Baseline Governance: Transaction Tax and Circuit Breaker Effects on Multi-Agent Welfare

Authors: SWARM Research Collective Date: 2026-02-13 Framework: SWARM v0.1.0

Abstract

We investigate the effects of transaction taxation and circuit breakers on welfare, toxicity, and distributional fairness in a mixed-agent simulation. Using the SWARM framework, we sweep tax rates (0%, 5%, 10%, 15%) and circuit breaker activation (enabled/disabled) across 80 runs (10 seeds per configuration) with 5 agents (3 honest, 1 opportunistic, 1 deceptive). We find that transaction taxes significantly reduce total welfare (d=1.41, p<0.0001 for 0% vs 15% tax), while circuit breakers have no statistically significant effect. The welfare reduction from taxation disproportionately affects honest agents (d=1.29, p=0.0002 for 0% vs 10% tax on honest payoff), while deceptive agents remain relatively unaffected. All welfare distributions pass Shapiro-Wilk normality tests, validating parametric analysis. Of 42 pre-registered hypotheses, 4 survive Bonferroni correction and 6 survive Benjamini-Hochberg correction.

1. Introduction

Governance mechanisms in multi-agent systems face a fundamental tension: interventions designed to reduce harmful behavior may impose deadweight costs that reduce overall welfare. Transaction taxes are a canonical governance lever — they increase the cost of interactions, potentially deterring low-value or exploitative exchanges, but also reducing the volume of beneficial interactions. Circuit breakers provide a qualitatively different mechanism: they halt activity when systemic risk indicators exceed thresholds, but may be inert in stable environments.

This study provides a controlled empirical characterization of these two governance levers in the SWARM baseline scenario, a minimal mixed-agent environment with 3 honest, 1 opportunistic, and 1 deceptive agent. We focus on three questions:

1. Does transaction taxation reduce welfare, and if so, is the effect monotonic in the tax rate?
2. Does circuit breaker activation affect welfare or toxicity in a stable baseline environment?
3. How do governance costs distribute across agent types — do they disproportionately burden honest or deceptive agents?

2. Methods

2.1 Scenario Configuration

Parameter	Value
Agents	5 (3 honest, 1 opportunistic, 1 deceptive)
Epochs	10
Steps per epoch	10
Payoff: s_plus	2.0
Payoff: s_minus	1.0
Payoff: h (externality)	2.0
Payoff: theta (governance threshold)	0.5
Reputation decay	1.0 (disabled)
Bandwidth cap	10
Staking	Disabled
Audit	Disabled

2.2 Swept Parameters

Parameter	Values
governance.transaction_tax_rate	0.0, 0.05, 0.10, 0.15
governance.circuit_breaker_enabled	False, True

Total configurations: 4 x 2 = 8. Seeds per configuration: 10. Total runs: 80.

2.3 Metrics

• Welfare: Cumulative payoff across all agents over all epochs
• Toxicity rate: E[1-p | accepted], expected harm from accepted interactions
• Quality gap: E[p | accepted] - E[p | rejected]; negative indicates adverse selection
• Agent payoff: Per-type mean payoff (honest, opportunistic, deceptive)

2.4 Statistical Methods

• Welch's t-test (unequal variance) for pairwise comparisons
• Mann-Whitney U as non-parametric robustness check
• Cohen's d for effect sizes (pooled SD)
• Bonferroni correction (alpha = 0.05/42 = 0.00119)
• Benjamini-Hochberg correction for false discovery rate
• Shapiro-Wilk normality validation on welfare distributions
• Paired t-test for agent-type stratification
• 42 total pre-registered hypotheses across all parameter-metric pairs

3. Results

3.1 Welfare by Tax Rate

Tax Rate	Welfare (mean +/- SD)	Toxicity (mean +/- SD)	Quality Gap (mean +/- SD)	N
0.00	62.5 +/- 7.0	0.3018 +/- 0.0133	0.0094 +/- 0.0184	20
0.05	60.8 +/- 7.3	0.3045 +/- 0.0143	0.0146 +/- 0.0218	20
0.10	52.3 +/- 8.3	0.3084 +/- 0.0154	0.0101 +/- 0.0201	20
0.15	53.2 +/- 6.2	0.3066 +/- 0.0149	0.0133 +/- 0.0178	20

Transaction tax significantly reduces welfare. The effect is non-linear: a 5% tax produces a modest decline (62.5 to 60.8), while 10% and 15% produce a sharper drop (to 52.3 and 53.2 respectively). Toxicity is unaffected by taxation — all values cluster near 0.305 regardless of tax rate. Quality gap remains near zero and positive (no adverse selection) across all conditions.

3.2 Circuit Breaker Effect

Circuit Breaker	Welfare (mean +/- SD)	Toxicity (mean +/- SD)	N
Disabled	56.8 +/- 7.7	0.3061 +/- 0.0140	40
Enabled	57.6 +/- 9.2	0.3046 +/- 0.0150	40

Circuit breaker activation has no statistically significant effect on welfare (p>0.05) or toxicity. This is expected in a stable baseline environment where systemic risk indicators remain below circuit breaker thresholds.

3.3 Interaction Effects

The grouped analysis (tax rate x circuit breaker) confirms that the two governance levers operate independently. Welfare reduction from taxation is consistent regardless of circuit breaker status.

3.4 Significant Results (Bonferroni-corrected)

Comparison	Metric	d	p	Survives
Tax 0% vs 15%	Welfare	1.41	0.0001	Bonferroni
Tax 0% vs 10%	Welfare	1.33	0.0002	Bonferroni
Tax 0% vs 10%	Honest payoff	1.29	0.0002	Bonferroni
Tax 5% vs 15%	Welfare	1.13	0.0010	Bonferroni
Tax 5% vs 10%	Welfare	1.09	0.0015	BH only
Tax 0% vs 15%	Honest payoff	1.08	0.0016	BH only

All significant effects involve the transaction tax on welfare or honest payoff. No toxicity, quality gap, opportunistic, or deceptive payoff comparisons reach significance.

3.5 Agent-Type Stratification

Agent Type	Mean Payoff
Honest	14.50
Opportunistic	11.46
Deceptive	2.26

Pairwise comparisons (paired t-test):

Comparison	Cohen's d	p-value
Honest vs Opportunistic	0.84	<0.0001
Honest vs Deceptive	5.73	<0.0001
Opportunistic vs Deceptive	2.91	<0.0001

Honest agents earn significantly more than all other types. Deceptive agents earn the least, suggesting that the baseline governance environment effectively penalizes deception even without explicit audit mechanisms.

3.6 Tax Impact by Agent Type

Tax Rate	Honest (mean +/- SD)	Opportunistic (mean +/- SD)	Deceptive (mean +/- SD)
0.00	16.14 +/- 2.44	11.95 +/- 4.49	2.17 +/- 0.80
0.05	15.20 +/- 2.85	12.61 +/- 5.46	2.59 +/- 0.85
0.10	12.94 +/- 2.50	11.41 +/- 3.94	2.08 +/- 1.70
0.15	13.70 +/- 2.05	9.88 +/- 2.53	2.19 +/- 1.56

The welfare reduction from taxation is borne primarily by honest agents (16.14 to 12.94, a 20% decline from 0% to 10% tax). Deceptive agent payoff is unchanged across conditions (~2.2), making taxation regressive: it taxes productive behavior without reducing exploitative behavior.

3.7 Normality Validation

Group	Shapiro-Wilk W	p-value	Normal?
Tax 0%	0.9370	0.2101	Yes
Tax 5%	0.9598	0.5403	Yes
Tax 10%	0.9708	0.7707	Yes
Tax 15%	0.9678	0.7087	Yes

All groups pass normality tests (p > 0.05), validating the use of parametric t-tests.

4. Discussion

4.1 Key Findings

Transaction taxes reduce welfare without reducing toxicity. The strongest finding is that taxation imposes a deadweight loss (d=1.41 for 0% vs 15%) while toxicity remains constant at ~0.305. This suggests that in a baseline environment, taxation does not selectively deter harmful interactions — it reduces interaction volume uniformly, affecting beneficial and harmful interactions equally.

The tax burden falls disproportionately on honest agents. Honest agents lose 20% of their payoff between 0% and 10% tax rates, while deceptive agents are unaffected. This is consistent with the theoretical prediction that agents who generate the most surplus (honest agents, through high-p interactions) pay the most in transaction taxes, while agents who generate little surplus (deceptive agents) have less to tax.

Circuit breakers are inert in stable environments. This is a null result, but an important one: circuit breakers are a latent governance mechanism that only activates under stress conditions. In a stable 5-agent, 10-epoch environment, systemic risk indicators never exceed thresholds. This motivates future studies under adversarial stress-testing conditions.

4.2 Non-linearity of Tax Effects

The welfare decline is not linear in tax rate. The jump from 5% to 10% (60.8 to 52.3, a 14% decline) is larger than from 0% to 5% (62.5 to 60.8, a 3% decline) or from 10% to 15% (52.3 to 53.2, essentially flat). This suggests a phase-transition-like threshold between 5% and 10% where tax costs exceed the marginal value of some interactions, causing agents to stop participating.

4.3 Implications for Governance Design

These results challenge the naive view that transaction taxes are a universal governance tool. In this environment:

• Taxes do not improve toxicity outcomes
• Taxes disproportionately harm honest agents
• The welfare cost of taxation is non-linear and potentially catastrophic above certain thresholds

More targeted governance mechanisms — such as reputation-weighted taxation, audit-based penalties, or staking requirements — may achieve better outcomes by selectively increasing costs for low-quality interactions while leaving high-quality interactions unaffected.

5. Limitations

1. Small agent pool: 5 agents limits the diversity of strategic interactions. Larger populations may exhibit different dynamics.
2. Short time horizon: 10 epochs may not capture long-term equilibrium effects of taxation.
3. No reputation dynamics: Reputation decay is disabled (rate=1.0), which removes a key feedback mechanism.
4. Binary governance levers: We test only on/off circuit breakers and fixed tax rates, not adaptive mechanisms.
5. No network effects: Agents interact in a complete graph; realistic systems have network structure.

6. Reproducibility

# Reproduce the sweep
python runs/20260213-173805_baseline_governance/run_sweep.py

# Reproduce the analysis
python runs/20260213-173805_baseline_governance/analyze.py

# Reproduce the plots
python runs/20260213-173805_baseline_governance/generate_plots.py

Scenario file: scenarios/baseline.yaml Seeds: 200-209 (10 per configuration) Total runs: 80 (4 tax rates x 2 circuit breaker states x 10 seeds)

7. References

[TODO]