REAP Calculator

What This Is

The REAP Calculator is a fiscal policy simulation tool that answers one question no existing tool answers:

Does this tax code actually make people's lives better?

Standard fiscal evaluation asks: "How much revenue does the State collect?" (R)

REAP asks: "How much of that revenue reaches its intended public purpose, at what cost to the economy, who bears the burden, and does the formal base grow or shrink?" (R_useful)

The difference between those two questions is the difference between collecting more and delivering more.

The Equation

Neither Surin nor Bertil had this equation alone. Surin gave us R = τ × α × GDP. Bertil added (1 − Φ). Together, it shows that a code collecting LESS but capturing LESS of what it collects can deliver dramatically MORE.

Origin Story

Two researchers working independently on Haiti's January 20, 2023 Tax Code (656 articles, 244 pages, effective October 1, 2026) arrived at overlapping conclusions from opposite directions:

Réginald Surin (Economist, former PM Economic Adviser, former MEF Economist) built a Computable General Equilibrium model calibrated on Haiti's Social Accounting Matrix. He found: the Code's calibration would contract GDP by 2.2%, hit the poorest 3.6x harder than the richest, and shrink the formal base from 30% to 23% of GDP.

Wesley Bertil (BARSS Research) built the Elite Extraction with Differential Targeting Model (EEDTM) across 25 cases spanning 200 years. Applied to the Code, he found: the reform changes everything affecting the powerless while leaving everything affecting the powerful unchanged. The allocation architecture (100% general fund) means 26% of collected revenue is captured through discretionary allocation, debt service to historical extractors, and political patronage.

They found each other on March 28, 2026. This calculator is the synthesis of their frameworks.

The Three Frameworks

MIFED (Surin, 2026)

Modèle d'Impact Fiscal en Économie Duale

Question: When you change the fiscal screw, what happens to an economy split between those who pay and those who don't?

Key insight: In an economy where 70% is informal, raising rates on the 30% formal sector without expanding the base is self-defeating (the "dual fiscal trap").

Equations contributed:

• R = τ × α × GDP (dual fiscal trap)
• τ_eff = τ / α (real fiscal pressure)
• Δα = -η × Δτ* (informal migration)
• μf = |ΔGDP| / ΔR (fiscal multiplier)
• Leontief propagation, LES consumption model, Harberger DWL

EEDTM (Bertil, 2024–2026)

Elite Extraction with Differential Targeting Model

Question: Who extracts, how much, from whom, and through what mechanism?

Key insight: Extraction follows predictable mathematical patterns. The same dynamics that operated through colonial debt operate through fiscal architecture.

Equations contributed:

• (1 − Φ) capture correction
• Symmetric migration: Δα = -η_out × Δτ* + η_in × Δ(B)
• Multi-dimensional Gamma (consumption, wealth, intergenerational, diaspora, reform)
• Selective reform test
• Double extraction identification

REAP (Bertil & Surin, 2026)

Revenue, Extraction, Allocation & Progressivity

Question: Does this tax reform actually improve lives?

Key insight: You don't need to collect more. You need to capture less of what you collect.

The synthesis:

Architecture

Class Structure

Data Flow

The 17 Equations

From MIFED (Surin, 2026)

1. Dual Fiscal Trap

Revenue is the product of rate × formal share × economy size. Raising τ without expanding α can shrink R.

2. Real Fiscal Pressure

What formal actors actually experience. Haiti: 5% apparent = 16.7% real (÷ 0.30).

3. Effective Integrated Rate

Multi-layer taxes stack multiplicatively. Five layers at 5% each ≠ 25%. They equal 22.6%.

4. Informal Migration

Higher rates push formal operators to informal. η = 0.20 (Schneider & Buehn 2009).

5. Fiscal Multiplier

GDP cost per unit of net revenue. Above 1.0 = reform destroys more than it collects.

6. Pass-Through

Share of tax increase absorbed by consumer vs producer. Rigid demand (food) → consumer absorbs 70-85%.

7. Leontief Propagation

A shock to one sector cascades through inter-sectoral linkages. Industry multiplier = 1.60.

8. Harberger Deadweight Loss

Pure waste. Benefits nobody. Scales with the SQUARE of the rate change.

9. Revenue Feedback

Each point of GDP lost costs 1.2-1.5 points of revenue. The trap feeds itself.

10. Regressivity

Consumption contraction ratio. Above 1.0 = poor hit harder. Below 1.0 = progressive.

From EEDTM (Bertil, 2024–2026)

11. Useful Revenue (The Composite)

What actually reaches public purpose after capture.

12. Symmetric Migration

Extends MIFED: adds formalization PULL from barrier reduction, not just rate PUSH.

13. Multi-Dimensional Gamma

Five dimensions: consumption, wealth, intergenerational, diaspora, reform selectivity.

14. Extended Multiplier

True cost per useful gourde. Higher than μf when Φ > 0.

15. Selective Reform Test

If only indirect taxes change while wealth/dividends stay frozen: Γ_reform = ∞.

16. Double Extraction Test

Population taxed for the consequences of having been extracted from.

REAP Composite

17. REAP Score

Five dimensions, each 0-1, weighted sum. Higher = better reform.

The SAM: Haiti's Economic Photograph

The model is calibrated on Haiti's Social Accounting Matrix (Cicowiez/IDB, 2013, rescaled to 2019). This is the ONLY publicly available SAM for Haiti.

What the SAM tells us

Three economic worlds:

• Agriculture (9%): Near-autarky. Buys almost nothing from other sectors. Low multiplier (1.22). Employs ~40% of population but almost entirely informal.
• Industry (27%): Most vulnerable. 41% of output is intermediate purchases. Highest multiplier (1.60). Every upstream tax amplifies through the chain.
• Services (64%): Dominant. Highest absolute impact from any shock because of sheer size. Multiplier 1.40.

The Technical Coefficients Matrix (A)

Read column-wise: Industry buys 6.63 centimes of agricultural goods, 19.36 centimes of other industrial goods, and 14.80 centimes of services for every gourde of industrial output.

The Leontief Inverse (I−A)⁻¹

Read column-wise: a 1-gourde increase in demand for industrial goods creates 1.57 gourdes of total output across the economy. That's the cascade.

Household Parameters (LES, from ECVMAS/IHSI 2012)

Q1 has 8 gourdes of margin on 100. When prices rise, those 8 gourdes melt. Q5 has 65 gourdes of cushion. The same tax produces wildly different impacts. That's structural regressivity. That's Gamma.

How to Use

Basic: Run built-in scenarios

# Terminal
python reap_calculator.py

Runs three Haiti scenarios (Their Code, Surin's Corrections, BARSS Code), prints results, comparison table, and REAP scores. Saves JSON.

Custom: Create your own policy

from reap_calculator import REAPCalculator, TaxPolicy

calc = REAPCalculator()

# Create policy
p = TaxPolicy("My Tax Reform")

# Set excise rates (product, old_rate, new_rate)
p.set_excise("beer_local", old=0.06, new=0.12)
p.set_excise("water", old=0.00, new=0.00)      # keep exempt
p.set_excise("tobacco", old=0.15, new=0.40)
p.set_excise("vehicle_gt10", old=0.10, new=0.50)

# Set direct taxes
p.set_direct("IS", old=0.30, new=0.28)
p.set_direct("TMS", old=0.02, new=0.025)
p.set_direct("IMF", old=0.00, new=0.003)

# Set TCA structure (flat or progressive)
p.set_tca(necessity=0.05, standard=0.10, luxury=0.12)

# Set capture rate (Φ)
p.set_capture_rate(0.15)

# Add extra revenue streams (M HTG)
p.add_revenue_stream("property_reform", 5_000)
p.add_revenue_stream("wealth_tax", 10_000)

# Set formalization barrier reduction (0-1)
p.set_barrier_reduction(0.20)

# Run
results = calc.run(p)
calc.print_results(results)

# Get REAP score
scores = calc.reap_score(results)
print(f"REAP: {scores['REAP']}")

# Compare with other policies
r2 = calc.run(another_policy)
calc.compare([results, r2])

# Save
calc.save(results, "my_results.json")

From JSON

# Save a policy
import json
with open("my_policy.json", "w") as f:
    json.dump(p.to_dict(), f, indent=2)

# Load a policy
with open("my_policy.json") as f:
    p_loaded = TaxPolicy.from_dict(json.load(f))

For another country

class MySAM:
    SECTORS = ["Agriculture", "Manufacturing", "Services"]
    X = np.array([...])   # Production by sector
    VA = np.array([...])  # Value added
    A = np.array([...])   # Technical coefficients 3x3
    # ... etc (see HaitiSAM for full structure)

calc = REAPCalculator(sam=MySAM())

Available Products

Products in the default Haiti catalog. Set rates for any subset.

REAP Scoring

Five dimensions, each scored 0-1:

Default weights: equal (0.20 each). Adjustable per context.

Scoring bands:

Known Limitations

1. 3-sector aggregation. Surin's full model has 22 sectors. Our replication uses the 3-sector aggregation from his published tables. Magnitudes are 40-60% of his full model. Relative comparisons between policies are valid.
2. Static, single-period. No dynamic effects. Year 3 and Year 5 projections are extrapolations. A dynamic CGE with period-by-period migration would sharpen the trajectory.
3. 14 of ~30 instruments modeled. The current Code has additional instruments (stamp duties, registration fees, zone franche regimes, fuel restructuring) not captured. This underestimates GDP contraction and DWL.
4. η_in (formalization pull) is estimated, not empirical. The inward migration elasticity (0.12) has no direct literature source. It's the most important unknown parameter. Survey data (specifically: "If registration were free and tax were 3%, would you formalize?") could estimate it.
5. Wealth tax behavioral response assumed. The 15B HTG wealth tax revenue assumes moderate compliance. Capital flight elasticity is unestimated. Range could be 8-22B.
6. Gamma precision. The simplified budget-category aggregation dampens the Q1/Q5 differential. Full product-level TCA disaggregation would sharpen Gamma, especially for the progressive TCA scenario.
7. Φ is a structural parameter, not empirically measured. The capture rate (26% for current architecture, 5% for BARSS 7-channel architecture) is estimated from budget analysis, not from tracked fund flows.

Design Philosophy

Why this tool exists

Standard fiscal analysis tools (IMF FARI, World Bank BOOST, OECD tax models) optimize R. They tell governments how to collect more. They don't ask whether collecting more makes anyone's life better. They don't model who bears the burden differentially. They don't account for institutional capture of revenue. They don't measure whether the formal base grows or shrinks.

In a country like Haiti — 70% informal, 26% captured, poorest quintile with 8% margin — optimizing R is optimizing the wrong variable.

REAP optimizes R_useful. It asks: after collection, after capture, after distortion, after migration — what's left that reaches people?

Why Python, why simple

This tool is a single file with one dependency (numpy). No frameworks. No cloud services. No API keys. No subscriptions.

Because the people who most need to evaluate fiscal policy are in places where bandwidth is expensive, electricity is intermittent, and institutional access to modeling tools is gatekept.

A tax code that affects 11.5 million people should be evaluable by anyone with Python and the publicly available parameters from an economist's published paper.

Why attribution matters

Every equation in this calculator traces to a specific person's work. Surin built the CGE methodology. Bertil built the extraction framework. The synthesis is joint. The attribution headers in the code, the copyright notices, and this README exist because intellectual work by economists in the Global South gets absorbed without credit by institutions in the Global North. Not here.

If you use this tool, cite:

• Surin, R. (2026). "Étude d'impact du nouveau Code fiscal haïtien." For any MIFED equation.
• Bertil, W. (2026). BARSS EEDTM Framework. For any EEDTM equation.
• Bertil, W. & Surin, R. (2026). REAP Framework. For the composite equation and scoring.

File Map

19. Sorceries/
├── reap_calculator.py              ← THIS TOOL
├── REAP_CALCULATOR_README.md       ← THIS FILE
├── haiti_cge_replication_v3.py     ← Hardcoded v3 (predecessor)
├── haiti_cge_replication_v2.py     ← Hardcoded v2 (predecessor)
├── haiti_cge_replication.py        ← Hardcoded v1 (predecessor)
├── datawrapper_build_haiti_tax.py  ← Chart builder (7 DW charts)
└── brattle_build_haiti_tax_code.py ← Report builder (8 SVG charts)

8. Research Reports/Haiti_Tax_Code_Reform/
├── 00_Project_Index.md             ← Master project index
├── MIFED_Framework_Analysis.md     ← Surin's framework reconstructed
├── Cross_Validation_MIFED_x_EEDTM.md  ← 9-cell cross-validation
├── Synthesis_Framework_REAP.md     ← Academic framework paper
├── Methodology_Paper_For_Surin.md  ← Working document for collaborator
├── REAP_Calculator_Results_Analysis.md ← Detailed results + gap analysis
├── reap_calculator_results.json    ← Machine-readable output
└── cge_replication_results_v3.json ← v3 output (predecessor)

The Punchline

Under the current Code:
  R = 164B collected × 74% useful = 121B reaching people

Under the BARSS Code:
  R = 190B collected × 95% useful = 180B reaching people

+48.5% more impact. Same economy. Less pain. Better architecture.

You don't need to collect more. You need to capture less of what you collect.

REAP Calculator Results: Three Tax Policies Compared

CGE Replication v3 + REAP Scoring + Gap Analysis

Tool Reference

Policy 1: Their Code (Décret 2023)

Identification

Effective Rates

Sectoral Impact (Leontief-propagated)

Consumption by Quintile

Base Dynamics

Revenue (M HTG)

Deadweight Loss

R_useful

REAP Score: 0.51

Policy 2: Surin's Corrections (R1+R2+R3+R6)

Identification

Effective Rates

Sectoral Impact

Consumption by Quintile

Base Dynamics

Revenue

R_useful

REAP Score: 0.55

Policy 3: BARSS Code

Identification

Effective Rates

Sectoral Impact

Consumption by Quintile

Base Dynamics

Revenue (M HTG)

R_useful

REAP Score: 0.81

Master Comparison Table

Gap Analysis

What the model captures well

What the model misses or underestimates

What ONLY emerges from the REAP framework (not from MIFED alone)

What ONLY emerges from MIFED (not from EEDTM alone)

Critical unresolved questions

Validation Against Surin (2026)

Appendix: Formulas Reference

MIFED Equations (Surin, 2026)

EEDTM Extensions (Bertil, 2024–2026)

REAP Score