Developing a Fair and Robust Energy Policy
Frank Blaskovich
Blaskovich Services, Inc.
Aptos, California, USA
ftblask@yahoo.com
Introduction
We used a system dynamics model and statistical analysis to find an energy tax policy
that is both fair and robust for all interested parties based on the assumptions made in this
study. Regret analysis techniques were developed and used effectively to find a policy
based on equivalent gains and losses for all stakeholders. The policy worked well under
a wide range of uncertain future conditions. Our approach could be particularly effective
when it is difficult or impossible to define or agree upon the probabilities of important
events.
The goal of this work was to expand on ideas that were developed in our previous papers,
as well as the work of other researchers in system dynamics, petroleum engineering and
other disciplines. We focused our efforts on defining a policy development process,
rather than on the potential complexities that might arise in real situations. However, we
believe the approach described here can be adapted to more realistic situations. That will
be the focus of future research.
Uncertain parameters (oil price, initial oil reserves) were varied for each future scenario
and also allowed to vary with time. The results of our analysis provided a clear definition
of the policy “playing field” bounded by the solutions that gave the stakeholders their
maximum benefits. Within this field, we also found a fair policy that resulted in
equivalent stakeholder gain and pain. Other policy alternatives could also be analyzed to
see where they are on the playing field and how fair they might be.
We hope that our techniques may eventually be used to find ways to encourage the
development of new energy resources in a way that is both fair and beneficial to all
interested parties.
Background
Sound energy policies are difficult to develop because it is hard to accurately define all of
the variables that can affect them. We chose to use a system dynamics approach to this
problem because this technique encourages us to include a wide range of parameters in
the model development process. In this paper, our model is an extended version of an
earlier model developed to study energy policy issues!*.
System dynamics and regret analysis have been used extensively*”’ in the last 10-20 years
to develop and evaluate policies in the presence of “deep uncertainty”, where the model
input data and/or structure is uncertain and probability distributions are difficult to define,
debatable and/or unknowable. This is particularly important when surprise events can
have a large impact on system behavior and results. We believe that the best policies are
those that are robust (work reasonably well even when bad surprises occur) and adaptable
(have the capability to accept and adjust to new information). Our efforts described in
this paper are an attempt to develop and test this type of policy in light of uncertain future
conditions.
As in our earlier work, we have assumed that there are two main stakeholders — the State
government and an oil producer (called Producer). The State collects fees (royalties and
taxes) from the Producer as a function of the volumes of energy resources available and
produced. In our model, these fees are the only source of revenue for the State and are
used to maintain and improve the conditions of the society that the State represents. The
Producer receives all of its revenue from one oil field and is solely responsible for the
costs of field development and operation, fees levied by the State, and final abandonment
of the field (e.g., restoration of the State lands to pre-development conditions).
We have also assumed that the State and Producer are completely separate entities and
have not included any societal or corporate needs in the calculations at this point in time.
This is an important simplification. For example, the State needs enough revenue to meet
the needs of the society and the Producer needs a positive net cash flow to remain in
business. A future version of our model might include the impact of population growth
on the societal needs and the impact of corporate growth requirements on the energy
provider’s needs.
We believe that our approach to a fair and robust energy policy could be extended to
address much more complex situations and help the State and Producer work together to
meet the energy needs of society well into the future. Our longer-term goal is to
investigate ways to encourage the development of untapped energy resources — both
fossil-based and renewable.
The Model
A system dynamics (SD) model* from previous work was used to represent the process of
developing and operating an oil reservoir that could typically exist in any oil-producing
region. The field has already been discovered when the model starts and is ready to be
developed.
Pipeline OPEX
Comeltion
= Macro
Ra
“* Pipsine oP
Costs
Fae F | Drain Cost Per
Well
| Facdty Cost
ae ees
State Royty Rate
=
et OF
4 Produce cf ater | Ta
state O1RSeme Talat
Discount Rte
: Rs nd
Poducepvt—
Figure 1 — System Dynamics Model Schematic — “model” view
Figure 1 shows the upper level design (“model” view) of the current SD model. This
view represents the oil production and economics of the field and the stakeholders. Oil is
produced from facilities built and wells drilled by the Producer, who pays a tariff to the
pipeline to transport the oil to market and fees to the State (as already described).
End of field life (EOFL) occurs when the Producer net cash flow (NCF) becomes
negative at least one year after development has ended. At that time, any remaining
depreciation from capital investments (CAPEX) is expensed and field abandonment costs
are incurred. The State only receives revenue as long as the Producer continues to have a
cash flow.
Depreciation Tangle CAPER
Lis oe
CCAPEX Expense
—Speta Tine
a
i
Welk Reguied
Last ¥
Fei Mbandonment
Cost
Undeprecited Coss
‘Normal Price User Input Sate
— Prod Tax Rate
Poly Optna
Ny fai — = \ \.
New Fat -— nae 8 ‘Pre sate od Ta
f “npn <9
Normal ovat Ue Py
Figure 2 — System Dy ics Model Sch ic - i view
Figure 2 shows the “planning” view from the SD model, which includes three different
calculations that are used in the “model” view described above.
The Wells Required variable (shown as a circle) is an external function that simulates the
Producer’s field development planning efforts with the objective of determining the
number of wells that should be drilled to maximize the Producer’s net present value
(NPV). Each time step, this calculation takes the currently available information about
the field (e.g., Initial Oil Reserves estimate) and other important variables (e.g., Oil Price)
and uses them to adjust the Wells Required result.
CAPEX depreciation, undepreciated CAPEX and field abandonment costs are also
calculated in the “planning” view and passed to the “model” view to complete the
economic calculations during each time step. These variables are shown in the upper
right corner of Figure 2.
The State Production Tax Rate (State Prod Tax Rate in Figure 2) is also calculated in the
“planning” view. Depending on the type of case being run, the user can input the State
Prod Tax Rate directly or the model can use correlations based on normalized curves and
endpoints as functions of important input variables (e.g., Oil Price, Initial Oil Reserves)
as shown in the lower right corner of Figure 2.
Figure 3 shows the “surprise” view from the SD model. This view contains variables that
can be used to simulate fluctuations in the variables that are particularly difficult to
estimate (e.g., uncertain probability distributions) and can vary with time or with other
events that can occur in the model during a scenario.
Initial O7 Reserves \
Noe Seed Avg Oi Price Est |
‘thee “\ Ss Ke rae
Sep iRe,
Chan
Pa ig average ea Price Initial Oil Reserves Fraction Remaining
Last Year To Pandere
Max Price Change Fractional Reserves /
Changes Due To Driling
Time i Last Year OFL Fly
Pa]
[ee |
Noise Seed. _Fractonal Reserves
mmaniia ont a
Changes Due To Driling
Initial Oi Price
Drie Well Drilled Last Welk R
Year Yeat
Figure 3 — System Dynamics Model Schematic — “surprise” view
In our current model, the Oil Price can change with time as a function of changes that
might occur outside the model. The Initial Oil Reserves can change as new wells are
drilled that confirm, raise or lower previous estimates. These parameters can have an
impact on the Wells Required and, as a result, on the total wells drilled in any scenario.
Parameter Sensitivity Analysis
The policy development process begins with identifying the most important model
variables that can influence the results we are interested in. The parameters are either
“drivers” (strongly affect outcomes, not under direct user control) or “levers” (strongly
affect outcomes, under user control).
State NPV Producer NPV
coomiciont
coomicient
Figure 4 - State NPV and Producer NPV Sensitivity Analysis
We assumed that our energy policy would be a function of NPV and identified the main
policy drivers and levers by using a single parameter sensitivity analysis technique. Each
parameter was assigned a likely range of values and sensitivity runs were made varying
each parameter one at a time. These SD model results were post-processed using an R°
script and the results are shown for the State and Producer NPV on Figure 4.
Longer bars indicate more important input parameters. Positive values suggest that the
parameter impact and NPV are positively correlated. Negative values suggest a negative
correlation.
As described in our earlier papers, the three State revenue sources (State Royalty Rate,
State Income Tax Rate and State Prod Tax Rate) are the primary levers that affect State
and Producer NPV. The three primary drivers are the Oil Price, Initial Oil Reserves and
the Max Oil Rate. To simplify our analysis in this paper, we used the State Prod Tax
Rate as our policy lever and Oil Price and Initial Oil Reserves as our policy drivers.
More variables will be included in our future research efforts and may require more
sophisticated analysis techniques" to identify the important drivers and levers.
Regret Analysis
Regret analysis is a powerful technique for finding policy strategies that can reduce the
impact of a “worst-case surprise” scenario. We developed these strategies by running our
SD model with numerous scenarios in which we varied the State Prod Tax Rate, the Oil
Price and the Initial Oil Reserves systematically. The tax rate varied from 0 to 100%, the
oil price varied from 0 to 300 $/stb (US dollars per stock-tank barrel), and the oil reserves
varied from 0 to 2 bstb (billion stock-tank barrels). In these cases, the variables were
held constant with time.
We post-processed the SD model results and calculated Regret for the State and for the
Producer as follows:
Model Regret(s,f) = Maximum Model NPV(f) — Model NPV(s,f)
Each future (f) was defined by an Oil Price and Initial Oil Reserves value. We generated
different scenarios (s) for every future by varying the tax rate and calculating the State
and Producer NPV values from the SD model. As expected, the maximum State NPV
required a large State Prod Tax Rate value. However, above a certain tax rate, the State
and Producer NPV became negative and both stakeholders failed. The Producer NPV
was maximized with a State Prod Tax Rate equal to zero.
The regret values for the stakeholders were then normalized as Relative Regret values
using the following relationship:
Relative Model Regret(s,f) = Model Regret(s,f) / Maximum Model Regret(f)
We believe that it is fairer to all stakeholders to compare fractional regret values since the
maximum obtainable NPV for each may be quite different.
From these results, we were able to calculate the Equivalent Regret as follows:
Equivalent Regret(s,f) =| Relative State Regret(s,f) — Relative Producer Regret(s,f)|
Equivalent Regret values close to zero reflected a compromise between the two extreme
cases and suggested a policy solution in which the State and Producer would experience
very similar NPV losses (i.e., deviations from their optimal NPV cases).
50 100 160 200 260 300 60 100 150 200 260 300
a ec Tato
{Hii oi Ti info int aaa: i
raaae
0.06
Fos
Loo 0.04
itor hit 01 ME int 01 it o1
Tax Rate (fraction)
pi
| 0.02
a a a Tt 0.00
50 100 150 200 250 200
Oil Price ($/stb)
50 109 150 200 250 300
Figure 5 — Equivalent Regret with Initial Oil Slices
We analyzed the Equivalent Regret solution space by plotting the 3-D results using 2-D
slices. Figure 5 shows the results using slices with various ranges of Initial Oil Reserves
values, highlighted as orange bars in each panel. Small reserves values are shown in the
oo 05 10 15 20 00 05 10 18 20
a a po
Oil Pica Oil Pice Oil Price mt Ol Price
08+ +
064 r
col eel | ee cl ee - 0.08
024 +
2 | Gilprice Oilaite Oil Ries, Oi Pree L
5 0.06
| bos
| + 06
3 | Pea tances! raed ie
e | Los | foo
5
J oil Price HM oil Price Gi Price WS Price
084 i.
0.02
064 r
044 r
024 +
* a ttt 0.00
00 05 10 18 20 00 05 10 15 20
Initial Oil (bstb)
Figure 6 - Equivalent Regret with Oil Price Slices
bottom left panel. Large reserves values are shown in the upper right panel. The color
scale represents Equivalent Regret values varying from 0 to 0.1. Values greater than 0.1
were not plotted to avoid unnecessary clutter and to help us identify important trends.
Similarly, we plotted the Equivalent Regret results with Oil Price slices in Figure 6. Low
oil prices are shown in the lower left panel and high oil prices are shown in the upper
right panel.
We used these model results to generate a correlation for use in the SD model to simulate
various tax policies, using normalized curves and endpoint correlations as functions of
Oil Price and Initial Oil Reserves. The SD model structure for this correlation is shown
in the “planning” view on Figure 2.
Model Results
We tested the model tax correlations by running several thousand cases with randomly
generated Oil Price and Initial Oil Reserves values (held constant with time) to see how
well the State Prod Tax Rate correlation would perform. The results from 10,000
possible futures are shown on Figure 7 as density and cumulative density plots in the
right and left panels, respectively. The upper panels show the results for the Producer
NPV and the lower panels show the results for the State NPV.
10 =
os ea ae
oa 8 F
z state —— 9 os ae State
z03 Prod —— § F Prod
6 02 4 BS / Net
at 5 02 le
ao oo |
0 2 4 6 8 Mw
prod nov
os 10
i oe
zoe sap — 8 as =
2 Prod 8 Prod ——
GB 006 Net a Oe a —
iB o2
oo : a0
rr) a 5 i 6
state npv state nov
Figure 7 - State and Producer NPV Distributions - No Time Variations
The tax policy based on Equivalent Regret (shown as Net in green) appeared to be fair to
both the State and the Producer. It generated better results for the State than the tax
policy based on the lowest Relative Regret for the Producer (shown as Prod in pink). It
also generated better results for the Producer than the tax policy based on the lowest
Relative Regret for the State (shown as State in blue). In effect, both parties gained and
lost by the same relative amounts. This is just one compromise solution that might exist
between the extreme State and Producer solutions that define the “playing field”.
The Net tax policy might also be called robust because none of the 10,000 cases resulted
in a failure, where failure is defined as an economic loss for either the State or the
Producer. The final State and Producer NPV values were greater than zero for all 10,000
futures modeled.
It is important to stress that robustness was defined here only as a positive model NPV.
In a more realistic model, the needs of the State and Producer should be included in our
definition. There may be many cases in which the final NPV is positive, but the
economic needs of either stakeholder may not be met. As stated earlier, the State needs
revenues to meet the needs of society. The Producer needs revenues to grow profitability
with time. This suggests that the tax policy developed here is only one component of a
more comprehensive energy program, which could include additional fossil fuel
development, renewable energy resources, conservation, etc.
Variable Uncertainties with Time
To more rigorously check the robustness of our calculations, we ran a series of additional
model runs with the State, Producer and Net tax policies in which we allowed the Oil
Price and Initial Oil Reserves to change with time. Figure 3 shows the model structures
used to generate synthetic Oil Price and Initial Oil Reserves variations.
a
State ——
/
State
Prod
Net
Density
State. ——
Prog
Density
2
Empirical CDF
&S
z
&
state npv slate npy
Figure 8 - State and Producer NPV Distributions — Time Variations
The results on Figure 8 show that the Net tax policy continued to do a reasonable job of
generating model results that were between those generated using the best State and
Producer policies over the entire range of NPV results obtained. The left panels show the
distribution of model results as densities and the right panels show results as cumulative
densities. Upper panels show results from the Producer’s perspective and lower panels
show results from the State’s perspective.
As for the cases shown on Figure 7 above, the Net policy met our fairness criteria
because it resulted in equivalent gains and losses for both the State and Producer.
However, all three policies were less robust than those in the earlier cases. Of the 10,000
futures studied, the best State policy failed (i.e., had a negative final NPV) 48 times, the
best Producer policy failed 25 times and the Net tax policy failed 28 times. Since the
Producer policy used a zero tax rate, it seems reasonable to conclude that most of the Net
tax policy failures were not caused by the tax rate. The overwhelming majority of Net
tax policy failures were due to field economics failures due to a rapid drop in the Oil
Price and/or drilling that found less Initial Oil Reserves than was originally expected.
Policy Implications
As mentioned above, the technique described here can effectively create a policy
“playing field” with boundaries defined by the best tax rates for the State and Producer
(i.e., those that result in their respective minimum Relative Regret values). A fair
solution is a policy that is based on equivalent pain and gain for all stakeholders, as
defined when the Equivalent Regret is zero. Any other policy can also be plotted on the
playing field to see how fair it is to the stakeholders. This suggests that our approach
could be valuable for policy negotiations and for developing stakeholder strategies.
Conclusions
Our simple model shows that an energy policy can be robust and flexible to account for
an uncertain or unknowable future. Traditional forecasting techniques can fail in light of
deep uncertainties. Regret analysis gives a clearer picture of the risks involved, likely
conflicts and possible solutions. In the studies conducted, the Net tax policy (based on
Equivalent Regret considerations) appeared to be both fair to the State and Producer and
robust. More sophisticated techniques may be required to identify sound policies when
there are numerous drivers and levers in the system.
Future Work
This research will be extended to study more realistic situations. For example, the SD
model and regret analysis could be used to look at ways to encourage the development of
untapped energy resources. It is possible that reduced taxes or other incentives might
help. Another issue might be the impact of changes in the society (e.g., population
growth) on the State’s needs for both revenues and energy supplies.
Mechanical failure and climate change surprises could be included in the SD model.
Multiple oil and/or gas fields (developed, undeveloped and undiscovered) could be
included to study ways to encourage exploration and development. These techniques
may also prove valuable for designing policies that smooth the transition to renewable
energy resources.
References
1. Blaskovich, F.T., “Oil Policy Regret Analysis With System Dynamics Models”,
P-1403, presented at 2012 International System Dynamics Conference, St. Gallen,
Switzerland, July 2012.
2. Blaskovich, F. T., “Energy Policy Using System Dynamics”, SPE 159195-MS,
Asia-Pacific Oil and Gas Conference and Exhibition, Perth, Australia, October
2012.
3. Blaskovich, F. T., “Energy Policy Analysis and Uncertainty”, SPE 165371-MS,
SPE Western Regional & AAPG Pacific Section Meeting, Monterey, California,
April 2013.
4. Bankes, S. C., “Exploratory Modeling for Policy Analysis”, Operations Research,
Vol. 41, No. 3, 1993.
5. Lempert, R. J., Groves, D. G., Popper, S. W., and Bankes, S. C., “A General
Analytic Method for Generating Robust Strategies and Narrative Scenarios”,
Management Science, Vol. 52, No. 4, April 2006.
6. Lempert, R. J., Popper, S. W., and Bankes, S. C., “Shaping the next one hundred
years: New methods for quantitative, longer-term policy analysis”, RAND MR-
1626-RPC, 2003.
7. Lempert, R. J., Popper, S. W., and Bankes, S. C., “Confronting Surprise”, Social
Science Computing Review, Vol. 20, No. 4, Winter 2002.
8. http://www.vensim.com is the main website for the system dynamics software
used in this study.
9. http://www.r-project.org is the main website for the R Project for Statistical
Computing.
10. Friedman, J. and Fisher, N., “Bump Hunting in High-Dimensional Data”,
Statistics and Computing, Vol. 9, No. 2, 1999.
