to 1981 was higher than the unemployment rate in any year in the prior period of 1948 to 1972 except for 1958, which had an unemployment rate of 6.8 percent. (Hamilton, 1983 p.228) Comparison of Post Embargo Economy with Baseline Hamilton uses as his baseline for comparison the prior period of 1960 to 1972. Growth decreased from 4 percent to 2.4 percent. Inflation more than doubled from 3.1 p ercent to 7.6 percent. The unemployment rate from 1948 to 1972 does not exceed the 6.7 percent average rate in any year except during the recession of 1958. The underperformance of the US economy concerned policy makers. This was an area of research for ec onomists seeking to develop theoretical explanations from examining the poor economic performance in the decade following the OPEC oil embargo. Oil Shocks precede recessions
11 Hamilton asserts that oil price shocks are often followed by a recessions in the US. James D. Hamilton demonstrates that there is Granger causality between the increase in oil prices during the period 1948 to 1972 and a decrease in GDP several quarters later. (Hamilton, 1983) The relationship between oil price spikes and the decrease in GDP is not a general law of economics, but rather dependent on specific structural factors that were operating during this period. Hamilton demonstrates a "cyclical tendency manifest over the last decade for oil price increases to be followed by recess ions has in fact characterized every recession in the United States since World War II with the single exception of the recession of 1960 61." (Hamilton, 1983, p. 229) Hamilton provides corresponding historical analysis, which attributes these oil spikes t o political events largely originating in the Middle East. (Hamilton, 1983, p. 229) Decline in GDP not accounted by Endogenous Variables The recessions Hamilton studied were not caused by endogenous economic factors, but rather by the exogenous shock of s pikes in crude oil prices. Hamilton did bivariate Granger causality tests using Sims's six variable approximation of the macroeconomy and found that the changes to GDP were not caused by endogenous economic factors. The six factors were two output variabl es: real GNP and unemployment, three price variables: price deflator for non farm business income, hourly compensation and import prices and M1 to represent the money supply. The output data shows: "a lack of evidence of any unusual
12 behavior in real outpu t which could have been used effectively to predict oil price changes over the 1948 72 period." (Hamilton, 1983, p. 234) Nor do the six macroeconomic factors taken as a group shows statistically significant predictive power. Correlation of Oil Price Incre ase to Declining GDP Hamilton uses data from1949 to 1972 excluding data after 1973 because it becomes nonstationarity from the arrangements and concerted action of OPEC nations. Hamilton's autoregression of oil price changes as a source of prediction of c hanges in GDP show that an increase in oil prices was followed by slower economic output after 3 to 4 quarters. The hypothesis that oil price changes do not predict GDP growth can be rejected at the .001 levels based on a four lag regression and also with an eight lag regression. Furthermore, the hypothesis that oil price changes do not predict future unemployment can be rejected at the .01 levels. Hamilton thus established a statistically significant correlation between changes in oil prices and decreasing GDP and rising unemployment. Hamilton presents a detailed description of the events during the oil price increases to lay the foundation for his claims that there was causality as well as correlation. (Hamilton, 1983 p. 234)
13 Doubt Cast on Post 1973 D ata on Oil Price to GDP Correlation Established by Hamilton Data after 1973 that was analyzed by Mark A. Hooker in his 1996 article: What happened to the oil price macroeconomy relationship? This article questions the oil price to GDP relationship discover ed by Hamilton. Hooker finds that the relationship loses its statistical validity after 1973 and that there is a fundamental decoupling of the relationship after 1981. Reconfirming of Hamilton's Earlier Results
14 Hooker's analysis of the sample data from 1948 to 1973 reconfirms the relationship that Hamilton discovered. He finds that the price changes of oil Granger cause both declining GDP and rising unemployment. Mark A. Hooker agrees that the embargo of 1973, which caused major increases in the price o f crude oil, is a reason for the recession that followed. His analysis also shows that the price increase of 1979 played a partial role in the recession of 1980 1982. However, he states that in response to the inflation caused by the oil price increases, the Federal Reserve implemented contractionary monetary policy that further hindered GDP growth Correlation Not Shown in Later Data Data values from 1973 to 1994 do not show the oil price to GDP relationship. According to Hooker's research for the full d ata sample (1948 1994), the price changes in oil fails to Granger cause any macro variable. Hooker makes a convincing argument: as crude oil prices have experienced more extreme swings, their impact on the US economy has decreased significantly. He backs up his view by testing the Granger causality of oil prices on the unemployment rate, real GDP, aggregate employment and industrial production. Hooker uses data from 1973 to 1995 and finds that oil prices do not Granger cause any of the macroeconomic indica tors. Hooker views the relationship between oil prices and the US economy as statistically insignificant over the full period that he examined. His data analysis shows that oil prices do not affect GDP growth after 1973.
15 Impulse Response Function Howeve r Hooker's data does show that there is a response in GDP to an oil price shock as reported in figure 1. (Hooker, 1996, p. 201) For a second sample of data from 1973 to 1994 Hooker conducts an impulse response function of GDP growth to an oil price shock. He uses the log level of oil prices (also known as the rate of change). His findings show that two periods after the initial shock resulted in the highest change in the GDP growth rate of about .8 percent. After four periods, the GDP growth rate was only a ffected .15 percent. The change in GDP growth rate due to oil price shocks returns to zero after nine periods. Mis measurement Not Responsible for Change in Oil to GDP Correlation Hooker proposes three hypotheses that are based on data mis measurement to explain why the oil price macroeconomic relationship has changed. One potential hypothesis he identifies is "that there were structural breaks in many US macro series around 1973, and that failing to account for this weakens the measure role of oil." (Hoo ker, 1996 p.196) Another hypothesis focuses on the endogeneity of crude oil prices and the U.S economy. Prior to 1973, oil prices were an exogenous factor of GDP but switched to an endogenous one after that era, possibly weakening the results of the Grange r causality test. The third hypothesis is that important asymmetries between increases and decreases in oil prices disrupted the relationship of oil prices to GDP. Hooker concludes that none of these hypotheses, which are each based on an explanation of mi smeasurement,
16 are valid. He concludes that the relationship of oil price to macroeconomy was fundamentally disrupted after 1981 and that mere data transformations are insufficient to resolve the breakdown in this relationship. As Hooker states, the point of "Granger causality tests are essentially for whether the unforcastable components of oil prices help predict macroeconomic indicators; the more information about future oil prices now contained in other variables the smaller is the unforcastable compon ent of oil prices. "(Hooker, 1996, p.202) Hooker elaborates on how it is difficult to determine the relationship between the economy and oil prices using an unrestricted VAR. He points out that the data show the absence of current correlations of oil pric es with the macroeconomy except that GDP growth and the increase in T bill rates: significantly add to prediction of oil prices at the log=levels" in the early sub sample 1948 to 1973 (Hooker 1996, p. 202). In the second sub sample, using data from 1973 1994, this macro relationship with oil prices is no longer statistically valid. Hamilton Responds to Hooker James D. Hamilton in his article: This is what happened to the oil price macroeconomy relationship, (Hamilton, 1996) confirms Hooker's conclusion that oil price data from 1986 to 1994 does not exhibit the linear relationship between oil prices and the macroeconomy as it did in the period from 1948 to 1973. However he asserts that since the oil price macroeconomy relationship is
17 complex, with numerou s variables interacting in a multifaceted way, the effectiveness of a simple linear or non linear model to capture the underlying economic dynamics is questionable. Hamilton continues to believe strongly in the negative impact of oil shocks on GDP even if it is more difficult to create statistically valid evidence. In his 2003 response to the Hooker article, Hamilton proposes a different methodology to measure oil price increases. Essentially, Hamilton adjusts the data so that only unambiguous increases ar e measured. Increases that occur after a price decrease and are corrections to make up previous declines are omitted. Hooker compares oil prices on a quarterly basis. Hamilton's approach is more complex and measures only bona fide increases. Hamilton views many of the oil price increases in the period that Hooker examined, as merely corrections of past price decreases. Hamilton suggests comparing price increases with the previous year's price rather than with the price in the previous quarter as a means of adjusting out what he considers to be price corrections rather than price increases. Hamilton's method compares the current price of oil with the previous four quarters. If a data point for a quarterly oil price is lower than the prices for the previous four quarters, that point is defined with a value of zero. If the data point is higher, then Hamilton computes the percentage price change from the previous year's highest price. Correcting the data in this manner Hamilton finds
18 that there is a statistical correlation for the full set of data analyzed by Hooker. The t statistic for the first subset was 3.0 and 2.2 for the second subset, which show statistical significance even for the later data set. However for the second data set the correlation is not statistically significant based on the test for stability of the coefficients. Hamilton hypothesizes that the reason for this result is the small sample size rather than the underlying data. Hamilton's hypothesis of the transmission mechanism is that sp ikes in crude prices dampen economic activity by decreasing demand for key consumption and postponing corporate investment in CAPEX spending in which energy costs represent a significant factor in the likely profitability of an investment. When there is a spike in oil prices, it can be harder to get profits higher than the hurdle rate. One might also hypothesize that increased oil costs could lead to lower profits in industries such as plastics or nylon production, in which higher costs could not be pass ed on to consumers. This would lead to lower stock prices, fewer and lower bonuses for employees which would decrease wealth and lower consumption according to Keynes' consumption function. (C = c0 + c1Yd) Lower profitability could also increase the cost of equity capital because firms would have to issue more stock to secure a comparable amount of cash. This could lead to a further dampening of the economic climate. These relationships suggest mechanisms for how higher oil prices transmit forces that decr ease the US GDP.
19 Hamilton's Research Post 2000 Hamilton concludes from his research post 2000, that the relationship of oil price changes to GDP is non linear. He continues to accept the importance of Hooker's research in the current literature, which s hows a weaker correlation in the oil price change to changes in GDP with post 1985 data. Hooker and Hamilton represent metaphorical opposite poles in the continuing economic research of this topic. In Hamilton's article: What is an oil shock? (Hamilton, 2 003), Hamilton states that a number of researchers have demonstrated "a clear negative correlation between energy prices and aggregate measures of output or employment" (Hamilton, 2003, p. 363). He cites the work of more than a dozen other economists in a ddition to his own research. Muellbauer and Nunziata predicted the US recession of 2001 from a multivariate analysis in which oil prices are given a large weight. Hamilton also cites microeconomic research, which demonstrates similar economic impacts from oil prices, on individual firms, industries or on labor. Importance of Form and of Function Hamilton attributes the weak correlation of oil prices to GDP growth from Hooker's research on the wrong specification of the form of the function. He cites appro ximately ten economists whose research comes to this same conclusion of
20 attributing Hooker's results to the form of the function he uses. One factor that this line of research mentions is that there have been changes in the process of generating oil prices through the creation of OPEC and the financialization of oil through speculation in the oil futures markets. According to this argument, the changes in the means of generating prices create the appearance of an unstable relationship even though there is a n underlying stable non linear relationship. There is a difficulty in determining which is the correct nonlinear specification to use. In an attempt to isolate the exogenous components of oil price changes from other functional factors, Hamilton examines five periods in which there was a pronounced curtailment of oil supply from conflicts in the Middle East. From a historical standpoint, it is hard to refute that military conflicts in the Middle East lead to disruption in production and consequently the s upply of oil. These disruptions, along with speculators expectations, can lead to sharp rises in the price of crude oil. An example of this is the Iraq invasion of Kuwait in 1990. The crude oil price jumped from $15 a barrel in the third quarter of 1990 to $27.9 in the fourth quarter. This oil price shock, as well as the one preceding the Iran Iraq war in the 80's, was followed by an economic recession in the US. One cannot say that these price shocks Granger caused the recessions, but it is hard to deny th at they didn't negatively impact the economic climate in which the recessions took place.
21 The result of Hamilton's studies confirms a nonlinear correlation that shows that these limits to the US oil supply cause poor economic performance. Hamilton argues that there is not only a correlation but also causality at work. He finds that the dynamic multipliers in the nonlinear relation are similar to linear relation with instrumental variables. Hamilton writes: "I conclude that the basic fact being summarized by the nonlinear relation analysis is the historical tendency of the U.S. economy to perform poorly in the wake of these historical conflicts." (Hamilton, 2003, p. 363) Hamilton's recent research reported in this article, shows that the relationship is as ymmetric in that increases in oil prices have, within certain parameters, a negative impact on GDP, and decreases have a statistically less significant impact on increasing GDP. Increases that measure recoveries from prior price declines have less of an i mpact then increases from a period of stable prices. When regressions are done with a data set of 50 values measur ing !Y t (the change in GDP) and !O t (the change in oil prices), only in cases where !O t is > than 0 using a ordinary least squares (OLS) regression the result is that both standard errors and the t statistic of 2. 66 provide statistically significant res ult. When positive and negative values for !O t are used, the size of the sample doubles but the result shows that oil prices have only the imputed effect and this effect is no longer statistically significant. The data before 1980 had few oil prices decr eases but after 1980 oil prices experienced increased fluctuations both up and down.
22 The early models linking Ot with Yt were based on a demand side effect of oil price increases. Under these assumptions, positive values for Ot would increase the price l evel and, given the Keynesian assumption of rigidity in the price of wages, this would lead to lower real wages and consumption, resulting in lower GDP. The price of oil impacts the price of gasoline and rises in the price of gasoline are shown to have a d irect impact on the size of automobiles that are purchased, increasing demand for high mileage small cars and decreasing demand for large, low mileage cars. Further there is significant research which shows that perceived uncertainty on future oil prices i n the wake of a price shock leads to the deferment of many big ticket purchases from automobiles, to appliances, to housing, to industrial equipment. If allocative disturbances are a major mechanism for impacting GDP, Hamilton argues that there is no reaso n to expect that there would be a symmetric linear relationship. (Hamilton, 2003, p. 365 366) Asymmetrical Relationship K.A. Mork, in his article, Oil and the macroeconomy when prices go up and down: An extension of Hamilton's results discovered that the relationship between the change in oil prices and the change in GDP is asymmetric. Mork's research confirms Hamilton's results, which show that there is a correlation between increasing oil prices and a decrease in GDP. However his research also
23 shows tha t when oil prices decrease there is not a corresponding increase in GDP. Mork's research complements Hamilton by adding data through 1988, including data points when oil prices declined. (Mork, p. 740 744) The disruptive affect of volatility J.P. Federe r, in his article entitled Oil price volatility and the macroeconomy: A solution to the asymmetry puzzle, developed the theory to explain the asymmetry based on the hypothesis that it is the volatility of oil prices, rather than price increases, which decr eases GDP. "This occurs through a disruption of business investment as well as curtailed spending on large ticket consumer items". (Federer, 1992, page 12). Federer in his 1993 article titled does uncertainty affect investment spending? Examines a wider ra nge of events that create uncertainty in the marketplace such as "major political events, changes in monetary and fiscal policy, oil price shocks, and financial crises raise uncertainty about cash flows or interest rates, we should expect that these events will depress economic activity by reducing the incentive to make irreversible investment expenditures." ( Federer, 1993, page 21) Before reading the article, Oil Price Shocks and the US economy: where does the asymmetry originate? By Nathan S. Balke, Ste phen P.A. Brown, Mine K.
24 Ycel, One might expect that since higher crude oil prices result in lower consumption and investment spending, then lower crude oil prices should lead to increased consumption and investment. However, an examination of the data b y Balke, Brown and Ycel shows that this is not the case. The relationship where it does exist is asymmetric. The authors believe that oil price volatility creates a level of uncertainty for firms such that even decreasing oil prices do not help stimulate economic activity. Most individuals are not responsive enough to quarterly changes in crude prices for those changes to have an impact on spending decisions due to the inelastic nature of gasoline. The authors' findings are that volatility in oil prices i s more statistically correlated with low GDP growth than steadily increasing crude prices. From linear to nonlinear In the article, Oil price shocks and real GDP growth: empirical evidence for some OECD countries by Rebecca Jimenez Rodriguez and Marcelo S anchez the authors use both linear and nonlinear vector auto regressive models to determine if oil price shocks have an effect on GDP growth of some OECD nations. They found that from the postwar period to the early 1970's, there existed a linear relations hip between oil prices and GDP. After 1980, only nonlinear models comparing oil prices to GDP growth exhibited statistically significant results. One method the authors use to examine the validity of their nonlinear models was to inspect the confidence ban ds of the impulse response functions.
25 The paths of transmission The authors built upon the research regarding the transmission of oil prices to the economy, done by Rotenberg and Woodford (1996), Kim and Loungani (1992) and Hamilton (1984), (1996). The s upply side affect occurred when firms lowered production due to the increased price of oil, a key input in many production processes. A positive oil price shock would also affect demand for investment spending on energy dependent goods. The authors also po int out that changes in oil prices can also be transmitted to the economy indirectly through inflation and exchange rates. Pre research expectations Prior to conducting their study, the researchers expected that price increases would bolster growth in cou ntries that were net oil exporters and hurt those nations that were net importers. However this commonsense reasoning did not hold true for all nations. Japan, a net oil importing country, does not experience a significant negative effect on GDP when oil p rices rise. The UK is another unique example of an oil exporting nation where increases in oil prices have a statistically significant negative effect on GDP and decreases in oil price have a statistically significant positive effect on GDP. The UK is incl uded in the early years of the time series as a net importer and later reclassified as a net exporter. This study does not differentiate between nations that export large or small quantities of oil relative to GDP.
26 Commodity prices in the long run The aut hors of the article How Persistent are Shocks to World Commodity Prices? By Paul Cashin, Hong Liang, C. John McDermott, examine time series data and compare two theories that explain commodity prices. The first theory hypothesizes that in the long run comm odity prices should increase due to the limited supply of natural resources from which they are extracted. Expounding a different view, Prebisch Singer wrote a paper in 1950 arguing that commodity prices would fall relative to manufactured goods due to a rapid increase in the supply of natural resources. Cashin, Liang and McDermott point out that primary commodities account for 25 percent of world merchandise trade and are a major source of export earnings for many developing nations and account for 50 pe rcent of the export earnings of these nations. They argue that "commodity markets play a nontrivial role in transmitting business cycle disturbances and in affecting inflation rates in industrial nations," citing The Macroeconomic Determinants of Commodity Prices." (Cashin, p. 178) When Cashin, Liang and McDermott measured commodity prices from 1957 to 1998 they found that neither theory held true for this period. In contrast, they found that commodity prices follow cyclical trends, alternately increasing and decreasing. Their research then focused on analyzing periods of pronounced increases in commodity prices and the duration of these commodity price shocks.
27 The persistence of commodity price shocks The authors use the median unbiased estimator to d efine an exact point and interval estimate for the autoregressive parameters in the commodity price data and calculate the length of time for the effects of commodity price shocks to dissipate. By their estimation, their econometric approach was superior t o least square unit root regressions, which are biased and have low power. The interval estimate provides useful information "as to whether a failure to reject the null hypothesis of a unit root is due to the null being true or due to the uncertainty of th e estimate of the autoregressive parameter." (p. 181) With their approach, they found that "on average shocks to commodity prices are very long lasting. For the majority of individual commodities it typically takes more than five years for half of the eff ect of the initial shock to dissipate."(p.182) This finding points to the high costs of commodity price stabilization schemes such as the Imp's CFF, Compensatory Financing Facility, and CCFF, Compensatory and Contingency Financing Facility, that were meant to smooth the effects of exogenous shocks to commodity prices when they fell below its medium term trend. In their analysis, the authors used an impulse response function, the cumulative response function, and the half life of a unit shock, to quantify th e persistence of shocks.
28 Chapter Two: The US Macroeconomy: An Overview This chapter provides information to accompany the Charts at the end of the chapter, on key measures for the US economy in the modern period, roughly extending from the Great Depress ion in 1929 to the present. The background information and data is on industrial production, oil price, unemployment, and inflation and GDP growth. Industrial production From the beginning of World War II to the end of the war industrial production double d in just those few years, from a measure of 10 percent to 20 percent of the 2002 baseline for industrial production. From 1960 to 1970 industrial production was in a trend of steady growth, rising from about 28 percent to 40 percent of the 2002 baseline. The data shows a lag between a recession and its effect on industrial production. Industrial production tends to reach a high point at the beginning of a recession and reaches its low point as the US economy moves out of the recession and begins to grow Changes in the level of industrial production, is an indicator of the US economic health and plays a significant role in the business cycle and has an impact on spot crude oil prices.
29 The chart (figure 4) measuring industrial production was created wit h data from the Board of Governors of the Federal Reserve System. The baseline, which is set at 100, is established with finished goods production in 2002. The data begins in the late 1930s and extends to 2010. Oil prices From 1950 to 1972 was the era of inexpensive and reliable oil supply. Prices were low and exhibited little price variation. After the 1973 OPEC oil embargo, prices increase by 125 percent. This price shock had serious negative effects on the US economy. Some economists blame it for prec ipitating the 1973 recession. During this recession, the data shows a decrease in the level of industrial production and GDP growth, coupled with an increase in the unemployment rate and inflation. The 1980s ushered in a new era for oil prices, this era w as fraught with a heightened level of price volatility, which is clear from figure five. In 1980 a barrel of the benchmark West Texas intermediate crude sold for $40. In 1986 the price fell to less than $20 a barrel. From 1999 to 2008, crude prices climbe d to historic highs of over $130 a barrel from just $15 a barrel nine years earlier. Crude prices then crashed to under $40 a barrel in 2009. Unemployment
30 Figure three measures the percent that are unemployed and receiving unemployment insurance. This pre sents a very conservative picture of the unemployment level, but has value as a business cycle indicator. From 1970 to 1974 there is downward pressure on the unemployment rate moving from just over 5 percent to a low of 2 percent. From 1974 to 1976 followi ng the mammoth oil shock that accompanied the OPEC boycott, the unemployment rate shoots up from 2 percent to just fewer than 8 percent. The highest levels of unemployment and lowest levels of industrial production tend to occur as the US economy is pullin g out of a recessionary period. GDP Growth In the past 80 years the US economy has experienced remarkable economic growth. In 1930, real gross domestic product was about $1.5 trillion rising almost 1000 percent to just under $14 trillion 80 years later. (Figure one) Inflation Figure two depicts the Consumer Price Index for all urban consumers from 1950 to 2010. The chart establishes the baseline of 100 as the average of the three years from 1982 to1984. In 1950 prices were 25 percent of the baseline, b y 1970 prices were 40 percent of the baseline. Starting in the 1970's the US economy experienced significant levels of inflation.
31 Charts Figure 1: Gross Domestic Product Figure 2: Consumer Price Index
32 The BLS (Bureau of Labor Statistics) sets the average index level, representing the average price level, for the 36 month period covering the years 1982, 1983, and 1984 as equal to 100. The BLS then measures changes in relation to that baseline. An index of 110, for example, indicates a 10 percent inc rease in prices since the baseline period; while an index of 90 represents a 10 percent decrease from the baseline. Figure 3: Civilian Unemployment
33 Figure 4: Industrial Production Industrial production of final goods with production in 2002 set as a baseline =100 Figure 5: Oil Prices 1940 2010
34 Figure 6: Oil Prices Quarterly spot Prices 1974 2009
35 Chapter Three: Methodology GARCH Estimates of Conditional Volatility The generalized autoregressive conditional het eroskedastcity (GARCH) model allows economists to use past values of conditional variance to forecast future levels of conditional volatility. This method is useful in determining the risk level for certain asset classes. For example an investment with hig h levels of conditional volatility would require higher rates of return to reflect the level of risk as it becomes increasingly difficult to predict future values. Therefore the conditional volatility can be a useful measure of both risk and required rate of return. Before the GARCH method existed, economists utilized similar models known as the ARCH model (autoregressive conditional hateroskedastcity) or the ARMA model (autoregressive moving average). The two earlier models laid the foundation upon which the GARCH model is built. In contrast to a "high order ARCH models...[a] GARCH (1,1) [has] less parameters to estimate and therefore lose[s] fewer degrees of freedom."(Asteriou 261) Consequently the GARCH is preferred to estimate volatility. The main difference between the earlier ARCH model and the GARCH approach is that ARCH utilizes decreasing linear lag values to measure past values of conditional variance. The GARCH is distinguished by incorporating not only the
36 square of the lagged residual but a lso the past values of conditional variance. The lagged squared residuals represent the past value of shocks, while the lagged values of conditional variance incorporate past variance. The GARCH method has a more flexible lag structure that enables it to be better at adaptive learning, therefore it can estimate high levels of variance during which it may be difficult to create accurate forecasts. GARCH (1,1) A simple GARCH (1,1) model is defined as: t 2 = 0 + 1 u t # 1 2 + 2 t # 1 2 Where t 2 is the conditional variance or volatility of crude oil, u t 1 2 is the squared error term from the previous time period, and t 1 2 is the square of the conditional variance from the previous period. This t hesis first used a GARCH (1,1) to measure the volatility of crude oil prices. This measure of conditional variance is then used in a standard OLS model to determine the effect of oil volatility on percent change of GDP growth, unemployment and inflation. A s a major driving force of the economy, oil prices may be expected to effect macroeconomic indicators, but the aim of this regression is to examine the effect of high volatility on these indicators. In order
37 to remove autoregressive errors the dependent va riables are lagged twice using an AR(2) model. The equations are as follows: CP I = 1 + 1 CP I t # 1 + 2 CP I t # 2 + 3 $ O I L 2 + G D P = 1 + 1 G D P t # 1 + 2 G D P t # 2 + 3 $ O I L 2 + U ne m p = 1 + 1 U ne m p t # 1 + 2 U ne m p t # 2 + 3 $ O I L 2 + (Bollerslev, 1986) (Gujarati, 2003,p.862) Stationarity and the Dickey Fuller Test By definiti on a set of stationery time series data will revert back to its mean value in the long run. If the data set has long run trends and is therefore not reverting back to its mean value, the data set is deemed nonstationary. Nonstationary data cannot be used i n econometric regressions. Much of the time series data used in this thesis was originally nonstationary. For example, the data on crude oil prices, GDP growth, and CPI when graphed followed an upward trend, which indicated the data was nonstationary. The data was then transformed into stationary data by taking the first difference. The data was then tested using the augmented Dickey Fuller test to see if there were unit roots as instructed by the literature to test for stationarity. The Dickey Fulle r test has three levels of confidence intervals; the 99 percent and 95 percent and the 90 percent. Unemployment, GDP growth, and crude oil
38 prices were found to not have a unit root at 99 percent confidence level. The consumer price index was the only data set, that the null hypothesis of having a unit root could only be rejected at the 90 percent confidence level. It is important to transform the series into stationary data in order to correctly estimate the VAR and not receive spurious results The unit root tests: Y t Y t 1 = Y t 1 Y t 1 + t Or Y t = Y t 1 + t A VAR Model Since there exists no theory dictating the way the variables in this model should be ordered, this thesis chose to use an unstructured VAR instead of the Choleski factorization method. The Choleski factorization method r equires theory to order the variables where as an unrestricted VAR does not weight or value certain variables more than the others. This is one reason why econometricians view VARs as such powerful, yet atheoretical tools. A vector autoregressive model (V AR) is used to estimate the equations below. The VAR was first introduced by Christopher Sims (1980) as a way to fix the
39 problem of previous simultaneous equations estimations that weighted some variables more than others and required that some variables t o be considered endogenous while other exogenous. The VAR allows for a simultaneous estimation without the distinctions between which variables should be deemed exogenous. Calculating the Number of Lags In order to obtain the correct number of lags the m odel was run with several lags and the lowest Akakike Information Criterion (AIC) was chosen. The AIC is utilized to obtain the optimum number of lags, because it imposes a penalty for the addition of further regressors. The AIC is calculated as: 2 l n l n 1 3 5 1 3 5 k R S S A I C = + # $ # $ % & % & Where k is the number of lag terms and 135 is the number of observations in the model, and RSS is the sum of the squared residuals. For three lags the AIC is 1.055, for four lags the AIC is 1.017, for five lags the AIC is 1.047. Therefore, the AIC is minimized at four lags; therefore the model incorporates four lags, such that k is equal to four. VAR: the number of lags=k=4 = the Stochastic error term
40 O i l t = 1 + j C P I t # j j = 1 4 $ + % j U N E M t # j + j = 1 4 $ & j O I L t # j + j G D P t # j + 1 t j = 1 4 $ j = 1 4 $ G D P t = 2 + j C P I t # j j = 1 4 $ + % j U N E M t # j + j = 1 4 $ & j O I L t # j + j G D P t # j + 2 t j = 1 4 $ j = 1 4 $ C P I t = 3 + j C P I t # j j = 1 4 $ + % j U N E M t # j + j = 1 4 $ & j O I L t # j + j G D P t # j + 3 t j = 1 4 $ j = 1 4 $ U N E M t = 4 + j C P I t # j j = 1 4 $ + % j U N E M t # j + j = 1 4 $ & j O I L t # j + j G D P t # j + 4 t j = 1 4 $ j = 1 4 $ From (Gujarati, 2003,p.849) The Impulse Response Function After selecting the opt imal number of lags for the VAR model, the next step is estimating the results. The most commonly used method to estimate the results of a VAR model is through the graph created by the impulse response function (IRP). This impulse response function depict s changes in the dependent variables when shocks are introduced to the model. An Impulse response function traces out the changes in a dynamic model over a given time frame. The shocks or error terms have an impact on the variables in the current time pe riod as well as future time periods. The time elapsed from the beginning of the shock to when the effect of the shock has dissipated is a useful measure for policymakers and analysts.
41 Chapter Four: Empirical analysis and results As discussed in the meth odology chapter, this thesis used two methods to analyze the interaction between shocks to the price of oil and the US business cycle. The variables that served as indicators for the condition of the macro economy in the analysis are US GDP growth, the lev el of industrial production, unemployment, and the consumer price index. One method to calculate the relationship between oil price shocks and the macroeconomic variables was an impulse response function generated from a VAR model. The other method was a n autoregressive distributed lag model, which incorporates the volatility of oil as calculated by a GARCH model. The VAR Results: The results from the impulse response function demonstrate a negative response from oil price increases to GDP growth. The i mpulse response function (Figure 7) simulates an oil shock and provides the range of responses for one standard deviation. Time is measured quarterly in all of the impulse response functions. From the first to the second period after the initial oil shoc k, GDP growth declines from .0 percent to .1 percent. By the third quarter the effects of the shock have almost returned to zero. The shock impacts GDP growth again in the third and fourth quarter this time falling slightly lower than .1 percent. Effects of the shock
42 begin to diminish after the fifth period and return to the original equilibrium level after the seventh period. Oil shocks have a very small positive effect on the level of unemployment. After a shock is introduced the unemployment level gr adually increases to just over .005 percent from the first period to the fifth period. It takes four more periods for the effects of the shock to fully dissipate. Although this is a very small change in the unemployment rate it does confirm that changes i n oil prices volatility does dampen the economy through different channels of transmission including unemployment. This slight change is statistically significant, however the magnitude of the change is itself not of much significance. There is also a po sitive response in the consumer price index to an oil shock. The data is labeled as CPI2 in the figure below and it serves as a proxy for inflation and represents the rate of change for consumer price data excluding energy. After the initial shock there i s a .01 percent effect on CPI2, which then slowly trends downward from period 2 to 3. But then continues its ascent until it reaches its maximum level of .015 percent at period four. Clearly oil shocks have positive impact on US inflation. The consumer price index returns to the equilibrium level in period six. There is a unique relationship between the level of industrial production and shocks to the price of oil in which the impact on industrial production changes
43 direction several times. When the s hock is introduced to the impulse response function there is a slight negative effect from period's one and two. Then the curve begins to increase past equilibrium level reaching a maximum positve level of .005 percent in the fourth quarterly time period. Industrial production then begins a gradual descent crossing the equilibrium level after time period seven. Again in this case the impact of oil price changes on industrial production while statistically significant produces a small magnitude of change tha t does not significantly impact the economy. AR (2) Model The other method used to analyze the data was an autoregressive model with each variable lagged twice, combined with the conditional variance of oil labeled GARCH01. The GARCH01, which is the condi tional variance of oil and serves as a proxy for oil volatility. After running the auto regressive model, certain interactions between the variables became clear. As expected all four variables were statistically impacted by the first lagged values in th e previous time period. The GARCH01 1 has a statistically significant negative impact on both GDP ( 1) with one lag and GDP ( 2) with two lags at a 99 percent confidence level. Granted the coefficient for the GARCH01 is a mere 0.000123. The sign on t he coefficient is negative as expected; oil volatility does negatively impact US GDP. But the magnitude is so 1 The GARCH01 is the series I created to represent oil volatility.
44 small that it does not provide evidence of a meaningful causal relationship between oil price shocks and the US GDP. I expected that oil volatil ity would dampen a firm's desires to increase their labor force and the analysis of the data provides extremely slight results to bolster my hypothesis. The GARCH01 has a statistically significant positive impact on the unemployment level. The coefficien t of the GARCH01 is only .00000497 percent with significance at a 99 percent confidence level. The magnitude of the impacted is quite small, that one could argue that the results might be more aptly interpreted as demonstrating no consequential effect. How ever the drivers that change unemployment levels are multifaceted and tend to have a delayed impact on the current rate of unemployment. The GARCH01 does not have a statistically significant impact on industrial production or the consumer price index. However the data used in this analysis is quarterly and it may take longer to see the effect on industrial production and CPI depending on the inventory level that individual firms maintain.
45 Table 2 Oil Volatility and Unemployment Oil Volatitily and GD P Variable Coefficient Std. Error Variable Coefficient Std. Error C** 0.034 0.015766 C*** 0.824466 0.162468 UNEMP( 1)*** 0.084 0.009326 GDP( 1)*** 0.371168 0.082187 UNEMP( 2)*** 0.090 0.009543 GDP( 2)** 0.18999 0.081759 GARCH01* 4.720E 06 2.03E 06 GARCH01*** 0.000132 4.06E 05 R squared 0.4521 R squared 0.3538 Adjusted R squared 0.4400 Adjusted R squared 0.3396 Prob(F statistic) 0.0000 Prob(F statistic) 0.0000 Oil Volatility and Industial Pr oduction Oil Volatility and CPI Variable Coefficient Std. Error Variable Coefficient Std. Error C 0.038704 0.037698 C** 0.053159 0.022919 IP( 1)*** 0.934063 0.084748 CPI2( 1)*** 0.578016 0.082287 IP( 2)** 0.164369 0.08472 CPI2( 2)*** 0.265527 0.081778 GARCH01 1.03E 05 1.84E 05 GARCH01 4.41E 06 6.59E 06 R squared 0.6536 R squared 0.676973 Adjusted R squared 0.6460 Adjusted R squared 0.669848 Prob(F statistic) 0.0000 Prob(F statistic) 0
46 Figure 7
47 Conc lusion In this thesis I set out to examine the impact of oil shocks and oil volatility on the US business cycle. Through the use of a GARCH (1,1) I calculated the conditional variance of oil which served as a proxy for oil volatility. With these results I then utilized an AR (2) model to see the interactions between oil volatility and the variables chosen to represent indicators for the US business cycle. Oil volatility as represented by the GARCH term did have a statistically significant negative impa ct on both lagged values of GDP growth. The unemployment level was positively impacted by the GARCH term in the AR (2), indicating that oil volatility increases the level of unemployment. When looking at the levels of industrial production and the consum er price index (CPI) the GARCH term was not found to have a statistically significant impact. The results from the VAR model showed that an oil shock resulted in a slight negative response to GDP growth. The model also showed a slight positive effect o n the unemployment level as well as the consumer price index. An oil shock initially elicited a negative response in industrial production but later shifts to having a positive impact. The analysis in this thesis indicates that oil shocks have a smal l, but statistically significant impact on the economy. These effects are perhaps too small to have
48 a sizeable impact on the business cycle, demonstrating there is not a statistical correlation between oil price shocks and the onset of recessions. Hami lton presents economic data centered on significant oil prices shocks, and explains the historical events that contributed to the price increase. He uses the resulting recessions in the US to illustrate the powerful negative effect oil shocks have on the b usiness cycle. Hamilton admits that his anecdotal evidence is not statistically significant but warns that the next oil price shock will result once again in a recession. Although the results from the model do not validate the linkage between oil shocks a nd recessions, it appears that many economists and policy makers accept this correlation. The famous economist Nouriel Roubini, who is widely credited with predicting in detail the unfolding of the 2008 financial crisis, has written an article warning that an Israeli attack on the Iran nuclear facilities could lead to a sharp spike in oil prices and another global recession. The Sub prime mortgage backed securities and the housing bubble have been assigned most of the blame for the recent "Great Recession however as Roubini points out the sizable increase in crude oil speculation and the increased volatility aided in destabilizing the global economy. During this time world economic growth and Chinese economic growth in particular, played a role increasing demand for oil. Speculative purchases on futures markets in what has been called paper barrels of oil increased from 4.5 times world oil demand in 2002 to 15 times world oil demand in 2009. It is beyond the scope of this thesis to
49 determine if oil specula tion is increasing the size of the shocks however this would be an interesting topic for future research.
50 Bibliography Asteriou, Dimitrios, and S. G. Hall. Applied Econometrics: a Modern Approach Using EViews and Microsoft. Basingstoke [Englan d: Palgrave Macmillan, 2007. Print. Awerbuch, S. , The Surprising Role of Risk and Discount Rates in Utility Integrated Resource Planning, The Electricity Journal Vol. 6, No. 3, (April), 20 33 --------, Do Consumers Discount the Future Corre ctly? A Market based Valuation of Fuel Switching, (with William Deehan), Energy Policy Vol. 23, No. 1 (January), 57 69 Balke, N.S., Brown, S.P.A., Ycel, M.K., , Oil Price Shocks and the U.S. Economy: Where Does the Asymmetry Originate?, The Energy Journal 23, 3, 27 52 Bollerslev, Tim. "Generalized Autoregressive Conditional Heteroskedasticity." Journal of Econometrics (1986). Web. Chaudhuri, K. , Long Run Prices of Primary Commodities and Oil Prices, Working Papers, The University of Sydney. Ciner, C. , Energy shocks and Financial Markets: Nonlinear Linkages, Studies in Nonlinear Dynamics and Econometrics October, 5 (3), 203 212 Energy Information Administration, Annual Energy Review 2008 Ferderer, J. P. , Oil Price Volatility an d the Macroeconomy, Journal of Macroeconomics 18 1, Winter, 1 26
51 Gisser, M.; Goodwin, T.H. , Crude oil and the macroeconomy: Tests of some popular notions. J. Money Credit Banking 18 1, pp. 95 103 Gujarati, Damodar N. Basic Econometrics Boston: McG raw Hill, 2003. Print. Hamilton, J.D. , This is what happened to the oil price macroeconomy relationship, Journal of Monetary Economics 38, 215 220 --------, What is an oil shock?, NBER Working Paper 7755 -------- Causes and Consequenc es of an Oil Shock of 2007 2008. Department of Economics, UC San Diego. Web. Hooker, M.A. , What happened to the oil price macroeconomy relationship?, Journal of Monetary Economics 38, 195 213 Hung, R.D.; Masulis, R.W.; Stoll, H.R. , Energy sho cks and financial markets, Journal of Futures Markets, 16, 1 27 Jones, C.M.; Kaul, G. , Oil and the Stock Markets, The Journal of Finance Vol LI, No. 2 Kammen, D., "An Energy Policy for the 21 st Century," Policy Matters (Fall 2005). Lee, K.; Ni, Shw an; Ratti, R.A. , Oil Shocks and the Macroeconomy: The Role of Price Variability, Energy Journal Vol. 16, 39 56 Loungani, P. , Oil price shocks and the dispersion hypothesis. Rev. Econ. Stat 68 3, 536 539. Mork, K.A., Olsen, O. and Mysen, H.T [ 1994] Macroeconomic responses to oil price increases and decreases in seven OECD countries. Energy Journal 15 4, 19 35.
52 Mork, K. , Oil and the Macroeconomy, When Prices Go Up and Down: An Extension of Hamilton's Results, Journal of Political Eco nomy, vol. 97, No. 51 Papapetrou, E. , Oil price shocks, stock market, economic activity and employment in Greece, Energy Economics Vol. 23 (5) September, 511 532. Rotemberg, J.J.; Woodford, M. , Imperfect Competition and the Effects of Energy Price Increases on Economic Activity, Journal of Money, Credit, and Banking Volume 28, Issue 4, Part 1 Nov., 550 577 Roubini, Nouriel. "A Global Tsunami By." Forbes.com 22 Apr. 2010. Web. Sadorsky, P. , Oil price shocks and stock market activity, E nergy Economics 2, 449 469 Sims, Chris. "STRUCTURAL VS. BEHAVIORAL MODELS." Time Series Econometrics (2002). Web. USA. EIA. Annual Energy Review (AER) 26 June 2009. Web. 20 Apr. 2010.