Transcript Analyzing the Macroeconomic Effects of Oil Price

Emmanuel Barnedo
Presentor
• Introduction
• Methodology
• Conceptual Framework
• Analytical Framework
• Results and Discussion
• Summary and Conclusion
• Policy Implications
• Limitations of the Study
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
“Crude oil and various petroleum product are crucial
in literally fueling the economy of a nation… If blood
is the lifeline of our body, then oil is the lifeline of the
economy…”
-Anakpawis Rep. Crispin Beltran (2008)
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
• During the oil crises in the 1970s, many
countries, experienced recession (Lee and Chui,
2009; Barsky and Kilian, 2001).
• In the 2000s, the Philippines proved once
more that it was indeed vulnerable to the
sustained increase in oil prices.
• The real Gross Domestic Product (GDP) had
declined considerably in 2007 until 2009 where
oil prices had reached its peak in 2008.
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
• The main objective of this study is to analyze
how changes in oil prices affect crude oil
consumption and some key macroeconomic
indicators in the Philippines.
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
• Specifically, the study aimed to accomplish
the following:
• To determine the effects of world and local oil
price changes in oil consumption and key
macroeconomic indicators, such as inflation
rate, investment, employment and real Gross
Domestic Product;
• To examine the time of disruption brought
about by the world and local oil price oil price
shocks;
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
• Specifically, the study aimed to accomplish
the following: (cont…)
• To compare the effects of these shocks in
terms of the pattern of disruption on the
domestic oil consumption and the key
macroeconomic indicators; and
• Lastly, to provide policy implications to
lessen the impact of oil price changes.
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
•
•
•
•
•
Conceptual Framework
Test of Stationarity
Vector Autoregressive (VAR) Model
Impulse Response Model
Sources of Data
Let:
• World oil price changes be D.DBOIL;
• Local oil price changes be D.DSOIL;
• inflation be INF;
• total oil consumption PPS;
• investment be FCF;
• total employment be EMP; and
• Gross Domestic Product be GDP.
• The Augmented Dickey Fuller (ADF) test is used.
𝑛
𝑌𝑡 = 𝜌𝑖 𝑌𝑡−1 + 𝛿𝑖
∆𝑌𝑡−𝑛 + 𝜀𝑖
𝑡=1
• ∆ is the differencing operator; 𝜀𝑖 is the white error term;
and 𝜌 and 𝛿 are the coefficients of the one period
lagged value 𝑌𝑡−1 and ∆𝑌𝑡−𝑛 , respectively, where
𝑛
𝑡=1 ∆𝑌𝑡−𝑛 = ∆𝑌𝑡−1 + ∆𝑌𝑡−2 + ⋯ + ∆𝑌𝑡−𝑛 (are higher
order autocorrelation) such that n is the optimum lag
length determined using sequential search method.
𝑛
𝑌𝑡 = 𝜌𝑖 𝑌𝑡−1 + 𝛿𝑖
∆𝑌𝑡−𝑛 + 𝜀𝑖
𝑡=1
Null Hypothesis: 𝜌 = 1 (𝑌𝑡 is non-stationary or there is a unit root)
Alternative Hypothesis: 𝜌 ≠ 1 (𝑌𝑡 is stationary or there is no unit root)
•
It follows the same asymptotic distribution as the DickeyFuller test so the same critical values can be used.
• Thus, if the computed absolute value of the tau statistic
(|τ|) exceeds the Mackinnon critical tau values, reject the
null hypothesis that 𝝆 = 1, the series is stationary.
Otherwise, fail to reject the null hypothesis, in such case,
the series is non-stationary (Gujarati, 2004).
• The augmented Dickey-Fuller tests for the variables under
study are:
𝐷. 𝐷𝐵𝑂𝐼𝐿𝑡 = 𝜌1 𝐷. 𝐷𝐵𝑂𝐼𝐿𝑡−1 + 𝛿1 𝑛𝑡=1 ∆𝐷. 𝐷𝐵𝑂𝐼𝐿𝑡−𝑛 + 𝜀1
𝐷. 𝐷𝑆𝐿𝑂𝐼𝐿𝑡 = 𝜌2 𝐷. 𝐷𝑆𝐿𝑂𝐼𝐿𝑡−1 + 𝛿2 𝑛𝑡=1 ∆𝐷. 𝐷𝑆𝐿𝑂𝐼𝐿𝑡−𝑛 + 𝜀2
𝐼𝑁𝐹𝑡 = 𝜌3 𝐼𝑁𝐹𝑡−1 + 𝛿3 𝑛𝑡=1 ∆𝐼𝑁𝐹𝑡−𝑛 + 𝜀3
𝑃𝑃𝑆𝑡 = 𝜌4 𝑃𝑃𝑆𝑡−1 + 𝛿4 𝑛𝑡=1 ∆𝑃𝑃𝑆𝑡−𝑛 + 𝜀4
𝐹𝐶𝐹𝑡 = 𝜌5 𝐹𝐶𝐹𝑡−1 + 𝛿5 𝑛𝑡=1 ∆𝐹𝐶𝐹𝑡−𝑛 + 𝜀5
𝐸𝑀𝑃𝑡 = 𝜌6 𝐸𝑀𝑃𝑡−1 + 𝛿6 𝑛𝑡=1 ∆𝐸𝑀𝑃𝑡−𝑛 + 𝜀6
𝐺𝐷𝑃𝑡 = 𝜌7 𝐺𝐷𝑃𝑡−1 + 𝛿7 𝑛𝑡=1 ∆𝐺𝐷𝑃𝑡−𝑛 + 𝜀7
• If the variable is found to be nonstationary in level form, it
must be stationarized thru differencing/detrending.
• The first VAR model used in the study with p-lag is given by:
𝒀𝒕 = 𝜶 + 𝝅𝟏 𝒀𝒕−𝟏 + 𝝅𝟐 𝒀𝒕−𝟐 + ⋯ + 𝝅𝒑 𝒀𝒕−𝒑 + 𝜺 𝒕
Where:
𝑌𝑡 = (𝐷. 𝐷𝐵𝑂𝐼𝐿𝑡 , 𝐼𝑁𝐹𝑡 , 𝑃𝑃𝑆𝑡 , 𝐹𝐶𝐹𝑡 , 𝐸𝑀𝑃𝑡 , 𝐺𝐷𝑃𝑡 ) denotes (nx1) vector
of (stationary/stationarized) time variables series ;
𝛼 is (nx1) vector of drift terms,
𝜋𝑖 is (nxn) coefficient matrix and 𝜀 𝑡 is (nx1) vector of white noise
error term; and
t=1,2,…,T; p=maximum no. of lags
*No. of lags were determined using Akaike Information Criterion
-A second VAR model was similarly specified for the local oil price
changes by replacing the world oil price changes (𝐷. 𝐷𝐵𝑂𝐼𝐿𝑡 ) with local oil
price changes (𝐷. 𝐷𝑆𝐿𝑂𝐼𝐿𝑡 )
VAR Model with world oil price changes
𝒑
𝒑
𝒑
d.𝒅𝒃𝒐𝒊𝒍𝒕 = 𝜶𝟏 + 𝒊=𝟏 𝜽𝟏𝒊 𝒅. 𝒅𝒃𝒐𝒊𝒍𝒕−𝒊 + 𝒊=𝟏 𝝐𝟏𝒊 𝒊𝒏𝒇𝒕−𝒊 + 𝒊=𝟏 𝜷𝟏𝒊 𝒑𝒑𝒔𝒕−𝒊 +
𝒑
𝒑
𝒑
𝜹
𝒇𝒄𝒇
+
𝝉
𝒆𝒎𝒑
+
𝟏𝒊
𝒕−𝒊
𝟏𝒊
𝒕−𝒊
𝒊=𝟏
𝒊=𝟏
𝒊=𝟏 𝜸𝟏𝒊 𝒈𝒅𝒑𝒕−𝒊 + 𝜺𝟏𝒕
𝑝
𝑝
𝑝
𝑝
𝑝
𝑝
𝑝
𝑝
𝑝
𝑖𝑛𝑓𝑡 = 𝛼2 + 𝑖=1 𝜃2𝑖 𝑑. 𝑑𝑏𝑜𝑖𝑙𝑡−𝑖 + 𝑖=1 𝜖2𝑖 𝑖𝑛𝑓𝑡−𝑖 + 𝑖=1 𝛽2𝑖 𝑝𝑝𝑠𝑡−𝑖 +
𝑝
𝑝
𝑝
𝛿
𝑓𝑐𝑓
+
𝜏
𝑒𝑚𝑝
+
𝑡−𝑖
𝑡−𝑖
𝑖=1 2𝑖
𝑖=1 2𝑖
𝑖=1 𝛾2𝑖 𝑔𝑑𝑝𝑡−𝑖 + 𝜀2𝑡
𝑝𝑝𝑠𝑡 = 𝛼3 + 𝑖=1 𝜃3𝑖 𝑑. 𝑑𝑏𝑜𝑖𝑙𝑡−𝑖 + 𝑖=1 𝜖3𝑖 𝑖𝑛𝑓𝑡−𝑖 + 𝑖=1 𝛽3𝑖 𝑝𝑝𝑠𝑡−𝑖 +
𝑝
𝑝
𝑝
𝛿
𝑓𝑐𝑓
+
𝜏
𝑒𝑚𝑝
+
3𝑖
𝑡−𝑖
3𝑖
𝑡−𝑖
𝑖=1
𝑖=1
𝑖=1 𝛾3𝑖 𝑔𝑑𝑝𝑡−𝑖 + 𝜀3𝑡
𝑓𝑐𝑓𝑡 = 𝛼4 + 𝑖=1 𝜃4𝑖 𝑑. 𝑑𝑏𝑜𝑖𝑙𝑡−𝑖 + 𝑖=1 𝜖4𝑖 𝑖𝑛𝑓𝑡−𝑖 + 𝑖=1 𝛽4𝑖 𝑝𝑝𝑠𝑡−𝑖 +
𝑝
𝑝
𝑝
𝛿
𝑓𝑐𝑓
+
𝜏
𝑒𝑚𝑝
+
𝑡−𝑖
𝑡−𝑖
𝑖=1 4𝑖
𝑖=1 4𝑖
𝑖=1 𝛾4𝑖 𝑔𝑑𝑝𝑡−𝑖 + 𝜀4𝑡
𝑝
𝑝
𝑝
𝑝
𝑝
𝑝
𝑒𝑚𝑝𝑡 = 𝛼5 + 𝑖=1 𝜃5𝑖 𝑑. 𝑑𝑏𝑜𝑖𝑙𝑡−𝑖 + 𝑖=1 𝜖5𝑖 𝑖𝑛𝑓𝑡−𝑖 + 𝑖=1 𝛽5𝑖 𝑝𝑝𝑠𝑡−𝑖 +
𝑝
𝑝
𝑝
𝛿
𝑓𝑐𝑓
+
𝜏
𝑒𝑚𝑝
+
5𝑖
𝑡−𝑖
5𝑖
𝑡−𝑖
𝑖=1
𝑖=1
𝑖=1 𝛾5𝑖 𝑔𝑑𝑝𝑡−𝑖 + 𝜀5𝑡
𝑔𝑑𝑝𝑡 = 𝛼6 + 𝑖=1 𝜃6𝑖 𝑑. 𝑑𝑏𝑜𝑖𝑙𝑡−𝑖 + 𝑖=1 𝜖6𝑖 𝑖𝑛𝑓𝑡−𝑖 + 𝑖=1 𝛽6𝑖 𝑝𝑝𝑠𝑡−𝑖 +
𝑝
𝑝
𝑝
𝛿
𝑓𝑐𝑓
+
𝜏
𝑒𝑚𝑝
+
6𝑖
𝑡−𝑖
6𝑖
𝑡−𝑖
𝑖=1
𝑖=1
𝑖=1 𝛾6𝑖 𝑔𝑑𝑝𝑡−𝑖 + 𝜀6𝑡
• It traces the responsiveness of the dependent variable in
the VAR system to a unit shock in error terms over time.
• But the error term must be nonautocorrelated (and
normally distributed) so that shocks can be represented
independently. Thus, non-autocorrelation and
normality of the distribution must be ensured first.
• The impulse response functions for this study are given
as follows:
𝑖𝑛𝑓𝑡 = 𝛼1 + 𝜋0 𝑤𝜀1𝑡 + 𝜋1 𝑤𝜀1𝑡−1 + 𝜋2 𝑤𝜀1𝑡−2 + ⋯ + 𝜋𝑘 𝑤𝜀1𝑡−𝑘 + 𝜇1
𝑜𝑐𝑜𝑛𝑡 = 𝛼2 + 𝛿0 𝑤𝜀1𝑡 + 𝛿1 𝑤𝜀1𝑡−1 + 𝛿2 𝑤𝜀1𝑡−2 + ⋯ + 𝛿𝑘 𝑤𝜀1𝑡−𝑘 + 𝜇2
𝑓𝑐𝑓𝑡 = 𝛼3 + 𝜖0 𝑤𝜀1𝑡 + 𝜖1 𝑤𝜀1𝑡−1 + 𝜖2 𝑤𝜀1𝑡−2 + ⋯ + 𝜖𝑘 𝑤𝜀1𝑡−𝑘 + 𝜇3
𝑒𝑚𝑝𝑡 = 𝛼4 + 𝜏0 𝑤𝜀1𝑡 + 𝜏1 𝑤𝜀1𝑡−1 + 𝜏2 𝑤𝜀1𝑡−2 + ⋯ + 𝜏𝑘 𝑤𝜀1𝑡−𝑘 + 𝜇4
𝑔𝑑𝑝𝑡 = 𝛼5 + 𝜃0 𝑤𝜀1𝑡 + 𝜃1 𝑤𝜀1𝑡−1 + 𝜃2 𝑤𝜀1𝑡−2 + ⋯ + 𝜃𝑘 𝑤𝜀1𝑡−𝑘 + 𝜇5
• The effects of such shock upon the VAR model over time are
graphed up to (k-1) lags with its confidence band.
• A second set of IRFs was also specified for local oil price
changes whose error terms were represented by 𝑑𝜀1𝑡 , 𝑑𝜀1𝑡−1 ,
𝑑𝜀1𝑡−2 , … , 𝑑𝜀1𝑡−𝑘 .
• The study covered the period 1991Q1- 2010Q4. The
variables included in the study were:
•
•
•
•
•
•
•
oil prices of Dubai Fateh (DBOIL) - IMF
pump prices for diesel oil (DSOIL) - DOE
inflation rate (INF) - NSO
petroleum products sales (PPS) - DOE
fixed capital formation (FCF) - NSCB
total employed people (EMP) - NSO
Gross Domestic Product (GDP) - NSCB
• There were some adjustments and estimations
made, such as:
• oil prices of Dubai Fateh (DBOIL); and
• quarterly data for petroleum product sales
Augmented Dickey-Fuller tests
Level Form
Optimal
Variables
• There were
some adjustments and estimations
Lag
t-stat
p-value
made, such
as:
a
length
• oil prices
of Dubai Fateh (DBOIL);
b
d.dboil
1
-10.505
stationary
• the first difference
(𝑌𝑡 − 𝑌𝑡−1 ) of0.0000*
DBOIL and DSLOIL
b
d.dsloilwas
2
-5.982
0.0000* in world
stationary
taken/used
to
represent change
oil
and
local oil price
and
PPS price (D.DBOIL)
2
-2.684
0.0769(D.DSLOIL);
nonstationary
petroleum product
INF • quarterly2 data for-6.143
0.0000* sales stationary
EMP
2
-0.488
0.8943
nonstationary
FCF
1
-2.603
0.9814
nonstationary
GDP
2
0.398
0.9814
nonstationary
a Optimal
lag length was determined through sequential search method.
* represents significant at 5% level.
Augmented Dickey-Fuller tests
Adjusted Variables
Variables
s4.ppsc
s4.fcfc
detrend_empd
s4.gdpc (with drift)
Optimal
Lag
lengtha
t-stat
p-value
1
1
1
1
-3.522
-3.487
-3.156
-2.511
0.0075*
0.0083*
0.0227*
0.0072*
Optimal lag length was determined through sequential search method.
b Adjusted using first difference (𝑌 − 𝑌
𝑡
𝑡−1 ) to represent ∆𝑌.
c Adjusted using fourth seasonal differencing (𝑌 − 𝑌
𝑡
𝑡−4 ).
d Adjusted using detrending approach.
* represents significant at 5% level.
a
stationary
stationary
stationary
stationary
• According to the Akaike Information Criterion
(AIC), the optimal lag length for the first VAR
model was three (3) while the second was two.
Effect(s) of World Oil Price
Shock
Effect(s) of Local Oil Price
Shock
dslshock2: D.dsloil -> inf
dbshock3: D.dboil -> inf
.6
.4
.4
.2
.2
0
0
-.2
-.2
0
10
95% CI
20
step
30
impulse response function (irf)
40
0
10
95% CI
20
step
30
impulse response function (irf)
• The initial reaction of inflation was positive that may be
attributed to the direct and indirect effect(s) of an oil price
shock.
40
Effect(s) of Local Oil Price
Shock
Effect(s) of World Oil Price
Shock
dbshock3: D.dboil -> S4.pps
dslshock2: D.dsloil -> S4.pps
200
0
0
-100
-200
-200
-400
-300
0
10
95% CI
20
step
30
impulse response function (irf)
40
0
10
95% CI
20
step
30
40
impulse response function (irf)
• Crude oil was said to be relatively inelastic. However, the
significant decline in oil consumption also signalled that it was
becoming less inelastic over time
Effect(s) of World Oil Price
Shock
Effect(s) of Local Oil Price
Shock
dbshock3: D.dboil -> S4.fcf
dslshock2: D.dsloil -> S4.fcf
500
1000
500
0
0
-500
-1000
-500
0
10
95% CI
20
step
30
impulse response function (irf)
40
0
10
95% CI
20
step
30
impulse response function (irf)
• The initial increase in investment on energy-efficient capital
may be relatively higher compared to the decrease (or
postponement) in the investment on other capital.
40
Effect(s) of World Oil Price
Shock
Effect(s) of Local Oil Price
Shock
dslshock2: D.dsloil -> detrend_emp
dbshock3: D.dboil -> detrend_emp
50
50
0
0
-50
-50
-100
-100
0
10
95% CI
20
step
30
impulse response function (irf)
40
0
10
95% CI
20
step
30
40
impulse response function (irf)
The slow recovery of employment may be attributed to: (1) the
industry-specific skills of labor (Loungani 1986) and (2) increase
in investment .
Effect(s) of World Oil Price
Shock
Effect(s) of Local Oil Price
Shock
dslshock2: D.dsloil -> S4.gdp
dbshock3: D.dboil -> S4.gdp
1000
500
500
0
0
-500
-500
-1000
0
10
95% CI
20
step
30
impulse response function (irf)
40
0
10
95% CI
20
step
30
impulse response function (irf)
The increase in GDP may be attributed to the increase in
investment. Such increase may have reduce the negative
impact on energy-intensive sectors, such as transport
40
• Conclusion
• Policy Implication(s)
• Limitation(s) of the Study
Variables
inf
A
+
B
-
C
0.304
D
0.089
s4.pps
-
n.a.
273.138
n.a.
s4.fcf
+
-
575.947
469.056
detrend
_emp
s4.gdp
-
n.a.
40.089
n.a.
+
-
661.222
622.406
E
14
Quarters
6
Quarters
15
Quarters
16
Quarters
16
Quarters
F
0.246
G
0.037
151
n.a.
49
219.636
31.211
n.a.
150
359.379
H
11
Quarters
9
Quarters
12
Quarters
14
Quarters
20
Quarters
A- Initial Response to the Oil Price Shock (the same for both)
B- Second Response to the Oil Price Shock (the same for both)
C- Magnitude of the Initial response (Max Value) (World Oil Price Changes)
D- Magnitude of the Second Response (Max Value) (World Oil Price Changes)
E- Length of Disruption of the World Oil Price Shock
F- Magnitude of the Initial response (Max Value) (Local Oil Price Changes)
G- Magnitude of the Second Response (Max Value) (Local Oil Price Changes)
H- Length of Disruption of the Local Oil Price Shock
• Although oil price shocks were found to be
disruptive, regulating the oil downstream
industry could create more distortions.
• Since the said shock is temporary, the
government can implement short-term
intervention, catered specifically to
particular sector.
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
• As a long term solution, the
government should promote the
investment on energy-efficient
technology/capital and the
production of indigenous energy
sources.
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
• Non-availability of quarterly data for
oil consumption
• The use of diesel oil price
Analyzing the Macroeconomic Effects of Oil Price Changes in the Philippines
End of presentation
Thank you very much!