Paul Bidanset - Modeling, Regression and GWR

The assessment profession stands at an inflection point. While regression analysis has been with us since the 1960s, fewer than 16% of offices actually use it today. Yet as Paul Bidanset notes in his recent conversation, we're experiencing "our industrial revolution for appraisal and mass valuation."

This gap between available technology and actual implementation tells us something important about our profession: having the tools isn't enough. We need to understand how to integrate them into our existing workflows without disrupting the essential work of fair and accurate assessment.

The Promise and Reality of Mathematical Models

When assessors first started using computers in the 1960s and 70s, we called it "computer assisted mass appraisal." As Bidanset wryly observes, "I think we assist the computers at this point." The terminology may be dated, but the core challenge remains: how do we harness computational power while maintaining the transparency and defensibility our work demands?

The benefits of regression-based approaches are clear. Put ten regression modelers in a room with the same dataset, and you'll likely see more consistency than with ten single-property appraisers working independently. This isn't about replacing human judgment, it's about scaling accuracy and equity across tens or hundreds of thousands of properties.

"Mass appraisal and specifically regression analysis really allows us to, as assessors to get to scale accuracy, to scale equity, get more accurate, more uniform values in a more cost effective manner," Bidanset explains. For jurisdictions facing annual revaluation cycles with massive property counts, this isn't just an efficiency gain, it's the difference between meeting statutory requirements and falling behind.

Geographic Intelligence: The Local Within the Mass

Perhaps no innovation better exemplifies the evolution of mass appraisal than geographically weighted regression (GWR). Traditional regression treats all properties in a jurisdiction equally, but we know markets don't work that way. Historic homes in the old town center appreciate differently than tract housing on the periphery.

GWR addresses this by incorporating location directly into the analysis. As Bidanset illustrates: "As properties become older, they're closer to the end of their economic life. There's a negative association between age and price." But in historic districts, "they're very old... there's a premium that goes along with them."

This isn't just about mathematical sophistication, it's about modeling markets the way they actually behave. GWR lets us "zero in and optimize this specific area and give me coefficients that'll allow me to price out properties with respect to this area."

Currently, most offices using GWR employ it for exploratory analysis rather than final values. New York City pioneered its use for residential assessments, but for most jurisdictions, it remains "a tool in the toolkit" for identifying submarkets and refining neighborhood boundaries.

The AI Frontier: Promise With Caution

The latest wave of machine learning algorithms, neural networks, gradient boosting, deep learning, offers tantalizing accuracy gains. But as Bidanset warns, accuracy isn't everything in our profession.

"When they're sitting down with a taxpayer who comes in and says how the heck did you calculate this... assessed value of my property. They can sit down and say, well you know, we've got bathrooms rated at, you know, this price per bathroom."

This explainability requirement creates a fundamental tension with black-box algorithms. While these models can achieve remarkable accuracy metrics, Bidanset mentions achieving a 5% coefficient of dispersion with gradient boosting, they often overfit the data. Applied to holdout samples, that impressive 5% can balloon to 35%.

Still, AI finds its place in supporting roles. Deep learning excels at property condition assessment through image analysis, identifying overgrown lots or structural issues that would take armies of field assessors to catalog. These tools enhance rather than replace traditional valuation methods.

Lessons from the Field: Moldova's Digital Transformation

Sometimes the most instructive examples come from unexpected places. Moldova's property tax modernization effort demonstrates what's possible when data infrastructure catches up with analytical ambitions.

"Even just from 2015 until now, a massive overhaul with their data. It's much more complete, it's much more accurate and data is the ingredient... for property valuations," Bidanset notes. The same models that struggled with incomplete cadastral data now perform remarkably better simply because the underlying information improved.

This underscores a crucial point: sophisticated algorithms can't overcome bad data. Moldova's success came from addressing fundamentals first, building comprehensive registries, ensuring data quality, then applying appropriate analytical tools.

The Path Forward

Perhaps Bidanset's most valuable insight addresses not technology but psychology: "Everybody thinks that they're the only one with a dirty house, and they don't realize that everybody's house is messy and everyone's behind cleaning it up."

This confession-booth mentality holds us back. Whether struggling with rising CODs, staff retention, or technology implementation, assessors often feel uniquely challenged. The reality? "Every problem that an assessor has, a million other people have that same problem."

The offices making progress aren't those without problems, they're those willing to seek solutions. As Bidanset encourages, "I've seen the most disadvantaged offices turn around and implement and benefit from it."

Key Takeaways

Technology adoption remains slow but steady. While regression analysis has existed for decades, most offices haven't implemented it. This isn't failure, it's recognition that sustainable change takes time.

Geographic intelligence matters. Tools like GWR acknowledge that real estate markets are inherently local. The challenge is incorporating this sophistication while maintaining transparency.

AI serves best in supporting roles. Deep learning for property condition assessment, benchmark models for validation, these applications enhance rather than replace core valuation work.

Data quality trumps algorithmic sophistication. Moldova's transformation shows that improving data infrastructure yields better results than implementing cutting-edge algorithms on poor data.

You're not alone. Every assessor faces similar challenges with technology adoption, staff resources, and maintaining accuracy. The successful offices are those willing to ask for help and share their experiences.

The revolution in mass appraisal isn't about replacing assessors with algorithms. It's about equipping professionals with tools that scale their expertise across entire jurisdictions while maintaining the fairness, transparency, and local knowledge that define good assessment practice.