Hotelling Lemma150319 PDF
Hotelling Lemma150319 PDF
Hotelling Lemma150319 PDF
Gérard Gaudet
Département de sciences économiques and CIREQ
Université de Montréal
June 2007
1 Presidential address delivered at the 41st annual meetings of the Canadian Economics Associa-
tion, Halifax, 2 June 2007. I would like to thank Didier Tatoutchoup for his assistance in collecting
the data and Isabelle Boudreau for bibliographic assistance. Very special thanks also to my friends
and long time collaborators Pierre Lasserre, Ngo Van Long, Michel Moreaux and Stephen Salant
for their comments and especially for the many enjoyable and profitable discussions about natural
resource economics we have had over the years.
1 Introduction
The year 2006 marked the 75th anniversary of the publication in the Journal of Political
research as this one has had on the economics of exhaustible resources.1 This anniversary
provides a good occasion to look back at how the famous Hotelling’s rule of natural resource
exploitation has evolved as a framework for understanding the functioning of natural resource
markets and to look forward at some unresolved issues, both theoretical and empirical, that
would help reconcile the theory and the facts about resource price behavior over time.
Hotelling’s motivation in writing that paper was twofold. He was reacting to the de-
mand for regulation of the exploitation of exhaustible natural resources by the Conservation
Movement, which had been particularly strong in the U.S. during the period 1890 to 1920.
But he also wanted to propose an economic theory of exhaustible resources to remedy the
inadequacy of what he called the “static-equilibrium type of economic theory” for analyz-
ing an inherently dynamic problem. His analytical approach to the problem of the efficient
use of exhaustible resources was similar to that which he had used in a wonderful paper
published in 1925, in the Journal of the American Statistical Association, titled “A General
Mathematical Theory of Depreciation” (Hotelling 1925). In that paper, he introduced the
definition of the depreciation of an asset as the decrease in the discounted value of its future
returns. Not only is the topic related to that of the depletion of a natural resource, but it
required the use of the calculus of variations, as did his economics of exhaustible resources.
Hotelling’s treatment of the economics of exhaustible resources did not get serious at-
tention before the 1970s. There are probably two explanations for this. The first is that
1
By exhaustible resources Hotelling means nonrenewable resources, as opposed to renewable resources.
The expression “exhaustible resource” is somewhat of a misnomer, since both type of resources are ex-
haustible. Hotelling recognizes that “a complete study of the subject would include semi-replaceable assets
such as forests and stocks of fish”[renewable resources], but chooses to limit himself to “absolutely irreplace-
able assets”[nonrenewable resources] (Hotelling 1931, p. 139–140).
1
economists, even academic economists, read the newspapers, and the newspapers’ attention
was directed to other more pressing issues in the middle of the Great Depression, as it was
during most of the war years that followed in the 1940s.2
The adequacy of natural resources for sustained economic growth became a concern after
World War II, particularly in the U.S. This concern gave birth to the President’s Materials
Policy Commission (Paley Commission), whose report was made public in 1952, and, as an
outgrowth, to Resources for the Future, the influential Washington think tank. Hotelling’s
contribution nonetheless went unnoticed throughout the 1950s. Resources for the Future
sponsored the publication in 1963 of Scarcity and Growth, by Harold Barnett and Chandler
Morse (Barnett and Morse 1963), which was the first systematic analysis of long run scarcity
measures of a number of natural resources. Barnett and Morse make just a passing reference
to Hotelling’s 1931 paper.3 So lack of public concern for the issues of natural resource use is
not sufficient to explain the lack of interest for Hotelling’s contribution before the 1970s.
The other reason why the paper did not get the early attention it deserved is that it was
a difficult paper, given the level of mathematical sophistication of the economic profession at
the time. In fact, according to Hotelling’s own later account, his paper was rejected by the
Economic Journal because its mathematics were too difficult (Arrow 1987, Darnell 1988).
Then came, in the 1970s, another period of intense public concern for natural resource
scarcity. This was triggered by Limits to Growth, the report for the Club of Rome published
in 1972 by Dennis Meadows and his co-authors (Meadows et al. 1972), predicting catastrophic
consequences for the early twenty-first century unless economic growth was seriously cut
back, and by the 1973 oil embargo and the ensuing energy crisis. This time academic
economists were ready to tackle the analytical framework put forth by Hotelling in 1931,
being equipped with the optimal control theory learned by working on optimal economic
2
The same volume of the Journal of Political Economy contained a paper titled “Some theoretical aspects
of stock market speculation” and another titled “Savings, investment, and the control of business cycles”
which were certainly more topical at the time.
3
They characterize Hotelling’s paper as “contribut[ing] usefully in a technical way [my emphasis] to the
question of whether, and why, the time distribution of resource use and investment under private enterprise
conditions may be socially erroneous.”(Barnett and Morse 1963, p. 46)
2
growth problems in the 1960s. William Nordhaus published in 1973 an important study
entitled The Allocation of Energy Resources which applies the Hotelling model to the problem
of how different energy sources should be used (Nordhaus 1973). He also published the next
year a critical analysis of Limits to growth in the American Economic Review (Nordhaus
1974). Robert Solow devoted his 1974 Ely Lecture to the American Economic Association
to the subject of natural resource economics (Solow 1974). The same year the Review of
Economic Studies published a Symposium devoted to the topic of exhaustible resources,
with contributions by Partha Dasgupta, Geoffrey Heal, Tjalling Koopmans, Joseph Stiglitz
and Robert Solow, among others. Natural resource economics began to acquire greater
prominence as a field of research. The now famous Hotelling’s rule was at the heart of it
2 Hotelling’s rule
Like the stock of any physical capital, a stock of natural resource in situ is an asset to its
owner. In a market economy, the value of this asset, like for any other capital asset, will be
Typically, the rate of return on a physical asset can be decomposed into three components:
• A first component is attributable to the flow of product generated by the marginal unit
of the asset — its rate of marginal productivity or dividend rate.
• A second component is due to the fact that the asset’s physical characteristics may
change over time, a factor which may or may not depend on the use being made of it
• A third component is the rate at which the asset’s market value changes over time.
This may be negative, as long as it is offset by some other positive components of the
return.
4
That is not to say that there were no significant contributions to the natural resource literature before
the 1970s. To mention a few: Gray (1914), Scott (1955b), Gordon (1967), Herfindahl (1967) as concerns
nonrenewable resources; Gordon (1954), Scott (1955a) and Smith (1968) as concerns renewable resources.
3
In order for the asset markets to be in equilibrium, this rate of return must be equal to the
rate of return that can be expected by its owners if they were to sell the marginal unit and
invest the proceeds elsewhere. I will assume for now that the alternative is a “bond” which
yields an exogenously given riskless rate of return and will call it the “rate of interest”.
In the case of some physical asset such as buildings, machinery and equipment, the first
component usually refers to the marginal product derived from its use as an input in the
production process. The second component refers to the rate of physical depreciation of the
asset, which enters negatively in the rate of return. It may be due to wear and tear from the
use of the asset or simply to obsolescence, and is often treated as a constant in theoretical
work. The third component, refers to the capital gains that can be had by holding this asset.
Suppose now the asset in question is a nonrenewable resource, such as a mineral deposit
or a stock of oil in the ground. Such assets, being non reproducible, have the property that
the size of their existing stock cannot be increased over time. Furthermore, holding such an
asset in situ yields no dividend: as long as it is left in the ground, it is totally unproductive,
services. Hence the first component is identically zero. As for the second component, there
is usually no exact equivalent in the case of resource stocks, in the sense that physical
deterioration will not occur from simply holding the asset in the ground. There is a sense
however in which keeping the marginal unit of the asset stored in the ground rather than
extracting it prevents the average quality of the remaining stock from deteriorating. This
second component therefore enters positively in the rate of return, rather than negatively.
Let me also set this component to zero for now and come back to it later. This leaves the
rate of appreciation in value as the only source of return on the stock of natural resource.
The value of the marginal unit of resource held in the ground is what it can fetch on the
flow market, net of the cost of taking it out of the ground. If p(t) is the current flow price
which the resource can fetch on the market once extracted and c(t) is the marginal cost of
4
extracting it at date t, then its marginal value in the ground must be:
π̇(t)
= r. (1)
π(t)
This is the famous Hotelling’s rule, derived as an equilibrium condition on the asset markets.
It states that the net price of the natural resource — its asset price — must grow at the rate
of interest.
If the marginal cost of extracting the resource is independent of the rate of extraction
and invariant over time, then this immediately yields a prediction as to the behavior of the
If this were a correct representation of reality, we should observe the price of nonrenewable
resources continuously growing at a rate which tends to the rate of interest as the share of
cost in price gets smaller and smaller over time and that of the scarcity rent gets higher and
higher.
This is not what we have been observing over the last century and more. Using U.S.
price data for the period 1870-2004 for copper, lead, zinc, coal and petroleum, 1880-2004
for tin, 1900-2004 for aluminium and nickel and 1920-2004 for natural gas, I have plotted
in Figures 1 to 10 the rate of change of price of each of those seven nonrenewable minerals
and three nonrenewable fossil fuels. The first thing that we notice is the high volatility in
the rate of change of those prices. But more significantly for the purpose at hand is the fact
that this volatility appears centered at zero. In fact, in none of the ten cases is the mean
rate of change of price significantly different from zero.5 It is very hard to detect any trend
5
The p-values range from 0.82 to 0.98.
5
in the actual price levels of those resources and certainly not the kind of positive trend that
be more to the return on holding nonrenewable natural resource stocks than simply the rate
of price appreciation. Factors related to extraction costs, to durability of the resource, to
market structure and to uncertainty must also be taken into account in order to correctly
characterize the rate of return on resources. In what follows I will discuss in turn how each
of those factors could potentially help bridge the gap between the underlying theory and the
evidence.
3 Costs
The first factor that comes to mind which is likely to attenuate the growth in the flow price
change was an important factor in the finding of Barnett and Morse that, except possibly for
forestry, the average cost of extractive resources had been following a declining trend over
For the sake of the argument, write the unit cost of extraction as c(t) = ce−αt , to reflect
exogenous technological progress at the rate α in the extraction technology. Then Hotelling’s
rule yields the following prediction for the rate of change in the market price p(t):
ṗ(t) c(t) c(t)
=r 1− −α . (3)
p(t) p(t) p(t)
The rate of change of the market price is thus the weighted sum of the rate of interest and
the rate of decrease in cost due to technological change. If the share of marginal cost in price
6
The analysis by Berck and Roberts (1996) of the actual prices of nine nonrenewable natural resources
up to 1991 (the same resources as in Figures 1 to 10, plus iron and minus nickel and tin) using time-series
methods leads them to conclude that there is at best “a weak supposition that natural resource prices will
rise” (Berck and Roberts 1996, p. 65). A quick glance at the evolution of those prices between 1991 and
2005 leads one to conjecture that adding those fourteen years of observations would make the supposition
weaker still.
6
is sufficiently high, the effect of the rate of technological change on cost dominates. The
reverse is true if the share of marginal cost in price is sufficiently low. This is consistent with
a price path that would be at first decreasing and then increasing (Slade 1982). It cannot
however justify forever flat or decreasing price paths. In the long run, as the share of cost
in price becomes negligible, the effect of the rate of interest must come to dominate and
the rate of growth of price must eventually become positive and again approach the rate of
interest over time.
There is another reason besides technological progress for the cost of extraction to vary
over time. It can be interpreted as having to do with the second component of the rate
of return on physical assets, which I have so far neglected. As the resource stock gets
depleted, we can expect the marginal cost of extraction to increase, due to the fact that the
resource tends to be less easily accessible and of lesser grade. A marginal addition to current
resource extraction not only uses up the resource stocks, but it uses up the cheapest available
and hence increases all future cost. This Ricardian effect has been called a “degradation
cost” (Solow and Wan 1976) and can be assimilated to physical depreciation that can be
prevented by not depleting the asset. To capture it, write the marginal cost of extraction as
c(t, X(t)) = e−αt f (X(t)), f 0 (X(t)) < 0.7 Hotelling’s rule must then be modified to reflect
where x(t) represents the rate at which the stock X(t) is being depleted. Thus leaving the
marginal unit of resource in situ provides an additional return, in the form of a lower cost
of extraction in the future. Hotelling’s rule now dictates that the net equilibrium price grow
at a rate which is less than the rate of interest. But since this stock effect directly enters
the net price through its effect on cost, the expression for the rate of growth of the flow
7
For careful analyses of the effect on the predicted time path of the asset price (or scarcity rent) of other
more general hypotheses concerning the extraction cost function, see Farzin (1992, 1995) and more recently
Livernois and Martin (2001).
7
price remains unchanged, except for the fact that the level of marginal cost now depends on
the remaining stock. Flow price behavior remains qualitatively the same. The prediction is
still that the rate of growth of the market price is the weighted sum of the rate of interest
and the rate of decrease of extraction cost due to technological progress. Unless extraction
becomes unprofitable before the stocks are exhausted, we should still observe that price will
eventually be growing at a rate that approaches the rate of interest as the share of cost in
price becomes smaller and smaller.
As just sketched, the depletion effect could be interpreted as being due strictly to physical
reasons, as if lower-cost resources had to be exploited first in order to gain access to the
higher-cost ones. This is often the case in any one resource pool. But at an aggregate level
it is realistic to assume that many resource pools that differ in their cost of extraction are
available for exploitation at any one time. Then it is economic efficiency and not nature that
dictates that, of any two resource deposits, the low-cost one be depleted before moving on
to the next highest-cost one (Herfindahl 1967, Solow and Wan 1976).8 As long as the stocks
in both resource pools are positive, asset markets equilibrium requires that Hotelling’s rule
hold for each. Hence the asset price of each must be growing at the rate of interest but with
the asset price of the high-cost deposit lower than that of the low-cost deposit, since the
flow price they can get on the market is the same for both deposits (I am assuming identical
products). The owner of the high-cost deposit will find it unprofitable to extract and will
hold on to his asset as long as the low-cost deposit is not depleted. When the switch to the
higher cost deposit occurs, there will be a jump up in the cost of extraction. Hence the asset
price will have grown on average at less than the rate of interest in the long-run (Dasguspta
and Heal 1979, p. 172–175). But the flow price itself will be continuously growing.
8
The problem of the efficient order of use of many resource pools has been generalized to resource pools
that may exhibit nonconvex cost of extraction over some range (Weitzman 1976), to resource pools that are
differentiated by their set-up cost as well as their marginal cost (Hartwick, Kemp and Long 1986) and to a
general equilibrium context (Kemp and Long 1980, Amigues, Favard, Gaudet and Moreaux 1998). It has also
been generalized to resource pools that are differentiated by their geographical location with respect to many
spatially distributed markets (Gaudet, Moreaux and Salant 2001) and by characteristics which distinguish
them in their efficacy in many different end uses (Chakravorty and Krulce 1994, Chakravorty, Roumasset
and Tse 1997).
8
4 Durability
When we think of nonrenewable resources, we often think of fossil fuels, such as oil, natural
gas and coal. But an often neglected fact is that many nonrenewable resources — in par-
ticular metals — are durable goods. In fact of the ten nonrenewable resources whose price
trend we just examined, seven can be considered durable resources, in the sense that they
(at least partly) a stock demand. Once above ground, this stock of durable good becomes a
conventional physical asset and the asset markets equilibrium requires that its rate of return
be equal to the rate of interest. This rate of return will include a dividend component (the
gains component, which corresponds to the rate of appreciation of the market price of the
good. Hotelling’s rule must still hold for the asset held below ground, which means that this
market price net of the cost of extraction must also be growing at the rate of interest. But it
does not follow anymore from Hotelling’s rule that market price will necessarily be growing.
In fact we can easily imagine a scenario where price would be always falling, even though
the net price (the price of the asset in the ground) is always rising at the rate of inter-
est (Levhari and Pindyck 1981). For this, simply assume that demand is stationary and
that the resource is perfectly durable, so that there is no physical depreciation. Then the
long-run desired stock of the durable above ground is a constant. If marginal extraction cost
is an increasing function of the rate of extraction, then this desired stock will be approached
smoothly, at a decreasing rate. If the desired stock exceeds the available reserves, then the
resource is scarce and its net price must follow Hotelling’s rules. But since demand is a
decreasing function of the stock already in circulation, market price must be decreasing.
Of course in the real world demand is not stationary and the rate of physical depreciation
is not zero and it may itself depend on price (Henderson, Salant, Irons and Thomas 2007).
9
But even if we relax those assumptions, the price path can be expected to exhibit a decreasing
phase before starting to increase. The rate of decrease and the length of the decreasing
phase will depend on the rate of depreciation and the rate of growth of demand (Levhari
and Pindyck 1981). Clearly, durability can be an important factor in reconciling Hotelling’s
5 Market structure
Another factor which can influence the rate of return on resource stocks is the presence of
imperfect competition.
If the resource stock is in the hands of a monopolist, the marginal value to the owner of
the stock of resource left in the ground will be equal to the marginal profit it can bring on
the flow market once extracted. To a monopolist, this is less than the net price. The asset
markets equilibrium condition will still require that the rate of return on the resource stock
be equal to the rate of interest. Only now the rate of appreciation of the in situ value is not
measured by the rate of change of the net price, but by the rate of change of the monopolist’s
marginal profit. If we neglect changes in the marginal cost over time, the prediction is then
There are two effects from the introduction of a monopolist. The first comes from the
fact that even if the proportion of marginal revenue to price does not vary over time (η does
not change), price will be growing at a slower rate than in a competitive market simply
because marginal revenue is smaller than price and hence the weight given to the rate of
interest is smaller. Thus even if the elasticity of demand remains unchanged over time, as
would occur if demand were isoelastic and stationary, the price path will appear flatter than
in a perfectly competitive world unless the marginal cost of extraction is negligible. The
second effect comes from the fact that the gap between price and marginal revenue can be
10
expected to change over time. If this gap is decreasing over time (η is growing), this will tend
to further flatten the price path. The difference between price and marginal revenue being
inversely related to the elasticity of demand, this will happen if the elasticity of demand
is increasing over time. This is what is to be expected as substitutes for the resource are
under monopoly and perfect competition at the date of exhaustion of the resource stock,
this means that the monopoly price must initially be higher and eventually become lower
than the competitive price. This reasoning assumes that the available stock of resource is
the same under monopolistic and competitive industry structures.9 If costly efforts must be
expended to determine the initial resource stock, then the monopolist may possibly further
exert his monopoly power by choosing a lower initial stock than would price-takers (Gaudet
and Lasserre 1988). This however does not change the fact that monopoly power will tend
himself remarked, a more realistic market structure for many nonrenewable resources is some
form of oligopolistic competition.11 The flattening effect on the price path just described
will of course still be felt, since it is due strictly to the presence of some market power. An
oligopolistic world is however much more complex than either perfect competition or pure
monopoly. It involves strategic interactions in an inherently dynamic context, since the re-
9
The fact that the monopolist’s price is flatter than the perfectly competitive price path means that the
monopolist will take a longer time than the competitive market to exhaust the same initial resource stock,
hence the saying that “the monopolist is the conservationist’s best friend”.
10
The well known special case of a stationary isoelastic demand, under which the monopolistic and com-
petitive price paths coincide (Stiglitz 1976), rests not only on an assumption of zero marginal cost, but also
on an assumption that the initial reserves are at the same exogenously given level in both situations (Gaudet
and Lasserre 1988).
11
Hotelling analyzes in some details the cases of perfect competition and monopoly. He provides no formal
analyses of oligopolistic market structures, other than a very sketchy discussion of duopoly. He does however
recognize that a market composed of “a few competing sellers” is “more closely related than either [monopoly
or perfect competition] to the real economic world” (Hotelling 1931, p. 171).
11
source stock evolves over time. The strategic considerations involved may themselves further
modify the way the rate of return on holding the stock of resource is to be characterized, as
we will see.
The pioneering works on oligopolistic markets for nonrenewable resources were motivated
by the post-1973 oil market and the presence of OPEC as a dominant player. Thus the first
approach to the problem (Salant 1976) treated the market as composed of a dominant cartel
and a competitive fringe and adopted the Nash-Cournot equilibrium concept, whereby each
player ignores the effects of his decisions on the strategies of his rivals. Under this approach,
in solving for an equilibrium it is assumed that the competitive fringe takes as given the
price path of the dominant player and chooses a production path, while the dominant firm
takes as given the production path of the competitive fringe and chooses a price path. This
type of market structure provided the framework for a number of subsequent analyses (for
instance Pindyck (1978), Salant (1982), Lewis and Schmalensee (1979) and Ulph and Folie
(1980).
A number of other studies have retained the cartel-versus-fringe assumption, but have
treated the cartel as a Stackelberg leader (Gilbert 1978, Newbery 1981, Ulph 1982, Groot,
Withagen and de Zeeuw 1992). Under this approach the cartel is aware of the effect of its
decisions on the decisions of the fringe and takes it into account. The fringe, as a follower,
chooses a production path in reaction to the output path of the cartel. The cartel, as the
leader, determines its own output path taking as given the fringe’s anticipated reaction.
Thus, unlike in a Nash-Cournot world, the cartel is in a position where it can manipulate
the fringe’s reaction.
Those models have provided extremely important insights into the functioning of imper-
fectly competitive markets under the dynamic conditions that follow from the non renewa-
bility of resource stock and they have been very useful tools of analysis. But they also raise
some questions as to the proper way to model nonrenewable resource markets. In particular,
one drawback from which they suffer is that the strategies are limited to being a function
12
only of time and the initial stocks of the resource — so-called open-loop strategies. This
implicitly assumes that the players can credibly commit at the outset to whole price or
output paths. In the absence of the capacity to credibly commit in this way, the equilibria
they generate are generally not robust to deviations of the vector of resource stocks from
its equilibrium level. To use the game theory jargon, the equilibria obtained are generally
not subgame perfect. In the case of the leader-follower approach, the derived equilibria may
not even be time-consistent, in the sense that the cartel may, in some cases, wish to deviate
from the initially chosen path at some future time if it is allowed to reconsider its plan
(Newbery 1981, Ulph 1982, Maskin and Newbery 1990, Karp and Newbery 1992, Karp and
Newbery 1993).12
An alternative is to assume that the strategies consist of decision rules that express a
producer’s rate of extraction as a function of the current stocks of the resource and possibly of
time.13 The resulting equilibria are of the feedback (or closed-loop) type and will be Markov
perfect: they will constitute a Nash equilibrium not only for the resource stocks along the
solution path, but also for all out of equilibrium resource stocks. Solving for Markov perfect
Nash equilibria in this context raises very difficult theoretical and computational issues and
much remains to be done. Only very recently (Groot, Withagen and de Zeeuw 2003) has a
solution to the feedback Stackelberg equilibrium been derived, for the case of linear demand
the oil industry is an exception rather than the rule in this respect.14 Although there have
12
An equilibrium may be time consistent but not subgame perfect: there may not be any incentives to
deviate from the initially announced plan if allowed to reconsider in the future, but still the decision may
not remain optimal if the vector of resource stocks deviates from its equilibrium value. Of course, if the
equilibrium is subgame perfect, then it is also time-consistent.
13
This type of strategy is called Markovian because it depends only on the current resource stocks and
not on the whole history of the stocks.
14
The diamond industry may be another such exception, the De Beers cartel having exercised strong
control over supply for many years. But the structure of that industry is also changing, since it appears
13
historically been numerous attempts to cartelize other natural resource industries, none have
been very successful at controlling the market (Eckbo 1976, Teece, Sunding and Mosakowski
1993, Suslow 2005, Benchekroun, Gaudet and Long 2006) and, when able to attain some
success in pushing up price, have been short-lived.15 Certainly, many resource markets can
extraction game as an open-loop dynamic game, can be viewed as a direct replication of the
Cournot framework in a dynamic setting (Lewis and Schmalensee 1980, Loury 1986, Polasky
1992). Indeed, in terms of information structure, open-loop dynamic games are formally
identical to static games, since in the open-loop game the player is allowed to condition his
decision only on the initial state and time. From a purely technical point of view, the only
difference lies in the fact that in a static game the decision space is the space of real numbers,
whereas in the open-loop dynamic game the decision space is the infinite-dimensional space
The fact that in a dynamic context the state evolves over time as a function of the
actions taken opens the door to more intricate analysis. Some issues which appear trivial
in a truly static context become very relevant in a dynamic context, even if formulated as
exogenous perturbation — for instance a tax which is imposed for a finite period of time
and then lifted. In a truly static world, once this tax is lifted, the production level can be
expected to revert to the previous static equilibrium level, since the state is unchanged by
the temporary effects of the tax on production. In a dynamic context however, such as a
natural resource industry, the temporary tax will leave the industry in a different state then
it would have been without the tax, since current production decisions have an effect on
that De Beers is having to loosen its grip over supply in the face of emerging competition and government
regulations (The Economist, 24 February 2007, p. 68).
15
Other than OPEC, the resource cartels studied by Pindyck (1977, 1978) (copper and bauxite) have since
completely ceased their activities.
14
the state, as measured by the resource stocks. If the producers are able to anticipate this
temporary perturbation and so are able to condition their open-loop paths on it, then we
can expect the open-loop equilibrium production path to be perturbed not only while the
tax is in effect, but also before the tax is applied and after it is lifted (Benchekroun and
Gaudet 2003).
This example highlights the most objectionable feature of modeling natural resource
oligopolistic markets as open-loop equilibria. In the open-loop approach to a nonrenewable
resource exploitation game, each producer is assumed to choose at the outset the whole time
path of extraction of its own resource stock, conditional only on its initial reserves and those
of its competitors. It is therefore assumed that the producers do not use information on
the evolution of the level of reserves which they observe after initially making their decision
on the production path. Thus if the above tax scheme was unanticipated at the time of
deciding on the production path, it is unlikely that the planned open-loop production would
still constitute a Nash-equilibrium once the reserves have adjusted to this exogenous shock.
Open-loop strategies are probably not a good description of how oligopolistic natural
resource producers make their decision. As in the sequential decision cartel-fringe model
just discussed, it seems more reasonable to assume that producers condition their production
decisions on the current state of reserves rather than simply on the initial reserves and time,
and thus use Markovian decision rules. This of course assumes that the producers observe
the evolution of reserves, which is a realistic assumption given that much information on
current stocks is readily available for most resources. We are then looking for a Markov
perfect oligopolistic equilibrium, with each producer choosing its decision rule as a function
15
condition their production decisions on the remaining reserves, then Hotelling’s rule needs
to be modified. Indeed, if a producer knows that his own resource holdings affect a rival’s
production decision and hence the rival’s reserves, which in turn enter into his own decision
rule, then there will be a strategic component to the return on holding reserves. This strategic
effect will depend, at any given time, on each producer’s remaining reserves and his rivals’
reserves (see Benchekroun and Gaudet (2003), for an example).
~
Let us write this strategic component s(X(t), ~
π(t)), where X(t) is the vector of stocks
remaining at time t. For the sake of the argument, ignore the stock effects on costs and
assume a symmetric equilibrium. We might then rewrite Hotelling’s rule as:
π̇(t) ~
+ s(X(t), π(t)) = r. (6)
π(t)
Unfortunately, so little is known about the properties of the decision rules in this type of
equilibrium that it is not possible to put a definite sign on this component of the rate of
return. It is therefore not possible to predict the effect it should have on the price path. It
certainly cannot be ruled out that it is a factor in explaining the flatness of the observed
price paths. But much theoretical work needs to be done before its empirical relevance can
be ascertained.
6 Uncertainty
I have so far assumed a world of perfect certainty in which wealth can be held either in
the form of known natural resource stocks or in the form of a bond, whose rate of return is
exogenously given. This obviously lacks realism if the objective is to explain long-run price
behavior. Resource stocks are not known with certainty and both extraction and exploration
16
If as extraction proceeds one gains information about the true size of the reserves, then the
net price of the resource will reflect the informational value of extraction. As a consequence,
it will not be rising at the rate of interest (Kemp 1976, Loury 1978, Gilbert 1979, Kemp and
Long 2007). Extraction will proceed faster than otherwise and this will impact on price.
Exploration is also an obvious way of gaining information about the true level of reserves.
With uncertain stocks it becomes profitable to carry out exploration activities in order
to reduce this uncertainty. Any resulting revision of the estimated reserves will change
the basis for the calculation of the asset price (Deshmukh and Pliska 1980, Arrow and
Chang 1982, Lasserre 1984). If the estimated reserves are repeatedly revised upwards as a
result of the exploration activity, then the predicted rise at the rate of interest of the net
price of the resource will be offset by repeated downward adjustments in the net price that
serves as a basis for this rise. Hotelling’s rule applies between such revisions of reserves,
as usual. But if over a long period of time successful exploration activities dominate the
unsuccessful ones, then the asset price may appear to be growing on average at less than the
rate of interest. This will of course reflect itself in a similar way in the time path of the flow
price, whose long-run increase will appear dampened by the successive drops in the asset
price.
Insecure property rights have also at times constituted another source of uncertainty
which has had an impact on the patterns of extraction and of price. Faced with a threat of
expropriation, owners of resource stocks are likely to act as if their discount rate were higher
than in the absence of that threat (Long 1975). Leaving the resource in situ becomes a risky
investment, since the insecure owner has no guarantee that he will be able to fully reap the
fruit of his investment. As a consequence, he will tend to shift production from the future
towards the present.17 It has even been argued that the quadrupling of oil prices between
1973 and 1977 should be attributed not so much to the cartelization of the market, but to
17
The “overexploitation” that occurs in such a situation is much like the overexploitation that occurs when
a resource is exploited in a common property regime, as is the case, for instance, when many oil producers
pump from the same oil pool (See Libecap and Wiggins (1984), Libecap (1989) and Gaudet, Moreaux and
Salant (2002))
17
the resolution of the property rights issue, which definitely shifted from the companies to
the host countries by 1973 (Johany 1979, Mead 1979).18 According to this argument, as the
possibility of nationalization became more and more imminent, the oil companies that held
concessions in the Middle East reacted by raising their rate of discount. Once the property
rights were secured in their favor, leaving the oil in the ground became a more attractive
investment for some countries than it would have been for the companies facing the threat
of expropriation.
More generally, exploitation of the stocks of natural resources and the decisions to invest
in such stocks takes place in a risky environment. To illustrate the possible effect of this
on the long-run resource price path, consider a stylized economy in which there are two
goods and three assets. One of the goods is a nonrenewable resource, which serves as an
input into the production of the second good, a composite commodity which can be either
consumed or accumulated. The accumulated stock of the composite good can be held either
in the form of physical capital or in the form of a bond, which yields the exogenously given
risk free rate of return, as above. The physical capital can be either devoted to extracting
the natural resource or combined with the flow of extracted natural resource to produce the
composite good. Both the production process and the extraction process, and hence the cost
of producing the composite good and the cost of extracting the resource, are stochastic.19
Total wealth in this economy is now held in the form of two risky assets (the nonrenewable
resource and the physical capital) and one risk-free asset (the bond). In addition to choosing
a level of consumption, the representative risk-averse consumer decides on the allocation
of his portfolio among those three assets. Hotelling’s rule will now take the form of the
familiar intertemporal equilibrium asset-pricing rule of portfolio theory, adapted to take into
account that the holdings of the nonrenewable resource stock cannot be increased (Gaudet
18
See also on this Salant (1982), pages 9 and 14. A closely related type of uncertainty is where agents face
the threat of a price collapse due to massive sale of government held resource stocks (such as gold) at some
unknown time in the future (Salant and Henderson 1978).
19
To be more precise, we might assume, as in Gaudet and Khadr (1991), that each of the production
processes depends on a parameter that represents the state of productivity in the sector and that the change
in those productivity parameters follows an Itô process.
18
and Howitt 1989, Gaudet and Khadr 1991) and yield no dividend.
This means that the equilibrium expected rate of change of the in situ value of the
resource will generally differ from the rate of interest, so as to reflect the relative riskiness of
the natural resource as an asset. Suppose unfavorable productivity outcomes in the resource
extraction sector are associated with unfavorable productivity outcomes in the production
of the composite commodity and similarly for favorable outcomes. Then on average the
two outcomes tend to reinforce each other, thus increasing the uncertainty in the economy’s
future capacity to produce the composite good, as perceived from today’s standpoint. To
the risk-averse representative consumer, this is an undesirable feature of holding the resource
On the other hand, suppose that unfavorable productivity changes in the resource sec-
tor are typically accompanied by favorable productivity changes in the production of the
composite commodity, and vice-versa. It is then expected that when the resource is least
easily accessible, it is also the most productive. The two outcomes therefore tend to offset
each other and thus to mitigate the uncertainty in the economy’s capability to produce the
composite commodity. The resource stock is then a relatively less risky asset and thus does
not need to exhibit a large expected appreciation in value relative to the rate of return on
For the purpose at hand, a convenient way of expressing the equilibrium asset pric-
ing condition for the resource is in its so-called consumption-beta form (Breeden 1979,
Ingersoll 1987). In this form, Hotelling’s rule modified to account for uncertain returns
becomes (Gaudet and Khadr 1991):
µπ − A(y)σπy = r (7)
where µπ is the expected rate of appreciation of the in situ price of the resource, y is
consumption, which can be interpreted in this representative consumer framework as per
capita consumption, A(y) is the Arrow-Pratt measure of relative risk aversion and σπy is the
instantaneous covariance between the rate of change of the in situ resource price and the
19
rate of change of consumption.
In a world with many risky assets, we can write this equilibrium condition in a different
form. If µM is the instantaneous rate of return on any portfolio that has a nonzero covariance
with changes in consumption, then we also have µM = r + A(y)σM y . Hence, by substitution
µπ − β[µM − r] = r, (8)
It is now apparent from (7) that the expected rate of appreciation of the in situ price
will differ from the rate of interest by a risk premium that takes the sign of the covariance
between the rate of change in the in situ price and the rate of change of consumption. A
positive covariance means that the return on the resource stock would tend to accrue in
states of the world where marginal utility of consumption is low. This makes the resource
a relatively risky asset and calls for a positive risk premium. But this covariance may well
be negative, in which case holding the resource stock constitutes a form of insurance against
adverse changes in consumption. The risk premium associated with holding the resource
stock is then negative and the expected rate of increase in the in situ price will be less than
the rate of interest. In fact, if the risk premium is sufficiently negative, we may even expect
where µp and µc denote respectively the expected rates of change of the flow price and the
expected rate of change of extraction cost. If the covariance between the rate of change in
the in situ price and the rate of change of consumption is sufficiently negative, both terms on
the right-hand side may be negative. Thus negative risk premia on the holding of resource
stocks can reconcile Hotelling’s rule with observed price paths that are relatively flat and
20
even decreasing. A direct test of this conjecture would require data on asset prices (π), as
opposed to flow prices (p). Unfortunately, only flow prices are observed.
The rare cases where tests of Hotelling’s rule have been carried out taking into account the
possibility of risk diversification (Young and Ryan 1996, Slade and Thille 1997, Roberts 2001,
Livernois, Thille and Zhang 2006) tend to support the view that this is an important element
in the determination of the return of natural resource stocks.20 Its role in explaining the
observed behavior of resource price paths certainly deserves serious empirical investigation.
7 Concluding remarks
In conclusion, I want to emphasize that the foregoing discussion should not be viewed as
an exhaustive survey of the field of natural resource economics, nor of the relevance of
Hotelling’s rule.21 I chose to concentrate on those theoretical developments that seem more
likely to reconcile the asset pricing rule that is Hotelling’s rule with the observed behavior
of market prices over time. In doing so, I have ignored important issues and topics for
which Hotelling’s rule is of particular relevance. Let me just mention resources and growth,
also ignored the large literature on renewable resources, which, contrary to nonrenewable
resources, renewable resources are assets to their owners and are subject to the same asset
markets equilibrium conditions, with the important difference that stocks of renewable re-
sources, such as fisheries or forestries, yield a dividend in situ in the form of natural growth.
Some renewable resources are also more prone to problems raised by difficulties in applying
20
Livernois et al. (2006) is the only study that actually uses direct observations of the asset price. Their
data comes from the sales of logging rights for old-growth timber in the U.S. Pacific Northwest. This resource
can be considered nonrenewable, since the old-growth timber in that region is several hundred years old.
Young and Ryan (1996) use aggregate average cost for four Canadian mining industries, while Slade and
Thille (1997) use cost estimates for 14 Canadian mining firms. Roberts (2001) uses the flow price to represent
the asset price, which amounts to assuming that cost are either zero or a constant proportion of the flow
price, so that the rates of change of the two prices are the same.
21
For a more exhaustive survey of the literature on nonrenewable resources see (Krautkraemer 1998).
21
private property rights, which leads to free access or common property exploitation, another
with respect to the use of energy resources is a perfect example of this. From a social point
of view, the net benefits will take into account the damage costs that result from the use
of those energy resources, in addition to the cost of extracting them. Hotelling’s rule then
will be flatter than in the absence of the externality, or even decreasing, in order to push the
lasting damage cost towards the future.22 To the extent that such environmental policies
were to be implemented, one should observe the same flattening effects on the actual market
price paths.
The impact of environmental policies on market equilibrium resource price paths is ob-
viously an interesting direction for further research. So is the application of Hotelling’s rule
to the more general analysis of what some perceive as the “new scarcity”(Simpson, Toman
and Ayres 2005), namely the limited capacity of the environment as a waste sink.
22
For an interesting first effort at modeling this problem, see Bartelme, Lim and Yee (2007).
22
Data sources
Copper, lead, and zinc: Manthy (1978) and U.S. Geological Survey, Mineral Commodity
Summaries
Tin: Potter and Christy (1962) and U.S. Geological Survey, Mineral Commodity Summaries
Aluminum, nickel and silver: U.S. Geological Survey, Mineral Commodity Summaries
Coal, natural gas: Manthy (1978) and U.S. Energy Information Administration, Annual
23
References
Amigues, Jean-Pierre, Pascal Favard, Gérard Gaudet, and Michel Moreaux (1998) ‘On the
optimal order of natural resource use when the capacity of the inexhaustible substitute is
limited.’ Journal of Economic Theory 80(1), 153–170
Arrow, Kenneth J. (1987) ‘Harold Hotelling.’ In The New Palgrave: A Dictionary of
Economics, ed. John Eatwell, Murray Milgate, and Peter Newman, vol. 2 (London:
MacMillan Press) pp. 670–671
Arrow, Kenneth J., and Sheldon Chang (1982) ‘Optimal pricing, use, and exploration of
Management 9, 1–10
Barnett, Harold J., and Chandler Morse (1963) Scarcity and Growth: The Economics of
Natural Resource Availability (Baltimore, MD: Johns Hopkins University Press for
Bartelme, Dominick, Lili Lim, and Chia Chye Yee (2007) ‘Optimal extraction,
Michigan (mimeo)
Benchekroun, Hassan, and Gérard Gaudet (2003) ‘On the profitability of production
Benchekroun, Hassan, Gérard Gaudet, and Ngo Van Long (2006) ‘Temporary natural
resource cartels.’ Journal of Environmental Economics and Management 52, 663–674
Berck, Peter, and Michael Roberts (1996) ‘Natural resource prices: Will they ever turn
Chakravorty, Ujjayant, and Darrell L. Krulce (1994) ‘Heterogeneous demand and order of
resouce extraction.’ Econometrica 62(6), 1445–1452
24
Chakravorty, Ujjayant, J. A. Roumasset, and K.-P. Tse (1997) ‘Endogenous substitution
Dasguspta, Partha S., and Geoffrey M. Heal (1979) Economic Theory and Exhaustible
Resources (Welwyn, England: James Nisbet and Co. and Cambridge University Press)
Deshmukh, Sudhakar, and Stanley R. Pliska (1980) ‘Optimal consumption and exploration
Company)
Farzin, Y.H. (1992) ‘The time path of scarcity rent in the theory of exhaustible resources.’
(1995) ‘Technological change and the dynamics of resource scarcity measures.’ Journal of
Gaudet, Gérard, and Ali M. Khadr (1991) ‘The evolution of natural resource prices under
Gaudet, Gérard, and Peter Howitt (1989) ‘A note on uncertainty and the Hotelling rule.’
Gaudet, Gérard, and Pierre Lasserre (1988) ‘On comparing monopoly and competition in
exhaustible resource exploitation.’ Journal of Environmental Economics and
Management 15, 412–418
Gaudet, Gérard, Michel Moreaux, and Stephen W. Salant (2001) ‘Intertemporal depletion
of resource sites by spatially distributed users.’ American Economic Review
91(4), 1149–1159
25
Gilbert, Richard J. (1978) ‘Dominant firm pricing policy in a market for an exhaustible
Groot, Fons, Cees Withagen, and Aart de Zeeuw (1992) ‘Note on the open-loop von
102(415), 1478–1484
Hartwick, John M., Murray C. Kemp, and Ngo Van Long (1986) ‘Set-up costs and theory
13, 212–224
Henderson, Dale W., Stephen W. Salant, John S. Irons, and Sebastian Thomas (2007) ‘The
Taxation, ed. Mason Gaffney (Madison WI: University of Wisconsin Press) pp. 68–90
Hotelling, Harold (1925) ‘A general mathematical theory of depreciation.’ Journal of the
American Statistical Association 20(151), 340–353
39(2), 137–175
26
Ingersoll, J.E. (1987) Theory of Financial Decision-Making (New York: Rowman and
Littlefield)
Johany, Ali D. (1979) ‘OPEC and the price of oil: Cartelization or alteration of property
rights.’ Journal of Energy and Development 5(1), 72–80
Karp, Larry, and David M. Newbery (1992) ‘Dynamically consistent oil import tariffs.’
Canadian Journal of Economics 25(1), 1–21
(1993) ‘Intertemporal consistency issues in depletable resources.’ In ‘Handbook of
Kemp, Murray C., and Ngo Van Long (1980) ‘On two folk theorems concerning the
(2007) ‘Extracting several resource deposits of unknown size: Optimal order.’ Working
Lasserre, Pierre (1984) ‘Reserve and land prices with exploration under uncertainty.’
Levhari, David, and Robert S. Pindyck (1981) ‘The pricing of durable exhaustible
University Press)
27
Libecap, Gary D., and Steven N. Wiggins (1984) ‘Contractual responses to the common
pool: Prorationing of crude oil production.’ American Economic Review 74(1), 87–98
Livernois, John, and Patrick Martin (2001) ‘Price, scarcity rent, and a modified r percent
rule for non-renewable resources.’ Canadian Journal of Economics 34(3), 827–845
Livernois, John, Henry Thille, and Xianqiang Zhang (2006) ‘A test of the Hotelling rule
using old-growh timber data.’ Canadian Journal of Economics 39(1), 163–186
Long, Ngo Van (1975) ‘Resource extraction under the uncertainty about possible
(Baltimore MD: Johns Hopkins Press for Resources for the Future)
Maskin, Eric, and David Newbery (1990) ‘Disadvantageous oil tariffs and dynamic
Mead, Walter J. (1979) ‘An economic analysis of crude oil : Price behavior in the 1970s.’
Journal of Energy and Development 4(2), 212–228
Meadows, Denis L. et al. (1972) The Limits to Growth: A Report for the Club of Rome’s
Project on the Predicament of Mankind (New York: Universe Books)
Newbery, David M. G. (1981) ‘Oil prices, cartels, and the problem of dynamic
Pindyck, Robert S. (1977) ‘Cartel pricing and the structure of the world bauxite market.’
28
Bell Journal of Economics 8(2), 343–360
United-States, 1870-1957 (Baltimore MD: John Hopkins Press for Resources for the
Future)
Roberts, Michael J. (2001) ‘Can portfolio theory explain flat and downward natural
resource price trends?’ Paper presented at the Second World Congress of Environmental
84(5), 1079–1093
(1982) Imperfect Competition in the World Oil Market : A Computerized Nash-Cournot
Salant, Stephen W., and Dale W. Henderson (1978) ‘Market anticipations of government
policies and the price of gold.’ Journal of Political Economy 86(4), 627–648
Salo, Seppo, and Olli Tahvonen (2001) ‘Oligopoly equilibria in nonrenewable resource
markets.’ Journal of Economic Dynamics and Control 25, 671–702
Scott, Anthony (1955a) ‘The fishery: The objectives of sole ownership.’ Journal of Political
Simpson, David R., Michael A. Toman, and Robert U. Ayres (2005) Scarcity and Growth
Revisited: Natural Resources and the Environment in the New Millennium (Washington,
29
DC: Resources for the Future)
Solow, Robert M., and Frederic Y. Wan (1976) ‘Extraction costs in the theory of
Stiglitz, Joseph E. (1976) ‘Monopoly and the rate of extraction of exhaustible resources.’
Suslow, Valerie Y. (2005) ‘Cartel contract duration: Empirical evidence from inter-war
Teece, David, David Sunding, and Elaine Mosakowski (1993) ‘Natural resouce cartels.’ In
Handbook of Natural Resource and Energy Economics, vol. III, ed. A.V. Knesse and J.L.
Ulph, A.M. (1982) ‘Modeling partially cartelized markets for exhaustible resources.’ In
30
Economic Theory 12, 351–364
Young, Denise, and David L. Ryan (1996) ‘Empirical testing of a risk-adjusted Hotelling
model.’ Resource and Energy Economics 18, 265–289
31
0,8
Copper
0,6
0,4
Rate of price change
0,2
0,0
-0,2
-0,4
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
0,8
Lead
0,6
0,4
Rate of price change
0,2
0,0
-0,2
-0,4
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
32
4,0
Zinc
3,0
2,0
Rate of price change
1,0
0,0
-1,0
-2,0
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
1,0
Aluminum
0,8
0,6
Rate of price change
0,4
0,2
0,0
-0,2
-0,4
-0,6
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
33
1,0
Tin
0,8
0,6
Rate of price change
0,4
0,2
0,0
-0,2
-0,4
-0,6
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
2,0
Nickel
1,5
Rate of price change
1,0
0,5
0,0
-0,5
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
34
1,0
Silver
0,8
0,6
0,4
Rate of price change
0,2
0,0
-0,2
-0,4
-0,6
-0,8
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
0,8
Natural Gas
0,6
0,4
Rate of price change
0,2
0,0
-0,2
-0,4
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
35
0,8
Coal
0,6
Rate of price change
0,4
0,2
0,0
-0,2
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
1,0
0,8
Petroleum
0,6
Rate of price change
0,4
0,2
0,0
-0,2
-0,4
-0,6
1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000
Year
36