The Importance of Economic
Variables in Crude Oil Price
Movements
Thesis MA Economics
University of Amsterdam
International Economics & Globalisation
Author: Michiel Nivard (5941776)
Supervisor: Maja Micevska Scharf
Second reader: Dirk Veestraeten
Abstract
A quantitative analysis of the relation between crude oil price volatility and three economic variables (i.e. GDP level, industry output and inflation) shows that only the crude oil price and US GDP level have a significant relationship in both directions. Further analysis indicates that the GDP level is a stronger predictor of crude oil price than vice versa. Additionally, the used VAR method predicted a rebound of the crude oil price in the coming year with a high of $91/b in the summer of 2016 notwithstanding any external shocks.
2
Verklaring eigen werk
Hierbij verklaar ik, Michiel Nivard, dat ik deze scriptie zelf geschreven heb en dat ik de volledige verantwoordelijkheid op me neem voor de inhoud ervan.
Ik bevestig dat de tekst en het werk dat in deze scriptie gepresenteerd wordt origineel is en dat ik geen gebruik heb gemaakt van andere bronnen dan die welke in de tekst en in de referenties worden genoemd.
De Faculteit Economie en Bedrijfskunde is alleen verantwoordelijk voor de begeleiding tot het inleveren van de scriptie, niet voor de inhoud.
3
Table of Content
1. Introduction ... 4
2. Crude Oil Price Dynamics ... 7
2.1. Crude Market... 7
2.2. Supply... 8
2.3. Demand ... 10
2.4. Technology ... 11
2.5 Conclusion ... 11
3. Literature Review: Oil Price & Macro Economy ... 13
4. Empirical Model... 16 4.1. Introduction... 16 4.2. Theory/Methodology... 16 4.3. Quantitative Analysis ... 18 4.3.1. Data ... 18 4.3.2. Methodology ... 19 4.3.3. Analysis... 24 5. Conclusion ... 35 6. References... 37
4
1. Introduction
Between 2010 and 2015 there has been a high spread within the price of crude oil.1 Peaks of $111,72/b were contrasted by lows of $44,12/b between April 26th 2011 and January 29th 2015.2 In the graph below (graph 1.1), a gradual increase in the price of crude oil from the 1970’s to 2015 can be identified. When adjusting for inflation (blue line) it is clear that in the mid-1970’s there was a peak in oil prices which repeated itself in the mid-2000’s. Or to be more precise: a double-spike can be seen in the more recent years where one peaked in 2007 and the other between 2011-2014 with a large drop in 2009.
Graph 1.2 shows both the percentage of petroleum use (blue line) and the absolute petroleum use (red line) of the US industry from 1949 to 20153. It shows that from 1949 to 1979 there has been a steep increase of the use of petroleum in the total US energy mix. Between 1979 and 2014 the percentage of petroleum in the total energy mix has been fluctuating between 37 and 45 percent. In the transport sector the percentage is even larger
1 Crude oil is the raw form of extracted oil and comes in several different forms (e.g. light, heavy, sour,
sweet). It can be used for many different purposes (e.g. petrol, diesel, gasoline, kerosene, lubricant, plastics). This thesis will use the generic term crude oil in order to prevent any unnecessary technical terminology.
2 Price in dollar per barrel. Data from EIA: http://www.eia.gov/dnav/pet/pet_pri_spt_s1_d.htm 3 Datasource: EIA
5 where petroleum accounted for 93 percent of total energy use.4 For the entire US economy, the total crude consumption in 2014 was 13.97 million barrels per day.5
Combining these two developments it is clear that crude is an important factor in the US economy. Hence the price fluctuation that is shown in graph 1.1, will have a significant impact on the overall state of the US economy. As petroleum is an important input in the US industrial production, any fluctuation in the price will affect its average costs and thus output. Similarly, the transportation sector (both commercial and private), which is heavily reliant on petroleum, will be sensitive to any crude oil price volatility. This relationship theoretically also runs the opposite way as a growing economy will demand more energy to fuel its activities
and hence translate into a higher crude demand. Simple economics tells us that the increased demand will reflect in the price of oil, meaning that there is a double relationship between crude oil and economic factors.
These developments and statistics create validity for a renewed attention on the – potential – relation between the oil price and the overall economy. Is it possible to identify a relationship between the recent oil price spikes and several economic factors? Is it possible to distillate a pattern for certain economic variables vis-à-vis a fluctuation in the price of oil and/or vice versa? This thesis will aim to investigate these questions and attempt to quantitatively analyse the effect of a shock in oil prices on selected economic variables. Some research on the potential economic effect of oil price volatility exists, but unfortunately these are somewhat outdated (see, for example, Darby (1982) on inflation, Bernanke (1997) on monetary policy and Yousefi (2004) on exchange rates). This thesis will hence aim to add more recent analysis to already existing literature.
4 Idem.
6 The main line of research in this thesis will be to investigate the oil price fluctuation and its effect on the US economy. The research question to be answered is: does the oil price volatility have a significant effect on US macro-economic factors? This question will be addressed with a VAR-model using data sets from the EIA (Energy Information Administration), economic data from US department of Labor Statistics, and US Federal Reserve. Economic variables that are investigated are the US industry yearly output, trade weighted US dollar value and the US GDP level. The VAR model will be used to analyse the potential relationship over time between the oil price and the selected overall economic indicators. This method is chosen, as it is best suited to address the time-effect, because the variables will interact over time.
When analysing the data and considering graph 1.1 and graph 1.2, the hypothesis of this thesis will be that there indeed will emerge a significant relationship between the crude oil price volatility and the overall economic factors in the US. The rationale is that indeed the exposure of the US economy to crude oil is large and hence the exposure to any price fluctuation will be significant. This exposure will have its effect on several economic factors.
The thesis is divided into 6 sections. After the introduction, the 2nd section will give a qualitative analysis on the dynamics of the oil price, as it aims to provide an insight in the drivers that affect the fluctuations of the price of oil. In the 3rd section an overview of existing theories regarding the relation of the oil price and the macro economy will be discussed. The aim is to establish an insight in the current theoretical standing on the impact of a volatile oil price, which will serve as a starting point for the quantitative research in section 4. The quantitative research will involve a data analysis of several economic variables and their relation with the volatility in oil prices using a VAR model. Section 5 will provide a conclusion and the 6th section will present the bibliography.
7
2. Crude Oil Price Dynamics
In this section a qualitative analysis regarding the crude oil price dynamics will be given, which will focus on specific drivers. First of all, the size of the oil market and the geo-political influence create specific characteristics that are unseen in other markets. For this reason the oil market itself will be discussed in section 2.1. Like any commodity, the supply and demand characteristics are important drivers for the crude oil price. These specific factors are the second and third driver, which will be discussed in section 2.2 (supply) and section 2.3 (demand). Lastly, innovation plays a large role in the production of crude oil. Section 2.4 will therefore describe the role of technology on the price dynamics of crude oil.
Before proceeding with the qualitative analysis of the drivers of the crude oil price it is important to acknowledge that there is a treat of causality running both ways. Demand fluctuations can trigger prices but the opposite is also plausible. Technology can impact prices but simultaneously the price can induce new technology. Therefore, it is important to acknowledge that the drivers are in fact dynamic and the causality might not be one directional.
2.1. Crude Market
As has been shown in the first section, the oil price has ranged from $44 to $112 within a few years. This $70 range is remarkable for a commodity that is not subject to trends or marketing. So what are the main reasons that crude oil is so volatile? What are the main market characteristics?
First, the specific characteristics of crude are important. The first important characteristic is that once crude oil has been pumped out of the ground it is very easily transported and/or stored.6 Crude is a non-perishable, liquid commodity and thus highly suitable for trading all over the world. The global infrastructure has been developed over the years and current US storage capacity amounts to 500 mb (with 60% utilization rate)7 and the global oil tanker fleet capacity in 2014 amounted to 782 Mdwt (or 112 mb)8.
This tradability potential gives traders the opportunity to employ price strategies. Hedging, speculating, storing, etc. are common phenomena in the crude oil market that all have an effect on the price (in)stability. This has led to the creation of trading hubs in New York (NYMEX) and London (ICE) where futures and options in the two main crude benchmarks are traded (i.e. WTI and Brent). In the recent decade, with the emergence of the Chinese economic power, the crude oil consumption has shifted to the east and
6 Crude oil is moveable either by pipe, boat, truck or rail. According to the API there is over 190.000
miles of oil pipelines in the US alone that transport crude oil 24 hours a day. The total international oil trade consists of 50 million barrels per day and mainly use oil tanker vessels for transportation.
7 Source: EIA (March 2015)
8 subsequently trade flows have redirected. This might lead to the establishment of trading hubs in the east and potentially the creation of an eastern crude oil benchmark.9
When assessing the figures of the crude oil trade, it is important to realize the total size of the global crude oil market. Average global production amounted to 88.7 mb/d in 2014.10 Or, phrased differently, the total crude oil market in 2014 was worth $2.77 trillion (i.e. at an average price of $85.6/b). Both size and liquidity have given the oil market increased opportunities for traders, users and producers resulting in the biggest commodity market in the world.
Besides the tradability of crude the wide variety of usages is of importance when assessing the crude market. Petroleum products that are derived from crude can be classified in three main categories: fuel, feedstock and other products. Its biggest component is fuel with examples like: gasoline, kerosene and diesel accounting for roughly 80% of petroleum products. Industrial feedstock (e.g. naphtha, coke, propane) take up roughly 15% and other products (e.g. lubricants, waxes, asphalt) account for the rest.11
The last important characteristic of the oil market that influences the oil price is the availability of proven reserves. Every oil company has a portfolio of basins/plays from which oil can be produced. However, these oil reserves will run out as the field gets depleted. For this reason, oil companies need to increase their portfolio of proven reserves in order to guarantee their future production. This urge has led to continuous exploration in order to maintain the possibility of future production (ergo consumption). If the proven reserves were to decease, today’s oil price will be affected.
Last potential reason for volatility is the derived trading market for oil commodities. Both at the London Stock Exchange (ICE) and the New York Mercantile Index (NYMEX), oil contracts are traded. The link with the physical oil market is strong and any trading fluctuation in the derivative market has an effect on the price of a barrel of oil. Carollo (2012) identifies this trading relationship as the main raison for oil price volatility in his book ‘Understanding Oil Prices’. In that sense, he reverses the causal relation as he argues that the virtual trading (which is derived from the actual production) is affecting the oil price rather than the opposite where the actual market forces affect the trading and thus the crude oil price.
2.2. Supply
One of the issues affecting the oil price is supply side driven. Simple economics tells us that as supply does not meet demand, the price will change. Whenever producers alter their production or external factors influence their capacities the price of oil will be affected. The security of supply is thus a major driver in the oil price.
9 Weber argues that there is a potential for an ‘East of Suez’ crude benchmark combined with a
trading hub in Oman/Dubai as a result of the eastward shift of oil markets.
10 Source: BP Statistical Review 2015 11 Source: EIA
9 Although cartel formation is illegal in most economic sectors, one of the most commonly known cartels is the Organisation of Petroleum Exporting Countries (OPEC). Total production of OPEC countries consists of 40% of world capacity and total OPEC exports account for 60% of internationally traded crude.12 These numbers give the OPEC influence on the international market. The leading OPEC member Saudi Arabia can use the organisation to influence the crude price levels by regulating the total OPEC production.13 The reduction of competition in the international oil market that the OPEC cartel has realized, explains some of the oil price volatility. In graph 2.1 (Source: EIA) both number 8 and 12 indicate a cut in OPEC production levels resulting in a sharp price increase.
These instances can be classified as OPEC´s effort to secure a minimum oil price. In other words, whenever the oil price decreases below a certain threshold the OPEC will alter its production in order to adjust the price to its preferences. This is named the ‘call on OPEC´ and has been a persistent market inefficiency throughout (most of) history.
When discussing the security of supply, an important feature that has a big impact is the occurrence of a war. In the recent years there have been several episodes of military conflict in oil producing countries that had a (in)direct effect on the oil price. Graph 2.1 shows the Iranian revolution (3), Iran/Iraq war (4), the Iraqi invasion of Kuwait (6) and the 9/11 attacks (9). In all instances these military conflict resulting from geopolitical disturbance led to increased oil prices as fear arose for the vulnerability of crude oil supply streams.
12 Figures of OPEC production and exports can be found on the EIA website and OPEC website. 13 Traditional Saudi strategy was to maintain a guaranteed (minimum) crude oil price via production, or
‘the call on OPEC’. When crude prices started to decline, expectations were that Saudi Arabia controlled OPEC would issue a decreasing production. However, in November 2014, OPEC decided to maintain current production and thus effectively dropping the price.
10 A last major geopolitical event that has affected the supply of crude oil is the Arab oil embargo on the west during the 1970s. Number 2 indicates this event in graph 2.1. This embargo led to major oil shortages and caused a major price increase of oil.
In more recent times the supply side in oil markets is affected by both the increased production by US shale firms and the pending status of a nuclear arm deal with Iran. The increased production (and thus supply) by US shale producers has led to lower oil prices. Iran on the other hand has been a stable crude oil producer/exporter but was banned from international market by sanctions imposed by the UN in response to Iran’s alleged nuclear arms program. As soon as sanctions are lifted, Iran is able to re-enter the crude oil market and will add additional supply to the market, potentially lowering the price.
2.3. Demand
The other important factor is the demand side of oil. High or low demand for oil can have major impact on the price. In graph 2.1 both 7 and 11 indicate a falling oil price due to international financial crises that led to a declining world demand for oil and subsequently to lower crude oil price. In the mid-1990s the Asian financial crisis led to a decrease of oil price and in the 2009 the global financial crisis caused the oil price to plummet. Obviously, increasing economic growth will have the reversed effect as demand grows and oil price increases.
The same effect can be identified in more recent years. The ever expanding Chinese economy in the first decade of the 21st century caused higher demand for energy. This increased demand has been the main driver for increasing oil prices during the first 10 years of this century. Equally, the recovery (albeit slow) of the US economy in the aftermath of the financial crisis has contributed to the increasing global demand and subsequently increasing oil prices.
Nevertheless, demand for oil can also decrease due to various reasons. After years of Chinese economic growth, China is now experiencing lower growth than it expected. Also, the Chinese stock market has been witnessing great losses which indicate a possible economic bubble. Fears of further negative Chinese economic news, combined with slow economic recovery in Europe, have been the main driver of the steep decline in the oil price in the first half of 2015.
Of course, the fact that China is a large country and that any (unforeseen) change in its outlook will affect oil demand is important. However, also crises in smaller countries with less exposure to the global demand can have a potentially large effect. For instance, the Greek crisis in Europe can cause widespread economic downturn even though the Greek economy is relatively small.
In general, increases (decreases) in economic growth will increase (decrease) the demand for crude and thus increase (decrease) the price of oil. If uncertainty of future economic consequences is in play, this will affect the oil price fluctuation. In other words, expectations of the (future) state of the economy – both demand and supply – play a large part in establishing oil prices. This connection also shows the relationship between overall GDP levels and the oil price: whenever GDP increases it will increase demand for crude oil
11 and henceforward increase the crude oil price. As such, it provides a solid theoretical base for investigating the relation between crude oil prices and the average GDP level in section 4.
2.4. Technology
One important distinction in oil producing industry is within the realm of the accessibility of oil reserves. Conventional oil reserves are more easily and cost effective to produce whereas the unconventional oil reserves are more difficult and expensive to extract.14 With rising crude prices it becomes profitable to devise methods for exploration and producing these unconventional oil reserves. This has led to innovation that is responsible for adding new known oil reserves and increased worldwide production. At the same time, the technological innovations are affecting the crude oil price in the sense that production becomes more cost effective.
A recent example is the fracking technology and horizontal drilling that has been used to extract the tight shale oil reserves in North America. This innovative use of extraction methods has led to a potential increase of the world oil reserves with 335 billion barrels.15 This is the equivalent of more than 10% of traditional world reserves. Even though the technology is not new,16 the high crude oil prices made the hydraulic fracturing (fracking) economically viable. The addition of these extra reserves and the production increase can both be considered as drivers for the crude oil price decrease as it has increased supply and secured future crude oil production.
Other major innovations that have impacted the industry are the rotary drill, offshore drilling and possibility to extract crude oil from tar sands. In general, technologic improvements have the capacity to reduce cost, increase efficiency and increase the producible crude oil reserves. Some of these innovative transitions are only possible when the average crude oil price reaches a certain level. Some of the innovations are gradual whereas others have more significant impact. Nevertheless, changes in the production process of crude oil can affect the price mechanism.
2.5 Conclusion
As some characteristics of the oil market have been established, a deeper look into the composition of the oil price can be attempted. The composition of the oil price consists of many factors and is too complex to discuss here. However, some conclusions could be drawn from the previous sections and combined in a possible theoretical framework. This framework might prove to be useful throughout the rest of this thesis and might explain the recent oil price fluctuation. It is arguable that the oil price is relatively constant at approximately the $75 mark. Around this price there is a natural range of approximately $20
14 Examples of unconventional oil reserves are tar sands, shale tight oil or located in (extreme)
deep-sea reserves.
15 According to a combined study of the EIA, DoE and ARI from 2013.
12 within the oil price can fluctuate due to natural market issues (supply and demand). This is indicated by the range of A in graph 2.2
However, outside of this bandwidth there exists another layer of price uncertainty that occurs on top of the regular price volatility. This is another $20 margin on both sides that only affects the real oil price when extreme situations occur. This is shown by area B in graph 2.2. These extraordinary situations can include war, technological innovations or other disruptive measures. This model provides a good explanatory structure of the recent oil price fluctuations. Apparently there has been a natural downward (upward) pressure on oil price combined with extraordinary negative (positive) pressures that created the extreme low (high) oil price.
Although graph 2.2 is not part of the traditional economic literature, it can be of some assistance in analysing the price dynamics of the crude oil. However, as has been shown in the previous sections, there are many factors influencing the crude oil price and therefore it remains difficult to provide a definite explanation. In the next section, an overview of the existing economic literature regarding the relationship of the crude oil price fluctuation and the overall economy is provided. This will help constructing a framework through which the quantitative model in section 4 can be constructed.
13
3. Literature Review: Oil Price & Macro Economy
In the previous section several drivers of the oil price fluctuation have been described. But how, if at all, does this volatility affect the macro economy or its individual components? In this section the existing literature on the relation between the price volatility and the macro economy will be investigated. The overview will provide the mainstream ideas and help in the further analysis of the relationship of the oil price and the macro-economy. It will act as a framework for the quantitative analysis to be performed in section 4.
The crude oil price is a widely researched topic. The size of the market and the deep penetration within all sectors of the economy makes it an important measure of the overall economic conditions. It is possible to distinguish three major treads of research within the literature investigating the oil price volatility and its effects on the economy. The first line of research is the actual empirical effect of the crude oil price fluctuation on several economic variables (e.g. GDP, exchange rate, gold price, stock markets, inflation). A second field of research aims its focus at some specific oil price shocks and their effects on the overall economy. These researches ‘use’ the shocks to find any anomalies in the perceived oil/economy relationship. Lastly, the third thread of investigation is directed at the fundamentals surrounding the relationship between the crude oil price and the economy. What assumptions are made? Is it possible to determine a definite causal relation that is one-directional?
Research that directly investigates the relationship between the crude oil price volatility and the overall economy has been focussing on several economic factors. A first economic variable that has a perceived relationship with the crude oil price is inflation. This relation has been studied by Darby (1982), Bernanke et.al. (1997) and Cologni & Manera (2007). An extensive research on the effect on inflation in the G7 countries has been done by Cologni and Manera (2007). They find that indeed a significant change in the price for crude oil will have an effect on the inflation levels of the specific country. However, the domestic monetary policy following the change in inflation differs per nation.
In an older article, Darby (1982) examines the aftermath of the oil price shock in the early 1970s and its effect on real prices and thus inflation. Although his research was limited in sense of the used data, his conclusion is that indeed the inflation is affected by the level of the crude oil price over time.
Monetary policy has been marked as an important channel in the oil price/inflation relationship by Bernanke et.al. (1997). Their research examined the channel through which the crude oil price fluctuation is affecting the economy. They argue that as the crude oil price fluctuates, the inflation figures will change due to changes in the production costs. In their view, the monetary policy has to counter this effect and thus balance the effect of the oil price impact albeit altering the interest rate. Therefore, the model by Bernanke et.al. has included the monetary policy as an endogenous factor as it is an important intermediary step in the oil price/economy relationship.
Other papers have investigated the relationship between the exchange rate and oil price (Yousefi & Wirjanto 2004 and Zhang et.al. 2008). Interestingly, both these papers look
14 at the effect of the exchange rate on the crude oil price, rather than the other way around. Yousefi concludes that OPEC crude oil price setting is affected by the US dollar exchange rate with individual OPEC members. On the other hand, Zhang et al.investigates the general effect of the US dollar exchange rate on the WTI crude oil price. Their conclusion is that the exchange rate is a significant driver for crude oil price fluctuation but the reverse relation is less significant.
While most scholars have focussed their attention on the US (or developed) market, Bhunia (2013) has investigated the relation between the crude price and the domestic gold price and financial variables in India. Bhunia concludes that indeed the several factors show a cointegration relationship, which means that crude oil price is related to the price of gold, exchange rate and the stock indices in India.
Several papers have investigated the effect of oil price shocks and their effect on the economy. However, several of these have highlighted the problematic assumptions underlying the research. In their 2004 research, Barsky and Kilian have not only included inflation as an economic variable but also investigated the effect of the oil price upon economic growth. Their conclusions are that the perceived relationship is problematic and that the direction of causality can run both ways. This results in flawed assumptions regarding oil price shocks and the presupposition of the effect of oil price shocks on the economy, whereas the opposite could also hold.
This argumentation, that the direction of the causal relationship might be bi-directional, is used by more scholars. Especially Kilian has addressed these issues in both of her articles. Kilian (2008) investigated the traditional theory of the relation between oil price and the economy. She argues that the oil price fluctuation should be regarded as endogenous and to incorporate the economic supply and demand effect into the oil price trend. Also, in her 2009 article, she rejects the idea of a thought experiment where the oil price fluctuates but all other factors are kept equal. In principle, she is making the case for a dynamic system where one change will set of multiple economic reactions and ultimately the causal relation becomes ambiguous.
When assessing the oil price shocks of the 1970’s and early 2000’s, Blanchard and Gali (2007) have researched the impact on these price shocks on the economy of several developed economies. The inflation of these countries was not largely affected and Blanchard and Gali conclude that this has to do with monetary policy, labour market dynamics, oil exposure and lack of other shocks. Indeed this provides another indication that there are more variables at play when assessing the economic effects of oil price shocks.
Lastly, the short polemic between Hamilton and Hooker (1996) regarding the actual economic effect of crude oil price fluctuation provides a useful insight. Hooker argues that the perceived economic effect of oil price volatility from the 1970’s is not significant in quantitative research and thus future research should be aware of the non-existent relation. In a reaction, Hamilton agrees with the overall conclusions of Hooker but does argue that major oil price shocks (like war driven supply shocks) have a significant impact.
When combining all existing literature on the oil price/economic relationship, it becomes clear that the empirics does not provide a clear picture. In other words, there remains some ambiguity surrounding the actual impact of the oil price on the economy. The
15 second takeaway is the fact that not only does the oil price affect the economy, but the oil price is also influenced by the economy itself. In general, it is important to note that due to the dynamic environment the causal relations may be murky, which should be incorporated in the data analysis.
16
4. Empirical Model
4.1. Introduction
Following the qualitative research and the literature review regarding the crude oil market and its effect on the overall economy, this section will focus on the construction of an empirical model that evaluates the anticipated relation between the crude oil price fluctuation and several economic factors. For this purpose several US data sources will be used to fit a VAR model, which has the capacity to use time series data and compare the mutual effect among the selected variables.
Section 4.2 will shortly describe the theory underlying the VAR model and how each variable is related to its own lags and the lags of other variables. Section 4.3 is the actual modelling of the acquired data. This section will discuss the data (section 4.3.1.), the methodology (4.3.2.) and the analysis (4.3.3.). The analysis is subdivided into the Granger-causality test (4.3.3.1.), the Impulse Response Functions (4.3.3.2.) and the forecasts (4.3.3.3.).
4.2. Theory/Methodology
The aim of the thesis is to investigate whether a relationship exists between oil price fluctuations and overall economic indicators like GDP level, Industry Output and Inflation. As has been shown in the previous sections, the oil price is incorporated in the overall economy but the exact relation is ambiguous as many factors may be at play. Also the fact that both oil prices and economic indicators change over time, makes that time should be incorporated as a factor and hence problematic to isolate the desired relationship. Therefore, this thesis will incorporate the timing effect in an effort to incorporate the influence of past periods on the overall trend.
Building on the oil price dynamics (section 2) and the literature review (section 3) above, the theory of this thesis will assume that, given the importance of oil in several economic sectors, the effect of a price fluctuation will be noticeable in fluctuations in these economic indicators. In other words, if the oil price moves, because of its penetration in the overall economy, other economic indicators will show an effect.
In order to compare the effect of the oil price fluctuation on the US economy, a Vector Auto Regression model will be used. This model allows for the analysis of time series data. It compares the effect of a variable on itself but also the effect of other variables. The model hereby incorporates the lagged effect of those variables. This is the general form of a VAR equation:
(1) 𝑌𝑡= 𝜇1+𝛼11𝑌𝑡−1+𝛼12𝑌𝑡−2+…+𝛼1𝑝𝑌𝑡−𝑝+𝛽11𝑋𝑡−1+𝛽12𝑋𝑡−2+…+𝛽1𝑝𝑋𝑡−𝑝+𝜀1𝑡
17 Formulas (1) and (2) indicate the effect of the lags of Y and X on either Y and X, where p are the number of lags that are included in the model, α and β are the respective coefficients, µ are the constants and ɛ are the uncorrelated error terms. In this case the economic variables of interest are not only affected by its own historic movement, but might also be influenced by historic price movements of the crude oil price.
To test whether there is a predictive relation among the variables, it is possible to test for Granger causality. The null-hypothesis is that the 𝛽’s are equal to 0. If the null-hypothesis can be rejected than it can be said that the crude oil price volatility Granger causes the economic variable fluctuation and hence that there is some predictive power in the oil price for the economic variable. In other words, if the coefficients of the historic oil price are not significantly different from 0 than the oil price will not Granger cause the movement of the respective economic variable (and vice versa).
In order to find a path of 𝑌𝑡 or 𝑋𝑡, it is possible to investigate the effect of the error term in the model, which can be interpreted as a shock to the system. This will generate the impulse response functions of the variables. However, it is first important to show that the error terms are not correlated, or that:
(3) 𝑐𝑜𝑣𝜀1𝑡,𝜀2𝑡=0
If equation (3) holds, then the effect of the shock on subsequent variables over time can be estimated. For this purpose the errors have to be orthogonalized. This means that the errors terms are substituted by another term and that the second error term incorporates a part of the first error term. If the errors are not orthogonalized then the errors are correlated and a single error will incorporate the combined shocks to several variables and the effect on separate equation is meaningless. The orthogonalization process is as follows:
(4) 𝜀1𝑡=𝜂1𝑡
(5) 𝜀2𝑡=𝛾𝜀1𝑡+𝜂2𝑡=𝛾𝜂1𝑡+𝜂2𝑡
This shows that indeed the error terms are uncorrelated and that their expectation is zero:
18 After the error terms are orthogonalized, the effect of the shock on separate variables 𝑌𝑡 and 𝑋𝑡 can be measured and the impulse response be measured. For this purpose, let the 𝑌𝑡−1 and 𝑋𝑡−1 be equal to zero and the orthogonalized error term 𝜂1𝑡 equal to 1. From this it is possible to reconstruct the time-path of both 𝑌𝑡 and 𝑋𝑡 because of the shocks, or in other words the impulse response functions can be calculated.
The impulse response functions that are created will provide some insight in the relation of the variables over time. The IRF shows the effect of the shock of the error term on the variables over time. It shows how the two variables are related and what the effect is of their respective behaviour over time including the other variable and its lags. With this information it is possible to analyse the research question and the trend that both variables have.
4.3. Quantitative Analysis
In this section the aim is to identify whether there exists a relation between the oil price fluctuation and several US economic indicators. The three economic indicators that have been used are (1) the US industry Production Index, which reflects the overall output of the US industry and therefor the state of US economy, (2) the Trade Weighted Value of the US dollar that provides an indication of the strength of the US dollar compared to other currencies and (3) the overall US GDP Index. In order to compare these figures, a VAR model will be used to compare the time series data of the oil price and the US economic indicators.
4.3.1. Data
The data that has been used for the oil price is the WTI (Western Texas Intermediate) price per barrel between January 1986 and May 2015. Since the investigation is focussed on the US, the oil price indicator will be the US oil benchmark. The WTI prices are denominated in real US dollars and thus need to be adjusted into current US dollars. Inflation figures over the same period (01/1986 – 05/2015) allows for adjusting the WTI prices into today’s dollars and were obtained from the US Census Bureau. This created the opportunity to convert the WTI prices into comparable US dollars. In all three graphs the adjusted WTI price is indicated by the blue line and the price range is shown on the left vertical axis. The overall pattern in WTI data is the increasing volatility over time although no apparent trend has occurred over the period (1986: $21 2015: $26).
For the US industry output, figures were obtained from the St. Louis Federal Reserve System (FED). These figures represent an index of the total value of the Industrial output from the United States in the period 01/1986 – 05/2015, with 2007=100. The top graph in Graph 4.1 shows the combined WTI and US Industry data. The red line (US Industry Output) indicates an upward trend from roughly 55 to 105. Some negative Industry output shocks (2001, 2008 and 2014) seem to coexist with the negative shocks in oil prices. However, more investigation needs to be done before any final conclusions can be drawn.
The middle graph in Graph 4.1 shows the index of the Trade Weighted US Dollar (in red). If anything, it shows a slight downward trend (120 to 90) but overall the volatility is more notable. In contrast to the (seeming) co-movement of the WTI price and US industry index,
19 the WTI price and US dollar index seem to move in antiphase. Whenever the value of the US dollar decreases (1990, 2002, 2008 and 2014) the WTI price moves in the opposite direction. The US GDP index is shown in the bottom graph in Graph 4.1. This data was obtained from the US Bureau of Economic Analysis but was provided in quarterly increments. In order to align this data with the monthly frequency of the other data, interpolation was performed. Between the quarterly points a straight line was drawn in order to maintain the trend in the existing data. Lastly the US GDP data was only available until March 2015, which was 3 months less compared to the other data sets. However, this will not pose a limitation for my research.
4.3.2. Methodology
In order to perform a VAR analysis, the statistical program Stata 13 will be used. The obtained data has been inserted in the program, converting it to the correct format. Since the data are time series, this has to be programmed so that the program recognizes this characteristic. Stata needs to acknowledge the time series character of the oil price, US industry output, value of the US dollar and US GDP data in order to include the possibility of lagged effects (the command ‘tsset’ is used). After these adjustments the inflation adjusted WTI oil prices are labelled as AdjP2, the Index for US Industry output is inserted as IndPro2, the trade weighted value of the US dollar as Dollar2 and the US GDP as USGDP. These labels will be seen in the Stata output below.
As has been explained in section 4.2., the VAR model is used to identify a relation between previous lags and a specific variable of choice. By combining two time series data of separate variables it is possible to determine the relation between a change (or shock) on the selected variable over time (so including lags). However, before this analysis can be performed some tests have to be done on the data itself to make sure the VAR analysis can be done and results will be
meaningful and valid. First it is important to determine whether the eigenvalues of the data are smaller than one. If this is the case than the VAR-model is stable. The stability of the VAR-model will also prove helpful when creating the IRF later on. To test whether this is true, Stata allows for a ‘varstable’ test. Graph 4.2 represents the results. It is clear that all the absolute values of the eigenvalues are smaller than one, which indicates that the VAR-model is stable.
20 As shown in equations (1) and (2), the VAR model investigates what the relation is between variables and their respective lags. However, there is no standard number of lags (P) that need to be included in the regression. However, it can make a great difference in the estimation of the model and a proper analysis of the number of included lags is needed. Therefore the second procedure that has to be executed is determining the optimal number of lags. For this purpose the Final Prediction Error (FPE) needs to be identified (for which a ‘varsoc’ test is conducted). Below, table 4.1 gives an overview of the output considering a different number of lags. All the available data (WTI Oil prices, Industrial Production, Dollar value and US GDP Level) were imported and the maximum number of lags was set to 10. In other words, within the range of 10 lags, the ‘varsoc’ test defines an optimal number of lags. varsoc AdjP2 IndPro2 Dollar1 USGDP, maxlag(10)
Selection-order criteria
Sample: 1986m11 – 2015m1 Number of obs = 339
Endogenous: AdjP2 IndPro2 Dollar2 USGDP Exogenous: _cons
Table 4.1
In order to select the optimal number of lags, the ‘information criterion’ (AIC or BIC) is used. These criteria depend on the log-likelihood of the model. In this case the BIC (Bayesian Information Criterion) is used as an determinant for the number of lags. The BIC formula is given below by formula (3), where L is the log-likelihood from the model, p is the number of lags, k is the number of variables and T is the sample size (number of observations):
(7) 𝐵𝐼𝐶= −2𝐿+𝑘+2𝑘𝑝ln𝑇
The Stata output in table 4.1 indicates the smallest BIC with a star. This indicates the model with the best fitted number of lags and prevents over- or under fitting. Performing this test indicates that this criterion corresponds with 2 lags (p = 2). In the further analysis the number of lags used will be 2 as is shown in formulas (8) and (9) below (p = 2):
lag LL LR df p FPE AIC HQIC SBIC 0 -4572.41 6.2e+06 26.9994 27.0174 27.0446 1 -1403.85 6337.1 16 0.000 .052276 8.40029 8.49024 8.62601 2 -1195.9 415.89 16 0.000 .016847 7.26786 7.42977 7.67416* 3 -1169.81 52.19 16 0.000 .015874 7.28831 7.44218 7.79518 4 -1136.79 66.042 16 0.000 .01436 7.10789 7.41372 7.87534 5 -1103.2 67.182* 16 0.000 .012949* 7.0041* 7.3819* 7.95214 6 -1095.73 14.923 16 0.530 .013624 7.05448 7.50423 8.18309 7 -1083.95 23.569 16 0.099 .013975 7.07935 7.60106 8.38854 8 -1074.57 18.765 16 0.281 .014543 7.11839 7.71206 8.60816 9 -1066.98 15.174 16 0.512 .015299 7.16802 7.83366 8.83837 10 -1059.68 14.242 16 0.581 .016142 7.22041 7.958 9.07133
21 (8) 𝑌𝑡= 𝜇1+𝛼11𝑌𝑡−1+𝛼12𝑌𝑡−2+𝛽11𝑋𝑡−1+𝛽12𝑋𝑡−2+𝜀1𝑡
(9) 𝑋𝑡= 𝜇2+𝛼21𝑌𝑡−1+𝛼22𝑌𝑡−2+𝛽21𝑋𝑡−1+𝛽22𝑋𝑡−2+𝜀2𝑡
Thirdly, it is necessary to test whether the variables are stationary. For this purpose a Dickey-Fuller test is performed, for which the results are presented below in tables 4.2, 4.3, 4.4 and 4.5. This test will show whether the differences of the coefficients are related and whether there is a trend among the variable over time. If this is not the case than the variables follow a random walk.
The null-hypothesis for the Dickey-Fuller test is that the variables are non-stationary (or contains a unit-root) and the alternative hypothesis is that the variables are stationary. In both tests, the number of lags will be 2, as has been established according to the results from the FPE test.
The results in table 4.2 show the possible stationarity of the WTI oil price that has been adjusted for inflation. The results indicate that the test statistic lies outside the 5% confidence interval and thus the null-hypothesis can be rejected at this level. Also the p-value of .0935 shows that the variable AdjP2 can be considered as non-stationary at the 5% level. dfuller AdjP2, lags(2)
Augmented Dickey-Fuller test for unit root Number of obs = 350
Interpolated Dickey-Fuller Test
Statistic 1% Critical Value 5% Critical Value 10% Critical Value Z(t) -2.597 -3.452 -2.876 -2.570 MacKinnon approximate p-value for Z(t) = 0.0935
Table 4.2
In table 4.3, the results for testing the stationarity of the US industry production is represented. In this test the general positive trend of US industry output over time (as seen in the top graph in Graph 4.1) has been incorporated in the test result. The test statistic is -1.563 and thus the null-hypothesis is rejected at a 10% confidence level. Also the p-value of .8065 indicates that the output of US industry over time is non-stationary.
22 dfuller IndPro2, lags(2) trend
Augmented Dickey-Fuller test for unit root Number of obs = 350
Interpolated Dickey-Fuller Test
Statistic 1% Critical Value 5% Critical Value 10% Critical Value Z(t) -1.563 -3.986 -3.426 -3.130 MacKinnon approximate p-value for Z(t) = 0.8065
Table 4.3
The other two variables, value of the trade weighted dollar and the US GDP index, are also tested for stationarity via the Dickey-Fuller test. The results are shown in tables 4.4 and 4.5. In both cases the null-hypothesis can be rejected at a 10% confidence interval as the test statistic is -2.566 in case of the value of the US dollar and -1.521 when assessing the US GDP index. Also the p-values (0.1003 and 0.8218) indicate a non-stationarity for both variables. Similar to the US industry production data, the US GDP has a clear positive trend (as can be seen in the bottom graph in graph 4.1), which has been incorporated in the Dickey-Fuller test.
dfuller Dollar2, lags(2)
Augmented Dickey-Fuller test for unit root Number of obs = 350
Interpolated Dickey-Fuller Test
Statistic 1% Critical Value 5% Critical Value 10% Critical Value Z(t) -2.566 -3.452 -2.876 -2.570 MacKinnon approximate p-value for Z(t) = 0.1003
23 dfuller USGDP, lags(2) trend
Augmented Dickey-Fuller test for unit root Number of obs = 346
Interpolated Dickey-Fuller Test
Statistic 1% Critical Value 5% Critical Value 10% Critical Value Z(t) -1.521 -3.986 -3.426 -3.130 MacKinnon approximate p-value for Z(t) = 0.8218
Table 4.5
After these tests have been conducted, there is enough evidence that further analysis can be done and the VAR analysis that will follow will generate valid results. The results are discussed in section 4.3.3.
24 4.3.3. Analysis
Tables 4.6i, 4.6ii and 4.6iii represent the VAR regression results of the variables Oil Price, US Industry output, US dollar value and US GDP level, where 2 lags have been used. It is clear that there are various signs of the coefficients indicating a alternating pattern. However, the significance of the coefficients does differ and a deeper analysis is needed. The following sections will provide this detailed reflection.
var AdjP2 IndPro2 Dollar2 USGDP, lag(1/2)
Vector autoregression
Sample: 1986m3 – 2015m1 No. of obs = 347
Log likelihood = -1219.587 AIC = 7.236812
FPE = .0163317 HQIC = 7.395819
Det(Sigma_ml) = .0132708 SBIC = 7.636165
Equation Parms RMSE R-sq chi2 P>chi2 AdjP2 9 1.84023 0.9771 14833.79 0.0000
IndPro2 9 .477602 0.9990 354945.3 0.0000
Dollar2 9 1.37019 0.9835 20643.74 0.0000
USGDP 9 .108564 1.0000 9822049 0.0000
25 Coef. Std. Err. z P>[z] [95% Conf. Interval] AdjP2 AdjP2 L1. 1.293896 .0515273 25.11 0.000 1.192985 1.394888 L2. -.3550099 .0518787 -6.84 0.000 -.4566982 -.2533296 IndPro2 L1. .0870236 .2097017 0.41 0.678 -.3239841 .4980314 L2. -.1458569 .2099916 0.69 0.487 -.557433 .2657191 Dollar2 L1. -.0767463 .0704003 -1.09 0.276 -.2147283 .0612358 L2. .0618457 .0680731 0.91 0.364 -.0715751 .1952665 USGDP L1. 1.48276 .6757176 2.19 0.028 .158378 2.807143 L2. -1.410148 .6767603 -2.08 0.037 -2.736573 -.0837219 _cons 1.143671 1.629866 0.70 0.483 -2.050807 4.338148 IndPro2 AdjP2 L1. .0183734 .0133731 1.37 0.169 -.0078375 .0445842 L2. -.024914 .0134643 -1.85 0.064 -.0513035 .0014756 IndPro2 L1. .9721724 .0544248 17.86 0.000 .8655017 1.078843 L2. .0132543 .0545001 0.24 0.808 -.0935639 .1200725 Dollar2 L1. .0083228 .0182713 0.46 0.649 -.0274883 .0441339 L2. -.0132462 .0176673 -.75 0.453 -.0478735 .0213811 USGDP L1. 1.335239 .1753721 7.61 0.000 .9915165 1.678962 L2. -1.322748 .1756427 -7.53 0.000 -1.667001 -.9784946 _cons .6479245 .4230064 1.53 0.126 -.1811529 1.477002
Table 4.6ii
26 Coef. Std. Err. z P>[z] [95% Conf. Interval] Dollar2 AdjP2 L1. -.081179 .038366 -2.12 0.034 -.1563749 -.0059831 L2. .079058 .0386276 2.05 0.041 .0033493 .1547666 IndPro2 L1. -.3279895 .1561388 -2.09 0.036 -.6331068 -.0210541 L2. .4077881 .1563546 2.61 0.009 .1013386 .7142375 Dollar2 L1. 1.279239 .0524183 24.40 0.000 1.176501 1.381977 L2. -.3260179 .0506856 -6.43 0.000 -.4253598 -.2266761 USGDP L1. .3163319 .5031229 0.63 0.530 -.6697708 1.302435 L2. -.3918254 .5038992 -0.78 0.437 -1.37945 .5957989 _cons 3.859861 1.213558 3.18 0.001 1.481331 6.238392 USGDP AdjP2 L1. -.0025552 .0030398 -0.84 0.401 -.0085132 .0034028 L2. -.0016856 .0030606 -0.55 0.582 -.0076842 .004313 IndPro2 L1. .0176087 .0123713 1.42 0.155 -.0066386 .0418561 L2. -.0207043 .0123884 -1.67 0.095 -.0449852 .0035766 Dollar2 L1. .0027186 .0041533 0.65 0.513 -.0054217 .0108588 L2. -.0034492 .004016 -0.86 0.390 -.0113204 .0044219 USGDP L1. 1.721871 .0398639 43.19 0.000 1.643739 1.800003 L2. -.7176714 .0399254 -17.98 0.000 -.7959237 -.6394191 _cons .1050545 .0961537 1.09 0.275 -.8834033 .2935123
Table 4.6iii
27 In this section two more detailed analyses will be performed. First, a Granger-causality test will be performed in order to test whether the 𝛼𝑖𝑡 & 𝛽𝑖𝑡 coefficients in equations (8) and (9) are significant. Is there evidence that the fluctuation of the crude oil price explains the different economic variables? Secondly, since the VAR turned out to be stable (see graph 4.2), it is possible to create Impulse Response Functions (IRFs). The IRFs will provide a means to interpret the pattern of the variables after a shock has occurred.
4.3.3.1. Granger-Causality Test
One of the research topics in this thesis was to investigate whether there is a relation between the fluctuation of the oil price and several economic variables over time. This can be done by testing the significance of the coefficients of the estimated model. Consider formula (10) below:
(10) 𝑌𝑡= 𝜇𝑡+𝑖=1𝑚𝛼𝑖𝑌𝑡−𝑖+𝑗=1𝑚𝛽𝑗𝑋𝑡−𝑗+𝜀𝑡
where 𝑌𝑡 is the economic variable US GDP index over time, 𝑋𝑡is the adjusted crude oil price over time and 𝛼𝑖 and 𝛽𝑗 are the respective coefficients. In order to test whether the adjusted crude oil price is of any effect on the US GDP index, the significance of the 𝛽𝑗coefficient has to be measured. When Granger-causality testing, a test is performed to test the hypothesis 𝛽𝑗=0. This method can be used for all three variables and hence the effect of the fluctuating crude oil price on the selected economic variables can be established. As has been seen in the described literature (section 3), the causality direction can be ambiguous, therefore the Granger-causality test can be performed both ways.
test [Dollar2]L.AdjP2 [Dollar2]L2.AdjP2
test [AdjP2]L.Dollar2 [AdjP2]L2.Dollar2
( 1) [Dollar2]L.AdjP2 = 0 ( 1) [AdjP2]L.Dollar2 = 0 ( 2) [Dollar2]L2.AdjP2 = 0 ( 2) [AdjP2]L2.Dollar2 = 0
chi2( 2) = 4.51 chi2( 2) = 1.64
Prob > chi2 = 0.1049 Prob > chi2 = 0.4415
Table 4.7
The first Granger-causality test has been performed by investigating the coefficient of the effect of the crude oil price fluctuation on the trade weighted dollar value (and vice versa). The results are shown in table 4.7. The null-hypothesis is that the crude oil price coefficients (𝛽𝑗) in both the first and second lag are equal to zero, as is seen in the left part of table 4.7. The reverse goes for the right part of table 4.7, as the first and second lag of the dollar value coefficient are equal to 0.
The test result (𝑐ℎ𝑖2) for the first test is equal to 4.51, which is equal to a F-statistic of 2.26. this means that the null-hypothesis cannot be rejected and that the crude oil price coefficient is not statistically different from 0. In other words, it cannot be concluded that a change in the price for crude oil is a significant driver for the change in the value of the US
28 dollar. Similarly, the value of 𝑐ℎ𝑖2 in the second test is equal to 1.64 (F-statistic = 0.82). Again, this does not provide enough evidence for a conclusion that the value of the dollar significantly affects the trend of the crude oil price and hence the null-hypothesis cannot be rejected. Also the p-values (10.5% and 44%) indicate that the null-hypotheses cannot be rejected at any confidence interval.
Table 4.8 shows the Granger-causality tests for the relation between the US GDP levels and the crude oil price. Again both directions of the causality have been tested and in both cases the null-hypothesis is that the respective coefficients are equal to zero in both lags. The results are provided in the table below:
test [USGDP]L.AdjP2 [USGDP]L2.AdjP2
test [AdjP2]L.USGDP [AdjP2]L2.USGDP
( 1) [USGDP]L.AdjP2 = 0 ( 1) [AdjP2]L.USGDP = 0 ( 2) [USGDP]L2.AdjP2 = 0 ( 2) [AdjP2]L2.USGDP = 0
chi2( 2) = 15.64 chi2( 2) = 9.94
Prob > chi2 = 0.0004 Prob > chi2 = 0.0069
Table 4.8
Opposed to the test-statistics in the table 4.7, the test-statistics regarding the US GDP index and the crude oil price are significant. When assessing the effect of the crude oil price fluctuation on the trend of the US GDP, the 𝑐ℎ𝑖2 is equal to 15.64 (or an F-statistic of 7.82), implying the null-hypothesis can be rejected and the coefficient explaining the crude oil price relation to the US GDP level is significantly unequal to 0. Also in the opposite direction, the coefficient that relates the US GDP level to the movement of the crude oil price appears to be significantly different from 0. With a F-statistic of 4.97 (𝑐ℎ𝑖2=9.94) and a p-value of .0069 the index of US GDP level is significantly impacting the level of the crude oil price.
The last relation that is investigated via the Granger-causality test is the effect of the crude oil price on the output level of the US Industry (and vice versa). The results are shown in table 4.9 and include the general assumptions that the null-hypothesis is equal to the coefficients of the explanatory variable in both lags being equal to 0.
The test statistic (𝑐ℎ𝑖2) for the coefficient of the crude oil price effect on the level of US industrial production is equal to 4.51. At this level the null-hypothesis cannot be rejected and the crude oil price fluctuation has no significant effect on the trend of the production level of the US industry. Equally, the reversed relation proves to have no significant evidence as the test statistic is 3.44.
29 test [IndPro2]L.AdjP2 [IndPro2]L2.AdjP2
test [AdjP2]L.IndPro2 [AdjP2]L2.IndPro2
( 1) [USGDP]L.AdjP2 = 0 ( 1) [AdjP2]L.USGDP = 0 ( 2) [USGDP]L2.AdjP2 = 0 ( 2) [AdjP2]L2.USGDP = 0
chi2( 2) = 4.51 chi2( 2) = 3.44
Prob > chi2 = 0.1047 Prob > chi2 = 0.1794
Table 4.9
After conducting the three Granger-causality tests it can be concluded that only the relation between the crude oil price and the US GDP levels has a significant coefficient that explains their mutual trend, even given a 1% confidence level (both p-values are .0000). When considering the results for the other two relations, perhaps only the effect of the crude oil price on both the value of the dollar and on the US industrial production level can be marked significant but only at a confidence level of 10.5% (p-values of .1048). However, these confidence levels are too low and not marked as such within the range of credibility pursued in this thesis.
4.3.3.2. Impulse Response Functions
After investigating the significance of the coefficients of the explanatory variables using the Granger-causality test, this section will use Impulse Response Functions (IRFs) to investigate the trend that the variables follow after a particular shock. As the Granger-causality test have only deemed the oil price/GDP relationship significant this is the main concern when constructing the IRF’s, however, also the other IRFs will be shown because of any information it will provide.
The IRFs shown in graph 4.3, indicate the effect over time of the four different variables on a shock at t=0. The 16 graphs represent the effect of all variables combined (including themselves) on a shock at t=0. The grey area surrounding the graph shows the 95% confidence interval indicating the range of probability. In general, the uncertainty increases as the number of periods increase. The axis of the separate graphs are different and this proves it difficult to compare the individual response functions. Not all graphs are equally relevant for the research in this thesis. Therefore, the focus will be on some particular relations that are of interest in this research. The focus will be on the impulse response functions where the adjusted WTI price is included in relation to the several identified economic factors.
30 The first impulse response function that shows an interesting result in graph 4.3, is the relation between the WTI price and the value of the dollar. Both impulse response functions have been highlighted in graph 4.4 below. This format makes comparison easier due to the appearance of the axis.
When analysing the results in graph 4.4, is becomes clear that there is a negative relationship between the oil price fluctuation over time and the response of the value of the dollar. This corresponds to the initial hypothesis stemming from graph 4.1, where the value of the dollar and the adjusted WTI price appeared to move in antiphase.
31 Upon closer examination of the impulse response functions, it is clear that both movements are different. The left graph in graph 4.4, indicates a response of the value of the dollar (and its trajectory) in response to a first period shock in the WTI price. Whenever the WTI price increases, the value of the dollar immediately jumps down and continues a downward trend. This trend stabilizes after 3 periods, after which the level of the dollar value returns to its ‘after-shock level’. When the impulse/response is reversed however, the WTI price tends react slower to a shock in the dollar value. Nevertheless, the WTI price does tend to remain negative and does not recover to its original value. Both response functions do indicate that the oil price and dollar value are inversely related.
In graph 4.5 the impulse response function of the US GDP level and the adjusted WTI price is shown. It should be noticed that the scales of the axes differs between both left and right graph and also from graph 4.4. Another interesting feature that differs from the previous graph is the fact that the left response function turns negative, while the right response function turns positive.
The fact that the sign of the relation differs, indicates that the order of impulse and response variable matters. In the left graph a shock in the WTI price at t=0 results in a gradual negative trend of the US GDP level. In other words, an increase in the price for oil translates in lower future GDP levels. This effect is supported by economic theory that is described in section 2, as higher oil price create higher commodity prices that leads to lower consumption and production.
Nevertheless, it has to be noted that for different countries this trend could be different. If, instead of the US, an oil producing (and exporting!) country would have been taken, the GDP level would probably have been proportional to the oil price. For instance, Saudi Arabia’s GDP level is most likely correlated to the oil price as a large part of its GDP is dependent on oil sales, which will be higher if the oil price increases. But, because the oil industry is not the main part of the US economy and because the US has an active crude export ban, the effect of the oil price on the US GDP is negative.
When the order is changed and a GDP shock is introduced to the system, the oil price experiences a positive shock. Again, this corresponds to economic theory as higher GDP level indicates a higher demand which will lead to a higher oil price. This trend is supported by the economic theory in section 2, as the overall demand is an important factor in global oil price levels, especially when it involves a large economy such as the US.
32 Another observation is regarding the scale on both vertical axes in graph 4.5: the GDP shock in the right graph triggers a larger shock in oil price (range = |0 – .5|) than the other way around as seen in the left graph (range = |0 – .15|). Reasons for this effect could be that the effect of economic demand is one of the main drivers for the oil price whereas the oil price is ‘just’ one of the drivers of GDP level (others would include consumer confidence, innovation, trade, etc.).
The last relation that is highlighted is the relation between the oil price fluctuation and the production level of the US industry. The impulse response functions of both shocks are represented in graph 4.6. Both graphs show a similar pattern where there is an initial increase of the response variable but this increase is gradually followed by a decrease which results in a negative longer term effect.
When focussing on the left graph in graph 4.6, the WTI shock in t=0 results in a small increase in the level of US Industry production. This result seems to contrast the findings in graph 4.5, where the increase in oil price decreases the over US GDP. One explanation for this divergence might be that the US industry production is measured in $value output. This would imply that due to the increasing oil price, the prices of products (which include oil) would also increase and hence the total value of the industry output would initially increase. Nevertheless, the increase is overturned at t=2, which can be explained because of a similar effect as the decrease of GDP levels in graph 4.5: as oil prices rise, the cost of production rises, causing lower demand and lower production.
The right graph in graph 4.6 shows the response of the oil price to a positive shock of the US industry production. Similar to the left graph, there is an initial increase in the oil price. A possible explanation would be that as oil is widely used as both feed stock and energy source in the industry, higher production means higher demand for oil which increases the price. However, after 3 periods the price increase declines resulting in a slightly negative price effect in t=8. This could be explained by the fact that increased demand for oil (and the subsequent increased price) would spur oil producers to increase their supply and hence putting a downward pressure on the oil price until a new equilibrium is reached.
Again, it has to be noted that the size and timeframe of the effect in both directions differs: the effect of an oil price shock is larger (|0.23|) compared to the effect of an Industry production shock (|0.07|). Also the initial increase is shorter and lower in the left graph as the peak in US industry production is 0.03 and reached in t=1 compared to the right graph where the peak is reached in t=3 at a high of 0.11.
