Master thesis, 15 ECTS
Master’s Programme in Economics/ Master thesis (II), 15 Credits
Oil Prices and Consumer Spending in Sweden
- Do Oil Price shocks affect Private Consumption?
Jesper Byström
Abstract
This thesis uses a Vector Autoregressive Model and an Impulse response function to examine the impact of rising oil prices on personal consumption expenditure in Sweden. In Sweden the use of fossil oil has been declining for several years, and by 2045, Sweden aims to have no net emissions of greenhouse gases into the atmosphere, although, there is still a lot of oil left in the energy system, not least in the transport sector. As Sweden has continued to reduce its oil dependency and carbon dioxide emissions, it is interesting to investigate whether the international oil price has an impact on the consumption in Swedish households.
In this thesis, oil price increases are found to have a negative impact on personal consumption expenditure.
The result from this thesis could be an important implication for policymakers deciding about laws and subsidies for renewable energies when facing a trade-off between the environment and private consumption.
Key words: Oil price shocks, Vector Autoregressive Model, Household consumption, Sweden.
Table of content
1. Introduction: ... 1
2. Theoretical framework and earlier research ... 4
2.1 Literature review: ... 4
2.2 Theoretical model: ... 6
2.3 Transmission Mechanism Channels ... 8
2.3.1 The real balance theory: ... 8
2.3.2 The monetary policy channel ... 8
2.3.3 Wealth transfer effect: ... 9
2.4 Hypothesis ... 9
3. Empirical Approach ... 10
3.1 Data ... 10
4. Statistical approach ... 12
4.1 Augmented dickey fuller test ... 12
4.2 Johansen test of cointegration ... 13
4.3 Vector autoregression model ... 14
4.4 Lag order selection model ... 14
4.4.1 Bayes information criterion (BIC) ... 15
4.4.2 Akaike information criterion (AIC) ... 15
4.5 Lagrange-multiplier test for autocorrelation ... 15
4.6 Granger causality test ... 16
4.7 Impulse response function ... 17
5. Results ... 18
5.1 Augmented Dickey-fuller test ... 18
5.2 Johansen test of cointegration ... 18
5.3 Vector autoregressive model ... 18
5.4 Granger-causality test ... 19
5.5 Impulse response function ... 20
6. Discussion ... 21
6.1 Drawbacks ... 22
7. Conclusion ... 24
7.1 Future research ... 24
8. Reference list ... 25
9. Appendix ... 27
1. Introduction:
Due to the fact that, Sweden has consistently attempted to reduce its dependence on fossil fuels and aims to have no net emissions of greenhouse gases into the atmosphere, by 2045.
This study will investigate if a relationship exists between oil price and household consumption.
If a relationship exists, this study will also try to investigate the effect of a shock in the oil price, on the private consumption. The purpose of this thesis is to analyze how oil price affects private consumption and provide policymakers with useful information, when they are deciding about laws and subsidies for renewable energies when facing a trade-off between the environment and private consumption.
Since the end of world war II, many researches have suggested that oil price fluctuations have severe consequences for macroeconomic activities. The effect of oil price increase typically reduced the world demand for goods and services since oil-intensive production becomes more expensive due to higher production cost for the firms (Hamilton 1983). The sequence of Hamilton’s studies opened the door for future studies in the field. These studies based on data on oil-importing countries suggested that oil shocks have a negative consequent on different economic activities (See Mork (1989), Jiménez-Rodrıguez & Sanchez (2005) and Hamilton (2003)).
As mention above, oil price shocks can have direct impact on the macroeconomics indicators such as export/imports, interest rate, exchange rates. Therefore, the impact on the economic growth can be measure. Mehra and Peterson (2005), investigate and analyze the oil shocks and the impact on household consumption in the U.S. The article finds empirical evidence indicating that oil price increases have a negative effect on consumer spending, where oil price declines did not have any significant effect on consumption.
A number of empirical studies have shown that one of the key mechanisms by which oil price increase affect aggregate output is through the consumption channel (Mehra and Petersen (2005), Hamilton (2003) and Odusami (2010).
Even though oil dependency among many countries have decreased since the late 80s, oil is still an important component for many countries, and consumption contributes for a large share of total GDP of an economy. For example, in Sweden approximately 50% of the GDP in the past five decades is related to household consumption (See appendix A).
This thesis will focus on the Swedish economy. The final use of energy in 2018 amounted to 373 TWh, of which 84 TWh was from fossil oil. The use of fossil oil has been declining for several years, although there is still a lot of oil left in the energy system, not least in the transport sector.
Graph 1, Final use of fossil oil products. Source: The Swedish energy agency
Historically, the decline in the use of oil has primarily occurred in industry and housing, as the use of oil has been replaced by other types of energy. In the housing sector, the trend of reduced oil consumption has continued in recent years, but at a declining rate. In the industrial sector however, the historical decline has ceased in recent years. Although biofuels have risen sharply in recent years, the transport sector is still dominated by fossil fuels (The Swedish energy agency).
Simultaneously, Sweden has consistently attempted to reduce its dependence on fossil fuels.
By 2045, Sweden aims to have no net emissions of greenhouse gases into the atmosphere. The Swedish government has introduced a number of measures to reduce emissions. Among other things, a bonus-malus system is introduced with a bonus for vehicles with low carbon dioxide emissions and higher taxes for those with high carbon dioxide emissions. In addition, the
0 20 40 60 80 100 120 140 160 180
198319851987198919911993199519971999200120032005200720092011201320152017 TWh
Final use of fossil oil products
Industry Transport Housing, service etc
government has increased the appropriations for rail maintenance and introduced a tax on air travel as well as invested more resources to promote cycling and has given authorities the task to have travel-free meetings to a greater extent (Government Offices of Sweden, 2018).
Sweden also aims that the emissions from domestic transport, in addition to domestic flights, will be reduced by at least 70 % by 2030 compared with 2010 (Swedish Environmental Protection Agency).
Since climate change is one of the most pressing challenges facing the international community.
A broad range of policy instruments can be used to curb carbon emissions, and economic instruments such as taxes and emissions trading are critical elements of any comprehensive mitigation strategy. Energy sources were first taxed in Sweden in the 1920s. A carbon tax was instituted in 1991, alongside an already existing energy tax, and it remains a cornerstone of Swedish climate policy. Over time, the carbon tax has increased in importance, contributing to a broad range of environmental and climate objectives. For example, the carbon tax provides incentives to reduce energy consumption, improve energy efficiency and increase the use of renewable energy alternatives (Government Offices of Sweden, 2020)
By increasing the tax level gradually and in a stepwise manner, households and businesses have been given time to adapt, which has improved the political feasibility of tax increases.
One example that can give a simple overview of how much the taxes are for the private consumer. The average price of gasoline at a pump station in April in 2020 was SEK 12,64 / liter, of which the tax share of the price was 72,9 percent (SPBI, 2020).
As oil consumption has decreased in Sweden, there is an interest in investigating whether oil prices affect private consumption. By adding the energy tax and carbon dioxide tax to the oil price, we can get an overview of how the total price affects consumption.
2. Theoretical framework and earlier research
2.1 Literature review:
Since the end of world war II, many researches have suggested that oil price fluctuations have severe consequences for macroeconomics activities. The sequence of Hamilton’s (1983) studies opened the door for future studies in the field. These studies based on data on oil-importing countries suggested that oil shocks have a negative consequent on different economic activities.
Mork (1989) developed Hamilton’s (1983) study. Hamilton (1983) did only examined periods in which all the large oil price movements were upward, therefore it left unanswered the question whether the correlation persists in periods of price decline. Mork (1989) investigate the effect of a price increase and decrease. The result show strongly a large negative effect of an oil price increase, however the coefficients for a price decline, was small and varying signs.
The coefficients were not significant, and the conclusion were that the data did not identify any significance effects of oil price declines.
Jiménez-Rodríguez & Sánchez (2005) investigate the effects of oil price shocks on the real GDP growth for nine countries. In particular, they fund that oil prices increase did have a larger impact on real GDP growth than that of oil price declines, with the latter being statistically insignificant in most cases. In the cases of oil importing countries, an oil price increase did have a negative impact on economic activity, in all cases except Japan.
Hamilton (2009) explore similarities and differences between the increased oil prices in 2007- 2008 and earlier oil price shocks, by looking at what caused the price increase and what the effect was on the economy. Historical oil price shocks were primary caused by physical distribution of supply, although the oil price increased in 2007-2008 was caused by a strong demand confronting stagnating world production. However, even though the causes were different, the consequences for the economy appeared to have been very similar to those observed in earlier crises.
As presented above, these results mainly focused on general macroeconomic activities, although Mehra and Peterson (2005) did take a different path and analyze the effect of oil shocks on the household consumption in the US. In their study they develop a model based on the life-cycle model of consumption (Modigliani and Brumberg 1954). Their study is based on
data from the US, and includes the variables income, wealth, oil price and interest rate. They find evidence that oil price shocks have an impact on private consumption in the short-run, but not in the long-run.
Kilian (2008) investigate the economic effects of energy price shocks, he investigates how consumer expenditures responds to a higher energy price. He estimates the energy-price elasticities of energy demand, and the results are expressed as elasticity with respect to the price of energy, evaluated at the average energy share. The elasticity estimate for total energy consumption was -0,45, however there are important differences across different forms of energy. The strongest response was observed for Heating Oil and Coal, and it could be observed that electricity and natural gas did have the weakest response. The point estimate is only -0,15, after one year, for electricity price elasticity of electricity demand. The larger response of heating oil and coal is likely due to household’s ability to store heating oil in tanks. This allows the household to purchase heating oil when the prices are low and delay the purchase of heating oil when the price is high. Kilian (2008) also estimate the energy-price elasticities of non-energy consumption. The overall elasticity for total consumption was -0,15. The effect is mainly driven by a reduction in vehicles purchases.
Wang (2012) use a logistic smooth transition model to examine the impact of rising oil prices on personal consumption expenditures in open and industrialized economies. In the face of uncertainty regarding future oil prices, consumers initially rationally postpone spending. In particular, the effects of rising oil prices on personal consumption expenditures are greater than those of falling oil price
Zhang and Broadstock (2014) extend the work done by Mehra and Peterson (2005) and investigate the impact of oil shocks on consumption expenditure in ASEAN and East Asia economies, but excluded interest rates in their model. They also find a negative relationship between oil price shocks and consumption.
Zaman (2019) investigate whether international oil price change has any impact on consumer spending for four countries, Canada, Germany, the UK and US. She finds evidence that international oil price shocks have a significant impact on consumer spending.
2.2 Theoretical model:
The theoretical approach that will be used in this thesis will follow the theoretical approach used by Mehra and Peterson (2005). The empirical framework begins with a general/standard macroeconomic specification of (per capita) household consumption, where the level of consumption in an economy, 𝐶!, is affected by the existing level of wealth 𝑊!, as well as current and discounted expected future income, 𝑌! and 𝐸(𝑌!"#), respectively, where 𝑖 = 1, … , ∞.
According to the Life-cycle Hypothesis (LCH), it is assumed that the consumer maximizes her/his consumption utility subject to her/his total lifetime earnings. Also, it is assumed that a consumer faces a budget constraint during the lifetime, and the budget constraints is derived from the theoretical household budget constraint:
𝑊!"$= (1 + 𝑟!)(𝑌!+ 𝑊!− 𝐶!) (1)
Where 𝑊!"$ is the wealth of the next period, and it equal to the discounted value of today’s income and wealth minus the consumption done by today. We also impose a condition that
#→&lim3($"))'!"#$4 = 0, and we assume that 𝑟! = 𝑟!"$= 𝑟, then, by repeated substitution of the budget constraint, current-period wealth is obtained as:
𝑊! = ∑&#,-($"))+!"$$− ∑&#,-($")).!"$$ (2)
According to Hall (1978), consumption follows a martingale process, gives that 𝐸(𝐶!"$) = 𝐶!. Then, taking the expectations of equation (2):
𝐶! = $")) ∑#,-& /(.($"))!"$$)+$")) 𝑊! (3)
We also assume that income have a constant growth rate of real income, g, then E(𝑌!"#) =
$"0
. ! + 𝜂!"$ , where 𝜂!"$ is a white noise process. Thus:
𝐶! = )20) 𝑌!+$")) 𝑊!+ ∑&#,$($"))3!"#$ (4)
The derivation to this point establishes that a long-run relationship exists between consumption, income and wealth. Mehra and Peterson (2005) refer this as the planned level of consumption, 𝐶!4, and it can be expressed in a similar form by first taking expectations of the error term and adding a constant term, leading to the estimated long-run relationship:
𝐶!4= 𝜔-+ 𝜔$𝑌!+ 𝜔5𝑊! (5)
Where 𝜔$ = )
)20 and 𝜔5 = )
$"), and the expected value of the error term in equation 5 is zero.
However, actual consumption differs from planned consumption for a multitude of reasons.
Campell and Mankinw (1989) show that the short-run dynamics of consumption can be conveniently written in the form of an error correction model:
∆𝐶! = 𝛼-+ 𝛼$(𝐶!2$4 − 𝐶!2$) + 𝛼5∆𝐶!2$4 + ∑87,$𝛼67∆𝐶!27+ 𝜇! (6)
By substituting equation (5) to (6):
∆𝐶!= 𝛼"+ 𝛼#(𝜔"+ 𝜔#𝑌!$#+ 𝜔%𝑊!$#− 𝐶!$#) + 𝛼%∆(𝜔"+ 𝜔#𝑌!$#+ 𝜔%𝑊!$#) +
∑(')#𝛼&'∆𝐶!$'+ 𝜇! (7)
Assuming that future income grows constantly relative to the current level, and also that the consumers have rational expectations, the expected value of accumulated and discounted future income streams is proportional to the current income. The model can be simplified to:
∆𝐶!= 𝛽"+ 𝛽#(𝐶𝑡𝑝− 𝐶!$#) + 𝛽%∆𝑌!$#+ 𝛽&∆𝑊!$#+ ∑(')#𝛽&'∆𝐶!$'+ 𝜇! (8)
Where:
𝐶!4= 𝜔-+ 𝜔$𝑌!+ 𝜔5𝑊! (9)
The model will also include oil and the short-term interest rate as exogenous variables in equation (8) and the finale model becomes:
∆𝐶! = 𝛽-+ 𝛽$>𝐶!4− 𝐶!2$? + 𝛽5∆𝑌!2$+ 𝛽6∆𝑊!2$+ ∑87,$𝛽;7∆𝐶!27 +
∑87,$𝛽<7∆𝑂𝐼𝐿!27 + ∑87,$𝛽=7∆𝐼𝑅!27+ 𝜇! (10)
And equation (9) becomes:
𝐶!4 = 𝜔-+ 𝜔$𝑌!+ 𝜔5𝑊!+ 𝜔6𝑂𝐼𝐿!+ 𝜔;𝐼𝑅! (11)
Equations (10) and (11) will be the main equations for the empirical analysis.
2.3 Transmission Mechanism Channels
In this section, the theory on how oil price increases may affect the real economy will be discussed. Therefore, the question is, how do oil prices, in theory, affect the macroeconomy?
A simple answer it that previous research does not offer any dominant theoretical mechanism (See Hoker (2002), Hamilton (2003), Wang (2012) and Odusami (2010).
2.3.1 The real balance theory:
The real-balance channel posits that oil price increase lead to inflation, lowering real money balances held by households in the economy and thereby depressing aggregate demand trough familiar monetary channels. Since a higher oil price leads to higher gasoline prices, which leads to a reduction in disposable income, since it is assumed that a consumer spends a certain proportion of their income on gasoline.
2.3.2 The monetary policy channel
The monetary policy channel becomes operative if the national bank tightens policy in response to inflation induced by oil prices, which may exacerbate further the negative output effect associated with oil shocks.
2.3.3 Wealth transfer effect:
Since Sweden is an oil-importing country, it is reasonable to expect a reduction in consumption as a result of an oil price increase, due to the transfer of income from oil-importing countries to oil-exporting countries. The main reason for this is that when a transfer of income from an oil- importing country to an oil-exporting country occurs, the disposable income in the oil-exporting country increases but falls in the oil-importing country, which in turn causes negative effects on the consumption of the oil importer. However, in relation to this effect, the wealth transfer effect requires a longer time to influence consumption, since it is assumed that it will take longer for wages to be adjusted compared to prices.
2.4 Hypothesis
To answer the question presented in section 1.1, the following hypothesis will be tested:
H-,$: There is no relationship between oil price shocks and consumption expenditure H?,$: There is a relationship between oil price shocks and consumption expenditure
H-,5: There is no Granger causality from oil price to consumer expenditure H?,5: There is a Granger causality from oil price to consumer expenditure
3. Empirical Approach
3.1 Data
The variables that will be used in the model is: Private final consumption of Households, Net financial assets of Household, Disposable income of the household sector, Short term interest rates and Brent oil spot prices.
The data for Private consumption, disposable income and the financial asset has been collected from SCB (2020). The short-term interest rate has been collected from the Swedish Riksbank and the price of Brent oil is collected from the Federal Reserve Bank of St Lois.
The data is collected on a quarterly level between the first quarter of 1987 to the last quarter of 2019. Private final consumption of Households, Net financial assets of Household, Disposable income of the household sector and the oil price have all been adjusted for inflation with the Swedish CPI index (OECD), to reflect inflation change. The Brent oil spot price have been converted to Swedish SEK with the exchange rate level from the Federal Reserve Bank (2020) and an energy tax and a carbon tax have been added to the price, from The Swedish Tax Agency (2020).
The variables, consumption, net financial wealth and disposable income are divided with total population level, since the model wants to capture the effect on the consumer level.
Furthermore, all variables except short-term interest rate are transformed into its natural logarithm. This transformation of these variables is done since the model wants to capture the relative change, and it is in line with previous research (see Mehra and Peterson 2005 and Kilian 2008).
Variable Abbreviation Explanation Private final consumption
expenditure of the household
dCONS_log Total private consumption
expenditure, SEK/capita.
Logarithm values Net financial wealth for the
Household
dINC_log The net financial wealth of household sector, asset minus liabilities, SEK/capita.
Logarithm values Disposable income for the
household sector
dWEALTH_log Net disposable income for the household sector – SEK/capita. Logarithm values
Short term interest rate dIR 3 months short term money market interest rate in Sweden
Brent oil spot price dOIL_log Brent crude oil spot price in SEK and an added carbon- and energy tax, logarithm values
Table 1, data description
Sources: SCB, Swedish Riskbank, Federal reserve bank of St.Louis.
4. Statistical approach
The model that will be used to analyze the relationship between oil price and private consumption is the Vector Autoregression model (VAR), a Granger causality test and an impulse response function.
The choice of using a VAR-model and not a VECM-model is based on the result from a Johansen test of cointegration. The test is done to investigate if there is cointegration in the theoretical model, which suggest that there is a long-run relationship between the variables in the model. The test will be presented more in detail in section 4.2 Hence, if there is no cointegration, a VAR-model is suitable, to investigate the short-run relationship. (Stock and Watson 2015)
A VAR model was also used by Jimenez and Sanchez (2005), Blanchard Gali (2007) and Zaman (2019). Even though, Zhang and Broadstock (2014) and Mehra and Peterson (2005) did use a VECM model, they did not find any long-term relationship between oil and consumption and could therefore only conclude that there exists a short-term relationship
4.1 Augmented dickey fuller test
If the future is like the past, then these historical relationships can be used to forecast the future.
However, if the future differs fundamentally from the past, then those historical relationship might not be reliable guides in the future. Therefore, a key concept of time series regression is that the data is stationary (Stock and Watson, 2015).
A time series 𝑌! is stationary if its probability distribution does not change over time, on other words, if the joint distribution of (𝑌!) does not depend on s regardless of the values of T, otherwise 𝑌! is said to be nonstationary.
The augmented Dickey-Fuller (ADF) test for a unit root test the null hypothesis Ho: 𝛾 = 0 , against alternative H1: 𝛾 ≠ 0 in the regression:
Δ𝑌! = 𝜇 + 𝛾𝑌!2$+ ∑@#,$𝛽#Δ𝑌!2#+ 𝜀! (12)
The lag length will be decided by one method presented by Ng and Perron (1995). The method
number of lags. If the t-statistic for testing H0: 𝛽8∗ = is greater than 1,6 then set k=K* and perform the unit root test. However, if the t-statistic is smaller than 1,6 then set k*=k*-1 and repeat the process. A common rule of thumb for determining k* suggested by Schwert (1989):
𝑘∗ = 𝑖𝑛𝑡 M12 ∗ 3 B
$--4
#
%P (13)
4.2 Johansen test of cointegration
The Johansen procedure consist of two tests, the trace-test which produce a trace eigenvalue statistic and the other test is the Maximum eigenvalue test which produce a max-statistic. Let r be the rank of Π, and this is the same as the number of cointegrated vectors. For both of the test statistics, the initial Johansen test is test of the null hypothesis of no cointegration against the alternative of cointegration. The tests differ in terms of the alternative hypothesis (Verbeek, 2008).
Maximum eigenvalue test:
The maximum eigenvalue test examines whether the largest eigenvalues is zero relative to the alternative that the next largest eigenvalue is zero. The null hypothesis is that 𝑟𝑎𝑛𝑘 (Π) = 0 and the alternative hypothesis is that rank 𝑟𝑎𝑛𝑘 (Π) = 1. If the rank of the matrix is zero, the largest eigenvalue is zero and it could be concluded that there is no cointegration and the tests are done. The test of the maximum (remaining) eigenvalue is a likelihood ratio test. The test statistic is:
𝜆CDE(𝑟-) = −𝑇 log(1 − 𝜆W)&"$) (14)
Where 𝜆CDE(𝑟-) is the likelihood ratio test statistic for testing whether 𝑟𝑎𝑛𝑘 (Π) = 𝑟- versus the alternative hypothesis that 𝑟𝑎𝑛𝑘 (Π) = 𝑟-+ 1.
Trace test
The trace test is a test whether the rank of the matrix Π is 𝑟-. The null hypothesis is that 𝑟𝑎𝑛𝑘 (Π) = 𝑟- and the alternative hypothesis is that 𝑟- < 𝑟𝑎𝑛𝑘 (Π) ≤ 𝑛, where n is the maximus number of possible cointegrating vectors.
𝜆!)DFG(𝑟-) = −𝑇 ∑8H,)&"$ln (1 −𝜆WH) (15)
For example, the hypothesis that rank 𝑟𝑎𝑛𝑘 (Π) = 0 versus the alternative that 𝑟𝑎𝑛𝑘 (Π) ≤ 𝑛.
If there exist cointegration among the variables in the model, there can be concluded that there exist a long-term relationship between the variables. Although, if no cointegration can be observed among the variables, there can be concluded that there exists no long-term relationship between the variables in the given model (Verbeek, 2008).
4.3 Vector autoregression model
To investigate if a relationship exists between the oil price and the household consumption in Sweden, a vector autoregression (VAR) test will be done. The VAR model is an extension from the univariate AR model which only regress one variable time series while the VAR model will list all variables of the time series. The VAR in this study will have five variables, C, O, W, D, IR and consist of five equations (Stock and Watson, 2015).
The standardized VAR-model over p periods of time will in this study become:
𝐶! 𝐷! 𝑊! 𝐼𝑅! 𝑂!
=
⎝
⎜⎛ 𝛾$ 𝛾5 𝛾6 𝛾; 𝛾<⎠
⎟⎞ +
⎝
⎜⎛
𝛽$$ 𝛽$5 𝛽5$ 𝛽55
𝛽$6 𝛽$; 𝛽$<
𝛽56 𝛽5; 𝛽5<
𝛽6$ 𝛽65 𝛽;$
𝛽<$
𝛽;5 𝛽<5
𝛽66 𝛽6; 𝛽6<
𝛽;6 𝛽<6
𝛽;; 𝛽;<
𝛽<; 𝛽<<⎠
⎟⎞
⎝
⎜⎜
⎛ 𝐶!24 𝐷!24 𝑊!24 𝐼𝑅!24
𝑂!24⎠
⎟⎟
⎞+
⎝
⎜⎛ 𝜀$!
𝜀5!
𝜀6!
𝜀;!
𝜀<!⎠
⎟⎞
(16)
4.4 Lag order selection model
It is important to decide the number of lags int the VAR model. Choosing the right lags of a VAR requires balancing the marginal benefit of including more lags, against the marginal benefit of having fewer lags. On the one hand, if the number of lags in an estimated autoregression is too low, the model will omit potentially valuable information contained int the more distance lags. On the other hand, if the number of lags is too high, the model will include to mane coefficients then necessary, which in turn introduces additional estimation error into the forecasts. The lag lengths in a VAR can be decided using either F-test or information criteria. In this study Information criteria will be used (Stock and Watson, 2015).
4.4.1 Bayes information criterion (BIC)
One way to select the right number of lags, is by minimizing an information criterion. This can be done though the Bayes information criterion estimate, also called the Schwarz information criterion:
𝐵𝐼𝐶(𝜌) = ln(det (∑gI)) + 𝑘(𝑘𝜌 + 1)JK (B)B (17)
Let ∑gI be a 𝑘 × 𝑘 covariance matrix of the VAR errors, and the ∑gI is the estimate of the covariance matrix where I, j element of ∑gI is $
B∑B!,$𝜇̂#!𝜇̂H!, where 𝜇̂#! is the OLS residual from the i:th equation and 𝜇̂H! is the OLS residual for the j:th equation. 𝑘(𝑘𝜌 + 1) is the total number of regression coefficients in the VAR. The BIC estimator of 𝜌, 𝜌 is the value that minimizes BIC(𝜌) among all the possible choices 𝜌 = 0,1…., 𝜌𝑚𝑎𝑥, where pmax is the largest value of p considered and p = 0 corresponds to the model that contains only an intercept. T is the amount of observations (Stock and Watson, 2015)
4.4.2 Akaike information criterion (AIC)
Another information criterion, that can be used to select the number of lags in a vector autoregressive model, is the Akaike information criterion:
𝐴𝐼𝐶(𝜌) = ln(det (∑gI)) + 𝑘(𝑘𝜌 + 1)B5 (18)
The difference between these two models is the term ln(T) in the BIC is replaced with a 2 in the AIC. This will imply that the second term in the AIC will be smaller than the BIC.
4.5 Lagrange-multiplier test for autocorrelation
A test that gives the proper weighting to the residual autocorrelations is a Lagrange multiplier test for residual autocorrelations. If we consider a linear model:
𝑌!= 𝑋′!𝛽 + 𝜀! t = 1, 2,…T, (19)
Where:
In the Lagrange multiplier test for residual autocorrelations, the null hypothesis corresponds to 𝜌 = 0 and the alternative hypothesis is that 𝜌 ≠ 0. If the null hypothesis does not get rejected, it is assumed that the model does not have any residual autocorrelation. (Verbeek, 2008).
4.6 Granger causality test
To investigate if the variables in the VAR-model have any predictive power of one another, a Granger causality test will be performed. The test will show the direction of the causality in the model, that variables X causes Y or not, and vice versa. Granger causality means that if X Granger-causes Y, then C is a useful predictor of Y. However, it should be mentioned that the Granger causality test has not so much to do with causality as one might think (Stock and Watson), while “Granger predictability” is a more accurate term then “Granger causality”, the latter has become part of the jargon of econometrics.
The granger causality statistic is the F-statistic that test the hypothesis that the coefficients on all values of one variable, for example X are zero. This null hypothesis implies that these regressors have no predicative content for Y beyond that contained in the other regressors, and the test of this null hypothesis is called the granger causality test. This can be written as:
Δ𝑌! = ∑$"%&𝛼"Δ𝑌!#" + ∑$'%&𝛽'Δ𝑋!#' + 𝜀&! (21)
Δ𝑋! = ∑$"%&𝛾"Δ𝑋!#" + ∑$'%&𝛿'Δ𝑌!#' + 𝜀(! (22)
Equation 21 shows that the current value of Y is affected by the past values of itself and the past values of X, and similar equation 22 shows that X is related to the past values of itself and the past values of Y. The null hypothesis for eq 21 is that 𝛽H = 0, and the null hypothesis for eq 22 is that 𝛿H = 0. If we reject the null hypothesis for eq 21, that will means that “Δ𝑋 does not Granger causes Δ𝑌" (Stock and Watson, 2015)..
4.7 Impulse response function
Using the results of the Granger causality test, one can analyze which of the variables in the model have a significant effect for predicting future values in the VAR model. Although there are significant results, the test does not say whether it is a positive or negative relationship.
To investigate if there is a positive or a negative effect, an Impulse Response Function (IRF) will be performed. The IRF simulate how a variable behave by “shocking” one variable with one standard deviation by another variable.
The impulse response function will take the form:
𝑦!)$ = ∑*" % ,𝜓"𝜖!)$#& (23)
Where
{𝜓$}",' = ./.0*,,-.
/, (24)
The impulse response function measures the response of 𝑦#,!"L to an impulse in 𝑦#,!, keeping all other variables dated t and before constant (Verbeek, 2008).
5. Results
5.1 Augmented Dickey-fuller test
To investigate if the private consumption is affected by an unexpected oil price increase, it is important to test if the variables are stationary or not. The result from the ADF test will show if the variables are stationary or not. If the variables are non-stationary at level, then the variables will be differentiation and test again if they are stationary or not. The result show that all variables are non-stationary at level and stationary after differentiation (see appendix B).
5.2 Johansen test of cointegration
The Johansen test of cointegration finds no evidence of cointegration and the null hypothesis of no cointegration is accepted. Therefore, the model will only be able to investigate the short- run effects and a VAR model will be used instead of a VECM model. The result is presented in appendix C.
5.3 Vector autoregressive model
Since the test from the Engle-granger test showed that there is no cointegration between the variables in the model, the error correction term in the theoretical model that was presented in section 2.2, drops out. In the VAR-model the relationship between the five variables are tested.
In this thesis, it will be focused on the model where the variable private consumption is set as dependent variable. The relationship from the VAR test can be seen in table 1.
Table 1. Vector autoregression model
Dependent variable Lag Consumption Wealth Income Stibor Oil
Consumption 1 -0,215* 0,014 0,25** -0,004** -0,18**
2 0,103 0,071** -0,160 -0,007*** -0,008
3 0,104 0,078** -0,112 -0,003* 0,014*
4 -0,246 0,068** -0,162 -0,001 -0,012*
𝑅%: 0, 9593
In table 1, it can be observed that coefficients from the lagged variable of oil, is insignificant for the first two lags, and significant in lag three and four. Consumption first lag is significant, Wealth is insignificant in first lag, however it is significant in all the others lags. Stibor is significant in the first three lags.
Further on, the result from the Lagrange multiplier test for autocorrelation (see appendix D it could be observed that the VAR(4) model have no presence of autocorrelation. Since the variables in our model take a turn being dependent and independent it will be hard to interpret the result and therefore it could not be drawing any conclusion from the result. Therefore, a Granger causality test and an Impulse Response function will be done, and the result is presented in the next section.
5.4 Granger-causality test
A Granger causality test has been done to determine if there exist any causality between the variables in the VAR-model. The test will also identify in which direction the causality is for the variables.
Table 2. Granger causality test
Y ß X P-value
Cons Inc 0,128
Cons Wealth 0,056*
Cons IR 0,000***
Cons Oil 0,091*
Cons All 0,000***
Inc Cons 0,062*
Inc Wealth 0,024**
Inc IR 0,000***
Inc Oil 0,279
Inc All 0,000***
Wealth Cons 0,048**
Wealth Inc 0,119
Wealth IR 0,000***
Wealth Oil 0,842
Wealth All 0,000***
IR Cons 0,003***
IR Inc 0,024**
IR Wealth 0,017**
IR Oil 0,051*
IR All 0,002***
Oil Cons 0,725
Oil Inc 0,921
Oil Wealth 0,927
Oil IR 0,859
Oil All 0,105
Table 2. shows the output of the Granger causality test. *,** and *** indicate the coefficients significant level at ***=1%. **=5% and *=10%.
The result from the Granger-causality shows a significant result at 10% level, that there is an association between oil and consumption. The result also shows the direction of the causality, and it goes from oil on consumption, and not the other way around.
5.5 Impulse response function
From the VAR(4) model and the Granger causality test, it could be concluded that an unexpected oil price increase affects the household consumption. A common approach to identify the shocks of a VAR model is to use orthogonal impulse responses (OIR). The result of an OIR might be sensitive to the order of the variables, therefor the order of the variable will follow the same approach as previous literature (See Mera and person, Zhang and Broadstock and Nabila Zaman).
Figure 2 show that one positive standard deviation shock to the oil price decreases the consumption expenditure in the first three periods, before recovering to earlier levels.
Figure 2. Orthogonalized Impulse response function. Where the vertical axis measures the percentage change of a shock in the Brent oil price and the horizontal axes measures time. One period corresponds to one quarter
In appendix E, the result of the other variables can be observed. A positive shock in the oil price does not have a significant effect on wealth and income, although a positive shock in the oil
6. Discussion
Since this thesis did not find any cointegration between the variables in the presented model, this thesis did therefore only investigate the short-term effects from an unexpected oil price increase in household consumption. This means that the suggested theoretical VECM model rule out and a VAR model is used.
From the VAR(4) model, it could be concluded that oil price shocks do affect household consumption in the short run. The result from the Granger-causality shows a significant result at 10% level, that there is an association between oil and consumption. The result also shows the direction of the causality, and it goes from oil on consumption, and not the other way around.
The result in this thesis, falls in line with previous result found from Kilian (2008), Mehra and Peterson (2005) and Zhang and Broadstock (2014), that there is a short-term relationship between oil price and household consumption, and that a higher oil price leads to a reduction in household consumption.
Jiménez-Rodríguez & Sánchez (2005) did find among oil importing countries, that oil price increases are found to have a negative impact on GDP growth, in all cases except Japan.
Jiménez-Rodríguez & Sánchez suggest that it is important to consider not just whether oil prices increase or decline (and by how much), but also the environment in which the movement take place. An oil shock in a stable price environment is likely to have a larger economic consequence than one in a volatile price environment.
One explanation for the results of this thesis can be linked to the Transmission Mechanism Channels, that is presented in the theoretical framework section. The real-balance channel posits that oil price increase lead to inflation, lowering real money balances held by households in the economy and thereby depressing aggregate demand though familiar monetary channels. Since a higher oil price leads to higher gasoline prices, which leads to a reduction in disposable income, since it is assumed that a consumer spends a certain
proportion of their income on gasoline.
Since Sweden is an oil-importing country, it is reasonable that expect a reduction in consumption as a result of an oil price increase, due to the transfer of income from oil-importing
from an oil-importing country to an oil-exporting country occurs, the disposable income in the oil-exporting country increases but falls in the oil-importing country, which in turn causes negative effects on the consumption of the oil importer. Jiménez-Rodríguez & Sánchez (2005) did find that the effect of oil shocks on GDP was positive for Norway, which is an oil export- country.
As Sweden has continued to reduce its oil dependency and carbon dioxide emissions, it is interesting to investigate whether the international oil price has an impact on the consumption in Swedish households. Although the results show that oil prices have a negative impact on consumption, it could not be said anything about what this will look like in the future. But it is important to remember that even if Sweden would reduce its oil dependency to a significantly lower level, closer to the target, and thus minimize the real balance effect from rising oil prices, oil shocks could have a continued impact on the Swedish economy. An example of this is if an oil shock causes an increase in the price level from countries from which Sweden imports the necessary goods, and thus Sweden's consumption level decreases.
6.1 Drawbacks
One problem with the chosen model is the notion of a representative individual. Together with Alan Blinder, Deaton was able to show that changes in aggregate consumption over time were predictable in a manner inconsistent with the PI model's modern formulation (Blinder and Deaton 1985). This could reflect two things: Either the model was wrong, or the assumption of rational expectations was wrong. A third possibility, to which Deaton would attach greater and greater importance, was that the assumption of a representative individual was wrong. This led him to continue working with the model and to study a number of implications that no one had previously realized.
For individuals, income fluctuates sharply; In addition, some individuals receive an income increase while others receive an income reduction. These counterbalanced fluctuations take each other out when calculating total income (or its average), so that you only see the slow fluctuations that give persistence in aggregated data. Only when you abandon the idea of the representative consumer, and derive optimal consumption for each individual consumer, and then sum them up to aggregate consumption, does the agreement between theory and data become really good. In order to study the decisions of individual individuals (or households)
years, as opposed to cross-sectional data, where each individual only shows up once. However, panel data is very expensive to collect and is in many cases not available at all. Therefore, the ability to use this method has not been used, since the ability to access the data that would be needed.
Furthermore, this thesis only analyzes how a positive shock in oil prices affects household consumption in the short run. Therefore, this thesis will not be able to draw conclusion of a decreasing oil price or some long-term relationship.
7. Conclusion
The purpose of this thesis where to investigate the effect of an oil price shocks on the household consumption within the Swedish economy and provides new evidence of how the relationship exist, when taxes are also included in the oil price.
Understanding the effect of oil price shocks on consumption behavior is an important key tool for policymaker in order to implement an effective decision to avoid recessionary periods. The finding from the VAR-model, Granger-causality test and the IRF-graph shows that an oil price increase do have a negative impact on household consumption in the short-run. This is also in line with previous literature, such as Mehra and Peterson (2005) and Zhang and Broadstock (2014).
7.1 Future research
An interesting area for future studies would be to examine the effect on how the price of oil would affect the private consumption expenditure, using the method previously presented by Deaton. By doing a combination between micro- and macrodata and see if the result would be different.
Furthermore, since this thesis only where able to investigate how an increase in the oil price affect the household consumption, it could be in interest to analyze how the household consumption responds on a negative shock in the oil price. Future studies could also examine how different subgroups in consumption are affected by a shock in oil prices.
8. Reference list
Blanchard, Olivier J., and Jordi Gali. 2007. The Macroeconomic Effects of Oil Shocks: Why are the 2000s so different from the 1970s? National Bureau of Economic Research.
Campbell, J.Y. and G. Mankiw. (1989), ‘Consumption, Income, and Interest Rates:
Reinterpreting the Time Series Evidence,’ NBER Working Paper No. 2436.
Hamilton, J.D., 1983. Oil and the Macroeconomy since World War II. Journal of Political Economy, 91(2), pp.228–248.
Hamilton, James D. 2003. “What is an Oil Shock?” Journal of Econometrics 113 (2): 363–98.
Hamilton, James. 2009. “Causes and Consequences of the Oil Shock of 2007–2008.”
Working paper, University of California, San Diego.
Jiménez-Rodríguez, R. & Sánchez, M., 2005. Oil price shocks and real GDP growth:
empirical evidence for some OECD countries. Applied Economics, 37(2), pp.201–228.
Kilian, L., 2008. The Economic Effects of Energy Price Shocks. Journal of Economic Literature, 46(4), pp.871–909.
Mehra, Y.P. and J.D. Peterson (2005), ‘Oil Prices and Consumer Spending,’ FRB Richmond Economic Quarterly 91(3), pp. 53-72.
Mork, K.A., 1989. Oil and the Macroeconomy When Prices Go Up and Down: An Extension of Hamilton's Results. Journal of Political Economy, 97(3), pp.740–744.
Modigliani, F. and Brumberg, R.H. (1954) Utility Analysis and the Consumption Function:
An Interpretation of Cross-Section Data. In: Kurihara, K.K., Ed., Post-Keynesian Economics, Rutgers University Press, New Brunswick, 388-436.
Ng and Perron(1995) Unit Root Tests in ARMA Models with Data-Dependent Methods for the Selection of the Truncation Lag, JASA
Odusami, B. O. (2010) ‘To consume or not: How oil prices affect the comovement of consumption and aggregate wealth’, Energy Economics, Volume 32, Issue 4, July 2010, pp.
857–867,
Phillips, P.C.B. & Ouliaris, S., 1990. Asymptotic Properties of Residual Based Tests for Cointegration. Econometrica, 58(1), pp.165–193.
Stock, J. H., & Watson, M. W. 2015. Introduction to Econometrics. 3rd ed. Harlow: Pearson Educated Limited
Schwert (1989), Test for Unit Roots: A Monte Carlo Investigation, JBES
Wang, Y. S. (2013) ‘Oil price effects on personal consumption expenditures’, Energy Economics, Volume 36, March 2013, pp. 198–204,
Zhang, D & Broadstock, David, 2014. Impact of International Oil Price Shocks on Consumption Expenditures in ASEAN and East Asia. , DP-2014-24.
U.S. Energy Information Administration, Crude Oil Prices: Brent - Europe [DCOILBRENTEU], retrieved from FRED, Federal Reserve Bank of St. Louis;
https://fred.stlouisfed.org/series/DCOILBRENTEU,
9. Appendix
Appendix A
Grahp Sweden - Household Consumption, percent of total GDP, world bank
Appendix B
Augemented dickey-fuller test (at level)
Variable Test statistic P-value
Consumption(8) -0.133 0,9461
Wealth(7) -1,056 0,7321
Income(9) 0,317 0,9781
Interest rate(8) -1,087 0,7203
Brent Oil(5) -1,387 0,5885
A common rule of thumb for determining k* suggested by Schwert (1989) = 12 Numbers of lags are in parentheses.
Augemented dickey-fuller test (first difference)
Variable Test statistic P-value
Consumption(6) -4,284 0,0005***
Wealth(7) -3,660 0,0047***
Income(8) -3,769 0,0032***
Interest rate(7) -4,885 0,0000***
Brent Oil(4) -6,170 0,0000***
Appendix C
Johansen tests for cointegration
Null hypothesis Alternative Trace statistic 1% critical value 𝐻-: 𝑟 = 0 𝐻-: 𝑟 = 1 66.4497* 66.52
𝐻-: 𝑟 ≤ 1 𝐻-: 𝑟 = 2 38.1015 45.58 𝐻-: 𝑟 ≤ 2 𝐻-: 𝑟 = 3 19.5875 29.75 𝐻-: 𝑟 ≤ 3 𝐻-: 𝑟 = 4 8.7834 16.31 Lag length p = 4 Intercept included T =128
Notes: * Denotes acceptance of the null hypothesis that there is no cointegration
Null hypothesis Alternative Max statistic 1% critical value 𝐻-: 𝑟 = 0 𝐻-: 𝑟 = 1 28.3482* 35.17
𝐻-: 𝑟 ≤ 1 𝐻-: 𝑟 = 2 18.5140 28.82 𝐻-: 𝑟 ≤ 2 𝐻-: 𝑟 = 3 10.8041 22.99 𝐻-: 𝑟 ≤ 3 𝐻-: 𝑟 = 4 8.5155 15.69 Lag length p = 4 Intercept included T =128
Notes: * Denotes acceptance of the null hypothesis that there is no cointegration
Appendix D
Lag order selection
Lag AIC BIC
0 -7,33 -7,22
1 -8,89 -8-20
2 -9,92 -8,65
3 -12,69 -10,84
4 -13,43 11,19*
5 -13,54 -10,53
6 -13,47 -9,89
7 -13,44 -9,28
8 -13,67* -8,93
9 -13,46 -8,14
Notes: * denotes the optimal lag order that minimizes the information criteria
Appendix E
Figure 3. Orthogonalized Impulse response function. Where the vertical axis measures the percentage change of a shock in the Brent oil price and the horizontal axes measures time. One period corresponds to one quarter
