Upload
others
View
8
Download
0
Embed Size (px)
Citation preview
How Consumers Responded to the 2014-2015 Oil Price Shock?Evidence from the Consumer Expenditure Survey
Patrick Alexander and Louis Poirier (2017)July, 2017
Presented by: Louis Poirier2017 CE Microdata Workshop20 July 2017
The views expressed in the presentation are those of the authors. No responsibility for them should be attributed to the Bank of Canada
Protected B
MotivationBetween June 2014 and December 2015, global oil prices fell by almost 50%. This decline was expected to help U.S. GDP growth to accelerate to 3.6 % in 2016 and
3.3 % in 2016 (IMF). However, U.S. economic growth was only 2.6% in 2015 and 1.6 % in 2016.
Where was the impact from lower oil prices on consumption? “...the decline of oil prices over the last two years hasfailed to deliver the usual economic benefits. ” (New York Times, 2016)
2
This paper
Therefore to find the source of the “missing stimulus”, we have looked into the consumer’s reaction to lower gas prices.
Data: Consumer Expenditure Survey (CE)
Question: Did total consumption increase more for households that consume gasoline than it did for people that do not consume it (after June 2014)?
This PaperContribution to the literature: The first study to examine this episode using:
– Representative microdata– Isolate the impact of lower gas prices with:
• Information on vehicle ownership status • Intensity of gasoline consumption
Summary of findings:– Evidence of significant divergence in consumption response among
vehicle owners and high gasoline spenders.– Suggests the consumption stimulus was sizable.
Handling the CE data
5
Cleaning the data
Consumer Expenditure Survey (CE) Interview Survey (FMLI), 2013-2015 Append each fmli file including the ones in Stata
• use "$inipath2013\fmli131x.dta"• append using "$inipath2013\fmli132.dta"• append using "$inipath2013\fmli133.dta"• append using "$inipath2013\fmli134.dta"• append using "$inipath2013\fmli141.dta"
Cleaning the data
1. Steps to transform the “quarterly” data in monthly space:2013Q4 2014Q1x
xyz_pq = 500$ xyz_cq =1000$
(1) (2) (3)Oct Nov Dec Jan Feb Mar
if qintrvmo =
Results:Dec_xyz_2013= 500$Jan_xyz_2014= 500$Feb_xyz_2014= 500$
Cleaning the data
2. “Weighting” the data to get a representative sampleResults:
Dec_xyz_2013_w = 500$* finlwt21
Jan_xyz_2014_w = 500$* finlwt21
Feb_xyz_2014_w = 500$* finlwt21
Cleaning the data3. Removing aberrant or irrelevant observations:
– CU spends nothing or has zero income– Delete CU who bought gasoline and report
having no vehicle– Remove 1 per cent lowest and 1 per cent
highest income percentile as in Coibion et al. (2012)
– Delete some data points that are extreme
Our microdata pretty much in line with CE aggregates
Micro CE Agg CE Micro CE Agg CE Micro CE Agg CETotal spending 48,993$ 51,100$ 51,031$ 53,495$ 52,777$ 55,978$ Total spending ex gas 46,332$ 48,489$ 48,497$ 50,966$ 50,644$ 53,885$ Non-discretionnary spending 34,424$ 33,345$ 36,215$ 35,004$ 37,762$ 36,747$ Food spending 5,060$ 3,977$ 5,165$ 3,971$ 5,238$ 4,015$ Shelter spending 9,863$ 10,080$ 10,283$ 10,491$ 10,411$ 10,742$ Transportation spending 6,303$ 6,392$ 6,454$ 6,605$ 7,277$ 7,414$ Baby care spending 311$ N.A. 327$ N.A. 384$ N.A.Health care spending 3,552$ 3,631$ 4,267$ 4,290$ 4,247$ 4,342$ Personal insurance spending 5,560$ 5,528$ 5,750$ 5,726$ 6,289$ 6,349$ Utilities spending 3,775$ 3,737$ 3,969$ 3,921$ 3,915$ 3,885$ Discretionnary spending 11,908$ 15,144$ 12,282$ 15,962$ 12,883$ 17,138$ Alcoholic beverage spending 378$ 445$ 406$ 463$ 462$ 515$ Apparel spending 909$ 1,604$ 933$ 1,786$ 988$ 1,846$ Entertainment spending 2,190$ 2,482$ 2,413$ 2,728$ 2,485$ 2,842$ Personal care spending 296$ 608$ 307$ 645$ 321$ 683$ Education spending 944$ 1,138$ 936$ 1,174$ 927$ 1,309$ Books spending 109$ 102$ 109$ 103$ 91$ 114$ Food away from home spending 2,290$ 2,625$ 2,441$ 2,787$ 2,555$ 3,008$ House expenses spending 1,598$ 3,331$ 1,609$ 3,387$ 1,786$ 3,782$ Miscelleaneous spending 3,194$ 2,809$ 3,127$ 2,889$ 3,268$ 3,039$ Gasoline spending 2,661$ 2,611$ 2,534$ 2,468$ 2,133$ 2,090$
2013 2014 2015
Theoretical approach
11
TheoryEdelstein and Killian (2009) describe three distinct types of responses to negative gasoline price shocks:1. Discretionary income effect:
– Suppose: MPC > 0, 0 > ndg > -1
– Increased spending on gasoline and other products. Back-of-the-envelope calculation for discretionary income effect:
– Pg fell by ≈ 25% – nd
g ≈ -0.42– MPC = 1– Average CU spend 250$/month on gas
Discretionary savings: (0.25)*(0.58)*(1)*(250) <= 36.25$
Theory2. Income switching effect
– Suppose: -1 > ndng
– Increased spending on gasoline-related products.3. Precautionary savings effect
– Suppose: incomplete insurance markets– Increased MPC due to higher future real income.
Implication: the upper bound of consumption response is unclear. However, with additional effects, max > 36.25$
– Edelstein & Kilian (2009) mentions the response could be 4 times as large than the discretionary income effect, when you take into account additional effects.
Empirical Strategy
14
Empirical strategy
15
To do a diff in diff, you should only capture one time-invariant shock between 2 groups (i.e.: impact of lower gas prices).
• One group is affected by the shock and the other one is not.
• Therefore, the only parameter that should impact the difference in consumption between your 2 groups and the 2 time periods is the shock from lower gas prices.
Jan13 Jul13 Jan14 Jul14 Jan15 Jul15
Treatment group Control group
Assumption for consumption for a diff in diff regression
Last observation: December 2015
Before the shock After the shock
2. Control group
Impact of lower gas prices
1. Treated group
Oil price shock (Jun 14)
Spending for treated group
Spending for control group
Empirical strategy
Changes in Medicare in 2014 ? This is not consumption. Spike in January 2013 ?
Remove some sub-categories of spending to create our core spending (Yngit ):
0
50
100
150
200
250
300
0
20
40
60
80
100
120
2013 2014 2015
Education Spending
150
170
190
210
230
250
270
290
310
100
110
120
130
140
150
160
170
180
190
200
2013 2014 2015
Health Insurance Spending
Average Household Expenditures (USD)
450
470
490
510
530
550
570
590
610
630
125
145
165
185
205
225
245
2013 2014 2015
Low gas spenders (LHS) Normal gas spenders (RHS)
Retirement Savings
Last observation: December 2015Source: Bureau of Labor Statistics
Empirical strategy
1. Difference-in-difference specification:
where highgasi = 1 if gas spending in top 80th percentile.2. Alternative difference-in-difference specification:
where vehownri = 1 if CU reports owning a vehicle.Identifying assumption: car ownership “sticky”, contemporaneously unrelated to gasoline prices
Yngit = Spending ex gas, health insurance, savings and education
Empirical strategy
Car owners Non-Car owners
• Annual Income $ 65,184.72 $ 21,160.21
• Average number of persons in the CU
2.6 1.7
• Average age of the CU head
51.3 54.4
• Is the CU headed by a woman?
52% 62%
• Is the CU living in an urban area ?
94% 96%
• Is the CU having a mortgage?
39% 5%
• Is the head of the CU working ?
68% 39%
High gas spenders
Low gas spenders
• Annual Income $69,578.92 $ 29,437.28
• Average number of persons in the CU
2.7 1.7
• Average age of the CU head
50.0 57.5
• Is the CU headed by a woman?
51% 61%
• Is the CU living in an urban area ?
94% 93%
• Is the CU having a mortgage?
58% 14%
• Is the head of the CU working ?
71% 41%
Our control variables and their average between sub-groups:
Empirical strategy
19
A visual inspection of the pattern between the treated and control groups before and after the shock shows an acceptable behaviour for normal versus low gas spenders.
• The slopes are still not that similar for car versus non-car owners.
1700
1800
1900
2000
2100
2200
2300
2400
3300
3400
3500
3600
3700
3800
3900
2013 2014 2015Normal gas spenders (RHS)
Slope for normal gas spenders (LHS)
Low gas spenders (LHS)
Slope for low gas spenders (RHS)
Core spending evolution pre- and post-shock between the 2 groups respects the diff in diff principles
Average Household Expenditures (USD)
Last observation: December 2015Source: Bureau of Labor Statistics
1400
1450
1500
1550
1600
1650
1700
1750
1800
3250
3300
3350
3400
3450
3500
3550
3600
3650
3700
3750
2013 2014 2015
Car owners (LHS)Slope for car owners (LHS)Non-car owners (RHS)Slope for non-car owners (RHS)
Core spending evolution pre- and post-shock between the 2 groups does not respect the diff in diff principlesAverage Household Expenditures (USD)
Last observation: December 2015Source: Bureau of Labor Statistics
Jan13 Jul13 Jan14 Jul14 Jan15 Jul15
Treatment group Control group
Assumption for consumption for a diff in diff regressionBefore the shock After the shock
Results
20
1. Results for high gas spenders with core spending
21
Table 1.Regression on core spending ex gas for high gas spenders as treated
Robust t statistics in parentheses * p<0.05 ** p<0.01 *** p<0.001 \ controls * = number of persons, age, sex, in the CU, mortgage, urban, working
Basecase IV on income Month fixed effects Cluster by CU
Income0.0220*** 0.0272*** 0.0220*** 0.0220***(95.78) (117.62) (97.75) (51.09)
After July 2014 72.59*** 60.55** 213.3*** 72.59*(3.87) (3.22) (7.79) (2.36)
high gasoline 535.2*** 444.0*** 534.6*** 535.2***(27.67) (23.16) (22.66) (16.29)
Combined effect 104.5*** 93.19*** 105.6** 104.5*(4.18) (3.72) (2.97) (2.56)
Constant Y Y Y YControls* X X X X
Observations 190852 190852 190852 190852
Adjusted R-squared 0.219 0.212 0.219
2. Subcomponents of consumption
22
Table 1.2. Regression with spending on essential products, by sub-group
Robust t statistics in parentheses * p<0.05 ** p<0.01 *** p<0.001 \ controls * = number of persons, age, sex, in the CU, mortgage, urban, working
Essential Food Shelter Transportation Auto Baby care HealthcarePersonal insurance Utilites
Income 0.0130*** 0.00111*** 0.00591*** 0.00373*** 0.00178*** 0.000438*** 0.000604*** 0.000336*** 0.000824***[121.80] [104.07] [150.32] [40.34] [19.85] [60.95] [40.74] [30.09] [108.89]
High gasoline 57.11* 7.471** 28.42** 24.37 14.11 -0.511 -6.637 -1.193 5.191**[2.34] [3.06] [3.16] [1.15] [0.69] [-0.31] [-1.95] [-0.47] [2.99]
After July 2014 336.8*** 48.93*** -36.03*** 220.5*** 127.3*** -11.59*** 34.13*** 6.518** 74.42***[16.03] [23.21] [-4.64] [12.06] [7.18] [-8.16] [11.65] [2.96] [49.77]
Combined effect 41.39 2.938 -22.45* 51.47* 44.09 5.418** 4.711 1.145 -1.844[1.51] [1.07] [-2.22] [2.16] [1.91] [2.93] [1.23] [0.40] [-0.95]
Constant Y Y Y Y Y Y Y Y Y
Controls* X X X X X X X X X
Observations 190852 190852 190852 190852 190852 190852 190852 190852 190852Adjusted R-squared 0.152 0.284 0.180 0.021 0.007 0.056 0.026 0.009 0.304
3. Subcomponents of consumption
23
Table 1.3. Regression with spending on non-essential products, by sub-group
Non-essential Alcohol Apparel Entertainment Books AppliancesHousehold expenses
Food away from home Miscelleaneous
Income 0.00906*** 0.000377*** 0.000788*** 0.00193*** 0.0000865*** 0.000146*** 0.00148*** 0.00170*** 0.00248***
[76.07] [63.36] [17.30] [40.91] [35.40] [12.74] [40.22] [82.54] [36.35]
After July 2014 15.48 2.658*** 4.684*** 10.61*** 0.672** 4.892*** 0.118 4.921** -8.330
[1.51] [4.71] [4.34] [3.85] [2.88] [5.03] [0.03] [2.66] [-1.02]
High gasoline 198.3*** 4.568*** 2.114 39.45*** 3.304*** 6.753*** 22.14*** 49.71*** 71.52***
[19.07] [8.37] [1.13] [11.89] [14.60] [7.29] [6.03] [26.82] [9.26]
Combined effect 63.06*** 2.613*** 8.890*** 9.712* -1.625*** -2.267 11.43* 9.335*** 21.70*
[4.97] [3.72] [3.90] [2.22] [-5.44] [-1.77] [2.56] [3.96] [2.36]
Constant Y Y Y Y Y Y Y Y Y
Controls* X X X X X X X X X
Observations190852 190852 190852 190852 190852 190852 190852 190852 190852
Adjusted R-squared0.149 0.100 0.024 0.043 0.027 0.006 0.035 0.159 0.036
Robust t statistics in parentheses * p<0.05 ** p<0.01 *** p<0.001 \ controls * = number of persons, age, sex, in the CU, mortgage, urban, working
6. Other specifications:Mortgage, urban, oil states, etc.
24
Boost to consumption from lower oil prices is seen more in CUs : Median income Without mortgage Living in non-oil states
Conclusion
25
Recap of Key Findings
26
Clear evidence of differential response for those highly exposed to the shock.– Benchmark case suggests a consumption boost of +\- 100 $ per
CU– In line with other studies
Proves the existence of the positive windfall on consumption from lower gas prices– Maybe did not get capture properly by aggregate macro data…
Questions
27
Thank you !
28
Back-up slides
29
1.1. Results for car owners with core spending
30
Table 2.Regression on core spending ex gas for car owners as treatedModel A Model B Model C Model D Model E
Income 0.0252*** 0.0251*** 0.0243*** 0.0243*** 0.0222***(124.74) (124.74) (117.82) (117.82) (97.26)
After July 2014 111.0*** 113.1*** 35.33* 48.43**(8.13) (8.31) (2.23) (3.08)
Car owners 813.6*** 770.4*** 586.4***(68.05) (48.51) (36.02)
Combined effect 85.63*** 92.24***(3.94) (4.27)
Constant Y Y Y Y YControls* X
Observations 190852 190852 190852 190852 190852
Adjusted R-squared 0.302 0.303 0.308 0.308 0.32
Robust t statistics in parentheses * p<0.05 ** p<0.01 *** p<0.001 \ controls * = number of persons, age, sex, in the CU, mortgage, urban, working
More spending on gas ?
Question: How large was the stimulus for non-gasoline spending?
1.5
2
2.5
3
3.5
4
50
55
60
65
70
75
80
85
Jan-13 Apr-13 Jul-13 Oct-13 Jan-14 Apr-14 Jul-14 Oct-14 Jan-15 Apr-15 Jul-15 Oct-15
Number of gallons
Number of gallons consumed per household (LHS) Dollars per gallon including taxes (RHS)
Chart 1: The number of gallons consumed is inversely proportionate to gasoline prices
Dollars per gallon
Existing literature
32
Previous findings on 2015-2015 oil shock stimulus. Farrell and Grieg (2016):
– Consumers spent 80% of savings on non-gas items.– Weakness: Non-representative
Gelman et al. (2016):– Consumers spent 65-75% of savings on non-gas items.– Weakness: Non-representative
Baumeister and Killian (2017):– Consumption stimulus was roughly 1.2% of real consumption.– Weakness: Uses aggregate data, no micro evidence– Criticized as “not theoretically sound” (Ramey, 2016)
Cleaning the data
1. Steps to transform the “quarterly data” in monthly space:2013Q3 2013Q4
xyz_pq = 500$ xyz_cq =1000$
(1) (2) (3)Jul Aug Sep Oct Nov Dec
if qintrvmo =
Results:Sep_xyz = 500$Oct_xyz = 500$Nov_xyz=500$
Data by sub-groupsAverage monthly consumption excluding gasoline
All sample With a car Without a car Urban Rural Oil Non-oil
Before shock (Jan2013-Jun2014) $3,948.45 $4,168.57 $1,820.03 $4,010.80 $2,917.92 $4,245.78 $4,069.19
After shock (Jul2014-Dec2015) $4,214.31 $4,454.26 $1,837.03 $4,295.63 $3,146.98 $4,564.08 $4,334.11
Difference (%) 6.7% 6.9% 0.9% 7.1% 7.8% 7.5% 6.5%Average monthly gasoline consumption
All sample With a car Without a car Urban Rural Oil Non-oil
Before shock (Jan2013-Jun2014) $218.75 $241.14 $ - $218.46 $229.12 $236.60 $215.22
After shock (Jul2014-Dec2015) $184.75 $203.36 $ - $184.84 $229.12 $206.63 $181.18
Difference (%) -15.5% -15.7% - -15.4% -19.2% -12.7% -15.8%Number of observations 191,002 173,484 17,518 179,238 11,764 21,241 134,250
1700
1800
1900
2000
2100
2200
2300
2400
3300
3400
3500
3600
3700
3800
3900
Jan1
3M
ar13
May
13Ju
l13
Sep
13N
ov13
Jan1
4M
ar14
May
14Ju
l14
Sep
14N
ov14
Jan1
5M
ar15
May
15Ju
l15
Sep
15N
ov15
The evolution of core spending before and after the shock
between the 2 groups respect the diff in diff principles
Normal gas spenders (LHS)
Slope for normal gas spenders (LHS)
Low gas spenders (RHS)
Slope for low gas spenders (RHS)
Empirical strategy
6080100120140
150
200
250
300
Jan1
3M
ay13
Sep
13Ja
n14
May
14S
ep14
Jan1
5M
ay15
Sep
15
Health insurance spending differs
Car owners (LHS)
Non-car owners (RHS)
Changes in Medicare in 2014 ?
60
80
100
120
140
420440460480500520540560
Jan1
3M
ay13
Sep
13Ja
n14
May
14S
ep14
Jan1
5M
ay15
Sep
15
Retirement savings is differnt
Car owners (LHS)
Non-car owners (RHS)
This is not consumption.
0
50
100
150
050
100150200250300
Jan1
3M
ay13
Sep
13Ja
n14
May
14S
ep14
Jan1
5M
ay15
Sep
15
Education spending exhibits weird patterns
Car owners (LHS)
Non-car owners (RHS)
Weird spike in January 2013 for non-car owners
Let’s remove these sub-categories to create a core measure of spending without gas.