Presentation at MIS Speaker's Series, Eller College of Management, The University of Arizona https://mis.eller.arizona.edu/events/06nov2018/mis-speakers-series-jungpil-hahn

1. Understanding Blockchain Governance Decentralization:
An Agent-based Simulation Model
Jungpil Hahn
National University of Singapore
jungpil@nus.edu.sg
Joint work with Christoph Müller-Bloch, JonasValbjørn Andersen and Jason Spasovski (IT University of Copenhagen)

2. Understanding Blockchain Governance Decentralization:
An Agent-based Simulation Model
Jungpil Hahn
National University of Singapore
jungpil@nus.edu.sg
Joint work with Christoph Müller-Bloch, JonasValbjørn Andersen and Jason Spasovski (IT University of Copenhagen)

3. the rise of algorithmic operations
efore talking about Blockchains

6. algorithms have real-world consequences that
are materialized in practice

7. algorithms have real-world consequences that
are materialized in practice
• unintended consequences
• opaque connection between process and outcome

8. … distributed ledger technology in the form of a distributed transactional database,
secured by cryptography, and governed by a consensus mechanism …

10. … distributed ledger technology in the form of a distributed transactional database,
secured by cryptography, and governed by a consensus mechanism …
tamper-resistance

13. • central operator not needed
• choose validator node
=> algorithmic governance via
consensus mechanism
(PoW, PoS)

14. … distributed ledger technology in the form of a distributed transactional database,
secured by cryptography, and governed by a consensus mechanism …
tamper-resistance

15. … distributed ledger technology in the form of a distributed transactional database,
secured by cryptography, and governed by a consensus mechanism …
tamper-resistance
decentralization of decision making power
requires

16. … distributed ledger technology in the form of a distributed transactional database,
secured by cryptography, and governed by a consensus mechanism …
decentralization of decision making power
is the intended outcome of
consensus mechanisms

17. … distributed ledger technology in the form of a distributed transactional database,
secured by cryptography, and governed by a consensus mechanism …
(de)centralization of decision making power
emerges from
network of nodes + interactions

18. algorithms have real-world consequences that
are materialized in practice
• unintended consequences
• opaque connection between process and outcome

19. Lastly, wealth in the crypto universe is even
more concentrated than it is in North Korea.
Gini(N. Korea) = 0.86
Gini(USA) = 0.41
Gini(Bitcoin) = 0.88

20. Complex Adaptive Systems (CAS)
/
Agent-based Modeling

21. Complex Adaptive Systems (CAS)
CAS Structure CAS Component CAS Component Elements
Generative
structure
(unobserved)
Model Rule The behavioral logic of the
model specified at the model
level
Elemental
structure
(micro-level)
Agent Identity
Attributes
Behavioral Rules
Interaction Connection
Flow
Environment Initial conditions, model
parameters and settings
Observed
structure
(macro-level)
Emergent
Property
Output observations at the
system level

22. Blockchain as CAS
CAS Structure CAS Component CAS Component Elements Equivalence in Blockchain Governance
Generative
structure
(unobserved)
Model Rule The behavioral logic of the
model specified at the model
level
Consensus mechanism an an algorithm for
chooseing validator nodes
Elemental
structure
(micro-level)
Agent Identity A public address that identifies nodes
Attributes Currency stake
Behavioral Rules Make a transaction
Interaction Connection Transactions on the blockchain, fee paid to
winning validator nodes
Flow Amount and volume of transactions between
agents
Environment Initial conditions, model
parameters and settings
Number of available validator nodes, initial
distribution of stakes
Observed
structure
(macro-level)
Emergent
Property
Output observations at the
system level
Distribution of decision making power,
structure of the validation network

23. Agent-based model of PoS consensus mechanism
• Blockchain network consisting of A potential validator agents
• Each agent a is assigned an initial currency balance ba
• At each t, a random number Vt of transactions of size sat (<ba) between
random pairs of agents take places and an agent is selected to be the
validator; sat = 1
• Likelihood of becoming the validator node depends on decision making
power pa which is proportional to bat
• Validator node receives transaction fee F and is added to ba

24. degree of (de)centralization
Gini coefficient =
Gini = 0; no inequality; fully decentralized
Gini = 1; full inequality; fully centralized
A
A + B

25. 0.5
0.6
0.7
0.8
0.9
1.0
0 20,000 40,000 60,000 80,000 100,000 120,000
GiniCoefficient
Blocks
NXT Observed
Simulated
model validation

26. simulation experiments
Parameter Baseline Experiments
Design
Parameters
Initial # of Validator Nodes 73
5, 10, 25, 50,
250, 500, 1000%
Transaction Fee 207
Behavioral
Parameters
Transaction Amount 39,434
Transaction Volume 2.54
Validator Network Growth 0.0293

27. Initial validator nodes
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
10%
25%
50%
250%
500%
1,000%

28. Initial validator nodes
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
1,000%

29. Transaction fee amount
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
10%
25%
50%
250%
500%
1,000%

30. Transaction fee amount
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
1,000%

31. Transaction amount
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
10%
25%
50%
250%
500%
1,000%

32. Transaction amount
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
1,000%

33. Transaction volume
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
10%
25%
50%
250%
500%
1,000%

34. Transaction volume
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
1,000%

35. Validator network growth
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
10%
25%
50%
250%
500%
1,000%

36. Validator network growth
0.5
0.6
0.7
0.8
0.9
1.0
0 50,000 100,000 150,000 200,000 250,000 300,000
GiniCoefficient
Blocks
Baseline (100%)
5%
1,000%

37. summary
Parameters Greater decentralization with …
Initial Network Size Larger initial networks
Transaction Fee Smaller transaction fees (marginal)
Transaction Amount Larger average transaction amounts
Transaction Volume Initially larger volume but marginal impact in the long-term
Validator Network Growth Slower growth rates but faster growth rates in the long term

38. scenario testing

39. maximal decentralization scenario
0.5
0.6
0.7
0.8
0.9
1.0
0 20,000 40,000 60,000 80,000 100,000 120,000
GiniCoefficient
Blocks
Baseline
Design Parameters Only
All Parameters

40. Conclusions / Contributions
• Well-intentioned designs of algorithmic governance may lead to
unexpected (undesirable) outcomes
• Identify model parameters (initial validator network, transaction fees) that
are likely to lead to (un)desirable levels of decentralisation
• Identify behavioural parameters (transaction volume, amount, validator
nework growth) that are likely to lead to (un)desirable levels of
decentralisation
• CAS and agent-based modeling as a useful tool to reduce opacity induced
by algorithm-mediated decision making

41. Limitations and Future Work
• What are “acceptable” and/or “critical” levels of (de)centralization
• Study interaction effects of model and behavioral parameters
• Other consensus mechanisms? (e.g., variants of PoS, PoW, PoA)
• Endogenous parameters