Abstract
Hunting leases are impacting rural land values. Using land sales transactions in northern Mississippi, this study quantified how hunting lease income affects forestland value, while controlling for sale attributes, county attributes, and ecosystem differences. Estimation procedures included traditional as well as spatial econometric modeling. A dollar increase in per acre hunting lease rate was associated with a 0.80% increase in per acre forestland value. Hunting lease income was capitalized into forestland value at a 7.55% rate. These findings should be helpful to forest landowners, appraisers, buyers, and government officials in land valuation and natural resource management issues. (JEL Q24, Q26)
I. Introduction
Expenditures on wildlife-associated recreation activities by U.S. households are impacting rural economies in important ways (Munn et al. 2010; Reeder and Brown 2005; Torell et al. 2005). In particular, wildlife recreation-related revenues (e.g., hunting leases, wildlife viewing, fishing) are increasing land values. Baen (1997) found that wildlife recreation income could enhance land values in Texas to the point that recreation becomes the highest and best use of rural land. Henderson and Moore (2006) demonstrated that farmland values in Texas were as much as $65.51 per acre higher in locations with more developed markets for wildlife recreation activities (hunting, fishing, wildlife watching). They estimated that a dollar increase in a county’s per acre recreation income of all types was associated with a 1.3% increase in per acre farmland value; while a dollar increase in a county’s per acre hunting lease rates specifically was associated with a 6% increase in per acre farmland value. Jones et al. (2006) reported that real estate appraisers estimated that 36% of the value of forestland sales in Mississippi was due to recreational opportunities the tracts provided.
While the above-mentioned studies highlight the significance of wildlife recreation-related income streams and their role in inducing changes in rural land values, more work is needed to explore how this relationship varies by recreation type (e.g., consumptive and nonconsumptive) and rural land type (e.g., forestland, cropland, and pastureland). Not all recreation activities impact land values the same way; consumptive recreation activities (e.g., hunting, fishing) are likely to have different impact than nonconsumptive activities (e.g., bird watching), because markets for consumptive recreation activities are relatively mature and resultant income streams, because of their greater reliability, are more likely to translate into higher land values. Moreover, most recreation activities (consumptive and nonconsumptive) are likely to occur on forestlands, which suggests that forestland values are affected more than values of agricultural lands or pasturelands.
It is important to determine how specific recreation activities are driving specific land values. When landowners, appraisers, buyers, and public officials engage in land transactions, they have specific land attributes with specific implicit prices in mind that equate the demand for and supply of land attributes in different uses (agriculture, forestry, housing schemes, etc.). The current study focuses on how hunting lease income is capitalized into forestland values, while controlling for other relevant factors. It is unique because unlike the work of Baen (1997) and Henderson and Moore (2006), it (1) focuses on the narrower relationship between hunting lease income and forestland values, (2) uses individual land sales and hunting lease income data across substate ecoregions rather than county averages, (3) determines the capitalization rate for hunting lease income, and (4) accounts for spatial dependence of land values based on spatial regression methods. Furthermore, unlike Jones et al. (2006), we focus strictly on hunting leases, not all recreational opportunities, and determine lease contributions to land values empirically instead of relying on expert opinion.
II. Background and Model Specification
Theoretical Model of Land Valuation
Earlier research on land valuation employed market equilibrium analysis involving appropriate specifications of the demand and supply of land, and estimated the underlying parameters based on simultaneous equations modeling (Devadoss and Manchu 2007). This approach to land valuation, however, fell into disfavor after it was realized that in any given period the supply of land1 is inelastic (Burt 1986). Since then, income capitalizationmodels have often been used in land valuation studies (Latruffe and Mouel 2009). In the context of land markets, studies have focused on the capitalization of government conservationpayments (Barnard et al. 1997), urban development pressures (Plantinga, Lubowski, and Stavins 2002), open spaces, rural amenities, and wildlife-associated income into land values. To our knowledge, the studies by Baen (1997) and Henderson and Moore (2006) are the only applications to wildlife-associated income.
According to the income capitalization approach, land value is determined as the discounted expected value of the stream of future net returns:
[1]where LV is the current per acre land value; NOIt is net operating income in period t; and rt is the time-varying discount rate in period t used to account for the time value of money. The capitalization rate or cap rate (Cr) refers to the ratio of first-year (expected) net operating income (NOI1) to land value (LV); Cr converts the expected income stream from land into an estimate of land value by dividing the net operating income stream by the capitalization rate. If the income stream is expected to grow at a constant growth rate (g) into the foreseeable future, the value of land is estimated as the present value of a perpetual stream of future net operating income using discount rate r:
[2]Thus, equation [2] essentially expresses land value as a linear function of current net returns, discount rate, and expected growth in net returns to land (Pope 1985). Rearranging [2], Cr equals the total required return on land less the expected growth rate in net returns to land:
[3]While interpreting Cr, it is important to note that (1) Cr is not an internal rate of return on investment because it does not consider changes in projected future income or changes in the value of land over time due to changes in the income stream; (2) Cr is agent specific, meaning that sellers aim at getting the highest price for land (or sell at the lowest cap rate possible) ,whereas buyers try to buy at lowest price possible for land because, from a buyer’s perspective, the higher the cap rate, the better.
Empirical Specification of Forestland Value
Forestland can be characterized as a differentiated factor ofproduction (Palmquist 1989). This motivates the use of the hedonic pricing method to estimate implicit valuations of forestland’s attributes. The method posits that the total price of a differentiated commodity can be approximated by the sum of implicit marginal valuations of its constituent attributes (Rosen 1974). Although the method deals with market outcomes, it is considered a nonmarket valuation method because differentiated commodities have qualities that are not explicitly priced and traded in markets (Jeanty et al. 2002). The hedonic price function is considered a market clearing function resulting from the simultaneous interaction of bid and offer functions for attributes. Since land supply is inelastic in any given period, bid functions are sufficient to derive equilibrium marginal implicit prices (Henning et al. 2001).
In this study, the reduced form of a hedonic price function corresponding to per acre forestland value was specified to depend on four sets of attributes: (1) per acre lease rate, (2) transaction features (year sold, distance to nearest primary road, acres of forestland sold, proportion forested, whether there were existing Wetland Reserve Program [WRP] or Conservation Reserve Program [CRP] easements on the property), (3) characteristics of the county in which the parcel was located (average timber value per acre, population density, population growth, percent of population with graduate or higher degree, adjacency to metropolitan counties, amenity index), and (4) ecoregion2 where the parcel was located (Table 1). So per acre forestland value forestland value (FLV) was modeled as a function of the per acre lease rate, transaction features, county characteristics, and ecosystem characteristics.
Definitions of Variables Included in the Analysis
Per acre forestland value is expected to be positively related to per acre lease rate, ceteris paribus, because investment in land for recreational purposes is one of the strongest factors boosting rural land prices nationwide. Gilliland and Vine (2009) report that rural land markets, once dominated by farmers and ranchers, are now driven by nonagricultural buyers who are best characterized as recreational investors, primarily motivated by a property’s recreational potential. Of the set of transaction features, the coefficient on the time trend is expected to have a positive impact on forestland values, whereas distance to nearest primary road, size of transaction, proportion forested, and whether or not the land is encumbered by CRP or WRP easements are expected to have negative impacts.
The time trend3 used to control for inflation is expected to have a positive coefficient because forestland values have been increasing due to the increase in the overall price level in the economy (Aronow, Binkley, and Washburn 2004; Kennedy et al. 2002). Instead of using a time trend variable, one could deflate the dependent variable using a GDP deflator. We did not follow this route, however, because we wanted to maintain consistency with Henderson and Moore (2006), who used a time trend in their analysis. Distance to nearest primary road affects access costs, which in turn affect the extent of the area within which forest products can be profitably produced (Alig and Plantinga 2004); an increase in the distance to nearest primary road is, thus, expected to negatively impact forestland values. Transaction size (acres sold) is expected to negatively impact forestland values for two reasons: (1) fewer buyers compete for large tracts than for small and medium size tracts (Kennedy et al. 2002), meaning the market for large parcels is thin and the transaction costs are high; and (2) tract size affects how the land will be used: residential or commercial uses would demand smaller sizes, whereas agricultural producers would prefer a larger tract size (Guiling 2007). WRP and CRP easements are expected to negatively influence forestland values because these programs preclude certain future uses that might have economic value. As Taff and Weisberg (2007) point out, if these contracts and their associated restrictions did not have any future negative economic consequences, how could one justify paying for them in the first place?
Of the county attributes, timber value per acre, population density, population growth, percent of population with higher education, and amenity index are expected to have a positive impact on forestland values, whereas the expected impact of proximity to urban areas is ambiguous. Timber value per acre, as a proxy for property-specific timber value, is expected to have a positive impact on forestland value; all else equal, more timber value increases property value (Aronow, Binkley, and Washburn 2004; Kennedy 2002). Population density and population growth are expected to have positive impacts because both of them induce a shift in the demand for land, and consequently a rise in the price of land. Percent of population with a bachelor’s or higher degree is expected to be positively related to land values because an educated labor force translates into higher incomes; as people become wealthier, they tend to demand more land for various uses. The natural amenity index measures the physical characteristics of a county that enhance its desirability as a place to live (McGranahan 1999) and serves as a proxy for the impact of scenic and environmental amenities on land values (e.g., land values tend to increase when located near amenities or streams and lakes). The impact of proximity to urban areas is probably the most complex to predict (Plantinga, Lu-bowski, and Stavins 2002); it can be negative if prospects of government regulations are great, but it can be positive if development prospects are high. In interface areas (e.g., metropolitan or urbanizing locations), the economic hierarchy of land uses suggests that development-related land use factors tend to strongly dominate forestry (Alig and Plantinga 2004).
Regional delineation by ecoregion accounts for differences in forestland values unaccounted for by the above-mentioned factors, whereby an ecoregion is a large unit of land or water containing a geographically distinct assemblage of species, natural communities, and environmental conditions. Ecoregions are complex patterns determined by climate, geology, and the evolutionary history of the planet (Omernik and Griffith 2008) and differ in the type, quality, and quantity of ecosystem services they provide. The ecoregions of northern Mississippi are the Delta, Loess Hills, North Central Hills, Black Prairie, and Tombigbee Hills (Chapman et al. 2004).4
As with all hedonic price models, some sale characteristics could not be included in the empirical model because of missing data. For instance, factors such as the potential for bioenergy production or carbon sequestration may also affect land values. However, markets for these commodities generated on forestland in Mississippi are tenuous at best. Bioenergy production on forestland is still in its infancy, and the market for sequestered carbon collapsed with the demise of the Chicago Climate Exchange.
III. Estimation Procedures
Functional Form
There is no theoretical economic basis for selecting the appropriate functional form of a hedonic price function (Cropper, Deck, and McConnell 1988). Depending on the nature of the differentiated commodity or factor under study (Jeanty et al. 2002), the hedonic price function may be concave (so that marginal implicit price declines with higher levels of an attribute), convex (implying increasing marginal implicit price), or linear (with constant marginal implicit price). Therefore, we tested several functional forms and selected the one with the greatest log-likelihood value. The Box-Cox transformation could not be used because numerous variables in the model take on nonpositive values; the log of a zero or negative number is undefined. The following functional form was ultimately chosen:
[4a]Expressed in logarithmic form, it is written as
[4b]Implicit Prices of Attributes
The implicit price of an attribute is obtained as the partial derivative of the hedonic price function with respect to that attribute. This derivative is a function of the attribute, possibly other attributes, and demand-side factors, unless the hedonic price function is linear. For explanatory variables measured in logarithms (e.g., transaction size and timber value per acre), implicit prices are given by the expression (Stewart 2005)
[5a]
[5b]For explanatory variables in levels (e.g., per acre hunting lease rate, population density), the implicit prices are given by the expression (Stewart 2005)
[5c]For discrete explanatory variables (e.g., WRP, CRP, and ecoregion dummies), implicit prices are given by the expressions (Stewart 2005; Kennedy 1981)
[5d]where SE is the standard error of γj.
Regression Diagnostics
To ensure that assumptions underlying ordinary least squares (OLS) estimation were met, a set of diagnostics tests was performed. To assess if colinearity was a problem, variance inflation factor (VIF) and condition number (CN) were computed. Colinearity is suspected in hedonic pricing models because of the seemingly similar but otherwise distinct explanatory variables. To detect heteroskedasticity, the White test (Stewart 2005) was used; heteroskedasticity was suspected because of the cross-sectional nature of land sales and county data. To assess the adequacy of the hedonic price specification, the Ramsey specification test (Stewart 2005) for omitted variable bias was used. Last, Moran’s test (Pisati 2001) was used to test the null hypothesis of no spatial autocorrelation. Spatial autocorrelation is a frequent phenomenon in land markets (Soto et al. 2004). Forestlandtransactions data could exhibit this feature because the price of a parcel of forestland is likely to be influenced by the prices of nearby parcels, either because appraisers, buyers, and landowners use similar sales in a neighborhood as a reference for determining a transaction price, or there are spillover effects (Can and Megbolugbe 1997) caused by management decisions of nearby landowners.
To test the null hypothesis of no spatial autocorrelation, this study used an inverse distance-based weight matrix (implying that closer neighbors exert greater influence). In view of the significance of the spatial weight matrix in spatial autocorrelation analysis (Soto et al. 2004), we experimented with various distance bands to select a spatial weight matrix that appropriately characterized the presumed spatial relationship between the price of a parcel of land and the prices of neighboring parcels. According to Anselin (1988), a spatial weight matrix should be based on the visualization of spatial patterns and of the interaction among spatial units. Only a careful observation of existing information in each case could reveal the ordering of the data in space and guide the choice of appropriate weights.
The Moran’s I was estimated based on a row-standardized 
inverse distance weight matrix (i.e., spatial correlation is assumed to decrease as the distance between observations increases). The elements of the weight matrix were defined such that wij = 1/dij if dij < c, and wij = 0 if i = j or dij > c; dij is the distance between land transaction i and j (ij = 1,...N); and c is the distance beyond which no spatial correlation is expected; c was set at 10 miles. Under the null hypothesis of no global spatial autocorrelation, the expected value of Moran’s I is given by E(I)= — I/(N — I). Statistical inference is based on z-values, computed by subtracting E(I) from I and dividing the result by the standard deviation of I (Pisati 2001): ZI= I— E(I)/sd(I). A positive and significant z-value for Moran’s I (low p-value) indicates positive spatial autocorrelation (meaning similar values are spatially clustered more than caused purely by chance), whereas a negative and significant z-value indicates negative spatial autocorrelation (meaning similar values are more spatially dispersed than expected under a random pattern).
Regression Analysis in the Presence of Spatial Autocorrelation
Depending on the nature of autocorrelation, a spatial lag model or spatial autoregressive error model may be appropriate. The spatial lag model formalizes the idea that nearby land parcels are more related than distant land parcels. This implies that a spatially weighted sum of neighborhood parcel prices should be included as an explanatory variable in the specification for hedonic price function as shown below:
[6]where ρ is spatial dependence parameter and WpLv is a standardized spatial weight matrix. The spatial weight matrix is row-standardized (every row of the matrix sums to 1) so that the spatial lag term can be interpreted as a spatially weighted average of neighboring parcels prices (Anselin 1988). The null hypothesis of no spatial error is stated as: Ho: ρ = 0. In the presence of spatial autocorrelation in the dependent variable, OLS is biased and inconsistent; the spatial dependence (ρ) parameter must, thus, be estimated simultaneously with other parameters using maximum likelihood (Anselin 1988).
The Spatial Error Model
If the data exhibit spatial autocorrelation in residuals, a spatial error model is needed to improve precision of parameter estimates. The spatial error model has the following form:
[7]where Wε is the weight matrix and λ is a spatial error parameter to be estimated jointly with the regression coefficients; μ is an independently and identically distributed (iid) error term. The null hypothesis of no spatial error is stated as: Ho: λ = 0; OLS estimation is inefficient in the presence of spatial autocorrelation in residuals because the assumption of independence among disturbances is violated (Anselin 1988). In the context of forestland markets, residuals may not be independent if important variables are omitted or measured improperly.
Data Sources and Variable Measurement
The data used in this study were compiled from various sources. Data on land sales (N = 113) with active hunting leases were obtained from the Federal Land Bank System and Mossy Oak Properties, a large, private real estate brokerage firm specializing in rural property. These data included key attributes of the transaction, specifically, year of sale (2004 through 2008), price per acre, number of acres, hunting lease rate per acre, land use pattern (acres in forests, agriculture, pastures and other uses), location (township and range), and whether the property was encumbered by WRP or CRP easements.
Using GIS (ArcView) the township and range statistics were used to estimate distance of the sales to the nearest primary road, determine latitude and longitude for use in spatial autocorrelation analysis, and delineate the study area (northern Mississippi) by ecoregion (Figure 1). Of the 113 sales, 9 sales were discarded because they did not have township and range information; another 8 sales were discarded as they did not have any forest cover (e.g., planted or natural pines, hardwoods). Data on county characteristics (population growth, population density, and percent of population with a bachelor’s or higher degree) for 2008 were obtained from the U.S. Census Bureau. Data on the adjacency of counties (with land sales) to metropolitan counties and amenity indices were obtained from the U.S. Department of Agriculture.5
Land Sale Locations in Mississippi by Ecoregion
To estimate average timber values per acre in each county, data on acreage and volume by species (softwood versus hardwoods) and product class (sawtimber versus pulpwood) for 2006 were obtained from the U.S. Department of Agriculture Forest Service Forest Inventory and Analysis (FIA) database (U.S. Forest Service 2011).6 The corresponding 2006 stumpage prices for northern Mississippi were compiled from Timber Mart-South reports (Norris Foundation 2006). These prices were $421.25 per MBF (Doyle log rule) for pine sawtimber, $20.03 per cord for pine pulpwood, $222.50 per MBF (Doyle log rule) for mixed hardwood sawtimber, and $16.45 per cord for hardwood pulpwood. County volumes by product class were multiplied by product prices, summed, then divided by the acreage of forestland in the county to provide the average per acre value of timber for each county.
IV. Empirical Results
Descriptive Statistics
To provide a sense of the distribution of raw data and likely statistical significance of explanatory variables, Table 2 reports descriptive statistics of the variables used in the regression analysis. Averaged across northern Mississippi ecoregions, per acre forestland value (the dependent variable) is $1,598, with a wide range ($550 to $4,300) that is reflective of significant substate regional differences. Among the explanatory variables, wide variations are exhibited by per acre lease rate, transaction size, percent forested, county timber value per acre, population growth, and proximity to urban areas (adjacent). The average county timber value per acre is $1,551 and ranges from $155 to $2,692. About 12% of the transactions have some land under WRP and another 23% have some land under CRP. Based on the adjacency variable, only 12% of the forestland transactions occurred in proximity of urban areas. Last, the majority of the transactions (42%) occurred in the Delta, distantly followed by the North Central Hills (22%). The shares of the other ecoregions range between 10% and 14%.
Descriptive Statistics of Variables Included in the Analysis
Distinguished by ecoregion (Table 3), average forestland value per acre is lowest in the Loess Hills ecoregion ($1,283) and highest in the Black Prairie ecoregion ($2,216), followed by the North Central Hills ecoregion ($1,796). The Tombigbee Hills ($1,491) and the Delta ($1,479) forestland values closely mimic the overall average forestland value per acre. The average annual lease rate per acre (the explanatory variable of major interest to this study) is $21.62, with a range of $5 to $65. Next, the average transaction size is 242 acres, with a range of 12 acres to 2,505 acres. On average, the properties are 80% forested, but actual forested acreage ranges from 7.8% to 100%. Average county timber values per acre are greatest in the Black Prairie, North Central Hills, and Tombigbee ecoregions, averaging around $2,100 per acre. Timber values are substantially less in the Loess Hills ($1,414 per acre) and lowest in the Delta ($993 per acre).
Land Value, Lease Rate, and Sale Size by Mississippi Ecoregion
The Scheffe multiple comparison test (Webster 1992) was used to determine which pairs of regional means were statistically different (Table 4). First, there were no differences between any pair of the ecoregions with regard to lease rate and transaction size. Second, average county timber values per acre were significantly less in the Delta and Loess Hills than in the Black Prairie, North Central Hills, and Tombigbee Hills. There were no significant differences between the North Central Hills, Tombigbee Hills, and the Black Prairie. The significantly lower county per acre timber values in the Delta and Loess Hills can largely be explained by differences in timber types (which vary by ecoregion) and associated product prices. Hardwood forest types dominate in the Delta and Loess Hills, while pine and mixed-pine types dominate in the other ecoregions, and prices for hardwood timber are substantially less than for pine timber; these differences are reflected in lower average per acre county timber values in the Delta and Loess Hills ecoregions. Finally, mean per acre forestland value was significantly greater in the Black Prairie region than the Delta and Loess Hills. No other pairs of regional forestland value means were significantly different.
Scheffe Multiple Comparisons for Testing Differences between Means
Where significant, differences in forestland values mirrored county-level timber values; however, there were instances where significant differences in timber values occurred without corresponding significant differences in forestland values. There were also instances where there were no significant differences in either forestland or county-level timber values. Thus, it appears that forestland values and timber values may move together in some ecoregions but not necessarily in all of them. In the past, forestland was used mainly for timber production, and thus, property-specific timber and forestland value covaried and, a priori, a strong relationship was expected between the two. The lack of correlation may be due to data issues. In the absence of property-specific timber values, the timber values reported here were computed using county timber inventory estimates as a proxy. Property-specific timber values can vary dramatically around county means, and the true relationship between timber values and forestland values could be masked. Aside from data issues, other factors can explain the weak relationship between these two values. Increasingly, forestland is valued for far more than traditional timber production (Binkley et al. 2006). Ecosystem services (e.g., enhancing water quality, carbon sequestration, wildlife habitat), consumptive (e.g., hunting, fishing) and nonconsumptive (e.g., wildlife viewing, trail riding) recreation, and nontraditional timber products (e.g., pinestraw, biofuels) are factors that could reasonably be expected to contribute to timberland value and could, arguably, vary by ecoregion in light of regional differences in the ability to produce these goods and services. Furthermore, alternative uses such as urban development and agriculture may also affect forestland values locally. Thus, forestland values that differ across ecoregions but that are not associated with similar differences in timber values is not an unreasonable finding.
Model Fit and Regression Diagnostics
The best model fit had forestland value (the dependent variable), transaction size, and timber value per acre (two of the explanatory variables) expressed in logarithms. All the other variables were expressed as levels, proportions, time trends, or dummies. To select the most appropriate model, additional diagnostics tests concerning multicolinearity, heteroskadasticity, specification bias due to omitted variables, and spatial autocorrelation were performed.
Hedonic pricing models are typically plagued by multicolinearity because of the inclusion of seemingly similar but otherwise relevant and distinct explanatory variables in the reduced form specification of the hedonic pricing function. However, the maximum variance inflation factor (VIF)7 of less than 10 (Baum 2006; O’Brien 2007) and condition number of 29.8 (Gujarati 1988) indicated that colinearity was not a major concern. Based on the White test, the null hypothesis of homos-kedasticity (constant variance) could not be rejected at the conventional levels of significance (χ2(95) = 96; p = 0.452). The Ramsey specification test (Stewart 2005) indicated that the null hypothesis of the adequacy of the hedonic price specification (Ho: no omitted variables) could not be rejected (F(3,75) = 2.05; p > F = 0.114). The Moran’s I (I =0.029, p = 0.006), estimated based on the row-standardized inverse distance weight matrix, rejected the null hypotheses of no spatial autocorrelation in per acre forestland value, lease rate per acre, and size of transaction at a 1% level of significance. Spatial lag and spatial error models were, thus, estimated to obtain desirable parameter estimates.
Parameter Estimates
Results of three final models are presented in Table 5. Evaluated in terms of maximized log-likelihood and root mean squared error, both the spatial lag and error models dominate the ordinary least squares regression. The significant lambda (λ) and rho (ρ) parameter estimates further reinforce the idea that spatial autocorrelation needed to be accounted for. Of the two spatial autocorrelation models, the spatial lag is the preferred model based on maximized log-likelihood criterion although in terms of root mean squared error the two spatial regression models are indistinguishable (both models have the same predictive power). Thus, we use the spatial lag model to analyze capitalization of hunting lease income into forestland values, and report implicit prices of explanatory variables for this model only.
Estimation Results of Ordinary Least Squares, Spatial Lag, and Spatial Error Regressions
Consistent with a priori expectations, the coefficient on per acre annual lease rate is highly significant across all models. As this variable is measured in levels, whereas the dependent variable is measured in logarithms, the coefficient is a measure of proportional change in land value given a unit change in lease rate. Accordingly, a dollar increase in per acre annual lease rate is associated with a 0.80% increase in per acre forestland value. Evaluated at the mean per acre forestland value of $1,598, this translates into an implicit price of $13.24 per acre and a capitalization rate (Cr) of 7.55% ( = 1/13.24 × 100). Considering a mean per acre lease rate of $21.62, this capitalization rate indicates that $286.36 ( = $21.62 × $13.24) of the total per acre forestland value, approximately 18%, can be attributed to hunting leases.
Of the set of predictors characterizing the transaction attributes, size of transaction and percent forested are not significant; all the other variables (distance to nearest road, time trend, WRP, CRP) are significant at 10% or better and have appropriate signs. The implicit price of proximity to nearest primary road indicates that per acre value is $56.51 less for forestlands located a mile further away from nearest primary roads than otherwise similar forestlands. The implicit price associated with the time trend (capturing the impact of macroeconomic factors) indicates that in the study area per acre forestland value has been increasing by $95.96 per year, a 6% annual rate. This finding is consistent with previous research, as forestland values have been increasing due to the increase in the overall price level in the economy (Aronow, Binkley, and Washburn 2004; Kennedy et al. 2002). The largest implicit prices are, however, associated with WRP and CRP contracts. Each of these contracts reduces forestland value by about $540 per acre.
Of the county-specific factors, only percent of population with a bachelor’s or higher degree and population growth are statistically significant and have the appropriate signs. A percentage increase in the percent of population with a bachelor’s or higher degree is associated with a $73 increase in per acre forestland value, and a percentage increase in the county population is associated with $139 increase in per acre forestland value. The remaining county-specific variables (per acre timber value, population density, adjacency to metropolitan areas, and amenity index) are not significant.
Last, the set of dummy variables aimed at capturing the impact of ecoregion differences exhibit significant effects. Compared to the Black Prairie ecoregion (base category), forestland values are significantly lower in the Delta ( — $545) and the North Central Hills ( — $508) ecoregions.
V. Conclusions and Discussion
Using a unique land sale dataset, this study quantified the capitalization of hunting lease income into forestland values in northern Mississippi. The results provide strong evidence about the role of hunting lease income in determining forestland values, while controlling for transaction features, county attributes where land sales occurred, and regional ecosystem differences. Specifically, in northern Mississippi, a dollar increase in hunting lease rate is associated with 0.80% increases in forestland values. Evaluated at the per acre mean forestland value of $1,598, this translates into a $13.24 increase in per acre forestland value for a dollar increase in lease rate per acre. This indicates an overall capitalization rate (Cr) of 7.55% for hunting lease income into forestland values. This capitalization rate is five times greater than the corresponding estimate of 1.53% ( = 1/65.51 × 100) reported by Henderson and Moore (2006) for Texas. The higher capitalization rate estimated by the current study is understandable for two reasons. First, the current study focuses on hunting leases, whereas Henderson and Moore include all recreational activities. This narrower focus is important because the hunting lease market is the most fully developed of all recreation markets (i.e., the market for hunting leases is well established, whereas the market for other recreational activities such as wildlife viewing or trail riding is sporadic at best) and thus its value is more likely to be fully captured in market transactions. Second, the current study focuses on forestland, whereas Henderson and Moore (2006) include all rural land, which can include a variety of land uses. By including all land uses in their analysis, the economic benefits of recreational activities are dispersed across all land use types even though some land uses do not support these activities. For example, forestland is ideally suited for hunting, whereas row crop land is not. Thus, this study clearly demonstrates that when analyzing the role of wildlife-associated recreation income in influencing rural land values, it is important to be explicit about the type of recreation income (e.g., wildlifewatching, fishing, or hunting) and the type of rural land (e.g., agricultural land, forestland, or pastureland).
Ecoregions reflect the diversity of parent materials, climate, biological factors, and topography in the state, inducing differences in game quality (Strickland and Demarais 2000), and game quality is an important predictor of hunting lease rates (Rhyne, Munn, and Hussain 2009). However, the findings of the current study do not corroborate with previous research as we do not find significant differences in lease rates across ecoregions. While future research is needed to advance our understanding of lease rate differences in northern Mississippi, it is possible for a single lease rate to prevail despite ecoregion differences emphasized by Strickland and Demarais (2000) and game quality differences (e.g., species featured such as deer, turkey ,and waterfowl; size of deer antlers), because while important, they are not the sole determinants of lease rates. Rather lease rates are determined by an array of forces underlying the demand for and supply of hunting access, many of which extend across ecoregion boundaries (Munn and Hussain 2010). One important determinant of demand, for example, is the population of hunters in the market for hunting leases. Given that all the study sites are within two to three hours’ driving time of each other—a bearable driving distance for hunters accessing hunting lands— the pool of hunters seeking hunting leases is not restricted by ecoregion boundaries.
In contrast to lease rates, forestland values do vary by ecoregion. After controlling for transaction features and county attributes where land sales occurred, Delta and North Central Hills forestland values were significantly lower than those for the Black Prairie ecoregion. Given that hunting lease values, which do not vary by ecoregion, determine only 18% of forestland values, other factors determine the remaining 82%. As previously discussed, ecoregions vary in their ability to produce ecosystem services, nonconsumptive recreation, and nontraditional forest products, as well as in their suitability for other land uses. With all these potential differences between ecoregions and the large majority of forestland value being determined by nonlease factors, significant differences in land values across ecoregions are possible even though lease rates are uniform.
Of the transaction-specific attributes, size of transaction and percent forested are not significant. The lack of significance of the coefficient on size of transaction is unexpected, but it could be reflective of the distinct northern Mississippi forestland market or a feature of the data used in this analysis. Previous research studies are, however, unanimous about the significant negative impact of size of sale on land values, regardless of whether estimation is based on OLS (Turner, Newton, and Dennis 1991; Snyder et al. 2007) or spatial autocorrelation models (Kennedy et al. 2002). The lack of significance of percent forested is understandable; while it was expected to have a significant negative impact, as forestlands tend to be less productive and hence less valuable than agricultural lands, the data used in the current study essentially concern forested parcels. The highly significant negative coefficients associated with WRP and CRP are consistent with a priori expectations (Latruffe and Mouel 2009; Taff and Weisberg 2007). Both severely restrict management options available to landowners. The relatively larger coefficient on WRP compared to CRP is also understandable, as WRP contracts impose more severe restrictions on land use and hence entail higher opportunity costs, but the difference between the two coefficients is not significant (F(1,78) = 0.35; p > F = 0.55).
The role of county attributes in influencing land values is well documented. The findings of this study, however, do not agree with previous research findings (Henderson and Moore 2006; Devadoss and Manchu 2007). Population growth and percent of population with a bachelor or higher degree (proxying county income level) are the only variables that have significant coefficients. Percent population with a bachelor or higher degree was included after county median income failed to contribute to a better fit. The coefficient needs to be cautiously interpreted because only two-thirds of the land buyers were
Mississippi residents, and the possibility exists that some of these may not reside in the counties where sales occurred. Further research on the role of county-level attributes in influencing forestland values is also needed. The current study employed a limited data set involving just 14 counties.
As a proxy for the property-specific timber value, the average per acre timber value in the county was expected to influence forestland value; however, this proved not to be the case. Since timber inventories can vary dramatically from property to property and it is the value of timber on the parcel in question, rather than countywide average, that matters to buyers and sellers, this finding is understandable. It underscores the importance of parcel-specific information (i.e., site productivity, species composition, age distribution, planting and management costs, cost share payments, and taxes) in addition to prices by product class (e.g., pine sawtimber, hardwood pulp) to have a refined estimate of the value of timber inventory, and quantifying its impact on forestland values.
The insights of this research should be valuable to Mississippi land appraisers and real estate institutions in their efforts to serve their clients; to forestland owners in deciding to provide fee-access hunting, sell their lands, or manage their lands for wildlife; and to state and federal regulators in their appraisal of regional development projects. More work is, however, needed to refine these findings. Developing markets such as carbon offsets may result in enhanced recreational value through delayed harvests (Hartman 1976). In contrast, developing markets for bioenergy may result in diminished recreational value due to shorter rotations and more intensive management (Meehan, Hurlbert, and Gratton 2010). Such factors need to be recognized as these markets become established.
Acknowledgments
We wish to thank the reviewers for thoughtful and very helpful comments throughout the review process. We also wish to thank Katarzyna Grala for assistance with our graphics. Funding for the study was provided by the U.S. Fish and Wildlife Service.
Footnotes
The authors are, respectively, research economist, The Wilderness Society, Anchorage, Alaska; professor, Department of Forestry, Mississippi State University; graduate student, Department of Wildlife, Fisheries, and Aquaculture, Mississippi State University; associate professor, Department of Wildlife, Fisheries, and Aquaculture, Mississippi State University; and assistant extension professor, Department of Forestry, Mississippi State University.
↵1 While the total supply of land is fixed, the supply of land for a specific use is positively sloped in the long run. The relevant issue concerns land supply to a particular use given a range of potential uses (agriculture, forestry, rangeland, etc.). In principle, land will transfer from use A to B if highest long-run return is in use B, and would depend upon the costs of transfer, perception of risk, and the institutional setting (e.g., government regulations involving land use constraints). For more details see Golub et al. (2006).
↵2 Ecoregions (areas of general similarity in ecosystems and in the type, quality, and quantity of environmental resources) are designed to serve as a spatial framework for the research, assessment, management, and monitoring of ecosystems and ecosystem components (Bryce, Omernik, and Larsen 1999). These general purpose regions are critical for structuring and implementing ecosystem management strategies across federal agencies, state agencies, and nongovernment organizations that are responsible for different types of resources within the same geographical areas (Omernik et al. 2000).
↵3 The inclusion of a time trend variable rather than annual dummies was necessitated by the relatively short time series and degrees of freedom considerations. See Taff and Weisberg (2007) for the use of annual dummies to account for inflation when the length of the time series is not a constraint.
↵4 A brief description of the ecoregions comprising northern Mississippi: (1) Tombigbee is a region of undulating, irregular plains and clayey soils with generally poor drainage. Although there are small areas of cropland and pasture in the valleys and on gently sloping ridges, the region is mostly forested with oak-hickory-pine forests. Ridge tops are often dominated by shortleaf pine, while hardwood forests are common on slopes. The presence of chestnut oak and Virginia pine distinguish this region in Mississippi. (2) Black Prairie used to have dominant trees of sweetgum, oak, and red cedar. Today, the area is mostly cropland and pasture, with small patches of mixed hardwoods, red cedar, and pines. Pond-raised catfish aquaculture occurs in some parts of this region. (3) North Central Hills was once a highly productive agricultural area in Mississippi. Many areas are now in pine plantations or have reverted to a mixed forest landscape. The gently rolling to irregular plains are a contrast to the more dissected bluffs of Loess Hills. The loess layer tends to be thinner than neighboring Loess Hills and thins more to the east. Streams and rivers tend to be low gradient and murky. Severe erosion in earlier years contributed heavy sediment loads to downstream reaches. (4) Loess ecoregion consists primarily of irregular plains, some gently rolling hills. Thick loess is one of the distinguishing characteristics. The bluff hills in the western portion contain soils that are very deep, steep, silty, and erosive. Oak-hickory and oak-hickory-pine were the dominant natural vegetation, but now there is a mosaic of forest and cropland. While oakhickory forest is the general natural vegetation type, some of the undisturbed bluff vegetation is rich in beech and maples. (5) Delta is mostly a broad, flat alluvial plain with river terraces, swales, and levees. Soils are typically finer-textured and more poorly drained, although there are some areas of coarser, better-drained soils. The region contained one of the largest continuous wetland systems in North America and is still a major bird migration corridor. Most of the original bottomland hardwood forest has been replaced with cropland of soybeans, cotton, rice, farmed wetlands, pastureland, or large catfish ponds. For complete details, see www.epa.gov/wed/pages/ecoregions/ms_eco.htm.
↵5 U.S. Department of Agriculture data available at www.ers.usda.gov/briefing/rurality/ruralurbcon/.
↵6 FIA data are collected in roughly five-year intervals. The 2006 data were the only inventory data available during our study period; however, since 2006 is the midpoint of our study period (2004-2008) it is a reasonable choice to represent timber volumes for the period.
↵7 A VIF of 10 indicates that (all other things being equal) the variance of the ith regression coefficient is 10 times greater than it would have been if the ith independent variable had been linearly independent of the other independent variable in the analysis. Thus, it tells us how much the variance has been inflated by this lack of independence (O’Brien 2007).
