Upward and Downward Trends of the Crude Oil Price: Their Main Causes and Driving Factors

(di Giovanni Coppola)


As soon as the first oil shocks occurred in the U.S. during the 1850s, economists and scientists have always tried to find specific tools that could enable to predict future changes in its price. At that time the crude oil began to be recognized as a valuable commodity and very early took the place of turpentine or whale’s oil in generating energy useful to enlight. Given its incredible success, it started to be commercialized and its production increased in huge amounts, from a half million barrels to more than 2 millions; the price quickly dropped until reaching 10 cents for a barrel. But that lasted only for some months, until the outbreak of the U.S. civil war provoked a general surge in prices of commodities, due to the cut-off of supplies. After the war however, a series of similar boom-bust cycles was registered, mainly due to the discovery of new oil-weels (especially in Pennsylvania), that resulted in an incredible increase in production and consequent decrease in prices.

With the new century’s coming, oil not only gained importance in terms of economic value but it became a fundamental “ingredient” in some key economic sectors such as automobile manufacturing and sales. Before the beginning of the II World War, U.S. suffered from a serious gasoline famine that obliged some businesses on the West Coast to shut down, prices once again augmented, until new earnings in production from Texas, California, and Oklahoma quickly eliminated the shortages of 1920 and re-established an equilibrium. During the same period, with the growth in supplies and the falling demand, the U.S. government started to focus its attention on how to efficiently manage oil reserves. It was quite clear the need for regulations, thus some regulatory agencies such as the Texas Railroad Commission (TRC) and the Oklahoma Corporation Commission were established for this purpose. In addition the states’ power was reinforced at the federal level by the National Industrial Recovery Act of 1933 and the Connally Hot Oil Act of 1935, which prohibited interstate shipments of oil produced in violation of the state regulatory limits.

The TRC in particular played a great role in the post-war era since it was responsible for forecasting product demand at the current price and set allowable production levels consistent with this demand, ensuring in this way a constant price from one month to another. A contribution in maintaining steady prices was also given by the burst of the Korean war, nevertheless after 1953 a sequence of events brought to the Suez canal crises. More than 40 ships sunk during the conflict between Israel (supported by France and England hoping to regain control over the canal) and Egypt which caused an enormous fall in the total world production of oil (10.1%) with the consequent surge in prices.

The following decades were characterized by almost permanent production shortages mainly due to the depletion of U.S oil fields and the termination of the rights of foreign central banks to exchange dollars for gold, establishing the end of Bretton Woods system. Other than that, a sequence of war conflicts, especially in the middle-east, firstly brought to an embargo on oil exports (as in the case of the Egyptian-Israeli war 1973-74), then to cutbacks in its production (as in the case of the Iran-Iraq war 1980-81 and the First Persian Gulf war 1990-91) with the easily forecastable increase in prices. This trend overturned only for the last years of 80s when the Arabs, in spite of the Iran-Iraq conflict, decided to increase their production.

The advent of the new millennium marked a drastic transformation in the people standards of living since many countries progress from the status of agricultural economies to industrial ones. This transition affected also the world oil’s market as confirmed by the huge rise in the global oil consumption (from around 17% in 1998 to more than 69% since then) and demand. The latter is mainly explained by some Asian countries such as China and Japan that after experiencing a currency crisis in late 90s, began to grow limitlessly.

This upward and downward pattern of crude oil’s price, as past history teaches us, is unpredictable and is still distinguishing our days, with the consequent tireless research about the reasons of those movements. In that sense, the main aim of this research project is to reply to two principal questions: “How do events like terroristic attacks, wars and economic crises influence the price of oil?” And “How does the price of oil, related to these events, influences the USA GDP?”

In order to answer them, it has been necessary to construct a specific model with two regressions, considering monthly data belonging to a temporal range going from the February of 1994 to the December of 2011.  

Literature Review

A great part of theoretical models of transmission of oil price shocks have focused on the implications of exogenous variation in the price of imported crude oil. An unexpected increase in the price of imported crude oil brings to a reduction in purchasing power of domestic households and increases the cost of producing domestic output to the extent that oil is a factor of production along with capital and labor, which is akin to adverse aggregate supply shock. These two effects are symmetric in increases and decreases. Recently evidence has been found showing that the domestic demand channel of transmission dominates in practice while domestic supply channel of transmission is weak (Kilian and Park, 2009). Due to that, we would expect an oil price shock to be deflationary of inflationary if it occurs in isolation. However, it is possible that the curves demand and supply shift together reinforcing the decline in GDP but offsetting the effects of the oil price shock on the price level. Nevertheless we don’t have an interpretation that matches the common perception that exogenous oil price shocks are recessionary or strong inflationary (Bruno and Sachs, 1982). The difficulty of explaining large recessions on the basis of exogenous oil supply shocks is a common problem in empirical work. For now, the central pillar remains the fact that the response of the economy is asymmetric in positive and negative oil price shocks, such that positive oil price shocks generate large recessions, while negative price shocks have little or no effect on the economy. Additional indirect effects of unexpected changes in the real price of oil are: reallocation effect, uncertainty effect. In case of reallocation effect, the responses of real output to oil are necessary asymmetric in unanticipated oil price increases and unanticipated oil price decreases, while the uncertainty effect is measured by the expected volatility of the real price of oil over the relevant investment horizon and prompts firm to delay investments, causing them to drop.

In approaching to this research project, a series of recent works regarding the possible factors affecting the price of oil have been analyzed. This review turned out to be a good and fundamental starting point in favor of choosing the variables of the regressions.

The first work taken into consideration is a research conducted in the first part of 2014 by EIA, the U.S. Energy Information Administration, an institution responsible for collecting, analyzing, and disseminating energy informations on some factors that affect crude oil price, availing of chart data updated monthly and quarterly.

From the first graph (Appendix A, Fig. 1), it can be seen how the crude oil price reacts to a variety of geopolitical and economic events. The record peak of WTI price has been reached at the beginning of the financial crisis in 2008.

Second thing highlighted (Appendix A, Fig. 2) is the simultaneous movement of the different indexes of oil prices due to arbitrage. For this analysis, the major oil indexes are taken into account and as it is shown, they move together along time.

A third effect is illustrated by another graph (Appendix A, Fig. 3): the crude oil prices are the primary driver of petroleum product prices.

Lastly, since the Saudi Arabian market is the biggest one in the world for crude oil, it often acts to balance the oil market: cuts in its production tend to lead to price increases (Appendix A, Fig. 4).

A second paper (Christos Kollias, Catherine Kyrtsou, Stephanos Papadamou, 2013) reports the effects created by wars and terrorist attacks on the covariance between oil prices and the indexes of four major stock markets which are S&P500 (American), DAX, CAC40, FTSE100 (European). The authors chose to develop a non-linear BEKK-GARCH models to examine the previous relationship and to discover if the European and American markets react in a different way to these events.

The model is constructed using covariance matrices and dummies with value 1 if a particular event occurred and value 0 if it did not; terrorist attacks have been classified in two broad groups: center for events that took place in western countries (US, Spain, Italy etc.) or the terrorists’ targets were of western interest; and periphery for events occurred in eastern countries (Russia, China etc.) or targets were of eastern interest.

Preliminary econometric analysis confirmed data to be non-stationary, with presence of autocorrelations, asymmetry and heteroskedasticity. The conclusions showed that in case of war conflicts the covariance between stock indexes and oil returns decreases, while terrorists’ actions caused only a simultaneous movement (decrease in covariance) of CAC40 and DAX indexes. S&P and FTSE100 remain unaltered in both cases. Other results found that wars’ effects are concentrated during the initial phases of the events: that enables market investors to adjust quickly to longer lasting effects. Terrorist attacks instead, favor diversification since they only affect some markets.

Another research, done by Lutz Kilian, studies the causes and consequences of oil price shocks along time. Some of the key insights of his paper have been founding that that the real price of oil is endogenous with respect to economic fundamentals and that oil price shocks do not occur ceteris paribus. It has been uncovered (as already said in the Introduction) that in U.S. the government has been the regulatory authority of the oil price until 1971 when the country ceased to be a net exporter of crude oil. The price was setted based on the forecasted oil consumption (Hamilton, 1983). Then during 80s the U.S. started to be heavily dependent on middle-east supplies and the WTI oil price has been allowed to converge to the global price. So, in the first period real price of oil has been exogenous respect to the global economy, while after 1971 it became endogenous (Alquist et al. 2013).

Exogenous oil supply shocks with respect to U.S. have also had an impact on WTI price but less than shocks in demand, as 1974 increase in price suggests (Hamilton, 2003; Killian, 2008).

Political events in the Middle East too, may have affected the real price of oil by shifting expectations about future shortfalls of oil supply relative to oil demand: that could have caused speculations, since a consistent part of people would have cut its production to store oil for the future selling at a higher price. Financial markets provided an alternative venue to trade on the basis of expectations but speculative pressures here were ruled out by the arbitrage condition (Killian and Murphy; Killian and Lee, 2013).

Another possible explanation to the price fluctuations is based on the role of OPEC in controlling the price of oil since 1980 (Smith, 2005), but the attempts to act as a cartel and to prevent oil price to fall vanished giving no evidence to what said.

Measures Section

For the study, it has been decided to adopt the WTI Crude Oil index (*fdWTICrudeOil) as the dependent variable, which is the main benchmark in oil pricing. The influencing factors taken into account are: the Standard&Poor 500 index (fdSP500), the OPEC Basket price index (fdOPECBasket), the Variation in Standard Demand of oil barrels (STDemandVar), the Variation in Standard Supply of oil barrels (STSupplyVar), the Price of Gas (fdPriceofgas), the happening of Geopolitical Events (GeopoliticalEvents), Wars in Middle East (WarsinMiddleEast), Wars in Afghanistan (WarsinAfghanistan), Terrorist Attacks (TerroristicAttack) and Economic Crises Cycles (EconomicCrises – Appendix B).

The role of the Standard&Poor 500 (an index that follows the gait of the 500 american most capitalized firms and that is calculated by multiplying the price of each stock in dollars by the number of share publicly available for each stock, dividing the whole for a divisor that measures the total market capitalization), the OPEC Basket price index (a weighted average of prices in dollars for petroleum blends produced by the OPEC countries), the Variation in Standard Demand and Supply of oil barrels (in gallons), and the Price of Gas (in dollars), is motivated by the strong influence that they exerts on the oil’s price, as it will become clear soon. The dummies Geopolitical Events, Wars in Middle East, Wars in Afghanistan, Terrorist Attacks and Economic Crises Cycles, have been exploited for understanding how much such qualitative events can cause changes in the explained variable. The latter in particular, has become the peculiarity of this work, since in the past literature there are no feedbacks about that kind of phenomenons.

Standard demand and supply variations variables are taken in account as logarythms since they represent an annual percentage variation: all their values are included between 0 and 1.

It has also been appointed to study how alterations in crude oil’s price affect the USA real GDP, the second regression presented has this scope. In this case the explanatory variables are represented by Personal Consumption

Expenditure (fdPCEbillionsofdollars), Government Investment (fdGovInvbillionsofdollars), Government Expenditure (fdGovExpBillionsofdollars) and Net Exports (fdNetExportsBillionsofdollars). For reasons of completeness and in order of relating the two regressions, some interactions such as WTIxTerroristic Attacks (WTIxTerroristicattacks), WTIxWars in Middle East (fdWTIxWarintheMiddleEast), WTIxWars in Afghanistan (fdWTIxWarinAfghanistan), WTIxEconomic Crises (WTIxEconomiccrises) and WTIxGeopolitical Events (WTIxGeopoliticalevents) have been included (Appendix B). The main reason behind the presence of interaction variables is due by the will of studying how the price of oil (represented by the WTI index), considered in the same framework of the happening of a particular event (such as wars or economic crises), influences the real value of the US GDP.

The sources used to collect the data, organized in a monthly frequency (from February 1994 to December 2011 as said), are trusty entities such as EIA, the World Bank, Yahoo Finance, OPEC, St. Louis Fed (for a detailed overview, see the links in the references‘ section).

* In parenthesis the names of the regressors used in STATA Tables.

Empirical Model and Data

The first regression outline looks like that:

WTICrudeOILt = β0 + β1 (S&P500)t + β2 (OPECBasket)t + β3 ln(STDemandVar)t + β4 ln(StSupplyVar)t + β5 (Priceofgas)t + δ6 (GeopoliticalEvents)t + δ7 (WarsinMiddleEast)t + δ8 (WarsinAfghanistan)t + δ9 (TerroristicAttacks)t + δ10 (EconomicCrises)t +ut

The second model overview:

USrealGDPt = β0 + β1 (PCE)t + β2 (GovInv)t + β3 (GovExp)t + β4 (NetExp)t + δ5 (WTIxTerroristicAttacks)t + δ6 (WTIxWarsinMiddleEast)t + δ7 (WTIxWarsinAfghanistan)t + δ8 (WTIxEconomicCrises)t + δ9 (WTIxGeopoliticalEvents)t + ut

Violations of the Assumptions

In order to implement correctly the regressions, to get unbiased and consistent OLS estimator and to be able to inference, it has been necessary to check the six classical linear model (CLM) assumptions for time series.

Before analyzing the possible violations regarding the two models, it is useful to state the assumptions:

-Assumption TS.1 (Linearity in Parameters): The stochastic process follows the linear model yt = β01xt1+…+ βkxtk+ut, where ut is called the error of disturbance;

-Assumption TS.2 (No Perfect Collinearity): In the sample, none of the independent variables is a linear combination of the others;

-Assumption TS.3 (Zero Conditional Mean of the error or Strict Exogeneity): For each t, the expected value of the error ut, given the explanatory variables for all time periods, is zero. E(ut|X)=0, t=1,2,…,n;

-Assumption TS.4 (Homoskedasticity): Conditional on X, the variance of ut is the same for all t: Var(ut|X)=Var(ut)=σ2, t=1,2,…,n;

-Assumption TS.5 (No Serial Correlation): Conditional on X, the errors in two different time periods are uncorrelated: Corr(ut,us)=0, for all t≠s;

-Assumption TS.6 (Normality): The errors ut are independent and identically distributed as Normal(0,σ2).

Under TS.1 through TS.3, the OLS estimators are unbiased (1). Under TS.1 through TS.5 (Gauss-Markov assumptions), the OLS estimators are BLUE (2). By adding also TS.6, the OLS estimators become normally distributed, conditional on X, and under the null hypothesis the t statistic has a t distribution and the F statistic has an F distribution.

It is straightforward to check TS.1 for both models: it is respected as can be seen by looking at the two regressions in the previous paragraph (3. Empirical Model and Data).

Assumption TS.2 is also confirmed in both regressions. At a first glance, this assumption is likely to be violated due to the presence, in both models, of lots of variables and especially dummies. In reality, all of them are very well specified. Moreover, in the final models some variables have been dropped due to their insignificance, and since these are not correlated with the ones still present in the equations, the problem of omitted variables bias can be ruled out.

From the tests regarding the zero conditional mean assumption, it turned out that, in the two regressions, the error terms are not correlated with the variables: it has been possible to state this conclusion after getting the residuals and correlating with all the x’s in the models (Appendix C, Tables 1 and 2) (3).

In order to test for heteroskedasticity of the error terms in the two regressions, two tests have been performed: the Breusch-Pagan Test (4) and the White Test (5). The former is used to find linear form of heteroskedasticity, the latter for nonlinear ones. They both rely on the null hypothesis of homoskedasticity, against the alternative of heteroskedasticity. It emerged that in both tests and in both regressions there is no evidence of heteroskedasticity, thus failing to reject the null hypothesis (Appendix C, Tables 3).

In order to be sure that the error terms have constant variance, it has been conducted a regression for both models using robust standard errors. Again, the previous result is confirmed from the fact that both the coefficients and the standard errors from the robust regression do not vary significantly, remaining almost unaffected from those obtained by the normal regressions (Appendix C, Tables 4 and 5). The homoskedasticity of the error terms can also be confirmed from the fact that there is almost no presence of outliers in the residuals (Appendix C, Fig. 6).

In addition, in both regressions there are no problems of serial correlation. To test for possible violations of assumption TS.5, the Durbin-Watson Test(6) has been used. The result of the test for the first regression is DW(5,214)=1.957593, which lies between du=1.82 and 2, meaning that there is no evidence of autocorrelation; while the outcome of the second regression is DW(8,71)=2.216928, which lies inside 2 and 4-du≈2.3, ending exactly with the same same result as before (Appendix C, Fig. 7) (7).

Concluding, also assumption TS.6 is not violated in both regressions. The normality in the error term can be ensured by serving of some graphical representations (Appendix C, Fig. 8). In particular, from the histograms on the distributions of the residuals, it is possible to notice the bell-shaped curve, typical of the Normal distribution.


(1)E(β^)= β: this means that after drawing infinite random samples from the population, computing an estimate at each time, and averaging these estimates over all the random samples, it is possible to obtain beta.
(2)Best Linear Unbiased Estimator.
(3)If Cov(ut,X)=0, then E(ut|X)=E(ut)=0.
(4)the Breusch-Pagan Test regresses the squared residuals on the explanatory variables and then performs an F-statistic of overall significance of all the coefficients;
(5)the White Test regresses the squared residuals on the explanatory variables, their squares and crossed product, and then perform an F-statistic of overall significance of all the coefficients;
(6)the Durbin-Watson Test regresses the residuals on one period lag residuals
ut= ρut-1+et and then uses the test to check the null hypothesis of no autocorrelation (H0: ρ=0);
(7)explanation of DW statistic can be found in Wooldridge (2010).


In pursuance of working with time series, first of all it has been necessary to detect stationarity and to correct it where needed. For studying it, the Dickey-Fuller test has been applied, thanks to whom the presence of stationarity emerged. In both the regressions, stationarity has been highlighted only in some variables, in particular in WTI CrudeOil, S&P500, OPECBasket and Price of gas in the first regression; and US real GDP, P.C.E., Government Investments, Government Expenditure, Net Exports, WTIxAfghanistan and WTIxWar in Middle East for the second one. For them, as Appendix D Fig. 1 shows, stationarity has been corrected taking the first-difference, generating a series of new variables, which are used to get the OLS estimators for the full regressions (Appendix D, Tables 2 and 3).

The first impression about the outputs obtained regressing the two models with STATA reveals that the variables are able to explain the 90% (R2=0.9010) of the variation in the WTI Crude Oil and almost 60% (R2=0.5899) of the US real GDP. Looking at the p-value of the coefficients it emerged that some of them could be not significant. For investigating this intuition, firstly it has been necessary to check for the correlation between the variables (Appendix D, Tables 4 and 5), and after that, an F-statistic for multiple exclusion restriction has been implemented. For the first regression the test is stated as follow:

H0: β345789=0
H1:H0 not true;

and for the second one the following:

H0: δ59=0
H1: H0 not true.

The first model resulted in having a value of F(6, 203) equal to 0.44, which is smaller than the 5% Critical Value of the F-distribution (2.10). Regarding the second regression, a value of F(2, 61) equal to 0.11 has been founded, which is smaller than the 5% Critical Value (3.15). Thus, in both tests the null hypothesis failed to be rejected, concluding the irrelevance of these variables. As a consequence, the two initial regressions have been reformulated, dropping the unnecessary explanatory variables and running the two final models

(Appendix D, Tables 6 and 7). The first one resulted to be:

WTICrudeOILt = β0 + β1 (S&P500)t + β2 (OPECBasket)t + δ6 (GeopoliticalEvents)t + δ10 (EconomicCrises)t +ut

while the second:

USrealGDPt = β0 + β1 (PCE)t + β2 (GovInv)t + β3 (GovExp)t + β4 (NetExp)tδ6 + (WTIxWarsinMiddleEast)t + δ7 (WTIxWarsinAfghanistan)t + δ8 (WTIxEconomicCrises)t + ut

In both regressions, the R2 and the adjusted-R2 do not have significant changes: regarding the first regression (Appendix D,Table 6), even though the R2 has a slight decrease, the adjusted-R2 slightly increased. Almost the same happens in the second regression (Appendix D, Table 7), meaning that the excluded variables are not so useful to explain the variation in either the WTICrudeOil and USrealGDP.

Noteworthy is the fact that OPECBasket and PCE, in the first and second model respectively, remain the two explanatory variables with greater magnitude, before and after dropping the regressors.

Concluding, for what concerns the initial issues at the center of this study, it can be said that the two big wars in which the United States engaged in the last two decades did not influence and are still not influencing the changes in the WTI Crude Oil price. The same can be stated for the terrorist attacks, while, as can be supposed, economic and geopolitical events have had a central role in showing the behavior of the dependent variable.

Regarding the second question, as someone can expect, the determinants of the GDP have a leading role in explaining its eventual changes. As shown in this research, also the price of oil during economic crisis and the two main wars contributed to influence the US real GDP.


At the end of this work, it is worthwhile to highlight the fact that the final results reached in analyzing the first regression are quite surprising. If we think about how past events such as wars affected oil production and caused movements in its price, it was quite unforecastable that terroristic attacks suffered and wars in which USA engaged are not very significant in explaining changes in the oil price. On the contrary it has been reinforced the concept that geopolitical events usually have an impact on crude oil price. Although that, the latter, together with terroristic attacks again, are useless in explaining contingent real GDP fluctuations: something that can thought to be logic. Furthermore, since in both regressions no problems of autocorrelation between variables came out, it is possible to say that each one of them, taken individually, has a leading role in explaining dependent variable’s movements. The proof that confirms this theory (at least for the 1st regression) is the very high R2 that measures variability.

Finally, this work tries to be as much complete and realistic as possible, but for a better model it would have been necessary to include also variables regarding price’s indexes of oil’s substitute energies and indexes regarding the exhaustion of the oil’s deposits into the first regression, and variables such as consumption, real wages and exchange rate to the second one. The reason of their exclusion is mainly due to technical problems in their research and to incompatibility with the time range on which I referred.



ABOUT NEWS. (2015) How Are Oil Prices Determined?. Available from:

APERGIS, N. and MILLER, SM. (2009) Do structural oil-market shocks affect stock prices?. Energy Economics. Available from:

ASCHE, F., GJØLBERG, O. and VÖLKER, T. (2003) Price relationships in the petroleum market: an analysis of crude oil and refined product prices. Energy Economics. Available from:

CHICAGO TRIBUNE. (2015) Falling crude oil prices’ impact on U.S. economy. Available from:

ELYASIANI, E., MANSUR, I. and ODUSAMI, B. (2011) Oil price shocks and industry stock returns. Energy Economics



HAMILTON, J. D. (2011) Historical oil shocks. Available from:

JI, Q. (2012) System analysis approach for the identification of factors driving crude oil prices. Computers & Industrial Engineering. Available from:

KILIAN, L. (2012) Oil Price Shocks: Causes and Consequences. Available from:

KILIAN, L., VIGFUSSON, R. J. (2014) The Role of Oil Price Shocks in Causing U.S. Recessions. Available from:

KOLLIAS, C., KYRTSOU, C. and PAPADAMOU, S. (2013) The effects of terrorism and war on the oil price–stock index relationship. Available from:

KRICHENE, N. (2002) World crude oil and natural gas: a demand and supply model. Energy Economics. Available from:

LAYIWOLA, S. (2010) Impact of US Crude Oil Inventory on West Texas Intermediate (WTI) Crude Oil Prices. Available from:

MOSTERT, F. (2010) Factors influencing the International Price of Oil in the Medium to Long-Term. Available from:

NATURAL RESOURCES CANADA. (2010) Review of Issues Affecting the Price of Crude Oil. Available from:


WORLD BANK GROUP. Available from:

YAHOO. Available from:

YANG, CW., HWANG, MJ. and HUANG, BN. (2002) An analysis of factors affecting price volatility of the US oil market. Energy Economics

ZHANG, X., YU, L., WANG, S. and LAI, KK. (2009) Estimating the impact of extreme events on crude oil price: An EMD-based event analysis method. Energy Economics. Available from: