Abstract
We study the impact of discrete versus continuous time on the behavior of agents in the context of a dynamic common pool resource game. To this purpose, we consider a linear quadratic model and conduct a lab experiment in which agents exploit a renewable resource with an infinite horizon. We use a differential game for continuous time and derive its discrete time approximation. In the single agent setting, we fail to detect, on a battery of indicators, any difference between agents’ behavior in discrete and continuous time. Conversely, in the two-player setting, significantly more agents can be classified as myopic and end up with a low resource level in discrete time. Continuous time seems to allow for better cooperation and thus greater sustainability of the resource than does discrete time.
Similar content being viewed by others
Notes
Note also that some authors such as Noussair et al. (2015) conduct their experiments in discrete time, while their theoretical model is in continuous time, which poses the question of to what theoretical predictions should we compare lab results: those from discrete or those from continuous time models? Moreover, Tasneem et al. (2019) study the ability of a single economic agent to exploit a renewable resource efficiently. To do that they test in the laboratory an optimal control problem with an infinite horizon in continuous time and show that extraction behavior results in a steady state of the resource only 56% of the time.
Battaglini et al. (2016) define dynamic free-riding this way: “an increase in current investment by one agent [which] typically triggers a reduction in future investment by all agents". In the context of a CPR, a decrease in extraction level can be seen as an investment to obtain a higher resource level, and thus greater benefits in the future.
One exception to this way of implementing continuous time in the laboratory is Calford and Oprea (2017). The authors propose a protocol where time freezes when one decision is taken in order to let the other player to react to this decision without delay in the game. This protocol is useful and easy to implement for timing games like that studied by Calford and Oprea (2017) but less appropriate for CPR games as we explain in Sect. 3.
Since the continuous time condition involves higher network traffic, we limited the number of participants per session to a maximum of 14, which explains the greater number of sessions for this treatment.
ORSEE (Greiner 2015) is the platform used by the LEEM to manage the subject pool.
ECU means Experimental Currency Unit.
The return flow coefficient is the quantity of water returning to the groundwater after each extraction.
In the sole agent continuous time treatment, subjects were able to change their extraction rate at any moment by simply moving the graduated slider displayed on their computer. Every second, the computer transmitted the slider value to the server, which then performed the computations (resource and payoff) and updated the values displayed on the computer’s graph and text interfaces.
In the two-player continuous-time treatment, player 2’s computer sent the cursor value to the server as soon as it changed, while player 1’s computer transmitted the cursor value to the server every second, triggering the server to continuously broadcast the updated values to both players. Thus, every second, the server took player 1’s current extraction and player 2’s most recent extraction (i.e., the last one transmitted by his computer). In this way, time was synchronized between the two members of the group, since only one player was triggering the continuous updating of the information. This also reduced network traffic because as long as the second player did not change his extraction, his computer did not transmit a new value.
Noussair and Matheny (2000) and Brown et al. (2011) show that behavior is not significantly different under random termination or continuation payoff in single-agent cases. Moreover, with a random ending in a two-player game, the players may have different beliefs about the last period or instant. Continuation payoff avoids this problem.
Notice that while discounting allows us to implement the continuation payoff here, it has limited impact on the payoffs that are accumulated within the 10 minutes of the game. Given our parametrization, the optimal extraction rate when R = 20 is equal to 0.56. At t = 18 (first instant/period that R = 20 with the optimal extraction path) it generates a payoff of 1.02 ECU while at t = 60 (the last instant/period) it generates a payoff of 0.82 ECU, a gap of only 20%.
To take a concrete example, instead of comparing the agent’s extraction w(t) to the conditional constrained myopic and conditional optimal extraction, \(w(t)_{my}^{c}\) and \(w(t)_{op}^{c}\), we could compare it to the temperature in Moscow and Istanbul from day 1 to day 600, and we would find that our agent’s extraction is closer to the temperature either in Moscow or in Istanbul, because one MSD will always be smaller than the other, even if completely irrelevant.
An alternative is proposed by Suter et al. (2012), who run a similar regression (without the constant term) and consider that an agent follows a given behavior if the coefficient is not significantly different from 1. A natural way to do this is to implement a Wald test with:
$$\begin{aligned} \left\{ \begin{array}{ll} H_{0} : \beta _{1} = 1\\ H_{A} : \beta _{1}\ne 1, \end{array} \right. \; and\;\; W = \frac{(\hat{\beta _{1}}-1)^2}{var(\hat{\beta _{1}})} \rightarrow F_{(1,300)} \end{aligned}$$In this case, a very imprecisely estimated coefficient \(\beta _{1}\) (very large \(var(\hat{\beta _{1}})\)) will lead us to reject \(H_{A}\) and classify the agent as myopic or optimal, although he follows neither an optimal or myopic path. This is why we propose the aforementioned alternative rule for classification.
We present regression results using 1 lags. Results using 5 and 10 lags are available upon request.
By "theoretical one", we mean the resource level at the end of the experimental stage if an agent played perfectly optimally or myopically during the whole experimental stage.
A more precise comparison of the results is not possible since the authors use a different empirical strategy.
The maximum group payoff is 240 ECUs, so we compute the individual efficiency ratio by halving this value. Nevertheless, it is possible to get "more than your own share". Obviously, if one of the two members of the pair extracts a very small amount of groundwater, the other member can obtain more than \(50\%\) of the total maximum payoff.
To make continuous and discrete time comparable, we take the difference between two decisions separated by ten seconds in continuous time.
In Fig. 7a we also see an increase in stability over time for both treatments. Note, however, that the greater instability in the beginning of the play time can be explained by the game setting. Indeed, players need first to either let the resource grow or deplete it before reaching a steady state, depending on their preferred equilibrium.
To make continuous and discrete time comparable, we use only one decision every ten seconds in continuous time.
The c.d.f. are statistically different according to the Kolmogorov–Smirnov test (p-value \(<0.05\)).
The average difference in extraction between players of the same group at the end of the game equals 0.18, while the average player’s extraction level equals 0.27.
Concentration (Gini) indexes are significantly different whether we use the standard, Erreygers or Wagstaff indexes (O’Donnell et al. 2016).
References
Battaglini M, Nunnari S, Palfrey TR (2016) The dynamic free rider problem: a laboratory study. Am Econ J Microecon 8(4):268–308
Bigoni M, Casari M, Skrzypacz A, Spagnolo G (2015) Time horizon and cooperation in continuous time. Econometrica 83(2):587–616
Bó PD (2005) Cooperation under the shadow of the future: experimental evidence from infinitely repeated games. Am Econ Rev 95(5):1591–1604
Brown M, Flinn CJ, Schotter A (2011) Real-time search in the laboratory and the market. Am Econ Rev 101(2):948–74
Calford E, Oprea R (2017) Continuity, inertia, and strategic uncertainty: a test of the theory of continuous time games. Econometrica 85(3):915–935
Chen DL, Schonger M, Wickens C (2016) oTree-an open-source platform for laboratory, online, and field experiments. J Behav Exp Financ 9:88–97
Fréchette GR, Yuksel S (2017) Infinitely repeated games in the laboratory: Four perspectives on discounting and random termination. Exp Econ 20(2):279–308
Friedman D, Oprea R (2012) A continuous dilemma. Am Econ Rev 102(1):337–63
Gardner R, Moore MR, Walker JM (1997) Governing a groundwater commons: a strategic and laboratory analysis of western water law. Econ Inq 35(2):218–234
Gisser M, Sanchez DA (1980) Competition versus optimal control in groundwater pumping. Water Resour Res 16(4):638–642
Greiner B (2015) Subject pool recruitment procedures: Organizing experiments with orsee. J Econ Sci Assoc 1:114–125
Herr A, Gardner R, Walker JM (1997) An experimental study of time-independent and time-dependent externalities in the commons. Games Econom Behav 19(1):77–96
Hey JD, Neugebauer T, Sadrieh A (2009) An experimental analysis of optimal renewable resource management: the fishery. Environ Resource Econ 44(2):263
Janssen MA, Holahan R, Lee A, Ostrom E (2010) Lab experiments for the study of social-ecological systems. Science 328(5978):613–617
Leng A, Friesen L, Kalayci K, Man P (2018) A minimum effort coordination game experiment in continuous time. Exp Econ 21(3):549–572
Mason CF, Phillips OR (1997) Mitigating the tragedy of the commons through cooperation: an experimental evaluation. J Environ Econ Manag 34(2):148–172
Negri DH (1989) The common property aquifer as a differential game. Water Resour Res 25(1):9–15
Noussair C, Matheny K (2000) An experimental study of decisions in dynamic optimization problems. Econ Theor 15(2):389–419
Noussair CN, van Soest D, Stoop J (2015) Cooperation in a dynamic fishing game: a framed field experiment. Am Econ Rev 105(5):408–13
O’Donnell O, O’Neill S, Van Ourti T, Walsh B (2016) Conindex: estimation of concentration indices. Stand Genomic Sci 16(1):112–138
Oprea R, Charness G, Friedman D (2014) Continuous time and communication in a public-goods experiment. J Econ Behav Organ 108:212–223
Rand DG, Greene JD, Nowak MA (2012) Spontaneous giving and calculated greed. Nature 489(7416):427–430
Rubio SJ, Casino B (2003) Strategic behavior and efficiency in the common property extraction of groundwater. Environ Resource Econ 26(1):73–87
Suter JF, Duke JM, Messer KD, Michael HA (2012) Behavior in a spatially explicit groundwater resource: Evidence from the lab. Am J Agr Econ 94(5):1094–1112
Tasneem D, Engle-Warnick J, Benchekroun H (2017) An experimental study of a common property renewable resource game in continuous time. J Econ Behav Organ 140:91–119
Tasneem D, Engle-Warnick J, Benchekroun H (2019) Sustainable management of renewable resources: an experimental investigation in continuous time. Appl Econ 51(35):3804–3833
Vespa E (2020) An experimental investigation of cooperation in the dynamic common pool game. Int Econ Rev 61(1):417–440
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The authors would like to thank Marc Willinger, Lisette Ibanez, and Nicolas Quérou for comments as well as the ANR GREEN-Econ [grant number: ANR-16-CE03-0005] for financial support and the Experimental Economic Laboratory of Montpellier for technical support. Corresponding Author: alexandre.sauquet@inrae.fr. “Online appendix” available at https://tinyurl.com/4y7shzuv.
Supplementary Information
Below is the link to the electronic supplementary material.
Appendices
Appendices
Appendix A: The Discretization of the Continuous Time Model
This section presents the procedure adopted to discretize the continuous time model. Let’s consider the following continuous time model:
For the discretization of the model above, let’s consider \(\tau\) as the discretization step and n as a period. Time is discretized into intervals of length \(\tau\), such that the differential equation and the payoff are approximated in each interval \(n\tau\), \((n+1)\tau\). Thus, the discretization of the objective function gives:
Using Taylor’s first order limited development of \(e^{-r\tau }\) gives :
Thus, the objective function becomes:
The discretization of the dynamics gives:
The discrete time problem can be defined as:
The discrete time model therefore converges towards the continuous time model when the discretization step \(\tau\) tends toward zero.
In order to see the degree of the approximations used in the experience, with the parameters chosen in the model, Fig. 10 shows the feedback trajectory in continuous time and in the discretizations (\(\tau = 0.1\) and \(\tau = 1\)).
Appendix B: Figures from Experimental Instructions
Rights and permissions
About this article
Cite this article
Djiguemde, M., Dubois, D., Sauquet, A. et al. Continuous Versus Discrete Time in Dynamic Common Pool Resource Game Experiments. Environ Resource Econ 82, 985–1014 (2022). https://doi.org/10.1007/s10640-022-00700-2
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10640-022-00700-2