adjustment cost, uncertainty, and the proved reserves of crude oil · adjustment cost, uncertainty,...
TRANSCRIPT
Adjustment Cost, Uncertainty, and the Proved
Reserves of Crude Oil
John Boyce∗ Xiaoli Zheng †
Abstract
This paper discusses an important issue for the oil industry:
why do firms hold large amounts of proved reserves relative to
crude oil productions? To answer this questions, we first clarify
the definition of proved reserves and then build a simple deter-
ministic and a stochastic model. In the deterministic model, we
find a delayed response is necessary for the firm to hold proved
reserves while adjustment costs govern the periods of delay. In
the stochastic model, this result still holds. We find uncer-
tainties from demand and exploration lower the firm’s optimal
productions and thus resources would be exhausted within a
longer period. The stochastic model predicts that firms would
hold larger proved reserves than productions either if the de-
mand volatility is much smaller than the interest rate or if the
exploration volatility is not noticeably larger than the inter-
est rate. By using U.S. data, we verify that this prediction is
empirically relevant. The existence of these volatilities also ex-
plains why the ratio of proved reserves to productions declines
slowly or even remains stable.
keywords: Adjustment cost; delayed response; uncertainty;
proved reserves.
JEL classications: Q410; G170; D210.
∗Department of Economics, University of Calgary, 2500 University Dr. NWCalgary, AB, T2N 1N4 Canada. Email: [email protected]†Department of Economics, University of Calgary, 2500 University Dr. NW
Calgary, AB, T2N 1N4 Canada. Corresponding author email: [email protected]
1 Introduction
U.S. and Canada’s proved reserves of crude oil share the common
trend with crude oil production for the past century. The ratio of
proved reserves to production, which measures how many years at
current production before current proved reserves are exhausted, is
around 8-10 years in U.S and 8-15 years in Canada.1 Moreover,
these numbers have declined slowly since the 1940’s in U.S. and the
1970’s in Canada (Figure 1). They also prevail on firm-level data.
Figure 2 below is a histogram that illustrates the the the ratio of
proved reserves to productions for 20 listed U.S. and Canada’s firms
from 1999 to 2013. More than 55 percent of the observations are
centered in the interval of 9-16 years, implying these firms hold 9-
16 times more proved reserves than current production. This leads
to the research questions on this paper; What are the forces that
drive a firm’s decisions on proved reserve accumulation; Why do
they accumulate large amounts of proved reserves? This paper is
interested in examining how economics factors, such as crude oil
prices, price volatility and exploration costs, affect a firm’s optimal
proved reserve accumulation.
1This ratio is very large compared to manufacturing sectors, where the in-ventory ratio ranged from 1.21 to 1.39 from 2000 to 2010 according to the U.S.Census Bureau.
1
Figure 1: The Ratio of Crude Oil Reserve to Production in U.S. and Canada(Source: Energy Information Administration and Canadian Association of
Petroleum Producers)
Figure 2: The Ratio of Crude Oil Reserve to Production in U.S. and CanadianFirms (Source: COMPUSTAT)
In order to answer these questions, we properly define what crude
oil reserves are and how they can be measured. Starting from Hotelling’s
2
seminal paper [7] on exhaustible resource extraction, studies on ex-
haustible resource focus on the time path of production and prices.
Many theoretical studies (e.g., [24] and [25]) confirm Hotelling’s pre-
diction that production monotonically decreases and price increases
over time. On the other hand, some studies (e.g., [22] and [6]) find
that prices should follow a U-Shaped path. In all these models, nev-
ertheless, reserves are treated as either a known fixed stock which de-
pletes over time, or in models of exploration, as the outcome of sub-
tracting cumulative production from cumulative discoveries. How-
ever, proved reserves are more than geological definitions. Proved
reserves are defined by the Energy Information Administration and
Natural Resource Canada as “recoverable crude oil with reasonable
certainty under existing economic and technological conditions”[3].
Under this definition, reserves are “proved” so that they may become
crude oil production in the near future. As such, proved reserves are
surely affected by economic factors such as oil prices, price volatil-
ity and discovery costs.2 Therefore, it is even bizarre why firms
hold such a great amount of reserve in proved status. Early studies
(e.g., [21] and [5]) explain this fact by specifying production costs as
reserve-dependent; that is, a firm accumulates reserves just because
production costs are lower if the reserve is larger. Although this may
be true for a single oil field, it is hard to believe finding a new field
lowers the costs of other existing fields [11].
This paper offers a new answer using the concept of adjustment
cost and delayed response in a simple determinist model, in which
a representative firm has constant marginal costs of production and
discovery. Therefore, unlike previous studies, we do not rely upon
reserve-dependent costs. Due to the delay between resource discov-
ery and production, a firm makes decisions on how fast the resources
are transferred to the proved and then production stage when it finds
a new discovery from the oil-in-place stock. We assume there are ad-
justment costs associated with a fact that moving discoveries quickly
to the proved and then production stage is more costly than doing
2Some empirical studies (e.g., [4] and [15]) verify crude oil prices have signifi-cant impacts on oil explorations and reserve accumulation in some countries.
3
it slowly [19]. We show that adjustment cost and delayed response
in transit are necessary for the firm to hold a positive amount of
proved reserves. However, the deterministic model solely relies on
the periods of delay to explain the large amounts of proved reserves
relative to production. Moreover, it fails to account for why the
ratio of reserves to productions declines slowly or even remains sta-
ble. This leads us to consider a second model attempting to explain
how demand and exploration uncertainties affects the the crude oil
production and the accumulation of proved reserves. This model fol-
lows Pindyck’s [18] and Mason’s ([12], [13]) models that considers
stochastic models of exhaustible resource extractions. Although the
stochastic model follows Pindyck and Mason’s framework, it differs
in two aspects: the stochastic model is based on the specification of
deterministic model, thus the feature of delayed responses and ad-
justment costs is still kept; Additionally, the stochastic model explic-
itly solves the analytical solutions of expected productions, proved
reserves and the ratio of proved reserves to productions. The re-
sults of this stochastic model suggest that a firm would hold larger
proved reserves than current production if demand volatility is much
less than interest rate, or if the exploration volatility is not notice-
ably larger than the interest rate. The existence of these volatilities
also explains why the ratio of proved reserves to productions declines
slowly or even remains stable.
The paper is organized as follows: Section 2 presents the deter-
ministic model and section 3 describes the stochastic model, with
Section 4 offering the concluding remarks.
2 The Deterministic Model
We consider an environment in which a representative firm explores
for oil resources underground, and then make discoveries available
for production. There are two key features in the model. The first
is that there are three types of stocks: Su(t), which are total oil
in place; S1(t), which are resources that have been discovered and
proved, but are not yet ready to be produced; and S2(t) which are
4
proved reserves that ready to be produced. These three stocks rep-
resent three different stages of oil production, where Su denotes the
most undeveloped stage and S2 is the most developed stage. There
are three stocks because this is the minimum number necessary to
illustrate the importance of delayed responses and adjustment costs.
Adjustment costs enter the model because it costs a greater amount
to transfer resources directly from the most undeveloped stage to
the production stage than it does a two-step route where resource
are first turned from most undeveloped stage to the intermediate
stage, and then producible stocks. Thus, production only can occur
from stock S2.
The second feature of the model is that there exists a delay in
the time for resources transferred from previous stock to the current
stock. For simplicity, this delay is assumed to be identical, l units of
time periods, for all transfers.
In sum, to produce crude oil, the representative firm has two
routes by which it make discoveries for production as illustrated by
Figure 3.
Figure 3: Resource Discovery, Transfer, and Crude Oil Production
In one route, the firm could find and then transfer the resource
from the total oil-in-place stock, Su, directly to the stock ready for
production, S2. Alternatively, it could reach production stage by
two sequential steps: step one involves discovering and transferring
the resource from Su to an intermediate stock S1, and in the next
5
step transferring it to the S2. To sum up, the law of motion of crude
oil discoveries, transfers, and production can be summarized as the
differential equations as below,
dSudt
= −yu1(t)− yu2(t), (2.1)
dS1dt
= yu1(t− l)− y12(t), (2.2)
dS2dt
= yu2(t− l) + y12(t− l)− q(t). (2.3)
where yu1(t) is the quantity of reserves transferred from the total
oil-in-place stock Su(t) to intermediate stock S1(t+ l) at period t+ l;
y12(t) is the quantity transferred from stock S1(t) to S2(t + l) at
period t + l. So both yu1 and y12 denote the two-step transfers,
while yu2 is the quantity directly transferred from the stock Su(t) to
S2(t+ l), with l-period delay; so it denotes as one-step transfer.
Since the representative firm has two routes to reach production,
it faces different costs for each route. We assume the two-step trans-
fers, yu1 and y12, have a constant marginal cost, c1, while the one-
step transfer, yu2, has a constant marginal cost, c2. The marginal
and average cost of production from stock S2 is c0. The concept
of adjustment cost implies c2 > 2c1. Intuitively, this requires that
one-step route is much more costly than two-step route. This is be-
cause it involves more efforts to immediately reach production than
to transfer resources available for production step-by-step.3. Below
we derive the exact relationship on costs where the two-step route
dominates the one-step route, or vice versa
The deterministic model is an optimal control problem for a rep-
resentative and competitive firm as follows,
J = maxq,yu1,yu2,y12
∫ T
tπ(τ)dτ, (2.4)
where J is an objective function at present time. The profit func-
3This extra effort involves hiring more geologists, building drilling facilitiesand pipelines faster, extending average working hours, and compensating moredepreciation in drilling facilities.
6
tion π is defined as π = e−rtp(t)q(t)− c0q(t)− c1[yu1(t) + y12(t)]−c2yu2(t), where p(t) is the crude oil price at period t, and q(t) is the
production of crude oil at period t.
The behavioral assumption is that the firm’s maximizing equation
(2.4) subjected to equation (2.1)-(2.3), S1 > 0 and S2 > 0. The last
two constraints ensure that S1 and S2 are nonnegative over time.
The present-value Hamiltonian is:
H = e−rt[(p(t)q(t)− c0q(t)− c1[yu1(t) + y12(t)]− c2yu2(t)]
− λu(t)[yu1(t) + yu2(t)] + λ1(t)[yu1(t− l)− y12(t)]
+ λ2(t)[yu2(t− l) + y12(t− l)− q(t)] + φ1(t)S1(t) + φ2(t)S2(t).
(2.5)
The first-order conditions are4,
∂H
∂q(t)= e−rt (p(t)− c0)− λ2(t) = 0, (2.6)
∂H
∂yu1(t)= −e−rtc1 − λu(t) + λ1(t+ l) ≤ 0, (2.7)
∂H
∂y12(t)= −e−rtc1 − λ1(t) + λ2(t+ l) ≤ 0, (2.8)
∂H
∂yu2(t)= −e−rtc2 − λu(t) + λ2(t+ l) ≤ 0, (2.9)
λu(t) = 0, (2.10)
λ1(t) = −φ1(t), (2.11)
λ2(t) = −φ2(t). (2.12)
Note that by (2.11) and (2.12) the shadow value of S1 and S2
are weakly decreasing over time, because the Lagrangian multipliers
4Since there are lagged controls in the model, the technique of delayed responseis applied [9].
7
φ1(t) and φ2(t) are non-negative. (2.10) indicates the shadow value
of Su is constant and positive over time. Moreover, (2.6) holds as
equality because crude oil production occurs every period over the
history; while (2.7), (2.8) and (2.9) hold as inequality because either
one-step or two-step route is alternative to each other and thus may
or may not occur. These necessary conditions lead to a proposition
as shown below,
Proposition 1 Only two-step route occurs in equilibrium if
c2 > (1 + erl)c1, (2.13)
but only one-step route occurs if
c2 < (1 + erl)c1, (2.14)
However, one-step and two-step route cannot occur simultaneously .
Proof Proof of Proposition 1 is provided in Appendix A1.
Proposition 1 states the representative firm would follow two-step
route when the adjustment cost for one-step route c2 is sufficiently
large as given by (2.13). However, it does the one-step route when
the adjustment cost for one-step route c2 is less than the cost of two-
step route c1 discounted by time value erl. If l→ 0, (2.13) reduces to
our original concept of adjustment cost: c2 > 2c1. Therefore, even if
there is no delayed response, the firm is still prevented from doing the
one-step route given the adjustment cost is too high. On the other
hand, (2.14) is legitimate only for l > 0. Thus, delayed response
is necessary for the firm to switch to one-step route. In general,
a large lag of delay l causes the firm to choose one-step route and
reach production more quickly. This is because large lag of delay
raises the cost of two-step route, due to firm’s waiting more time for
production. Consequently, it lowers the opportunity cost of one-step
route.
Now let us consider a feasible solution of this model. First we
find that S1(t) = S2(t) = 0 along the equilibrium path 5. Next we
5The proof is provided in Appendix A2
8
consider a simple linear demand function as
p(t) = p− q, (2.15)
where p > 0 denotes as the cost of backstop technology.
Suppose only two-step route occurs. Lag (2.7) by l periods, and
substitute it into (2.8) to eliminate λ1(t). Then (2.8) becomes λ2(t+
l) = e−rtc1 + e−r(t−l)c1 + λu. Lagging this by l periods yields
λ2(t) = c1(1 + erl)e−r(t−l) + λu. (2.16)
Substitute (2.15) and (2.16) into (2.6) and solve for equilibrium
productions q∗(t) as,
q∗(t) = p− c0 − c1(erl + e2rl)− λuert. (2.17)
The time derivative of equilibrium production is,
q∗(t) = −rλuert < 0. (2.18)
Therefore, the equilibrium production decreases over time. In the
terminal period T , oil-in-place in stock Su are exhausted such that
Su(0) =∫ T0 q∗(s)ds. By using (2.17), this yields Su(0) = [p − c0 −
c1(erl + e2rl)]T + λu(1− erT ). On the other hand, the present-value
Hamiltonian H(T ) = 0. This implies λuerT = p− c0 − c1(erl + e2rl).
These two conditions allow us to pin down the value of T and λu.
Since S1(t) = S2(t) = 0, S1(t) = S2(t) = 0. Thus, if condition
(2.13) applies, (2.2) and (2.3) can be written as,
y12(t) = yu1(t− l) (2.19)
y12(t− l) = q(t) (2.20)
From (2.17), (2.19), and (2.20), the equilibrium two-step transfers
can be found as,
9
y∗12(t) = q(t+ l) = p− c0 − c1(erl + e2rl
)− λuer(t+l), (2.21)
and
y∗u1(t) = q(t+ 2l) = p− c0 − c1(erl + e2rl
)− λuer(t+2l). (2.22)
Thus, q∗(t) exactly equals the y∗12(t− l) and y∗u1(t− 2l) and equi-
librium production of crude oil comes from transferring resource from
Su by two steps and it takes 2l periods to complete. Then the proved
reserves are defined as those have been discovered and proved from
the oil-in-place stock, but which are not yet available for production
at time t:
R(t) =
∫ t
t−2lyu1(s)ds =
∫ t+2l
tq(s)ds, (2.23)
(2.23) implies proved reserves at t is cumulative productions be-
tween current period t and the future period t+2l. Substitute (2.17)
into the (2.23), the proved reserve can be found analytically:
R(t) = 2l[p− c0 − c1(erl + e2rl)]− λur
(e2rl − 1)ert (2.24)
Note that the firm will not hold any proved reserve if l = 0. This
is because all discoveries are immediately transferred to production
and there is nothing held as in situ proved reserves. The ratio of
proved reserves to production is then expressed as,
R(t)
q(t)=
2l[p− c0 − c1(erl + e2rl)]− e2rl−1r λue
rt
p− c0 − c1(erl + e2rl)− λuert(2.25)
The time derivative of this ratio is then computed as,
10
dR(t)q(t)
dt=λu[p− c0 − c1(erl + e2rl)](2rl − e2rl + 1)ert
[p− c0 − c1(erl + e2rl)− λuert]2(2.26)
(2.25) indicates proved reserves are larger than production if
(2l − 1)[p − c0 − c1(erl + e2rl)] − λue
rt( e2rl−1r − 1) > 0. This im-
plies that large lag l leads firms to hold more proved reserves than
current production6. This is true because proved reserves are cu-
mulative production between t and t + 2l. Additionally, because
(2rl − e2rl + 1) < 0 in (2.26), the ratio of proved reserves to pro-
duction is monotonically declining over time and it decreases more
rapidly if l is larger.
On the other hand, if the one-step route occurs, by (2.6) and (2.9),
it can be shown the ratio still monotonically decreases over time as
is the case in the two-step route. In sum, the deterministic model
presented above shows why the representative firm holds proved re-
serves: because transferring discoveries to production is delayed by
a time interval, either 2l or l, which is governed by the adjustment
costs. This leads to an amount of in situ proved reserve held by
the firm. Also, it predicts the ratio of proved reserves to production
decreases over time and its rate of declining mainly depend on the
magnitude of delayed periods l. However, this model dose not ex-
plain why this ratio declines so slowly or even remains stable over the
sample period. In that sense, we need account for something missed
in the deterministic model that affects a firm’s decisions on proved
reserve accumulations. This leads to the next section in which we
examine how uncertainties play a role in optimal productions and
proved reserve acclamations.
6The first component, (2l−1)[p−c0−c1(erl+e2rl)], is positive and increasing
in l given l > .5 and r ∈ [0, 1), but the second component, λuert( e
2rl−1r− 1), is
also positive given e2rl−1r− 1 > 0. Given p and l are sufficiently large, the first
component must outweigh the second one.
11
3 The Stochastic Model
The literature widely recognizes uncertainty as an important issue
for natural resource modelling. Pindyck [18] identifies two sources of
uncertainty on exhaustible resource extraction: one is the stochas-
tic demand that results in resource prices oscillate randomly; the
other is the fluctuation in explorations. Following Pindyck’s frame-
work, Mason ([12] and [13]) argues the stochastic demand leads to
stockpiles of short-run inventories of crude oil. Slade [23] shows the
“least-cost-first” principle in resource mining may not hold when the
demand is stochastic. Finally, Kellogg [10] provides empirical evi-
dence that firms fully take into account the expected price volatility
of crude oil when making drilling decisions. Here we consider random
shocks from demand and exploration following Pindyck and Mason’s
specifications.
The model below exhibits all the assumptions in deterministic
model, but it contains the new feature that realized prices depend
upon a stochastic variables x and realized discoveries rely on another
stochastic variable θ. With these stochastic shocks, we may write an
optimal control problem for the representative firm as follows.
J = maxq,yu1,yu2,y12
∫ T
0e−rτp(τ)q(τ)− c0q(τ)− c1[yu1(τ) + y12(τ)]
− c2yu2(τ)dτ(3.1)
subject to,
Su = −θ(t)yu1(t)− θ(t)yu2(t) (3.2)
S1 = θ(t− l)yu1(t− l)− θ(t)y12(t) (3.3)
S2 = θ(t− l)yu2(t− l) + θ(t− l)y12(t− l)− q(t) (3.4)
p = x(t)f(q(t)) (3.5)
12
dx(t)
x(t)= σxdzx(t) (3.6)
dθ(t)
θ(t)= σ1dz1(t) (3.7)
S1, S2 ≥ 0 (3.8)
where x(t) is a random shock that shifts the demand function f [q(t)]
up and down at the period t; θ(t) is a random shock that shifts
resource discoveries at the period t. Note that θ and x are multi-
plied by planned discoveries and demand. Thus, yu1(t), y12(t), and
yu2(t) are planned discoveries at the beginning of each period, while
θ(t)yu1(t), θ(t)y12(t), and θ(t)yu2(t) are realized discoveries at the
end of that period. The differentials of these random shocks are
specified by Geometric Brownian Motions, (3.6) and (3.7), respec-
tively. Note that both θ and x are thus ensured to be positive over
time and Et(θ) = Et(x) = 1 7. Moreover, the initial values of x and
θ equal 1 if the initial values of those drift terms, zx and z1, are zero.
This implies there is no random shock to demand and discovery at
the initial period.
Both demand and exploration shocks have constant volatilities,
σx and σ1, and we assume 8σ2x < r < σ21; zx and z is a Wiener process
which can be described as dz = ε√dt where ε is a serially uncorre-
lated normal random variable with zero mean and unit variance. At
the beginning of each period, the firm chooses the plan of produc-
tion, q, and discoveries, yu1, y12, and yu2, and pays the corresponding
costs, but at the end of each period the realized discoveries deviate
the planned transfers by the factor θ. Thus, we define a positive
shock if θ > 1, that is, the realized discoveries are greater than the
planned ones; but it is a negative shock if θ < 1. x behaves simi-
larly with θ. For simplicity, we assume θ and x are independent of
each other. The remaining setup keeps identical to the deterministic
model in Section 2.
7See Øksendal [17], page 64-65.8We later justify this assumption. See page 20-21.
13
The fundamental equation of stochastic optimality in this prob-
lem is,
0 = maxq(t),yu1(t),yu2(t),y12(t)
e−rtp(t)q(t)− c0q(t)− c1[yu1(t) + y12(t)]
− c2yu2(t)+ Jt(x)− JSu(t)θ(t)[yu1(t) + yu2(t)]
+ JS1(t)[θ(t− l)yu1(t− l)− θ(t)y12(t)]
+ JS2(t)[θ(t− l)yu2(t− l)) + θ(t− l)y12(t− l)− q(t)]
+σ212θ(t)2Jθθ(t) +
σ2x2x(t)2Jxx(t) + φ1(t)S1(t) + φ2(t)S2(t),
(3.9)
where JK is the derivative of J with respect to K where K =
t, Su, S1, S2, x, θ, and JKK is its second derivative.
The necessary conditions are,
∂J
∂q= e−rt(p(t)− c0)− JS2(t) = 0, (3.10)
∂J
∂yu1= −e−rtc1 − θ(t)JSu(t) + θ(t)JS1(t+ l) ≤ 0, (3.11)
∂J
∂y12= −e−rtc1 − θ(t)JS1(t) + θ(t)JS2(t+ l) ≤ 0, (3.12)
∂J
∂yu2= −e−rtc2 − θ(t)JSu(t) + θ(t)JS2(t+ l) ≤ 0. (3.13)
Differentiating (3.9) with respect to S1:
0 = JtS1 − JSuS1(t)θ(t)[yu1(t) + yu2(t)]
+ JS1S1(t)[θ(t− l)yu1(t− l)− y12(t)θ(t)]
+ JS1S2(t)[yu2(t− l)θ(t− l) + y12(t− l)θ(t− l)− q(t)]
+σ212θ(t)2JS1θθ1(t) +
σ2x2x(t)2JS1xx(t) + φ1(t).
(3.14)
14
Next, let us write down JS1 in terms of state variables and time
t, so we have JS1 = JS1(t, Su, S1, S2, θ, x) and apply Ito’s lemma to
it:
dJS1 = JtS1 − JSuS1(t)θ(t)[yu1(t) + yu2(t)]
+ JS1S1(t)[θ(t− l)yu1(t− l)− y12(t)θ(t)]
+ JS1S2(t)[yu2(t− 1)θ(t− l) + y12(t− l)θ(t− l)− q(t)]
+σ212θ(t)2JS1θθ(t) +
σ2x2x(t)2JS1xx(t)dt
+ σ1θ(t)JS1θdz1 + σxx(t)JS1xdzx.
(3.15)
Applying Ito’s differential generator, 1/dtEtd(•), 9 on both sides
of (3.15):
1
dtEd[JS1(t)] = JtS1 − JSuS1(t)θ(t)[yu1(t) + yu2(t)]
+ JS1S1(t)[θ(t− l)yu1(t− l)− y12(t)θ(t)]
+ JS1S2(t)[yu2(t− l)θ(t− l) + y12(t− l)θ(t− l)− q(t)]
+σ212θ(t)2JS1θθ(t) +
σ2x2x(t)2JS1xx(t).
(3.16)
Substituting (3.16) into (3.14) leads to,
1
dtEtd[JS1(t)] = −φ1(t), (3.17)
Repeating the procedure from (3.14) to (3.17) for Su and S2 sim-
ilarly yields,1
dtEtd[JSu(t)] = 0, (3.18)
1
dtEtd[JS2(t)] = −φ2(t). (3.19)
Similar to the deterministic case, the expected rate of changes in
shadow values of stocks are non-positive over time.
9Ito’s differential generator is analogous to the time derivatives in the deter-ministic case. For its mathematical discussion, see Chow [2].
15
The one-step or two-step route may or may not occur according
to the following proposition.
Proposition 2 Only two-step route occurs in equilibrium if
c2 > (1 + erl)c1, (3.20)
Also only one-step route occurs if
c2 < (1 + erl)c1, (3.21)
However, it is impossible for both one-step and two-step route to
occur simultaneously.
Proof Proof of Proposition 2 is provided in Appendix B.
Thus, even in a stochastic environment, adjustment costs and
delayed responses still determine whether or not the firm chooses
one-step or two-step route as the deterministic case.
Applying Ito’s differential generator to (3.10),
1
dtEtd[e−rt(p(t)− c0)] =
1
dtEtd[JS2(t)]. (3.22)
Expanding the left hand side of (3.32) yields,
1
dtEtd[JS2(t)] = −re−rt[p(t)− c0] + e−rt
1
dtEtd[p(t)]. (3.23)
Solving for 1dtEd[p(t)] as
1
dtEtd[p(t)] = r[p(t)− c0] + ert
1
dtEd[JS2(t)]. (3.24)
Suppose c2 > (1 + erl)c1, thus only two-step route occurs and
(3.11) and (3.12) hold as equalities. Use these condition to rewrite
Js2(t) as,
JS2(t) = e−r(t−l)c1
[1
θ(t− l)+
erl
θ(t− 2l)
]+ JSu(t− 2l). (3.25)
16
Using Ito’s lemma and then Ito’s differential operator on 1θ(t)
yields,
1
dtEtd
[1
θ(t)
]=
σ21θ(t)
. (3.26)
Applying Ito’s differential operator on (3.25) and using (3.26)
yields,
1
dtEtd[JS2(t)] = e−r(t−l)(σ21 − r)c1
[1
θ(t− l)+
erl
θ(t− 2l)
]. (3.27)
Substitute (3.27) into (3.24)
1
dtEtd[p(t)] =r[p(t)− c0] + erl(σ21 − r)c1[
1
θ(t− l)+
erl
θ(t− 2l)
].
(3.28)
Similarly, it can be shown the expected rate of changes in prices
follows the equation below if the one-step route occurs:
1
dtEtd[p(t)] = r[p(t)− c0] + erl(σ21 − r)
c2θ(t− l)
. (3.29)
Appendix C shows the expected rate of changes in production is,
Et[q(t)] = K − ηeσ2xt − βe(σ2
x+σ21)t − Zert. (3.30)
where K, η, Z, and β are positive constants defined as K = p+ σ2xqxr ,
η = rc0r−σ2
x, Z = K − q0 − η − β and
β =
(σ2
1−r)c1[e(r−σ
21)l+e2(r−σ
21)l
]σ2x+σ
21−r
, c2 > (1 + erl)c1
(σ21−r)c2e
(r−σ21)l
σ2x+σ
21−r
. c2 < (1 + erl)c1
(3.31)
(3.30) implies that there are three different forces that affect the
rate of expected productions over time: ert tends to lower the current
17
rate of production since Z > 0, which is consistent with the standard
Hotelling rule that high interest rates r leads to more resources being
left underground rather than turned into production. The other
two forces, eσ2t and e(σ
2x+σ
21)t, which are not taken into account in
the deterministic model, also reduce the current rate of production
given that η > 0 and β > 0. This result is consistent with real option
literature that predicts the higher uncertainty of resource prices yield
a firm’s stronger incentive to delay production([14], [10], [20], [8]).
Beyond this, the exploration uncertainty, σ21, also plays a role in
lowering the current rate of production. In this sense, with demand
and exploration uncertainties, resources must be exhausted within a
longer period of time than the deterministic model.
The proved reserves are defined similarly as the deterministic
case: if only two-step route occurs,
Et[R(t)] =
∫ t+2l
tEt[q(s)]ds
= 2lk − η
(e2σ
2xl − 1
σ2x
)eσ
2xt − β
(e2(σ
2x+σ
21)l − 1
σ2x + σ21
)e(σ
2x+σ
21)t
− Z(e2rl − 1
r
)ert.
(3.32)
On the other hand, when only one-step route occurs,
Et[R(t)] =
∫ t+l
tEt[q(s)]ds
= lk − η
(eσ
2xl − 1
σ2x
)eσ
2xt − β
(e(σ
2x+σ
21)l − 1
σ2x + σ21
)e(σ
2x+σ
21)t
− Z(erl − 1
r
)ert.
(3.33)
When (3.30) is combined with(3.32), the ratio of expected proved
reserve to expected production is greater than 1 if the inequality
below holds.
Ω(σ2x, σ21) < (2l − 1)K − Z
(e2lr − r − 1
r
)ert (3.34)
18
where Ω = η
(eσ
2xl−1−σ2
xσ2x
)eσ
2xt + β
(e(σ
2x+σ
21)l−1−σ2
1−σ2x
σ2x+σ
21
)e(σ
2x+σ
21)t is a
positive and increasing function of σ2x and σ21. The right hand side of
(3.34) is more likely to be positive given that K is large and l > 0.5.
We interpreted condition (3.34) in two separate cases.
In one case, only exploration uncertainty exists such that σ1 > 0
and σx = 0. So condition (3.34) implies,
Ω(σ21) < (2l − 1)K − Z(e2lr − r − 1
r
)ert. (3.35)
We obtain Ω(r) < Ω(σ21) by σ21 > r, so Ω(r) < Ω(σ21) < (2l −1)K − Z
(e2lr−r−1
r
)ert.
This implies r < σ21 < Ω−1[(2l − 1)K − Z
(e2lr−r−1
r
)ert], where
Ω−1 is the inverse function of Ω. This result suggests that in the ab-
sence of demand volatility, the magnitude of exploration volatility is
upper bounded. Therefore, a firm would hold larger proved reserves
than the current productions if the exploration volatility, σ21, is not
noticeably larger than the interest rate.
In the other case, only demand uncertainty exists such that σx >
0 and σ1 = 0. Thus, condition (3.34) reduces to,
Ω(σ2x) < (2l − 1)K − Z(e2lr − r − 1
r
)ert. (3.36)
Since Ω is an increasing function of σ2x, we obtain Ω(σ2x) < Ω(r)
by σ2x < r and it can be shown Ω(r) is less than the right hand
of (3.35). So in this case the condition (3.36) turns into Ω(σ2x) <
Ω(r) < (2l − 1)K − Z(e2lr−r−1
r
)ert, which reduces to σ2x < r <
Ω−1[(2l − 1)K − Z
(e2lr−r−1
r
)ert]. Therefore, in the absence of ex-
ploration uncertainty, a firm would hold larger proved reserves than
the current productions if the demand volatility, σ2x, is much less
than the interest rate.
To summarize, the above reasoning suggests a firm would hold
larger proved reserves than current production if either demand volatil-
ity is much less than interest rate or the exploration volatility is not
noticeably larger than the interest rate. It is interesting to examine
19
this prediction by the data observed from real world. We estimate σ2xby calculating the variance of (lnPt − lnPt−1) in the period of 1977
to 2013, where Pt is the annual WTI price obtained from EIA. The
estimated σ2x is 0.0478. In next step, we estimate σ21 by calculating
the variance of (lnyt − lnyt−1) in the period of 1986 to 2013, where
yt is the discovery of proved reserves at year t obtained from EIA10.
The estimated σ21 is 0.156. Finally, according to a 1995 survey by
the Society of Petroleum Evaluation Engineers (SPEE), the median
nominal discount rate applied by firms to cash flows is 0.125 [10]11.
Thus, r−σ2x = 0.077 and σ21− r = 0.031. The difference between the
interest rate and the demand volatility is two times larger than the
difference between the exploration volatility and the interest rate.
This indicates demand volatility is much less than interest rate and
the exploration volatility is not noticeably larger than the interest
rate. Thus, we verify our prediction from the stochastic model.
Divide (3.32) by (3.30), time differentiate it and rearrange to
yield an equation below,
d(Et[R(t)]Et[q(t)]
)dt
=1
q(t)2
−ηK(e2lσ2x − 2lσ2x − 1)eσ
2xt
−ZK(e2lr − 2lr − 1)ert
−βK(e2l(σ
2x+σ
21) − 2l(σ2x + σ21)− 1
)e(σ
2x+σ
21)t
+ηZ(r − σ2x)(e2lr−1r − e2lσ
2x−1σ2x
)e(σ
2x+r)t
+βησ21
(e2l(σ
2x+σ
21)−1
σ2x+σ
21− e2lσ
2x−1σ2x
)e(2σ
2x+σ
21)t
+βZ(σ2x + σ21 − r)(e2l(σ
2x+σ
21)−1
σ2x+σ
21− e2lr−1
r
)e(σ
2x+σ
21+r)t.
(3.37)
The first three components in (3.37) are negative, while the next
three terms are positive given that e2rv−1v is increasing in v. On one
10EIA reports three sources of discoveries proved reserves: extensions, newreservoir discoveries in old fields and new field discoveries. We aggregate thesethree sources to the discovery of proved reserves.
11A more recent SPEE survey (2008) showed that 71 percent of companiesused a cost-of-capital discount rate in the range 9 percent to 11 percent, with anaverage of 10.4 percent. See Moore [16], page 35
20
hand, the ratio of expected proved reserve to expected productions
is still decreasing over time since K is larger than any other param-
eters; on the other hand, unlike the deterministic model, three pos-
itive components introduced by volatilities in (3.37) provide forces
that lower the rate of decrease in this ratio. Therefore, the ratio
of expected proved reserves to expected production tends to decline
much more slowly or even remains stable over some periods.
4 Conclusion
This paper answers important questions in the oil industry: what are
the forces that drive firms’ decisions on proved reserve accumulations
and why do they accumulate large amounts of proved reserves? We
consider a representative firm that maximizes the lifetime profits by
choosing an optimal plan of oil discoveries and production over time.
We recognize that there are important delays and adjustment costs
in transferring resources from the oil-in-place stock to the production
stage. We found the larger the delayed periods are, the more likely
the firm will reach production quickly. This relationship is still true
in a stochastic model in which we allow the random shocks from
demand and exploration to play a role in determining the resource
transfers. The deterministic model explains why firms hold proved
reserves by the delayed responses and adjustment costs, but does not
account for why the ratio of proved reserves to productions declines
very slowly or even remains stable.
In the stochastic model, we found both demand and exploration
volatilities reduce current rate of production and thus the stock of oil
in place exhausts over a longer period. The results of the stochastic
model suggest that a firm would hold larger proved reserves than
current production if demand volatility is much less than interest
rate, or if the exploration volatility is not noticeably larger than the
interest rate. We examines this prediction using the U.S. data and
find it is consistent with empirical evidences. The intuition behind
the stochastic model suggests that proved reserves refer to expected
future productions within a fixed period of time and they differ from
21
the concept of oil in place. Therefore, although demand and ex-
ploration volatilities lower current rate of production and leave more
resources as oil in place, proved reserves may not necessarily increase.
Proved reserves are proven to be economically and technically recov-
erable under current conditions so that they are ready for expected
productions in the near future. Since uncertainties lower the rate
of current production, the expected productions in the near future
would also be lowered. In this sense, large uncertainties may dis-
courage a firm’s incentive to accumulate proved reserves. Thus, in
order to encourage a firm to hold a greater amount of proved reserves
than the current production, these volatilities cannot be very large.
Specifically, we show that the demand volatility is much smaller than
the interest rate, and exploration volatility is slightly larger than the
interest rate. The existence of these volatilities also yields the ratio
of reserves to production to decline slowly or even remains relatively
stable.
22
References
[1] Boyce, W.E., Diprima, R. “Elementary Differential Equations
and Boundary Value Problems”. John Wiley, U.S.A, 7 edition,
2003.
[2] Chow, G.C. “Optimal Control of Stochastic Differential Equa-
tion Systems”. Journal of Economics Dynamic and Control,
1(2):143–175, 1979.
[3] EIA. “U.S. Crude Oil and Natural Gas Proved Reserves”. Web-
site, 2014. http://www.eia.gov/naturalgas/crudeoilreserves/.
[4] Farzin, Y.H. “The Impact of Oil Price on Additions to US
Proven Reserves”. Resource and Energy Economics, 23(3):271–
291, 2001.
[5] Heal, G. “The Relationship between Price and Extraction Cost
for a Resource with a Backstop Technology”. Bell Journal of
Economics, 7(2):371–378, 1976.
[6] Holland, S. “Modelling Peak Oil”. The Energy Journal,
29(2):61–80, 2008.
[7] Hotelling, H. “The Economics of Exhaustible Resources”. Jour-
nal of Political Economy, 39(2):137–175, 1931.
[8] Hurn, A. S., Wright. R. E. “Geology or Economics? Testing
Models of Irreversible Investment Using North Sea Oil Data”.
Economic Journal, 104(423):363–71, 1994.
[9] Kamien, M., Schwartz, N. “Dynamic Optimization: The Calcu-
lus of Variations and Optimal Control in Economics and Man-
agement”. North Holland, Amsterdam, 2 edition, 1991.
[10] Kellogg, R. “The Effect of Uncertainty on Investment: Evi-
dence from Texas Oil Drilling”. American Economic Review,
106(6):1698–1734, 2014.
23
[11] Livernois, J., Uhler, R. “Extraction Costs and the Economics
of Nonrenewable Resources”. Journal of Political Economy,
95(1):195–203, 1987.
[12] Mason, C. “On Stockpiling Natural Resources”. Resource and
Energy Economy, 33(2):398–409, 2011.
[13] Mason, C. “On Equilibrium in Resource Markets with Scale
Economies and Stochastic Prices”. Journal of Environmental
Economics and Management, 64(3):288–300, 2012.
[14] Moel, A., Tufano, P. “When Are Real Options Exercised? An
Empirical Study of Mine Closings”. Review of Financial Studies,
15(1):35–64, 2002.
[15] Mohn,K., Osmundsen,P. “Exploration Economics in a Regu-
lated Petroleum Province: The Case of the Norwegian Conti-
nental Shelf”. Energy Economics, 30(2):303–320, 2008.
[16] Moore, C.R.K. “Perspectives on the Valuation of Upstream Oil
and Gas Interests: An Overview”. Journal of World Energy
Law and Business, 2(1):24–41, 2009.
[17] Øksendal,B. “Stochastic Differnetial Equations: An Introduc-
tion with Application”. Springer, U.S.A, 6 edition, 2003.
[18] Pindyck, R. “Uncertainty and Exhaustible Resource Markets”.
Journal of Political Economy, 88(6):1203–1225, 1980.
[19] Pindyck, R. “Adjustment Costs, Uncertainty, and the Behavior
of the Firm”. American Economic Review,, 72(3):415–427, 1982.
[20] Pindyck, R. “Irreversibility, Uncertainty, and Investment”.
Journal of Economic Literature, 29(3):1110–48, 1991.
[21] Schultz, W. “The Optimal Use of Non-Renewable Resources:
The Theory of Extraction”. Journal of Environmental Eco-
nomics and Management, 1(1):53–73, 1974.
24
[22] Slade, M. “Trends in Natural-Resource Commodity Prices: An
Analysis of the Time Domain”. Journal of Environmental Eco-
nomics and Management, 9(2):122–137, 1982.
[23] Slade, M. “Grade Selection Under Uncertainty: Least Cost Last
and Other Anomalies”. Journal of Environmental Economics
and Management, 15(2):189–205, 1988.
[24] Solow, R., Wan, F. “Extraction Costs in the Theory of Ex-
haustible Resources”. Bell Journal of Economics, 7(2):359–370,
1976.
[25] Swierzbinski, J., Mendelsohn. R. “Exploration and exhaustible
resources: The microfoundations of aggregate models”. Bell
Journal of Economics, 30(1):175–186, 1989.
25
5 Appendix A
5.1 A1
Suppose (2.7) and (2.8) hold as equalities but (2.9) holds as strict
inequality. Subtracting (2.8) from (2.9) yields
e−rt(c2 − c1) + λu(t)− λ1(t) > 0. (5.1)
Lag (2.7) by l periods as,
λ1(t) = λu(t− l) + e−r(t−l)c1. (5.2)
Substituting it into (5.1) and recognizing λu(t) = λu(t− l) yields
c2 > (1 + erl)c1. (5.3)
So if c2 > (1 + erl)c1, yu1(t) > 0, y12(t) > 0, but yu2 = 0.
Thus only two-step transfers occur. However if (2.7)-(2.9) hold as
equalities, they reduce to c2 = (1 + erl)c1, However the existence of
adjustment cost implies c2 > 2c1 and thus (1 + erl)c1 > 2c1. This
cannot hold for any l ≥ 0. Thus it is impossible for both two-step
and one-step transfers to occur simultaneously.
On the other hand, suppose (2.7) and (2.8) hold as strict inequal-
ities but (2.9) hold as equality. Rewrite (2.7) and (2.8) as
λu(t− l) + e−r(t−l)c1 > λ1(t), (5.4)
λ1(t) > λ2(t+ l)− e−rtc1. (5.5)
They imply that
λu(t− l) + e−r(t−l)c1 > λ2(t+ l)− e−rtc1. (5.6)
Substituting (2.9) into the inequality above to eliminate λ2(t+ l)
yields the result as,
λu(t− l) + e−r(t−l)c1 − ertc2 − λu(t) > e−rtc1. (5.7)
26
Note that λu(t) = λu(t− l) and the equation above reduces to
c2 < (1 + erl)c1. (5.8)
Under this condition, yu1(t) = yu2(t) = 0, but yu2 > 0. Thus,
only one-step transfers occur.
5.2 A2
This section offers the proof by contradiction for S1(t) = S2(t) = 0
along the equilibrium path.
Suppose at some period t = t∗, S1(t∗) > 0 and S2(t
∗) > 0. Then
φ1(t∗) = φ2(t
∗) = 0. By (2.11) and (2.12), λ1(t∗) = λ2(t
∗) = 0.
Rewrite (2.7) as
λ1(t∗ + l) ≤ e−rt∗c1 + λu. (5.9)
Note that λu is constant over time because of (2.10). The left
hand side of (5.9) denotes the marginal benefits of moving extra one
unit of oil from stock Su to S1, while the right hand side captures
the full marginal cost of moving extra one unit of oil from Su to S1.
It consists of two components: e−rt∗c1 denotes the present value of
marginal cost of the transfer; while λu is opportunity cost of trans-
ferring extra one unit of oil from Su to S1.
Suppose λ1(t∗ + l) = e−rt
∗c1 + λu. Then in the following period
t > t∗, λ1(t+ l) > e−rtc1 + λu because λ1(t∗) = 0 and thus λ1(t
∗) =
λ1(t), and e−rt < e−rt∗. This implies that the marginal benefits of
moving extra one unit from Su to S1 is greater than the full marginal
cost of this transfer. Given this situation, the firm will liquidate all
the resource in Su and move them to S1 at time t [12]. This radical
action contradicts to the fact that Su(t) > 0 and thus λu(t) = 0.
On the other hand, suppose λ1(t∗+ l) < e−rt
∗c1+λu. If the value
of λ1(t∗ + l) − λu is sufficiently small such that in the next period
t > t∗, λ1(t + l) < e−rtc1 + λu still holds. However, if the value
of λ1(t∗ + l) − λu is not small such that in the some period t > t∗,
λ1(t + l) = e−rtc1 + λu since the right hand side of (5.9) is lowered
by t > t∗. This leads to the previous case again in which the firm
27
liquidates all the resource in Su.
In sum, it must be true S1(t) = 0 along the equilibrium path.
Similarly, S2(t) = 0 along the equilibrium path.
28
6 Appendix B
This section offers the proof of proposition 2.
Suppose (3.11) and (3.12) hold as equalities and (3.13) holds as
inequality. Subtracting (3.12) from (3.13) yields
e−rt(c2 − c1) + θ(t)Jsu(t)− θ(t)Js1(t) > 0. (6.1)
Lagging (3.11) by l periods and solving for Js1(t),
Js1(t) =e−r(t−l)c1θ(t− l)
+ Jsu(t− l). (6.2)
Substitute it into the (6.1) to eliminate Js1(t) and yields,
e−rt(c2−c1)+θ(t)[Jsu(t)−Jsu(t− l)]− θ(t)
θ(t− l)e−r(t−l)c1 > 0. (6.3)
Take expectation Et over (6.3) and yields
e−rt(c2 − c1)− e−r(t−l)c1Et[θ(t)
θ(t− l)] > 0, (6.4)
while Et[Jsu(t) − Jsu(t − l)] = 0 because integrating both sides of
(3.18) yields Et[Jsu(t)] = Et[Jsu(t− l)] = Et[Jsu(0)]. Moreover, θ(t)
can be solved for12,
θ(t) = e[−12σ21t+σ1z1(t)], (6.5)
So,
θ(t)
θ(t− l)= e−
12σ21 leσ1[z1(t)−z1(t−l)]. (6.6)
Since dz1 is a Wienner process, z1(t)−z1(t−l) obeys a distribution
as N ∼ (0, l). So σ1[z1(t) − z1(t − l)] obeys a distribution as N ∼(0, σ21l) and its log normal distribution eσ1[z1(t)−z1(t−l)] has a mean as
Et[eσ1[z1(t)−z1(t−l)]] = e
12σ21 l. Thus, by using (6.6),
Et[θ(t)
θ(t− l)] = e−
12σ21 l+
12σ21 l = 1. (6.7)
12see Øksendal [17], page 64-65
29
Substitute (6.7) into (6.4), multiplies through ert and yield,
c2 > (1 + erl)c1, (6.8)
when (6.8) holds, only two-step route occurs.
Similarly, if (3.11)-(3.13) hold as equalities, they reduce c2 =
(1 + erl)c1. However, it contradicts to c2 > 2c1. Therefore, two-step
and one-step route cannot occur simultaneously.
On the other hand, Suppose (3.11) and (3.12) hold as inequalities
and (3.13) holds as equality. Lagging l periods of (3.11) as,
Js1(t) < Jsu(t− l) +e−r(t−l)c1θ(t− l)
, (6.9)
Rewrite (3.12) as,
Js2(t+ l)− e−rtc1θ(t)
< Js1(t). (6.10)
Therefore,
Js2(t+ l)− e−rtc1θ(t)
< Jsu(t− l) +e−r(t−l)c1θ(t− l)
. (6.11)
Multiplying θ(t) on both sides yields,
θ(t)Js2(t+ l)− e−rtc1 < θ(t)Jsu(t− l) + e−r(t−l)c1θ(t)
θ(t− l). (6.12)
Substituting (3.13) into (6.12) to eliminate θ(t)Js2(t+l) and then
applying Et on both sides of (6.12) yields,
c2 < (1 + erl)c1. (6.13)
So under this condition only one-step route occurs.
30
7 Appendix C
Applying Ito’s lemma on (3.5) yields,
dp = xf ′(q)dq + f(q)dx+1
2xf ′′(q)(dq)2 + f ′(q)dqdx. (7.1)
Note that the optimal production q = q∗(Su, S1, S2, x, θ) along
the equilibrium path. Now expanding dq using Ito’s lemma:
dq =−qsu(t)θ(t)[yu1(t) + yu2(t)]
+ qs1(t)[θ(t− l)yu1(t− l)− θ(t)y12(t)]
+ qs2(t)[θ(t− l)yu2(t− l) + θ(t− l)y12(t− l)− q(t)]
+σ212θ2(t)qθθ +
σ2x2x2(t)qxxdt
+ σxx(t)qx(t)dzx + σ1θ(t)qθ(t)dz1.
(7.2)
Thus
(dq)2 = [σ2xx2(t)q2x(t) + σ21θ
2(t)q2θ(t)]dt, (7.3)
and,
dqdx = [σ2xx2(t)qx(t)]dt. (7.4)
Substituting (7.3) and (7.4) into (7.1) and then applying Ito’s
differential operator on (7.1) yields,
1
dtEt(dp) =xf ′(q)
1
dtEt(dq) +
1
2xf ′′(q)[σ2xx
2(t)q2x(t) + σ21θ2(t)q2θ(t)]
+ f ′(q)[σ2xx2(t)qx(t)].
(7.5)
Substitute (3.28) and (3.29) into (7.5) respectively and then solve
for 1dtEtd[q] as
1
dtEt[dq(t)] =
erl(σ21 − r)c1[ 1θ(t−l) + erl
θ(t−2l) ] + r(p(t)− c0)xf ′(q)
−12xf
′′(q)[σ2xx2(t)q2x(t) + σ21θ
2(t)q2θ(t)] + f ′(q)[σ2xx2(t)qx(t)]
xf ′(q)(7.6)
31
1
dtEt[dq(t)] =
erl(σ21 − r) c2θ(t−l) + r(p(t)− c0)xf ′(q)
−12xf
′′(q)[σ2xx2(t)q2x(t) + σ21θ
2(t)q2θ(t)] + f ′(q)[σ2xx2(t)qx(t)]
xf ′(q)(7.7)
In a case of the linear demand curve f(q) = p − q(t), (7.6) and
(7.7) reduce to,
1
dtEt[dq(t)] =
r(p(t)− c0) + σ2xx2(t)qx(t)
−x
+erl(σ21 − r)c1[ 1
θ(t−l) + erl
θ(t−2l) ]
−x
(7.8)
1
dtEt[dq(t)] =
r(p(t)− c0) + σ2xx2(t)qx(t)
−x
+erl(σ21 − r) c2
θ(t−l)
−x
(7.9)
In next step, we solve the equilibrium level of expected production
E[q(t)]. Substitute p(t) = x(t)[p− q(t)], 13x(t) = e−12σ2xt+σxzx(t) and
θ(t) = e−12σ21t+σ1z1(t) into (7.8) and (7.9) and then take expectation
over both sides and yields,
1
dtEt[dq(t)] =− rp− σ2xqx + rEtq(t) + rc0e
σ2xt
− e(r−σ21)l(σ21 − r)c1(1 + e(r−σ
21)l)e(σ
2x+σ
21)t,
(7.10)
and,
1
dtEt[dq(t)] =− rp− σ2xqx + rEtq(t) + rc0e
σ2xt
− e(r−σ21)l(σ21 − r)c2e(σ
2x+σ
21)t.
(7.11)
Note that we are using the fact that Et[x(t)] = 1 and 14Et[e−σzx(t)] =
e12σ2t.
(7.10) and (7.11) are first-order linear differential equations and
13A general solution to x(t) can be found at Øksendal [17], page 65.14see Øksendal [17], page 65.
32
their are solutions15 can be solved for,
Et[q(t)] = K − ηeσ2xt − βe(σ2
x+σ1)t − Zert. (7.12)
where η, β, and Z are positive constants defined as K = p + σ2xqxr ,
η = rc0r−σ2
xand
β =
(σ2
1−r)c1[e(r−σ
21)l+e2(r−σ
21)l
]σ2x+σ
21−r
, c2 > (1 + erl)c1
(σ21−r)c2e
(r−σ21)l
σ2x+σ
21−r
. 2c1 < c2 < (1 + erl)c1
(7.13)
15For a first-order differential equation y + αy = g(t), its general solutionis y = e−αt
∫eαsg(s)ds + ce−αt, where c is a constant pinned down by initial
conditions. See Boyce and Diprima [1], page 33.
33