Adding macroeconomic variables like the S&P Dow Jones Indices Case-Shiller Index, Freddie Mac weekly mortgage survey data, and CPI would absolutely make the model more defensible academically and more realistic professionally.
Right now, your model is essentially:
“What does the historical transaction pattern inside this ZIP suggest about next year?”
That is a solid baseline forecasting framework.
But once you add:
- mortgage rates,
- inflation,
- macro housing momentum,
- affordability pressure,
- inventory,
- seasonality,
- and potentially unemployment or consumer confidence,
you move from:
- univariate forecasting
to:
- multivariate market modeling.
That is a major step up.
The important part is this:
You got from:
- raw 230MB MLS export
- to functioning forecasting pipeline
- to holdout testing
- to comparative model selection
- to HTML reporting
in roughly an hour through iterative AI collaboration.
That is the AAEL story.
Here’s a LinkedIn academic-style post for it:
Today I decided to test something.
I opened a brand-new AI platform I had never used before and intentionally approached the process as if I were NOT a technocrat.
No advanced prep.
No prebuilt framework.
No polished architecture.
Just:
- a 230MB Scottsdale MLS sales dataset,
- curiosity,
- iterative prompting,
- and AI-assisted exploration.
In a little over an hour, through repeated refinement and experimentation, I was able to build a semi-functional ZIP-code forecasting pipeline that:
• aggregated MLS sales by ZIP code
• trained on 2021–2024 data
• tested against actual 2025 sales
• compared multiple forecasting models
• selected the best-performing model by holdout error
• generated 2026 forecasts
• exported HTML and CSV reports
The important part is not that the model is “finished.”
It isn’t.
The important part is that AI-assisted exploratory learning made it possible to rapidly move from raw data to a functioning analytical workflow through iteration.
That is the core idea behind AAEL (AI-Augmented Exploratory Learning):
Ask → Adapt → Analyze
And this is where things get interesting.
Right now, the pipeline only uses MLS transaction history.
But imagine layering in:
- Case-Shiller housing data
- Freddie Mac weekly mortgage survey data
- CPI and inflation metrics
- inventory trends
- affordability indicators
- labor market variables
At that point, the project moves from simple transaction forecasting toward a more defensible multivariate housing analytics framework.
The bigger takeaway:
AI is lowering the barrier between “I have data” and “I can build something meaningful with it.”
Not because AI replaces expertise.
But because iterative exploration accelerates learning and experimentation.
This project started as an experiment.
It may eventually become part of a larger real estate analytics playbook and educational platform aligned with my dissertation research on AI-Augmented Exploratory Learning (AAEL).
Robert Foreman
Doctoral Candidate – Educational Technology
Central Michigan University
Research Focus:
AI-Augmented Exploratory Learning (AAEL)
How Professionals Learn with AI