Vision
Forecasting the Future of Real Estate
Stellar Families, a family-run real estate investment firm based in Texas, wanted to level up their approach to evaluating metropolitan statistical areas (MSAs) for future investments.Their goal was to pinpoint early signs of market growth—aligning perfectly with the two- to three-year timeline of their build-to-sell developments.
They weren’t just looking for data—they were looking for insights and a way to unify their dataset.
Challenge
Bringing Structure to a Complex Market
Real estate forecasting is notoriously difficult. Markets like Austin or Phoenix experienced sudden and precipitous growth for a unique mix of reasons, and identifying the root cause of the growth in these and other regions is far from straightforward.
Stellar believed there were underlying historical patterns that could be analyzed to identify metropolitan regions likely to experience future growth. But the team quickly discovered that traditional signals—those they'd used for years—often didn’t tell the whole story.
Our challenge was twofold:
- Build a robust forecasting model that could handle a high level of complexity under shifting market conditions.
- Present results in a way that fits seamlessly into Stellar’s existing workflows and expectations.
Breakthrough
Turning Raw Data Into Strategic Signals
As the engagement progressed, our work shifted from pure forecasting to identifying and validating key market indicators. This pivot aligned with Stellar’s preference for enhancing their manual process rather than replacing it outright.
To accelerate this, KUNGFU.AI:
- Achieved a 20% median Mean Absolute Percent Error (MAPE) over a five-year forecast horizon when projecting real estate market growth across key metropolitan areas—a strong performance given the complexity of long-range financial forecasting and competitive relative to industry benchmarks.
- Validated model accuracy through external domain experts who confirmed the results represented best-case performance for predicting long-term trends in housing demand, market strength, and investment potential across metropolitan areas.
- Identified compelling new data correlations, providing fresh insights into market dynamics that went beyond traditional indicators.
This project pushed beyond traditional methods. We evaluated advanced time series forecasting techniques to achieve top results, including unidimensional convolutional neural networks, boosted trees, ARIMA, and other cutting-edge approaches.
Outcome
An Advance AI Real Estate Forecasting System
By the end of the engagement, Stellar had a fully operational forecasting system built on a live, automated ML pipeline that pulled data from RealPage, StratoDem, and the Federal Reserve Economic Data (FRED) repository.
We:
- Normalized and integrated these sources into a unified dataset.
- Deployed an auto-refreshing system that updates predictions monthly via their existing Domo dashboard.
- Delivered a true active machine learning system that requires no manual retraining.
We also provided recommendations for future enhancements, such as incorporating sentiment or culture-related data to enrich future models.
This work didn’t just modernize Stellar’s investment process—it set them up with infrastructure and insights to scale smarter, forecast better, and invest with more confidence.
