[VIP] Don’t Tell Yourself Pretty Lies, Part 2: Opendoor Sales Analysis

In Part 1, I looked at the economics of iBuyers generally, and concluded that we need to analyze them as market makers, using market maker economic models, rather than as investors or brokerage companies. In this part 2, I wanted to get a bit more detailed and granular, using some data I was fortunate enough to have a reader share with me on the condition of anonymity. Thank you reader — you know who you are!

What we have here is a granular look at some of Opendoor’s activities in Florida. I can’t list the addresses, but they’re not important for our purposes anyhow. I will also point out that I had very little desire to go find all of the data for the hundreds of data points here so limited myself to ~100 properties for more research. What I wanted to look at was what Opendoor paid for the home, then subsequently sold it for, and if that property was originally listed for sale with a real estate agent or not. Some of the information is not available publicly, so I limited myself to public sources, like Zillow, Redfin and Realtor.com’s sold data.

With those limits/caveats out of the way, let’s take a look at some data.

Opendoor’s Purchase and Sales

We start first with a table of the top 15 properties by price bought and sold by Opendoor in the last couple of years. I was curious to see how Opendoor does with more expensive properties, since the common narrative is that iBuying might work for cookie-cutter, low-cost homes, but not for higher end properties. Opendoor purchased these homes for prices ranging from $298,900 to $469,700 and sold them for between $311,000 and $479,000.

Now, that’s not “luxury” or “high end” in many markets, but from what I’m seeing and from what I know of that area of Florida, those are on the upper end.

As you can see (you might have to scroll a bit), the average purchase price of these 15 homes was $362,440. Opendoor then turned around and listed them for an average of $376,867 for an average List-to-Purchase of 104.1%. It then sold these homes for an average of $365,388, or 97.0% of Sold-to-List ratio, and ended up with an average Sold-to-Purchase (what it bought the house for vs. what it sold the house for) of 100.9%.

What that strongly suggests is that Opendoor paid market price for these homes. After all it’s said and done, Opendoor managed a spread of just 0.9% over what it paid for the house. Without the 7% Service Fee (assumed as an average, based on recent data from Mike Delprete), Opendoor likely loses money, taking costs of maintenance, taxes, and transaction costs into account.

What about more recent sales? I was simply curious if Opendoor’s operations had gotten better. Here’s that table.

The averages for the most recent 15 transactions are:

• Purchase price: $218,580
• List Price: $234,800
• List to Purchase: 107.7% (vs. 104.1% above)
• Close Price: $226,980
• Sold-to-List: 97.1% (vs. 97.0% above)
• Sold-to-Purchase: 104.6% (vs. 100.9% above)

There is improvement across the board here. Is it because of recency, implying that Opendoor is getting better at pricing properties in this market? Or is it because of price, where Opendoor is simply better at pricing homes in the 200K range vs. 300K or 400K ranges?

One might-be-relevant factoid from these 30 transactions is that the 8 transactions with a negative spread (meaning, Opendoor paid more for the house than it sold for) have purchase dates from August of 2018 to November of 2018. I might do more sold data analysis later and see if there’s a real trend here, but that might be something to consider, that Opendoor has a positive spread on homes purchased after November of 2018. (And in a larger dataset I took these 30 properties from, I’m seeing a bloodbath in September and October of 2018, which was a strange market. Maybe Opendoor adjusted its algorithms accordingly, or perhaps the market just changed.)

The Seller’s Economics

The final interesting thing is to look at whether the seller got screwed, and if so, by how much. Since Opendoor says closing costs like title, escrow, etc. are like traditional sales, I won’t consider that. We also don’t know what repair costs were, if any, and there are many reports from the field that sellers are often hit with very large repair concession requests after the initial offer, leading to a feeling of bait-and-switch. Nonetheless, this is the data we have on the 15 most recent sales:

As you can see, the average seller would have netted $10,082 more by selling with a real estate agent or roughly 4.6% of the price at which they sold to Opendoor. In three cases, the seller actually made more money selling to Opendoor than by the normal process, even taking the service fee into account — one has to assume here that Opendoor simply made a wrong bet on those properties and overpaid, to the seller’s benefit.

It has to be noted that that conclusion is likely incorrect, for two main reasons: renovation/maintenance costs and holding costs. We have no data to go on as far as renovation/maintenance goes, on either side: what did Opendoor ask for in terms of concessions from the seller at purchase, and what would have the seller spent on renovations/maintenance under a normal list-and-sell process? We also don’t know what kinds of holding costs the sellers would have incurred (mortgage interest payments, taxes, fees, etc.) for the holding period, which average 4.2 months between the time they sold to Opendoor and the time that Opendoor sold that house.

What we can ask, however, is whether $10K is worth saving 4.2 months of FUD (Fear, Uncertainty, Doubt) for the seller as well as the inconvenience of trying to sell one’s home. The answer depends on the seller, of course, but I can say without a doubt that I would pay $10K every day and twice on Sunday to save 4.2 months of the pain and hassle of trying to sell my home.

Furthermore, if you look at the price history of the homes that were once listed with a real estate agent, we see a consistent pattern of dropping the price in order to generate interest. For example, the property originally listed on 6/7/18 for $279,900 saw three price drops, prior to the sale to Opendoor: $900 on 6/19, $5,000 on 6/22, and $4,100 on 7/10. The seller then took the property off the market on 8/18 and sold to Opendoor on 10/3 for $253,500. If the seller had stayed on the market, it seems likely that he would have had to drop the price again and again until it got to the market price around $253,000 (considering that Opendoor itself sold that house for $252,500 after six months on the market).

I know I’m making a lot of assumptions, most of them counterfactual, but it’s something to think about.

Opendoor’s Pricing Engine is Very Accurate

Note that the Sold-to-List ratio for Opendoor averages 97% from the list of data points above. I don’t know how that compares to real estate agents in the relevant market during the relevant period, but I did ask Sunny the recovering-broker about that. Her view is that unless that was a red-hot market and the average was significantly higher, 97% Sold-to-List ratio is really quite excellent.

Here’s why that’s interesting: who priced the homes for Opendoor? Looking at most of the listings here, the listing agent of record for Opendoor is Brandon Lavallee, who works in Acquisitions for Opendoor. The record I have of him most prominently is as a real estate agent in Las Vegas. It likely means that Brandon works out of Las Vegas, but is the listing agent of record for the MLS in Florida.

Which means that Brandon did not tour the home and price it for sale. We knew that already, right? So who did? The answer is likely a proprietary algorithm at Opendoor, together with data points that may have been gathered from the seller, local Opendoor contractors or agents or others. But a computer program priced these homes, and in all likelihood, a computer program suggested the price drops along the way, until an offer came in.

A 3% error rate by a piece of software in pricing homes is not to be overlooked. In my Red Dot on iBuyers, discussing how accurate or inaccurate AVMs are, I wrote:

So at least the Zestimate is less accurate than a human being estimating via both the initial list price and the final list price closer to the transaction.

…

But the median absolute percent errors for the final list price by a local expert is 4.5%. The same for Zestimate is 6.3%. That means that 50% of the final list price was within 4.5% of the ultimate sale price, while 50% of the Zestimate was within 6.3% of the ultimate sale price. We’re talking about a 1.8% difference between the estimate and the ultimate sale price, in either direction high or low.

I continued, after discussing how Redfin takes the listing price into account in its AVM:

Not only were 63.66% of Redfin’s estimates within 3% of the ultimate sale price, 79.69% of their estimates were within 5% of the ultimate sale price.

Note that this was not comparing Redfin’s listing agents and their initial and final list price vs. an AVM. This was comparing Redfin’s AVM to Zillow and Homes.com’s AVMs. And the real takeaway for us is this:

• Redfin’s median error rate of 2.06% is better than the 4.5% median error rate of listing agents in the TEGoVA study above.
• The 79.69% of Redfin’s estimates that were within 5% of the ultimate sale price is better than the 56% of the final list price in the TEGoVA study above.

Therefore, we might conclude (or at least suggest that there is room to believe) that Redfin’s listing-price informed AVM estimate is superior to a local expert’s final listing price. [Emphasis in original]

And now we have some data from Opendoor’s actual buying and selling operations to back that up. The research from TEGoVA (The European Group of Valuers’ Association) on AVMs cited in my Red Dot shows that a human being, a local real estate agent, has median error rates of 5.6% in the Initial List Price. Opendoor’s algorithm has a 3% error rate in its Initial List Price.

This is, to put it mildly, an issue for real estate agents.

Other Interesting Tidbits

There are a few other interesting tidbits from this dataset.

First, looking at the 15 most recent Opendoor transactions, the estimated profit per transaction is $1,580 without taking into account their holding costs, maintenance costs, renovation costs, transaction costs, etc. but also not taking into account whatever seller concessions Opendoor got at purchase. So even without those important factors, that profit margin is thin, thin, thin: 0.72% of the cash outlay on purchase. I assume that margin gets slimmer once other costs are taken into account.

Second, in just about every transaction I have, there is a buyer’s agent involved when Opendoor ultimately sells the house. Not a single buyer of an Opendoor home was unrepresented. Furthermore, while we don’t know what Opendoor offered the buyer’s agent in cooperating compensation, given Opendoor’s stance that they fully cooperate and compensate, one assumes that Opendoor paid out between 2.5% and 3% of its closing price on every deal.

On the data I have, Opendoor paid out $1.1 million in buyer agent compensation over 164 transactions. You decide how big a threat to the industry that is.

Third, at least from the data I’ve looked at, the vast majority of sellers appear to go directly to Opendoor. A few were listed, either FSBO or with an agent, and then after months of not selling the home, at some point the homeowner decides to sell to Opendoor. In most cases, however, there is simply a record of a sold transaction on such and such a date, for such and such a price. We don’t know whether such “direct” sellers were in fact represented by an agent or not, but it’ll be interesting to see over time as word about Opendoor and other iBuyers spreads.

Takeaways

Lest this get too long (I know, I can’t believe I typed those word just now either), let me summarize the main takeaways for me from this exercise in data analysis.

1. Opendoor pays market price for its acquisitions. This reinforces the point I made in Part 1 (and really, throughout my commentary on iBuyers) that the iBuyer/Market Maker model is not to be confused with investor/home flipper models.
2. Opendoor is making a profit, albeit a tiny tiny sliver, on each transaction… as far as we know, because we don’t have Opendoor’s unit economics numbers as we do for Zillow.
3. The seller, on these numbers, loses money selling to Opendoor. But the amount is $10K on average, to save 4.2 months of Fear, Uncertainty, Doubt and Inconvenience that is the modern home selling experience. Whether that’s a price people are willing to pay for convenience and certainty or not is the $640 billion Question.
4. Opendoor’s algorithm is scarily accurate. 97% Sale-to–List is superior to the 5.6% error rate of the average local real estate agent.

In the next part of this series, we’ll discuss Differentiation to some extent, leading towards what traditional brokerages and agents need to think about as the iBuyer model continues to grow.

-rsh