Systemic Analysis in Legislating: Modeling the “Cash for Clunkers”
Stimulus
Tao Wang
Emst & Y oung
225 Asylum Street, 14th floor
Hartford, CT 06103
taow90@ gmail.com
860-712-5953
Chester S. Labedz, Jr., J.D., Ph.D.
Central Connecticut State University
Robert V ance Academic Center 458
1615 Stanley Street
New Britain, CT 06050
labedzchs@ ccsu.edu
clabedz@ gmail.com
401-524-7711
Abstract
Legislating often may lead to unintended consequences and fail to achieve intended
consequences due to the complexity of political and social environments. In this article, the
authors build a system dynamics model focused on the American 2009 “cash for clunkers”
legislation. The authors identified dynamic hypotheses of both intended and unintended
consequences in legislative history and political commentary. Unintended consequences were
suggested: distortions in new vehicle sales and production, used vehicle supply and consumer
driving behaviors. Causal loop and stocks and flows models were developed. Using a Vensim
simulation, the authors tested for significant statistical differences in automobile related variables
with and without the legislation’s eight-week sales subsidy. The study found only short-lived
effects on used car dealers, charitable donation programs, and sales of new cars. The reasoning
and technique presented in this case study suggests a systematic and learning-intended
alternative to the prevailing “art” of political decision making.
Key Words
Prospective Legislative Impact statements
Unintended statutory consequences
Legislative Learning organization
Cash for Clunkers legislation
1 of 24
Introduction
Few fields of human endeavor promote intended activities while triggering unintended
consequences more commonly than does political governance. To legislators, regulators and
executive officeholders, promotion of intended outcomes seems their raison d’étre and principal
focus. By comparison, their attention to discerning and avoiding prospective unintended (and
undesirable) consequences of political action appears generally unsystematic and incidental.
Using decision-making processes in the governmental policy-making arena that are sensitive
to systemic interconnectedness would seem vital, given the gravity of many of the issues faced.
Citizens rightly may expect policy-makers to shape it in a sphere of rational analysis, allegiance
to truth, and pursuit of the general welfare. Expectations are that policy makers will be wise,
acting upon experience gained in enacting or implementing prior laws or through knowledge
generated and accumulated, even in other jurisdictions. The overall quality of knowledge-in-use
in the decision-making process in the legislative system should presumably increase over time
and, with it, the quality of lawmaking.
Stone (2002) explores these expectations at length in the public policy domain and then
dismisses them. She identifies a profound policy paradox between expectations and practice and
then justifies it through an alternative logic of lawmaking, which seems to be the reality of
policy-making in many developed, democratic nation states. However, practice of the “political
art” offers rich opportunities for the development of unintended and undesirable civic
consequences. And thus does the practice of politics, “the art of the possible” (von Bismarck,
1867), become the enemy of systems thinking, the integrative fifth discipline of essential
organizational learning (Senge, 2006).
In previously published research (Labedz, Cavaleri and Berry, 2011), the second author and
colleagues contested the inevitability of such dysfunctional behavior. We argued that such policy
making ills were preventable and remediable through a mechanism called a prospective
legislative impact statement (“P.L.I.S.”), if developed with respect to lawmaking under
consideration. We suggested that the P.L.I.S. process would incorporate the disciplines of both
systems thinking and system dynamics, and offered a recent A merican statute (with intemational
forebears) as a test case of our proposal. We proposed that the enactment and implementation of
the 2009 U.S. Car Allowance Rebate System (““C.A.R.S.”), colloquially known as “cash for
clunkers”, serve as the case study. The 2012 research was space-constrained to fully employ
systems thinking and system dynamics techniques with respect to that law, and thus omitted
testing through a system dynamics simulation of certain dynamic hypotheses proposed there.
This paper addresses those gaps. In the interest of brevity, and to maximize focus here on the
simulation, we move ahead to the case study, the dynamic hypotheses, and model construction
and simulation. The reader is directed to the prior article for the development and justification of
the P.L.LS. response to the “political art” and the contribution of this continued research to
knowledge management theory.
Case Study and Dynamic Hypotheses: Cash for Clunkers
The C.A.R.S. policy was enacted and implemented by the United States government in mid-2009
(C.A.R.S., 2009) as an eleventh-hour addition to supplemental war appropriations legislation,
2 of 24
with initial Federal funding of $1 billion. As introduced in the Congress (H.R. 2751, 2009), it
formally stated a small number of intended “results,” to employ one of Stone’s (2002)
classifications, including the provision of incentives to replace high polluting automobiles with
new, fuel efficient, less polluting automobiles. The bill’s sponsor rhetorically wrapped it in pro-
consumer, family-friendly, support of public services, and buy-A merican trappings as well.
C.A.R.S. gained early support from industrial trade associations, automobile manufacturers, new
car dealers and recyclers, and auto-related labor unions (111 Cong. Rec. 6348, 2009).
Only certain enumerated models of vehicles in private hands and under 25 years old could be
traded in, and upon trade-in and purchase of a new qualifying vehicle the participant would
receive a $3000 or $4000 taxpayer-financed payment, the amount based on the fuel efficiency
improvement in estimated miles per gallon from traded vehicle to new purchase. The program
started officially on July 1°. Due to unexpectedly high participation in this program, an
additional $2 billion quickly was authorized under the program. The official ending date of this
program was August 25, 2009, still months earlier than expected, when the additional funding
was exhausted.
Because the proposed law would authorize Federal spending, its pre-enactment analysis by
the nonpartisan Congressional Budget Office (“CBO”) of its anticipated fiscal impact was legally
mandated. The CBO analyzes spending and revenue effects of such bills, but in accordance with
a mandate to provide objective and impartial analysis, its reports are devoid of policy
recommendations (USCBO, 2009a) and of any analysis of broader dynamic effects of such
proposals (McBride, 2013). The CBO developed specific estimates of the costs of the C.A.R.S.
program based on an assumed use of 625,000 vouchers: administrative costs of about $55
million, and overall program cost of about $2.6 billion per year during 2010 through 2014. It
provided no other analysis of dynamic consequences of such alaw. The CBO and Congress
acted urgently: the estimates and assumptions regarding C.A.R.S. were only nineteen days old
when Congress approved this legislation. Other stakeholders had just a few days in which to
examine and question the C.A.R.S. bill before it became law.
Arenas (2012) provides a useful definition of a dynamic hypothesis in the context of
proposed action: the dynamic outcomes of a given action, expected by those who have taken the
decision to initiate that action. The benefits predicted (above) by the legislation’s sponsors
provide the first dynamic hypotheses to be examined here through systems thinking model
development methods and system dynamics simulations — as we propose they would be in the
P.L.LS. process. By examining the Congressional Record relating to C.A.R.S. and other
materials (Blinder, 2008), we translate the advocates’ arguments into four hypotheses of intended
(“i”) consequences of the stimulus:
H1i. It would accelerate motor fuel savings nationwide and provide incentives to registered
owners of high polluting automobiles to replace such automobiles with new fuel efficient and
less polluting automobiles or public transportation.
H2i. It would support jobs in automotive and related industries, get customers back into the
automotive showrooms, help [American] dealers move cars, and improve the environment.
H3i. It would represent a direct income transfer to the owners of clunkers, who are mostly low-
income people, and who would almost certainly spend the cash they receive, thereby giving
the economy a much-needed boost.
3 of 24
H4i. It would stimulate the demand for new cars as people trade up from used vehicles and as
millions of old cars would be permanently removed from service.
We tum now to some of the stated concems about cash for clunkers, translating them into
hypotheses of unintended and undesired consequences. Some analysts soon identified
prospective unintended environmental consequences. Greater fuel efficiency might lead to car
owners’ greater willingness and financial ability to drive more miles, thereby eroding the
intended environmental gain (Glaeser, 2009; McCullagh, 2009).
H1. By decreasing the fuel cost expended per mile of driving, through the program’s
substitution of fuel-efficient new cars for clunkers, C.A.R.S. might lead to car owners’
greater willingness and financial ability to drive more miles.
Once C.A.R.S. was enacted and its implementation commenced, other stakeholders
(including economists, charitable organizations, and other elements of the automotive industry)
weighed in with objections. The earliest objections arose from stakeholders whose interests were
excluded entirely from the legislative process. Used-car dealerships were an ignored stakeholder
in the cash for clunkers deliberations. In the aftermath of the C.A.R.S. program, fewer trade-in
vehicles would augment used-car inventories, because all vehicles traded in under C.A.R.S. were
destroyed. Used-car dealers’ demand for used vehicles might not be met for several years due to
the C.A.R.S.-lessened net new car prices of 2009.
H2. By rendering a large number of used cars permanently inoperable, cash for clunkers would
reduce the remaining supply of used vehicles available to prospective buyers, and by
inducing such scarcity would decrease the availability of used vehicles and harm the interests
of used-car dealers whose business depended on them.
Westley (2009) soon claimed that C.A.R.S. would raise the prices of remaining vehicles in
the used car secondary market and increase price levels in general through monetary inflation.
H3. By rendering a large number of used cars permanently inoperable, cash for clunkers would
reduce the supply of used vehicles available to prospective buyers, and by inducing such
scarcity would increase prices demanded for the remaining used vehicles and thereby harm
the interests of low income individuals who most needed to obtain them.
Automakers, their employees and their union representatives raised their concern that
demand and consumption were merely accelerated (Pethokoukis, in Fu, 2009), so that levels of
production and labor hours following the stimulus period would be trimmed.
H4. While encouraging consumers’ substitution of fuel-efficient new cars for clunkers, C.A.R.S.
would create little or no sustainable appetite for new car purchases, but instead would
increase demand during the stimulus period but promote decreased buying thereafter.
H5. While bringing forward in time new car purchases by some individuals who instead would
have purchased at later dates, without restoring the stock of prospective car buyers at those
later dates, C.A.R.S. likely would diminish labor hours demanded at later times.
Charities lamented that C.A.R.S. would reduce the supply of used vehicles that taxpayers
otherwise might have donated for community charitable support (Shogren, 2009).
H6. By rendering a large number of used cars permanently inoperable, cash for clunkers would
reduce the supply of used vehicles that remained available for donation by their owners,
4 of 24
thereby reducing funding of non-profit organizations and the community services they
provide.
Prior experiences of several foreign governments with respect to similar stimulus programs
also were available for consideration by U.S. legislators, had they wanted realism instead of
thetoric as they advanced the C.A.R.S. bill. Auto repair shops, used-car dealers and retailers in
other industries already had experienced damaging effects when Germany subsidized new car
sales in its so-called A bwrackpramie program (Ewing, 2009). These unintended consequences
and others already were visible in at least five European nations that had implemented such
programs by the time C.A.R.S. was adopted in the U.S. Miravete and Moral (2009) discuss
European precedents. Lessons that could have been learned from international experience with
similar programs were ignored, due in large measure to the polis approach that Stone (2002) has
described and the absence of P.L.L.S. rigor.
Depiction and analysis of system structures that lead to such intended and unintended
consequences are primary contributions made by systems thinking and its calculus-based
analytic engine, system dynamics. Note that correspondingly-numbered hypotheses with and
without the “i” designation essentially negative one another. In describing and analyzing claims,
we focus below on the warnings made testable by the hypotheses of unintended consequences.
We note two points about the modeling that follows. First, while it tests specific hypotheses
about specific legislation, the viability and possible contribution of the P.L.1.S. proposal itself is
more broadly under scrutiny. Second, the modeling here represents the efforts of the authors
relying on limited public records, rather than such modeling by a government office which
would have access to considerably greater input from interested parties. If P.L.I.S. analysis were
mandated by law and a federal legislative learning organization with greater resources
administered it, the modeling and its inputs undoubtedly would be more robust than those
presented here.
The Models
Modeling is a dynamic process, and causal loop drawings are an important tool for representing
the feedback structure of systems (Sterman, 2000). The prior research used such a drawing
(reproduced below as figure 1) to display the mostly-common structure that lays behind the
contrasting predictions of hypotheses 1 and 1i. In it, the solid loops at center depict the long-
standing twin policy aims of auto regulators and environmentalists. The miles-per-gallon
(“mpg”) policy has aimed since 1978 to increase the fuel efficiency of the installed automobile
base through gradually-more-stringent government mandates that affect each automaker’s new
vehicle production (NHTSA 2011). The Drive Less policy encourages reduced use of private
vehicles, and thereby reduced fuel consumption and consequent environmental pollution,
through ride-sharing, high-occupancy vehicle, mass public transit, bicycling, pedestrian transit
and other initiatives.
5 of 24
+ Efficiency i.
~~
a Reduced Cost
Promote Greater per Mile
Fuel Efficiency \
ry MPG. \
policy
Fundamental Goak oS :
ay rw Environmental Total Automotive
Reduced Automotive :
Drissions Gap _+ Emissions Greater Mileage at
i Same Expense
*\\ Consumers’ policy
Tesistance
‘ ie Drive Less
‘Stimulate Consumer Policies to Reduce policy
Demand Driving Greater free*
x ii
Sal
Facilitate - ii
Consumer Usage Automobile
+ Usage
Fig. 1. CLD presenting policy aims of, and policy resistance, to C.A.R.S.
C.A.R.S. aimed clearly to support the mpg policy, because the installed base of automobiles
would achieve a higher average fuel efficiency level simply as new, higher-mpg vehicles
permanently replaced old and less efficient “clunkers” on a one-for-one basis. C.A.R.S aimed
too to stimulate consumer purchases of new automobiles which would provide transport that,
mile-for-mile, was less expensive to drivers in their private vehicles. The two intended
influences of C.A.R.S. are depicted through the dashed (lighter) arrows at the left of figure 1. Its
right-hand arrows however describe dynamic hypothesis 1. Here, greater fuel efficiency permits
drivers to drive additional miles for no additional fuel cost, miles they otherwise would have
avoided if still driving their more-expensive-to-operate clunkers.
The “textbook” process of testing these dynamic hypotheses next calls for data collection
from various sources to permit the creation of stocks and flows (“S&F”) structures that
incorporate and extend figure 1 and mathematical testing of the S&F model for the predicted
behaviors. If the U.S. Congress already had imposed on itself the system thinking approach and
discipline called for in the P.L.I.S. proposal, a governmental legislative learning organization
would have been positioned in 2009 to accept such inputs of data and modeling suggestions as
any and all stakeholders would submit. The authors’ sources and resources are fewer than the
government’s, however. Therefore, we do not create the S&F model relating to the
environmental effects stated in hypotheses 1 and 1i. Instead, we developed a causal loop
drawing (figure 2) and resulting stocks and flows model (figures 3 through 5) that permits our
testing of some others of the dynamic hypotheses: H2i and H2 (used car dealers), H4i and H4
(overall demand), H5 (auto workers’ labor hours), and H6 (charitable donations).
6 of 24
Tax incentives in Chast incentives — CARS. incentive to eee o
donate ‘ a and sho} i: forlabor
hous
Charities’ donation. (a oni bet rd Late se vehicles ne z* ‘ te
Auto aa Sy Fix +
Ha be
t OMe Ne Ua on lat che! \ Ge ) OY 5 5 ‘ Shicle
‘Siting inventive to ew vehi recast deman¢{for
‘accep donations hae? financial _w ies on vehicles
“p= Public demand nal US.
+ used vehicles Durable aie supply (M3)
demand P
Eavorabilly of Durable personal
economiccontiions_— Sf, Cumentdomaned ct] consumption U.S. population
Fig. 2. CLD presenting non-policy-intended consequences of C.A.R.S.
In figure 2, the fears of used cars dealers are presented centrally and those of charities at left.
For each of those stakeholders, a new, large government-sponsored incentive to scrap used
vehicles augured a reduction in the inventories upon which each depends for resales and the
resulting income. These concerns, and Westley’s fear of a resulting increase in used car prices
due to their newly-induced scarcity, have been discussed above. These are the sort of systemic
issues which stakeholders would bring to a government learning office that would be charged
with developing systems-sensitive P.L.I.S. for policymakers’ consideration. The right side of
figure 2 suggests the labor-perceived fear that P.L.I.S.-subsidized new car sales would merely
effect temporary reductions in dealer inventories, without triggering any need for their
replenishment or any resulting demand for increased labor hours. [For readability, we supply a
larger version of figure 2 at the end of this document.]
We tur now to the S&F model that will be used to test the identified hypotheses. Because
of its size, we discuss and display it incrementally in figures 3 and 4, and completely in figure 5.
The stocks and flows passage of automobiles within the United States is described in figure 3
by an aging chain structure (Sterman, 2000). It sets forth the sequencing of vehicle production,
private ownership, and vehicle retirement in a structure similar to that employed in another
recent impact assessment of the automotive industry (Walther et al., 2010). Automakers
periodically place production orders (“MGT’s Order”) based on such factors as current and
forecasted market (“MGT’s MKT expectation”) and economic conditions (“Economic Factor”),
new car demand and inventories, historical sales data, and other factors that affect their industry.
(Shahabuddin, 2009). As a part of the luxury and durable goods component of the economy,
automobile sales have been highly cyclical: sales typically are high for certain months of each
year and predictably lower in other months. “Inventory gap” is the difference between the
targeted inventory level and the current one (“new vehicles inventory”), and “MGT’s MKT
expectation” is a lookup table based on seasonal patterns of new buyer behavior. We assigned
weightings to the influences of the economic and cyclical factors, so that the production and
current inventory levels would better replicate historical behavior. We assumed that targeted
inventory level equals two months of current level of market demand (“prospective buyers”) and
have used “inventory gap” as the aggregate targeted production level.
These orders lead, after a production and shipment delay (“Targeted time”) to deliveries to
dealers’ “New Vehicles Inventory” stock. These new cars remain there until sold, at which point
7 of 24
they flow into “Late Model Year Vehicles on Road”. Autos remain in that stock on average for
just over five years. Automobiles older than that are traced in the stock of “Older Model Year
Vehicles, privately held”. We choose to distinguish vehicles younger and older than 66 months
because that was the average finance term of new car loans in 2009 (Bird, C., 2010), reasoning
that individuals were less likely to sell or replace such vehicles until those loans were repaid.
Autos leave the “on Road” stocks in three ways. Their chief (and permanent) outflow occurs
through vehicle scrapping, in which reusable parts are salvaged from them before their remains
are smelted, processed and recycled to some extent. This model assumes that late model vehicles
are scrapped in negligible numbers and that older vehicles are scrapped at a rate which maintains
the total number of vehicles on the road in equilibrium over the model period, absent the effect
of C.A.R.S. Because of H6, our model provides an outflow for the temporary removal of older
cars from the road when their owners donate them to tax-exempt charitable organizations. These
organizations often refurbish the donations and then resell them to drivers, usually retuming
through the auctioning inflow (General Accounting Office, 2003). Finally, the model recognizes
(for the sake of completeness, through an instantaneous return flow) the private resale of used
vehicles, in which vehicles remain as Older Model Y ear Vehicles on Road but their legal
ownership has changed. The buy-sell trading of used cars among owners and dealers does not
change the total quantity of cars in the market or on the road.
Used car
ake Fractional rate for
older vehicles trade in
Usual faction to fac
eae Usual fraction
; Tale Model [Older Model’ sual fraction for
New Vehicles YearVehicles donating
‘Autos aging
Vehicles, privately
ne Sc
Production | laventories |" Driving fom | onRoad Tapping
Showroom 2
Auto aging rate a
vain 2 ee
me vA 0 car trade in mi
Donating
fractionto |
auctoring —\
Donated Vehicles at
tax exempt charities
Fig. 3. S&F structure of automobiles’ aging chain
Figure 4 models the corresponding demand side for new private vehicles. The simulation of
new car buying behavior in the system consists of an inflow of individuals beginning to look,
thereby becoming become part of the stock of “prospective buyers”, and then flowing out of that
stock chiefly when they buy new vehicles. (We supply a secondary, temporary outflow for
consumers who become discouraged or who briefly “walk away from deals”, only to return
weeks later.) The inflow of prospective buyers comes principally as consumers trade in their
older automobiles, and to a lesser extent from individuals who own newer “late model” cars. The
earlier-discussed flow, “driving from showroom” is set identical to the “buying” flow here,
because this model makes the simplifying assumption that each new car buyer acquires just one
vehicle at a time. To avoid double counting, we limit the number of individuals who begin
8 of 24
looking for new cars to the number of old vehicles scrapped, even though the individuals in
question may not be the same persons.
Fractional rate for
Tecovel
Fractional a
discouraging rate
Discouraged <Time>
buyers
i peal abe Usual fraction for looking
i by older vehicle owners
D ‘overing rate © OWNeIS
Dis ing rate Ownership Ratio
Usual fraction to new
vehicle purchases
Prospective
Vehicle |
buying Buyers looking:
Fig. 4. S&F structure of prospective new car buyers
Figure 5 combines the partial models of figures 3 and 4 and adds one other sector. At the far
right, we formulate the model element “Statutory behavioral incentives under C.A.R.S.” as a
PULSE function that takes the values of zero or one. When it takes the latter, it affects the six
newly shown “change in fraction” elements to which it is causally tied, as explained in the next
paragraph.
discouraging Fractional at for
due CFC
Frocioml recovering
decourgeg mie | a
Eonmni ior Dawe | ~
i if
insu hi ‘Usual fraction of loo}
wet / pe | Petes cenpeptenin
a, a ering rate | Owners: by older vehicle owners
\ ‘Statutory behavioral
( Ownersip Ratio ines under C.A.RS
4 itn factonel og >
king by elder model
—
Change in action of
a ae
|
Froctional rate for |
— stevia |)
| /|
a pts cartes Useless
\Ca ea
rs oe ail
a a a
‘ fort ey
cae ——
‘eewto | —
Production a domat
= {wtal vehicles on- ‘Domated Vehicles at
road tax exempt charities
9 of 24
Fig 5. The complete S&F model for simulation
The dynamic hypotheses will be tested through the stocks and flows model depicted in figure
5 across a 312-week simulation period, calendar years 2006 through 2011, using the Vensim
DSS (Ventana Systems, 2003) application. [For readability, we supply a larger version of the
complete S&F model at the end of this document.) Although most of the figure 5 model
elements depict the automotive sector generically, with or without the 2009 legislation, the six
underscored “change in fraction” factors trace directly from the “Statutory behavioral incentives
under C.A.R.S.” element in the upper right. Weeks 183 through 190 of the simulation period
correspond to the operational period of the stimulus, from July 1 through August 25, 2009.
Framed as an eight-week PULSE function in Vensim, that “switch” temporarily triggers those
six change fractions in simulating actual developments. If that pulse is zeroed out, however, the
model produces for comparison and hypothesis testing behavior as if C.A.R.S. had never been
enacted.
Government records (Council of Economic A dvisors, 2009) indicate that 677,081 matching
trade in and buy transactions occurred during the eight week stimulus period. Because of the
subsidy offer, interest in looking for new vehicles increased and sales of new cars increased
compared to the months before the subsidy. To promote its environmental aims, C.A.R.S.
required that older vehicles be rendered inoperable, so older car scrapping temporarily increased
to reflect each trade-in under the program. Within the used car market, because of the decrease
in “privately held older year vehicles” due to this scrapping, a subsequent temporary decline in
used automobile trading can be expected. These effects are captured among those six
underscored change fractions.
The Data
To develop credible simulation results, modelers should observe a number of validation and
model testing protocols, many of which Sterman’s chapter 21 (2000) summarizes. One requires
that the simulation reproduce actual behavior within the real-life system of interest. We focused
on three model elements for which reference mode values could be obtained or developed. The
reference elements are the stock of “New Vehicles Inventories” and the flows named
“production” and “buying” of new vehicles. They serve as the check points to verify the
correspondence of simulation results to historical data. It was difficult to find historical data to
compare at other points in the simulation, as the elements either are abstract, proprietary and
confidential, or are by nature estimated.
Several private firms offer various data sets presenting different slices of the American
automotive sector, at various price points to the researcher. Official govemment sources offer
other data. Within our research budget, we purchased access to U.S. new car sales and inventory
data by model line for 2006 through 2011 from Ward’s Automotive Group and Automotive
News (Crain Communications, Inc.). We used U.S. government-supplied Bureau of Economic
Analysis (“Bureau”) and General Accounting Office data as well.
Ward’s Automotive Group is a research organization that has covered the automotive
industry for over 85 years. The Bureau develops economic statistics, including monthly
automobile market data with related economic adjustment factors, within the U.S. Department of
Commerce. The Bureau’s cyclical automobile sales factors are based on Ward’s monthly sales
data, so Ward’s data are considered the most definitive in this model. The GAO data (2003)
10 of 24
reported to Congress the structure and scope of private automotive donations to charities, which
are relevant in testing H6.
Automotive News is a weekly automotive newspaper published for industry participants.
The data we purchased was sorted by manufacturer, model and make. Due to different collection
methods, the commercial sources’ data are not identical, but their overall trends are similar and
consistent. The data from Automotive News would be more important if the model were
designed to go to more detailed levels, such as to a make/model level of disaggregation.
For our purposes, the commercially-provided data contain a common defect. Due to the
global nature of the automotive industry, the same model and make of a vehicle may be
produced in factories the world over. Linking production and sales data becomes distorted by
transnational imports and exports; it is difficult to trace in research whether automobiles sold in
the U.S. are produced there or elsewhere. Thus, in establishing the reference mode for
production, we use the sales and inventory reference modes as inputs to simulate production
behavior. The model uses the following formula to back into the production reference mode: -
Production = Ending Inventory +Sales - Beginning Inventory.
Model Validation and Testing
In developing and examining the S&F model prior to hypothesis testing, we undertook a number
of commonly-stipulated tests and measures (Sterman, 2000; Rahmandad and Sterman, 2012).
We extended our textual research beyond the claims made by sponsors and proponents of the
legislation to other stakeholders whose perspectives extended both our CLDs and S&F drawings.
This broad casting of the model boundary net is consistent with the P.L.I.S. methodology that the
second author’s 2011 research suggests. We conclude that an addition of further elements will
not significantly change the behavior of focal model elements. We cite US population as an
example. Although we believe it influences sales of vehicles over the long run, we deem it
insignificant across our simulation period, as it was relatively stable from 2006 through 2010.
The model’s boundary might expand to include other factors like buying behavior, buyer
demographics, and manufacturers’ marketing efforts, but we do not believe that the model’s
overall conclusion and recommendation will change materially through expanding it.
The model’s structure is consistent with the general description of the U.S. automotive
industry. The CLD and S&F provide for key market elements, the C.A.R.S. program, and
stakeholders’ reaction to that government intervention. The S&F model consists chiefly of two
partial models: the aging chain of vehicles from production to scrapping, and the consumer’s
new car buying decisions from initial interest to eventual ownership. New car production
decisions are based on feedback from the most recent sales data and current levels of inventory.
The model then conserves each vehicle from manufacture through ownership changes until final
disposition through disassembly for scrap materials. Auto aging and scrapping rates are
constrained through first-order negative feedback loops so that their associated stocks cannot
take on negative values. New car buying interest is tied to the late model and older model year
owner cohorts, so that new car buying is similarly, realistically constrained. (There cannot be
more buyers than current owners, as we assume a simplifying one vehicle per owner policy, and
in fact will be many fewer buyers at any time.) Even as simplified, the partial models thus
logically replicate physical laws and conserve material (vehicles, drivers) as appropriate.
11 of 24
The S&F model passes Vensim’s “check model” and “units check” consistency tests without
use of arbitrary scaling factors. Parameter values are consistent with relevant descriptive and
numerical background information, and have real world meanings. Estimated parameters were
calculated to better replicate historical behavior prior to using the S&F model to test hypotheses.
We also conclude that different integration methods available through the V ensim software will
lead to very similar simulation results, and that integration error is not a risk.
Prior to hypothesis testing, we conducted extreme condition tests relating to scrapping rate
and new car purchasing. In one pair of tests, we set the usual fractional scrapping rate
successively to one and zero values. When set to one, all older year vehicles were scrapped. As
a result, only vehicles under age five would be on the road. Owners who held vehicles of ages 5
to 15 years scrapped their vehicles at the beginning of the modeling period, leading to peak
buying behavior at the beginning and a slowing down afterwards. When set to zero, no vehicles
would be scrapped during the studied period. The number of prospective buyers will keep rising,
as people owning older model year vehicles keep increasing. However, actual buying would not
increase dramatically because people are using older cars for a long time without scrapping. This
soon will lead to inventory overbuild, but due to the feedback loop from the sales level and the
current over-built inventory level, production will stabilize afterwards. In the second pair of tests,
we changed the usual fraction to new car purchases from its stipulated 80% value to 100% and
zero. The change to 100% would shock manufacturers, who planned on the basis of the 80%
rate, and would cause decrease in inventory. When set to zero, interested buyers would “kick the
tires” but not purchase. This should lead to overbuilding vehicles and maximizing production,
and the trial simulation bears it out.
The pending P.L.I.S. argument for formally applying systems thinking and system dynamics
methods prior to legislating relies on the S&F model to trace unintended consequences of the
government’s C.A.R.S. intervention. These include “what if” scenarios, examining alternate
results if the government had not implemented “cash for clunkers”, “what more” scenarios, in
which the intervention spawned unintended consequences including policy resistance, and “what
else” scenarios in which the alternative stimulus designs might produce more satisfactory
outcomes. Behavioral reproduction testing seems most important in assessing the utility of the
S&F model for these purposes. In assessing the simulation’s reproduction of real-world data, we
tested the three key variables for which we had government-supplied or purchased historical
data: automakers’ new car production, and dealers’ new car inventories and sales. These
reference mode data extend 182 weeks before the eight-week incidence of the C.A.R.S. stimulus
and continue for 122 weeks after it. Figure 6 presents the historical reference mode data for
these variables and their base case values produced through simulating the S&F model.
12 of 24
140,000
120,000
100,000
80,000
60,000
40,000
20,000
0
a
mHR aD
ann st
AMOR OAaANMHR AGM OR
ONnBDHDONMTHOEMDHDOH
aAagaeaeaea gaan
229
adaon
PuoOR
ANNAN
— — — Ref Mode Simulation
a
oO
nN
a
3
cy
Fig. 6a. US new car sales: reference mode and simulation results
1,400,000
New Inventory
1,200,000 *
= sae N
1,000,000 >
800,000
7
600,000
400,000
200,000
0
AMORDIAMHRAAMHRAGMHADIAMHRAG
SARFSRSBASAHATHSRZASGATHSRSS
SSSSSSSSRNAAARAAAM
— — — Ref Mode Simulation
Fig. 6b. US new car inventories: reference mode and simulation results
13 of 24
140000 .
| Production
120000
100000
80000
60000
40000
20000
0 ITT TTT TUTTI row rm ATTA
AMYNRAAMHARDAMHER ANAM HORAAMHROS
ANHMDTORADSAMTHOERACHNATHOERAS
SAHARA A ANNAN ANN
— — — Ref. Mode Simulation
Fig. 6c. US new car production: reference mode and simulation results
Graphically, the model tracks the historical time series data, reflecting the insertion of the
C.A.R.S. stimulus in weeks 183 through 190, reasonably well. It overestimates both sales and
production in the months immediately preceding that mid-2009 stimulus period but then traces
the actual results well during those eight weeks. Comparison of the new car inventory graphs
reveals one prominent underestimation (weeks 147 through 184), matched with two later periods
of overestimation of new car inventory.
We report in Table 1 summary statistics of the goodness of fit of the base model. Mean
Absolute Percentage Errors (MA PE) values, as calculated between the empirical data and the
base case simulation output, range from 7.13% to 13.55 %. These too suggest an adequate
tracking by the simulation of the reference mode values of the three focal variables. Theil (1996)
inequality statistics permit the decomposition of these MA PE values into three components:
model bias, unequal variation, and unequal covariance. The table presents those values too. Per
Sterman (2000), the Theil values for sales and production chiefly confirm phase shifts between
simulated and historical data, which fluctuate with similar means, amplitudes and frequencies.
The Theil values indicate unsystematic error in the case of inventory, as simulation output tracks
actual data except for an error term with a zero mean. As seen through the U™ value in Table 1,
however, the mean values of the reference mode and the simulation output for inventory are
quite close.
Table 1. Behavioral reproduction test statistics for base case of simulation
R2 Mean Root u" us us
Absolute Mean (model (unequal (unequal
Percent Square bias) variation) covariance)
Error Error
Sales 0.43631 10.72% 12,403 7.98% 0.52% 91.50%
Inventory 0.75826 7.13% 76,752 0.53% 34.16% 65.31%
Production 0.41655 13.55% 14,529 5.02% 0.84% 94.14%
14 of 24
Test Results for Dynamic Hypotheses
Completion of these preliminary model tests now permits the testing of the dynamic hypotheses
relating to unintended consequences of C.A.R.S. on used car dealers (H2), new car demand (H4),
auto workers’ labor hours (H5), and charitable donations (H6). The S&F model depicts the
behavioral incentives of C.A.R.S. in the upper right of figure 5. In the base nun, the model
element Statutory Behavioral Incentives under C.A.R.S. took on the value of one only in months
183 through 190, thereby directly activating the six “change in fraction” elements to which it ties
through causal arrows. To test the extent of these unintended consequences, we ran the
simulation again, now with the eight-week C.A.R.S. pulse zeroed out, too. In testing the
hypotheses, the two sets of simulation-produced values are compared for all weeks after the
week in which the subsidy came into direct effect. For most elements, the direct effect started
after week 182, but for production, due to the model’s built-in information delay, the direct effect
began at week 185. Specifically, we contrast the values of the model elements “used vehicle
resale”, “Prospective Vehicle Buyers”, “total production hours” and “donating” respectively in
testing the four hypotheses. We used paired-samples t tests (Norusis, 1997) to test short term
direct effects during and after the subsidy period, in which our null hypotheses predicted the
absence of differences between the mean values of each focal variable with and without the
introduction of the C.A.R.S. subsidy, during and after the subsidy period. We used az test to test
the 2-year long term effect from the implementation of C.A.R.S. Table 2 presents the results of
testing of the four dynamic hypotheses.
Table 2a. Test results for hypotheses relating to C.A.R.S. influence on used car held at dealers
and on donations
Direct effect from C.A.R.S. Effect after C.A.R.S.
(8 Weeks) (96 weeks) 2-year effect
Confidence Confidence Confidence
tscore level z score level z score level
Used cars
Trade-in 152.402474 99.95% 0.96309 83.15% 2.2168775 98.65%
Donation 318.916441 99.95% 0.9631 83.15% 2.634906 99.59%
Table 2b. Test results for hypotheses relating to C.A.R.S. influence on prospective new car
buyers and on new car production hours
Direct effect from C.A.R.S.
(8 Weeks) Direct effect after C.A.R.S. 2-year effect
lengths Confidence Confidence
tscore Confidence level (weeks) tscore level z score level
Prospective
buyers -10.8476 99.95% 5 4.2723 99% -2.635 99.59%
Production hours -13.4832 99.95% 28 2.3053 99.95% -1.055 85.31%
15 of 24
H2 examined the fears of used car dealers that C.A.R.S. would induce scarcity of used
vehicles, thereby harming their business interests. The simulation permitted comparison of the
“Used vehicle resales” flow with and without the trade-in-and-destroy stimulus during the
stimulus period and after the stimulus expired. Table 2a confirms that used car sales declined
markedly after introduction of the stimulus (z=2.2168, confidence level = 98.65%). On closer
inspection however, the statistically significant effect occurred during the eight weeks of the
subsidy (t=152, confidence level = 99.95%) and not thereafter. H2 is supported, but only as to
the eight weeks of the stimulus.
H4 considered the fears of workers and their union representatives that C.A.R.S. would
create little or no sustainable appetite for new car purchases, but instead would increase demand
during the stimulus period but then promote decreased buying thereafter. The model’s “buying”
flows, after the stimulus versus without it, are compared here to test this claim. The results
suggest that cessation of the stimulus likely created a surge in Discouraged buyers unable to
participate in the program, and that this surge only eroded exponentially over time. The t-scores
and high confidence levels for comparisons of Prospective buyers during the eight weeks of
C.A.R.S. stimulus and the five weeks immediately thereafter support that conclusion. Table 2b
confirms a significant difference in hours, and H4 is supported.
As aresult of the predicted change in “buying”, H5 proposed that reductions in labor hours
authorized by automakers would occur in months after the stimulus, as those workers feared.
Comparison of stimulus and no-stimulus values of “Total production hours” after week 185 test
this hypothesis. Table 2b indicates an absence of confidence (85.31%) in suggesting any long-
term effect of C.A.R.S. on production hours. H5 is rejected.
Finally, H6 tested the claims made by non-profit organizations that feared reductions in
charitable donations of used cars which needed to be destroyed in order for car owners to claim
their stimulus payouts. The focal model element here is the “Donating” flow. As with Used
vehicle resales, discussed above, Table 2a confirms that charitable donations declined markedly
after introduction of the stimulus (z=2.6349, confidence level = 99.59%). On closer inspection
however, the statistically significant effect occurred during the eight weeks of the subsidy
(t=318, confidence level = 99.95%) and not thereafter. H6 is supported, but only as to the eight
weeks of the stimulus.
Figure 7 graphically compares the with- and without-C.A.R.S. behavior of the focal variables
of table 2. In this figure appear the temporary scarcity of used cars and of donated vehicles, the
surge and decay in discouraged buyers, and the merely temporary effects on production hours, all
as tested for significance in table 2.
16 of 24
Figure 7a Comparison of used vehicle resales
during and after C.A.R.S.
Figure 7b Comparison of prospective buyers
during and after C.A.R.S.
500000 +
400000 ~
300000
200000
100000
0
Used vehicle resales
axgeseaeneeserg
BaSSaHamst OR w
AA NANNNNANNANAAN
- — — Without C.A.R.S. —— With C.A.R.S.
150000. Prospective buyers
100000 DN
50000
- — — Without C.A.R.S.
Figure 7c Comparison of production hours
during and after C.A.R.S.
Figure 7d Comparison of donating during and
after C.A.R.S.
5000000 5
4000000
3000000
2000000
1000000
0
Production hours
2PHartsaereyes
00 Sanam tHONn
AANNANNNANANAN
Donating
Without C.A.R.S. —— Wit!
Fig. 7 Comparison of effects on table 2 variables, with and without C.A.R.S. subsidy
Discussion, Limitations, and Future Research
Through the S&F model, we tested four claims (dynamic hypotheses) relating to unintended
consequences of C.A.R.S., as suggested by stakeholders in the U.S. automotive field while the
subsidy legislation was under lawmakers’ consideration or during the eight week period in which
its stimulus was available to consumers. We found that concerns regarding vehicle shortages
emerging due to C.A.R.S., raised by used car dealers (H2) and charitable organizations (H6) that
received car donations, likely were unfounded once the stimulus period expired. The simulation
lent its support to concerns (H4) that C.A.R.S. would not create any sustainable appetite for new
car purchases, but instead would merely increase demand during the stimulus period but depress
it thereafter. Finally, the modeling did not support concerns (H5) that reduced new car
production hours would result from the C.A.R.S. program.
The model’s parameters and structure likely can be improved, as discussed below.
Nonetheless, the approach, equations and reference modes developed and the results obtained to
date suggest greater benefits may yet be achieved through refining the model and asking further
17 of 24
questions of it. For example, as mentioned above, Kiley suggested alternatives in designing the
C.A.R.S. incentives that he claimed would better support environmental protection aims or
concerns. With additional inputs to model structure and data, such alternative “what more”
versions of proposed lawmaking likely could be examined in advance of legislative action.
A principal limitation of this research lay in our limited ability to retrieve empirical data that
likely would have refined the models and improved upon and increased the number of reference
modes with which the simulation was compared. While we are grateful to the university sources
with which we have been associated for their grants provided to purchase auto industry data,
others likely have access to more numerous and perhaps higher quality data. In a sense, this
helps to support the argument made by Labedz et al. (2011) that prospective legislative impact
statements be developed under government auspices. Undoubtedly, the greater resources of a
dedicated legislative learning office and of interested private stakeholders would support more
complete data retrieval, develop more comprehensive models, and make even greater
contribution to knowledge management among lawmakers.
Clearly, we chose not to test each hypothesis laid out in the Introduction. Our data access
constraints suggested that building and testing our limited model was a prudent program,
sufficient for a first test of the P.L.I.S. proposal. We would welcome the opportunity to build
more, and hope to test H1 in a next stage.
Our limited access to automakers’ decision processes about inventories and production plans,
let alone their then-current financial, operational and competitive considerations (as also
experienced by Shahabuddin, above), posed another challenge in developing model structure
relating to the left side of figure 5. Much of this information is confidential or proprietary, but
likely it is more knowable or better estimable by a government office acting under lawmakers’
authority. Much of it is variable across the set of automakers too, and access to additional
proprietary databases likely would lead to improvements upon our version.
Such limitations notwithstanding, we conclude that the P.L.I.S. approach to legislative
knowledge management, including its application of systems thinking and system dynamics
approaches, is viable and deserving of future research, support and funding. As the second
author and his co-authors observed, P.L.I.S. would be central in an iterative feedback process
that carefully assembles a dynamic systems model from the claims of interested parties as
legislation is proposed, then traces the operation of the system in order to validate (some of)
those claims in the years following its enactment. Lessons leamed through this recursive process
would be available to guide subsequent amendments of that law, but more importantly could
guide systemic discussion of future legislative proposals that bear similar designs, as for example
with other targeted economic stimulus measures. Legislative models may be developed
transnationally, so that a government may call upon other nations’ prior experience, systemic
learning, successes and shortcomings while crafting its own laws. In a specific case, the
American experience with C.A.R.S. might add to models of the A bwrackpramie and other
European auto substitution precedents mentioned in the Introduction, and might have learned
from their systemic models if those had been available.
The Senate of the United States has just approved an early stride into the use of broader
feedback analyses. In an approved amendment to the concurrent budget resolution for the
government’s 2014 fiscal year, it would require the CBO to estimate revenue changes in
18 of 24
connection with certain bills “that incorporates the macroeconomic effects of the policy being
analyzed” (S.CON.RES.8, 2013).
Not only legislators and their constituents would benefit from the P.L.I.S. approach. It offers
to provide invaluable “what if’, “what else” and “what more” guidance to guide legislators in
better lawmaking. It offers to challenge the sense of civic powerlessness before powerful
interests, causes and symbols that Stone validates within the “art” of public policy making. It
goes without saying (but still we will) that holistic thinkers and integrative scientists and
professionals would enjoy greater opportunities to deploy their talents in global support of
systems thinking, that integrative fifth discipline of essential organizational leaming. The Tax
Foundation certainly sees these possibilities. “Additionally, Congress should look to outside
groups ... to independently estimate the effects of tax changes. An open discussion of the
various models, and their underlying assumptions, would greatly improve the tax writing
process.” (McBride, 2013).
19 of 24
References
Arenas F, 2012. Stocks of knowledge and organizational performance. In Proceedings of the
30th International System Dynamics Conference, St. Gallen, Schweiz.
Bird C. 2010. Car Loans: How Long Should They Be? Cars.com. July 9, 2010, retrieved March
3, 2013.
Bismarck OEL v. 1867. Interview (11 August 1867) with Friedrich Meyer von Waldeck of the
St. Petersburgische Zeitung. Reprinted in First Bismarck: neue Tischgesprache und Interviews,
1: 248.
Blinder A. 2008. A modest proposal: eco-friendly stimulus. The New York Times. July 27, 2008.
"C.A.R.S.". 2009. United States. Supplemental A ppropriations Act (Pub. L. 111-32, H.R. 2346,
enacted June 24, 2009). Title XIII, Consumer Assistance to Recycle and Save Program.
Congressional Budget Office. ("CBO"). 2009. United States Congress. Cost Estimate for H.R.
2454, American Clean Energy and Security Act of 2009. June 5, 2009, retrieved November 23,
2009 from http://www.cbo.gov/ftpdocs /102xx/doc10262/hr2454.pdf.
Congressional Record. 2009. United States. Congress. House of Representatives. Remarks of
Congresswoman Betty Sutton, D-OH. 111 Cong. Rec. H6348.
Council of Economic A dvisors. United States. 2009. Economic Analysis of the Car Allowance
Rebate System ("Cash for Clunkers"). September 10, 2009.
Crain Communications, Inc. 2012. United States automotive industry new vehicle inventory and
sales data, Ward's Automotive Group custom data. 2005-2012 US Sales by model line and
2005-2012 US Inventory by model line. Purchased A pril 26, 2012.
Ewing J. 2009. Car-scrapping plans -- Germany's lessons. Der Spiegel Online. May 7, 2009,
retrieved March 24, 2010.
Fu L. 2009. Cash for clunkers: bad for environment?, My Fox Chicago. August 4, 2009, 8:39 PM
CDT, retrieved August 12, 2009.
General A ccounting Office. United States. 2003. Report to the Committee on Finance, U.S.
Senate, Vehicle Donations: Benefits to Charities and Donors, but Limited Program Oversight.
Glaeser E. 2009. Program has clunky reasoning. Boston Globe. August 8, 2009.
Global System Dynamics and Policies project, (2009), GSD partner directly informs Estonian
parliamentary policy. http://www.globalsystemdynamics.eu/index.php. Retrieved 2010-12-06.
20 of 24
H.R. 2751. United States. Congress. Consumer Assistance to Recycle and Save Act (as
introduced in House) 111th CONGRESS 1st Session June 8, 2009.
Joyce C. 2009. Critics say ‘Clunkers' program isn't very green. NPR's All Things Considered.
August 3, 2009, retrieved August 11, 2009.
Kiley D. 2009. How to rewrite cash for clunkers legislation. Bloomberg Business Week. Posted
July 29, 2009, retrieved March 15, 2013.
Labedz CS, Cavaleri SA, Berry GR. 2011. Interactive knowledge management: putting
pragmatic policy planning in place. Journal of Knowledge Management 15(4), 551-567.
Lindblom CE. 1980. The Policy-Making Process. Prentice-Hall: Englewood Cliffs, NJ.
Lyneis J. 2000. System dynamics for market forecasting and structural analysis. System
Dynamics Review 16(1): 3-25.
McBride W. 2013. Senate votes to require CBO to use macroeconomic revenue analysis.
TaxF oundation.org. March 25, 2013. Retrieved May 29, 2013, from
http://taxfoundation.org/blog/senate-votes- require-cho-use-macroeconomic-revenue-analysis.
McCullagh D. 2009. Cash for clunkers does little to help the environment. CBSnews.com.
August 5, 2009. Retrieved August 22, 2009, from http://www.cbsnews.com/blogs/2009/08/05/
business/econwatch/entry5217824.shtml.
Miravete EJ, Moral MJ. 2009. Qualitative effects of Cash for Clunkers programs. Centre for
Economic Policy Research Discussion Paper DP7517.
National Environmental Protection Act ("NEPA"). 1970. United States. Pub. L. 91-190, 42
U.S.C. 4321-4347, January 1, 1970, as amended by Pub. L. 94-52, July 3, 1975, Pub. L. 94-83,
August 9, 1975, and Pub. L. 97-258, § 4(b), Sept. 13, 1982.
National Highway Transportation Safety A dministration ("NHTSA"). 2011. United States.
Department of Transportation. Summary of Fuel Economy Performance. NVS-220. April 28,
2011.
Norusis MJ. 2007. SPSS 9.0 Guide to Data Analysis. Prentice Hall: Upper Saddle River, NJ.
Rahmandad H, Sterman JD. 2012. Reporting guidelines for simulation-based research
in social sciences. System Dynamics Review 28(4): 396-411.
S.CON.RES.8. United States. Section 416 of S.CON.RES.8 qais® Congress), the Concurrent
Budget Resolution passed by the Senate. March 23, 2013.
Senge PM. 2006. The Fifth Discipline: The Art and Practice of the Learning Organization.
Doubleday: New Y ork.
21 of 24
Shahabuddin S. 2009. Forecasting automobile sales. Management Research News 32(7):
670-682.
Shogren E. 2009. Charities lament ‘Cash For Clunkers' program. NPR All Things Considered.
July 14, 2009, retrieved August 19, 2009.
Sterman JD. 2000a. Learning in and about complex systems. System Dynamics Review 10(2-3):
291-330.
Sterman JD. 2000b. Business Dynamics: Systems Thinking and Modeling for a Complex World.
McGraw Hill Higher Education: Boston, MA.
Stone D. 2002. Policy Paradox: The Art of Political Decision Making. WW Norton & Company,
Inc: New Y ork.
Ventana Systems. 2003. Vensim: The Ventana Simulation Environment. Harvard, MA.
Walther G., Wansart J., Kieckhafer K., Schnieder E. Spengler T.S. 2010. Impact assessment in
the automotive industry: mandatory market introduction of alternative powertrain technologies,
System Dynamics Review 26(3): 239-261.
Westley C. 2009. The “I Hate the Poor” Act of 2009: Who Really Benefits from Cash for
Clunkers Program. Foundation for Economic Education. Retrieved August 22, 2009 from
http://fee.org/articles/hate-poor-act-2009/.
22 of 24
Fig. 2 CLD presenting non-policy-intended consequences of C.A.R.S. [restated here for readability]
<Favorability of
ge w onomic conditions>
Taxincentves to Chartable incentives CARS. iain
donate Va sho} fm for labor
a hours
Charities' donation Older used eee Late maida vehicles New vehicle 4
‘Older used vehicles
mw _— - Cp pp ately hel rar privately held>
Auto donations
{ GN f ) Git, bt: <Late model
Used car dealers’ Li? a vehicles on road>
Charities' incentives to ; inventories \,. vehicle Forecast demand for i: for
accept donations St i va inventories new vehicles
Public demand aa US.
(+ used vehicles Durable industrial cn supply (M3)
demand
Favorability of carl ( Durable personal
economic conditions Crent demand fot consumption U.S. population
new vehicles <Favorability of asa
economic conditions>
23 of 24
Figure 5. The complete S&F model for simulation
Further discot
due to CFC
Fractional rate for
recovering
Fractional
discouraging rate <Time>
Discourac
Economic factor
Usual fraction of looking
{ by late model year
y \vering rate owners
Usual fraction for looking
y older vehicle owners
Statuiory behavioral
incentives under C.A.R.S.
MGT's av
» ier Change in fraction of
expectation looking by late model Change in fraction for
year owners used vehicle sling
Used car held at] ,
Information delay Weight of economic denies Fractional rate for
‘Targeted time a wi son factor to consumer <Time> older vehicles trade in
behavior
Inventot Ov h Ki <Eco ic
y ra ms hip y foetonl sas _ IH Used car tree is Woe frectiontp <Loonon
Expected factor A rapping
\ Tate Model ] [Older Model Ye:
New Vatices] |e chaes Vehicles, ‘
= tel Change in fraction
Production | laveniones | Drivagtom |_onRoad | Auosacing | Scrapping fr crepcing
MGT's pre-crash___—" Showroom
overbuild Change in fraction
MGT's Total production ee for donating
post stimulus cut hous Donating
eae Usual fraction for
roduction donating
Tale total vehicles on- Donated Vehicles at
Toad tax exempt charities
24 of 24
