Towards a resource-based economy: Modeling valuing systems and ownership functions in a simulated economy

(1)

Towards a resource-based economy: Modeling valuing

systems and ownership functions in a simulated economy

Max L. de Raad 10760970

Bachelor thesis Credits: 18 EC

Bachelor Opleiding Kunstmatige Intelligentie University of Amsterdam Faculty of Science Science Park 904 1098 XH Amsterdam Supervisor dhr. dr. R. (Roberto) Valenti Faculty of Science University of Amsterdam Science Park 904 1098 XH Amsterdam June 30th, 2017

(2)

CONTENTS CONTENTS

A resource-based economy is an economy which focuses on effective resource al-location to efficiently satisfy all individual needs. All resource are declared public and pricing is managed digitally and reward is based on contribution. In this re-search, these concepts were incorporated in a simulated digital economy. A pricing mechanism based on real-time demand was contrasted against a fixed price market economy, and the effects of revenue interest rates through ownership were contrasted against an economy with all resources declared public. Ownership was detrimental to the efficiency of the economy, and a on-demand pricing mechanisms was marginally better, especially in equality of resource distribution. Overall, the results were in favour of a resource-based economy, and the findings in this research give reason for experimenting electronic economic management in the real economy.

1 Introduction

A resource-based economy (hereinafter: RBE), proposes a new, sustainable economic sys-tem for exploiting the earth’s natural resources in a sustainable manner to satisfy all human demand and needs. In such a economy, a monetary valuing system is replaced by a system based on effective resource allocation, and rewards are be based on contribution rather than possession. The notion of property is replaced by ’access’, since there is no need to own a certain product, but simply to utilize it. The idea of a RBE has been incorperated in the Zeitgeist Movement (McLeish, Berkowitz, Joseph, & eds., 2014) and the Venus project (The Venus Project , 2017; Fresco & Meadows, 2002).

Such an economy could be accomplished using a digital network where valuing and allocation of resources could be managed with a supercomputer, as has been proposed in Project Cybersyn: an attempt to implement such a network in the Chilean economy (Medina, 2011). It could also be accomplished by using a decentralized network with a more recent technique such as blockchain, on which the Bitcoin is based. (Kosba, Miller, Shi, Wen, & Papamanthou, 2016). With this technique a network of computers would manage fair distribution of resources, without the need for a ’middle-man’ like a govern-mental institution or a bank. A popular blockchain designed for resource sharing and valuing is the Ethereum network (Ethereum, 2017).

Effective resource distribution is also relevant in peer-to-peer file sharing networks. Vishnumurthy, Chandrakumar, and Sirer (2003) have proposed a ’Karma’ framework to complement peer to peer networks for effective and fair resource sharing. They tackle

(4)

1 INTRODUCTION

problems of freeloaders and malicious attacks of the system in a peer-to-peer file sharing network by introducing a unique ’balance’ of ’karma’ for every user in the network. It works as a distributed, decentralized system, just like a blockchain. The framework gener-ally ensures a fair exchange of the ’karma’ when requesting or offering files. A model like this could also be used for efficient and fair resource distribution in an economy. Bauwens has written different articles arguing in favor of the effectiveness of a peer-to-peer pro-duction, trancending traditional industrialism and capitalism, and ways to integrate such production systems into the political system(2009, 2005).

To accomplish a resource based economy, a call for a new valuing system of resources thus arises. Resources should be distributed fairly among the population, while preventing over-consumption or an unequal distribution. This could be accomplished by a valuing system of tokens or ’karma points’, which can be earned based on an individual’s con-tribution to society. Prices of items will be based on demand, and working on the most demanded item will be the most rewarding. The purpose of this research is to investigate the effects of such constraints on the valuing system on the durability and equality in a simulated economy. This is captured in the follow research question:

In a simulated digital economy, what set of agent strategies, incentives and valuing system leads to a balanced and sustainable economy as emergent behavior?

The aim here is to simulate a multi-agent system where a stable economy emerges from the incentives of agents combined with a specific type of valuing system. The incen-tive of the agent could for example be to maximise earnings while minimising the time spent (work) to achieve this. The results should be a form of emergent behaviour, where there are no explicit rules instructing the agents what to do. The goal is to reach a stable wealth distribution on the long term, as this indicates the general well-being of all agents in the system and the durability of the economy, and to evaluate the effects of the valuing systems and incentives as emergent behaviours: the agents should exhibit the behaviour uninstructed and thus implicitly. This implies that there does not need to be any psy-chological incentives for agents to behave optimally, which in realistic situations might be impossible to realize for all people, but the rule-set of the economy constraints behaviour in a way that it conforms to the emergence of a sustainable society. Besides modeling a valuing system, this research will also focus on the effect of ownership of resources. Whereas in a RBE all resources are declared public, in a market economy (hereinafter: ME) resources can be owned leading to an increased earnings by the owner. This might also lead to higher welfare gaps between agents. Comparing these systems should prove the effectiveness of a RBE against a ME.

(5)

1 INTRODUCTION

Besides providing insight into system dynamics in an artificially intelligent environ-ment, this research can also provide insights into the scientific field of economics. Firstly, removing the mentioned notion of ownership, besides being conceptualized in the RBE, aligns with more recent, sharing-economy trends. Peerby (for resource exchange), Airbnb and Uber (although still using money as exchange) are examples of such open-source sharing-economy initiatives. Secondly, this research approaches the classic field of welfare economics, which is generally concerned with the maximization of the welfare of a society (De Scitovszky, 1941). It is also concerned with the efficient utilization of resources, like the simulation in the current research will. This research also offers a potential extension to earlier research into Cybersym by Haber and Valenti (2015). In this research, it was evaluated whether a digital economy could be self sustaining, and was found to be suc-cessful. This research adds to the findings by attempting to contrast a digital economy with a classic, market driven economy.

Firstly, in the following section, knowledge on simulation involving multiple agents will be elaborated from the field of artificial intelligence. This literature review is followed by a broad description of the methodology of this research: The conceptual design of the simulation will be described first, followed by the technical implementation, the choice of fixed and free parameters, and finally the methodology of evaluating the simulation performance. In the fourth section, the results will presented: Both graphs showing the development of welfare statistics in the simulation and numbers will be presented, evaluated, and analyzed. From the results on, in the section ’Conclusion and Discussion’, final remarks will be made on the results obtained, and the implications for science and society are discussed more broadly. In the final chapter ’Future work’, suggestions for both extensions of the current research or other relevant research are made.

(6)

2 LITERATURE REVIEW

2 Literature review

A simulation is using a computational model to gain insight into a (complex) system’s behaviour in a synthetic environment (Bandini, Manzoni, & Vizzari, 2009). In the case of a RBE, which has never been fully realized, neither in project Cybersyn, simulating a model of this economy is the only possibility of proving its effectiveness scientifically. There is no possibility of conducting inductive or deductive reasoning in this case. Simulation, however, has been described as a new, third way of conducting science (Goldspink et al., 2002).

2.1 Distributed Artificial Intelligence

To build a simulation of an economy where agents individually decide on the most effective strategy, knowledge on distributed artificial intelligence (DAI), a subfield of artificial intel-ligence, is required. This field of study, according to Kraus, Wilkenfeld, and Zlotkin(1995), ”is concerned with how automated agents can be designed to interact effectively”, and can be divided into two main classes: Distributed Problem Solving (DPS) and Multi-Agent Systems (MAS). DPS is mainly concerned with a system design where there is optimal cooperation between individual nodes or agents in a network. For this research, it is relevant in the sense that the economy should be designed optimally to stimulate cooper-ation. However, this is not the main goal of this research, because cooperation might not be required for a stable and durable economy by definition, and also cannot be assumed, as modeling behavioural components is out of the scope of this research. However, there is no explicit notion of cooperation in a MAS, and it can be seen as a more general form of DAI. Competition between agents is possible in this case as opposed to a DPS. Because of this a MAS can closely resemble a real economy, and will be mainly considered in this research. The concept of a MAS will be further elaborated on from the next paragraph on.

2.2 Simulating a Multi Agent System

For simulating a digital economy, as mentioned, designing a multi-agent system (MAS) is favorable. A MAS consists of three key parts: agents (in the case of this research: people in an economy), the environment (land with resources) and mechanisms of inter-actions (trading resources) (Bandini et al., 2009). There are 2 general types of MAS’s: multi-agent decision systems, where agents collectively make decisions, and multi-agent simulations, where a real-world domain is simulated (Luck, McBurney, Shehory, Will-mott, et al., 2005). Since the decisions of individual agents are not of key importance in

(7)

2.3 Multi Agent Resource Allocation 2 LITERATURE REVIEW

this research, but rather the emergence of behaviour, the focus will be on the latter type. More specifically, the field of multi-agent resource allocation (MARA) is interesting for an economy including resource distribution. Generally, MARA covers the distribution of items among agents in a simulated network. The concept is broadly discussed in a paper by Chevaleyre et al.(2006), and relevant concepts for the simulations in this research will be further elaborated in the following paragraph.

2.3 Multi Agent Resource Allocation

An example of a MARA system is a distribution of resources with assigned values. Social welfare in a simulated economy can be measured with a collective utility function, with the sum of the utilitarian welfare of the individual agents as the most common measure. This utilitarian welfare of a single agent could for example be the sum of the values of the resources an agent possesses. However, utilitarian welfare can also be measured as the utility of the agent in the network which is worst off, which might be important to consider to measure fairness of resource distribution in an economy. One agent might for example posses an high amount of resources thus making the sum of welfare of the system apparently high, while all other agents could still be poor. Both measurements of the system performance of the simulation will be considered in this research.

Using the knowledge of a MAS and MARA, there is a knowledge-basis for building a simulation. In the following section a model will be proposed and the implementation will be described.

(8)

3 METHODS

3 Methods

3.1 Conceptual design

3.1.1 People and consumption

The simulation will be a MAS in the form of a fictive economy where agents need to con-struct words from individual letters (hereinafter: resources) to have their needs satisfied and to be able to ’consume’ words (hereinafter: products). The individual words can be compared to real products like a meal where all individual letters are ingredients to construct this meal, so a ’pizza’ requires a p, an i, two z’s and an a. These words have to be ’written’ sequentially by the agents from a part of a book which is assigned to each of them. Writing a book is a representation of a person’s changing consumption pattern in a realistic economy. It includes a main similarity: unique individual needs. The scope of these needs is limited to 26 unique products, which are all letters from the alphabet. Like in a realistic situation, personal needs can vary day by day. Consumption and trade of other, more realistic products could be implemented in the system, but it would not make a significant difference to measure the performance of the system. Writing a book with its letters as resources is simply more abstract. Moreover, the main advantage of using a book is that it contains a natural form of resource distribution. Therefore, scarcity of prod-ucts does not need to be manually determined, which would make room for bias and error.

The agents can obtain resources by buying them from mines, and receive a starting fund at the start of the iteration to achieve this. Mines in this context represent factories, shops, or even services in a real economy. The mines and agents compose the ’agent’ component of a MAS in this simulation (Bandini et al., 2009). Besides agents buying resources from the mine, they can work on the mines to gain income from buyers. This is a form of retribution to the system, comparable to work in a realistic economy. If a letter is bought from a mine, the price paid will be distributed among all workers. In this way the funds will always circulate within the system, without any funds ’leaking’ out of the system or ’flowing’ in. Working at mines and buying for mines compose the mechanism of interaction in the MAS, and the environment will simply be represented as a grid where all agents are placed and actions take place. The following figure illustrates the conceptual design of the simulated economy:

(9)

3.1 Conceptual design 3 METHODS Ball Next word 1… Next word 2… Cab Next word 1… Next word 2… Call Next word 1… Next word 2…

b, a

b, a, l

l, l

a

b

c

l

Inventory Wish Agent Requests Mine

Figure 1: The conceptual design of the simulated economy. All agents have an assigned product list (from a book), from which they will construct each product sequentially (wishes). They buy individual letters from mines, and these letters will be stored in their inventory, until they can construct the word.

3.1.2 Pricing and ownership

In a classic market economy the price of the letters is determined by demand and offer as it would be on a macro-level by external factors as scarcity in the current market economy (ME). However, the market prices don’t change very often and are fixed for a substantial period of time. For this reason, the prices of letters in the ME simulation are fixed based on general letter frequency in the book. In a resource based economy (RBE), instead, the price is adjusted on a micro-scale, and changes real-time by demand. This is an important concept in a RBE: resources should be produced locally and on-demand. In a real economy, this demand could be tracked by using a blockchain system, which adjusts prices accordingly.

The second parameter which will be alternated is the notion of ownership. In the ME the closest agent to mine will be the owner, as in a ME resources could be possessed by predisposition or heritage. The owner will receive a part of the income achieved by other agents from working at the mine. This parameter, in a simplified form, can capture the effects of hierarchy, randomness and predisposition which might induce wealth gaps, unequal distribution of resources or poverty, as in a real ME.

(10)

3.2 Implementation 3 METHODS

Working hours, hierarchy and other factors which might influence income or produc-tivity are mostly left out in this simulation to reduce the parameterization. This will result in a isolated, closed environment simulation. For the implementation of more ad-vanced models it should be relatively easy to add more parameters, but reducing (partially random) parameters helps with a clear measurement of the system performance.

This simulation is a ’black box’ approach to test the research based economy. Since the parameters are tweaked but the outcome is unknown, the behaviour can be considered ’emergent’, as is stressed in the research question. In this research, both the choice of ei-ther a RBE or ME and the notion of ownership or the absence of this notion are considered binary variables. This leads to a total of four combinations of both variables. However, a notion of a RBE in combination with ownership is arguably not not compatible, as a RBE implies that all resources are declared public. Because of this, the three remaining pairs will be tested: the resource based economy, the market economy non-ownership (hereinafter: ME-N), and the market economy with ownership (hereinafter: ME-O). The interest ratio (for the owner of a mine) in the ME-O can however also be considered a continuous variable, which will also be tweaked. This will be described in more detail in the evaluation subsection.

3.2 Implementation

To build the simulation, the Repast Simphony (Repast, 2016) suite is chosen. The soft-ware is implemented in- and supporting Java, which the author is familiar with. It is a customization of the Eclipse Java IDE and is fully modular. It also offers features like a data sink, real time data graps and a visual representation. Besides, the Repast Simphony software is also used the previous Cybersym research (Haber & Valenti, 2015), and the classes are available on Github to be reused. Also, built-in Java classes of the Simphony Suite all other Java packages are accessible.

In the Java project for the current simulation, there are classes for ’agents’, ’products’ and ’mines’. Each class has a constructor which assigns basic properties to each object. The amount of objects for each class in the simulation is variable: the amount of agents and mines in the simulation can be indicated. All mines and agents are placed on a ras-terized grid in the environment. The ’working’ network shows at what mine agents are currently working. The following figure shows the simulation in Repast Simphony:

(11)

Figure 2: The simulation in Repast Simphony. Blue circles represent agents, which are labeled with their number and balance. Red triangles represent Mines, tagged with their letter, price, and amount of workers on the mine.

3.2.1 Initiation of simulation

To initiate the simulation, the environment with a grid is created and the desired amount of agents and mines is placed on the grid. Next, a book1 (as a .txt file) is read in and parts of the book are assigned to the different agents. Using a specified limit batches of the words of the book will be assigned to the agents. If, for example, this limit is set to 1000, the first agent will receive the first 1000 words, the second agent will receive words 1001-2000, and so on.

1

(12)

In the case that ownership is an enabled parameter, for each mine the distance to all agents will be measured, and the closest agent will become the owner of the mine. Next, if the ME is enabled, all letters will be counted and the count of each letter will be divided by the total amount of letters. This leads to a percentage distribution of all letters in the book. This directly defines the fixed price for the simulation. In a RBE the price will be updated on every tick, as will be described in the ’Mine’ section. The following graph will illustrate the letter distribution for the book:

7,94 1,58 2,66 4,27 12,24 2,04 1,90 6,36 7,10 0,150,80 3,88 2,85 6,71 8,19 1,72 0,11 5,52 6,40 8,81 3,09 0,95 2,39 0,17 2,13 0,04 0 2 4 6 8 10 12 14 a b c d e f g h i j k l m n o p q r s t u v w x y z Fr equency (% )

Occurence of letters in book

Figure 3: The letter distribution in the book ’The Adventures of Sherlock Holmes’ by Sir Arthur Conan Doyle. These numbers divided by 100 will also make up the price for each resource in the ME.

From the initiation on, the agents will execute actions with every ’tick’, an iteration in the simulation. Mines will adjust parameters accordingly. The following subsections will elaborate on the behaviour of Agents and Mines.

3.2.2 Agent

On every tick, an agent will start by updating its parameters in updateParameters(). In case that he constructs and consumes a product in the previous tick, he will update its wishes to the next word required, with its individual required letters. Then it will try to complete a product (word) if he has all required letters in consumeProduct(), and it’s turn for the tick is over. If this is not possible for the agent, the agent can try to buy a required resource (letter) from a mine, in buyResourcefromMine(). Here, the agent iteratively browses through all mines, finds a mine which has at least 1 worker (which is a condition to extract from a mine), he can afford the letter from, and then chooses

(13)

inventory.

Also, on every tick, the agent will determine the best mine to work in

determineBestMineToWork(). Here, it will calculate the utility of every mine to work there. The utility is calculated by the value of the resource (letter) of the mine divided by the number of agents who already work there (reason for this will be clarified in the next subsection). In the case of a ME, the agent also considers the demand for the specific letter, and multiplies the utility by this demand. This is not necessary for a RBE, as the demand is already incorporated in the price. Without knowledge of this demands agents in the simulated ME would be significantly disadvantaged. It should however be consid-ered that in this knowledge might not be available in a ME, and this already proposes a main advantage of an open market RBE compared to a ME. The agent will pick the mine with the highest utility value.

A summary of the steps an agent undertakes for each tick: 1. updateParameters()

2. consumeProduct() OR buyResourcefromMine() 3. determineBestMineToWork()

3.2.3 Mine

On every tick, the mine will start by updating the price of its letter if the RBE is enabled in the current simulation. In the RBE, the value of a letter is determined by summing up all individual request for the specific letter in determineValueByDemand(). For the ME, the price of each letter is based on the general frequency of the specific letter in the book in determineValueByFreq(), where these prices are fixed at the start of the simulation. To ensure a fair comparison between both economies, the sum of the price of the letters should be equal in both economies on any given tick. To establish this, the relative price of a letter is fixed in the ME, but the height of the absolute price is normalized between 0 and 1. In the RBE this is accomplished by dividing the demand for an individual letter by the total amount of demands. In the ME, for each letter, the occurances of a specific letter in the book are counted, and then divided by the total amount of letters in the book. The following figure should clarify the pricing mechanism for both economies:

(14)

• RBE:

• Price A=0.75

• Price B=0.25

• ME:

• Price A=0.7

• Price B=0.3

A

B

_A

A

B

Figure 4: For this example, assume only A’s and B’s exist and there are 4 agents in the system. A’s occur 70% of the time, and B’s occur 30% of the time. Accordingly, in a ME the price of A’s will always be 0.7, and of B’s 0.3. In a RBE, if 3 agents request an A, and 1 requests a B, the total amount of requests will be 4. The price of a letter will be (demand/total demand). For A this is (3/4) = 0.75, for B this is (1/4) = 0.25.

3.2.4 Parameters

The following parameters will be fixed:

• The agents all start with a fund of 1, this is the sum of all letter prices combined. • There are 26 mines in the simulation (1 for every letter in the alphabet).

• There are 30 agents in the simulation. • All agents receive 3000 words to satisfy.

• In the case of ownership, the interest rate for the owner of a mine is 30%

Two more parameters will be tested interactively. Firstly the ownership interest ratio will be tested and compared iteratively for 0 (ME-N), 10, 20 and 30%. A final constraint in the simulation is that the actions per tick to be executed by an agent is restrained to 1 (consumeProduct() or buyResourcefromMine()). Intuitively, this number cannot be justified, as the actions per tick are relative. A stock trader could make a 100 transactions in a minute, or a consumer could buy 20 products per day on average. Therefore, these restraints will be removed in a final experiment so that agents can infinitely buy resources and consume products in a tick until they run out of money. Here the ME-N and the RBE will be compared concisely.

(15)

3.3 Evaluation 3 METHODS

3.3 Evaluation

The main variable by which the system’s performance is measured is the amount of wishes(products) an agent has satisfied. The reasoning here is that this variable best resembles the general wealth or well-being of individual agents. At a fixed chosen tick the satisfaction of wishes of all agents is measured from multiple simulations while keeping all parameters constant. Afterwards, the system is changed from the type of economy, and this process is repeated to obtain another sample. The final parameter which will be tweaked will be the parameter of ownership. This results in three different sample groups:

• A RBE non-ownership (RBE) • A ME non-ownership (ME-N) • A ME with ownership (ME-O)

Since there are 3 sample groups, a One-way ANOVA test in combination with a Tukey-HSD test is the most suitable measurement to prove statistical differences in the means of wishes satisfied between the groups. Multiple t-tests would increase the chance of a Type-I error. Besides the main output variable of amount of wishes satisfied, other variables will be evaluated more subtly: variance and fluctuations of fund posession and the variance of wishes satisfied within systems. For this comparison a fixed measure-point of 10.000 ticks is chosen. Also, graphs are reported showing the development of satisfaction of wishes in different systems. Afterwards, the ownership function will be tested iteratively as described in the previous ’Parameters’ subsection. The result section will present both graphs which can be interpreted qualitatively, and numbers which will be also be analyzed statistically.

(16)

4 RESULTS

4 Results

4.1 Wishes satisfied

The following graphs shows the mean values of the minimum, maximum and average wishes satisfied of 5 separate simulations of the three different economies:

0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 # W ishes satisfi ed # Ticks Wishes satisfied RBE (Mean) RBE (Min/Max) ME-N (Mean) ME-N (Min/Max) ME-O (Mean) ME-O (Min/Max)

Figure 5: Mean values of the average, minimum (agent with the least amount of-) and maximum wishes satisfied of 5 iterations of the three different economies (from the 30 agents)

The velocity of wish-satisfaction appears to be constant on average for all economies, since the lines are straight. This indicate that agents are effective in acquiring and spend-ing their funds. The velocity of wish-satisfaction in between economies however show differences: The most evident effect in the graphs is the negative effect of ownership. It is also noticeable how the maximum wishes satisfied, for a fortunate agent, is barely higher in the ownership system, and the poorest and the wealthiest agent differ greatly in amount of wishes satisfied compared to the other economies. The RBE and ME-N are much closer together, but on all variables (min, max and mean) the RBE is still the superior economy.

The average wishes satisfied on tick 10.000 for each system is specified in the following table:

(17)

4.1 Wishes satisfied 4 RESULTS

Type Iter. 1 Iter. 2 Iter. 3 Iter. 4 Iter. 5 Mean SD RBE 1773.1 1771.8 1760.7 1761.6 1764.1 1763.8 5.8 ME-N 1657.8 1656.5 1657.6 1673.7 1671.4 1663.4 8.4 ME-O 481.4 671.5 683.8 439.9 674.70 590.3 119.3

Table 1: Average wishes fulfilled for different systems in 5 iterations

Because of the high variance within the ownership group versus the other groups the data does not meet the assumption of equal variance. When still performing a Tukey-HSD test, it results in a very significant effect (p<0.01) for the ME-O groups against the other groups, and no significant difference between the ME-N and RBE group. It is also noticable how the variance within the RBE group is lower then in the other systems, but only slightly lower than the ME-N group.

Since the variance of the ME-N and RBE are comparable, it is interesting to investi-gate this effect more closely. This could demonstrate the effect of an on-demand valuing system against a ME. For both economies, the amount of iterations will be increased to 10, and an unpaired t-test will be performed to investigate statistical differences.

Type Iter. 6 Iter. 7 Iter. 8 Iter. 9 Iter. 10 Mean SD RBE 1754.8 1758.9 1755.9 1764.0 1772.8 1763.8 6.8 ME-N 1634.1 1591.6 1648.7 1638.5 1648.1 1647.8 23.4

Table 2: Average wishes fulfilled in the ME-N and RBE in 5 more iterations. Means and SD are of all 10 iterations.

The results of the t-test with above data are t = 15.0492, df = 18 and sd = 7.707. The effect is extremely significant with p<0.0001 and the effect size is -115.9840 (The mean of MEN minus RBE). Considering the data, the average number of wishes satisfied at tick 10.000 in different simulations of the RBE and ME-N show a low standard devia-tion, with the RBE again being the superior economy in terms of stability. The standard deviation in the ME-O is much higher, which can be explained with the ownership prop-erty: The ownership ensues a higher randomization than the other economies, because the ownership of different mines may or may not be spread more evenly among the agents.

For a more in-depth analysis of individual satisfaction of wishes, the following graphs illustrate the iterative satisfaction of individual wishes over 10.000 ticks in one iteration for each economy:

(18)

4.2 Fund balance 4 RESULTS 0 500 1000 1500 2000 0 2000 4000 6000 8000 10000 # W ishes sa tisfied # Ticks Individual wishes satisfied (RBE)

0 2000 4000 6000 8000 10000

# Ticks

Individual wishes satisfied (ME-N)

0 2000 4000 6000 8000 10000

# Ticks

Individual wishes satisfied (ME-O)

Figure 6: Mean values of the average, minimum (agent with the least amount of-) and maximum wishes satisfied of 5 iterations of the three different economies (from the 30 agents)

Again, the graphs show a slight difference between the wealth distribution of agents in the RBE and ME-N: the spread of the velocity of wishes statisfaction (wealth gap) is higher in the ME-N. Again, the effect in the ME-O is more drastic: while many agents still have a normal satisfaction of wishes, there are some with a crucially lower satisfaction of their wishes.

4.2 Fund balance

The following graphs illustrate the fluctuations in fund balance over the first 500 itera-tions of a simulation of each economy:

0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 0 100 200 300 400 500 # F unds po ses sed # Ticks RBE 0 100 200 300 400 500 # Ticks ME-N 0,00 5,00 10,00 15,00 20,00 0 100 200 300 400 500 # Ticks ME-O

Figure 7: Fluctuations in fund balance over the first 500 iterations of a simulation of each economy. Note the adjusted scale on the y-axis for the ME-O.

As the graphs show, the RBE is again the most stable economy in terms of variance and fluctuations of funds. Although the fluctuations are quite extreme in all economies, being least extreme in the RBE typically in the 0-3 range, all economies seem to self-correct these extremes, except for the ME-O. In this iteration for the ME-O, there is a single agent who profits from the ownership, while there may be more agents profit-ing from ownership in other iterations, while effectively drainprofit-ing funds from other agents. This single agent might be the owner of the mine ’e’ since this letter has the highest value.

(19)

4.3 Iterative comparison ownership function 4 RESULTS

4.3 Iterative comparison ownership function

Based on the very significant negative effect of a 30% interest rate for owners of mines, it is chosen to examine effect of lower interest rates. This might show that lower interest rates, while still advantaging a small group of agents, preserves the durability of an econ-omy. Therefor, this subsection will provide a comparison of an interest rate of 0% (the ME-N), 10%, 20% and 30% (the ME-O).

The following graphs illustrate the average value of 3 iterations of the minimum, mean and mean wishes satisfied of all agents for an ownership interest rate of 10% and 20%, the dotted lines represent the data from the first subsection of the average wishes fulfilled for a 0% interest rate (ME-N), and a 30% interest rate (the former ME-O constant):

0 500 1000 1500 2000 0 2000 4000 6000 8000 10000 # W ishes sa tisfied # Ticks

Wishes satisfied (10% Interest)

Mean Min/Max Mean 0% Mean 30%

0 2000 4000 6000 8000 10000

# Ticks

Wishes satisfied (20% Interest)

Mean Min/Max Mean 0% Mean 30%

Figure 8: The average value of 3 iterations of the minimum, mean and mean wishes satisfied of all agents for an ownership interest rate of 10% and 20%,

The number of iterations for the 10% and 20% interest rate is now increased to 5 for statistical analysis. The average wishes satisfied on tick 10,000 for the four ownership interest ratios is specified in the following table:

Type Iter. 1 Iter. 2 Iter. 3 Iter. 4 Iter. 5 Mean SD 0% (ME-N) 1657.8 1656.5 1657.6 1673.7 1671.4 1663.4 8.4

10% 1108.9 1300.0 1311.3 1300.3 1160.3 1236.2 94.6

20% 923.6 562.6 841.6 883.8 884.3 819.2 146.4

30% (ME-O) 481.4 671.5 683.8 439.9 674.70 590.3 119.3

Table 3: Average wishes fulfilled for different ownership interest ratios in 5 iterations

(20)

4.4 Infinite actions 4 RESULTS

When still performing a Tukey-HSD test, all groups statistically differ from each other with a significance of p<0.01, except for the groups with an ownership interest ratio of 20 and 30%, which differ with a significance of p<0.05. When considering the deviation in the means of wishes fulfilled, the greatest effect is between the ME-N and ME-O(10%) groups (-427.2), followed by the deviation of means between the O(10%) and ME-O(20%) groups (-417.0). The final, much smaller deviation is between the ME-ME-O(20%) and ME-O(30%) groups (-228.9).

The graphs and these results indicate a linear correlation between the average wealth of an economy and the percentage of ownership of mines. However, it seems the first percentiles (0-10% and 10-20% increases) have a larger effect than the later percentile (20% to 30% increase). From these findings it can be concluded that any form of ownership, even in small amounts, is very undesirable for the sustainability of an economy.

4.4 Infinite actions

A final parameter which is important to consider is the amount of actions an agent can make per tick. In the simulation this was so far constrained to one action; either purchas-ing a resource or consumpurchas-ing a product. Arguably, it is not very plausible that an agent would spend all its money in a tick, but it could execute multiple actions. To investigate if the system exhibits the same behaviour when agents can have multiple actions in a tick, the constraint of the amount of actions is removed. Now, an agent can execute actions as long as he has sufficient funds to do so.

Allowing agents for infinite actions per ticks had a tremendous effect on the velocity of wishes satisfaction. The first agents to satisfy all their wishes would be at around tick 1000 for both the RBE and the ME. The following graph illustrates the effect of the allowance of infinite actions per tick. Again it is the average value of 5 iterations of the minimum, mean and mean wishes satisfied of all agents:

(21)

4.4 Infinite actions 4 RESULTS 0 500 1000 1500 2000 2500 0 100 200 300 400 500 # W ish es sa tisfie d # Ticks

Average wishes satisfied (Infinite actions)

RBE (Mean)

RBE (Min/Max)

ME (Mean)

ME (Min/Max)

Figure 9: The average value of 5 iterations of the minimum, mean and mean wishes satisfied of all agents when infinite actions per tick are allowed

Here, the ME appears to perform slightly better. However, the variance seems to be bigger in the ME as well. Again, a unpaired t-test will be performed to investigate statistical differences.

Type Iter. 1 Iter. 2 Iter. 3 Iter. 4 Iter. 5 Mean SD ME-N 1548.4 1600.8 1846.4 1661.6 1616.1 1654.7 114.6 RBE 1547.5 1581.5 1544.9 1549.4 1560.8 1556.8 15.1

Table 4: Average wishes satisfied for ME-N and RBE in 5 iterations with infinite actions

The results of the t-test with above data are t = 1.8937, df = 8 and sd = 51.680. The effect is significant with a confidence p<0.1 and the effect size is 97.87 (The mean of ME-N minus RBE). This shows that the MEN on average performs slightly better. However, the economy seems less stable with a much higher variance of averages between simulations. More specifically, the gap between the wealthiest agent and the poorest agent was much larger in individual simulations of the ME-N compared to the RBE, as can be seen in the following graph of the satisfaction of individual wishes in two simulations where infinite actions are allowed:

(22)

4.4 Infinite actions 4 RESULTS 0 500 1000 1500 2000 0 100 200 300 400 500 # W ishes sa tisfied #Ticks

Individual wishes satisfied (RBE)

0 100 200 300 400 500

#Ticks

Individual wishes satisfied (ME)

Figure 10: The individual satisfaction of wishes in a simution of the RBE and ME, with infinite actions allowed.

This graphs shows how the real-time pricing mechanism is more effective at stabilizing the demand and satisfaction of the wishes of individual agents, which results in a low wealth-gap. The velocity of demand satisfaction is also more constant in the RBE, as the lines in the graph of the ME are more curved while the lines in the RBE are approximately straight. This effect is more evident here than in the earlier graphs since the scope of iterations is lowered to 500, thus short term fluctuations are better visible.

(23)

5 CONCLUSION & DISCUSSION

5 Conclusion & Discussion

The simulation has shown a very significant positive effect for satisfaction of individual wishes in a resource-based economy compared to the market economy, but the effect size was relatively small. This effect slightly reversed when infinite actions were allowed for an agent. However, it is disputable whether this is a realistic scenario in a real econ-omy, as it would imply that all people spend their funds maximally. The variance, which can be considered an indicator to the wealth gap in a real economy, was also smaller in simulations the RBE, also in the case of allowance of infinite actions. Finally, in the RBE, the velocity of wishes satisfaction was on average more constant for all agents. This shows how knowledge of current demand can improve general satisfaction of needs on a larger scale, in a closed environment, and exposes opportunities for the management of the pricing in a economy by a supercomputer or a blockchain. It also shows how resources are distributed more evenly and constantly, as can be concluded from the lower variance in the RBE, the more constant satisfaction of wishes, and the higher stability of funds possessions. The effect of a high fund variation can be explained with the fixed prices which exist in the ME: an agent can continuously work on a certain letter. An agent can start working at a letter with a very high utility, he might keep working at this letter for an extensive period, while the chance of other agents joining in is smaller (because the mine is already occupied). This could boost the fund of this agent, while automatically draining funds of other agent, and thus their purchasing power. In the RBE, the value of letters and thus the utility of working on a certain mine, is more dynamic, which disen-gages the utility of permanently working on a single mine.

The research has also clearly shown the detrimental effect of ownership of resources, where there was not really any profit for any of the agents in the economy: The ’richest’ agent did not have significantly more wishes satisfied than the richest agents in the other systems. This while other agents greatly suffered from a lack of funds to purchase any letters. Evidence for this effect can be found in both fund possessions and product con-sumption. This greatly impaired the functioning of the system and resulted in a huge gap in wealth and access to resources. Even low ownership interest-rates have shown to have a big negative effect, where there is a linear correlation between the interest rates and negative impact on the economy. Apparently the economy is very sensitive to disruptions in flows of funds, if it were to ’leak’ to a selected group of advantaged agents.

While the effect size is not consistent througout the parametrization, it can be con-luded that the quality of the RBE seems to be higher overall. The system is more stable

(24)

5 CONCLUSION & DISCUSSION

in resource distribution. The research question was: In a simulated digital economy, what set of agent strategies, incentives and valuing system leads to a balanced and sustainable economy as emergent behavior?. It appears that when agents are solely concerned with their own utility, and try to maximize this using a uniform strategy, a RBE provides an overall balanced and sustainable solution to manage the economy efficiently. Morover, no agent should be declared ’owner’ or gain other advantages over other agents concern-ing resource distribution and reward, since it is both inefficient and destructive for the economy. This shows that a RBE where all resources are declared public has potential for a future restructuring of the current political and economical systems, towards a new, sustainable society. This would not even require manipulation of the incentives of the participants of society, but simply a restructuring of the societal ’rule-set’.

(25)

6 FUTURE WORK

6 Future Work

The system investigated in this simulation is a closed, isolated system. In a real-world situation, there are evidently many more variables to be investigated. This research did not focus on the individual decision making process, but assumed a uniform agent strategy. In future work, the psychological/sociological component could be modeled in more depth. Future MARA research could also consider real-world resources and a data stream of their real value, using stock values for example. The implementation of a mar-ket economy could also be expanded in future research, implementing banks, taxes and governments. Also, the distance component could be modeled more extensively. In this research, it is only utilized to determined ownership, but in future research it could be used for a more extensive individual predisposition determination. An example would be: distance to a mine creating greater difficulty/cost to extract from it. In a more dynamic system agent could also be allowed to move toward mines for a more optimal agent posi-tioning, which the Repast Suite also allows.

Another opportunity for improving the effectiveness of an on-demand pricing system, like implemented in this simulation, could be to use machine learning to predict demand, and adjust prices accordingly. This is beyond the scope of this research, but is certainly interesting for future research, also because implementing smart prediction functions us-ing machine learnus-ing could be implemented in a digital economic management system like a blockchain. This would also account for other heuristics for optimal price determina-tion, and could theoretically even be applied for managing agent incentives.

The conceptual assumptions of this research are based on concepts which also need more future research. The concept of a resource based economy is incorporated into dif-ferent organizations such as the Venus Project and the Zeitgeist movement, but literature on the concept is scarce and in the view of the author it’s assumptions are not very clearly defined or sometimes insufficient in the sense that they do not provide verifiable/falsifiable hypotheses. In this sense, this research could also contribute to the conceptualization of a resource based economy. It has showed the detrimental effects of ownership in the form of interest, with no one profiting, and the potential of a real-time demand-driven pricing system. Therefore, the conclusions of this research can only promote the revive of Cybersyn, the growth of blockchain and the extention of other sharing initiatives like peerby into our society. Such experiments might help further induce the start of a digital, sustainable and fair global economy.

(26)

References References

References

Bandini, S., Manzoni, S., & Vizzari, G. (2009). Agent based modeling and simulation: an informatics perspective. Journal of Artificial Societies and Social Simulation, 12 (4), 4.

Bauwens, M. (2005). The political economy of peer production. CTheory, 1 .

Bauwens, M. (2009). Class and capital in peer production. Capital & Class, 33 (1), 121–141.

Chevaleyre, Y., Dunne, P. E., Endriss, U., Lang, J., Lemaitre, M., Maudet, N., . . . Sousa, P. (2006). Issues in multiagent resource allocation. Informatica (Slovenia), 30 (1), 3–31.

De Scitovszky, T. (1941). A note on welfare propositions in economics. The Review of Economic Studies, 9 (1), 77–88.

Ethereum. (2017). Retrieved from https://www.ethereum.org/

Fresco, J., & Meadows, R. (2002). The best that money can’t buy:(beyond politics, poverty and war). Osmora Inc.

Goldspink, C., et al. (2002). Methodological implications of complex systems approaches to sociality: Simulation as a foundation for knowledge. Journal of Artificial Societies and Social Simulation, 5 (1), 1–19.

Haber, J., & Valenti, R. (2015). Project cybersym: Modeling distributed cybernetic demand management for resource based economies.

Kosba, A., Miller, A., Shi, E., Wen, Z., & Papamanthou, C. (2016). Hawk: The blockchain model of cryptography and privacy-preserving smart contracts. In Security and privacy (sp), 2016 ieee symposium on (pp. 839–858).

Kraus, S., Wilkenfeld, J., & Zlotkin, G. (1995). Multiagent negotiation under time constraints. Artificial intelligence, 75 (2), 297–345.

Luck, M., McBurney, P., Shehory, O., Willmott, S., et al. (2005). Agent technology: Computing as interaction. A Roadmap for Agent Based Computing. University of Southampton on behalf of AgentLink III .

McLeish, B., Berkowitz, M., Joseph, P., & eds. (2014). The zeitgeist movement defined, realizing a new train of thought. TZM Lecture Team.

Medina, E. (2011). Cybernetic revolutionaries: technology and politics in allende’s chile. MIT Press.

Repast. (2016). Simphony. Retrieved from https://repast.github.io/

The venus project. (2017, April). Retrieved from https://www.thevenusproject.com/ en/about/resource-based-economy

(27)

References References

framework for peer resource sharing. In Workshop on economics of peer-to-peer systems (Vol. 35).

Towards a resource-based economy: Modeling valuing systems and ownership functions in a simulated economy