## Abstract

We build a workable game of common-property resource extraction under rational Bayesian learning about the reproduction prospects of a resource. We focus on Markov-perfect strategies under truthful revelation of beliefs. For reasonable initial conditions, exogenously shifting the prior beliefs of one player toward more pessimism about the potential of natural resources to reproduce can create anti-conservation incentives. The single player whose beliefs have been shifted toward more pessimism exhibits higher exploitation rate than before. In response, all other players reduce their exploitation rates in order to conserve the resource. However, the overall conservation incentive is weak, making the aggregate exploitation rate higher than before the pessimistic shift in beliefs of that single player. Due to this weakness in strategic conservation responses, if the number of players is relatively small, then in cases with common priors, jointly shifting all players’ beliefs toward more pessimism exacerbates the commons problem.

This is a preview of subscription content, access via your institution.

## Notes

Debates about acid rain (see, for example, [16, pp. 178–181]), or about a large-scale biodiversity deterioration ([16, pp. 249–257]) are related to shifts in fundamentals regarding the natural ability of resources to reproduce. While environmental changes are not disputed among experts as facts, the magnitude of such changes is a vivid topic of disagreement among experts. For example, the article by Rörsch et al. [20] documents the opposition against [16] book by a part of the scientific community. What we keep from this debate among experts for the purposes of this study is that investor uncertainty about environmental fundamentals is a plausible working hypothesis for analyzing investments in natural resources.

Despite that we analyze a specific example, the stochastic structure of our Bayesian-learning setup is general, i.e., it is not restricted by conjugate-priors assumptions, as it will be clearer in the model’s outline below. This generality allows for a wide range of applications, extending even to the analysis of jump processes. This versatility of the example’s stochastic structure may stimulate further theoretical and empirical investigation, not only because of the theoretical ideas which are demonstrated here, but also because the example may be fitted to actual data itself. Here, we restrict our analysis to demonstrating the example’s mechanics theoretically.

For example, collecting and processing data on each winter’s temperature inform scientists and investors in natural-resource markets about the validity of their prior beliefs regarding a structural break due to “global warming.”

This heterogeneity in prior beliefs is justified by differences in non-informative priors, which are priors formed before any data are available. Ad-hoc differences in non-informative priors can persist and can play a crucial role in the transitional dynamics of a game. For an introduction to non-informative priors and references on alternative criteria for choosing non-informative priors see, for example, [7, pp. 61–66].

Focusing on Markov-perfect-Nash-equilibrium strategies may seem restrictive. Yet, even this simplified equilibrium concept involves considerable technical complexities. It was several years after the seminal example by Levhari and Mirman [15] that [21] proved equilibrium existence for the more general deterministic version. Dutta and Sundaram [6] were the first to extend analysis to the stochastic version of the game under rational expectations, while Amir [2] contributed a remarkably general characterization of such stochastic games offering existence and uniqueness results. For a general survey of natural-resource games see [17].

Kalai and Lehrer [11] show a key result related to our rational-learning formulation. When Bayesian updating is envisaged by each player, then Bayesian updating of collected information will lead in the long run to accurate prediction of the future play of the game and rational expectations as a limit of behavior with probability one, as time goes to infinity. This consideration on the side of players, which learning will be completed in the long run, and its impact on strategies, is a key distinctive feature of rational learning from other forms of learning. (The way to make players envisage Bayesian updating in our setting that focuses on Markov-perfect Nash strategies, is to incorporate Bayes’ rule in the Bellman equation of each player.) In their survey paper, [4] explain this distinction as well.

A recent paper that also combines strategic interaction extending the [15] model with rational Bayesian learning in a similar fashion to [13] is [19]. A key difference of the present paper from [19] is that here heterogeneous beliefs are allowed, and the impact of belief changes of one player can be characterized using a general stochastic structure for learning.

The proof regarding the solution of the linear system of multiple Nash-equilibrium necessary conditions, which appears as Lemma 3 in the Appendix of this paper, and which relies on the “ matrix determinant lemma,” first appeared in Koulovatianos [12, pp. 27–29] and enabled the identification of the solution in both [1, 12].

We use a player’s index \(i\in \left\{ 1,\ldots ,N\right\} \) as subscript for variables and as superscripts for functions, and we drop it whenever it is redundant.

While Proposition 1 holds for cases in which \(\theta \) is a multidimensional vector, the comparisons performed in this section rely upon the concept of stochastic dominance. Avoiding the technicalities of multidimensional stochastic dominance is the main reason for assuming that \(\theta \) is a single parameter in this section.

For the definition of strict FOSD, see, for example, [10, p. 6].

Propositions 1 and 2 can facilitate such an analysis under heterogeneous beliefs. Yet, belief heterogeneity makes it difficult to obtain intuitive generalizations about how beliefs affect the intensity of the commons problem. Under heterogeneous beliefs, one would need to compare equilibrium extraction rates with a social-planner’s solution under heterogeneous beliefs. This cumbersome task is beyond the scope of this paper.

The commons problem is straightforward to verify from Eq. (19). Increasing \(N\), i.e., the number of players with the same priors, always leads to higher aggregate exploitation rate, \(G\left( N,c^{o}\left( \xi \right) \right) \).

Recall from our analysis above that such a shift in priors toward pessimism is captured by \(\underline{\xi }\prec _{\mathrm{{FOSD}}}\xi \), in which \(\xi \) is the initial priors, and \(\underline{\xi }\) is beliefs after the shift. The result is \(c^{o}\left( \underline{\xi }\right) >c^{o}\left( \xi \right) \).

Notice that, according to Eq. (19), each player’s individual exploitation rate is,

$$\begin{aligned} \frac{G\left( N,c^{o}\left( \xi \right) \right) }{N}=\frac{c^{o}\left( \xi \right) }{1+\left( N-1\right) c^{o}\left( \xi \right) }, \end{aligned}$$which is increasing in \(c^{o}\left( \xi \right) \).

This is verifiable from Eq. (19), which implies that \(\partial ^{2}\left[ G\left( N,c^{o}\left( \xi \right) \right) /N\right] /\left[ \partial N\partial c^{o}\left( \xi \right) \right] <0\).

This proof first appeared in [12, pp. 27–29], which is an early version of this paper.

I am indebted to an anonymous referee and the Editor for motivating me to provide the argument that leads to the condition given by (11).

## References

Agbo M (2013) Strategic exploitation with learning and heterogeneous beliefs, SSRN working paper, No 1973793

Amir R (1996) Continuous stochastic games of capital accumulation with convex transitions. Games Econ Behav 15:111–131

Antoniadou E, Koulovatianos C, Mirman LJ (2013) Strategic exploitation of a common-property resource under uncertainty. J Environ Econ Manag 65:28–39

Blume L, Easley D (1993) Rational expectations and rational learning, economics working paper archive (EconWPA), series: game theory and information, no. 9307003. http://129.3.20.41/eps/game/papers/9307/9307003.pdf

Cogley T, Sargent T (2008) Anticipated utility and rational expectations as approximations of bayesian decision making. Int Econ Rev 49:185–221

Dutta PK, Sundaram RK (1992) Markovian equilibrium in a class of stochastic games: existence theorems for discounted and undiscounted models. Econ Theory 2:197–214

Gelman A, Carlin JB, Stern HS, Rubin DB (2004) Bayesian data analysis, 2nd edn. Chapman & Hall, London

Guidolin M, Timmermann A (2007) Properties of equilibrium asset prices under alternative learning schemes. J Econ Dyn Control 31:161–217

Harville DA (1997) Matrix algebra from a statistician’s perspective. Springer, New York

Jackson MO, Rogers BW (2007) Relating network structure to diffusion properties through stochastic dominance. B E J Theor Econ 7(1):1–13. http://www.bepress.com/bejte/vol7/iss1/art6

Kalai E, Lehrer E (1993) Rational learning leads to nash equilibrium. Econometrica 61:1019–1045

Koulovatianos C (2010) A paradox of environmental awareness campaigns. University Library of Munich, MPRA Paper 27260

Koulovatianos C, Mirman LJ, Santugini M (2009) Optimal growth and uncertainty: learning. J Econ Theory 144:280–295

Koulovatianos C, Wieland V (2011) Asset pricing under rational learning about rare disasters. Centre for Economic Policy Research (CEPR) discussion paper no. 8514.

Levhari D, Mirman LJ (1980) The great fish war: an example using a dynamic Cournot–Nash solution. Bell J Econ 11:322–334

Lomborg B (2001) The skeptical environmentalist: measuring the real state of the world. Cambridge University Press, Cambridge

Long NV (2011) Dynamic games in the economics of natural resources: a survey. Dyn Games Appl 1:115–148

Mawhin J (2009) Bounded solutions: differential vs differene equations. Seventh Mississippi State: UAB conference on differential equations and computational simulations. Electron J Differ Equ Conf 17:159–170. http://ejde.math.txstate.edu or http://ejde.math.unt.edu

Mirman LJ, Santugini M (2013) Learning and technological progress in dynamic games. Dyn Games Appl 4(1):58–72

Rörsch A, Frello T, Soper R, De Lange A (2005) On the opposition against the book the skeptical environmentalist by B. Lomborg. J Inf Ethics 14:16–28

Sundaram R (1989) Perfect equilibrium in a class of symmetric games. J Econ Theory 47:153–177

## Acknowledgments

I thank the Editor, Georges Zaccour, and three anonymous referees, whose comments greatly improved this paper. I also thank Johannes Hörner, Ramon Marimon, Marc Santugini, and Daniel Seidmann for useful discussions and suggestions. An earlier version of this paper has been circulated in 2010, under the title “A Paradox of Environmental Awareness Campaigns” , University Library of Munich, MPRA Paper 27260.

## Author information

### Authors and Affiliations

### Corresponding author

## Appendix: Proofs

### Appendix: Proofs

The proof of Proposition 1 relies on Lemmata 1 through 3, which is separate results. We state and prove Lemmata 1 and 3 below, while the proof of Lemma 2 appears in the literature. So, we state Lemma 2 and provide an appropriate citation for its proof.

###
**Lemma 1**

For any function \(h:\mathbb {R}\rightarrow \mathbb {R}\) and any given prior distribution \(\xi _{0}\) , for all \(t\in \left\{ 1,2,\ldots \right\} \) and \(\left\{ \xi _{s}\right\} _{s=0}^{t-1}\) generated through

from any sequence \(\left\{ \eta _{s}\right\} _{s=0}^{t-1}\) of independent draws of the shock \(\eta \), the conditional expectation

assuming that appropriate integrability conditions hold in (24) for all \(t\in \left\{ 1,2,\ldots \right\} \).

###
*Proof of Lemma 1*

We express \(\xi _{t-1}\) as a function of \(\xi _{t-2}\), according to the Bayesian update of beliefs given by (23), and we substitute it into the LHS of (24),

i.e., \(\xi _{t-1}\) has been canceled from the expression. Continuing in this way up to period \(0\), the LHS of (24) becomes

and because \(\eta \)’s are independent over time, this last expression is equal to the result given by (24). \(\square \)

###
**Lemma 2**

(matrix determinant lemma) Let \(A\) be an \(N\times N\) nonsingular matrix, and \(x\),\(~y\) be any \(N\times 1\) vectors. Then,

###
*Proof of Lemma 2*

See [9, p. 416, Theorem 18.1.1 and Corollaries 18.1.2 and 18.1.3]. \(\square \)

###
**Lemma 3**

Let the \(N\times N\) linear system

in which,

with \(a_{i}\ne 1\) for all \(i\in \left\{ 1,\ldots ,N\right\} \) . Then, the solution to this system (denoted by a star) is unique with,

###
*Proof of Lemma 3*

Simple substitution of (26) into (25) proves the validity of the solution. The condition \(a_{i}\ne 1\) for all \(i\in \left\{ 1,\ldots ,N\right\} \) guarantees linear independence among all rows (and columns) of \(A\), which implies the nonsingularity of \(\mathbf {A}\), which, in turn, proves the uniqueness of (26), proving the Lemma. Yet, below we provide the steps that lead to identifying this solution, which may prove useful for characterizing similar results in other resource-extraction games with similar structure.^{Footnote 18}

Pick any \(i\in \left\{ 1,\ldots ,N\right\} \), and notice that (25) implies,

in which \(\mathbf {1}_{N}\) is an \(N\times 1\) vector of ones, and \(\mathbf {0}_{ \mathbf {1}_{i}}\) is an \(N\times 1\) vector of zeros, with the sole exception that its \(i\)-th element is equal to \(1\). The *matrix determinant lemma (Lemma 2)* implies that

Setting,

and combining (27) with (28), we obtain

Equation (29) implies that, in order to characterize \(x_{i}^{*}\) , we must first characterize \(\det \left( \mathbf {A}_{i}\right) \) and \(\det \left( \mathbf {A}\right) \). We start from characterizing \(\det \left( \mathbf {A}\right) \). Let

which implies that \(\tilde{\mathbf {A}}\) is a diagonal matrix. Denoting the \( i \)-th diagonal element of \(\tilde{\mathbf {A}}\) by \(\mathtt {diag}\left( \tilde{\mathbf {A}}\right) _{i}\), it is,

Applying again the *matrix determinant lemma (Lemma 2)*,

which implies

In order to characterize \(\det \left( \mathbf {A}_{i}\right) \), we use the definitions of \(\mathbf {A}_{i}\) and \(\tilde{\mathbf {A}}\), noticing that \( \mathbf {A}_{i}-\tilde{\mathbf {A}}=\) \(\mathbf {1}_{N}\cdot \left( \mathbf {1} _{N}^{T}-\mathbf {0}_{\mathbf {1}_{i}}^{T}\right) \), which implies,

in which \(\mathbf {1}_{\mathbf {0}_{i}}\) is an \(N\times 1\) vector of ones, with the sole exception that its \(i\)-th element is equal to \(0\). Combining (32) with the *matrix determinant lemma (Lemma 2)*,

which gives,

Equation (26) is derived after combining (29) with (33), (31), and (30). The formula given by Eq. (26) holds for all \(i\in \left\{ 1,\ldots ,N\right\} \), since the choice of \(i\) was arbitrary. \(\square \)

###
*Proof of Proposition 1*

Our solution approach follows Levhari and Mirman (1980). We start from deriving RLMPNE in the finite-horizon setting. Then we use the finite-horizon RLMPNE results in order to generalize them to the infinite-horizon case. The approach of Levhari and Mirman (1980) for proving the result helps in exhibiting the informational structure of the problem.

*The static problem (0-period-horizon problem)*

The Nash-equilibrium solution is not unique in this case, but without loss of generality we can set,

in which \(\kappa _{i}^{\left( 0\right) }\) is some constant with \(\kappa _{i}^{\left( 0\right) }\in \left[ 0,1\right] \) for all \(i\in \left\{ 1,\ldots ,N\right\} \), with \(\Sigma _{i=1}^{N}\kappa _{i}^{\left( 0\right) }=1\). (We are solving this problem recursively, so we denote the \(n\)-th iteration by a superscript “ \(\left( n\right) \)” wherever this is applicable.) In order to keep each player’s problem well defined in next iteration, so as to comply with the requirement that value functions are well defined (notice that logarithmic utility implies that zero consumption in one period implies a value function equal to minus–infinity, and zero-consumption choices prevail in a Markov-perfect solution) we focus on a solution with \(\kappa _{i}^{\left( 0\right) }\in \left( 0,1\right) \) for all \(i\in \left\{ 1,\ldots ,N\right\} \), and with \( \Sigma _{i=1}^{N}\kappa _{i}^{\left( 0\right) }=1\). So, the value function of the agent in the static problem is,

which is the only case with the value function not depending on \(\Xi _{0}\).

*The 1-period-horizon problem*

The decision of player \(i\) is determined by the Bellman equation,

so, using (35),

with first-order condition,

which implies that, for all \(i\in \left\{ 1,\ldots ,N\right\} \), \(C^{i,\left( 1\right) }\left( k;\Xi _{0}\right) \) is of the multiplicatively separable form,

in which \(\left\{ c^{i,\left( 1\right) }\left( \Xi _{0}\right) \right\} _{i=1}^{N}\) is the unique solution to the linear system,

with

Lemma 3 applies to the linear system given by (38), and it implies that (38) has a unique solution with \(c^{i,\left( 1\right) }\left( \Xi _{0}\right) \in \left( 0,1\right) \), and \(\Sigma _{i}c^{i,\left( 1\right) }\left( \Xi _{0}\right) \in \left( 0,1\right) \). Substituting \( \left\{ C^{i,\left( 1\right) }\left( \Xi _{0}\right) \right\} _{i=1}^{N}\) of the multiplicatively separable form given by (37) into (36 ), the Bellman equation, leads to a value function of the form,

for all \(i\in \left\{ 1,\ldots ,N\right\} \), in which \(\kappa ^{i,\left( 1\right) }\left( \Xi _{0}\right) \) is a constant that does not affect optimization in future steps. Unlike \(V^{i,\left( 0\right) }\left( k;\Xi _{0}\right) \), the value function \(V^{i,\left( 1\right) }\left( k;\Xi _{0}\right) \) depends on \(\Xi _{0}\). Yet, we have an explicit form regarding the way \(V^{i,\left( 1\right) }\left( k;\Xi _{0}\right) \) depends on \(\Xi _{0}\). Most interestingly, in Eq. (40) the coefficient of \(\ln \left( k\right) \) depends on \(\xi _{0}^{i}\) only and not on the beliefs of other individuals.

*The 2-period-horizon problem*

The decision of player \(i\) is now determined by the Bellman equation,

so, using (40),

subject to,

What is crucial to observe here is the notation about the timing of shocks. In the problem expressed by (41), each player is deciding upon a strategy in period \(0\), expecting both a shock \(\eta _{0}\) in period \(0\), after the decision has been made, and a shock \(\eta _{1}\) in period 1. Yet, it is the shock \(\eta _{0}\) which will determine how the prior distribution \( \xi _{0}^{i}\) will evolve to \(\xi _{1}^{i}\), which is an element that the analytic form of (41) allows us to see explicitly. So, with the time horizon being expanded, we can see how prior beliefs determine what type of information is expected to arrive and also how this information is expected to be exploited.

To simplify notation, we can re-write (41) as,

with

as in equation (39) above, and,

in which \(\xi _{1}^{i}\left( \theta _{1}|\eta _{0}\right) \) is given from ( 42). The first-order conditions of (43) are given by,

which implies that \(C^{i,\left( 1\right) }\left( k;\Xi _{0}\right) \) is of the multiplicatively separable form,

for all \(i\in \left\{ 1,\ldots ,N\right\} \), where \(\left\{ c^{i,\left( 2\right) }\left( \Xi _{0}\right) \right\} _{i=1}^{N}\) is the unique solution to the linear system,

in which

Again, Lemma 3 guarantees that (45) has a unique solution with \( c^{i,\left( 2\right) }\left( \Xi _{0}\right) \in \left( 0,1\right) \), and \( \Sigma _{i}c^{i,\left( 2\right) }\left( \Xi _{0}\right) \in \left( 0,1\right) \), while substitution of \(\left\{ C^{i,\left( 2\right) }\left( \Xi _{0}\right) \right\} _{i=1}^{N}\) as given by (44) into the Bellman equation given by (43) gives a value function of the form,

where \(\kappa ^{i,\left( 2\right) }\left( \Xi _{0}\right) \) is a constant that does not affect optimization in any future step. At this point we have seen enough of the problem’s structure to be able to deduce the formulas of the \(n\)-period horizon problem.

*The *
\(n\)
*-period-horizon problem*

The strategy of player \(i\) is determined by the Bellman equation,

with \(V^{i,\left( n\right) }\left( k;\Xi _{0}\right) \) being of the form,

in which \(\kappa ^{i,\left( n\right) }\left( \Xi _{0}\right) \) is a constant, and

Moreover, players’ strategies are of the form

in which \(\left\{ c^{i,\left( n\right) }\left( \Xi _{0}\right) \right\} _{i=1}^{N}\) is the unique solution to the linear system,

with

To calculate \(E_{0}\left( \eta _{t}|\xi _{0}^{i}\right) \) we rely on Lemma 1. From Eq. (24) of Lemma 1, after setting \(h\left( \eta \right) =\eta \), the identity function, we obtain,

and from (6) it is,

Substituting (49) into (46) we obtain,

*The infinite-horizon problem*

Notice that since \(\mathcal {H}\subseteq \left( 0,1\right) \), \(\mu \left( \theta \right) \in \left( 0,1\right) \) for all \(\theta \in \Theta \). After taking the limit when \(n\rightarrow \infty \), (50) gives,

so, \(\mu \left( \theta \right) \in \left( 0,1\right) \) for all \(\theta \in \Theta \) guarantees that \(V^{i}\left( k;\Xi \right) \) is well defined. Since \(\xi _{0}^{i}\) is a density function, \(\int _{\Theta }\xi _{0}^{i}\left( \theta \right) \)d\(\theta =1\), which leads to,

(Subscript “ \(0\)” of \(\Xi _{0}\) has appeared in order to remind that \(\Xi _{0}\) denotes prior beliefs in period \( 0\). In the infinite-horizon setup this timing does not matter any more, so subscript “ \(0\)” can be dropped. So, we drop it throughout the rest of the proof.) Moreover, the solution is again of the multiplicatively separable form

and (48) is generalized to,

in which

Since \(\mu \left( \theta \right) \in \left( 0,1\right) \) for all \(\theta \in \Theta \),

Inequality (52) guarantees that Lemma 3 applies, which implies that the unique solution to (51) is given by Eq. (9).

Regarding the transversality condition given by (3), notice that according to (4) and (9),

Combining the expression on the left-hand side of (3) with (53) gives,

proving that the transversality condition given by (3) indeed holds.

In order to guarantee that \(V^{i}\left( k;\Xi \right) \) is well defined, it remains to verify condition (11), which can prove important for identifying conditions on \(\phi \left( \cdot \left| \eta \right. \right) \) and \(\Xi _{0}\) that guarantee the boundedness of \(\kappa ^{i,\left( \infty \right) }\left( \Xi _{0}\right) \) in applications using specific functional forms.^{Footnote 19} Specifically, let us take a guess on the functional form of the value function of player \(i\in \left\{ 1,\ldots ,N\right\} \),

in which \(\kappa ^{i}\left( \Xi \right) =\kappa ^{i,\left( \infty \right) }\left( \Xi _{0}\right) \) is an unknown functional and \(f\) is given by

which is the same for all players \(i\in \left\{ 1,\ldots ,N\right\} \), since it is independent from specific beliefs \(\xi ^{i}\). Substituting Eq. (54) into the Bellman equation (5) gives,

which becomes,

First-order conditions based on (56) give,

so,

and if we make the additional guess that the strategies of all other players are of the multiplicatively separable form \(C^{j}\left( k;\Xi \right) =c^{j}\left( \Xi \right) \cdot k\), then (57) reconfirms that player \(i\) is also of multiplicatively separable form, namely,

Substituting (58) into (56) we identify terms that are multiplicatively separable expressions of \(\ln \left( k\right) \) and terms that do not depend on the state variable \(k\). Starting from the terms that do not depend on \(k\), and which do not affect optimization, these should satisfy the recursion,

Isolating all terms that are multiplicatively separable expressions of \(\ln \left( k\right) \) these must satisfy,

which is an expression that complies with function \(f\left( \theta \right) ,\) given by (55), verifying the validity of (10). Substituting the expression given by (9) into (59), verifies Eqs. (11) and (12), proving the proposition. \(\square \)

## Rights and permissions

## About this article

### Cite this article

Koulovatianos, C. Strategic Exploitation of a Common-Property Resource Under Rational Learning About its Reproduction.
*Dyn Games Appl* **5**, 94–119 (2015). https://doi.org/10.1007/s13235-014-0113-3

Published:

Issue Date:

DOI: https://doi.org/10.1007/s13235-014-0113-3

### Keywords

- Renewable resources
- Resource exploitation
- Non-cooperative dynamic games
- Bayesian learning
- Stochastic games
- Commons
- Rational learning
- Uncertainty
- Beliefs