LVT Cape Town Project: Kick-off Meeting

RIO Mentorship Programme

Author

Peter Courtney

Published

February 2, 2026

1 Welcome & Introductions (15 min)

I want to start by saying that I am here primarily to support your ambition on this project. I have outlined some interesting research questions and provided the data; I am happy to take this as far as you want to go. This is your project, and I am here to guide and enable you.

Note on today’s meeting: I cannot attend this meeting in person, but this will be the only time I cannot meet in person. I have a strong preference for meeting face-to-face. I am currently at the Blavatnik School of Government as a Recognised DPhil Student; we can meet there or anywhere else in Oxford that works for you.

1.1 About Me

I’m Peter Courtney, a joint PhD candidate in Economics at VU Amsterdam (Tinbergen Institute) and Stellenbosch University. I am currently at Oxford as a Recognised DPhil Student at the Blavatnik School of Government. My research focuses on development economics and education policy in Africa.

Relevant background for this project:

  • Hold the 2025 C. Lowell Harriss Dissertation Fellowship from the Lincoln Institute of Land Policy (specifically for property valuation and taxation research)
  • Emergent Ventures Grant for “Machine Learning Land Valuation in Cape Town”
  • Published with Oxford University Press and Brill; policy work for UNU-WIDER and South African National Treasury
  • Previous positions at J-PAL Africa and UN FAO
  • I wrote an op-ed in News24 Business arguing that “Land value tax is SA’s best shot at land reform”

Why I care about this project: Cape Town is one of the most spatially unequal cities on Earth, a direct legacy of apartheid planning. LVT offers a path towards addressing South Africa’s land question without confiscation. The missing piece for political viability may be showing voters their personalised “shadow bill”; whether this actually benefits the median voter is an empirical question we need to investigate.

Why Cape Town specifically: Cape Town is a very innovative city, ranked Africa’s most efficient metro, with a reform-minded administration. Using Cape Town as a case study could have outsized impact, both for South Africa and as a demonstration for cities worldwide.

1.2 Team Introductions

Let’s go around and share:

  1. Your name
  2. One sentence on why you applied to this specific project
  3. Something interesting about you that isn’t on your application

1.3 Schedule the Next Meeting

First priority: Before we leave today, we need to agree on a recurring meeting time and schedule our next meeting.

2 Project Understanding Check (15 min)

2.1 Exercise: Explain the Project

Task: In your own words, briefly explain (2-3 sentences each):

  1. What is a Land Value Tax and why is it considered efficient?
  2. What is the specific problem we’re trying to solve with this project?
  3. Who is our target audience for the outputs?

Key points to confirm:

  • We are building a tool that shows every household in Cape Town how their taxes would change under LVT
  • Whether most homeowners would pay more or less is an empirical question we need to investigate; we hypothesise that the burden would shift towards speculators and owners of underutilised prime land, but this must be tested
  • A critical step is backing out current estimates of how much people pay in rates, so we can calculate how their bill might change under different scenarios
  • Our outputs include an interactive map, a policy brief for Cape Town municipality, and a dataset of estimated land values

2.2 Common Misunderstandings to Address

  • This is NOT about increasing total tax revenue; it is revenue-neutral
  • We are separating land value from improvement value (buildings, renovations, etc.)
  • We do not presume outcomes; we investigate them empirically

2.3 A Note on Rents vs Land Values

An important conceptual distinction:

  • Stock of land values: The capitalised market price of land at a point in time (what the land would sell for)
  • Flow of rents: The annual rental income or imputed rent that land generates

These require different calculation approaches. The flow approach may be more relevant for tax policy (since property taxes are annual) and can be more robust to fluctuations in discount rates and market sentiment. We should calculate both and compare them, as each has advantages for different purposes.

3 Questions from the Team (10 min)

What questions do you have about:

  • The project scope?
  • The methods we will use?
  • Cape Town specifically?
  • Anything from the background reading?

4 Expectations & Time Commitment (10 min)

4.1 My Expectations of You

  1. Proactive communication: If you are stuck, confused, or behind, tell me early. No surprises.
  2. Quality over quantity: I would rather you do less work well than rush through tasks poorly.
  3. Intellectual honesty: Flag uncertainties. Say “I don’t know” when you don’t. Question my assumptions.
  4. Collaborative spirit: Support each other. Share resources and insights. This is a team project.
  5. Professional standards: Meet deadlines. Cite sources. Write clearly.
  6. Reproducibility as a golden rule: Everything you produce must be “one click” reproducible. This means clean code, clear documentation, and no manual steps that cannot be scripted.

4.2 Your Time Commitment

Based on your applications:

Team Member Hours/Week Notes
Ami 6-8 hours Potentially more depending on the week
Annika ~5 hours
Guilherme 7-8 hours

Total team capacity: Approximately 18-21 hours/week

4.3 My Commitment to You

  • Response time: I will respond to messages within 5 hours on weekdays
  • Meetings: I will meet with you once a week, or however frequently you would like to meet. I have a strong preference for meeting in person.
  • Feedback: I will provide constructive feedback on all work submitted
  • Reading: I will read anything and everything you send to me, provided you have read it entirely and are willing to do a mini reading group on it with me
  • References: I will write references and support your career goals where I can
  • Tools: I will pay for £20 worth of Cursor credits for each of you. This is not a lot, so I suggest you use the Composer model to stretch your credits further. Please download and set up Cursor.
  • Time and effort: I am very happy to put a lot of time and effort behind this project. Your ambition is the limiting factor, not mine.

6 Project Methodology (20 min)

6.1 Phase 1: Predicting Total Property Values with XGBoost

The first econometric piece will be attempting to estimate total property prices and outperform the city’s own valuations. We will use XGBoost for this.

Why this matters: If we can predict property values better than the city, we establish credibility and understand the drivers of property values in Cape Town.

First step: Creating extensive descriptive tables (using the summarytools package in R) and maps (using leaflet). This exploratory data analysis is essential before any modelling.

6.2 Phase 2: Alternative Land Valuation Methodologies

The second major piece is deriving alternative paths to land valuation and seeing if we can get them to align. Different approaches include:

  1. Spatial matrix of empty land sales: Using actual sales of vacant land to estimate land values directly
  2. Hedonic models with residual/spatial approach: Estimate total property value using hedonic methods, then either subtract estimated improvement value to get the residual as land value, or use the spatial component of the model to estimate land values based on location

A key goal is to compare these methodologies and understand where they agree and disagree.

6.3 Phase 3: Tax Incidence Simulation

Once we have land value estimates, we simulate household-level tax incidence under different LVT reform scenarios. This requires:

  • Backing out current rates payments from the data
  • Modelling revenue-neutral rate structures
  • Generating “shadow bills” for each property

6.4 Phase 4: Policy Communication

We need to be very clean with the policy work. Our approach:

  1. Start with an academic white paper: Rigorous methodology, full documentation, peer-review ready
  2. Distill into policy notes: Accessible summaries for policymakers and the public

6.5 Stretch Goal: Historical Research on the Transvaal

For interested parties, there is a potential stretch goal project: creating a research proposal to study the impact of shifting away from land value taxation towards total value taxation in the Transvaal. The key legal instrument was the Local Authorities Rating Ordinance of 1933.

This would require archival and historical work. It is only suitable if the econometric work is not a perfect match for someone’s interests and skills. Let me know if this appeals to you.

7 Potential Audiences for This Work (5 min)

If we produce good work, there are people who want to see it:

  • Lars Doucet and Greg Miller at the Center for Land Economics: They are working on a startup regarding LVT and mass appraisal. I can likely get an audience with them for this work.
  • Francis DiTraglio: He is also interested in this work and could be a valuable contact if we create something of quality.
  • Young Urbanists and Development Action Group (DAG): Potential organisations for networking, but this must be done with extreme caution and my guidance.

The intellectual work must take precedence. Getting an audience with the right people is a secondary challenge that we can address once we have solid results.

8 Main Assumptions & Risks (15 min)

8.1 Ways This Project Could Go Wrong

Let’s discuss these honestly:

8.1.1 Data Risks

  • Data quality: Property records may have errors, missing values, or inconsistent geocoding. Data cleaning will be critical.
  • Land vs improvement values: Splitting these accurately is technically challenging; this is what the project is about.

8.1.2 Methodological Risks

  • Model accuracy: Machine learning for property valuation has known limitations. What is our acceptable error margin?
  • Legibility of ML vs traditional models: XGBoost and similar ML models can be “black boxes.” We need to think carefully about how to make our estimates legible and interpretable, compared to traditional spatial econometric models where coefficients have clear meanings.
  • Shrinkage estimators and bias: Many ML approaches introduce bias through regularisation (shrinkage). In sensitive policy settings like taxation, we need to understand what it means to introduce bias and whether this is acceptable.
  • Assumptions in simulation: Tax incidence modelling requires assumptions about revenue neutrality, rate structures, and exemptions.
  • Generalisability: Will our Cape Town model be useful for other cities?

8.1.3 Political/Impact Risks

  • Getting an audience with the right people: Networking with organisations such as Young Urbanists and Development Action Group (DAG) could become part of the project, but this must be done with extreme caution and my guidance before you jump into anything.
  • Premature communication: Because the City of Cape Town must approve final publications, we need to be careful about what we share and with whom before approval.
  • Oversimplification: Policymakers and homeowners may not understand the nuances.
  • Cherry-picking: Results could be misused by advocates on either side.

8.1.4 Team Risks

  • Skill gaps: We may encounter technical challenges none of us have faced before.
  • Coordination: Working across different schedules.
  • Scope creep: Trying to do too much with limited time.

Discussion: Which of these risks concern you most? What others can you think of?

9 Team Backgrounds, Strengths & Goals (20 min)

9.1 Quick Profiles

Note: These are based on your applications. Please let me know if you would like to make any edits or corrections.

9.1.1 Ami Rubiés

  • Programme: Economics & Philosophy (2nd year), Wadham College
  • Key strengths: Advanced R (biostatistics, Monte Carlo simulations), LaTeX, ggplot2, data visualisation, data journalism experience at The Times
  • Relevant interest: Attended Invisible College (Works in Progress); influenced by Seeing Like A State
  • Career goal: Masters in Development Economics → ODI Fellowship → World Bank Young Professionals Programme

9.1.2 Annika Burman

  • Programme: International Studies & Asian Studies (3rd year), University of Michigan (visiting Oxford)
  • Key strengths: Python, R, SQL, agent-based modelling, policy analysis, professional writing (State Department), EA leadership
  • Relevant experience: Fish Welfare Initiative internship (working with uncertainty), Brainomix (South Africa healthcare research)
  • Career goal: Political nonprofit space → eventually founding her own nonprofit(s)

9.1.3 Guilherme Lopes

  • Programme: History & Economics (2nd year), St John’s College
  • Key strengths: Python (web scraping, data cleaning), consulting experience (Capitox), policy briefs, journalism (25,000+ views)
  • Relevant experience: Central Bank of Portugal internship (price data scraping for inflation analysis)
  • Career goal: Journalism, economic research, or international organisations; connecting research to real-world impact

9.2 Discussion Questions

For each team member:

  1. What aspect of this project are you most excited about?
  2. What is a skill you are hoping to develop through this project?
  3. What is something you are nervous about or feel less confident in?

10 Roles & Division of Labour (10 min)

Important: These proposed roles are completely random at this point and need extensive discussion. Consider them a starting point for conversation, not assignments.

10.1 Proposed Initial Roles (For Discussion)

Area Primary Support
Literature Review Guilherme All
Data Cleaning & Exploration Ami Guilherme
Spatial Analysis & GIS Ami Annika
ML Modelling (XGBoost) All (collaborative learning)
Alternative Valuation Methods TBD
Simulation & Tax Incidence TBD
Academic White Paper All
Policy Brief Writing Guilherme, Annika Ami
Visualisation & Interactive Tool Ami Guilherme
Stakeholder Networking (with guidance) Guilherme Annika
Historical/Archival Research (stretch) TBD (if interested)

10.2 Additional Data Acquisition (If Needed)

Once you start exploring the data, you may identify gaps that require additional data. Potential sources:

  • Deeds Office: Historical transaction records
  • Property24: Listing and sales data
  • Google Earth Engine: Remote sensing data (e.g., detecting improvements between two time periods)

To acquire additional data, submit a proposal explaining why it is valuable. I will then decide whether to purchase it.

11 Essential Skills to Develop (5 min)

Through this project, you will learn:

  • Programmatic GIS: Geospatial analysis in R/Python (not just point-and-click software), including the exactextractr package for zonal statistics
  • Google Earth Engine: Remote sensing and satellite imagery analysis
  • Machine learning for valuation: XGBoost and related methods
  • Quarto for reproducible reporting: All reports should be written in Quarto
  • Working with large administrative datasets
  • Translating academic economics into policy-relevant outputs

A note on programming languages: I have a strong preference for R over Python, but you should use the tools that suit you best. The important thing is that your work is reproducible and well-documented.

12 Further Reading by Background (5 min)

12.1 Core Reading (Everyone)

12.2 Ami: Econometrics & Spatial Methods

Given your strong quantitative background:

12.3 Annika: Policy Analysis & Implementation

Given your policy and agent-based modelling experience:

12.4 Guilherme: Communication & Political Economy

Given your journalism and consulting experience:

13 Logistics (15 min)

13.1 Meeting Schedule

Proposed: Weekly meetings, [DAY] at [TIME]

  • What day/time works best for everyone?
  • Are there any weeks you already know you will be unavailable?
  • I have a strong preference for meeting in person. We can meet at the Blavatnik School of Government or anywhere else in Oxford.

13.2 Preparation Expectations

  • Submit your weekly work at least 24 hours before our scheduled meeting
  • This allows me to review and prepare feedback
  • “Work” includes: code, drafts, questions, or a brief update on progress

13.3 Communication

My preference: Slack for everything. Set up a Slack workspace and add me.

Reporting: Please learn how to use Quarto for all reporting. This ensures reproducibility and professional output.

Response time expectations:

  • I will respond within 5 hours on weekdays
  • Urgent issues: flag with “URGENT” and I will respond ASAP

13.4 Tools & Platforms

  • AI coding assistant: Cursor (I will provide £20 credits each; use the Composer model to stretch your credits)
  • Code: GitHub repository (I will set this up)
  • Documents: Google Drive shared folder
  • Reporting: Quarto
  • Communication: Slack
  • References: Zotero shared library (optional but recommended)
  • GIS/Remote sensing: Google Earth Engine (free, requires setup); exactextractr for zonal statistics

14 Project Timeline (10 min)

14.1 Proposed 8-Week Timeline

Week Focus Key Deliverable
1 Kick-off, background reading, data exploration Project understanding confirmed; descriptive tables and initial maps
2 Literature review, continued data exploration Annotated bibliography
3-4 Data cleaning, initial spatial analysis, Phase 1 modelling Clean dataset, preliminary XGBoost model
5-6 Phase 2: Alternative land valuation methods Comparison of methodologies
7 Tax incidence simulation, visualisation Draft interactive tool, draft white paper
8 Policy brief, final presentation Final outputs

Discussion: Is this timeline realistic given your other commitments? Where do we have flexibility?

14.2 Key Dates

  • Final presentation: TBD
  • Policy brief deadline: TBD
  • Exam periods/conflicts to note:

15 Next Steps (10 min)

15.1 Defining “Good Enough”

For each task, we will be explicit about:

  1. Minimum viable output: What is the baseline that counts as “done”?
  2. Stretch goals: What would “excellent” look like?
  3. Person responsible: Who owns this deliverable?
  4. Deadline: When is it due?

15.2 This Week’s Tasks

Task Owner Deadline “Good Enough”
Schedule next meeting All Today Date and time confirmed
Complete core reading (4 articles) All [DATE] Notes/summary shared
Set up Slack workspace [TBD] [DATE] Everyone added
Set up Cursor with credits All [DATE] Working installation
Sign Data User Agreement and NDA All [DATE] Signed documents returned
Initial data exploration: descriptive tables Ami [DATE] summarytools output shared
Initial data exploration: maps [TBD] [DATE] Basic leaflet maps of property values
Draft annotated bibliography outline Guilherme [DATE] Structure and 5+ key sources identified
Review Municipal Property Rates Act Annika [DATE] 1-page summary of relevant provisions

16 Final Questions & Reflections (5 min)

  1. What is one thing you are taking away from this meeting?
  2. Is there anything you are still unclear about?
  3. What support do you need from me to succeed?

17 Contact Information

Peter Courtney

Team:

18 Appendix: Glossary of Key Terms

For reference throughout the project:

Term Definition
LVT Land Value Tax, a tax on the unimproved value of land
Split-rate taxation Taxing land and improvements at different rates
Deadweight loss Economic inefficiency caused by taxation
Valuation roll Municipal register of property values
Shadow bill Personalised estimate of how taxes would change under reform
Georgism Economic philosophy based on Henry George’s work, advocating land value capture
Municipal Property Rates Act Local Government: Municipal Property Rates Act 6 of 2004 (South Africa)
GIS Geographic Information System, software for spatial analysis
Mass appraisal Techniques for valuing many properties simultaneously
XGBoost Extreme Gradient Boosting, a machine learning algorithm
Hedonic model Statistical model that estimates property value based on characteristics
Flow of rents Annual rental income or imputed rent from land
Stock of land values Capitalised market price of land at a point in time
Shrinkage estimator Statistical method that introduces bias to reduce variance

Document prepared for the RIO Mentorship Programme kick-off meeting