Shamus: Hello and welcome to The iBuyer.com podcast. I’m your host as always, Shamus Samerdyke. And today, we’ve got a very special guest, one of my favorite people. We actually work together. We have Mike Casale. He’s one of the head data scientists in our AVM, expert and really the pioneer on helping develop our own asset valuation model and knows more than just about anybody that I know or I’ve heard of about what’s really going on with valuation in the iBuyer marketplace. So how are these different iBuyers, large and small valuing homes in order to determine these cash offers? So Mike, if you want to maybe give everybody just a quick 30 seconds of your background, some of the different things you’ve done and we’ll really, we’ll dive right in.
Mike: Sure! Thank you. Thank you for having me. And so, I guess, I got into valuations via just, just data analytics mostly through eCommerce and some different projects that I’ve been involved with over the years. And getting into machine learning and some of the recent developments in terms of the technology that most companies are now using for these kinds of predictions has progressed quite a bit over the last few years. And so, just using it all across the board for everything, including home valuations now and by machine learning we mean algorithms that you don’t specifically tell what to do, but over tens of thousands of different examples, they learn and get a little bit better and a little bit better until eventually, you know, they know your thoughts before.
Shamus: Before you think them, right? Perfect! So, to that point, you know, with everybody getting more and more involved in AI and machine learning, pretty much all of these companies at this point, the Zillow’s, Opendoors, Offerpad, iBuyer.com are all using these AVMs, Asset Valuation Model. So could you break down a little bit how, and of course, remember to the layman’s, can you break it down as simply as possible to how these AVMs work and how the machine learning helps predict home values better and faster?
Mike: Sure! So I think for, you know, you’d have to ask an appraiser or somebody who was there before the automatic valuation models came into pretty much standard practice. So what they used to do, I don’t know, but since I’ve been involved with it, all the property records are digital and so we have data providers and data aggregators, so they’re taking source data from County records, from local sources throughout the country, nationwide. And they’re bringing these disparate data sources together. So we’ve got, you know, assessors, recorders, tax collectors, real estate people, etc. And so aggregating all this data together and in a digital format where we can keep track of everything from backpacks history to you know, whether this lot has been rezoned and when things were built, updated. And really how it works in terms of machine learning is we take all this data and we string it out and we say on a parcel or property basis, there is, you know, 30 different data inputs.
And so you have bedrooms, bathrooms, square feet, etc. And the more data inputs that you have, the more refined the models can become. And so we have like, I’ll speak not just about iBuyer here but in general, the typical property record data that we can get would have somewhere around 400 data inputs.
Mike: Just on a per project basis. So this goes back anywhere, 7, 10 years, all the different sales history that the tax records and so forth, and then, flatten all this data in and in over tens of thousands of transactions. Then we tell an algorithm like, okay, well you’re going to go through these 50,000 transactions 5,000 times and you know, whatever the case may be. And throughout that, it’s just fine tuning, getting a little bit better each time, learning to recognize what’s an outlier, what can we basically push to the side, what’s important.
Shamus: So just to make it, let me simplify this just a little bit. So basically what it does is it takes these tens of thousands of transactions, okay? It runs through all of them and attempts based on your algorithm to predict what the outcome of that transaction would have been exact. Right. And it goes through all of these transactions and sees, okay, the machine predicted X but the actual transfer amount was $10,000 higher. So it’s going to do that same transaction again with slightly different inputs until it gets as close to that answer as possible. And when it gets as close to that answer as possible, it determines based on the 5,000 times I’ve done this same algorithm and modeling these 20 characteristics where the most important and got me to that answer the most times out of all of them.
Mike: Yeah. And, so you basically give it a strategy to say like, okay, well how you want to determine accuracy or error or how do you want to measure the objective that you’re trying to predict, right? And so, is it doing better or worse and depending on where you want it to do, these algorithms can be used and they are used in real estate for everything, including image recognition. Now Zillow is doing a bunch of stuff with, we’ve done some experimenting with, and you know, to classifying different things. And so maybe you want to say, what is the probability that the price is going to fall within some certain range.
Shamus: Say between 250 and 280,000.
Mike: Yeah. Exactly. Any range, any ranges you want and just say, okay, well now we have however many different iBuyer records in our case specifically, and then of these 50,000 records, we can come up with pretty good approximations for like what is the probability that it’s going to sell between, you know, this range and that range or it can also be, you know, more of a regression prediction which would be like predict an exact amount instead of a range.
Shamus: Okay. And so as a regression prediction, does that mean that it can continually changes until it gets exactly that number?
Mike: Yes, it’s a continuous number.
Shamus: Got you. Okay. Interesting.
Mike: No ranges.
Shamus: So my brain hurts just a little bit, which is why we have Mike with us. So, now we know how the AVMs work.
Shamus: We have an understanding of how the AVMS work. Let’s talk about how they’re working in the marketplace? Right? So let’s take like Zillow for example. So Zillow’s done, you know, comparatively small number of transactions when we’re looking instead of, compared to Opendoor and Offerpad and whatnot. However, there’s been a lot of research lately on, especially from, we’ll reference the Mike DelPrete article specifically in the show notes. But Mike DelPrete who is right now one of the leading sources of information on iBuyers in the marketplace, which will be us soon. He just did a research or a report recently showing that, it appears that Opendoor and Zillow are making pretty darn close to what a seller would walk away with offers on his home. Is that right?
Mike: In very general terms, yes. So, it’s not, there are some differences.
Shamus: Can you go over some of those differences?
Mike: Yeah. I don’t want to get into too much detail or muddy water, but essentially you’re going to take an area like Phoenix where you have a very heavily, all the big iBuyers are there and a large portion of the transactions that are happening are actually iBuyer transaction.
Shamus: Right. Maybe something like 10% now.
Shamus: Something like 10%, I mean like a crazy high number.
Mike: Right! It would seem like that would make sense at least.
Mike: And so in an area like that where you have a very large influence from iBuyer specifically, now you’ve got to just step back and take a look at the situation as a mom and a pop buyer going to be as, or is it going to be the same sort of a transaction as a whole team of data scientists from Zillow or Opendoor or somebody else in engineering transactions for some mom and pop or husband and wife who really aren’t looking at the data, they’re just, I want to move or whatever.
And so, I think that logically or objectively, you’ve got to admit, well, a team of data scientists who are really working for an investment type of operation, that price point is just going to be a little bit different than your mom and pop. But in an area like Phoenix where you have a heavy mix of iBuyers, we do see that the market value is pretty much coming into line a lot, a lot closer than it would be in an area where there really are very few.
Shamus: So it’s almost as if the more interaction and presence had by iBuyers in a marketplace, the more the market itself starts to shift towards the iBuyer’s pricing?
Mike: I would say they don’t control the markets yet, you know?
Shamus: So do you think that they have an effect on it? Is that what you’re trying to say? Or are you saying that their pricing has gotten so good because there’s so many and so competitive that it’s at market pricing?
Mike: I think that there’s a lot of opportunities, where there are a lot of iBuyers, there are a lot of opportunities to take a cash offer, which is faster. There’s a lot of benefits involved that you are not entirely monetary.
Mike: And so the more opportunity for people to make these decisions, I think the more it happens, and on the whole…
Shamus: And the more the competitive you think the offers get as well?
Mike: I think the offers get competitive. I think that, you know, like there’s a case to be set right now in the data that Zillow maybe the most competitive in terms of giving you the highest amount potentially. That’s not true everywhere, every time. On the whole, they’re pretty good. And then within a reasonable range you’re going to see, you know, plus or minus 5% of like what people might call a standard market value as opposed to like a cash or for market value. And perhaps there’s a little bit of a debate over whether there is a difference in those or how much that difference is. My thoughts are from looking at the data is that there is a little bit of a difference.
Shamus: Do you think it’s as big as 5%? I mean, a $250,000 transaction, that’s twelve and a half thousand dollars.
Mike: Right. So there’s a lot of different things that you’ve got to look into in terms of the 6% commission fee, things like that.
Shamus: 2% for closing costs, all those additional things.
Mike: All those things are not really getting figured into the transfer knowledge.
Mike: But like on the whole, if you look at the zestimate amount versus this Zestimate lower bound range, so it’s not the top of the page on Zillow, but if you go on Zillow, you can look and see the lower bounds zestimate. Generally you’re going to see somewhere between that lower bound and their middle value is, if you took like a halfway point, there would be a pretty good estimate on average. And so you’re getting somewhere between that top amount that Zillow is going to show you and the lower bound that they give you, more often than not, it’s going to be pretty close to in the middle of there.
Shamus: And you’re saying that’s what they’re offer and it comes out to be?
Mike: Yeah, they certainly different. This is not something, you know, like Zillow would be like, no, we don’t do that.
Shamus: Right! It’s not across the board.
Mike: It’s not a rule.
Shamus: But based on what you found or what you’ve looked at is you see it tend to fall in that lower core tile almost you could say.
Shamus: Right. So here’s a question and we’re probably going to get in trouble for asking this.
Shamus: But I know you’ve looked at a lot of the transactions throughout the United States and some of the history of Zillow specifically and there’s zestimate, has there or has not the Zestimate pretty much across the board reduced over the last year or so? Based on what you’ve seen? Now, this isn’t, okay. Listen, so we’re not saying anything negative or anything about Zillow, based on the data that Mike has looked through and seen, there has been a pretty much across the board reduction and the Zestimate, is that correct? Or am I way overshooting my bounds there?
Mike: So I would throw a cautionary flag that it’s not entirely clear to me what exactly has happened. It does seem like if we compare, there are automated valuation models, not only the Zestimate, there are many others, right?
Mike: There’s house Canary, there’s data providers, there is our own internal data.
Shamus: Everybody sort of have their own spin on it.
Mike: Right! And in fact, Mike CP, he works at the first American AVN. He is the one that he’s used in his analysis here. And it’s interesting to me, that comparing some of his results, I went back and I did a little bit of an analysis. Well does our data match up to what Mike is saying here? And, we’re able to see like you’re saying that it would seem like over time that the Zestimate, specifically in an area like Phoenix is perhaps changing or perhaps because they’re more involved with the iBuyer scene in that area, then…
Shamus: When you say change, is it changing or is it trending down? Trending up?
Mike: I’m going to point you to this right here. So this is a purchase price to AVM 2019.
Shamus: And we’re referencing, this is the Mike DelPrete article as well, which we’ll give you guys in the show notes.
Mike: Right. In Phoenix, he’s saying here, you’re looking at, the transfer amount is on average, 98% of what their AVM says.
Shamus: So basically what was paid for the house on paper is within 2% of what their AVM asset valuation model says it’s worth. Okay. So, that’s in Phoenix.
Mike: That’s in Phoenix, right. Now, you can see in an area like Florida where there’s not as many iBuyers and all these different things, they’re saying, well on average the actual transfer amount is 99.9% of their AVM predicted value.
Shamus: Got you.
Mike: And so right there, that sort of highlights what we’re talking about in an area with tons of iBuyer action like Phoenix versus a whole state like Florida, which is not predominant, you know, there’s not a huge influence from the iBuyer sector. And so, but in any case, I went back and I looked, okay, well in Phoenix, what is the Zestimate compared to the actual transfer amount in cases where the purchase was a known iBuyer. So if Opendoor, Offerpad or Zillow themselves actually buys something in the Phoenix area, what is the Zestimate amount in comparison to this versus Mike? Mike is using a different AVM not the Zestimate.
Mike: But doing the same thing with the Zestimate. And I think what we find in that case is that the Zestimate is actually much closer in an area like Phoenix as opposed to this AVM that Mike here is looking at. It’s actually further away in Phoenix and it’s closer in Florida. Whereas when you look at the Zestimate, you almost see an exact reversal of that, which is interesting if nothing else, I couldn’t tell you exactly the reasons.
Shamus: So do you think that it’s more like the Zestimate is more geared towards the potential iBuyer marketplace in an area where iBuyers are more heavily present? Potentially.
Mike: Yeah. Potentially.
Shamus: Yeah, that’s what I’m saying. Interesting.
Mike: Right. I think on the whole, you know, one of the takeaways is that, it certainly doesn’t look like at this point in time anyway, home sellers are getting ripped off by iBuyers. I would say that it’s actually a pretty…
Shamus: Absolutely not. Yeah.
Mike: It’s a pretty good deal from any of the big iBuyers. They are going to see and maybe that changes over time as they have more price controls like OPEC, you know, OPEC owns the supply of all oil… If Zillow ends up owning the single family home supply in the US I think maybe that doesn’t work out so well for consumers at that point. But right now I think it’s working out pretty well.
Shamus: Right. And I would say, you know, even based on what we’ve seen is that the amount of money that a seller will walk away with in a traditional transaction, so once you take out the realtor fees, the closing costs, any potential repairs, maintenance, etc. That final number that they walk away with based on the information nationwide of these iBuyer transactions, we’ve seen that it’s within 0.2, that’s not 2% that is 0.2% of what the final typical, average iBuyer offer comes out to be for that same property. So that means that the iBuyers are less than 1% on most circumstances away from what you would actually walk away with once you back out those realtor fees and all of that. And you wouldn’t necessarily have to use a realtor with the iBuyer. Right?
Shamus: So looking at it that way, based on, you know, I was trying to get Mike to tell us that the AVM in certain areas weren’t working as well, but he’s so good with the data, he was able to refute all of that. So, long story short, what I wanted, the reason I wanted to bring Mike on today was to really dive into the data a little bit more to discuss how the AVMs work and to point out based on the data that realistically people are afraid that iBuyers are shafting them and they’re not getting the best deal possible. But based on the data.
Mike: I would say I have much more faith in iBuyers than I do in like used car buyers. And also just another thing to throw out is that if you look at Zillow’s, you know, most recent third quarter earnings report, they say themselves that they’re on average losing $4,000 per transaction. So that’s probably because they want to build a level of trust with consumers and they’re willing to take a loss to give you some more services down the pipeline somewhere and maybe make that up. But certainly it speaks to, they’re actually suffering a little bit of a loss on per transaction basis in order to give the customer a better experience at this time, it seems.
Shamus: Right. So they’re very focused on helping the customer, the seller as much as possible, even to the detriment of their own earning statement.
Mike: Presumably, they’re looking to monetize elsewhere in their pipeline.
Shamus: Elsewhere in the pipeline or down the line, right? So they built up their brand, their trust and whatnot, and they know once they hit a certain amount of transactions, it’ll all work itself out. But either way, it means that they’re still staying committed to a certain level of customer experience. And we’ve seen that across the board with iBuyers as a whole. So I think that’s a great one to end it on. First, Mike, thank you very much for coming on today. I know how busy you are, so I appreciate you coming on and helping some of us layman’s people understand how all this works. It’s a little bit better. So if people are interested in finding out more about what it is that you do with data science and some of the things that’s going on with real estate valuations, is there any resources that they can look into?
Mike: We can put some out there maybe.
Shamus: So here’s what we’ll do, we’ll type some on the show notes, some of the places that you can go and some information on, you know, how machine learning is being used to utilize or in the world of asset valuation in terms of real estate, all that link in the show notes. As always, please don’t forget to subscribe. We really appreciate you guys coming and listening to us every week and we look forward to seeing you next time. That’s all we’ve got. Take care.
Mike: Thank you.