The Impact Of Oil Price Changes On The Macroeconomic Performance Of Ukraine

Following existing literaturethe focus is on six macroeconomic variables: nominal foreignexchange rate, CPI, real GDP, interest rate, monetaryaggregate M1 and average world price of oil.

THE IMPACT OF OIL PRICECHANGES ON THEMACROECONOMICPERFORMANCE OF UKRAINE

byOleg Zaytsev

A thesis submitted in partialfulfillment of the requirements

for the degree of

Kyiv School of Economics

Abstract

THE IMPACT OF OIL PRICECHANGES ON THEMACROECONOMICPERFORMANCE OFUKRAINE

by Oleg ZaytsevThesis Supervisor: Professor Iryna Lukyanenko

In this research we investigate the impact of oil pricechanges on Ukrainian economy Following existing literaturethe focus is on six macroeconomic variables: nominal foreignexchange rate, CPI, real GDP, interest rate, monetaryaggregate M1 and average world price of oil Adhering toCologni and Manera (2008) we allow for interconnectionbetween the variables to exist and adopt SVAR/VECMapproach for this purpose In particular, we choose betweenthe two closely related model types based on cointegrationproperties of the data We succeed in detecting long-runequilibria, estimate VECM and further perform innovationaccounting We find that oil price increases tend todeteriorate real economic activity in the short run (thoughwith one month lag) as opposed to the long run The reactiongoes through indirect effect, namely downward demandeffect, which is characterized by contraction of aggregatedemand in response to adverse oil supply shock Based on

the results of IRF we further numerically confirm the validity

of this channel Finally, we check if the asymmetry effectbetween oil price changes and real GDP response asdiscovered by Mork (1989) is present in Ukrainian data Wefind sustaining evidence in favor of symmetric response ofreal GDP to oil price increases/decreases in the short run

TABLE OF CONTENTS

Pa ge

CONCLUSION

……….………

46BIBLIOGRAPH

I sincerely thank my family, especially parents andgrandparents, the nearest people I have in life, for their love,endless support and spiritual encouragement Additionalgratitude is to Vladimir Vysotsky and Boris Grebenshchikovfor their immortal songs and philosophical nourishment Allthis helped me a lot while being a KSE student

Crude oil A naturally occurring, flammable liquid consisting

of a complex mixture of hydrocarbons of various molecularweights, and other organic compounds, that are found ingeologic formations beneath the Earth's surface

C h a p t e r 1

INTRODUCTION

Public attitude towards oil is ambiguous Some people thinkthat oil is the major threat to enduring economicdevelopment, social equality, environment and peace Theirviewpoint is most accurately expressed in the article byMichael Hirsh, Newsweek’s national economicscorrespondent, where his personal perception of ‘CrudeWorld: The Violent Twilight of Oil’, the hue and cry bookwritten by Peter Maass, is given Hirsh claims the following:

‘Oil is the curse of the modern world; it is “the devil’sexcrement,” in the words of the former Venezuelan oilminister Juan Pablo Pérez Alfonzo, who is considered to bethe father of OPEC and thus should know Our insatiableneed for oil has brought us global warming, Islamicfundamentalism and environmental depredation It hasturned the United States and China, the world’s biggestconsumers of petroleum, into greedy, irresponsibleaddicts that cannot see beyond their next fix With a fewexceptions, like Norway and the United Arab Emirates, oildoesn’t even benefit the nations from which it is extracted

On the contrary: most oil-rich states have been doomed to

a seemingly permanent condition of kleptocracy by a few,poverty for the rest, chronic backwardness and, worst ofall, the loss of a national soul’ (Hirsh, 2009)

Others believe that the usefulness of oil in modern economicsetting can not be questioned George W Bush, for instance,calls oil ‘a fluid without which our civilization would collapse’(Hirsh, 2009) Michael Schirber sets off his article ‘TheChemistry of life: Oil’s many uses’ by claiming that ‘besideswater, there's no liquid that humans rely on more thanpetroleum’ (Schirber, 2009) And to certain extent both ofthem are right Oil is the central production factor of theworld economy According to EIA estimates, between 1996and 2006 the total share of oil as a source of commercialenergy constituted 56% from total energy use ‘Black gold’powers machines and automobiles, and is the basic materialfor a wide range of products such as lubricants, asphalt, tars,tires, solvents, plastic, foams, bubble gums, DVDs,deodorants, crayons, to mention just a few The amount of oiland derived products an economy consumes depends uponnumerous factors Following Bacon and Kojima (2008) suchfactors as the level of GDP, industrial sector structure of aneconomy, the availability of choices among fuels that permitsubstitution, level of technological progress are the mostimportant ones All taken together they describe the stage ofeconomic development the country is in It is important torealize that in principal the use of crude oil after it isremoved from the ground is limited But the situation isabsolutely reversed for the products, which become availableafter extracted oil is refined Oil products, mainly fuels, areimportant for different sectors of an economy with a special

emphasis being assigned to transportation, construction,industrial and power-producing sectors Moreover, householduse of oil is overwhelmingly important for low-incomecountries, where the power-producing sector is in infantilephase.

Due to the fact that oil is widely used across all sectors ofUkrainian economy with no effective cost-beneficialsubstitute available, and taking into account that its pricedynamics has been relatively volatile in recent years, I wouldlike to investigate if the conventional hypothesis, aspioneered by Hamilton (1983), that oil price fluctuations mayadversely affect country’s macroeconomic performance holdsfor Ukraine The variables of interest (i.e endogenousvariables) are seasonally adjusted real GDP and nominalforeign exchange rate, interest rate, monetary aggregate M1and inflation level The choice of variables is mainly driven bysimilar studies, in particular Cologni and Manera (2008) isused as a benchmark, which have been conducted fordeveloping countries and is in accord with economic theory.The set of variables considered in this research may bedecomposed into control and state ones The former groupincludes M1, interest rate and foreign exchange rate, whichare used as leverage by the government In other words,their values are manipulated, so that the desirable economiceffect is obtained The latter group of variables counts thetwo main indicators of economy’s health, i.e inflation leveland real GDP The period under consideration is chosen to be

01/1996-12/2006 with data frequency being one month Theresearch differs from the others conducted on Ukrainian data

in that it considers and estimates the main economicvariables under study in a system framework viaSVAR/VECM approach, which allows for variable’scontemporaneous and lagged interconnection, whittles awayendogeneity problem In addition, we investigate if theasymmetric relationship between oil price changes and realGDP, which is characterized by unequal responses of thelatter to up- and downside movements of the former variable,

is present in the data To accomplish the latter goal,approach introduced in Mork (1989) is utilized

The topic is relevant for Ukraine as Ukraine suffers fromshortage of internal energy resources, including oil This fact

is supported by available statistics1, which reveals that evennow there is a huge gap between consumption andproduction sides of oil in Ukraine, hence to overcome theseimbalances the country will still heavily rely on import,mainly from Russian Federation and Kazakhstan, andcontinue to be vulnerable to external factors such as oil pricefluctuations, i.e Ukraine is exposed to oil price risk Theformal proof of the above statement is provided in the study

by Bacon and Kojima (2008), where the authors quantify oilprice risk exposure of any country by referring to itsvulnerability to oil price increases, which is defined as ‘the

1 In 2008 Ukraine consumed the amount of oil (370 bpd) almost four times of what it produced (95.17 bpd) (EIA).

ratio of the value of net oil imports (crude oil and its refiningproducts) to GDP’ (Bacon, Kojima, 2008) With regards toUkraine, its vulnerability measure amounted to 5.2% in 1996,5.0% in 2001, and 5.3% in 2006 In other words, during theconsidered years the country’s average net import of oilconstituted 5.1% of its GDP It may be inferred that change invulnerability between 1996 and 2006 constituted 0.1% Thischange is decomposed via refined Laspeyres index as follows(focus is mainly on consumption effects): oil price effectthrough consumption is 11.7, oil share in energy effect is -1.1, energy intensity effect is -4.9, real exchange rate effect is-4.6, total consumption effect is 1.0, total production effect is-0.9 What should be emphasized is extremely high value ofoil price effect through consumption if compared to otherfactors, which can be interpreted as a one percentage pointincrease in oil price will result in 11.7 percentage increase incountry’s vulnerability index during the considered periodgiven other factors are held constant This is indicative ofreduced aggregate demand, further drop in output andreduced economic activity.

The rest of this research paper is organized as follows: in thenext section brief overview of existing literature,methodologies used and channels representing oiltransmission mechanism is given, next particularmethodology applied is described, then comes datadescription, interpretation of the estimation results, andcorollary section completes the research To conclude,

according to the theory, oil price increases are expected tonegatively affect net oil importers through rising import bills,which in turn effect other prices, and lead to rising inflation,reduced macroeconomic demand (output level) and furtherunemployment (Bacon, Kojima, 2008) The scale of economicdecline is different for different countries and varies with theextent to which countries are dependent on oil.

C h a p t e r 2

LITERATURE REVIEW

Nowadays there exists a great mass of academic literaturefocusing on the economic properties of oil, its impact on theaggregate world economy and specifically on economies ofdifferent types (say, net exporters or net importers of oil,emerging or developed ones etc) Some of the papersconsider the impact of oil on particular economic variables,i.e estimate oil price pass-through into, say, exchange rate,inflation or unemployment; others estimate the system ofequations via appropriate econometric techniques to accountfor interrelationship between the included variables as well

as external (exogenous) ones and do innovation accounting,i.e compute impulse responses to oil price shocks, evaluateits significance, determine its magnitude, speed ofconvergence to the long-run value as measured by the time ittakes for the reaction to disappear Another block of papers,which should be highlighted separately, is the one where thequestion of asymmetric relationship between the level of oilprice and economic activity is investigated There are alsosome studies, which focus entirely on Ukraine and thus are ofparticular interest In addition, in order to justify the choice

of variables mentioned in the introduction, oil transmissionmechanism is considered in details The structure of theliterature review will be consistent with the aforementioned

strands of existent literature, while only the most importantarticles will be discussed.

The vivid example of the first block of papers is a researchundertaken by Chen (2009), where the author’s mainintention is to generalize a series of papers, which focus onthe issue of declining oil price pass-through into inflation,phenomenon, which is confirmed empirically by now.Moreover, an additional step is undertaken, so that thefactors, which may explain this decline, are formallydetermined Employing data from 19 industrialized countriesand estimating augmented Phillips curve model with errorcorrection term, the author proceeds by checking thestability of the estimated short-run pass-through coefficientsvia one-time structural break test proposed by Andrews(1993) The results of the test indicate that the majority ofcountries under study fail to reject the null of no structuralbreak In the next step a dummy variable is added into thecore estimation equation for the purpose of dividing thesample period into two parts: before and after the breakdate This time estimation results turn out to be different,with short-run pass-through coefficients being significantlylower in the post-break period The core innovation behindthis study is that rather than assuming a discrete number ofstructural breaks in the pass-through2, the author suggestsaccounting for its smooth change over time, i.e treating it as

a random variable following a martingale (random walk) This

2 Tests proposed by Bai and Perron (2003) may be useful for this purpose.

is a plausible assumption, since ‘it would be difficult tobelieve that a sudden innovation or shock would exist thatmakes the degree of pass-through change dramatically’(Chen, 2009) Moreover, for the majority of countries it issupported by the results of Hansen stability test The coremodel is modified so that it incorporates this change and isestimated via state space method Empirical conclusionreached is that for industrialized countries underexamination there appears to be a downward trend, i.e.gradual decline, in the oil price pass-through into inflationduring the covered period The major factors, which explainthis phenomenon, as determined by the results of fixed effectpanel regression with dependent variable being a series ofshort-run pass-through coefficients, are the declining share ofenergy consumption in the economy, favorable exchange ratemovements, higher degree of trade openness andaccommodative monetary policy The first two factors areexactly those used by Bacon and Kojima (2008) in thedecomposition analysis of oil price vulnerability index andhence empirically confirm its appropriateness.

Another paper, which perfectly fits into this category, is theone by Chen (2007) Relying on the fact that real shocks arethe primary causes of exchange rate fluctuations asconfirmed by Clarida and Gali (1994), the author considersthe long-run nexus between oil price and real exchange rateusing a monthly panel data for G7 countries In other words,cointegration properties of oil price and exchange rate time

series are examined using panel cointegration techniques.The reason for doing panel analysis is that country-by-country Johansen test results produce mixed evidence infavor of cointegration existence For some countriescointegration vector may be identified (Germany, Japan), forsome not (Canada, France, Italy, UK) Thus, based on theseestimates, the main hypothesis, which says that the realexchange rate is positively related to the real oil price, cannot be supported empirically, though theoretically it is true.

To overcome this collision, the author pools the data for allcountries under examination and after the implementation ofthe Fisher-type ADF and Phillips-Perron tests for panel unitroot, runs the panel cointegration residual-based testproposed by Pedroni (2004) This time obtained statisticsindicates fairly strong support for the hypothesis ofcointegration relationship between the two variables Therobustness check of this result is performed via likelihood-based cointegration test proposed by Larsson (2001), whichallows for the possibility of multiple cointegrating vectorsand thereby provides stronger evidence of cointegration At1% significance level the results obtained from the Larssontest are in complete accord with those of Pedroni test Theimportant issue is that these results continue to holdregardless of what type of oil price measure one uses:average world oil price, Dubai oil, Brent oil or WTI oil price.Given the evidence that the two variables of interest arecointegrated, the author proceeds by estimatingcointegrating coefficients and applies between-dimension

panel fully modified OLS method Estimation results suggestthat for G7 countries a rise in real oil price depreciates realexchange rates in the long run Despite the fact that thefocus of this research is mainly on developed countries, one

is not prohibited from extrapolating the results on developingones, mainly those incorporating floating exchange rateregime, i.e to state that for this group of countries oil pricecan adequately capture permanent innovations in the realexchange rate in the long run as well At least the theoreticalmodel behind this research does not distinguish betweencountry types

The research, which is a good example of the second block ofpapers mentioned above, is the one conducted by Cologniand Manera (2008), where the economic effects of oil priceare estimated by means of a structural cointegrated VARwithin open macroeconomic model for G7 countries Theauthors pay particular attention to monetary variables andincorporate interest rate and money aggregate M1 into study

in order to further understand how the latter respond toexogenous oil price shocks as well to capture the interactionbetween real and monetary shocks in affecting economicbehavior The authors define two long-run equilibriumconditions, i.e cointegrating vectors, through conventionalmoney demand function and the relationship that equatesexcess output, as measured by the difference between actualand potential output levels, to inflation rate, exchange rate,interest rate and the price of oil The short-run dynamics of

the model is represented via six equations, describing moneymarket, domestic goods market equilibria, exchange ratemovements and oil price shock mechanism, which is assumed

to be self-generating and contemporaneously independent ofany other component in the system The authors’ findings arequite predictable and indicate that for majority of countriesunder study one standard deviation shock in oil price onaverage causes inflation level to increase In addition, outputlevel is negatively affected, though this effect is lagged innature On the monetary side, interest rate increases mainlydue to inflationary and real (output) types of shocks, which is

an indicator of a tightening in monetary policy Moreover,empirical results suggest that a shock in oil price does nothave any contemporaneous effect on the exchange rate Toclear the issue, this finding does not fully contradict the oneobtained by Chen (2007) due to different methodologies used

by the researchers, i.e Cologni and Manera consider thisproblem separately for each country as opposed to poolingthe data and implementing the analysis on panel level.Finally, innovation accounting is performed to numericallyassess the oil price pass-through into the variables understudy

Approach similar to the previous study, i.e VECM, has beenapplied by Rautava (2004) in the research on the role of oilprice and the real exchange rate fluctuations of rouble onRussian fiscal policy and economic performance The resultsobtained indicate that the Russian oil-exporting economy is

influenced significantly by fluctuations in the aforementionedvariables through both long-run equilibrium conditions andshort-run effects More precisely, the author reports that

‘over the long-run, a 10% permanent increase (decrease) ininternational price of oil is associated with a 2.2% growth(decline) in the level of Russian GDP’ (Rautava, 2004) Onemore worthwhile example concentrates on four large energyproducers and addresses the issue of the effects of oil priceshocks on real exchange rate, output and inflation level viaSVAR methodology (Korhonen and Mehrotra, 2009).Theoretical explanation of the empirical model is provided by

a dynamic open economy Mundell-Fleming-Dornbusch model,augmented with an oil price variable Using this approach, aset of over-identifying restrictions on the matrix of structuralinnovations is imposed The authors proceed in usual fashion

to estimate the model and obtain the results similar toRautava (2004) In addition, they find that oil price shocks donot account for a large share of movements in the realexchange rate, as measured by FEVD, although for somecountries under study they appear to be significant Thewhole thrill about this research is that in case of Venezuelalinearity tests proposed by Teräsvirta (2004) produce someevidence of nonlinear relationship between the output and oilprice series The authors suggest overcoming this obstaclevia estimation of a (logistic) smooth transition regression,which allows for explicit modeling of the asymmetricrelationship between the variables, i.e takes nonlinearity intoaccount

The whole problem of asymmetric relationship between theaforementioned variables was initiated after inability of the

1986 oil price plunge, mainly caused by preceding oil glutand unstable situation in Middle East, to produce aneconomic recovery as opposed to economic downturnprovoked by 1973 artificial oil price surge This phenomenonhas been empirically justified by Mork (1989), who showedthat if one were to extend the period under consideration byincluding data from 1986 oil price plunge, the oil price-macroeconomy relationship, as established by Hamilton(1983), would collapse In fact, Hamilton3 obtained apersistent negative correlation between oil price changes andGNP growth using the US data of 1948-1972, and claimedthat ‘oil shocks were a contributing factor in at least some ofthe US recessions prior 1972’ In principal, the conclusionreached turned out to be correct, but did not tell the wholestory The problem was that the study focused on the period

in which large oil price declines did not occur, and hence onecould not extrapolate its results in this kind of environment

It was not clear, if the correlation between the two variableswould persist and so the ability of oil price declines tostimulate the economic activity was questioned What Morkactually did, was that he modified Hamilton’s research byseparating oil price increases and decreases into different

3 In his paper, Hamilton mentions an interesting observation that ‘all but one of the U.S recessions since World War II have been preceded, typically with a lag

of around three-fourths of a year, by a dramatic increase in the price of crude petroleum’ (Hamilton, 1983).

variables and re-estimated the model This time estimationresults produced mixed evidence with real oil price increasesbeing negative and highly significant at each lag level, thussupporting Hamilton’s conclusion, as opposed to real oil pricedecreases, which turned out to be positive though small andonly marginally significant To confirm the appropriateness ofthe method used, the author implemented two types of tests,accounting for the stability of model’s asymmetricspecification as well as pairwise equality of oil pricecoefficients The tests were successfully passed and strongevidence in favor of asymmetry hypothesis was obtained Thisfinding has been empirically verified for a number ofindustrialized countries Interested reader is encouraged tocheck for Mork and Oslen (1994) for additional evidence

The asymmetry issue requires sound theoreticalargumentation as well as complicates the procedure ofconducting empirical studies by asking for more advancedeconometric techniques to be used As a consequence, it isworth mentioning the paper by Huang et al (Huang, Hwangand Peng, 2005), in which multivariate (two-regime)threshold model proposed by Tsay (1998) is exploited toinvestigate the relationship between oil price change, itsvolatility component (estimated via GARCH (1,1) model), andeconomic activity as measured by the level of industrialproduction and real stock returns The US, Canada and Japanmonthly data (1970-2002) are employed for this purpose As

a reference point, the authors heavily criticize the study by

Sadorsky (1999) for its inability to take into account thatcountries may exhibit different threshold values for an oilprice impact, i.e ‘the amount of price increase beyond which

an economic impact on production and stock prices ispalpable’ (Huang et al., 2005) Eliminating this drawbackconstitutes the core contribution of this article The authorsintentionally incorporate into study the aforementionedcountries, since in case of USA and Japan, which are netimporters of oil, the threshold value is expected to be muchlower (2.58% for both) if compared to Canada, net exporter

of oil (2.7%) At the first stage of the analysis, one-regimeVAR model, augmented with a dummy variable, whichreflects structural break date, as determined by the approachsuggested in Bai et al (1998), is estimated Failure to takethe latter into account, i.e splitting the sample into twosubgroups, may result in biased results One-regime VARapproach, though it provides theoretically justified estimates,encounters several drawbacks, among which low explanatorypower of oil price shocks, ‘the average-out problem emanatedfrom positive and negative changes in the price of oil’ andinability to account for different levels of oil dependencebetween countries The problem is solved via implementation

of multivariate threshold autoregressive model (MVTAR).This time the value of threshold is calculated via the gridsearch procedure proposed by Weise (1999); estimationresults confirm the presence of asymmetric relationshipbetween the variables, which is reflected in that ‘responses ofeconomic activity are rather limited in regime I, but become

much more noticeable in regime II, where oil price changeexceeds the threshold level’.

Another methodology has been realized by Lardic andMignon (2006) in the article, which studies the long-termrelationship between oil price and GDP time series using datafor G7 and Euro Area countries To account for existingasymmetry, the authors adopt the approach, developedamong others by Schorderet (2004), which is based onasymmetric cointegration Intuitively, the latter may bedetermined after one decomposes two integrated time seriesinto positive and negative partial sums, and furtherconstructs a linear combination, which is stationary This isexactly how the authors proceed in their study As a result,they manage to affirm asymmetry hypothesis for the majority

of countries Among its possible causes, ‘monetary policy,adjustment costs, adverse effect of uncertainty on theinvestment environment’ are mentioned

Aliyu (2009) investigates oil price shocks effect on themacroeconomic performance of Nigeria between 1980-2007via VAR model using different asymmetric transformationsfor oil price variable, among which Hamilton’s (1996) NOPI4

and Mork’s (1989) specification The latter survives a number

of post-estimation tests, such as Wald and block endogeneity(Granger causality), which support the significance of oil

4 NOPI=max{0, o(t)-max{o(t-1), o(t-2), o(t-3), o(t-4)}} Main focus is on oil price increases (Aliyu, 2009).

price coefficients in the model Moreover, case of Nigeria isinteresting, since on its example one may observe how theasymmetry effect is reflected in oil-exporting economy Thistime, ‘evidence is found of more significant positive effect ofoil price increase, than adverse effect of oil price decrease onreal GDP’ (Aliyu, 2009).

In general, economists come up with different explanations ofthe asymmetry phenomenon For example, Lee et al (Lee, Niand Ratti, 1995) conduct a research study, where they claimthat ‘an oil price change is likely to have greater impact onreal GNP in an environment where oil price movements havebeen stable, than in an environment where oil pricemovements have been frequent and erratic’ In other words,one should account for the variability of real oil pricemovements prior to assessing the relationship between thetwo variables The authors propose to augment the VARmodel of Hamilton (1983) and Mork (1989) with theunexpected oil price shock variable normalized by a measure

of oil price variability ‘This ratio can be thought of as being

an indicator of how different given oil price movement isfrom its prior pattern’ To construct this variable changes inreal oil price are assumed to be exogenous to any othervariable present in the model, and thus depend entirely on itsown lagged values with error term following GARCH (1,1)process The variable is then defined as the ratio of the

‘unexpected part of the rate of change in real oil price’ to thesquare root of conditional variance of the error term In other

words, given that certain value of the unexpected part(numerator) is calculated, its impact on real economicactivity will diminish the higher is the value of conditionalvariance (denominator), i.e it will be treated as a temporarychange To draw an analogy, one may think of conditionalvariance as something describing general environment ofsome geographical area, say the desert of Judea, and theunexpected part representing average number of rainy daysduring the year If the latter figure is trifling, it will not makeone change the common perception that desert is an aridplace On the other hand, assume that rainy weatherprevailed during the whole year, undoubtedly, one will bepuzzled with this observation and his/her perception may beaffected The authors provide the following economicjustification of their approach: ‘a rise in real oil price that islarge relative to the observed volatility will result inreallocation of resources and the lowering of aggregateoutput, but during periods of high volatility, since current oilprice contains little information about future, rational agentswill be reluctant to reallocate resources in the presence ofreal costs of doing so, thus aggregate output remainsunchanged’ (Lee et al., 1995) The empirical results supportthis line of reasoning Moreover, if one distinguishes betweenpositive and negative normalized oil price shocks the issue ofasymmetry arises: the coefficients on positive ones are allstrictly negative and highly significant, the coefficients onnegative ones are of different signs and insignificant Themodel proposed by Lee et al survives a number of

robustness checks, including the one proposed by Hamiltonthat ‘the relationship between the impulse response of realGNP growth obtained from a nonparametric kernel estimateand normalized unexpected oil price shock be examined’.

Addressing the issue of oil price uncertainty, I would like tobriefly overview the research by Elder and Serletis (2009) Inthis paper the authors investigate the effects of oil priceuncertainty in Canada via two-variable (industrial output andoil price) structural VAR model, assuming, as in the previousstudy, that error terms are heterosc(k)edastic and followGARCH (1,1) process The proposed measure of uncertainty

is a conditional standard deviation of oil, i.e ‘a standarddeviation of the one-month ahead forecast for oil price,obtained from the multivariate variance function, in whichthe volatility of industrial production and the volatility of oilprice depend on their own lagged squared errors and laggedconditional variance’ (Elder, Serletis, 2009) The VAR model

is constructed in such a way that oil price uncertainty, which

is treated as an exogenous variable in the system, enters theequation for industrial production only The model isestimated by maximum likelihood (joint maximization overconditional mean and variance parameters) The estimationresults say that the coefficient in front of oil price uncertaintymeasure is negative and statistically significant, proving thatoil price volatility has tended to depress industrial output inCanada during the considered period Moreover, this termbrings about asymmetry, in the sense that ‘unanticipated oil

shocks, whether positive or negative, will tend to increasethe conditional standard deviation of oil, which will tend todepress output growth’ (Elder, Serletis, 2009) In otherwords, abrupt oil price declines may lead to contraction ofoutput due to increased uncertainty Finally, the relevance aswell as explanatory power of the uncertainty measure isrevealed after implementation of innovation accounting Inparticular, impulse-response analysis indicates that the lattervariable ‘strengthens negative response of output to oil priceshock’ Though this study considers the problem ofasymmetry from slightly different angle, the results obtainedare in complete accord with those of Lee et al (1995) andMork (1989).

Thus far the studies, which mainly used relativelyhomogeneous econometric techniques to account forasymmetric relationship between oil price shocks and outputlevel, were considered Another way to think aboutasymmetry is to explicitly assume that the relationshipbetween the two variables is nonlinear This issue has beenthoroughly investigated by Hamilton (1999), where hedeveloped a new framework for determining whether a givenrelationship is nonlinear, what the nonlinearity looks like,and whether it is adequately described by a particular model.What is unusual about the proposed approach is that at thefirst stage the nonlinear part of the regression equation is notdefined explicitly, remains unknown and is treated as theoutcome of a random process In the further research,

Hamilton applies the methodology he proposes to UShistorical data, and estimates the relation between oil priceand GDP growth (see Hamilton (2003)) The results, heobtains, prove the appropriateness of the method used, aswell as the nonlinearity hypothesis Just to mention that thesame methodology has been used by Zhang in his research

on Japan, where he shows that once a nonlinear asymmetriceffect is accounted for, a considerably larger coefficient onthe oil shocks can generally be obtained, providing evidence

of misspecification of a simple linear regression model (D.Zhang, 2008)

Alternatively, it should be mentioned that not all economistsbelieve in the existence of asymmetric (nonlinear) oil priceshock effect In particular Tatom (1988) blames monetarypolicy for the asymmetric response of aggregate economicactivity to oil price shocks, implying that in fact economyresponds symmetrically

As it was already mentioned, there exist some studies, whichinvestigate the impact of oil price fluctuations on Ukrainianeconomy One example is a research carried out byMyronovych (2002) The author adopts methodologyproposed by Gisser and Goodwin (1983) and estimates threeseparate St Louis-type equations5, where oil pricesimultaneously with monetary and fiscal policy variables

5 St Louis-type equations describe the impact monetary and fiscal policies have

on economic activity (Cologni, Manera, 2008)

influence real GDP, inflation and unemployment Theestimation method used is error correction mechanism(ECM) with Newey-West standard errors, which are used forthe purpose to eliminate serial correlation andheterosc(k)edasticity problems Clear dependence betweenoil price fluctuations and the first two macroeconomicindicators (GDP and inflation) is found, though the impact onunemployment level is not statistically significant at any laglength Moreover, the study reports that in case of GDPmonetary policy has the largest positive counterbalancingeffect among all the regressors, though it is lagged in nature.This finding is also supported by the observation made inPeersman and Robays (2009) Finally, Myronovych reportsthat one percent increase in the growth rate of real oil pricewill likely decrease next quarter GDP growth rate by 0.126percents and its effect is still increasing after In addition,contemporaneous increase in the quarterly growth rate ofinflation by 0.27% will also take place.

To conclude the literature review section and to justify thechoice of macroeconomic variables chosen for this research, Iconsider the channels through which oil price fluctuationsmay affect economy of a given country Following the existingliterature, I refer to those channels as oil transmissionmechanism According to Peersman and Robays (2009) oilprice increases are accompanied in the first place by a rise inconsumer price index, which conventionally includes energygoods such as petroleum, heating fuels etc This effect is

known as direct or short-run effect and its magnitudedepends on the weights assigned to energy goods inaggregate CPI Direct effect is assumed to be rapid andcomplete, though the question of completeness may bedistorted in case of high level of competition in retail energysector Indirect effect comes next and is more persistent andconsiderably larger in magnitude It captures long-runincreases in CPI, which on purpose excludes energy prices.Indirect effect is more relevant for policy makers, becausemonetary policy, whose effect is delayed in nature, is pursuedwith the reference to aggregate CPI changes This effect may

be decomposed into cost effect, second-round and demandeffects The intuition behind the cost effect, which ismeasured by changes in import deflator, is that higher price

of oil inevitably pulls production costs and terminal goods’prices up, finally resulting in increased CPI Second-roundeffect occurs since rising CPI is associated with decreasingpurchasing power of money, hence employees become worseoff and have an incentive to demand higher nominal wages torestore their real income This is possible to accomplishthrough wage indexation mechanism As a result, firm’s costsare subject to even further increase The firms respond to itvia additional increase in prices, which again will lead to anincrease in CPI, and so on and so forth It turns out thatsecond-round effect is cyclical and may result in even higherinflation level As Peersman and Robays (2009) mentioncorrectly: ‘the existence of second-round effects will depend

on supply and demand conditions in the wage-negotiation

process and the reaction of inflation expectations’ Turing todemand effect, one is supposed to remember theconventional textbook supply-demand graphs Exogenous oilprice shock shifts supply curve upward along aggregatedemand curve This results in an increase in the price leveland decrease in output The more elastic the demand curve

is, the lower the impact of oil price shock on the price levelwill be Moreover, for the country, which is a net oil importer,oil price increases are accompanied by the downward shift inthe aggregate demand curve, which reflects reducedcomposition of demand and even additional drop in output.The latter is associated with the following sub-channels:precautionary savings, uncertainty and monetary policyeffect As regards monetary policy effect, its legitimacy issupported by the results of Leduc and Sill (2001) study,where the authors construct a DSGE monetary model withinmonopolistic competition framework, and find out that ‘easyinflation policies are seen to amplify the impacts of oil priceshocks on output and inflation’ (Leduc, Sill, 2001) Thisconclusion is supported empirically in the study by Hamiltonand Herrara (2000)

Finally, oil price increases are also expected to negativelyaffect country’s terms of trade through negative currentaccount As a result, the Central bank will not be able tointervene into foreign exchange market endlessly to meetdemand, and will have to let the exchange rate to depreciate

C h a p t e r 3

METHODOLOGY

As opposed to the research conducted by Myronovych (2002),

I am inclined to estimate the system of equations viaSVAR/VECM framework, which allows for contemporaneousand lagged interconnection between the variables of interest

to exist, thus should provide more qualitative estimates Allthe series are considered in levels as opposed to growthrates The advantage of using SVAR/VECM approach is thatboth enable a researcher to perform innovation accounting(IRF and FEVD), i.e to numerically assess how one standarddeviation shock in the error term of oil price variable willaffect a set of endogenous variables included in the model aswell as determine what proportion of the forecast errorvariance of a particular variable is due to oil price shock

Following the theory, we choose between the following twoclosely related types of models: SVAR and VECM Decisioncriterion is order of integration of the series at hand and thepresence of long-run equilibria, which are supposed to bestationary6 To decide between the two options available,stationarity properties of the data are examined first Phillips-

6 Stochastic process, which has finite mean and variance (1 st and 2 nd moments, respectively) is called covariance stationary Additional condition is the dependence of covariance between two time periods on the distance or lag between those periods and not on the actual time at which the covariance is computed (D N Gujarati)

Perron (PP) and Augmented Dickey-Fuller (ADF) tests areemployed for this purpose The two are almost identical withthe only difference being the approach used to tackle aproblem of serial correlation in the residuals: the former test(PP) uses Newey-West standard errors while the latteraugments the core equation with lagged values of thedependent variable

In case the variables are found to be I(0), SVAR model ispreferred and estimated But if the variables turn out to beI(1), we may not directly proceed by estimating SVAR indifferences, since there may be a linear combination betweenthe variables, which is stationary In this case, the variablesare referred to as CI(1,1) and the coefficients in front of themconstitute cointegrating vector, which in turn determineslong-run equilibrium relationship7 Cointegrating vector isnot unique and in general if there are n variables in themodel, integrated of the same order, one may expect to get

at most n-1 stationary combinations, which constitutecointegrating rank If cointegration is detected, it should beincorporated into the model, since its omission entailsmisspecification error As Enders (2004) points out: ‘aprincipal feature of cointegrated variables is that their timepaths are influenced by the extent of any deviation from long-run equilibrium After all, if the system is to return to thelong-run equilibrium, the movements of at least some of thevariables must respond to the magnitude of the

7 Be reminded that for economic theorists and econometricians the meaning of the word ‘equilibrium’ is different In particular, the latter group understands it

as ‘any long-run relationship among nonstationary variables’ (Enders, 2004)

disequilibrium’ In other words, system’s short-run dynamics

is determined, among others, by its steady state, and hencethe latter should be incorporated into the model exogenously

We proceed by checking for the existence of cointegration(long-run equilibria) employing Johansen procedure8 (in

i trace r T

1

)ˆ1ln(

)

) ˆ 1 ln(

It is worth of emphasizing that one may obtain VECMrepresentation of the conventional SVAR model using thefollowing transformation (in fact, another name for VECM is

Y   (reduced form VAR),

For simplicity assume that the number of lags equals three aswell asQA1BI:

t t t

t

Y    1 1  2 2  3 3   ,

t t t

t

Y    1 1  ( 2  3 ) 2  3  2   ,

t t t

t t t

i

n i j j t

n i i

n i j j

П , (,  are adjustmentand cointegration matrices respectively9)

Irrespectively of model type estimated, the order of variablesassumed throughout the research is as follows:

 t t t t t tT

Y  1 The order of variables isimportant mainly for innovation accounting In case of SVARthe matrix of contemporaneous effects is defined in fashionsimilar to Cologni and Manera (2008) by A SVAR and in case ofVECM conventional Cholesky restrictions,A CHOLETSKY, areimposed, i.e its elements satisfy the following the followingcondition: a  A CHOLETSKY i j

0 1

0 0 1

0 0 0 1

0 0 0 0 1

0 0 0 0 0 1

0 1

0 0 1

0 0 0 1

0 0 0 0 1

0 0 0 0 0 1

CHOLETSKY

A

9 Number of long-run equilibria is exactly the rank of matrixП

According to the information filed in A SVAR, the variables,which have contemporaneous effect on the demand formoney balances (M1)d are assumed to be interest rate, realGDP and inflation rate; interest rate dynamics is describedvia real GDP, inflation level, exchange rate and oil pricevariables; real GDP is dependent upon inflation level,exchange rate and oil price changes; inflation level isinfluenced by exchange rate and oil price movements;exchange rate is directly influenced by oil price shocks;finally, oil price is considered as contemporaneouslyexogenous to any other variable in the system As regards

CHOLETSKY

A , it is used in case of VECM estimation, since due

to software limitations we are not able to manually definecontemporaneous coefficients’ matrix But in fact, thedifference between the two is not that crucial and constitutesonly two additional variables added to explain demand formoney balances, (M1)d=(M1)s in equilibrium In case of A SVAR

we have (M1)d advocated by Keynes’ liquidity preferencetheory, M 1 f(i,rgdp,cpi), and in case of A CHOLETSKY we havemore extended version of the former, M 1 g(i,rgdp,cpi, fx,oil).The logic behind adding extra variables in the demand formoney equation is the call to control for foreign influencegiven that Ukraine is a small, open, highly dollarizedeconomy Both matrices are theoretically elegant and hence,the difference between the two will not crucially underminefinal estimations Estimation method applied to both SVARand VECM is OLS, which said to be efficient one After anymodel is estimated, a number of tests are performed to check