Information Rigidities and Farmland Value Expectations

1. Introduction

Farmland markets are typically associated with high transaction costs, private valuations, and low and seasonal turnover (Melichar 1984; Featherstone and Baker 1987; Falk 1991, 1992; Just and Miranowski 1993; Shiha and Chavas 1995; Falk and Lee 1998; Chavas and Thomas 1999; Lence and Miller 1999; De Fontnouvelle and Lence 2002; Sherrick and Barry 2003; Deaton and Lawley 2022). These potential inefficiencies are concerning for two reasons: farmland accounts for a large share of the value of the sector’s assets (USDA 2024), and the U.S. farmland market experienced pronounced periods of boom and bust in the 1910s and 1980s. As a result, farmland owners are often concerned about the potential for a large-scale decline in farmland market prices (Zhang and Tidgren 2018). Sherrick and Barry (2003, 46) suggest that “Government policy could play a role in enhancing market efficiency in farmland markets through information channels.” The authors argue that market inefficiencies could be overcome by “more extensive and higher-quality publicly available information.” The goal of this article is to examine the degree to which information inefficiency influences farmland market dynamics. We follow the empirical approach of Coibion and Gorodnichenko (2015), who examine how economic agents incorporate new information in updating their expectations of inflation. The authors show that agents may not fully incorporate new information into their inflation expectations because they are unable to distinguish the signal from the noise or they update their information sets infrequently.

Coibion and Gorodnichenko (2015) examine a series of repeated expectations of inflation for various economic agents. The authors argue that the agents’ expectations in aggregate should become more accurate over time due to increased information. Their empirical test examines the correlation between expectation accuracy and the degree to which agents adjust their expectations over time. When this correlation is statistically significant and positive, agents fail to fully incorporate new information efficiently. Coibion and Gorodnichenko (2015) label this positive correlation as “information rigidity” because agents fail to fully adjust their expectations to new information or they update their information sets infrequently. The key to this approach is observing repeated expectations of the same terminal event.

We exploit a series of repeated price expectations of Iowa farmland market professionals since 1964. We apply the testing framework from Coibion and Gorodnichenko (2015) to examine the degree to which information inefficiency influences farmland price expectations, which according to theory are a key determinant of realized market prices (Just and Miranowski 1993). This research represents an important step toward a better understanding of the role of information in farmland market efficiency.

Our results detect information rigidities in the price expectations of Iowa farmland professionals, in the sense that they either update their information sets infrequently or underweight new information. Further, we show that the degree of information rigidities varies by location and over time. The accuracy of the expectations of Iowa farmland professionals is not fully explained by information rigidity. These findings should be interpreted with some limitations in mind. First, farmland market professionals who attend Iowa State University’s Soil Management and Land Valuation conference (SMLV) are our source of farmland expectations. SMLV participants have on average more than 20 years of experience in Iowa farmland markets (Zhang, Lence, and Kuethe 2021), but we do not know whether they purchased or sold farmland in any particular year. Second, we measure observed farmland market prices using the subjective values collected by the Iowa Land Value Survey (ILVS), rather than aggregate market transaction prices. While these survey values may differ from market transaction prices (see Scott and Chicoine 1983; Barnard and Wunderlich 1984; Zakrzewicz, Brorsen, and Briggeman 2012; Bigelow, Ifft, and Kuethe 2020), Stinn and Duffy (2012) find no statistical difference among average Iowa farmland transaction prices and the aggregate valuations of ILVS. In addition, ILVS county land values are the only annual estimates available each year, and they are used widely by county assessors in reviewing property tax assessments. Nonetheless, our results should be interpreted as suggestive evidence for the role of information in farmland market expectations.

This study may have direct implications for understanding how information, through expectations, affects farmland prices. Consistent with prior studies, we find that farmland market professionals may not fully incorporate new information when forming price expectations (Tegene and Kuchler 1991; Kuethe and Ifft 2013; Kuethe and Hubbs 2017; Kuethe, Brewer, and Fiechter 2022). In the information rigidity framework, we find that farmland market professionals may be unable to differentiate the signal from the noise or may update their information set infrequently. The information rigidity framework suggests a muted response to information shocks. As a result, information shocks are unlikely to lead to widespread mispricing. In the event that farmland market prices exceed market fundamentals, new information is unlikely to correct market prices quickly. In the spirit of Sherrick and Barry (2003), our results point toward the continued need for market information. The information rigidities framework suggests the need for higher-quality information so that market participants can better distinguish signal from noise and more extensive and timely information to encourage frequent updating.

2. Data

Iowa’s farmland market has been studied extensively by agricultural economists (Miranowski and Hammes 1984; Falk 1991; Just and Miranowski 1993; Falk and Lee 1998; Lence and Miller 1999; Zhang, Lence, and Kuethe 2021). The Iowa farmland market is the largest of all Midwestern states, valued at $216 billion in 2020 (USDA 2021a). Further, Iowa State University maintains the longest-running expert opinion survey of farmland market conditions, dating back to 1941.

Since November 1941, Iowa State University’s ILVS collects subjective valuations of contemporaneous, county-level farmland values from land market professionals, including agricultural lenders, realtors, rural appraisers, and farm managers.¹ Beginning in May 1964, Iowa State University collects expert land value expectations from the farmland market professionals who attend the SMLV. The SMLV is an annual extension and outreach event for land market professionals begun in 1927. Iowa State extension educators conduct a voluntary survey of participants. The respondents provide their subjective expectation of Iowa farmland market prices six months later, in November of the current year, which corresponds to the collection period of ILVS. In addition, the respondents provide their subjective expectation of Iowa farmland market prices 18 months later for the subsequent November.²

The repeated expectations follow the same structure as those studied by Coibion and Gorodnichenko (2015). We can examine the relationship between the accuracy of the six-month farmland price expectations and how the expectations evolved between the 18- and six-month horizons to test for information rigidity in the price expectations of farmland market professionals. Coibion and Gorodnichenko (2015) stress that their regression-based test must be applied to aggregate expectations of individuals. Thus, our analysis considers the expectations collected at the SMLV because only the aggregate expectations are available dating back to 1964. There are some important differences between the expectations of SMLV and the market outcomes of ILVS that we must consider. As documented in the survey instruments in the Appendix, the SMLV survey collects respondents’ subjective expectations for farmland prices that are representative of their county. These expectations are aggregated by the mean of all respondents across the state of Iowa and in four regional quadrants. In contrast, the ILVS collects respondents’ subjective current valuation for farmland prices that are representative of their county across three quality bands: high, medium, and low. Respondents provide the percentage of their county that represents all three quality grades. These percentages are used as weights to create a weighted average of farmland prices representative of that county. We measure aggregate realized farmland prices as the weighted average of the subjective prices because this value is likely to correspond to the mean across individual responses. In addition, these weighted averages represent the county-level average land values published by Iowa State University annually since 1950.

Although the sample of respondents may differ between the ILVS and SMLV surveys, we are relatively confident that the expectations correspond to the assumed outcome measure. The SMLV is an extension event designed to provide the continuing education of land market professionals. As shown in Appendix Figure A3, the participants report six- and 18-month expectations for the land values in their primary county. Participants receive the five-minute survey in their registration package. Since 2016, the voluntary expectation survey also asks for the years of professional experience in the Iowa farmland market. This question was introduced to mirror a similar question in the ILVS. As shown in Table 1, the mean experience of SMLV participants closely corresponds to that of ILVS respondents. Panel B compares the number of respondents by occupation for both SMLV and ILVS. As expected, SMLV has a larger share of respondents who are frequently involved in land market transactions, including farm managers, rural appraisers, brokers, and lenders. The ILVS includes additional professions, including farmers and landowners, as well as government professionals.

The solid line in Figure 1 plots the representative farmland price for Iowa from the ILVS between 1964 and 2021, using the previously described weighted average. The dotted line represents the aggregate expectations of the same year collected from the SMLV participants in the previous year at an 18-month expectation horizon. The dashed line represents the aggregate expectations collected from the SMLV participants in the same year, at a six-month expectation horizon. For example, the aggregate 18-month expectation peaked for 2014 at $10,042 per acre, collected in May 2013. In May 2014, the aggregate expectation of November 2014 land prices fell to $8,568 per acre. Finally, the November ILVS reported an aggregate value of $7,943 per acre for farmland across Iowa. The empirical test of Coibion and Gorodnichenko (2015) examines the correlation between the expectation error at the shortest horizon and the revision of the expectations across horizons. In our case, the expectation error would be represented by the percentage difference between the dashed and solid lines in Figure 1. The revision of expectations would be represented by the percentage difference between the dotted and dashed lines. The intuition of this empirical test is that expectations should be more accurate at shorter horizons than at longer horizons, as a result of information gains. The resulting correlation is expected to be positive.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Observed and Expected Value of Iowa Farmland

Source: Iowa State University Soil Management and Land Valuation Conference and Iowa Land Value Survey.

3. Baseline Estimation of Information Rigidity

Coibion and Gorodnichenko’s (2015) regression-based test of the correlation between expectation revision and error is as follows:

1

This test examines the role of information across the expectations formation process, beginning at the longest horizon. The first iteration of expectations is represented by F_t−18, which is expected percentage change from the most recently observed farmland price (, where f_t−18 is the expected price in levels and a_t−24 is the most recently observed farmland price in levels). The second iteration of expectations is similarly calculated at the sixth-month horizon (), and the subsequently observed or actual farmland price is represented by . Thus, the left side of equation [1] represents the percentage expectation error at the shortest horizon, and the right side represents the revision in expectations between the 18- and 6-month horizons. The correlation between the expectation revision and error is captured by β.

Our baseline estimation of equation [1] using ordinary least squares are reported in Table 2. Coibion and Gorodnichenko (2015) show that when β>0, agents do not incorporate new information efficiently when revising their expectations. As shown in Table 2, is statistically significant and positive.

Coibion and Gorodnichenko (2015) offer two ways to interpret . One, given a particular time period, agents update their information set every λ, where ()). Our estimation results suggest that Iowa farmland professionals update their information sets approximately every four months (). Two, new information is given a weight of G% when agents are updating their information set, where (). Our estimation results suggest that Iowa farmland professionals place nearly two-thirds weight on new information (). Coibion and Gorodnichenko (2015) stress that these interpretations of β are only valid when the expectations fail conventional tests of information efficiency. To provide a concise presentation of our results, Appendix Table A1 shows efficiency tests from Nordhaus (1987). These tests show that Iowa farmland professionals’ expectations meet the necessary condition of inefficiency.

It is important to note that there are alternate interpretations of β>0. As Nordhaus (1987) highlighted, β>0 can be interpreted at “forecast smoothing.” Coibion and Gorodnichenko’s (2015) interpretation of smoothing is the intentional dampening of revisions for the sake of reputation and the use of aggregated expectations should alleviate intentional smoothing by individual respondents. This interpretation was applied for USDA forecasters by Goyal and Adjemian (2023) in response to prior evidence of smoothing by USDA forecasters (Isengildina, Irwin, and Good 2006, 2013; Isengildina-Massa, Karali, and Irwin 2017). Goyal and Adjemian (2023) suggest that β>0 is more likely related to information rigidity than to forecast smoothing. Alternatively, β>0 could be interpreted as evidence of asymmetric loss, where respondents place a different penalty on overprediction relative to underprediction (Capistrán and Timmermann 2009). Previous research suggests that farmland market professionals in Indiana may place a greater weight on overprediction relative to underprediction for farmland price expectations (Kuethe, Brewer, and Fiechter 2022). Further, previous research suggests that there are a limited set of circumstances under which asymmetric loss leads to β>0 (Capistrán and Timmermann 2009). We cannot rule out the presence of asymmetric loss, but this paradigm would be conditional on information rigidity.

4. Additional Dimensions of Information Rigidity

Given the finding of information rigidity in the farmland price expectations from Iowa farmland professionals, we examine three more dimensions under which information rigidity may vary. First, we investigate whether information rigidity fully explains information inefficiency in farmland price expectations. Information efficiency implies that all contemporaneous information is contained in expectations. As a result, conditional on information rigidity, contemporaneous information should hold no relationship with expectation errors (Coibion and Gorodnichenko 2015). We examine the degree to which expectation errors are related to contemporaneous information, which should be part of the information set of farmland market professionals. Second, we explore the degree to which information rigidity is state-dependent. Coibion and Gorodnichenko (2015) show that information rigidity in professional forecasters’ inflation expectations is higher during periods of low U.S. GDP volatility. We suspect a similar relationship may be observed among farmland market professionals with respect to volatility in key drivers of farmland market prices, such as commodity prices, interest rates, and whether farmland prices are appreciating or depreciating. Last, we examine the degree to which information rigidity varies by location. Given that farmland prices vary by local market dynamics, agents’ relationship with information may also vary by location.

Contemporaneous Prices

As highlighted by Coibion and Gorodnichenko (2015), information rigidity may not fully explain information inefficiency. Information inefficiency implies that some information is associated with expectation errors. The model established by equation [1] suggests that agents’ behavior in information acquisition and processing can fully explain expectation errors. However, some information that is associated with expectation errors may not fully be reflected in the revision process captured by equation [1]. As a result, Coibion and Gorodnichenko (2015) suggest adding any information that is correlated with expectation errors to equation [1] to examine the degree to which information rigidity may not fully explain information inefficiency.

Following Coibion and Gorodnichenko (2015), we examine the correlation between expectation errors and a number of economic factors that theory suggests play an important role in farmland price determination: corn price, soybean price, interest rates, and inflation.³ The commodity prices are chosen because Iowa is the largest U.S. producer of corn and second-largest producer of soybeans (USDA 2021b). Interest rates are included because of their indirect relationship to farmland prices through costs of mortgages (Featherstone and Baker 1987). Finally, inflation is positively correlated with farmland market prices (Feldstein 1980; Just and Miranowski 1993).

If the correlation between expectation error and any of these variables is statistically significant, we add them as an additional regressor in equation [1] as follows:

2

where x_t−1 represents the additional regressor. The parameter δ captures the correlation between contemporaneous information and expectation error. When δ≠0, the contemporaneous information (x_t−1) holds predictive power for expectation error and information rigidity does not fully explain information inefficiency.

Coibion and Gorodnichenko (2015) stress that this interpretation of δ is only valid when the expectations fail the conventional test for full information efficiency. To provide a concise presentation of our results, the full information efficiency tests suggested by Coibion and Gorodnichenko (2015) are shown in Appendix Table A2. These tests show that corn prices, soybean prices, and interest rates are not correlated with farmland price expectation errors; as a result, we do not estimate equation [2] for these variables.⁴ The only variable correlated with expectation errors is inflation as measured by personal consumption expenditure (PCE).

Table 3 reports estimation results of equation [2] when PCE is represented by x_t−1. Our estimate is statistically significant, which suggests an increase in general price levels in the year before SMLV is associated with underpredicting farmland price changes. This finding is consistent with a positive correlation between farmland prices and inflation (Feldstein 1980; Just and Miranowski 1993). Thus, the inefficiency of farmland price expectations of market professionals may not be fully explained by information rigidity.

View this table:

• View inline
• View popup

Table 3

Ordinary Least Squares Estimates of Inflation (PCE) and Information Rigidities in Iowa Farmland Price Expectations, 1964–2021

State Dependence

Coibion and Gorodnichenko (2015) show that professional forecasters update their information sets or weight new information differently based on broader economic conditions. Specifically, during the “Great Moderation” (1980s–early 2000s), professional forecasters updated their information sets infrequently or only partially used new information. Coibion and Gorodnichenko (2015) use a variety of empirical techniques to show that β in equation [1] may vary over time.

We suspect that farmland market professionals also update their information sets or weight new information differently based on economic conditions in the agricultural sector. We similarly test whether β in equation [1] differs between periods of high and low volatility with respect to corn prices, soybean prices, and interest rates, as well as between periods when farmland prices are rising or falling. During periods of high volatility or falling farmland prices, information may be more valuable. As a result, farmland market professionals may be expected to update their information sets more frequently or place a greater weight on new information. We test the degree to which information rigidity is state-dependent with

3

where D is a dummy variable that takes the value of one for periods with high volatility or increasing farmland prices. State dependence in information rigidity is evaluated using the Wald test (H₀ : β₁=βH₂).

Table 4 reports estimates of equation [3]. Each column represents a test for state dependence across five measures of economic conditions in the agricultural sector. The dummy variable D takes the value of one when (1) the standard deviation of monthly percentage changes in corn prices for the 12 months before the SMLV conference (May) is greater than the median annual standard deviation of corn prices for 1964 to 2021,⁵ (2) the standard deviation of monthly percentage changes in soybean prices for the 12 months before the SMLV conference (May) is greater than the median annual standard deviation of soybean prices for 1964 to 2021, (3) the standard deviation of monthly interest rate for the 12 months before the SMLV conference (May) is greater than the median annual standard deviation of interest rate for 1964 to 2021,⁶ (4) the annual change in farmland price in the previous year was positive, and (5) the annual change in farmland price is greater than the average annual change in farmland price from 1964 to 2021.⁷ With these five defined states, we examine whether the ordinary least squares estimates of β₁ and H₀ : β₁=β₂ are distinguishable from one another.

View this table:

• View inline
• View popup

Table 4

Ordinary Least Squares Estimates of State Dependence of Information Rigidities in Iowa Farmland Value Expectations, 1964–2021

The only variable for which information rigidity differs between periods of high and low volatility is the effective Federal Funds rate. Our empirical estimates suggest that Iowa farmland market professionals are more responsive to new information during periods of low-interest rate volatility. These professionals update information about every six months (λ=0.473) or weight new information by about 50% (G=0.527) for periods of high-interest rate volatility. In times of low-interest rate volatility, these professionals update information about every two and half months (λ=0.219) or weight new information by about 80% (G=0.781). Farmland market professionals may perceive interest rates as a key driver in farmland price determination, because farmland is frequently purchased through mortgages.⁸ Thus, our results suggest that during periods of volatile interest rates, new information may be particularly valuable.