Real Estate AVMs

Abstract

The breakthrough in secure multi-party computations offered by Nil Message Compute (NMC) enables the creation of new data fields for Automated Valuation Models (AVM) in real estate. Nillion’s cryptographic leap called NMC is based on a new mathematical innovation which allows nodes in a permissionless, decentralized network to compute information in a secure way. This paper outlines how the breakthroughs of NMC could greatly improve pricing accuracy, efficiency, and profitability of Automated Value Models (AVMs) through the computation of active buyer offers with existing public real estate data.

Introduction

An automated valuation model is a service used by a range of professions to help determine real estate values. Mortgage lenders, consumer search portals, appraisers, investors, and many others rely on AVMs as a key input to their real estate decisions. AVM reports combine mathematical or statistical modeling with databases and are typically based on several data inputs including; the properties sales history, the tax assessor’s value, the basic features of a property, and the sales history of similar properties.

Popular consumer search portals such as Zillow use AVMs to provide an estimated price for any home in the US; this popular feature has helped Zillow receive approx 200 million monthly visitors. Increasingly, AVMs have also been used by real estate brokerages and companies that engage in iBuying. These companies, which rely on algorithms to purchase homes en masse and then sell to consumers for a profit, have attracted billions of dollars in VC investment and have seen a huge increase in residential real estate market share since their inception. With enormous stakes involved in property appraisals, mortgage underwriting, and especially iBuying, where flawed modeling can lead to disastrous outcomes; AVMs are a highly valuable service.

Data accuracy and dependability is critical for the ability of an AVM to effectively track and predict the housing market. As noted by the European AVM Alliance (EVV), “When the number of observations becomes too low, the chance of a bias sample not representing the whole market in a given time period becomes very high…As a consequence of this, various AVMs are perceived as playing catch up during periods of price movement.” However, by including active offers as a data input, AVMs have the potential to track home pricing in real time.

In the US, only the buyer, seller and their agents know the details of an active offer. Additionally, the home seller and their agent know all of the active offers including the offer that’s ultimately accepted. This knowledge can give top agents in a localized area a leg up to an AVM in their ability to accurately price the market in real time. This insight into real time market conditions is part of the service agents provide their clients and why local knowledge is often considered the foremost reason to choose one agent over another. Agents work for their clients however, and offers do not always reflect true value. By computing every offer that is submitted, a far broader scope of the active market can be achieved compared to any agent or singular brokerage could hope to capture.

For the variety of companies that use an AVM, attempting to include active offers in their model is impossible. Details around the offers as well the winning offer cannot be factored in. The AVM can only compute publicly available data such as the advertised price of the property and the eventual sold price. This is only a portion of data available to buyers submitting offers and sellers that ultimately accept one. Furthermore, there is a gap in time between when a property is listed for sale and when it closes and the sold price becomes public. This gap typically lasts 30-60 days and is known as the escrow period.

An AVM will not know the price of the property in the escrow period. The public data will look something like this:

The property was listed in December at $3.375m. It most likely received several offers and the seller accepted one of those offers in February. The home then officially changed hands at the end of March for $3.262m; 3.3% below the originally listed price. Only the buyer and seller and their respective agents knew the actual value of this home and it was a full month and half before any AVM would be able to include that data point. In today’s historically fast real estate market, that length of time is eons.

It would be an enormous improvement for AVMs to have the ability to include active offer data. This information is private of course. Buyers and sellers would not be able to negotiate very well if all their offers were public. Buyers would be wary of other parties swooping in with a better offer and sellers would be worried about low offers causing their property to be viewed as overpriced. There is no incentive for either party.

Nillion and its groundbreaking NMC technology can change this. By computing offer input data and combining it with historical databases it is possible to both factor in real time offers while keeping the property details a secret and thereby preventing any impact on the ongoing negotiation. Agents would be able to both keep tabs on the active market and share this pivotal information with their clients while at the same time sharing no information about any specific property or the parties involved.

Nillion: A Secure Processing Layer

The first step is to create a front end submission portal for offers. Importantly, the front end can only be accessed by licensed agents. In the US, every active licensed agent has a license number as well as an Multiple Listing Service (MLS) number. The MLS is an organization with a suite of services that real estate brokers use to establish contractual offers of cooperation and compensation (among brokers) and accumulate and disseminate information to enable appraisals. Agents use their MLS number to log into the MLS. This number could also be used to login to the offer portal via the user authorization operation offered by the NIL Service Layer.

This method ensures that the agent’s MLS number (corresponding to an individual agent) is not in a centralized server or even on the user device. Knowing the individual agent that is submitting an offer would make it possible to reveal information about the client the agent represents. It is therefore critical this information remains hidden.

Once logged in, the agent has two pieces of information to submit: the address of the property and the offer details. These would be two separate secrets that nodes in the NMC network would transform into particles using a One-Time Mask (OTM) and then further distributing those particles among Nillion nodes as shares using Linear Secret Sharing (LSS).

This accomplishes two things; first it provides an internet theoretic secure way of storing these secret offers in a decentralized manner. Real Estate governance boards such as the National Association of Realtors have strict guidelines regarding personal consumer information being disclosed to unaffiliated third parties. The major benefit of NMC is that the network can perform computations without any knowledge of what is being computed. Therefore, there is no disclosure to a third party since nothing is ever disclosed.

The second thing accomplished through NMC is the ability to run computations of offer details with public property data without ever disclosing the subject property. Public property data is available through multiple sources and feeds such as CoreLogic. By taking advantage of the rich historical property data already available, it is possible to compute the offer details to help develop a real time snapshot of market conditions.

Current publicly available market snapshots typically contain information regarding the most recently sold data to provide consumers with a gauge of where the market is heading. Graphics like the one below found on Rocket Homes are readily available but only provide part of the story.

By utilizing the offer details this data can be enhanced to include real time information:

This provides consumers with a better gauge on the marker than what past sold date can provide.

NIL Token Incentives

One consideration is how to incentivize an agent to submit their clients’ offer details. There would not be enough offers willfully submitted by agents if the only benefit was a more accurate AVM or improved consumer data. This can be solved by taking advantage of the NIL token, Nillion’s native utility token. Each time an offer is submitted by an agent, that agent receives NIL tokens as a reward for contributing data to the network. Furthermore, an offer that is submitted by both the buyers agents and the sellers agents could receive a greater reward as that offer could be deemed as “verified”. Only agents that have submitted to the network can access the computed data made available incentivizing more agents to contribute offers. In order to access the network they must deposit some of the tokens they have received as a reward.

As the network grows and the data becomes increasingly more valuable, 3rd party entities will need to purchase NIL in order to access the network. By rewarding valid data inputs and then selling the outputs for X+1 the amount awarded for inputs, a perpetuating cycle can be developed where the more an agent contributes the more valuable the network becomes and the greater rewards they can receive.

Conclusion and Considerations

The breakthrough in secure multi-party computation offered by NMC allows for property offers to be securely submitted and computed with existing property data to provide a data output that could revolutionize existing automated valuation models. This improvement could greatly increase the accuracy and efficiency of residential real estate markets by vastly increasing the pricing models used by all services involved with property transactions.

Some vital considerations before pursuing this use case would be quantifying how critical offer data is for an AVM. Additionally, figuring out what the optimal ways to combine the offer details with existing property data are in order to be most effective for an AVM. Significant selling of the idea to real estate agents and their brokerages would be necessary and regulatory hurdles would need to be navigated with the industries governing bodies and State governments. This onboarding process could take time and ultimately prove to be too difficult. Lastly, mechanisms would need to be put in place to discount fake offers from the data output and discourage agents from submitting offers that don’t exist. This could be regulated by real estate boards with severe punishments levied against any agent that submits a fake offer. However, this too would take significant time and institutional buy-in.

2 Likes