Explore farmland values and trends across the Midwest Heartland region. This analysis looks at real transaction data between Q1 of 2020 and Q2 of...
Farmland Valuation: Why Automated Valuation Models Miss the Mark
Take a closer look at the inefficiencies in valuing farmland and the limitations of automated valuation models (AVMs).
Farmland is one of the largest real estate asset classes in the U.S., with approximately $3.5 trillion spread across America’s rural communities. To put things into perspective, that implies the value of U.S. farmland real estate is roughly equivalent to the size of all U.S. multifamily housing stock. Yet, despite the overall size of the farmland market, it operates as a fairly inefficient marketplace with low asset turnover, poor price transparency, and inefficient capital structures.
While there are several reasons the farmland market is inefficient, this article will focus on one: the complexity of farmland valuations and why automated valuation models fall short.
Farmland Valuation and Market Inefficiencies
The traditional complexity in valuing farmland real estate is one of the main drivers of inefficiency in the market. Farms are not homogenous and they resist easy classification, meaning specialized knowledge is often required in their valuation, work often performed by an appraiser. Rural property appraisers require years of licensing, certification, and additional training offered through professional societies like the American Society of Farm Managers and Rural Appraisers.
While a professional appraisal service is often the preferred method of valuation, farmland owners may find this option to be a lengthy and cost prohibitive process. Cost and wait times have been driven up by a shortage of rural appraisers, with valuations frequently taking months to complete as appraisers work through a backlog of requests. These delays can disrupt the normal financing and transaction process, creating friction that reduces overall market efficiency.
AVMs Are Not the Solution for Fast Farmland Valuation
It is no surprise then that attention has recently turned to the role of automation, machine learning, and artificial intelligence (AI) as a means to speed up the process of valuing farmland real estate. The practice of integrating these tools within an automated valuation method (AVM) has received broad acceptance in commercial and single family real estate. These models utilize millions of property records and statistical models to estimate real estate values, a process that some claim leads to faster, cheaper, and more accurate property valuations.
Despite the growth of AVMs elsewhere, farm owners and rural property professionals have yet to fully embrace the technology as a reliable method of valuing farmland. Early farmland AVMs, some of which date back to 2014, have suffered from perceived inaccuracies in their valuations, increasing skepticism amongst potential users. Agricultural lenders, the largest market for valuation products, still rely on formal appraisal or desktop valuation tools for underwriting the majority of their loans.
Today, several companies advertise AVMs for valuing farmland, including Acrevalue and Grower’s Edge. These models are generally focused only on farmland values in eight midwestern states - Iowa, Illinois, Indiana, Minnesota, Ohio, Michigan, Wisconsin, and Nebraska. These states have been the target for early AVM efforts due to several similarities amongst them, namely higher land sale activity, generally good county record systems, a focus on corn and soybean production, and a lack of complex water rights and infrastructure.
Yet even in these ‘best case’ scenarios we find in midwestern land markets, AVMs struggle to accurately estimate land values. Estimations frequently represent an obviously erroneous figure or a range so broad as to be practically useless. Our current models simply cannot account for the multitude of idiosyncratic valuation issues that an asset class as diverse as American farmland presents. Future refinements in these AVMs are sure to improve the accuracy of results over time, but there will remain unique challenges when valuing farmland that will be difficult to overcome. We’ll examine those challenges—and their potential solutions—in the next section.
Why AVMs Struggle to Value Farmland
AVMs depend on huge quantities of accurate and recent data for their calculations. Here we encounter our first major issue, which is the relative paucity of rural property data available in the United States. Farmland, just like most other forms of real estate, may be valued through three basic valuation approaches—a comparable sales approach, an income approach, or a cost approach—but as we further explore here, these methods rely on data that is often unavailable or difficult to find.
Comparable Sales Approach
The comparable sales approach is the most common method utilized by AVMs, and relies on examining recent nearby transactions to estimate the value of similar properties. There are three main issues in aggregating large amounts of land transaction data to feed AVM models. First, annual asset turnover in farmland is estimated to be 1-2%, compared to 4-5% in some single family home markets, creating a deficit of transactions to feed the models. Total quarterly farmland transactions in some rural counties can number in the low single digits.
Second, there remain major data availability problems when working with rural county assessors. Generally speaking, rural counties have invested less in digitizing and modernizing their property records, meaning even when farmland sales exist, the results may not be accessible. Dozens of rural counties across the U.S. still require visiting physical archival records to retrieve transaction and parcel data, with no digital services available.
Finally, there is no multiple listing service (MLS) for farmland. In other real estate markets, the MLS serves as a critical professional resource detailing property listings and past transactions via user generated data. While these MLS databases are hyperlocal in nature, their aggregated data feeds many of the major data platforms we recognize today (i.e Zillow, Realtor.com, Redfin). AVMs must rely on incomplete public records without a consolidated database of professional property data for farmland.
Income Based Approach
Income based valuations rely on capitalizing net operating income from a property, and is most commonly used when evaluating real estate as an investment property. Obtaining accurate farmland rental data at a parcel or even regional level is complicated by the fact that relatively little farmland is rented in the U.S. The last major census by the USDA found that only 31% of farms are rented from non owner-operator landlords.
For those farms that are rented, varying forms of tenancy further muddy the picture. That’s because in addition to simple cash rental arrangements, many farmers utilize ‘share’ rents or ‘flex’ rents that vary greatly year to year depending on crop conditions or market pricing. Special arrangements between landlords and tenants frequently distort ‘real’ rents, such as agreements to improve a property, utilize specific farming practices, or allow the landowner to sublet parts of the property for recreational purposes. Even when true market rent can be determined, these data points are seen as confidential business information by farmers. Early efforts to publish cash rental rates for farmland have already been met with frustration and anger by farmers who resent the publicizing of their business data.
Cost approach valuations consider the value of bare ground and the value of any improvements (minus depreciation). This method is less commonly used in farmland valuations, largely because many farms represent bare ground only, without significant improvements or fixtures. But for those farms that are improved by drainage or irrigation, for example, there often exists a significant market premium that cannot be captured by AVMs. Estimating the level and value of improvement on a given farm presents perhaps the greatest challenge to utilizing AVMs, as the public record for this data is non-existent.
The nature of land improvements varies greatly by region, reflecting diverse land uses and crop requirements. In the Midwest, installing subsurface tile drainage is a common method of improving cropland, but the value of these tile installations varies greatly based on the age, condition, spacing, construction, and even the type or existence of an outlet for this drainage. Similarly, farms in the South and Delta states will frequently irrigate their crops using pivot irrigators, flood irrigation, or furrow irrigation. The type of irrigation equipment used, its age, and the manner of land grading (zero grade, leveled to a grade) will result in radically different land values.
In the irrigated West, the picture only becomes more complicated due to crop diversity and water rights issues. For Western farms, we may consider both the surface water and the groundwater districts it resides within, in addition to complex rights such as shares of private ditch companies. In places where perennial crops are planted, AVMs need to also consider the type of tree or vine, its age, its rootstock, and other minute differences such as method of trellising. Orchards in the West may be further improved by the addition of wind machines and surface water retention ponds for irrigation purposes.
The Limits of Automation in the Farmland Market
The challenge present in estimating the value of on-farm improvements reflects the broader issue AVMs face regardless of valuation approach—a glaring lack of meaningful rural property data. Public systems of record simply do not contemplate all of the nuanced data points that drive farmland values, and to date no private or professional database exists to fill the gap. Without the ability to identify, collate, and ingest this data at scale, AVMs will be unable to provide accurate farmland valuations.
The unique problems farmland presents for AVMs have only briefly been explored here, and there remains an immense amount of local knowledge and experience that cannot be easily automated or categorized. Farmland markets are incredibly diverse, and values may differ greatly within the same counties, water districts, or sometimes even on the same road. The value of a farm, while obviously tied to larger macroeconomic considerations such as interest rates, is still driven very much by local concerns. For a more accurate understanding of the value of a given property, there is still no replacement for a local farmer or valuation professional.
Looking forward, there is no doubt we will see further investments in AVMs for the purpose of valuing farmland. Broader advances in AI and machine learning will change our understanding of what is possible and will unlock rapid analysis of massive amounts of data. There is also important work that can be done to improve future AVMs, such as modernizing county assessors’ records and leveraging the expertise of land professionals to build out rural property databases (something our team at Acres is working on). Until then, the promise of AVMs will remain largely unfulfilled as they struggle with limited coverage, inaccuracies, and a skeptical consumer.