Spring 2014 Senior Distinction Papers-Class of 2013 · smartphones and tablets. Such diffusion has...
Transcript of Spring 2014 Senior Distinction Papers-Class of 2013 · smartphones and tablets. Such diffusion has...
1
Omicron Delta Epsilon International Honor Society in Economics
Beta Chapter: St. Olaf College
Executive Board 2013-2014 President: Kelly Tomera and Nick Evens
Vice President: Erik Springer
ODE Journal Executive Editor: Rebecca Gobel
ODE Journal Associate Editor: William Lutterman
About Omicron Delta Epsilon
Omicron Delta Epsilon is one of the world’s largest academic honor societies.
The objectives of Omicron Delta Epsilon are recognition of scholastic
attainment and the honoring of outstanding achievements in economics, the
establishment of closer ties between students and faculty in economics within
colleges and universities, the publication of its official journal, The American
Economist, and the sponsoring of panels at professional meetings as well as the
Irving Fisher and Frank W. Taussig competitions.
Currently, Omicron Delta Epsilon has 578 chapters located in the United
States, Canada, Australia, the United Kingdom, Mexico, Puerto Rico, South
Africa, Egypt, France, and the United Arab Emirates. With such a broad
international base, chapter activities vary widely, ranging from invited
speakers, group discussions, dinners, and meetings, to special projects such
as review sessions and tutoring for students in economics. Omicron Delta
Epsilon plays a prominent role in the annual Honors Day celebrations at
many colleges and universities.
Senior Distinction Papers-Class of 2013 Spring 2014
2
Omicron Delta Epsilon International Honor Society in Economics
Beta Chapter: St. Olaf College
Executive Board 2013-2014 President: Kelly Tomera and Nick Evens
Vice President: Erik Springer
ODE Journal Executive Editor: Rebecca Gobel
ODE Journal Associate Editor: William Lutterman
About Omicron Delta Epsilon
Omicron Delta Epsilon is one of the world’s largest academic honor societies.
The objectives of Omicron Delta Epsilon are recognition of scholastic
attainment and the honoring of outstanding achievements in economics, the
establishment of closer ties between students and faculty in economics within
colleges and universities, the publication of its official journal, The American
Economist, and the sponsoring of panels at professional meetings as well as the
Irving Fisher and Frank W. Taussig competitions.
Currently, Omicron Delta Epsilon has 578 chapters located in the United
States, Canada, Australia, the United Kingdom, Mexico, Puerto Rico, South
Africa, Egypt, France, and the United Arab Emirates. With such a broad
international base, chapter activities vary widely, ranging from invited
speakers, group discussions, dinners, and meetings, to special projects such
as review sessions and tutoring for students in economics. Omicron Delta
Epsilon plays a prominent role in the annual Honors Day celebrations at
many colleges and universities.
St. Olaf College’s Beta Chapter of Omicron Delta Epsilon aims to build a bridge
between the economics faculty and students, actively providing input and
assistance as needed to improve departmental events; they also publish an in-
house economics journal, encouraging, reviewing, selecting, and publishing
original work from economics students at the college.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
1
The St. Olaf College Economics Department’s
Omicron Delta Epsilon Journal of Economic Research
___________________________________________________________
Contents Spring 2014
___________________________________________________________
Shannon Cordes: Stuck in the Middle:
The Substitution Effect of Digital Technology on Middle Skilled
Jobs…….…….…….………………………………………..………....2
Kelly Tomera: Combating Obesity in America- Fat Tax Is Not the
Way To Go…………………..………………………………………..42
Ryan Johnsrud: US Stock Market as a Leading Indicator of the US
Econom…….……………………………………………………..…...54
Annabel Ansel: Obamacare: Healthcare Reform and the Ailing
Labor Market………………………….........................................................98
Spring 2014
2
Stuck in the Middle:
The Substitution Effect of Digital Technology on Middle Skilled Jobs
Shannon Cordes
Abstract
Since the 1980s, middle-skilled occupations have experienced a
steady decline in the share of U.S. employment, a phenomenon often
attributed to advances in digital technology. Among the explanations
reported in the economic literature, the Autor Levy Murnane (ALM)
hypothesis suggests that routine processes are most vulnerable to digital
substitution – digital technology substitutes for routine occupations and
compliments non-routine occupations. Tests have involved a division of
occupations as routine vs. non-routine, which are subdivided further as
manual, cognitive or analytic.
Unlike existing literature that examines the effect of digital
technology on employment, this paper analyzes its effect on the
unemployed. Using data from the Current Population Survey and the
Dictionary of Occupational Titles, I find that routine cognitive workers are
more likely to be unemployed than non-routine cognitive workers, thus
reinforcing the ALM hypothesis. However, the effect of advancements in
digital technology on the unemployment gap between routine cognitive
and non-routine cognitive occupations depends on the type of technology.
Using VAR techniques, I find that the net effect of advances in hardware
technology on the unemployment gap is zero, while the net effect of
advances in software technology is positive.
I. Introduction
Since the financial crisis in 2008 from which the U.S. economy
spiraled into the deepest recession since The Great Depression, politicians
and citizens await ‘job creation’. Hoping that new jobs will be
forthcoming, many also expect former jobs to return once the economy
recovers. But what if those former jobs will not return? What if the labor
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
3
market has structurally changed? Such that a job once occupied for over
ten years has been filled not by someone else, but something else – digital
technology.
The scope of digital technology is not limited to the
implementation of personal computers, but also encompasses the
application, diffusion and replication of software on computers,
smartphones and tablets. Such diffusion has created online marketplaces –
such as progromatic bidding – classrooms and applications that have
reduced skills at the hands of dozens of workers into one screen. This
paper aims to determine whether advancements in digital technology have
historically led to the displacement of workers employed in routinized
occupations.
It may be questioned whether advancing technology causes
permanent net job shifts. The corollary to job displacements by
technology is increased productivity and new job creation – per growth
models ranging from Solow’s to the Real Business Cycle model. Isn’t this
just another technology shock pushing aggregate demand outward?
In the last two decades, a vast array of literature has emerged citing
one of two forces to structural changes in the labor market: Skill-Biased
Technological Change (SBTC) and the Autor Levy Murnane (ALM)
hypotheses – the latter emerging from the inadequacy of the former to
“Stuck in the Middle” by Shannon Cordes
4
explain the rising job polarization in Western labor markets. According to
the SBTC hypothesis, technology increases the demand for skilled labor
and reduces the demand for unskilled labor, resulting in a higher wage
compensation for skilled relative to unskilled labor. Acemoglu (1998)
and Kramarz (1998) establish the correlation between skill acquisition and
technological change. Machin and Van Reenen (1998) supplement the
SBTC hypothesis with empirical evidence that indicate faster skill
upgrading is associated with higher industry research and development.
Yet, if the SBTC hypothesis were true, then in the last decade, one
would expect positive growth rates of employment for middle & high-
skilled labor, as well as a decline in low-skilled employment. However,
this is not the case. As Autor, Katz and Kreuger (1998), Autor, Levy and
Murane (2003), Goos & Manning (2007), Nedelkoska (2012) have
stipulated, Western labor markets exhibit a polarizing trend that can only
be explained in part by the SBTC hypothesis. Indeed, employment growth
of high-skilled labor has increased, but so too has the growth of low-
skilled employment. The individuals losing out are not unskilled workers,
but ‘middling workers’ employed in routinized labor. Consequently, the
SBTC hypothesis cannot fully explain the twin peaks phenomena of the
labor market. Instead, the ALM hypothesis offers a nuanced view of the
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
5
relationship between digital technology and the labor market, one that
surpasses the binary categorization of labor as skilled and unskilled.
II. The SBTC & ALM Hypotheses
Technological shocks can stimulate economic growth, yet the
impact of these shocks on the labor market requires closer scrutiny. In the
short run, the effect of a change in technology – such as the invention of
the railroad, computer or web browser – is palpable within the subsequent
boost in real GDP. However, the impact of a technological change on the
labor market reveals itself over a longer period of time. Often, the true
impact – as with computerization – does not emerge until decades after the
initial invention.
With increasing vigor, economists and citizens alike have raised
alarm about widening job polarization in the U.S. economy, which they
attribute to advancements in digital technology. The SBTC hypothesis
generalizes the impact of technology upon all skill types as equivalent.
Yet occupations within industries are characterized by varying types of
skills. So much so that the Dictionary of Occupational Titles (DOT)
assigns skill descriptions to twelve thousand unique jobs.
Contrary to the SBTC hypothesis, which categorizes occupations
as skilled and unskilled, the ALM hypothesis applies two tiers of skill
“Stuck in the Middle” by Shannon Cordes
6
decomposition. First, the tasks of labor are segmented into routine and
non-routine skills. Within these branches, tasks can be further broken
down into cognitive and manual skills. The ALM hypothesis predicts that
technology replaces routine cognitive and manual tasks, but complements
non-routine cognitive tasks.1 Once stated, this declaration seems obvious,
especially since the assembly line epitomizes the replacement of manual
labor by machines. However, we must reconsider our standard conception
of a routine task. Routine tasks have become synonymous with manual
tasks where the laborer repeats the same motion, whether that be
smoothing the surfaces of ceramic toilet bowls or individually wrapping
Galvadier chocolate truffles for packaging. Routine tasks, though, also
include routine cognitive skills.
Prior to the invention of the computer, tasks that required repetitive
information processing fell strictly within the mind’s domain. With the
invention of the computer, a machine whose primary function is to process
information, the mind’s domain has been encroached upon and in many
cases usurped by another domain – the network domain. Computers and
the bundles of software programs and communication capabilities
packaged with them have expanded the replacement capabilities of
technology to include human cognition. As Autor et. al (2003) articulate:
1 For non-routine manual tasks, the ALM hypothesis predicts that digital technology is
a weak, or limited, compliment.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
7
As symbolic processors–machines that store, retrieve, sort,
and act upon information–computers augment or supplant
human cognition in a vast set of information processing
tasks that were historically the mind’s exclusive
dominion…Computers have increasingly substituted for the
information processing, communications, and coordinating
functions of bookkeepers, cashiers, telephone operators and
other handlers of repetitive information processing tasks.
(5)
Computers have expanded the types of tasks replaced by technology to
include not only routine manual skills, but also routine cognitive skills.
Thus, computers function as substitutes for routine manual and cognitive
tasks and complements for non-routine cognitive tasks by increasing
productivity.
Over the past three decades, the share of employment has shifted
in favor of non-routine cognitive labor, as displayed in Figure 1.1.
Furthermore, while the employment share of routine cognitive labor has
fallen since 1970, the share of non-routine manual labor remains constant
over time. If the SBTC hypothesis was true, we would expect the share of
the lowest skilled occupations to decline as technology advances. Yet – as
the ALM hypothesis predicts – since computers function as a limited
complement of non-routine manual labor, its share of employment remains
stable as digital technology advances.
“Stuck in the Middle” by Shannon Cordes
8
Figure 1.1
As with all substitution effects, the primary component that drives
the replacement of one process for another is cost. In the decision to hire,
a firm must choose whether the marginal product of labor of hiring an
additional worker is equal to or exceeds the worker’s marginal cost. A
firm must also consider whether substitutes are available that offer a lower
marginal cost for the same or greater marginal product of the worker. As
substitutes, the price of computers and the quantity of routine cognitive
labor employed are positively related. Conversely as compliments, the
quantity of non-routine cognitive labor employed and the price of
computers are inversely related. Given this relationship between the price
0%
10%
20%
30%
40%
50%
19
68
19
70
19
72
19
74
19
76
19
78
19
80
19
82
19
84
19
86
19
88
19
90
19
92
19
94
19
96
19
98
20
00
20
02
20
04
20
06
20
08
20
10
20
12
Share of total
employment
Year
Employment Share by Skill Measure, 1968-2013
Nonroutine Cognitive Routine Cognitive Nonroutine Manual
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
9
of computers and its substitutes and compliments, as the price of
computers declines, the quantity demanded of routine cognitive labor will
decrease while the quantity of non-routine cognitive labor will increase.
The corollary of this statement indicates that as computer prices decline,
the number of individuals employed in routine cognitive labor will either
become unemployed or switch occupations. In order for this effect to be
true, the price of computers must have declined since 1977. Indeed, as
Bresnahan (2000) cites, the quality-adjusted price of computers has
declined at a compound rate of twenty percent per year through the mid
1990s.2
These two functions of computers – as substitutes and
compliments –are the central claim of the ALM hypothesis. If the ALM
hypothesis models the interaction between the labor market and digital
technology, we would expect within the data that routine cognitive
workers represent an increasing share of the unemployed.
III. Hypotheses: Adding the Lens of Job Loss
The majority of economic literature concerning the displacement
of routinized workers by digital technology focuses on measuring the
share of employment differential that can be attributed to computers (Goos
2 Figure 6.1 in the Appendix displays the steady decline of the computer price index
since 2005.
“Stuck in the Middle” by Shannon Cordes
10
and Manning, 2007, Autor, Levy and Murnane, 2003, Autor, Katz and
Kreuger, 1998). Such work attempts to determine the degree to which
computers modify the tasks of a given occupation by examining those
currently employed. In part, this is a result of the nature of the data, since
most surveys only ask employed individuals whether they use a computer
at work and for what purpose. Yet, by only measuring the effects of
technology displacement on the employed, existing literature neglects the
most important people of interest – the unemployed.
What happens to the individuals whose routinized jobs are usurped
by digital technology? Do they find work elsewhere? Are they more
likely to become unemployed? Must they accept a lower wage if changing
occupations? Ljubica Nedelkoska attempts to answer this very question
for the case of another western economic power: Germany. Nedelkoska
(2012) attempts to track the adaptation process of German workers whose
occupations require routine tasks. An individual whose skills become
obsolete faces two choices: unemployment or occupational change.
Nedelkoska concludes that workers performing routine tasks incur a
higher probability of becoming unemployed and switching occupations.
Similar to Nedelkoska’s conclusions, I predict that workers performing
routine cognitive tasks face a higher probability of unemployment
compared to those performing non-routine cognitive tasks.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
11
A question that remains, however, pertains to the causality
assertion. Is technology to blame for the decline of occupations requiring
routine cognitive tasks? Nedelkoska concludes that if such directional
causality exists, then it is a weak causal relationship in production and
manufacturing, and for coding technologies other than computers, the
relationship can even be complementary, proving that not all digital
technologies interact uniformly with the labor market. Autor, Levy and
Murnane (2003) conclude opposite results. Instead, within the most
highly computerized industries, the trend exhibits an increase in labor
input for non-routine cognitive skills and a decrease in labor input for
routine cognitive skills. As with most relationships in economics, direct
causality remains elusive. Association, at best, can be attained. Thus the
second hypothesis to be tested is whether advancements in digital
technology are historically followed by an expansion of the
unemployment gap between routine cognitive and non-routine cognitive
occupations.
IV. Data Methodology
In order to measure the likelihood of unemployment for non-
routine cognitive occupations compared to routine cognitive occupations
and to test for causality between technological advancement and the
unemployment gap, we require measures of skill task for occupations for
“Stuck in the Middle” by Shannon Cordes
12
sectors from which the unemployed work. The National Academy of
Sciences and Committee of Occupational Classification and Analysis
aggregated Dictionary of Occupation Titles (DOT) characteristics for the
574 occupation categories of the 1970 U.S. Census. In the COC-DOT
aggregation, Census Occupation Codes (COC) are assigned a score for
General Education Development, Aptitudes and Temperaments measured
in the DOT.
Using the same methodology as Autor, Levy and Murnane (2003),
five characteristics indicate the degree to which an occupation is non-
routine or routine.
Non-routine Cognitive-Analytic: Mathematical General Education
Development (GEDMATH) captures an occupations quantitative
and analytical reasoning skills.
Non-routine Cognitive-Interactive: Directional, Control, Planning
(DCP) measures an occupation’s communication and management
skills.
Non-routine Manual: Eye-Hand-Foot Coordination (EYEHAND)
takes on high values for occupations requiring a high degree of
physical agility and spatial recognition.
Routine Cognitive: Set Limits, Tolerances or Standards (STP)
indicates a worker’s ability to adapt to work requiring set limits,
tolerances or standards.
Routine Manual: Finger Dexterity (FINGDEX) captures the level
of motor skills an occupation requires.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
13
Based on the score of the DOT skill measures in the COC-DOT
aggregation, I calculated a weighted DOT mean task in order to assign one
of the five skill measures to each of the 574 census occupation codes. In
other words, I defined each occupation as non-routine cognitive-analytic,
non-routine cognitive-interactive, non-routine manual, routine cognitive or
routine manual.
Although U.S. Census Data served to classify occupational codes
by skill measure, I drew from the Current Population Survey (CPS) to
analyze unemployment by skill measure. Since the U.S. Census and the
CPS code occupations differently, I created a crosswalk using the CPS
translation page to assign a skill measure to the CPS occupation codes
with base year of 1970.3 By using the 1970 base year occupation codes,
occupations are comparable over time. However, in using the base year,
we assume that the task requirement of occupations remains constant
overtime. Although this assumption could limit the regression results
since technological advancement leads to changes in an occupation’s
tasks, the benefit of the assumption outweighs its limitations. By
classifying each CPS occupation with a skill measure, characteristics of
3 The Current Population Survey provides the translation page. The page categorizes
CPS occupation codes by U.S. census occupation codes for all years. The U.S.
census occupation codes are more detailed than the CPS occupation code. Thus to
create the crosswalk, I matched the U.S. census codes to their CPS counterpart as
defined by the translation page.
“Stuck in the Middle” by Shannon Cordes
14
the occupation – instead of the individual – can be tracked overtime, in
particular, unemployment status.
Probit Model
To measure the likelihood of unemployment of an individual with
a routine cognitive or non-routine cognitive-analytic task, I created a
probit model using CPS data from 1972-2013.4 The probit model is
defined as follows:
1. ( | ) ( )
where is the likelihood of unemployment of individual i at year T,
is a vector of coefficients for a vector of characteristics that include
age, sex, race, income, education and industry and is a vector of
coefficients for a vector of dependent variables that include dummy
variables defining the skill measure of an occupation.5
Vector Auto-regression (VAR) Model
Testing for Granger Causality through Vector Auto-regression is
one method of determining whether a causal relationship exists between
4 The sample consists of individuals between the ages of 18-65 who are in the labor
force. For each year, T, the number of observations is between 40-90k individuals.
See Table 1.2 in the Appendix for the probit model output. 5 Income is measured using an individual’s wage. All measures of income are
inflation adjusted using the CPI less food and energy with 2007 as the base year.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
15
two forces. A casual relationship between digital technology and the
unemployment rate of non-routine cognitive and routine cognitive jobs has
yet to be established or refuted. If the probit model supports the
hypothesis that individuals employed in routine cognitive occupations
incur a greater likelihood of unemployment than those employed in non-
routine cognitive work, then the question remains as to why this disparity
exists. The VAR model tests whether this difference in unemployment
rates can be attributed to advancements in digital technology.
In order to create a VAR model, I converted the existing micro
data from the CPS into time-series data. From 1972-2013, I calculated the
annual unemployment rate for routine cognitive and non-routine cognitive
jobs within the aggregate economy. Since the variable of concern is the
disparity between the unemployment rate of routine and non-routine
cognitive occupations, the dependent variable is the difference between
the two unemployment rates. (Since I subtracted the unemployment rate
of non-routine cognitive jobs from the unemployment rate of routine
cognitive jobs, we would expect this difference to be positive).
For the primary explanatory variable, a measure of digital
technology must be chosen. Investment in digital technology will serve as
a proxy for technological advancement in the VAR model. The Bureau of
Labor Statistics provides aggregate and industry level measures of digital
“Stuck in the Middle” by Shannon Cordes
16
technology investment in the National Income and Product Accounts
(NIPA). Categories of investment include PC (personal computer),
printers, hard drives, user-owned software, licensed software etc. I
aggregated these sub-categories into two umbrella categories: hardware
and software investment.
Although investment in durable and non-durable computer goods
commenced during the same time period, the respective growth rates of
investment vary significantly. Figure 1.2 and 1.3 displays the level of
economic wide investment in hardware and software technology for the
U.S. from 1972-2011. Accompanied with the graphs is a timeline that
tracks the milestones of invention for the respective technologies.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
17
Figure 1.26
U.S. Hardware Investment, 1972-2011
6 The timeline of hardware and software technological advancements was compiled
from the Computer History Museum.
“Stuck in the Middle” by Shannon Cordes
18
Figure 1.3
U.S. Software Investment, 1972-2011
Investment in both hardware and software increase over time – as
expected – yet for hardware investment, the growth rate remains fairly
constant while software investment exhibits varying rates of exponential
growth. Within the contours of these growth rates, the history of the
digital technology revolution resides.
In 1989, Sir Timothy John Berners Lee invented the World Wide
Web (it was released in 1990), but this invention alone did not spark the
fastest rate of computer hardware investment from 1992-1996. Without a
format to navigate, read and post content, the World Wide Web was
inaccessible to widespread users. Once CERN uploaded the first website
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
19
on August 6, 1991, the Internet became universally user-friendly, which
sparked the highest growth rate of hardware investment. Software
investment, however, did not respond in the same manner as hardware
investment to the creation of the first website. Not until 2002 – following
the DOT-COM bubble and the invention of the web-browser – did
software investment finally takeoff.
Since hardware and software investment responded with varying
growth rates to technological disturbances, they may also exhibit differing
effects on unemployment for routine cognitive and non-routine cognitive
occupations. Hence, the VAR model does not aggregate hardware and
software investment.
The VAR model is as follows:
2. ( )
( ) ( )
where ( ) is the difference in the unemployment rate at time T between
occupations with skill measures i and k, I is a vector consisting of the log of aggregate
investment in hardware and software, Y is the log of real GDP and is the log of the
CPI less food and energy.
“Stuck in the Middle” by Shannon Cordes
20
V. Results
Probit Model
The first hypothesis predicts that individuals employed in routine
cognitive work incur a higher likelihood of unemployment than their
counterparts in non-routine cognitive occupations. If this hypothesis
proves to be true and if we suspect that advancements in digital
technology contribute to the result, then we would expect the shift of
unemployment likelihood in favor of non-routine cognitive jobs to occur
after 1977, when the Apple computer made its debut.7 Indeed, the probit
model shows that routine cognitive workers become more likely to be
unemployed than non-routine cognitive workers after the invention of
computers.
The coefficient of the dummy variable for routine cognitive labor,
all else constant, becomes consistently significant in 1983, two years after
the first IBM personal computer arrived on the market and at the
beginning of the decade in which computer usage rapidly expanded across
all industries. Figure 1.4 presents the marginal effect of the coefficient for
routine cognitive occupations from 1968-2010.
7 Source: Autor (1998).
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
21
Figure 1.4
From 1968-1983, the coefficient alternates between significance and non-
significance in the probit model, indicating that skill measure is not a
conclusive factor in predicting the probability an individual will be
unemployed from one year to the next. Post 1983, individuals employed
in routine cognitive occupations have a consistently higher probability of
unemployment by 1-2 percentage points. In other words, the
unemployment rate for routine cognitive occupations is 1-2 percentage
points higher than for non-routine cognitive occupations (both analytic
and interactive). This percent difference in the unemployment rate is
consequential – especially considering the unemployment rate fluctuates
within the bounds of 6 and 8 percent.
0
0.005
0.01
0.015
0.02
0.025
0.03
0.035
Year
Marginal effect of the routine cognitve coefficient
in the probit model, 1968-2011
“Stuck in the Middle” by Shannon Cordes
22
The only anomaly within the data occurs in the year 2000, when
the marginal effect of routine cognitive occupations is not statistically
significant. On March 10, 2000, the dot-com bubble crashed, sending the
job status of all Americans, regardless of skill type, into a tailspin of
uncertainty. CPS data is collected at the end of March. Thus in 2000,
CPS data was collected within weeks of the dot-com crash, which explains
why all jobs, regardless of skill measure, incurred the same probability of
unemployment.
The results of the probit model are consistent with the results of
Nedelkoska (2012), where German individuals experienced a higher
likelihood of unemployment given employment within a routine
occupation. Nedelkoska had access to a panel data set in Germany, which
allowed her to measure the probability of occupational changes of
individuals across time. As expected, individuals occupied in routine
work were more likely to switch occupations than those occupied in non-
routine work. When comparing the likelihood of unemployment vs. the
likelihood of occupational changes, individuals were significantly more
likely to switch occupations than to become unemployed. Here lies the
limitation of the CPS dataset: individuals cannot be tracked over time and
consequently, occupational changes are not captured in the probit model.
Although the probit model cannot capture the probability of occupational
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
23
changes by skill measure, it must be recognized that advancements in
digital technology do not necessarily displace a worker, but instead force
an individual to change occupation.
The question remains, however, whether the 1-2 percentage point
difference in the likelihood of unemployment between routine cognitive
and non-routine cognitive occupations is a result of advancement in digital
technology. It cannot be assumed on the basis of skill measure alone that
digital technology causes workers of routine cognitive occupations to face
a higher unemployment rate than non-routine cognitive workers. The
VAR model attempts to supplement the results of the probit model by
determining whether a statistical link exists between the unemployment
gap and digital technology.
VAR Results
The VAR model measures whether advances in digital technology
are predictive of an expansion in the unemployment gap between routine
cognitive and non-routine cognitive occupations. Digital technology has
been categorized as hardware and software investment, since we expect
the effect on the unemployment gap to depend on the type of technological
advancement. In the VAR model, the unemployment gap is defined as the
difference in the unemployment rate between routine cognitive and non-
“Stuck in the Middle” by Shannon Cordes
24
routine cognitive-analytic occupations. Figure 1.5 provides a visual
representation of the unemployment gap.
Figure 1.5
Although one can imagine that other unemployment gaps exist –
such as the gap between routine cognitive and non-routine cognitive-
interactive or routine manual and non-routine manual unemployment – to
avoid overcomplicating the empirical analysis, the unemployment gap will
only pertain to the difference in routine cognitive and non-routine
cognitive-analytic unemployment. Since Autor et. al (2003) concluded
that technological advancements had the greatest impact on the share of
employment of routine cognitive and non-routine cognitive-analytic
occupations, I will restrict my analysis to the unemployment gap between
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
25
those two skill measures. Furthermore, the empirical results of the other
unemployment gaps did not generate significant findings.
Table 1.1 in the Appendix presents the results of the Granger
Causality test ordered by Cholesky Factorization for the estimated VAR
model.8 Before examining the model’s estimated effects of hardware and
software investment, we must first determine whether the model exhibits
expected relationships consistent with macroeconomic theory. In the
VAR model, GDP is exogenous and significantly effects inflation,
meaning that an unexpected rise in GDP is associated with rising inflation.
Figure 2.1 displays the impulse response of the CPI to a shock in GDP.
Figure 2.1
The impulse response generates a shock at time zero in order to measure
the response of one variable to an unanticipated rise in another variable,
8 The VAR model passes the unit root test and is stable. Furthermore, the lag
exclusion test recommends the use of two lags.
“Stuck in the Middle” by Shannon Cordes
26
thus simulating the dynamic between the two variables. In Figure 2.1, an
unexpected stimulus in GDP is followed by a rise in the CPI. Contrary to
expectations, hardware and software investment do not function as
complements. Instead, the model indicates that the primary driver of
hardware and software investment is computer prices. This result is
consistent with the conclusions of Autor et. al (2003). As demonstrated in
the impulse response in Figure 3.1 and 3.2, an unexpected rise in prices is
historically followed by a sudden decrease in software and hardware
investment over the short run (approximately two years).
Figure 3.1
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
27
Figure 3.2
Once prices stabilize in the medium run, hardware and software
investment return to their previous levels. Since the VAR model exudes
the expected macroeconomic relationships supported by theory and
existing empirical analysis, the VAR model has the potential to capture the
dynamics between advancements in digital technology and the
unemployment gap.
As stated by the ALM hypothesis, computer technology serves as
substitutes for routine cognitive occupations and complements the work of
non-routine cognitive-analytic occupations. If this hypothesis is true, then
we would expect hardware and software investment to be associated with
an expansion of the unemployment gap. The results of the Granger
Causality test show that hardware and software investment, GDP and
inflation are all significant contributors to the unemployment gap between
“Stuck in the Middle” by Shannon Cordes
28
routine cognitive and non-routine cognitive-analytic occupations over
time.
The impulse response of the unemployment gap to a shock in
GDP indicates that the gap responds in alignment with the expansions and
contractions of the economy. During an economic expansion, the
unemployment gap contracts and the reverse is true during an economic
contraction. Moreover, the unemployment gap also exhibits the trade-off
between unemployment and inflation predicted by the Phillips Curve.
Figure 4.1 and 4.2 display the impulse response of the unemployment gap
to a shock in GDP and inflation, respectively.
Figure 4.1
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
29
Figure 4.2
However, the unemployment gap cannot solely be attributed to
the cyclical nature of the economy. Technological advancement
contributes to the gap’s persistence, yet not all technologies exacerbate the
unemployment gap to the same degree. Contrary to expectations, an
unanticipated rise in hardware and software investment leads to a
contraction of the unemployment gap in the short-run (from year zero to
year two), as displayed in Figure 5.1 and 5.2.
“Stuck in the Middle” by Shannon Cordes
30
Figure 5.1
Figure 5.2
However, in the medium run, the effect of hardware and software
investment on the unemployment gap differs. In response to a shock in
hardware investment, the unemployment gap marginally rises above zero
during the medium run (from year four to year seven). In contrast, a shock
in software investment leads to an expansion of the unemployment gap in
the medium run that exceeds the initial contraction in the short run. Thus
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
31
the net change in the unemployment gap following a shock in software
investment is positive, which means that advancements in software
technology are historically followed by increased unemployment for
routine cognitive workers and/or decreased unemployment for non-routine
cognitive-analytic workers.
In order to account for the differing effects of hardware and
software investment on the unemployment gap, two possible explanations
arise. First, unlike automated devices responsible for replacing routine
manual labor following the industrial revolution, the machine itself – the
personal computer – may not fully substitute workers for routine cognitive
tasks. The computer cannot produce output without software (hence they
are complementary goods). Perhaps the substitution value lies not in the
machine itself, but the application of the machine realized through
software.
The computer acts as a mechanism upon which software – the true
use-function of the computer – can run. Without the creation of software,
the widespread accessibility and usability of the machine would not have
been realized. For example, during the infancy of the World Wide Web, its
usability was diminished without a means of navigating its terrain. Until
Netscape created the first user-friendly web-browser, the webpages and
“Stuck in the Middle” by Shannon Cordes
32
information contained upon those pages went unread, like lone signs along
an unpaved highway.
Although the complementary nature of hardware and software
lends itself to producing a good or service, digital technology’s greatest
contribution is its connective power – the ability to connect individuals
and to create a seamless interlay of all units of a firm. As Bresnahan
(2000) explains, advancement in digital technology alters the cost-
effective structure and organization of a firm. For example, Business
Workflow software fundamentally changed what was considered the most
cost-effective scale of a firm, which led to large organization changes.
According to Bresnahan, large organizational changes, which often
include the decentralization of decision-making, lateral communication
and a greater emphasis on the need for autonomous workers, have a larger
effect on the acquisition of higher skilled labor than the technological
change alone.
Such mass organizational re-structuring does not come without a
price – not only money, but also time. Restructuring a firm in order to
incorporate advances in digital technology requires time. Consequently, a
time-delay effect occurs that postpones the efficiency gains following the
implementation of the organizational changes. The time-delay effect is
precisely what occurs in the impulse response of the unemployment gap to
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
33
a shock in software investment. As Figure 5.2 displays, the expansion of
the unemployment gap does not occur until two years after the initial
shock. Thus, the substitution effect of advances in digital technology on
routine cognitive workers is not immediate. Only after firms achieve the
implementation of new digital technology do computers begin to replace
routine cognitive workers.
The substitution effect captured in the VAR model most likely
underrepresents the magnitude of the actual substitution effect occurring
in the U.S. labor market. Since the model only includes the number of
unemployed workers for each skill measure, those workers partially
replaced by digital technology are not included. ‘Partially replaced’ refers
to occupations in which computers do not replace the entire worker, but
only a subset of the worker’s skills. Bresnahan (2000) labels this partial
replacement effect as the ‘limited substitutability’ of digital technology.
Due to the nature of the dependent variable, the model does not capture
those routine cognitive workers who experienced a subset of their tasks
replaced by computers. If the VAR model could capture those routine
cognitive workers partially replaced by digital technology, we would
expect the impulse response of the unemployment gap to be significantly
larger.
“Stuck in the Middle” by Shannon Cordes
34
VI. Conclusion
The replacement effect of technology upon labor is nothing new,
as this effect has occurred since the industrial revolution, during the
creation of the assembly line and now, through advancements in digital
technology. Previously, the Skill-Biased Technology Change hypothesis
provided a widely accepted explanation for the effect of advancing
technology upon the labor market: technology increases the demand for
skilled labor and decreases the demand for unskilled labor.
Yet, this explanation fails to explain the recent decline of middle
skilled labor in the last three decades – a decline that existing economic
literature has attributed to advancements in digital technology. The Autor
Levy Murnane (ALM) hypothesis provides a more nuanced view of the
effect of technology on the labor market by categorizing labor within
routine and non-routine occupations. Routine and non-routine
occupations can be further segmented by manual and cognitive
occupations. The ALM hypothesis states that digital technology behaves
as substitutes for routine cognitive occupations and as compliments for
non-routine occupations. Insofar as the ALM hypothesis accurately
explains the dynamic of the labor market and digital technology, the
substitution effect of digital technology impacts a large share of the labor
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
35
market, since routine cognitive occupations are concentrated within the
middle class.
I tested two hypotheses: 1) Individuals employed in routine
cognitive occupations incur a higher probability of unemployment than
individuals employed in non-routine cognitive occupations and 2) To what
extent the unemployment gap between routine cognitive and non-routine
cognitive occupations can be attributed to advancements in digital
technology. The probit model confirmed that since the 1980s, routine
cognitive workers are more likely to become unemployed than non-routine
cognitive workers. In order to determine whether the unemployment gap
between routine cognitive and non-routine cognitive occupations can be
statistically linked to advancements in digital technology, a Vector
Autoregression Model (VAR) measured to what extent expansions in the
unemployment gap can be explained by advancing digital technology.
Using hardware and software investment as a proxy for digital
technology, the results of the VAR model conclude that the effect of
digital technology upon the unemployment gap depends on the type of
technology. Advancements in hardware technology are statistically
significant, but the net effect on the unemployment gap over time is zero.
In contrast, an advancement in software technology leads to a contraction
of the unemployment gap in the short run, but an expansion of the gap in
“Stuck in the Middle” by Shannon Cordes
36
the medium run that exceeds the initial contraction. Thus, the net effect of
software technology on the unemployment gap is positive, which means
that advancements in software technology are historically followed by an
increase in unemployment for routine cognitive occupations and/or a
decrease in unemployment for non-routine cognitive occupations.
The implications of these results indicate that as software
technology advances, a greater number of routine cognitive occupations
will either be fully displaced or partially displaced by digital technology.
Partial displacement refers to the replacement of a subset of skills required
within an occupation. Future research should be concerned with how to
transition workers with middling skills towards higher skilled occupations
that are complimented – not substituted – by digital technology. Such
transition efforts in the form of job training, education and skill-upgrading
programs will be of greatest importance not for new entrants of the labor
market (such as college graduates), but for existing laborers. As digital
technology –especially software – advances at an increasing rate, our
cultural expectation of a ‘lifetime’ career may be subject to evolution. In
the near future, the norm may no longer be to remain in one occupation
until retirement, but rather to reinvent our careers multiple times in order
to adapt to a labor market, economy and world buffeted by constant waves
of digital technological advancements.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
37
References
Acemoglu, Daron. "Changes in Unemployment and Wage Inequality: An
Alternative Theory and Some Evidence." The American Economic
Review 89.5 (1999). JSTOR. Web. 9 Jan. 2014.
Acemoglu, Daron. "Why Do New Technologies Complement Skills?
Directed Technical Change and Wage Inequality." The Quarterly
Journal of Economics 113.4 (1998). JSTOR. Web. 9 Jan. 2014.
Arico, Fabio R. "Both Sides of the Story: Skill-biased Technological
Change, Labour Market Frictions, and Endogenous Two-Sided
Heterogeneity." Scottish Institute for Research in Economics
(2009). Web. 6 Jan. 2014.
Autor, David H., Lawrence F. Katz, and Alan B. Krueger. "Computing
Inequality: Have Computers Changed the Labor Market?" The
Quarterly Journal of Economics 113.4 (1998). JSTOR. Web. 6 Jan.
2014.
Autor, David H., Frank Levy, and Richard J. Murnane. "The Skill Content
of RecentTechnological Change: An Empirical Exploration." The
Quarterly Journal of Economics 118.4 (2003). JSTOR. Web. 24
Sept. 2013.
Bresnahan, Timothy F., Erik Brynjolfsson, and Lorin M. Hitt.
"Information Technology, Workplace Organization, and the
Demand for Skilled Labor: Firm-Level Evidence." The Quarterly
Journal of Economics 117.1 (2002). JSTOR. Web. 6 Jan. 2014.
Caselli, Francesco. "Technological Revolutions." The American Economic
Review 89.1 (1999). JSTOR. Web. 9 Jan. 2014.
Goos, Maarten, and Alan Manning. "Lousy and Lovely Jobs: The Rising
Polarizationof Work in Britain." The Review of Economics and
Statistics 89.1 (2007).
JSTOR. Web. 23 Sept. 2013.
“Stuck in the Middle” by Shannon Cordes
38
"Graph: Consumer Price Index for All Urban Consumers: Personal
computers and peripheral equipment." Economic Research:
Federal Reserve Bank of St. Louis. Federal Reserve Bank,
n.d.Web. 24 Feb. 2014.
<http://research.stlouisfed.org/fred2>.
Hornstein, Andreas, Per Krusell, and Giovanni L. Violante. "The
Replacement Problem in Frictional Economies: A Near-
Equivalence Result." Federal Reserve Bank of Richmond. Federal
Reserve Bank of Richmond, Apr. 2005. Web. 29 Oct. 2013.
Katz LF, Autor DH. Changes in the Wage Structure and Earnings
Inequality. In: Ashenfelter O, Card D Handbook of Labor
Economics, vol. 3A. ; 1999. pp. 1463-1555.
Kramarz, Francis. "Computer's and Labour Markets: International
Evidence." The United Nations University World Institute for
Development Economics Research (1998). JSTOR. Web. 10 Nov.
2013.
Machin, Stephen, and John Van Reenen. "Technology and Changes in
Skill Structure:Evidence from Seven OECD Countries." The
Quarterly Journal of Economics 113.4 (1998). JSTOR. Web. 9 Jan.
2014.
Miriam King, Steven Ruggles, J. Trent Alexander, Sarah Flood, Katie
Genadek, Matthew B. Schroeder, Brandon Trampe, and Rebecca
Vick. Integrated Public Use Microdata Series, Current Population
Survey: Version 3.0. [Machine-readable database]. Minneapolis:
University of Minnesota, 2010.
Nedelkoska, Ljubica. "Occupations at risk: The task content and job
stability." Jena Economic Research Papers (2012). Web. 24 Feb.
2014. <www.jenecon.de>.
Rubart, Jens. "Heterogeneous Labor, Labor Market Frictions and
Employment Effects of Technological Change: Theory and
Empirical Evidence for the U.S. and Europe." Darmstadt
Discussion Papers in Economics 158 (2006). Web. 9 Jan. 2014.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
39
"Timeline of Computer History." Computer History Museum. Ed. Ganna
Boyko and Edward Lau. N.p., 2014. Web. 11 Feb. 2014.
<http://www.computerhistory.org/>.
U.S. Bureau of Economic Analysis, “Consumer Durables, ” www.bea.gov
(accessed January 14, 2014).
“Stuck in the Middle” by Shannon Cordes
40
Appendix
Table 1.1: VAR Granger Causality Test Results
The table presents the p-values for the Granger Causality Test.
Independent Variable
log(GDP) log(CPI) log(Hardware) log(Software)
URC - UNR-C/A
log(GDP) *** - - - -
log(CPI) 0.0256** *** - - -
log(Hardware) - 0.0015*** *** - -
log(Software) - 0.0091*** - *** -
URC - UNR-C/A 0.0643* 0.0536* 0.0434** 0.0252** ***
Adjusted R-Squared 0.74 Observations 34
URC - UNR-C/A refers to the difference in the unemployment rate between routine cognitive (RC) and non-routine cognitive-analytic occupations.
Figure 6.1
Source: Federal Reserve Bank
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
41
Appendix Table 1.2: Probit Model Output
Year Observations McFaddenR-squared
1968 42775 0.1137
1969 43535 0.1072
1970 42267 0.0959
1971 42831 0.1145
1972 41590 0.1214
1973 41453 0.1013
1974 41266 0.1065
1975 40798 0.1315
1976 43277 0.1238
1977 52652 0.1129
1978 52062 0.1207
1979 52991 0.1090
1980 63459 0.1093
1981 64275 0.1286
1982 57580 0.1267
1983 57608 0.1355
1984 64670 0.1301
1985 65713 0.1367
1986 64571 0.1214
1987 64573 0.1263
1988 64831 0.1256
1989 60767 0.1120
1990 66902 0.1095
1991 66737 0.1072
1992 66042 0.1177
1993 65414 0.1131
1994 63469 0.1205
1995 63356 0.1048
1996 55167 0.1049
1997 56402 0.1144
1998 56709 0.1100
1999 57021 0.0904
2000 53425 0.0988
2001 87056 0.0937
2002 85818 0.0818
2003 83782 0.0761
2004 81627 0.0866
2005 80067 0.0864
2006 79311 0.0799
2007 78511 0.0842
2008 78099 0.0906
2009 77982 0.0913
2010 76535 0.1025
2011 73380 0.1027
“Stuck in the Middle” by Shannon Cordes
42
Combating Obesity in America- Fat Tax Is Not the Way To Go Kelly Tomera
Given the alarming increase in the rate of obesity in America, it is
no surprise that it has become one of the main concerns of the healthcare
industry. Since many expensive procedures result from obesity among
middle age or elderly Americans, the anticipated future Medicare costs are
of large concern (Daviglus 1). Furthermore, at this late stage in life,
policies that change habits to a healthier lifestyle are extremely hard to
implement, forcing many policy makers to take a stab at preventative
measures like taxes (Daviglus 2). The media has commonly titled these as
“fat taxes.” While they have not been implemented in the U.S. yet (NYC’s
soda tax has been stalled), many health advocates strongly endorse them
(Petrecca 2). The main argument for a national fat tax is that taxes are the
most efficient way to alter our behavior (Miao). They hurt us where we
feel it most—our wallet. While this may be true for most goods, due to
historically weak elasticities of demand among targeted fatty foods,
variations on what constitutes “bad” food, and nonexistent national
support for fatty food taxes, combating obesity using a fat tax will not be
successful in the United States.
Recent history has demonstrated the weak elasticities of fatty
foods. Other nations have implemented fat taxes under different names.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
43
Most notably, Denmark was the first European nation in 2011 to enact a
fat tax (Kliff 1). They did not tax certain brands or specific food groups.
Rather, they across the board taxed anything with a saturated fat content
greater than 2.3% (Kliff 1). Products affected ranged from fancy cheeses
to bacon. The tax was not very effective for two reasons. First, Danes
merely drove across the boarder a few times a year to load up on fatty
foods in neighboring countries. Thus, their purchasing power was not
affected; instead, the local stores in Denmark felt the burden of the tax,
and the economy suffered (Kliff 2). Although most Americans would not
have this luxury, unless they lived near Canadian or Mexican borders, the
fact that the tax still generated revenue suggests that people continued on
their unhealthy diets. Secondly, the tax had varying results when it came
to fatty meats. The tax was imposed logically per carcass. However, this
resulted in higher prices for fatty burgers as well as lean steaks (“A Fat
Chance”). Although the tax was abandoned a year later, we can speculate
that since both meat groups were affected, consumers could not afford to
buy leaner high-end meats that were now even more expensive. Instead
they purchased even more of the poor quality cheaper cuts, having an
adverse affect on their health.
After a year of discontent from the people, Denmark’s government
caved and repealed the tax. It is interesting that the tax was so short lived
“Combating Obesity in America” by Kelly Tomera
44
since Europeans are used to much higher tax rates than the United States.
The European Average tax-to-GDP ratio is 38.8%, and Denmark’s is at
47.7% (European Commission). Compare this to the mere 25.2% tax-to-
GDP ratio in the U.S (European Commission). Still, the tax seemed just to
be another nuisance with no long-term effects. For a nation with a
population of 5.5 million people, the tax raised $200 million in revenues,
which shows that people still bought fatty foods (“Denmark” & “Taxation
Trends In The European Union”). To put this in perspective, Denmark’s
tax revenue from products and imports generated $3.8 billion Euros that
same year (“Taxation Trends In The European Union”). Thus, in less than
a full year, the fat tax comprised roughly 4% of this revenue category for
Denmark. Undeniably it would have comprised an even larger sector if
the majority of people had not spent their money in neighboring countries.
This is saddening, as the purpose of the tax was not to fund the Danish
government, but to instill healthier incentives in the population.
Denmark’s failed fat tax, demonstrates the strong purchasing power of
consumers when it comes to enjoyable food.
Furthermore, Denmark is not the only country with poor results
after creating a fat tax. France implemented taxes on sweetened beverages
in 2011 as well, and on some foods such as Nutella in later months
(Watson 2). Researchers Olivier Allais, Patrice Bertail, and Veronique
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
45
Nichele found that this fat tax in France had little effect on purchases
made by French households (4). For purchasing power to be affected,
recent studies have shown that taxes must be very large. Some articles
suggest the tax must be 20% or more to impact behavior on unhealthy
foods (Campbell 1). This large amount is hard to implement on goods,
since severe changes in industries like food are difficult for society to
accept. These drastic findings paired with the historical failures of
Denmark and France, show that for many countries fat taxes are simply
not effective.
While elasticities of demand seem to be low for food consumption,
what makes a fat tax even more difficult is the varying opinion on what
constitutes unhealthy nutrition. There really are no bad foods. Yes, some
nutrients should be eaten in moderation, but even fats can good for our
diets. Humans need various foods in order to have a balanced diet, and
nutrition needs to be taken from a more holistic approach. In the 1980s the
fat free movement started, yet, in the same decade, more Americans
became overweight (Nestle, Goldberg, Willet, and Taubes 2). Food had
more processed carbohydrates, and people replaced fat with these. Thus,
“fat free” on a label does not necessarily imply good nutrition. Imposing a
fat tax could potentially shift production of foods in America even further.
“Combating Obesity in America” by Kelly Tomera
46
Companies may begin replacing fat with ingredients that are possibly even
worse, like high caloric sugars.
Defining “bad” foods will never be easy, as it varies by person.
This is one reason why fat taxes are so disputable. Not only were there
unexpected mishaps with Denmark’s fat tax, but also it may have gone too
far. Some saturated fats can be good for you. Saturated fats are made up
of short chains, so they do not stack up in your blood easily and can be
beneficial for your metabolism. There is a difference however between
plant based saturated fats, and animal based saturated fats, which can raise
cholesterol (Kounaves 2). Plant based saturated fats are now heralded by
some as beneficial. In a study published by the American Medical
Association, researchers found that randomized trial groups that ate more
plant fats had lower lipoprotein levels (LDL) than those eating minimal
saturated fats. LDL is associated with cardiovascular disease, and low
levels are desired (Foreman). Due to studies such as this, coconut oil, for
instance, has been gaining recent hype among health fanatics. This is just
one example of the varying opinions on what constitutes “bad” food.
Even leading health organizations cannot agree on food standards.
The U.S. Institute of Medicine conducted a review stating that “25% of
energy from caloric sweeteners is acceptable”, while the World Health
Organization concluded that only 10% of energy from caloric sweeteners
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
47
is acceptable for nations with high income levels such as the United States
(Popkin 5). These notable organizations draw from much of the same data
and leading scientific studies, yet arrive at different conclusions. This
exemplifies the varying opinion regarding the latest recommendations for
nutrition that leads to much confusion among consumers. This confusion
about what qualifies as bad nutrition, poses a challenge when justifying fat
taxes.
While a lack of consensus on nutrition makes preventing obesity
difficult, the lack of national support in the United States on the
seriousness of obesity makes tackling the subject even harder. National
support for fat taxes is virtually nonexistent due to the political
atmosphere surrounding the issue. The government grants many resources
towards fatty food sectors, and with much bias. For instance, 60% of all
agricultural subsidies go towards meat and dairy production—foods
notoriously high in fat content—while less than 1% of farming subsidies
are granted to fruit and vegetable growers (Physicians Committee For
Responsible Medicine 1). A national fat tax would affect several
industries already stimulated by the government, resulting in intense
lobbying from these large corporations.
Not only is lobbying at the federal level dominating the discussion,
but also legislation at the state level is opposed due to their constituents.
“Combating Obesity in America” by Kelly Tomera
48
When New York City mayor Bloomberg attempted to pass a large soda
ban in his state, Mississippi (a state with the highest obesity rate in the
U.S) signed a bill prohibiting any future laws from limiting what
Mississippians eat or drink (Ford, Sutton, and Yan 2). The New England
Journal of Medicine captures this libertarian spirit perfectly stating, “The
American emphasis on the value of individual responsibility creates a
reluctance to intervene in what are seen as personal behavioral choices”
(Schroeder). Furthermore, when it comes to taxes, states have the most
influence. For example, taxing cigarettes as well as banning smoking in
public places in the United States, were sparked at the state level, not the
federal (Klein & Dietz 391). The already growing fear from states such as
Mississippi as well as others manifests the lack of support for national fat
taxes in the U.S.
Fat taxes are also concerning to states and people because they are
a regressive tax and the economic tax incidence falls on workers, whether
they are obese or not. In the United States as well as many countries, those
in lower income brackets purchase more fast food (Popkin 4). Fatty food
tastes better and more importantly is significantly less expensive.
Imposing a fat tax would hurt this group’s disposable income
considerably, and therefore can never gain full support without large
subsidies granted on healthier fruits and vegetables by the government.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
49
In addition as a regressive tax, fat taxes can hurt those who do not
even purchase fatty foods. The 328,000 employees at Nestle, 142,000
employees at the Coca-Cola plant in Georgia, 297,000 employees at
PepsiCo, 34,000 employees at Dole Food Company, 35,000 employees at
General Mills, and the other numerous workers at food corporations in
America would all suffer (9). Taxing fatty foods creates a decreased
demand for these products and hurts the economic prosperity of the
workers at these plants.
Along with disrupting these markets, taxing fatty foods faces
another problem: there is no negative externality associated with obesity.
Other than maybe the occasional inconvenience of being squeezed into
your airplane seat because an obese person is too large to fit comfortably
in their own, there really are no dire impacts of obesity on others.
Although some suggest that obese people place a burden on others with
their large health care costs, it is difficult to pinpoint a disease on obesity
when other factors may be at play. As a counter example, energizing the
nation against preventative measures for smoking was easy because
breathing in second hand smoke harms others. Economists Jonathan Klein
and William Dietz, emphasize in their paper “Childhood Obesity: The
New Tobacco,” the fact that successful health movements such as tobacco
9 These statistics can easily be found at the respective companies websites
“Combating Obesity in America” by Kelly Tomera
50
control measures were successful because they had the “mobilization of
grass-roots” to establish a threat within the nation (Klein & Dietz 389).
Although there are some movements in the United States to decrease
obesity, the majority are captured by the private sector. Take
WeightWatchers and Curves for example, both of which are lucrative
private businesses. Our government has not supplied these services, since
the public has not demanded them—at least not on a large-scale level. In
fact, the government is funding organizations that help ease acceptance of
obesity. The National Association to Advance Fat Acceptance (NAAFA)
was established in 1969; implemented into society early on as a founding
step towards social acceptance of obese people. As well, attitudes in
America have changed from fawning over thin supermodels, to demanding
more plus sized models (Olson 1). Dove’s campaign on “Real Beauty”
over the past few years has emphasized the curves and fat of everyday
woman and portrayed them as both lovely and alluring. Today, popular
groups on Pinterest and Tumblr are popping up under the name “Fat
Power”, encouraging women to embrace their love handles with catchy
and comical phrases. Groups and organizations such as these demonstrate
that there is not a strong social disapproval towards obesity in America.
In order to initiate disapproval, Americans need solid science to
back up policy. The research on what foods cause obesity, as mentioned
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
51
earlier, is simply not there. Imposing a fat tax will never bear the urgency
it requires, without the scientific research first. The prevention of
smoking was not successful until a plethora of scientific studies were
released (Klein, Jonathan, & Dietz 1). Causality between foods and
obesity have not been solidified, evoking an irresolute attitude within the
nation. Currently, America simply does not have the national support
when it comes to obesity measures, making fat taxes inherently faulty
measures to adopt at this time.
The World Health Organization ranked the U.S. as the 8th
most
obese nation, and we are one of the few countries at the top without
cultural fattening ceremonies (“World's Top 10 Most Obese Countries”).
America is starting to realize that obesity is a problem, yet it is not
perceived as an immediate threat. Historically weak elasticities among
targeted fatty foods, variations on what constitutes “bad” food, and
nonexistent national support for fat food taxes, demonstrate that
combating obesity using a fat tax will not be successful in the United
States. These reasons evince the need for reforms other than taxes when it
comes to preventing and reducing obesity rates in America.
“Combating Obesity in America” by Kelly Tomera
52
Sources
“A Fat Chance.” The Economist. 11. 17 (2012) : 1-4. Print.
Allais, Olivier, Veronique Nichele, and Patrice Bertail. "The Effects of a
Fat Tax on French Households' Purchases: A Nutritional
Approach." American Journal of Agricultural Economics 92.1 (2010) : 1-
10. Print.
Campbell, Denis. "'Fat Tax' on Unhealthy Food Must Raise Prices by 20%
to Have Effect, Says Study." The Guardian. Guardian News and Media,
12 May 2012. Web. 12 May 2013.
Daviglus, Martha L. "Health Care Costs in Old Age are Related to
Overweight and Obesity Earlier in Life." Health affairs 24 (2005):
W5R97-100. ProQuest. Web. 3 Apr. 2013.
“Denmark.” Central Intelligence Agency. N.p., n.d. Web. 9 May 2013.
European Commission “Taxation Trends In The European Union: Data
For EU Member States, Iceland, /7 Norway.” Eurostat. (2013) : 4-99.
Print.
“Fat Power.” Pinterest. N.p., n.d., Web. 3 May 2013
Foreman, David. “Foods to Lower Cholesterol.” Wall Street Journal. 24
Aug 2011. Web. 4 May 2013.
Ford, Dana, Joe Sutton, and Holly Yan. "No Soda Ban Here: Mississippi
Passes 'Anti-Bloomberg' Bill." CNN. Cable News Network, 01 Jan. 1970.
Web. 9 May 2013.
Klein, Jonathan D., and William Dietz. "Childhood Obesity: The New
Tobacco." Health affairs 29.3 (2010): 388-92. ProQuest. Web. 3 Apr.
2013.
Kliff, Sarah. “Denmark Scraps World’s First Fat Tax.” Washington Post.
13 Nov. 2012. Web. 20 Apr. 2013.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
53
Kounaves, Samuel. “Eliminating Animal Fats Lowers Cholesterol.”
Chemical & Engineering News. 40.16 (2012) : 1-5. Print.
Miao, Zhen. "Three Essays on Tax Policies Addressing the Obesity
Epidemic and Associated Calorie Intake." ProQuest. Web. 16 May 2013.
Nestle, Marion, Walter Willet, Gary Taubes, Dean Ornish, and Jeanne
Goldberg. "Did the Low Fat Era Make Us Fat?" PBS. PBS, 8 Apr 2004.
Web. 16 May 2013
Olson, Hannah. "Beyond “Plus Size”: Why The Natural Model Movement
Matters For Everyone." Blisstree RSS. N.p., 11 May 2013. Web. 12 May
2013.
Petrecca, Laura. "Judge Blocks NYC Large Soda Ban; Bloomberg Vows
Appeal." USA Today. Gannett, 11 Mar. 2013. Web. 4 Apr. 2013.
Popkin, Barry M. "Will China's Nutrition Transition Overwhelm its
Health Care System and Slow Economic Growth?" Health Affairs 27.4
(2008): 1064-76. ProQuest. Web. 15 May 2013.
Schroeder, Steven. “We Can Do Better — Improving the Health of the
American People.” New England Journal of Medicine. 377.10 (2007) :
1221-1228. Web. 14 Apr. 2013.
"Taxing America's Health: Subsidies for Meat and Dairy
Products." PCRM. N.p., May 2011. Web. 16 May 2013.
Watson, Leon. "France Approves Fat Tax on Sugary Drinks Such as Coca-
Cola and Fanta." Mail Online. N.p., 29 Dec. 2011. Web. 16 May 2013.
"World's Top 10 Most Obese Countries (PHOTOS)." GlobalPost. N.p.,
n.d. Web. 10 May 2013.
“Combating Obesity in America” by Kelly Tomera
54
US Stock Market as a Leading Indicator of the US Economy
Ryan Johnsrud
Abstract
The most recent stock market crash began slightly prior to the
recession of late 2007. Did the stock market crash predict the heavy
decline in economic activity? This paper deals with this question more
broadly, asking if the stock market is a leading indicator of the economy.
There are two main arguments in favor of the stock market as a leading
indicator of economic activity. The first is that stock prices already factor
in expectations about the future and so give hints about what is to come.
The second is the wealth effect theory, which states that when stock prices
rice, investors are wealthier and spend more, increasing consumption and
thus GDP. Using carefully selected macroeconomic data to present an
overall picture of US economic activity and the S&P 500 index to
represent the broader US stock market, this study also attempts to show
what state the economy would have been in between 2008 and 2011 had a)
the stock market decreased less violently, or b) the stock market not
crashed at all. A vector autoregressive (VAR) system is estimated with
this data, and subsequent conditional forecasting is performed using
different scenarios of changes in the S&P 500. This paper finds that the
stock market is indeed a leading indicator of the US economy, and the
Great Recession would not have been so great had the stock market not
crashed so intensely.
1. Introduction
The Great Recession forced macroeconomic and stock market
information to the front pages of newspapers and websites everywhere, to
the forefront of the 2008 presidential debate, and into everyday
conversations across the country. The National Bureau of Economic
Research (NBER) announced that the recession officially began in
December 2007 and ended in June 2009. Another economic data source,
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
55
Fedreal Reserve Economic Data (FRED) reports that the Standard and
Poor’s (S&P) 500 Index declined 57% from its peak in October 2007 and
bottomed out in March of 2009.10
Families saw college and retirement
funds dwindle in value, corporations saw their market capitalizations
decrease substantially, unemployment levels climbed rapidly. In sum, US
economic growth stalled.
Because the stock market (represented by the S&P 500 index)
began to quickly decline from its October ’07 peak, should economists,
politicians, corporations, and average Americans have known the
recession was coming? In other words, if the stock market crashes, is
heavy economic contraction imminent? This paper attempts to answer the
question: is the stock market a leading indicator of US economic activity?
The paper also examines Granger causality of different macroeconomic
and financial measures, and out-of-sample forecasts of US economic
activity by a simple model constructed using coefficients from the
estimated VAR. Conditional forecasting is used to provide a view into
macroeconomic health if 1) the stock market not crashed in late 2007 and
2) the stock market crash not been as severe.
Much has been written about the stock market as a leading or
lagging indicator of US economic expansion/contraction. Many of the
10 See Appendix A for an explanation as to why S&P 500 is used to represent the whole
stock market.
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
56
arguments as to why stock price movements are a leading indicator center
around two key points:
1) valuation of stock prices based on future dividend
expectations, and
2) the “wealth effect.”
The Dividend discount model (Gordon 1959) values a stock price by
estimating the value of all future dividends (cash returned to shareholder)
and calculating their present value. This simple way of valuing a share
price factors in future dividend expectations. Increasing profitability
indicates increasing cash balances in corporate America, and is often
followed by increased dividend payouts. Since stock prices reflect
expectations about future profitability, changes in stock prices are thought
to reflect changes in future profitability, and profitability is directly linked
to economic activity. However, dividends are never certain until the
shareholder receives the money, so expectations surrounding future
dividends are subject to human error. Some US corporations have upwards
of 30 equity research analysts at financial firms who are paid (very well in
some cases) to cover the company’s every move and recommend its stock
with a “buy,” “hold,” or “sell” rating. So dividend expectations can be
reliable, but not perfect, and thus stock market movements have the
potential to mislead the direction of the US economy.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
57
There are also professionals around the world who are paid (also
very well) to analyze the broad US stock market and make predictions
about what will happen in the near- and long-term future. Financial firms
then sell this advice to fund managers, portfolio managers, and the general
public alike. Despite all of this coverage on the stock market, investors are
the ones who drive the market. Average, everyday investors buy and sell
stocks as do managers of large endowments and mutual funds. This is
where the human error concept comes in. The psychology of investing is
very complex; fear and greed are the two key emotions that motivate many
trades (both smart and foolish).11
This human error is why expectations of
future share price of overall stock market movements, while generally
correct, can be misleading. Comincioli cites the stock market crash of
1987 as evidence. The drop in stock prices falsely predicted the fate of
economic growth, as the economy continued to grow until the early 1990s.
Similarly, this author cites a study by Robert J. Barro in 1989 which
revealed that stock prices predicted three recessions (1963, 1967, 1978)
that did not occur. The stock market is not a perfect leading indicator.
Pearce (1985) cites the “wealth effect” as another argument as to
why the stock market could potentially lead the direction of the economy.
Theory suggests that as stock prices increase, the value of investors’
11 For more information on fear and greed in the financial markets, see Bernstein (2009).
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
58
portfolios increases, thus investors become wealthier and spend more.
This causes a decrease in consumption (C), and (because consumer
spending is a hefty portion of total GDP) in economic activity. On the
other hand, when stock prices decrease, investors become less wealthy,
therefore they spend less, and the economy contracts. Poterba and
Samwick (1995) investigated this theory, and found clear evidence that
stock price changes foreshadow growth in consumer spending, in
particular spending on consumer durables (goods that don’t wear out
quickly: cars, furniture, household appliances, cellphones).
This paper uses US economic and stock market index data to
investigate the stock market as a leading indicator of the broad US
economy. A vector autoregressive (VAR) model is used to determine
Granger causality. Simple models are estimated, and then conditional
forecasts are made to compare the baseline scenario to what would have
happened had the stock market not crashed in late 2007. Section two
discusses prior literature on leading indicators (including the S&P 500
index and the NYSE Composite index), section three outlines econometric
methods employed, section four describes data used, section five outlines
results and interpretations, and section six concludes.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
59
2. Literature Review
There has been substantial econometric research performed in
order to gauge the appropriateness and accuracy of several economic and
financial measures as leading indicators of the US economy as a whole.
This type of research, and the models that emerge from it can be extremely
helpful in forecasting the economic performance of the country. Some
authors include discussions about multiple leading indicators combined
into one indicator with very good predictive accuracy, while others merely
consider one economic indicator, whether it’s stock prices, money supply,
weekly manufacturing hours, interest rate spreads, etc. For the purposes of
this paper, the focus is on past literature that concerns stock prices as an
economic indicator alone or combined with other economic or financial
measures.
The papers reviewed in this section use different econometric
models and concepts to investigate stock prices as a leading indicator of
future economic activity. Estrella and Mishkin (1998) utilize a probit
model, where the dependent variable can take on only two possible values.
In this probit model, the two possible values are
0 – not in recession
1 – in recession.
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
60
The authors also develop a forecasted probability of a future recession; a
probability that is of great interest to politicians, policymakers,
corporations, small-business owners, college students, and families across
the country. On the other hand, Stock and Watson (1989) utilize a VAR
(Vector Autoregressive) system and Granger causality to test multiple
financial and economic measures (including stock prices) as leading
indicators. Comincioli (1996) also uses the causality test originally
proposed by C.J. Granger in 1969 to see if changes in stock prices
“Granger cause” changes in economic activity. Comincioli focuses solely
on the stock market as a leading indicator, ignoring the appropriateness of
other factors as leading indicators.
Nearly every piece of literature relating to the movement of stock
prices as a leading indicator for the economy comments on the reasons
why stock prices are a possible leading indicator. Most papers comment
on the traditional valuation of stock prices (the expected discounted values
of future dividend payments) and the wealth effect, but Stock and Watson
offer another reason as to why the stock market could be a leading
indicator of economic activity. They discuss the role of stock prices as a
determinant of the cost of capital. This theory suggests that changes in a
corporation’s share price leads to changes in its capital structure, which in
turn can either increase of decrease the cost of capital. If a company
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
61
decreases its equity financing relative to its debt financing, the overall cost
of capital decreases, because equity is more expensive than debt. If capital
is less expensive, business spending increases, and the economy most
likely expands. And the leading indicator in this case would be a change in
stock prices.12
Similar results are achieved in the three papers discussed above.
Comincioli employs a Granger causality test to determine the leading
indicator effect. First, the author establishes a relationship between GDP
and the S&P 500 by regressing the % change in GDP over the percent
change in the S&P 500 lagged 6 quarters. 3 lags are statistically
significant, and positively related to GDP (the economy). Next, the author
estimates an unrestricted and a restricted OLS model and uses an F-test to
determine if the lagged S&P 500 terms belong in the regression when the
lagged GDP terms are in the equation with the % change in GDP as the
dependent variable. The conclusion is that the coefficients on the lagged
S&P 500 terms are not zero, and that stock prices do Granger cause the
economy. After interchanging the S&P 500 terms with the GDP terms and
running the same F-test, the F-statistic is not large enough to reject the null
hypothesis that the %GDP coefficients are 0. This result suggests the
economy does not Granger cause the stock market.
12 Benninga (2006).
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
62
Stock and Watson conclude that the growth of the S&P 500
Granger causes changes in the “state of the economy.”13
However, this
result is obtained from a simple regression of the state of the economy
variable on 12 lags of the growth in the S&P 500 index. Using a VAR
including other elements of the LEI (index of Leading Economic
Indicators), growth in the S&P 500 has no marginal predictive content for
the state of the economy. So the stock market is in fact a leading indicator
of the state of the economy, however examining other factors can capture
the expectations inherent in the stock market.
Estrella and Mishkin focus on variables predicting a recession
rather than predicting quantitative measures of future economic expansion
or contraction. Using their probit model, the authors come to similar
conclusions to those of the previous two papers, examining out-of-sample
results. As an example, take Q2 1971. The model is estimated using data
from the beginning of the sample up until the second quarter of 1970, and
then a forecast of recession or not is made for the second quarter of 1971.
The authors conclude that stock prices are indeed useful predictors,
particularly one to three quarters ahead. Estrella and Mishkin explain that
they are not proposing the stock market as a replacement to more complex
13 A measure of the CEI – the Coincident Economic Indicator index: industrial
production, personal income, manufacturing and trade sales, and employee-hours in non-
agricultural establishments.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
63
macroeconomic forecasting models, but that the stock market can serve as
a simple check.
3. Data
United States economic data is tracked very closely by various
government and non-government organizations. Investors, corporate
boardrooms, small business owners, college students, and more anxiously
await the release of weekly, monthly, quarterly, and yearly data, as it
offers a glimpse of the current health of the economy. Indices and
measurements of housing prices, unemployment insurance claims, GDP,
etc. are frequently reported. This data is readily available (some data
requires a small fee) from online databases. Federal Reserve Economic
Data (FRED), maintained by the Federal Reserve Bank of St. Louis is a
database that contains data obtained from agencies like the U.S. Census
and the Bureau of Labor Statistics (BLS). The US Bureau of Economic
Activity is another government agency that provides timely and relevant
economic data to its users, who then report it, use it in forecasting models,
and compare how their forecast compares to the actual number for the
reported period.
Stock market data is also very widely and readily available. A visit
to Yahoo! Finance offers archived stock prices and index values many
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
64
years into the past. The S&P 500 index is one of the most quoted indices
in the United States and in the world, because is serves as a national index
for the US stock market. S&P 500 index data was retrieved from the same
place as all of the macroeconomic data.
Data used in the VAR systems for this paper is all conveniently in
one place and was fairly easy to compile. The St. Olaf College Economics
department purchases data from Haver Analytics in the form of an EViews
database. The U.S. Economic Statistics (Haver) database contains
economic and financial data provided by government, public, and private
organizations. This database contains over 40,000 time series for the US
economy, everything from housing and construction to international trade
and business cycle indicators, dating back to 1947 in some cases.
Economic data used in the estimation of the VAR systems and
conditional forecasting is quarterly. Included in this data are GDP (Gross
Domestic Product), unemployment rate, CPI (Consumer Price Index),
trade-weighted value of the USD (United States dollar) versus major
currencies, exports of goods and services, M2 (money supply), and finally
the S&P 500 index. This combination of economic measures was chosen
to present an appropriate overall picture of the economy. Some papers
may use GDP alone to represent economic activity, but in this analysis
more data was deemed necessary for proper estimation.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
65
GDP was included in the estimation because it is an overall
measure of economic activity. Recessions and depressions are
characterized by certain movements of GDP over time, and therefore it is
a vital measure of US economic health. Many aspects of the economy are
reflected by GDP in some way, as the formula includes Consumer
Expenditures, Investment, Government Expenditures, and Net Exports.
Unemployment rate was used to represent the health of the job
market. This is a big topic for college students as they enter the workforce.
They see first-hand how difficult it was to find a job during and
immediately following the recession of late ‘07 – early ‘09. The Consumer
Price Index was included to get a glimpse into price levels in the US and
see what all this means for inflation.
The trade-weighted value of the USD versus major currencies is a
good barometer of how much the national currency in this country is
worth when compared to those of other major players in the world
economy. In calculating this value, more weight is assigned to currencies
of nations with whom the US engages in more export/import relations.
Major currencies in this comparison belong to the European area, Canada,
China, Japan, Mexico, United Kingdom, and Australia. Along with the
value of the US dollar compared to other major currencies, exports were
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
66
included in the VAR to help represent external views of and relations with
the United States economically.
M2 was included in the model to gauge the monetary policy
activity by the Federal Reserve. This is especially interesting in the
conditional forecasting model, examining how M2 reacts based on the
estimated VAR when the S&P 500 is manipulated. If the stock market had
not crashed in 2007, the money supply would have been much different
because there would most likely have been no stimulus package or
quantitative easing (buying financial assets from banks to increase the
money supply – M2) necessary. Finally, the S&P 500 index is used in this
econometric analysis to represent the broad United States stock market.
This index and its representation of the US market are explained in further
detail in Appendix A.
All of these variables, except for one, are measured in USD. The
unemployment rate (lr) is already a percentage. For ease of interpretation
and comparison, each variable except lr has been logged. For a small
change (d) in a variable (x), ln(x+d) – ln(x) = d/x (approximately). By
logging the variables, VAR and conditional forecasting are actually
estimated using (approximate) % change in each of the data sets described.
Variable y (used in VAR estimation) is actually log(gdp), variable m is
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
67
actually log(m2) and so forth. Variables in the model are defined as
follows:
y = log(GDP)
p = log(CPI)
x = log(exports)
fx = log(trade-weighted value of USD)
m = log(m2)
sp = log(S&P 500)
lr = unemployment rate
Conditional forecasting with the estimated VAR provides an
interesting view of many of these variables and how their trajectory would
have been altered had the stock market crash not occurred beginning in
late ’07. The next section describes how this data was used in estimating a
VAR and performing conditional forecasting.
4. Econometric Methods
This section describes estimation and models used en route to
results and interpretation. First, a VAR system was estimated using all
seven variables described above. A VAR system contains a set of
variables, and expresses each of these variables as a linear function of
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
68
itself and all of the other variables in the set. Here is a simple example of a
VAR:14
So it’s similar to an OLS, except that Z, A, and epsilon are vectors of
endogenous variables, coefficients, and residuals, respectively. P is the lag
length. The lag length in the VAR for this paper was selected using the
rule of thumb to capture at least a full cycle of data, (# periods per year + 1
extra period) 4 quarters in a year + 1 = lag length 5. The system was
estimated over a span of 32 years (1975q1-2007q4). The results of the
Granger causality tests are included in Appendix B (all output discussed is
included here). From this, the causality ordering is as follows, starting
with the most exogenous variable and continuing through to the most
endogenous:
sp→x→y→lr→p→m→fx
SP X Y LR P M FX
SP --
X * -- *
Y ** -- **
LR ** * * --
P * --
M * ** * -- *
FX * *** *** --
14 Wooldridge (2012).
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
69
After the re-ordering, Lag Exclusion Tests were performed, and the 5th
lag
cannot be excluded, as the p-value on fx is 0.031516 and we reject the null
hypothesis that Lag 5 can be excluded. Also this VAR is stable, as the
greatest autoregressive root is 0.996732. As shown in the results, no
variable statistically significantly Granger causes changes in the S&P 500
index. SP is does not lag the US economy in this time period based on
these variables and this economic data. Because it is clearly exogenous,
another VAR is estimated with the 4 lags of the variable sp as exogenous
variables. The same ordering is used with the remaining exogenous
variables in this new VAR. This new VAR has a joint p-value for the 5th
lag exclusion test of .061419, so the lag length remains at 5. As far as
stability, the new VAR with sp as an exogenous variable also satisfied the
stability condition as its largest AR root is 0.982445.
Next, conditional forecasting was done with a simple model made
from the new VAR with lags of sp as exogenous variables. This model
uses all the coefficients from the new VAR to estimate a value of each
variable as the dependent variable. Forecasts are made from 2008q1 to
2011q2 in order to compare with the actual data. The first case is the
baseline test, where the model simply forecasts as it is. Then the
conditional forecasts are made. Scenario 1 is where .1 is added to the
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
70
variable sp in each quarter from 2008q4 to 2011q2. This means that the
stock market would still decrease as it did in early 2008, but the decrease
would be 10% less than what it actually was. Thus, in scenario 1, the stock
market crashed, but not as bad as what actually transpired between ’07 and
’09. Scenario 2, on the other hand, forecasts what would have happened if
the S&P 500 continued to grow at the average growth rate from 2002q1 to
2007q4. Here sp is regressed on a constant and a trend variable in the
given period to get the average growth rate. (This regression is shown in
Appendix 2) Then forecast sp using this simple OLS model from 2008q1
onwards. This way, the stock market never crashed, it continued on its 5-
year historic path of growth. Note the graph below of the baseline
scenario, scenario 1, and scenario 2 of sp for visual clarity.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
71
5. Results and Interpretation
From the original VAR, after performing the Granger causality
tests and discovering that not one of the logged macroeconomic variables
Granger causes changes in the S&P 500 index, it is rather safe to say that
the S&P 500 does not lag the US economy. Further, it can be said that the
S&P 500 index may be a leading indicator of the US economy, as it
adding in lags of the logged S&P 500 creates a statistically significant
6.6
6.8
7.0
7.2
7.4
7.6
7.8
2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
SP (Scenario 1)
SP (Scenario 2)
SP
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
72
effect on three fairly exogenous variables themselves (y – GDP, lr – the
unemployment rate, and x – exports,). From the original VAR it is
apparent that changes in the S&P 500 are completely exogenous from
changes in the six macroeconomic variables in that VAR system. Next, the
effects of changing the conditions of the S&P500 index on GDP, the
unemployment rate, and exports are discussed.
Effect of different S&P 500 conditions on y
Note that in scenario 1, where the stock market decreased but it was a 10%
less awful than the crash that actually occurred, y (the log of GDP) only
9.44
9.46
9.48
9.50
9.52
9.54
9.56
9.58
9.60
I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2007 2008 2009 2010 2011 2012
Y Y (Baseline)
Y (Scenario 1) Y (Scenario 2)
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
73
declines between 2008q4 and 2009q2, so there still may have been a
recession, but it would not have been as substantial or as lengthy as the
Great Recession. In scenario 2, if the stock market had continued growing
at the actual average growth of 2002-2007, GDP would have continued its
upward trajectory as well and there would have been no recession. This
provides more evidence for the stock market (S&P 500) as a leading
indicator for the economy.
Effects of different S&P 500 conditions on lr
Unemployment spiked during the recession, shown by the blue line which
represents the actual unemployment rate. In conditional scenario 1,
4
5
6
7
8
9
10
11
I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2007 2008 2009 2010 2011 2012
LR LR (Baseline)
LR (Scenario 1) LR (Scenario 2)
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
74
similar to Gross Domestic Product, unemployment would still spike, but
not as much as the recessionary unemployment spike. In scenario 2, the
unemployment rate stays fairly steady from its 2007 level, and actually
slowly decreases. This represents a healthy job market if the stock market
had not crashed in late 2007.
Effect of different S&P 500 conditions on x
By now the reader can probably anticipate how scenario 1 and 2 will
affect exports. In scenario 1 exports stay above recessionary levels, and in
scenario 2 exports continue on their long-term growth trajectory. All these
7.25
7.30
7.35
7.40
7.45
7.50
7.55
7.60
I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2007 2008 2009 2010 2011 2012
X X (Baseline)
X (Scenario 1) X (Scenario 2)
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
75
effects together form a fairly strong argument in favor of the US stock
market as a leading indicator of the US economy.
6. Conclusion
These interpretations are not to argue that the stock market alone
caused the recession, and that the stock crash had no other causes. Those
topics are a whole other subject of debate, and much has been written and
analyzed with respect to causes of the Great Recession. What this paper
argues is that the stock market is exogenous with respect to
macroeconomic variables, and that had the stock market not crashed in
late 2007, the economy would have followed the stock market’s lead and
continued on its long-term growth trajectory. According to this VAR
model and its conditional forecasting, the recession would have been
shorter and less severe had the stock market decreased 10% less than it
did. Additionally, and fairly intuitively, the recession would not have
happened had the stock market continued on its average growth rate
trajectory between 2002 and 2007. This study concludes that the stock
market (represented by the S&P 500 Index) is a leading indicator for US
economic activity.
This is but a simple analysis; there are many topics for further
investigation in this field. Some of these include matters such as:
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
76
Is the S&P 500 Index the best representative of the broader
US stock market, or should a different index be used?
How many times has the stock market given a false signal
about the direction of the economy?
What is the stock market’s success rate at predicting US
recessions?
How does the stock market compare to other measures that
are referred to as “leading indicators” of the economy?
And many more. The results in this paper may be helpful to economists
and policymakers as they attempt to build more sophisticated models. This
model is not meant to replace any forecasting models, because there are
other leading indicators as well which forecast as well or better than the
stock market. This simple model can be used within another model, or
serve as a simple check to the more complicated models.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
77
Appendix A: S&P 500 Index
The Standard & Poor’s 500 Index is a stock market index that
tracks the 500 most widely held stocks on the New York Stock Exchange
(NYSE). It is generally considered to be representative of the total United
States stock market, as it reflects the risk-return characteristics of the
largest companies in America. Standard & Poor’s indicates that the
committee responsible for maintaining (addition/removal decisions, etc.)
the S&P 500 index attempts to ensure that the index remains a leading
indicator of US stocks.15
Informally, the S&P 500 is a good representative
of the health of corporate America. Although it does not directly reflect
the performance of small- and medium-sized corporations or any private
businesses, a vast majority of these entities sell goods and services through
larger companies and thus are indirectly reflected in the S&P 500 which
tracks the performance of shares of the largest US companies. The largest
holdings in the index include Apple Inc., Exxon Mobil Corp., General
Electric Co., Google Inc., Pfizer Inc., etc.
15 S&P 500 Factsheet (http://us.spindices.com/documents/factsheets/fs-sp-500-
ltr.pdf?force...true%E2%80%8E)
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
78
Appendix B: Model Output and Charts
1. Granger Causality for original VAR with all variables as
endogenous
VAR Granger Causality/Block Exogeneity Wald
Tests
Date: 12/18/13 Time: 14:00
Sample: 1975Q1 2007Q4
Included observations: 127
Dependent variable: SP
Excluded Chi-sq df Prob.
P 9.195607 5 0.1015
M 1.627866 5 0.8979
LR 2.862956 5 0.7211
Y 0.740531 5 0.9807
X 7.882102 5 0.1629
FX 4.250607 5 0.5139
All 38.35387 30 0.1409
Dependent variable: P
Excluded Chi-sq df Prob.
SP 5.235294 5 0.3878
M 3.564758 5 0.6136
LR 10.44379 5 0.0636
Y 2.930824 5 0.7107
X 3.000720 5 0.6999
FX 4.942534 5 0.4229
All 48.66480 30 0.0170
Dependent variable: M
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
79
Excluded Chi-sq df Prob.
SP 4.578384 5 0. 4695
P 9.796535 5 0.0812
LR 13.40563 5 0.0199
Y 10.90745 5 0.0532
X 2.191492 5 0.8221
FX 10.57508 5 0.0605
All 76.62607 30 0.0000
Dependent variable: LR
Excluded Chi-sq df Prob.
SP 12.01836 5 0.0345
P 5.178249 5 0.3945
M 5.524131 5 0.3553
Y 9.679729 5 0.0848
X 10.03173 5 0.0743
FX 8.316988 5 0.1396
All 56.93654 30 0.0021
Dependent variable: Y
Excluded Chi-sq df Prob.
SP 11.08655 5 0.0497
P 4.669482 5 0.4575
M 1.021198 5 0.9608
LR 14.86652 5 0.0109
X 6.397807 5 0.2694
FX 1.966302 5 0.8538
All 53.99197 30 0.0046
Dependent variable: X
Excluded Chi-sq df Prob.
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
80
SP 9.877321 5 0.0788
P 1.510621 5 0.9118
M 6.643644 5 0.2485
LR 4.349350 5 0.5003
Y 3.551835 5 0.6156
FX 12.34010 5 0.0304
All 68.64448 30 0.0001
Dependent variable: FX
Excluded Chi-sq df Prob.
SP 9.099344 5 0.1052
P 21.26463 5 0.0007
M 18.20451 5 0.0027
LR 9.727446 5 0.0833
Y 8.663596 5 0.1233
X 4.451813 5 0.4864
All 44.17647 30 0.0460
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
81
2. Lag Exclusion Tests for original VAR
VVAR Lag Exclusion Wald Tests
DDate: 12/18/13 Time: 14:32
SSample: 1975Q1 2007Q4
IIIncluded observations: 127
CChi-squared test statistics for lag exclusion:
Nnumbers in [ ] are p-values
SP X Y LR P M FX Joint
Lag 1 137.2269 117.6186 125.4653 162.8454 218.9442 228.3195 139.5034 1027.956
[ 0.000000] [ 0.000000] [ 0.000000] [ 0.000000] [ 0.000000] [ 0.000000] [ 0.000000] [ 0.000000]
Lag 2 8.436880 9.534107 11.87405 7.399297 26.98292 35.99694 17.73660 111.8699
[ 0.295650] [ 0.216548] [ 0.104778] [ 0.388521] [ 0.000336] [ 7.26e-06] [ 0.013217] [ 8.05e-07]
Lag 3 12.02972 8.569125 5.054153 2.808760 23.48394 20.55103 8.986765 79.48111
[ 0.099586] [ 0.285088] [ 0.653355] [ 0.902112] [ 0.001403] [ 0.004495] [ 0.253608] [ 0.003812]
Lag 4 6.605649 9.195626 2.885306 2.289260 12.99945 11.15129 11.63935 56.33660
[ 0.471060] [ 0.238914] [ 0.895406] [ 0.942113] [ 0.072122] [ 0.132154] [ 0.113060] [ 0.219567]
Lag 5 6.001909 10.69869 4.389959 4.308822 1.923263 8.289551 15.37199 60.25364
[ 0.539526] [ 0.152313] [ 0.733924] [ 0.743599] [ 0.963965] [ 0.307755] [ 0.031516] [ 0.130074]
df 7 7 7 7 7 7 7 49
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
82
3. AR Roots for original VAR
Roots of Characteristic Polynomial
Endogenous variables: SP X Y LR P M FX
Exogenous variables: C
Lag specification: 1 5
Date: 12/18/13 Time: 14:31
Root Modulus
0.996732 0.996732
0.968921 - 0.096761i 0.973740
0.968921 + 0.096761i 0.973740
0.949862 - 0.040872i 0.950741
0.949862 + 0.040872i 0.950741
0.930411 - 0.186064i 0.948833
0.930411 + 0.186064i 0.948833
0.806539 - 0.354352i 0.880949
0.806539 + 0.354352i 0.880949
0.466003 + 0.618485i 0.774392
0.466003 - 0.618485i 0.774392
0.555095 - 0.508580i 0.752851
0.555095 + 0.508580i 0.752851
-0.077458 + 0.745182i 0.749197
-0.077458 - 0.745182i 0.749197
-0.184835 + 0.710848i 0.734485
-0.184835 - 0.710848i 0.734485
0.333427 - 0.634092i 0.716412
0.333427 + 0.634092i 0.716412
-0.699853 0.699853
-0.356626 + 0.591189i 0.690425
-0.356626 - 0.591189i 0.690425
0.012601 - 0.683634i 0.683750
0.012601 + 0.683634i 0.683750
-0.564365 - 0.379819i 0.680272
-0.564365 + 0.379819i 0.680272
-0.443059 - 0.494033i 0.663604
-0.443059 + 0.494033i 0.663604
0.640587 + 0.063885i 0.643765
0.640587 - 0.063885i 0.643765
-0.559658 - 0.312044i 0.640772
-0.559658 + 0.312044i 0.640772
0.286815 + 0.432876i 0.519273
0.286815 - 0.432876i 0.519273
0.141799 0.141799
No root lies outside the unit circle.
VAR satisfies the stability condition.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
83
4. sp forecast for scenario 2
Dependent Variable: SP
Method: Least Squares
Date: 12/17/13 Time: 20:11
Sample: 2002Q1 2007Q4
Included observations: 24
Variable Coefficient Std. Error t-Statistic Prob.
C 4.554757 0.254770 17.87790 0.0000
@TREND 0.020974 0.002128 9.854230 0.0000
R-squared 0.815290 Mean dependent var 7.061120
Adjusted R-squared 0.806894 S.D. dependent var 0.164250
S.E. of regression 0.072178 Akaike info criterion -2.339719
Sum squared resid 0.114611 Schwarz criterion -2.241548
Log likelihood 30.07663 Hannan-Quinn criter. -2.313674
F-statistic 97.10586 Durbin-Watson stat 0.605764
Prob(F-statistic) 0.000000
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
84
5. Remaining conditional scenario graphs
m
8.80
8.85
8.90
8.95
9.00
9.05
9.10
9.15
I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2007 2008 2009 2010 2011 2012
M M (Baseline)
M (Scenario 1) M (Scenario 2)
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
85
fx
4.32
4.36
4.40
4.44
4.48
4.52
I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2007 2008 2009 2010 2011 2012
FX (Scenario 2) FX (Scenario 1)
FX (Baseline) FX
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
86
p
5.78
5.80
5.82
5.84
5.86
5.88
5.90
5.92
5.94
5.96
I II III IV I II III IV I II III IV I II III IV I II III IV I II III IV
2007 2008 2009 2010 2011 2012
P P (Baseline)
P (Scenario 2) P (Scenario 1)
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
87
References
Benninga, Simon. Principles of Finance with Excel. New York: Oxford
University Press, 2006.
Bernstein, William J. The Investor's Manifesto: Preparing for Prosperity,
Armageddon, and Everything in Between. 2009.
Comincioli '95, Brad "The Stock Market as a Leading Indicator: An
Application of Granger Causality," The Park Place Economist: Vol. 4
Estrella, Arturo, and Frederic S. Mishkin. "Predicting US recessions:
Financial Variables as Leading Indicators." Review of Economics
and Statistics 80, no. 1 (1998): 45-61.
Gordon, Myron J. "Dividends, Earnings, and Stock Prices." The Review of
Economics and Statistics 41, no. 2 (1959): 99-105.
Pearce, Douglas K., "Stock Prices and the Economy," Federal Reserve
Bank of Kansas City Economic Review, November 1983, pp. 7 22.
Poterba, James M., Andrew A. Samwick, Andrei Shleifer, and Robert J.
Shiller. "Stock Ownership Patterns, Stock Market Fluctuations,
and Consumption." Brookings papers on economic activity 1995,
no. 2 (1995): 295-372.
Stock, James H., and Mark W. Watson. "New Indexes of Coincident and
Leading Economic Indicators." In NBER Macroeconomics Annual
1989, Volume 4, pp. 351-409. MIT Press, 1989.
Wooldridge, Jeffrey M. Introductory Econometrics: a Modern Approach.
Cengage Learning, 2012.
“US Stock Market as a Leading Indicator” by Ryan Johnsrud
88
Obamacare: Healthcare Reform and the Ailing Labor Market
Annabel Ansel
President Obama’s Affordable Care Act (ACA), has not only fallen
short of the lofty goals of affordable, universal health insurance coverage,
but has left ruinous impacts on the labor market. Hoping to revamp what
he and his supporters viewed as the inadequate and inefficient healthcare
system of the United States, President Obama signed the Patient
Protection and Affordable Care Act into law on March 23, 2010 (The
Kaiser Family Foundation). Contrary to supporters’ expectations, the
outcomes of this already controversial legislation, commonly known as
Obamacare, has caused, and will continue to cause, problems for
individuals and the economy as a whole. While the health of U.S. citizens
was supposed to be improved by the ACA, their economic wellbeing is
more likely at risk when the legislation’s implications for labor are
considered. Strikingly, initial reports show minimal progress in enhancing
healthcare accessibility and affordability, underscoring the validity of
predictions that “Obamacare’s distortions to the labor market will
outweigh any growth from lowering health costs (“Health Reform and
Employment”). Due to the negative impact on labor demand, the ensuing
reduction in employee wages, and ultimate decrease in the supply of labor,
Obamacare should be promptly defunded and rescinded. This conclusion
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
89
is supported not only by concrete economic theory and empirical findings,
but through ethical reasoning as well.
To fully understand its detrimental effects on the labor market, a
basic explanation of the policies embodied in the Affordable Care Act is
critical. Generally, the Act aims to expand healthcare coverage, minimize
healthcare costs for individuals, and ease the process for purchasing
insurance plans. The individual mandate requires all citizens who can
afford coverage, with few exceptions, to purchase it or face a fine.
Similarly, the employer mandate forces all businesses with fifty or more
full-time employees (those working thirty or more hours a week) to offer
health insurance to these employees, or face a large penalty. Other
significant points include forbidding companies from denying coverage or
charging higher premiums based on preexisting conditions, and the
allowance of children up to age twenty-six to be covered by their parents’
plan. The ACA outlines various minimum benefit requirements on
healthcare coverage, and implements a number of spending cuts and tax
increases to fund its programs (The Kaiser Family Foundation). All of
these programs and mandates have already unfavorably impacted the labor
market, leading to the conclusion that although Obamacare was intended
to help people in many ways it has and in the future will, hurt them.
“Obamacare” by Annabel Ansel
90
The Affordable Care Act’s various policies and taxes have
increased immediate and expected future costs for businesses, producing a
diminishing demand for full-time employees, and labor in general. New
costs confronting businesses first come in the form of taxes. In order to
fund the multitude of programs, subsidies, and benefits provided by the
ACA, businesses and individuals will pay an additional 0.9% tax on
taxable income and an additional 3.8% on larger capital gains (Obamacare
Facts). As with any operation, future expenses are a vital factor in firm
production decisions. Thus, the large expense of the employer mandate,
requiring companies to offer preapproved health insurance plans to their
full-time employees or pay 60% of worker premiums, will inevitably
influence business strategy. If companies fail to comply, they will face a
fine of $2,000-$3,000 (Blase). The higher anticipated cost of production
will discourage expansion and reduce full-time labor demand (The
Manhattan Institute). Further, according to the National Health
Expenditures Survey, insurance premiums will rise by 7.9%, which is 4.1
% higher than without Obamacare in place (Tanner). Therefore, even
those employers already providing healthcare to employees will face
added, burdensome costs due to the reform. New costs of production
cannot easily be absorbed by businesses, especially those that are small or
struggling to stay afloat with current costs in the unstable economy. Even
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
91
large corporations, such as Delta Air Lines, expect an upsurge in expenses
that will affect hiring and compensation decisions (Thomas). It is
predicted by the Congressional Budget office (CBO) that the ACA will
cost businesses $52 billion in tax penalties from 2014-1019 (Blase). The
long list of new expenses imposed by Obamacare will ultimately alter
output and revenue because labor is a variable cost. While some firms
will respond by cutting benefits, many are cutting employment. With the
employer mandate, and general taxes, businesses are discouraged to hire
due to the new costs that each new worker represents, regardless of labor
productivity.
Interestingly, some economists argue that although wage cuts will
occur, unemployment will be a more common result as businesses search
for possible ways to offset costs. Because of substantial worker resistance
to wage reductions, as well as prohibitory minimum wage laws on firms,
the decline in wages that is possible, will not be enough to recoup the
significant amount of lost revenue due to the ACA (Tanner). Notably, a
recent National Association of Business Economics (NABE) survey
showed that in the Service Sector, 1 in 5 businesses believe health care
reform cost pressures have hurt employment in their firms in the last 3
months (Lange). There is clear disincentive for growth and thus loss of
future employment opportunities due to Obamacare and its policies.
“Obamacare” by Annabel Ansel
92
This negative impact of the ACA on labor demand is illustrated
through a basic supply and demand graph (Figure 1). When the employer
mandate and other ACA policies are implemented, the additional costs
imposed on employers is viewed as part of the price in employing an
additional unit of labor. As labor costs rise, labor demand will fall, shifting
the curve from DL0 to DL
1, and causing the amount of labor hired at
equilibrium to be reduced from L0 to L1. This undesired decline in labor
demand, generated by the ACA is also observed on an isoquant and
isocost diagram (Figure 2). The slope of the isocost is the ratio of the price
of labor (PL), to the price of capital (PK). The price of labor not only
includes the monetary wage (W) for the unit of labor employed, but
implicitly, any benefits, such as healthcare, as well. With an increase in
the relative cost of labor, demonstrated by the steeper sloped isocost (I2),
the amount of labor employed by the firm, at the least-cost combination of
inputs, decreases from L1 to L2. Obamacare raises the unit price of labor
by way of taxes, penalties, and mandates on the firm. The adjustments
within Figures 1 and 2 depict the unfortunate effects this healthcare
legislation has on employment. Finally, consistent with the theory, the
Congressional Budget Office predicted that the amount of labor being
used by the economy will decrease by about half a percent, which is
approximately 7,000 additional Americans unemployed (Blase).
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
93
As discussed, the healthcare system reform inevitably leads to a
fall in the total demand for labor. Additionally though, because the
employer mandate only requires coverage for full-time employees, part-
time employment is not only falling to a lesser extent, but perhaps even
increasing, relative to full-time labor. Because shifting to a primarily part-
time workforce allows the similar, if not the same level of production for
many firms, and, “Companies don’t want to pay for health care
unnecessarily if they can avoid it”, they will do just that (Jargon, Cronin,
and Needleman). For the entire U.S. workforce, in 2013, the addition of
part-time employees seasonally adjusted averaged 93,000 a month, while
the addition of full-time employees averaged 22,000. The reverse was true
in 2012, as more full-time than part-time jobs were added. (Jargon,
Cronin, and Needleman). Also, 15% of service sector businesses plan to
shift to more part-time workers as result of healthcare reform costs,
according to an NABE survey (Lange). This shift in businesses hiring
predominantly part-time workers to avoid paying for insurance plans is
harmful to both the labor market and the individual employee. The
individual is an ‘involuntary part-timer’, underemployed at a suboptimal
indifference curve (I2), with more leisure (L2) than preferred (Figure 3).
Countless workers struggling to make ends meet will be forced to
moonlight, taking another part-time job. Also, these part-time workers are
“Obamacare” by Annabel Ansel
94
usually not covered by their employer for health insurance, and thus pay
out of pocket, reducing their income further. In general, even though part-
time labor demand has increased to some extent, the result is contrary to
the intent of the Act: fewer people are provided coverage through
employers because they are part timers, and the cost of health care
coverage for those who once were full timers goes up, not down, making
health care less affordable for those people under the Act. In addition, the
large accumulation of costs now facing businesses has caused the overall
demand for labor to fall. These are fundamental signs that the Affordable
Care Act is doing more harm than good.
The second key point necessitating the dissolution of the ACA is
its effective drop in the market equilibrium wage for labor. A surge of
non-wage compensation, like the mandated employee healthcare, will
necessarily call for a cut in wages in order for the firm to remain on the
same isoprofit line, and continue business as is. As seen in Figure 4, this
downward movement along the isoprofit line will leave the worker at a
suboptimal combination of wages (W1) and fringe benefits (F1), on a lower
indifference curve (I2). One study, by Jonathan Gruber, in which
companies were required to provide health insurance with specific
childbirth benefits to their staff, “found strong evidence that employers
reduced wages to pay for the benefits”(Tanner, 9). These empirical
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
95
findings heavily support the theory outlined above. Because employers
pay their workers through a variety of forms besides wages, when one
mode of compensation increases, the others must decrease.
Obamacare’s costly implications for employers, which cause the
contraction in labor demand, is ultimately the reason for the falling wage.
While employment will be reduced in avoidance of the employer mandate,
Mercer, a financial consulting agency says that many firms, for one reason
or another will end up covering their employees. The cost increases will
force businesses to pay their employees less though, and one study found
that “every extra dollar spent on insurance comes out of wages” (“Health
Reform and Employment”). As explained by the basic Law of Supply and
Demand, when labor costs rise, labor demand DL0 will shift left to DL
1.
Equilibrium market wage paid to workers will consequently decrease from
W0 to W1 (Figure 1). Costs will especially surge for employers not
currently offering healthcare, but all employers will face costs regardless,
whether it be through taxes, penalties, compliance with the mandates or
simply higher premiums. Overall, due to the increase in the marginal tax
rate from Obamacare and parallel programs, Casey B. Mulligan estimates
a 17% fall in “the reward for working”, and a dampening of consumer
spending (Mulligan, p. 22). This fall in labor income is not only bad for
individual employees, but also for the consumer-driven U.S. economy in
“Obamacare” by Annabel Ansel
96
general. Consumer spending is critically important to growth of the
economy and derived labor demand. Other compensation, besides pure
monetary wages alone, are being cut as well. For example, UPS is a
company for which Obamacare will increase costs by 4% for 2014, in
addition to health care inflation, resulting in another 7.25% increase in
costs. Along with similarly large companies, UPS, in response to the cost
increase, has been forced to reduce employee benefits such as coverage for
spouses (Thomas). The ACA will lead to lower wages and reduced
benefits across industries, as employers pass the reform’s costs onto their
employees.
As a third and final argument, multiple factors resulting from
Obamacare have caused a contraction in the aggregate supply of labor, as
seen through the actions of various groups of workers. Because of the
Affordable Care Act’s expansion of mandatory benefits included in the
new health insurance plans, premiums have increased for a majority of
individuals, even when the extra benefits included, such as maternity care
and the provision for dependents, are not needed. Due to the individual
mandate, those who do not qualify for Medicaid or government-subsidized
plans must pay out of pocket for their healthcare, thus reducing their
income significantly. This has led many to intentionally work less in order
to decrease their annual earnings, and qualify for government assistance
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
97
(de Rugy). This substantial subsidy results in an income effect for the
individual, as their earnings are essentially larger, and thus they are
incentivized to work less. A typical income/leisure diagram shows a labor
force participant’s choice to purchase more leisure (L2), shifting from I1 to
I2, because of the government transfer’s disincentive to work (Figure 5).
Many people are expected to leave the labor force, and older people will
be more likely to retire earlier (“Health Reform and Employment”). The
Congressional Budget Office agrees, stating, “The expansion of Medicaid
and the availability of subsidies through the exchanges will effective
increases beneficiaries’ financial resources. Those additional resources
will encourage some people to work fewer hours or to withdraw from the
labor market.” Moreover, the CBO found that the legislation would reduce
the total amount of labor by half a percent (Blase).
In addition to those choosing to work less, many individuals who
are already in the low-income position to qualify for government
subsidized healthcare coverage will automatically leave the workforce.
Craig Garthwaite, Tal Gross, and Matthew J. Notowidigdo calculated, “a
decline in employment of between 530,000 and 940,000”, due to their new
subsidy eligibility (Garthwaite, Gross, and Notowidigdo, p. 33). A factor
in this exit from the market can be explained, at least partially, by a
phenomenon known as ‘employment lock’. This is when individuals stay
“Obamacare” by Annabel Ansel
98
in their jobs mainly because of the healthcare provided to them by their
employer. With the heavily subsidized public health insurance newly
available to many, workers feel less inclined to remain working
(Garthwaite, Gross, and Notowidigdo). The income/leisure graph in
Figure 6 shows this as well.
Although Yaa Akosa Antwi, Asako S. Moriya, and Kosali Simon’s
analysis of the Survey of Income and Program Participation (SIPP) reveals
the encouraging fact that more adults ages 19 to 25 are insured as
compared to before the healthcare reform, the ACA’s policy for
dependents will not be completely advantageous (Antwi, Moriya and
Kosali). Obama’s healthcare legislation, by allowing children to stay
under their parents plan until the age of twenty-six, will ultimately
decrease the supply of labor. Before Obamacare, many young people were
incentivized to find work that offered health insurance as part of the fringe
benefits package, as it would be too expensive otherwise. With the new
legislation in place, there is less need and incentive to find a job offering
these benefits, or a job at all. Further, even once these recent graduates are
employed, the individuals will be at a suboptimal position on the Fringe
Benefits diagram. The smaller slope of the indifference curve relative to
the isoprofit curve (I1) reveals their relatively greater valuation of wages,
due to the need to pay off large education debts (Figure 7). The dependent
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
99
provision in the ACA simultaneously reduces the supply of and the utility
of the worker. Casey B. Mulligan even argues that, “it is unlikely that
labor market activity will return even near to its pre- recession levels as
long as the ACA’s work disincentives remain in place” (Mulligan, p. 25).
A reduction in Labor supply is unquestionably dangerous for the well-
being of the fragile U.S. economy.
While the facts and figures all advocate for the termination of
Obamacare, when discussing the well-being of a nation, the ethics of a
situation must also be examined. Through an understanding of
Libertarianism, Utilitarianism, and the teachings of Martin Luther, it is
even more apparent that Affordable Care Act is unacceptable.
Fundamentally, Libertarians emphasize limited government, voluntary
association and political freedom. Robert Nozick, an influential American
philosopher, stressed the importance of a minimal state, with entitlement
only owing to one’s own production, or voluntary transfers. The
Affordable Care Act enables the U.S. government to be a more extensive
state, as it redistributes income and services, hence violating principles of
justice in acquisition and transfer. Specifically, the employer mandate is
an “intrusion into voluntary arrangements made between employer and
employee”(Blase). Similarly, Nozick opposes any “time slice” or “end-
state” patterned distributions, as they focus solely on the final distribution
“Obamacare” by Annabel Ansel
100
of goods to an individual, rather than consider the justice in procedure of
distribution (Nozick, p. 228). Obamacare is unethical as it is patterned and
end-state in nature, guaranteeing universal healthcare, regardless of the
negative impacts on the labor economy. Also, from the perspective of
Nozick, the ACA’s tax increases are unjust, as they are equivalent to
forced labor.
Another prominent Libertarian, Ayn Rand, embraces similar ideas
as Nozick, believing fully in the superiority of Laissez-faire capitalism.
Rand argues against collectivism, altruism, and any form of redistribution,
urging that individuals act with “rational self-interest” for their own
happiness (Rand, p. 8). Obamacare should be defunded as it embodies a
collectivist mindset, giving the responsibility of providing health
insurance for dependent individuals to the government and businesses.
Also, like Nozick, Rand believes that a transfer concerning more than one
person must involve “voluntary consent of every participant,” which the
taxes, employer and individual mandates, and even additional benefits in
new healthcare plans, certainly do not entail (Rand, p. 93). Normative
understanding of Libertarian ideals undoubtedly shows that the labor costs
on business, and unemployment caused by, the Affordable Care Act is
unacceptable.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
101
The Utilitarian view on the Affordable Care Act is simplistic, yet
clear. Although Obamacare does make some individuals better off, the
reform does not result in the greatest good for the greatest number, or the
greatest good overall. There is disutility to employees, as they are not
allowed to work as much as they would like due to the firm’s profit
maximizing decisions. Even people who may be receiving enhanced
benefits with the new “minimum essential coverage” requirements, may
not be as satisfied with their coverage, because these new benefits could
“involve tradeoffs in terms of consumer preferences, moral choices, and
cost” (Tanner, p. 7). Finally, firms face increasing costs, hurting expansion
and hiring, ultimately reducing profits of the firm.
A final aspect in the normative analysis of Obamacare’s impact on
labor is the ethics of Martin Luther. Although a strong proponent of
salvation by faith alone, Luther insisted on the importance of using
authority for justice and the good of others. He advised against unjust
economic practices including cheating, immoral forms of pricing and
asymmetric information, all of which emphasize protection of the
vulnerable (Luther). Obamacare, although intended to help everyone,
leads employers to ignore certain duties of their secular or temporal
offices. They fail to live by Christian virtues, acting in a self-interested
manner due to cost hikes. Firms fail to follow the principle of loving one’s
“Obamacare” by Annabel Ansel
102
neighbor as they fire people, reduce hours, reduce wages, and drop
benefits. The Affordable Care Act also leads to the exploitation of the
vulnerable, which Luther explicitly forbade. Employees, exposed to and
dependent upon their employers, face higher taxes, lower wages, and
unemployment, while employers, vulnerable to the policies of the
government, are exploited by mandates, higher taxes, and increasing
healthcare premiums. Finally, those at most risk for losing their jobs in the
process of cost cutting, are minimum wage workers-the most vulnerable
people in the labor market (The Manhattan Institute). As demonstrated by
the philosophy of Libertarianism, Utilitarianism, and Martin Luther,
Obama’s lofty healthcare reform is not only unsuccessful, but unethical,
and should be brought to an end.
The Affordable Care Act is costly to businesses, individuals and
the government. It is inefficient in providing affordable and universal
healthcare coverage. The labor market, although only a single facet of a
national economy, inevitably dictates other market outcomes. It is
unmistakable however, that through the drop in labor demand and wages,
as well as the fall in labor supply, Obamacare has, and will continue to,
adversely impact labor. The various injuries to the labor market discussed
in this paper would be of concern in a prosperous, stable economy, but
with current economic conditions, and the slow, jobless recovery the U.S.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
103
faces, the nation’s labor market should be of high concern. The health of
the people and the health of the labor market should go hand-in-hand. It is
imperative that the Affordable Care Act and its provisions be ended
promptly.
“Obamacare” by Annabel Ansel
104
Bibliography
Antwi, Yaa Akosa, Asako S. Moriya, and Kosali Simon. "Effects of
Federal Policy to Insure Young Adults: Evidence from the 2010
Affordable Care Act Dependent Coverage Mandate." The
National Bureau of Economic Research (2012). JSTOR. Web.
18 Nov. 2013.
Blase, Brian. "Obamacare and the Employer Mandate: Cutting Jobs and
Wages." The Heritage Foundation: Leadership for America. The
Heritage Foundation, 19 Jan. 2011. Web. 17 Nov. 2013.
de Rugy, Veronique. "How Obamacare Will Shrink the Labor Supply."
National Review 14 Oct. 2013. Web. 17 Nov. 2013.
Garthwaite, Craig, Tal Gross, and Matthew J. Notowidigdo. "Public
Health Insurance, Labor Supply, and Employment Lock." The
National Bureau of Economic Research (2013). JSTOR. Web. 17
Nov. 2013.
"Health Reform and Employment: Will Obamacare Destroy Jobs?" The
Economist 21 Aug. 2013. LexisNexis Academic. Web. 20 Nov.
2013.
Jargon, Julie, Brenda Cronin, and Sarah E. Needleman. "Restaurant Shift:
Sorry, Just Part-Time." The Wall Street Journal 14 July 2013.
Web. 28 Oct. 2013.
Klein, Ezra. "11 Facts About the Affordable Care Act." The Washington
Post 24 June2012. Web. 28 Oct. 2013.
Lange, Jason. "Analysis: Little evidence yet that Obamacare costing full-
time jobs." Reuters 22 Oct. 2013. Proquest Newsstand. Web. 18
Nov. 2013.
Mulligan, Casey B. "Average Marginal Labor Income Tax Rates under the
Affordable Care Act." The National Bureau of Economic Research
(2013). JSTOR. Web. 19 Nov. 2013.
St. Olaf College’s Omicron Delta Epsilon Journal of Economic Research
105
Nozick, Robert. "Anarchy, State, and Utopia." Economic Justice in
Perspective: A Book of Readings. Ed. Jerry Combee and Edgar
Norton. Englewood Cliffs: Prentice Hall Inc., 1991. 222-47.
Print.
"ObamaCare Watch: The Employer Mandate." Economic Policies for the
21st Century. The Manhattan Institute, 2013. Web. 18 Nov. 2013.
Rand, Ayn. The Virtue of Selfishness. New York City: The Penguin
Group, 1961. Web. 21 Nov. 2013.
<http://marsexxx.com/ycnex>.
"Summary of the Affordable Care Act." The Kaiser Family Foundation.
The Henry J. Kaiser Family Foundation, 2013. Web. 27 Oct.
2013.
Tanner, Michael. "The Patient Protection and Affordable Care Act: A
Dissenting Opinion." Journal of Family and Economic Issues
34.1 (2013): 3-15. EBSCO . Web. 20 Nov. 2013.
Thomas, Alexandra. "Truths about Obamacare: Will firms cut benefits?"
Headline News 4 Oct. 2013. Web. 18 Nov. 2013.
<http://www.hlntv.com >.
"What is the Cost of Obamacare?." Obamacare Facts: Dispelling the
Myths. Obamacare Facts, Web. 27 Oct. 2013.
“Obamacare” by Annabel Ansel
106
Omicron Delta Epsilon
St. Olaf College: Beta Chapter
Class of 2014 Annabel Ansel
Shannon Cordes
Nick Evens
Rebecca Gobel
Duy Ha
Ryan Johnsrud
Mark Lee
Jane Meyer
Apoorva Pasricha
Michael Tillman
Kelly Tomera
Gina Tonn
Gabriel Trejos Durán
Rachel Turbeville
Class of 2015 Sara Anderson
Alex Everhart
Erik Gartland
Audrey Kidwell
William Lutterman
Camille Morley
Bjorn Thompson
Erik Springer
Sarah Stevens
Leah Voigt
107
Omicron Delta Epsilon Journal of Economic Research
St. Olaf College: Beta Chapter
Executive Editor Rebecca Gobel
Associate Editor William Lutterman
Spring 2014 Papers Shannon Cordes
Kelly Tomera
Ryan Johnsrud
Annabel Ansel
108