Use of panel time-series data with cross-section dependence in evaluating farmland valuation: a cautionary note

ABSTRACT

The note highlights how misspecification of cross-section dependence structure in panel time-series data can lead to erroneous conclusions on farmland valuation. Combining the sample information from time-series and cross-section dimensions by using panel time-series data can improve inference on the net present value hypothesis for farmland. However, cross-section dependence must be addressed to take advantage of the additional information from this type of data. We consider three classes of panel unit root models that account for cross-section dependence through (1) common factor extraction, (2) block bootstrapping and (3) spatial dependence to explore whether farmland values can be explained by their economic fundamentals, given that the appropriate cross-section specification is implemented in testing. Results show that only spatial dependence approach accurately characterizes cross-section dependence in the Iowa panel time-series data, highlighting the importance of model selection when using data with cross-section dependence. Once the econometric model is specified with the underlying spatial cross-section dependence structure, the market valuation of Iowa farmland is mainly determined by fundamentals as predicted by the net present value model.

KEYWORDS:

• Farmland values
• net present value
• panel unit roots
• cross-section dependence
• spatial correlation

JEL CLASSIFICATION:

• Q14
• C23

Data availability statement

The data that support the findings of this study are openly available in Ag Decision Maker, Iowa State University Extension and Outreach at https://www.extension.iastate.edu/agdm/wholefarm/html/c2-72.html (reference number c2-72) and https://www.extension.iastate.edu/agdm/wholefarm/html/c2-11.html (reference number c2-11).

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplementary material

Supplemental data for this article can be accessed here.

Notes

¹ See the ‘Data Availability Statement’.

² Selection of the number of common factors, $K$, based on Bai and Ng (2004) Information Criterion (IC) are presented in the Supplemental Online Material, Table A3.

³ Moran’s I statistics indicate persistent global spatial correlation between the Iowa counties over the sample period (Supplementary Online Materials, Table A4); however, they do not provide information on the location of a spatial cluster.

⁴ The blue and red areas (i.e. ‘hot spots’) contribute to significant positive spatial autocorrelation, while the pink and pale blue regions (i.e. ‘cold spots’) contribute to negative global spatial autocorrelation. The areas that do not show spatial correlation are in white.

⁵ The statistical significance of a LISA value is determined by the Monte Carlo randomization procedure with 999 permutations.