Evaluation of Consumer Surplus with Dynamic Demand

Author(s): J. M. Dargay and P. B. Goodwin

Source: Journal of Transport Economics and Policy , May, 1995, Vol. 29, No. 2 (May,
1995), pp. 179-193
Published by: University of Bath

Stable URL: http://www.jstor.com/stable/20053072

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

University of Bath is collaborating with JSTOR to digitize, preserve and extend access to
Journal of Transport Economics and Policy

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
 Evaluation of Consumer Surplus
with Dynamic Demand

By J. M. Dargay and P. B. Goodwin*

1. Introduction
Traffic growth, its effects on congestion and pollution, and the adoption of new policies
to cope with these problems, have prompted a reconsideration of some long-standing
established ideas in demand analysis and policy appraisal in transport studies. We argue
that an important part of this reconsideration is the adoption of dynamic methods in which
the response to policies is seen as a process which takes a lengthy period. This means
abandoning the assumption that an observed situation can necessarily be interpreted as an
"equilibrium" between demand and supply factors. Although interest in the idea of an
eventual equilibrium state of a system may remain useful, it will have to be accompanied
by consideration of the path and speed by which the system approaches that state.
Such work as has been carried out on dynamic approaches, in theory and practice, has
mainly focused on the empirical analysis of response of demand to changes in policy or
travel conditions. A recent literature review by Goodwin (1992) cited thirteen studies in
which the effect of fuel price on fuel consumption had been calculated (with a short-term
elasticity of approximately -0.25 to -0.3 and a long-term elasticity of -0.7 to -0.8);eleven
studies in which the effect of fuel price on traffic levels had been calculated (with results
of -0.16 for a short-term effect and about -0.3 for a long-term effect, and other studies not
specifying the time period showing results of about -0.5). Sterner et al. (1992) estimated
gasoline demand elasticities for twenty-one countries, separately. The mean result for the
preferred model form gave a short-run elasticity of -0.24 and a long-run elasticity of
-0.79. A review of Australian evidence by Luk and Hepburn (1993) cited twenty-eight
studies, and came to the conclusion that the long-term elasticity of traffic levels with
respect to fuel costs was -0.1 in the short run and -0.26 in the long run. Such findings are
mirrored in similar patterns on public transport fares elasticities.
It seems reasonable now to say that the key result of such work ? that long-term
transport demand elasticities are different from, and usually bigger than, short-term

* Transport Studies Unit, University of Oxford. This paper is a contribution to the project "The real effects
of environmentally friendly transport policies" of the UK Economic and Social Research Council, and a
precursor to TSU's research programme as the ESRC designated research centre in transport. An earlier
version (which includes details of the model specification and testing) was presented to the 11th Annual
Conference on Transport Research, Linkoping, Sweden, January 1994, and a summary of this to the European
Transport Forum, PTRC, September 1994.

179

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
May 1995 Journal of Transport Economics and Policy

elasticities ? is an established research finding, and is not substantially under challenge

in principle. Clearly further work continues to be necessary, especially aimed at under
standing the behavioural responses at an individual level which underpin such aggregate
responses: this will involve use of disaggregate time-series data, especially panel surveys
with repeated observations on the same individuals.
However, this paper focuses on a different implication of the empirical findings about
elasticities ? an implication which, as far as we have been able to discover, has been the
subject of little or no research, at least as applied to transport. What, if any, are the
consequences for the formal business of economic evaluation?
There has been research on related questions of potential bias in evaluation procedures.
For example, Williams and Yamashita (1992) considered the bias in evaluation of road
schemes resulting from ignoring induced traffic in congested conditions. It is obvious that
if the demand forecasts are wrong any economic evaluation based on those forecasts will
also be wrong. This applies whether the errors are due to a mis-specified equilibrium
model, or to the adoption of an equilibrium model when a dynamic one would have been
more appropriate. It is not so obvious whether the structure of evaluation itself is affected,
in particular the estimation of "consumer surplus" which plays a central role in social cost
benefit evaluation.
In this paper we explore the measurement of consumer surplus for demand changes
which take a significant period of time to be completed, both in theory and with a specific
application relating to car ownership and use.

2. The Case for Dynamic Modelling

An important task of transport methodology is the adoption of dynamic methods in which
the response to policies is seen as a process which takes place over time. An expanding
body of literature demonstrates that long-term demand elasticities are different from, and
usually bigger than, short-term elasticities, and behavioural research demonstrates that
adjustment to new conditions can take several years, or even decades. What happens to
consumer surplus if we replace analysis of an equilibrium demand curve by a time
dependent adjustment process?
Assume that demand is influenced only by prices and personal income, and compare
two different equations:
Dt=firPI?) (1)
Dt =/(/>? PM, Pt_2 ... Pt_? Ip /M, It_2 ... /^) (2)
In the first equation, demand in each year t is in equilibrium with the prices and incomes
in the same year: for any given price or income change, demand responds instantaneously
or at least so swiftly that complete adjustment is accomplished within one year. The
equation has no memory: the level of prices or incomes in previous years, whether prices
and incomes have risen or fallen to their current level, or the level of demand before price
or income changes, make no difference to the estimation of demand in the current period.
In the second equation, demand in each year t is not only, as before, influenced by the
prices and income in that year, but also by the history of the preceding levels. In this case,

180

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
Evaluation of Consumer Surplus with Dynamic Demand J. M. Dargay and P. B. Goodwin

the equation has some memory?it does make a difference to the demand in one year what
prices and incomes have been doing in previous years. The equation allows people to have
some delays in responding to changing circumstances.
While in principle there is an infinitely large number of different functional forms that
could be used for such delays, in practice it is often assumed that there will be a systematic
decay of the effect, so that the most recent years have a larger impact than the more distant
past. One particular functional form that has this property enables a widely-used
simplification to be adopted, namely
D, = /(/>,/,DM) (2a)
This equation has the form of a sort of inertia ? people's behaviour this year is still
influenced by what they did last year?but it is also a shorthand way of dealing with the
more complicated longer-term process, since demand in year (M) was itself influenced
by the prices and income in that year and the demand in the year before that (r-2) and so
on. This forms a chain: in other words, demand in any year is the result of the whole history
of previous prices and incomes, as in equation (2).
This is the simplest case of a dynamic model. Its first claim for support is that it
corresponds a little more closely with the realities of behaviour ? people do have
memories, and there are delays in their responses to price or income changes. One
rationalisation for this delay is the idea of adjustment costs. Consumers do not respond
immediately to price or income changes because there are costs of doing so. Typical costs
may relate to changes in the capital stock (for example, buying or selling cars) or acquiring
the necessary information to make utility-maximising decisions. Another reason for lags
in behaviour has to do with expectations: consumers will not react immediately to price
or income changes that may not be considered permanent. Still another justification lies
in the notion of permanent income: consumers do not necessarily maximise their
consumption patterns with respect to their current income, but instead to some concept of
income over a longer period of time.
If such dynamic effects are important, but we ignore them in specifying a static model,
we are not only depriving ourselves of useful information, but we are also risking bias and
misleading results even in the information the model does give us.
In the case of a series of data over time, it is a matter of choice whether we use one
approach or the other ? the data requirements are the same, and in general if we can do
one sort of analysis, we can do the other with the same data, though the answers will be
different and therefore our policy or commercial actions may be also.
This is seen even more strongly with models which are often used in transport studies,
based on disaggregate data about the behaviour of individuals. In this case the choice is
not so free; we can build a static or equilibrium model with just one survey at one point
in time, but a dynamic model is more data-hungry as it will need some sort of panel of
individuals who are surveyed on two or more occasions. The two approaches can give
quite different results.
In a dynamic model the coefficients give us information not only about the size of the
impacts, but about their dynamic qualities. This is shown in the following example, which
is the logarithmic form of equation quite often employed in practical estimation:

181

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
May 1995 Journal of Transport Economics and Policy

LnD, = a + bLnPt + cLnF, + dLnDt_x (3)

1-d = Proportion of total effect within first time period;
b = Short-run elasticity;
bl{\-4) = Long-run elasticity;
[log(^-l)]/[log d\ = Time for proportion/of total effect.
From such relationships, we can also calculate how much of the effect happens in five
years, how long before 90 per cent of the effect is realised, and so on.1
The second important advantage of dynamic models is that they tell us when an effect
will occur. The results of static models, by contrast, either have to be interpreted as along
term equilibrium with nothing at all to say about when it is reached, or tell us that the total
effect is virtually instantaneous, which is so indefensible that it is usually not mentioned.

3. Consumer Surplus Measurement:

Implications of Dynamic Price Elasticities
The benefits to travellers in period t from a discrete change in travel costs (whether this
is the generalised costs usually employed, or the special case of money price as used here)
at time 0 can be measured by the change in consumer surplus, measured by the area under
the demand curve at time t9 D{t)9 where D{i) =ft{c). Assuming D is integrable, the change
resulting from a change in costs from c to c is given by

B(0 = ?zX0<fc (4)

If D is a linear function of costs, this integral is evaluated as:

jC D{t) dc = -V2[/,(c')+ /,(c)](c'- c) (5)

where/,(c') is the level of demand in period t given a change in costs from c to c in time
0, and/,(c) = D{0)9 the level of demand at the initial price, c, which is the same for all t9
given that D(0) is in equilibrium. Although equation (5) is exact only for linear functions,
this same formula, known as the "rule of ahalf ' is a reasonably good approximation to the
actual integral for marginal changes in c even if D is a non-linear function of c. This is the
calculation carried out as routine in the UK Department of Transport for the evaluation
of road projects, using generalised costs which include the money value of changes in the
time spent travelling as well as money costs, butusing assumed equilibrium relationships.
Since demand does not respond instantaneously to changes in costs, we have a series
of demand curves, D{t)9 representing the level of demand in different time horizons from
t=0 to the time of complete adjustment, or equilibrium, r*, following a change in costs in
time 0. By definition, consumer surplus B{t) resulting from the change in costs is then also
dynamic and changes over time. The intertemporal profile of demand and benefits is
1 We note that an anonymous referee for an earlier draft of this paper commented *The algebra of the
difference between short- and long-run elasticities is known to every second-year undergraduate." We trust
this to be true, but have not found a single case where traffic forecasts or economic evaluation of aroad scheme
has been carried out using such knowledge.

182

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
Evaluation of Consumer Surplus with Dynamic Demand J. M. Dargay and P. B. Goodwin

generally also affected by non-price factors, so that D{t) =ft{z,c). However, in order to
simplify the exposition, we will assume that all other factors remain constant, so that
D{t+i) = D{t) V i if c is constant. The price elasticity is defined as:

rtrt.PXO-fl??] c _ao(0 c .
Kl)~ AC{0) D{0)~dC{0)D{0)' w
Here AC(0) is a change in costs from c to c occurring at time 0, D{0) is the level of
demand at time 0 given the initial price c, and D{t) - D{0) is the change in quantity
demanded after a time interval /. We would expect this elasticity to grow monotonically
with increasing time and approach some asymptotic value e*, the long-run elasticity, as
D{t) approaches its equilibrium value, D*. It is possible to have other adjustment patterns,
such as oscillation, but in most economic situations le(01< le(tt-l) |foralW<i*,ande<0,
so that D{t) < D{t+l) for price falls and D{t) > D{t+\) for price rises, assuming that D(0)
is in equilibrium. This implies that B{t) < 5(i+l) for price falls andB(0 >B(H-1) for price
rises. This is shown in Figure 1.
It is interesting to compare Figure 1 with the somewhat similar presentation which is
sometimes used in road project evaluation, to distinguish between the evaluation of
changes in consumer surplus due to reassignment (that is, route changes) and changes in
the total volume of traffic. The UK Department of Transport (1993) has assumed in most
cases that the total volume of traffic is unaffected by changes in the cost or speed of travel,
and therefore uses a vertical demand curve similar to that labelled D{0). It would be
possible to argue that changes in route choice are likely to be quite rapid, but that changes
in mode or destination choice, or trip frequency, would take longer to apply. If this were
true, then the vertical demand curve could be thought to apply, equivalently, to assignment
or to short-term effects, and the sloping curve to other responses or to longer-term effects.
It should be said that there is no empirical evidence known to the authors in support
of the contention that changes in route choice are completed so rapidly, and it is also
challengeable that even if they were, route changes can be treated as a vertical demand
curve themselves. Nevertheless, if established practice changes to include the broader
range of behavioural response, then it will in any case involve a shift in emphasis from the
vertical to the sloping demand curve. The analysis in Figure 1, however, does not seek only
to replace one equilibrium demand curve by another more sensitive one, but also to
consider the time-scale over which short-term responses are converted to longer-term
ones.
In order to assess the total costs or benefits to travellers of a given change in travel costs,
it is normal practice to calculate the present value of the future stream of costs/benefits.
Assuming a discount factor, r, and a time horizon, N9 the present value can be written in
discrete time as

N [mac
If demand does not respond instantaneously to the change in costs, the present value
will thus depend on the dynamic profile of this response. (Note that this is not comparable

183

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
May 1995 Journal of Transport Economics and Policy

COST
Fall in Costs from
c to c'

DEMAND d(0) d(t) d*

COST
Rise in Costs from
c to c'

C'

DEMAND d(t) d(0)

Figure 1
Over- or Underestimation of Consumer Surplus

184

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
Evaluation of Consumer Surplus with Dynamic Demand J. M. Dargay and P. B. Goodwin

to the superficially similar practice of summing and discounting a stream of future benefits
from a changing demand level over the years where the changes result from other
extraneous factors such as economic growth, not delayed responses to the initial cost
changes.) When the underlying demand responses are calculated from some form of
equilibrium model in which the time-scale of response is indeterminate, and implicitly
treated as instantaneous, this is equivalent to using a single-value static demand elasticity
to calculate the consumer surplus, rather than a time-dependent dynamic elasticity, so that
e(?) = ts V t > 0, where es is a constant. This assumption of a static elasticity also implies
that D{t) = Ds and thus that B{t) = Bs V / > 0, so that the (undiscounted) benefit is the same
for all t and the PVbecomes:

"W-^?iy- TO
The question that arises is: what happens to the calculation of present value when we
erroneously use a static elasticity?
If the source of the static elasticity is a good long-term equilibrium model, then its
implied elasticity is sometimes perceived as the long-term asymptote of a dynamic
process. On the other hand, if the source is some sort of non-lagged time-series analysis,
it is likely that the elasticity will be close to the initial or short-term value of a dynamic
elasticity. (Neither of these dynamic interpretations of static elasticities is always valid,
because of mis-specification bias, but it is convenient to take these as the limiting cases
at this stage of the argument.)
It emerges that the nature of the error caused depends on the relationship of the static
elasticity to the dynamic elasticity, the speed of adjustment to equilibrium, the discount
rate r, and the time horizon, N. In the first example, the long-run elasticity, e*, is used
instead of the dynamic elasticity. In this case I e* I > I e{t) I Vt<t*. This implies that D* >
D{t) and 5* >B{t) for price falls andD* <D{t) andB* <B{t) for price rises. Thus the benefit
to consumers of price decreases is overestimated and the (negative) benefit or loss entailed
by price increases is underestimated, as compared with the true situation where initial
small responses build up over a period into bigger responses. In the assessment of
intertemporal benefits, each term in the summation in the present value calculation is over
or underestimated as shown above, so that the sum of the discounted benefits will include
the sum of these errors, and thus be over- or underestimated in the same fashion. The
slower the speed of adjustment, the greater will be the discrepancy. However, as the time
horizon increases, N approaches or exceeds i*, so the error becomes smaller. Also, the
higher the discount rate, the less the weight to future benefits, which are more nearly
accurate, so the error increases.
In the second case, suppose we erroneously use a short-run elasticity, e(l), where
|e(l) |< |e(0 |Vf>l,insteadofthedynaiiiicelasticitiese(O.Thenforr>l,D(l)<D(Oand
5(1) < B{t) for price falls and D{\) > D{t) and S(l) > B{t) for price rises, so the benefit of
price decreases is underestimated and the loss of price increases is overestimated. Again
each term in (7) with the exception of fi(0) and B{\) is over- or underestimated, as is the
discounted stream of future benefits. The effects of differences in the speed of adjustment,
discount rate and time horizon are the opposite to the case above.

185

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
May 1995 Journal of Transport Economics and Policy

The biases described are not due essentially to the mistake of using an elasticity which
is erroneously higher or lower than the true value (although this is important for other
reasons). They arise from using a single value, whatever it is, instead of a time-dependent
function. However, given that the two limiting cases discussed have opposite patterns of
bias, it is interesting to suggest that there is likely to be some notional value of an elasticity
? somewhere between the short-term and the long-term value ? whose use as a static
elasticity would give the same value for consumer surplus as a full calculation of the path
of response over time. This value will be in the nature of a pivot representing the point of
transition from underestimate to overestimate of benefits, and for convenience we call it
the "Pivot Elasticity". The Pivot Elasticity will depend on the intertemporal response
pattern, the discount rate and the time horizon chosen.
In order to examine this, we will look at a simple model, based on the monotonically
increasing elasticity discussed above. One commonly used rationalisation of this is
provided by the partial adjustment or habit persistence model. Assume that there exists a
desired (or equilibrium) level of demand at time t9 D*(i), which is determined by the level
of costs at time t and other factors. Assume also that in any given time period, consumers
only manage to move aproportion, <(>, of the difference between the desired demand at time
t9 D*(0, and the previous value of demand D(/-l). The proposed adjustment process may
be specified as shown in equation (9).
[D{t) - D(r-l)] = <|>[D*(i) - D(i-l)] (9)
where 0<<|><1 is the adjustment coefficient. The closer (|) is to unity the greater is the
adjustment made in the current period and the shorter the time required for complete
adjustment. If (|> = 1, adjustment is instantaneous, so that D{t) = D*{f) V t9 and demand is
always in equilibrium. Solving for D{t)9 equation (9) becomes:
D{t) = <t>D*{t) + {\-?)D{t-\) (10)
The intertemporal demand pattern resulting from a change in costs at time 0 is given
by continual substitution into the lagged values of D{i)

D{t) = 0D*'? (1 - <l>)s + (1 - 0)'D(O) (11)

where D*(r) = D* Vi, assuming a stationary equilibrium. Summing the geometric series
gives
D{t) = D*[l - (1-^yi + (l-<|>yD(0) (12)
so that the time profile of demand is a weighted average of the equilibrium value, D*, and
the initial value, D(0). It is easily seen that the greater the time elapsed from the price
change in time 0, the nearer to D* one becomes. The dynamic elasticities can be derived
from this equation:

e(0 = eV?(l-#
S=0 = e*[l- (1-0/] (13)
so that as t increases e(i) increases in absolute value and the effects during each time period
decline geometrically over time.
Given this adjustment pattern, the change in consumer surplus at time t9 B{t)9 resulting
from a change in costs from c to c at time 0, can be expressed as a weighted average of

186

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
Evaluation of Consumer Surplus with Dynamic Demand J. M. Dargay and P. B. Goodwin

the consumer surplus at equilibrium, B*, and the change in costs at the original level of
consumption, 5(0):

B{t) = P D{t)dc = [Ml-*?] f D*dc + (1-0?ZXO)JT dc = [M1-4/1B* + (1-^)^0) (14)

In other words, B{t) is a weighted average of the consumer surplus calculated at elasticity
?* and e(0) = 0. If instead of using the dynamic elasticities, e(0, a single, static elasticity,
e(0) < e{s) < e* in absolute value, is used for all time periods, B will be overestimated for
some time periods and underestimated for others.
The discounted present value of the future stream of benefits in equation (13) can be
written:

There will exist one elasticity e(m), 0 < m < f*, such that the present value evaluated at e(m)
will be equal to the present value calculated on the basis of the fully dynamic demand
function. This is the Pivot Elasticity. The present value calculated by using the Pivot
Elasticity is:

^.?tiizStp.mtlkt?.
r = 0 (1+r) * = 0 (1+r)' (16)
Setting PV{Bm) from (16) equal to PV{B) from (15) and cancelling common terms,
gives:

%<??-&Wf 07)
which is solved for m in terms of <)>, N and r, by summing the geometric series and taking
logs:

[(l-^^-lKr^)
m~ log(lHfr) (18)
A simpler form can be obtained by allowing N to be equal to infinity and summing the
infinite geometric series. This gives:

m=1^l (19)
In the special case where r - <|>,
loi
Hi)
m=toi(ii?) <20?
which defines the median lag, or the time required for one half of the total adjustment to
be completed. In this case m will be equal to that time period corresponding to half of the
long-run response, so that e(m)=e*/2. Thus in this case, the pivot elasticity is defined from
the demand conditions when the adjustment process is half completed, which makes
intuitive sense.

187

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
May 1995 Journal of Transport Economics and Policy

Table 1
Values of m for Various <f> and Times Required for Adjustment

= 0.1 0.3 0.5 0.7 0.9

N=30 6.8 4.1 2.7 1.8 1.0

N=oo 7.7 4.4 2.9 1.9 1.1

%e* 50 75 80 85 90

Median lag 6.6 1.9 1.0 0.6 0.3

90% e* 21.9 6.5 3.3 1.9 1.0

More commonly, however, r is smaller tha

will be greater than the median lag. Table 1
coefficients, (|>, under the commonly made
for m for N equal to infinity are also shown, a
relating to m, median lags and the time require
completed.
As ((> increases, the speed of adjustment increases, while the median lag decreases as
does the time for total adjustment. Thus the value of m also decreases. For example, if
adjustment is relatively slow, so that 0 = 0.1, we would need to use the elasticity for time
period m = 6.8 years, or 50 per cent of the long-run elasticity. On the other hand, if
adjustment is very rapid, so that <|) = 0.9, we should use the elasticity for time period m =
1, which in this case would be 90 per cent of the long-run elasticity. Clearly if adjustment
is very rapid, there is little difference between the short- and long-run elasticities, so that
the inaccuracy in the present value calculations in using one or the other would be minimal.

4. Estimation of Dynamic Elasticities for Car Ownership and Use

We have seen above that the dynamics of adjustment are important in determining the
change in consumer surplus resulting from a change in costs. It was also demonstrated that
in the general case, the use of a static elasticity to determine consumer surplus will result
in an over- or underestimation of the benefits or losses entailed. Despite this, it is general
practice to resort to using static elasticities in the assessment of the effects of various
transport policies and projects. Unfortunately, the importance of dynamics and the use of
dynamic models in transport applications is still not fully accepted, although the
arguments supporting the idea of differences between short- and long-run elasticities
seem self-evident.

188

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 1 01 Jan 1976 12:34:56 UTC
All use subject to https://about.jstor.org/terms
Evaluation of Consumer Surplus with Dynamic Demand J. M. Dargay and P. B. Goodwin

Theoretically, cross-section data will capture long-run relationships only if the

observations themselves are in equilibrium. As household income and make-up are
changing over time, this is hardly likely to be the case. However, there are other
disadvantages. First, such estimates give us no indication of how long it takes to reach
equilibrium or of the adjustment path. Obviously, it makes a difference to consumers'
welfare as well as to traffic forecasts whether the estimated response to a price change
takes two or twenty years. Secondly, some variables of interest which change over time
will be constant, or very nearly so, for all observations in a cross-section, or may be
unavailable, so that their influence cannot be estimated. Travel costs are a good example.
Although cross-section studies provide a basis for understanding the differences in travel
behaviour amongst individuals, they are less useful in forecasting travel demand or in
analysing the effects of transport policy, both of which are generally concerned more with
aggregate effects, rather than individual behaviour, and which presuppose some sort of
time dimension.
It would thus seem preferable to base such analyses on time-series data. However,
interpreting estimates of a static model on the basis of such data is not at all clear cut: the
common assertion that such estimates represent short-run values is not grounded in fact.
Under certain conditions the long-run relationship can be obtained from the static
regression (if adjustment is instantaneous or if the variables concerned are non-stationary
and cointegrated). Under other conditions the effects of ignoring the dynamics will result
in mis-specification and biased estimates which will be difficult to relate to any particular
time period.
However, even the estimation of a dynamic model on the basis of time-series data is
not unproblematic. In most instances, such studies are based on aggregate data; instead of
having the wide-range of individual behaviour characteristic of survey data, only
aggregate, or "average", behaviour of all individuals over a given period of time is
available. This entails not only a serious loss of information, but also gives rise to
aggregation problems. The theory of consumer demand is derived from the utility
maximisation of individuals, and the properties of these demand functions are not
necessarily carried over to the aggregate functions unless rather strong assumptions are
made, for example, that all consumers have the same preferences or that the individual
utility functions are closely distributed around some "average" function.
In order to avoid the problems of both purely cross-sectional and aggregate time-series
data, one would ideally base analysis of transport demand on observations of individuals
over time: on panel data. Unfortunately such data are not often available for the long
periods of time needed to investigate behavioural changes which may take years to
complete.
Empirical estimates of income and cost elasticities, available in a fuller paper from the
authors (Dargay and Goodwin, 1994) are now used to illustrate the implications of using
the dynamic formulations for consumer surplus calculations. The estimates are based on
a double-logarithmic partial adjustment model, in which both car ownership and use are
assumed to be determined by income, the purchase price of cars, car running costs and the
cost of public transport, all in real terms. Although the model is a relatively simple one,

189

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
May 1995 Journal of Transport Economics and Policy

.?| 4 I_I_I_I_I_I_I_I_I_I_L_J_I_I_L_J_I_I_L__J_I_I_I_L_J_I_I_I_I_L.
' 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
YEAR AFTER COST CHANGE

Figure 2
Dynamic Cost Elasticities

it is one commonly used in aggregate consumer demand studies and generally results in
reasonable estimates. The dynamic running cost elasticities for ownership, use per car and
traffic are shown in Figure 2.
In the short run, changes in running costs affect car use much more than they affect car
ownership, whereas in the long run the impacts are more or less identical: half occur
through changes in use per car and half through changes in the number of cars. In all cases,
the long-run elasticities are rather large, over -0.6 for both car ownership and car use, and
-1.3 for total car kms. It should be pointed out, however, that our definition of running
costs includes much more than petrol prices, so we would expect the elasticity to be higher
than those obtained for petrol prices alone. On average over the period petrol costs
accounted for around 50 per cent of total running costs. If the elasticities were the same
for petrol prices as for other running costs our results would imply a long-run elasticity
of car traffic to petrol prices of around -0.6. The difference in the speeds of adjustment
are apparent from the figure: car use adjusts relatively quickly, car ownership slowly, and
total car traffic somewhere in between.
Given the double logarithmic functional form we can calculate the change in consumer
surplus resulting from a change in costs from c to c by integrating the demand curve for
car traffic, T. The change in consumer surplus at time t is given by:

(21)
B(r)=?rwc=r(0)^|^^
190

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
Evaluation of Consumer Surplus with Dynamic Demand J. M. Dargay and P. B. Goodwin

Table 2
Percentage Error in Consumer Surplus Using Static Instead of Dynamic Elasticities

Static Elasticity Percentage Change in Running Costs

Increase Decrease

10 20 50 10 20 50

SR -0.4 3 5 12-3-6 -18

LR -1.3 -1-2-5 1 3 9

The change in consumer surplus at time / is thus determined by the initial level of
demand, the change in price and the dynamic elasticity, zTRC. From this relationship we
can calculate the percentage error in the estimated present value of benefits or costs
resulting when a static elasticity is incorrectly applied instead of the (correct) dynamic
one. For a given discount rate and time horizon, it can be shown that for this particular
model specification, the percentage error will depend only on the elasticity and the
direction and magnitude of the relative change in price. In Table 2 an example is given
using the estimates of the elasticities for total car traffic with respect to running costs. A
discount factor of 8 per cent and a time horizon of 30 years are used for the present value
calculation. The errors resulting for cost increases and decreases of 10 to 50 per cent are
shown for two separate cases. The first uses the estimated short-run elasticity, while the
second uses the long-run elasticity.
The errors in Table 2 are expressed as a percentage of the total change in consumer
surplus. For practical purposes, they will usually need to be converted into a percentage
of net present value, which will always be greater than the above figures : for the interesting
policy decisions where benefits are within 20 per cent of costs, for example, biases of this
scale in consumer surplus will normally substantially influence the rate of return and may
reverse the decision.
We see that for small changes in price, the error is marginal regardless of which
elasticity is used. However, the error increases as the price change becomes greater, and
for large price changes the error is quite substantial. Further, we see that using the short
run elasticity overstates the loss incurred by a price rise, while using the long-run elasticity
understates it. For a price fall, the opposite occurs: the short-run elasticity understates the
benefits gained while the long-run elasticity overestimates them. However, the effect is
only symmetric for rising and falling prices when the price change is marginal. As the
percentage change in price becomes larger, the asymmetry increases. In this case, the error
is greater for price falls than for price rises. This asymmetry arises because of the non
linear functional form used for the demand function; with a linear demand function such

191

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
May 1995 Journal of Transport Economics and Policy

asymmetries would not exist. Also, in this particular case the error is much greater using
the short-run elasticity than the long-run elasticity. Although this need not be the case, it
will generally be so unless the chosen time horizon is very short. Finally, as shown earlier,
we can find a static pivot elasticity which will yield zero error. In this case, the pivot
elasticity is about -1.0.

5. Conclusions and Policy Implications

Despite the rather simple models used and the preliminary nature of some of our
econometric results, a few conclusions can be drawn. Most importantly, using static
elasticities for transport policy assessment could lead to erroneous estimates of the impact.
It could be argued that for small cost changes the errors will be small and could be ignored.
For large changes, however, the errors can be significant.
Finally, we should move cautiously from theory to practical implications. Here, the
direction of bias will depend on the specific context of current practice. For example,
consider the case of a pricing policy in which the price of road use is increased (by petrol
prices or a road-pricing scheme), the revenue being used to support public transport.
Current assessment of such a policy would be influenced by petrol price elasticities and
public transport fares elasticities which, being drawn from unlagged analysis of time
series data, are significantly less than empirically estimated longer-term figures. Moving
to median or "pivot" elasticities would in both cases require an increase in the elasticity
assumed.
It may therefore be suggested that the policy bias in current practice leads to an
overestimate of the welfare loss in raising prices (in this example, for road use), and an
underestimate of the welfare gain in reducing prices (in this example, for public transport).
The static evaluation of consumer surplus changes would therefore seem to make the
policy appear less worthwhile than it really is.
On the other hand, some policies are evaluated using transport models, usually
calibrated on cross-section data, which claim to give a long-run equilibrium forecast of
the effects of the policies under test. This is especially so in the case of evaluation of major
infrastructure investments, in which travel time rather than money price is commonly the
largest element of benefit. If the model claims are deserved, then the implied elasticities
will be those that apply in the long run, and will be greater than median or pivot values.
In this case, the bias is the reverse: the value of reducing costs will be overestimated, and
the disbenefit of increasing them will be underestimated.
A last, and particularly important, example would be the evaluation of a "package" of
policies, some of which have short-term effects, and others longer-term effects. It would
cause distortion if the package as a whole is evaluated in terms of its final equilibrium,
without consideration of the different time-scales of the various components.
In summary, therefore, we propose that policies or projects to be compared should be
evaluated with explicit treatment of the time-scale of the behavioural response which will
apply. The assumption that evaluation can be carried out by reference to an assumed
equilibrium will lead to biased choices.

192

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms
Evaluation of Consumer Surplus with Dynamic Demand J. M. Dargay and P. B. Goodwin

6. Postscript
When this paper was written, it was the view of the authors that explicit consideration of
the dynamic properties of behavioural adjustment was a logically necessary development
in economic evaluation of transport projects, but that the long-established domination of
equilibrium models might make this difficult to achieve in practice. In December 1994,
the report of the Standing Advisory Committee on Trunk Road Assessment (SACTRA,
1994) recommended that the additional traffic induced by extra highway capacity should
be included in trunk road appraisal. In accepting this advice, the UK Department of
Transport (1994) issued a technical guidance note on how this should be done, which
includes a table of recommended generalised cost elasticity values. The note says:

"These values represent short-term elasticities and should be used for sensitivity
tests of opening year flows. Larger values of up to twice the above should be used
for sensitivity tests of design year flows."

The "design year" is usually the fifteenth year after opening. It will be obvious that if the
elasticities used for the fifteenth year after opening are about twice that used for one year
after opening, all the questions of an adjustment path discussed in this paper become
relevant, and indeed it would be impossible to calculate a net present value unless such
a path is assumed. It may be that official procedures for evaluation will need to "become
dynamic" more swiftly than we had expected.

References
Dargay, J. M. and P. B. Goodwin (1994): "Transport Evaluation in a Disequilibrium World: Some Problems
in Dynamics". 11th Annual Conference on Transport Research, Linkoping, Sweden.
Department of Transport (1993): COBA 9 Manual. (Revisions up to 15.11.1993). London: Department of
Transport, Highways and Economics Division.
Department of Transport (1994): Guidance on Induced Traffic. HETA Division.
Goodwin, P. B. (1992): "A Review of Demand Elasticities with Special Reference to Short and Long Run
Effects of Price Changes". Journal of Transport Economics and Policy, vol.26, no.2, pp. 155-70.
Luk, J. and S. Hepburn (1993): "New Review of Australian Travel Demand Elasticities". Victoria: Australian
Road Research Board.
Standing Advisory Committee on Trunk Road Assessment (1994): Trunk Roads and the Generation of
Traffic. London: HMSO.
Sterner, T., C. Dahl and M. Franzen (1992): "Gasoline Tax Policy, Carbon Emissions and the Global
Environment". Journal of Transport Economics and Policy, vol.26, no.2, pp. 109-20.
Williams, H. C. W. L. and Y. Yamashita (1992): "Travel Demand Forecasts and the Evaluation of Highway
Schemes Under Congested Conditions", Journal of Transport Economics and Policy, vol.26, no.3,
pp.261-82.

Date of receipt of final manuscript: August 1994

193

This content downloaded from

14.169.16.27 on Tue, 07 Sep 2021 13:21:23 UTC
All use subject to https://about.jstor.org/terms