Creating and Maintaining a Strong AVM Model for High Performance
May 05, 2012
Featuring Darius Bozorgi Veros' president and chief executive officer
May, 2012 Podcast Transcript - Duration: 11 minutes
Moderator: Welcome to Veros Real Estate Solutions May podcast. Darius Darius, our president and Chief Executive Officer is here to provide some prospective on creating and running a good AVM.
Darius: Thanks, Emily. I am very excited to be here. Many people don’t understand all that is involved and what we refer to as the care and feeding of an AVM. It’s a continuous process. It involves multiple different factors and different team members to produce optimal results. Veros just went through a rather significant and major change to how we were managing that process, our AVM process in 2011. As we continually revisit that process the results have been very, very positive.
Moderator: Can you provide some insight on the importance of data sources, data management models, and do diligence to an AVM?
Darius: Sure but let’s go even beyond that. When you talk about the care and feeding of a well performing AVM model it involves multiple layers, you have to look at your team. Who’s involved from a modeling perspective, who’s involved from data and data management prospective? You have to look at your data management process which includes your data sources. One of the big ticket items that people discuss today is MLS data and how that gets incorporated and the best way to deal with that. You also have to look at your model management processes, how many different models or methods are you employing, then separately how do you intend to reconcile those different methods.
Outlier management is a huge issue as well. So regardless of what methods you’re using behind the scenes, how you reconciled those methods. What processes do you have in place to identify potential outliers and then how do you deal with those once those are identified. Then you have to look into what kind of supporting data you use both behind the scenes as well as what you present to the customer. Finally you have issues of functionality, how are you going to assess transactional business verses portfolio business, et cetera. But there are a lot of considerations that you have to pay careful attention to on a very regular basis, it’s not a onetime shot.
Moderator: Good. Will you explore the care and feeding concept a little bit more, how does it provide consistent and superior AVM performance even for Veros?
Darius: Well I’ll put it this way, so I just went through kind of a list of different categories of care and feeding but you can really boil them down into two primary categories at the 30,000 foot level. One of those being data and data management and everything associated with the management of that data. The other side is the modeling category if you will, and that gets into model, model management, model reconciliation, outlier management, et cetera. You have to be firing in both categories. You cannot just spend all of your attention on the data side of the equation and ignore the modeling management side.
You will not produce optimal results consistently. The same holds true in the reverse. You cannot ignore the data side of the equation and just spend all your time on modeling. Certainly principals of garbage in, garbage out would apply. And you will similarly lose so to speak. You really have to pay attention to both and the other thing I would underscore is you have to pay attention to both on a very regular process.
Care and feeding is a continuous process, it’s not something you do just once a quarter or a couple of times a year you revisit that. You have to continuously be following that.
Moderator: Good, now here’s a two part question for you. What are your expectations in future AVM performance and what do you think AVM analytics team should be focused on in the future?
Darius: In terms of my expectations for future AVM performance I think that for the industry as a whole I think is very positive. And mortgage stake holders should be very encouraged about where AVMs are headed in general. So Veros is very proud that – to have been what we consider the first, second generation at AVM that was back when we came on the scene almost ten years ago and we really advanced AVMs into that second generation status. Not only in terms of putting a premium on accuracy but in also providing associated analytics such as confidence scores that were actually tied to statistical definitions of accuracy. The AVM industry today is really poised to move into what will be the third generation status of AVMs and once again Veros is very proud to be leading that next jump into that next plateau. Much of that is focused around data management at least from the data side, data sources already referenced in the incorporation of the MLS data. On the modeling side a lot of that is related to outlier identification and outlier management. In the end what’s happening in the third generation status of AVMs that are able to make the leap is we’re producing a much higher percentage of accurate evaluations.
In other words, we’re delivering higher hit rates while increasing our accuracy at the same time. Prior to this time those two concepts, hit rate and accuracy, were very much inversely correlated so you couldn’t deliver a higher hit rate without also delivering a lower accuracy and vice versa. In the third generation status the successful AVMs who can make that leap are able to do both deliver higher hit rates at higher levels of accuracy. And we’ve actually proved that with our latest releases of VeroVALUE.
Moderator: Good. Thanks, Darius, we’re starting to run out of time, is there anything else you’d like to add?
Darius: The only other thing that we haven’t discussed which I think is really important. The other aspect of AVM management is a company’s, an AVM company or AVM developers own internal due diligence processes. So all of us in the space spend a lot of time in numerous external or third party AVM validations, and that’s an important part of the AVM process. We could have a whole other discussion on that because oftentimes third party validations are either not independent or objective as is expected or required from the regulators.
Or can lead to results that are not representative of what the user would get in a production environment which ultimately is the key. So a lot of attention needs to be spent on that whole other topic that I think folks will have to be thinking about, which they historically haven’t, is how to build AVM cascades. I think that’s something that is lost on most people to be frank. And then finally that internal due diligence so I see a lot of AVM companies that merely rely on those external and potentially flawed AVM validations and they do very little internal due diligence. One of the major changes we have made in Veros is develop a very robust internal and almost real time dynamic due diligence platform, which gives us some very granular insight into how our models are performing. Where we need to focus attention has been a big part of the ability, our ability to significantly improve our results. So that’s something I would encourage anybody, AVM developer or AVM user to make sure they’re aware of and they’re addressing in their AVM decisions.
Moderator: Darius, we certainly appreciate your insight into AVM models and look forward to learning more about them as they evolve. For our listeners, additional information can be found in our news room at Veros.com or follow our Twitter handle @VerosRES. Also please submit questions or feedback to email@example.com. Thank you.
Darius: Thanks, Emily.