Spatial effects and endogeneity in a Ricardian model of climate change: an application to a Mediterranean region

ABSTRACT

The relationship between land values, climate and landscape diversity is investigated for a region of the Mediterranean where climate impacts are expected to be particularly severe. The Ricardian approach is applied, and spatial correlation and endogenous selection of farm type are accounted for. The analysis is at the farm level, and it is completely geocoded. It is found that farmland value is affected by both climate and landscape agrobiodiversity, and that their impacts differ across models. However, landscape agrobiodiversity is extremely significant, and it greatly contributes to sustaining the mean level of land value. The results show that spatial correlation and the endogenous nature of adaptation substantially affect impacts assessment and suggest that spatial dependence and adaptation should not be overlooked in Ricardian models.

KEYWORDS:

• spatial econometrics
• Ricardian model
• landscape diversity
• climate change

JEL:

• C31
• Q12
• Q54
• Q57

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Notes

1 Although the Ricardian cross-sectional approach is vulnerable to potential bias from omitted variables (Deschenes & Greenstone, 2007), the lack of observations prevents a panel analysis from being performed. If we performed a panel analysis, we would have reduced the sample size of more than 30%.

2 The values of those attributes are calculated using Analytical Tools Interface for Landscape Assessments (ATtILA) of ArcGIS ESRITM and based on a 5 × 5 km size grid.

3 The agrobiodiversity index is calculated using CORINE land cover inventory 2012 and including only agricultural land use. It is calculated using ATtILA of ArcGIS ESRITM on a 5 × 5 km size grid.

4 Interpolations and cross-validation were executed using the ESRI ArcGIS Geostatistical Analyst.

5 Appendix A in the supplemental data online tests whether the results are robust to the inclusion of annual degree-days. We find that annual degree-days are not more accurate than annual temperature in explaining farmland values and that the results are robust.

6 Sicily is characterized by mild winters, which allow the growth of perennials and crops such as, for example, winter wheat and forage, thus making a four-season model of precipitation more appropriate. A seasonal effect is tested for temperature, but the results reject this hypothesis.

7 Table B1 in Appendix B in the supplemental data online shows the estimates of alternative spatial models.

8 A spatial weights matrix should satisfy one of the following two conditions: (1) the row and column sums of the matrix W and $[I-\delta \mathit{W}{]}^{-1}$ before W is row-normalized should be uniformly bounded in absolute value as N (the number of samples) approaches infinity; or (2) the row and column sums of W before W is row-normalized should not diverge to infinity at a rate equal to or faster than the rate of the sample size N.

9 Among all explanatory variables, only distance from cities and agronomic potentiality show a significant spatial autocorrelation. Since the spatial model that includes only the interaction effects of distance from cities and agronomic potentiality has both a greater fit and greater parsimony than the spatial model including the interactions effects of all exogenous variables, we account only on those two effects.