Using Home Price Indexes for Time Adjustments

As we round the corner to the February 4th date when the GSE's will require market support for “time” adjustments, there have been a lot of posts by several appraisers all trying to help their colleagues with meaningful and practical ways to support their adjustments.? Kudos to all of you who have been helping on LinkedIn, YouTube, and other (more toxic) social media platforms.

Admittedly, I am not a huge fan of home price indexes (HPIs) if you have other tools that will work.? Paired sales (gasp), regression analysis, grouped sales data sets, and other techniques have already been covered. This post is intended to address HPIs and how you can use them. Those in thin markets with very few sales may want to perform more in-depth analysis, but they cannot make up data.? So, what are the rural and smaller market appraisers to do?? There is no exemption from the requirement to provide market support.

On the pages that follow, I will provide a background on indexes and then show how I developed models using artificial intelligence to handle the heavy lifting.? The results are two easy to use “websites” you can ask me for in the comments. These models can be used to your hearts content to perform the calculations and demonstrate graphically and tabularly what is happening in your market. Hopefully it’s worth a quick read.

TRIGGER WARNING: For those easily offended, the next few pages mention the "Z" word multiple times.?I get it, but go grab your thunder blanket and wrap up as you follow along.

Appraisers can use local property price indexes to support time adjustments, especially when local sales data is limited or a broader market perspective is needed. These indexes, such as the S&P CoreLogic Case-Shiller Index, the Freddie Mac House Price Index, or the Zillow Home Value Index (ZHVI), can be used to determine how values have changed over a specific period and can be applied as an adjustment factor to comparable sales prices.

Here are some key points regarding the use of indexes:

Consistent with Fannie Mae policy: The use of home price indices (HPIs) to support time adjustments is consistent with Fannie Mae policy.

Broader Market Perspective: Indexes can provide a broader market view, which is particularly useful when local sales data is scarce.

Quantifiable Adjustments: Indexing allows for a data-driven approach to adjustments, which is a Fannie Mae goal.

Various acceptable methods: Fannie Mae guidelines state that adjustment rates can be determined through statistical analysis, modeling, paired sales, or other commonly accepted methods. Indexing is one such acceptable method.

Transparency and Support: Appraisers must clearly document the data sources, tools, and techniques used when applying time adjustments using indexes.

Supplement to other methods: Indexing can validate other methods of time adjustment by providing a secondary source of evidence to bolster conclusions.

Specific Indexes:

• S&P CoreLogic Case-Shiller Index and the Freddie Mac House Price Index: These indexes are useful for understanding market trends at both national and metropolitan levels. They are a reliable benchmark for changes in property values over time and is helpful when recent comparable sales are limited.
• Zillow Home Value Index (ZHVI): This index represents the "typical" home value for a region, calculated as a weighted average of the middle third of homes in a given region. It is available in a variety of geographies or as they call it, cuts, consisting of different home types and adjustments. The most-cited ZHVI is the all homes, middle tier, smoothed and seasonally adjusted cut. A raw, non-smoothed version is also available and shows shifts in market conditions more quickly, though it may also show more "noise" in results from month to month. Zillow’s ZHVI should be cited as the “typical home value for the region,” and not as the median home value.
• As much as so many don’t care for the reliability of Zillow, at least they provided geographic specificity down to the zip code level.? Not ideal for a lot of reasons, but possibly better than the city, county, MSA/CBSA, or state level minimums of other indexes.? And I guess if FHFA is going to use this data set to say appraisers are biased, it is likely going to be very difficult for the GSEs to tell appraisers its not an appropriate source.

How to apply: An appraiser may use a local index to determine the average change in property values over a certain period and then apply that change to a comparable sale based on the time difference between the contract date of the comparable sale and the effective date of the appraisal. For example, an appraiser may find that property values have increased by an average of 0.5% per month over the past six months, and use that to support a 3% market condition adjustment for a comparable sale that occurred six months prior to the effective date of the appraisal.

Not the only method: While the use of indexes is acceptable, Fannie Mae does not require appraisers to use any specific method of time adjustment and has said that they're not looking for a specific type of graph. The goal is to use data to create an adjustment that reflects the market's reaction to the passage of time.

Data-Driven Adjustments: Fannie Mae emphasizes that adjustments must be based on market changes between the contract date of the comparable sale and the effective date of the appraisal. The adjustments should be based on data and not on a set of rules or arbitrary limits.

Support and Documentation: Appraisers must summarize the evidence, describe the data source, and describe the tools and techniques they used when applying time adjustments.

Using Generative AI to Develop Models

So now with that out of the way, how about something useful?? For the models I will happily send if you request them in the comments, I used Claude.ai.? Pick your favorite Generative AI model and likely they can do the same. But I wanted to show Claude some attention this week after experimenting the last few days with DeepSeek R1 (Oh my! A Chinese company that resulted in a $1 trillion dollar loss to the stock market this week).? What follows is not a complete step by step, but the general process I followed.?Frankly, this took about an hour to finish, but I had a working model after the very first prompt (Note: learn how to prompt if you want success with Gen AI!).? I just kept tweaking.

How to Work with Claude to Develop Custom Web Tools: A Step-by-Step Guide

Introduction

This guide explains how to collaborate with Claude to create custom web-based tools, in this case using the Zillow Home Value Index (ZVHI).? Again – you are in a safe space so don’t get upset.? As noted, I used this source because it is the only source I could find that reported information down to the smallest geographic area (zip code). I have also provided a similar tool using the Freddie Mac House Price Index (FMHPI?), but its limitation is that it only goes down to the CBSA level.? ?While you don't need to be a programmer to follow this process, having a clear vision of what you want to achieve is helpful.

The Development Process

Step 1: Gather Your Requirements

Before starting with Claude:

• Collect any necessary data files (in this case, the ZHVI CSV file) which you can get from Zillow (free, and I know appraisers like free) and it is updated every month on the 16th.
• Write down the key features you want (like dropdowns, charts, tables)
• Note any specific styling preferences or branding requirements
• Identify any external sources or attributions needed

Step 2: Start the Conversation with Claude

Begin by explaining your project at a high level. In this case, I:

1. Shared a portion of the ZHVI data file (I couldn’t share the whole thing because it was too large.? But I shared about 100 rows of data, so it had “real” data to work with.
2. Explained I wanted to create a calculator to determine changes over time and wanted it ultimately to be a stand-alone type product.
3. Specified the key features we wanted (Zip code selection, time slider, graph)

Tip: You don't need to have everything figured out at the start. Claude can help refine your ideas through conversation.

Step 3: Iterative Development

I used an iterative approach, which worked well:

1. First Version: We got a basic working version with: File upload Zip Code selection Time period slider Basic graph
2. Refinements: I then made improvements one at a time: Added a data table below the graph Included attribution text Customized the colors to match UT's burnt orange because frankly what other color would a reasonable person choose?

Key Point: Make one change at a time. This makes it easier to review and adjust each modification.

Step 4: Testing and Feedback

After each change:

• Review what Claude provided
• Test if it meets your needs
• Provide specific feedback
• Ask for adjustments as needed

Tips for Working with Claude

Be Clear and Specific

Good example: "Can you add a table below the graph that shows dates in the first column, index values in the second column, and monthly percentage adjustments in the third column?"

Provide Context

• Share relevant files at the start
• Mention any existing tools or examples you're drawing from
• Explain your preferences (like branding colors)

Iterate Gradually

Instead of requesting many changes at once:

• Make one change at a time
• Review the results
• Then move on to the next change

Ask for Explanations

If you don't understand something, ask Claude to:

• Explain what it did
• Clarify how something works
• Provide more details about specific features

Common Challenges and Solutions

Challenge: Unclear Requirements

• Solution: Start with a basic version and refine it through conversation
• Example: I started with a simple graph and added features like the data table later

Challenge: Technical Details

• Solution: Let Claude handle the technical implementation while you focus on what you want it to do
• Example: I specified what I wanted in the table, and Claude handled the calculations

Best Practices

1. Document Sharing Share all relevant files at the start Make sure files are in their final format Mention any specific data requirements
2. Feature Requests Be specific about what you want One feature at a time Provide examples when possible
3. Feedback Be clear about what works and what doesn't Specify exactly what needs to change Confirm when something is working well
4. Final Steps Review all features Test the tool Ask for any documentation needed

Example Conversation Flow

Here's how a typical development conversation might go:

1. Initial Request: "I have this ZHVI data and want to create a web tool to visualize it."
2. Feature Addition: "Can we add a dropdown to select different metropolitan areas?"
3. Refinement: "Could you make the graph use this specific color scheme?"

Conclusion

Working with Claude or any Generative AI model to develop custom tools is an iterative process. You don't need to be a technical expert - focus on clearly communicating what you want to achieve, and Claude can help with the implementation details. Take it one step at a time, be specific with your requests, and don't hesitate to ask for clarification or adjustments.

Remember: The key to success is clear communication and iterative development. Start simple, then build up more complex features as you go.

How do I get these Models?

Simply send a comment to me with your email address and I will send the two .html files to you. One use the Zillow Home Value Index data set (which is a separate download) and one use the Freddie Mac House Price Index data set (also a separate download). ?Sample images shown below. Basically, open the .html file I'll send you, chose the appropriate file you previously downloaded, type in the zip code (Zillow model) or use the drop down to select your CBSA, and be amazed at the results.

Zillow model images:

Freddie Mac model images

?

?