December 11, 2014

Using Degree Days to Value Farmland

A final draft of the working paper joint with Robert Mendelsohn and Shun Chonabayashi is available here.

In this paper we carefully review the use of degree days in the hedonic literature to value farmland and we are not able to confirm the hypothesis of Schlenker, Hanemann and Fisher (2006).

Here is the abstract:

Farmland values have traditionally been valued using seasonal temperature and precipitation. A new strand of the literature uses degree days over the growing season to predict farmland value. We find that degree days and daily temperature are interchangeable over the growing season. However, the way that degree days are used in these recent studies is problematic and leads to biased and inaccurate results. These new findings have serious implications for any study that copies this methodology.

The Appendix to the paper has a careful comparison of the weather data that we use (NARR) and data used by Schlenker and Roberts (2009). For those not familiar with the acronyms, NARR is the North American Regional Reanalysis generated by climatologists at the NOAA. It provides temperature and other climatic measurements over a 32x32 km grid at three hour time intervals from 1979 to present day.

Wolfram Schlenker took great care in examining previous drafts of our work (here, here and here). He has compared NARR data to his dataset and he has found that NARR data is inferior. But that comparison was not correct.

Instead of using NARR 2 meter air temperature, Schlenker used NARR  surface level temperature. This is like comparing pears and apples because Schlenker and Roberts (2009) - SR2009 - is based on weather stations. Weather stations record temperature at about 2 meters. All temperature data that is used in this literature is 2 meter air temperature and we never used surface temperature data. Surface temperature is the temperature of the “skin” of the planet. Surface temperature reflects different soil types. Daily maximum temperature can be very high in the NARR surface temperature dataset (try touching your concrete driveway on a summer afternoon). For this reason the NARR data seems inferior.

In short, both NARR and SR2009 data confirm that Schlenker, Hanemann and Fisher (2006) (SHF2006) greatly overestimate the number of degree days above 34°C. SR2009 and NARR data are quite similar (Wolfram Schlenker kindly gave us his weather data). In some tests NARR data performs better than SR2009 data, but I would not overstress this. The relevant fact is that with both NARR data and SR2009 data we reject the main hypothesis in Schlenker, Hanemann and Fisher (2006) - SHF2006. Why?

We find two problems with SHF2006. First, the weather data used in that paper is not as accurate as in the NARR and the SR2009 datasets and this may have misled the authors. Second, SHF2006 misinterprets agronomic research: farmers and agronomists do not use degree days to predict yields (and thus overall agricultural productivity). The argument in favor of degree days is based on a misreading of an agronomy result showing a linear function rising to 32°C and then abruptly falling (Figure 2-3 in Ritchie and NeSmith, 1991). However, the cited figure does not describe yield but rather the inverse of the time it takes a maize plant to develop a fifth leaf. The figure shows how degree days affect timing. A separate figure in the Ritchie and NeSmith paper reveals the traditional hill-shaped relationship between yield and temperature. In fact, farmers and agronomists use degree days to predict the duration of different stages of plants' growth, not to predict yields.

Finally, we also checked if using hourly temperatures instead of daily temperature to calculate degree days makes a difference and we are able to confirm our results, but this is technical stuff and all the details are in the Appendix.