background.docx

Boulder Housing Affordability & Displacement Risk: Predictive Modeling Project Plan
Research Question
How can we predict housing affordability trends and quantify neighborhood-level displacement risk in Boulder, Colorado using integrated economic and housing market data?
Model Overview
We will develop a hybrid predictive framework consisting of two complementary components:
Time Series Forecasting Model: Predicts aggregate housing affordability metrics for Boulder 6-12 months ahead using ARIMA/SARIMAX with exogenous economic variables from FRED
Spatial Risk Model: Assigns displacement risk scores to Boulder neighborhoods using logistic regression with economic stress indicators and housing supply dynamics
Technical Approach
Language: Python
Key Libraries:
Data: pandas, numpy, fredapi
Modeling: statsmodels, scikit-learn, prophet
Spatial: geopandas, folium
Visualization: matplotlib, seaborn, plotly
Data Sources:
FRED API: Denver-Boulder MSA economic indicators (income, employment, CPI-housing)
Boulder Open Data / County Assessor: Housing prices, construction permits
US Census Bureau: Demographic data at census tract level
Why This Matters
Policy Impact: Boulder city planners need forward-looking tools to anticipate affordability crises and identify vulnerable neighborhoods before displacement occurs.
Methodological Rigor: By combining time series econometrics with spatial risk modeling, we create a statistically robust framework that is reproducible and audit-ready.
Equity Considerations: Quantifying displacement risk with transparent metrics allows policymakers to prioritize interventions in communities most at risk, promoting equitable housing outcomes.
Step-by-Step Implementation Plan
1. Data Acquisition & Preprocessing
Set up FRED API access and pull monthly time series (2015-2024) for:
Median household income (Denver-Boulder MSA)
Unemployment rate
CPI for housing
Employment growth by sector
Acquire Boulder housing data:
Median home prices by zip code/census tract
Construction permits issued (monthly counts)
Rental vacancy rates if available
Merge datasets on common temporal and spatial keys
Handle missing values, outliers, and ensure stationarity for time series components
2. Feature Engineering
Calculate derived affordability metrics:
Price-to-income ratio (housing affordability index)
Percentage of income required for median mortgage/rent
Create displacement risk indicators:
Rolling 12-month price appreciation rate
Income growth gap (housing price growth - wage growth)
Supply pressure (new permits / existing housing stock)
Generate lag features for predictive power (3, 6, 12-month lags)
Spatial features: distance to Boulder downtown, transit access, proximity to employment centers
3. Exploratory Data Analysis
Visualize time trends in housing prices, income, and affordability ratios
Identify structural breaks (e.g., COVID-19 impact)
Examine spatial heterogeneity across Boulder neighborhoods
Correlation analysis between economic indicators and housing metrics
Test for stationarity using ADF/KPSS tests
4. Model Development - Time Series Component
Train ARIMA/SARIMAX model for Boulder aggregate housing affordability index
Include exogenous variables: unemployment rate, CPI-housing, construction permits
Perform grid search for optimal (p, d, q) parameters
Validate using rolling window cross-validation (train on 2015-2022, test on 2023-2024)
Generate 6-month and 12-month ahead forecasts with confidence intervals
5. Model Development - Spatial Risk Component
Define displacement risk binary outcome:
"High risk" = neighborhoods where price growth >15% AND income growth <5% over past 12 months
Train logistic regression model with features:
Economic stress (unemployment × price appreciation)
Supply constraints (low construction activity)
Demographic vulnerability (if census data available)
Use cross-validation to tune regularization parameters
Generate predicted probabilities for each Boulder census tract
6. Model Validation & Diagnostics
Time series model: Check residuals for autocorrelation, normality, heteroskedasticity
Spatial model: ROC curve, precision-recall, calibration plots
Sensitivity analysis: How do predictions change with ±10% shifts in key assumptions?
Out-of-sample performance metrics: RMSE, MAE for time series; AUC, F1-score for classification
7. Assurance Dossier Construction
Technical Documentation: Jupyter notebook with full reproducible workflow
Statistical Report: LaTeX document covering methodology, validation, limitations
Data Provenance: Clear documentation of all data sources, transformations, and assumptions
Bias & Ethics Assessment:
Identify potential biases in data (e.g., missing data on informal housing)
Discuss who might be harmed by model predictions
Propose mitigation strategies (e.g., human-in-the-loop for policy decisions)
Interactive Dashboard: Plotly/Streamlit app showing forecasts and risk maps
8. Visualization & Communication
Time series plots: Forecasted affordability trends with confidence bands
Interactive risk map: Boulder neighborhoods color-coded by displacement risk score
Comparative metrics: Boulder vs. Denver vs. National averages
Scenario analysis: "What if" visualizations (e.g., 20% increase in construction permits)
9. Deliverables
Python codebase (GitHub repository with clean structure)
Technical report (15-20 pages, LaTeX)
Executive summary for policymakers (2-page brief)
Interactive dashboard for exploration
Presentation slides summarizing key findings
Success Criteria
Time series model achieves RMSE <5% of mean housing affordability index on test set
Spatial risk model achieves AUC >0.75 for predicting high-risk neighborhoods
All code is reproducible with clear documentation
Deliverables are accessible to both technical and non-technical audiences
Ethical considerations are explicitly documented and addressed
Timeline Estimate
Data acquisition & preprocessing: 1 week
EDA & feature engineering: 1 week
Model development & validation: 2 weeks
Assurance dossier & deliverables: 1 week
Total: 5 weeks for complete project
Risk Factors & Mitigation
Risk: Data availability issues for Boulder-specific metrics Mitigation: Use Denver MSA data as proxy; supplement with Zillow or Redfin APIs
Risk: Model overfitting on limited Boulder data Mitigation: Use regularization; consider pooling with similar-sized Colorado cities
Risk: Ethical concerns about perpetuating displacement through prediction Mitigation: Frame model as early warning system for intervention, not deterministic forecast
Next Immediate Action
Set up Python environment and pull initial FRED data for Denver-Boulder MSA to validate data availability and structure.