Tor N. Tolhurst

University of California, Davis

PhD Candidate

tntolhurst@ucdavis.edu

Expected Graduation 2020.

Interests. Economics of Food, Agriculture, and the Environment; Econometrics.

Distinctions. Social Sciences and Human Resources Council of Canada Doctoral Fellowship (2018-2019), James Beard Foundation Taste America Scholarship (2016), UC Davis Provost's Entrance Fellowship (2015), Canadian Agricultural Economics Society Outstanding Master's Thesis (2014).

Job Market Paper

Model-free detection of a speculative asset bubble: Evidence from the world market in superstar wines.

Paper (June 2019)

Bubbles occur when an asset price deviates from its fundamental value. Economists have proven asset bubbles are consistent with neoclassical theory and can arise in a variety of laboratory settings; yet cogent, direct evidence of a bubble in an actual market has remained elusive. The challenge for the empiricist is that fundamental values are necessarily unobservable. I propose a bubble test for one of two nearly identical assets assuming the difference in their fundamental values is a latent exchangeable random variable. Their close relationship implies the difference in fundamentals is bounded, bounds which I can find using a concentration inequality. Using this test, I find strong evidence of a bubble in the price of the Bordeaux wine Lafite Rothschild, relative to other fine wines exchanged on global secondary markets. In at least three consecutive periods the upper bound is violated: the probability of this being a random, false detection is less than 0.1 percent. This is the first direct evidence of a bubble which is both consistent with rational bubble theory and independent of a structural valuation model.


Working Papers

Are farmers good neighbors? Evidence from pesticide applications near schools and daycares in California.

with Christopher DeMars, Karen Klonsky, Rachael Goodhue, and Minghua Zhang (lead author); Presentation (Last update: May 2018)

Children are disproportionately susceptible to the toxicity of pesticides, vulnerable to exposure due to frequent hand-to-mouth activity, and at risk of drift exposure when their school or daycare is located on the urban-rural interface. Farmers can reduce the possibility of drift by voluntarily conducting applications near sensitive sites on evenings and weekends, but may not internalize the additional costs to do so. We estimate the extent to which California farmers self-regulate pesticide applications near schools and daycares using administrative records and a quasi-experimental approach. The data record each pesticide application from 2009 to 2014---over 1.8 million unique applications---with farmer-field identifiers, which we merge with the distance from each field to the nearest school and daycare. In our preliminary results we focus on fungicide applications on almond orchards, visible applications typically involving volatile compounds, where we would expect to find a relatively high degree of self-regulation. With 2,761 unique farmers averaging 6.8 fields with applications per year in this subset, we find producers do self-regulate their applications near schools; for example, the propensity to conduct a weekday application decreases from 58.6 percent to 51.4 percent when there is a school within a quarter-mile of a field. However, interestingly, daycares appear less salient: the analogous effect of a daycare is statistically indistinguishable from zero. These basic results are robust to a number of different specifications, applications are trivially less likely on fields near schools, and the propensity to spray near schools does not change significantly during the summer.


Environmental justice and self-regulation of pesticide applications: Evidence from California.

with Rachael Goodhue, Christopher DeMars, and Minghua Zhang (lead author); Presentation (Last update: July 2019).

Pesticides are potent pollutants. Volatile organic compound emissions deplete ozone and produce smog, while runoff contaminates ground and surface water. Exposure can have severe consequences: chronic exposure is known to cause prenatal birth defects, cancer, and nerve tissue damage; acute exposure can be lethal. Children who live or attend school near the rural-urban interface experience the greatest risks. In this paper we investigate environmental justice concerns in the context of pesticide applications near schools and daycares in California. Specifically, we test the null hypothesis that the extent of self-regulation is homogeneous across the demographic and socioeconomic characteristics of the adjacent community. To do so, we merged the California Department of Pesticide Regulation's Pesticide Use Report with census data and confidential field-level spatial information. The preliminary results find evidence consistent with producers self-regulating their applications near schools, but also some evidence consistent with environmental justice concerns. We focus our preliminary results on January to March (pre-bloom) applications of fungicides on almonds, which are prophylactic and thus unlikely subject to intensive margin self-regulation (opting out of conducting any application) which would bias our self-regulation estimates towards zero. First, we find evidence of self-regulation for applications near schools: applications are 11.3 percent less likely during schooldays on fields that are within a quarter-mile of a school compared to those that are not. Second, we find a tightly estimated null effect for environmental justice concerns with respect to the Hispanic population. Third, we do find statistically significant evidence of environmental justice concerns with respect to the proportion of poor children in the surrounding community: applications near a school on schooldays are more likely (i.e. less self-regulation) when the proportion of poor children is higher. The magnitude of this effect is about one-sixth the overall self-regulation effect at the average proportion of poor children. These preliminary results include a number of controls, including for distance to the nearest urban building, application site size (acreage), application intensity (pounds of active ingredient per acre), and dummy variables for aerial and chemigation application methods.


Cheybychev-Cantelli inequality with estimated mean and variance.

Working paper (Last update: May 2019).

I propose a Cheybychev-Cantelli inequality with estimated mean and variance along the lines of Saw, Yang and Mo (1984). The extension is non-trivial because generating an arbitrary mean zero, unit variance vector with the maximal extent of asymmetry does not appear to have a closed form solution. Instead, I use nested grid searches, which guarantees I find the maximum number of "large" positive deviations in the arbitrary vector. For a weakly exchangeable sample of any length, the result of the nested grid searches divided by the length of the vector is the Cheybychev-Cantelli inequality for that sample.


Innovation and climate induced yield volatility in agriculture.

with Alan P. Ker (lead author); Working paper (Last update: July 2017).

When production and income are stochastic---and not a repeated game---volatility matters. We examine how year-to-year volatility in agricultural yields changed over the past half-century, a time of significant innovation. Our findings challenge a number of common views: volatility is increasing asymmetrically such that downside risk is increasing; climate contributes to the increase, but innovation is the dominant driver; and innovation interacts with unobserved heterogeneity (i.e. soil quality) across space. Our findings are important because yield volatility exhausts significant public funding in developed countries, contributes to poverty traps in developing countries, and imperils food security in both.


Measuring cold temperature stress and sowing date uncertainty to estimate climate impacts on agriculture.

Working paper (Last update: May 2015).

Many of the recent advancements in modeling the complex relationship between climate, innovation and yields have focused on maize production in the midwest United States; understandable given its share of world production and readily availability, high-quality data. However, it is unknown to what extent these models can be used in northern latitudes---latitudes that may prove crucial for successful adaptation to higher global temperatures. Northern latitudes introduce two unique but interdependent modelling challenges: (i) accounting for damage from exposure to cold temperatures; and (ii) accounting for significantly greater year-to-year variability in (unobserved) planting and harvest dates. To address the first issue we introduce two new measures of cold temperature damage which, analogous to growing and harmful degree days, are easy to calculate and predictive of yields. For the second we use Bayesian model averaging methods which allows for the length of the growing season to vary from one year to the next.


Additional Work in Progress

Does the crema of the crop raise all boats? The impact of the "Best of Panama" coffee bean auction on Panama's rural economy.

with Jessica Rudder (authorship shared equally).

A central theme of the resource curse (Auty, 1993) is that the resource in question is nonrenewable; but what happens when the valuable natural asset is not minerals or hydrocarbons, but terroir? To examine this question, we propose a case study of a luxury agricultural product, specialty coffee beans from Panama. The most expensive lot of coffee beans in the 2017 Best of Panama auction sold for $601/lb., the median lot for $47.20/lb. By any measure, these are extraordinary prices: $601/lb. is roughly $17/cup wholesale, for just beans. The non-trivial rents available in the auction could have significant spillovers for the rural economy---pushing up wages, stimulating investment and innovation, or capitalizing into land values---or create a "resource curse." We are seeking funds to initiate a project evaluating the impact of this unique auction on Panama's rural economy.


Partially identified distribution of treatment effects with a finite mixture model.

In progress.

The promise of partially identified distributional treatment effects---arguably the ultimate representation of heterogeneity---has led to a great deal of effort in the literature (Williamson and Downs, 1990; Manski, 1990, 1993, 1997, 2003, 2007; Fan and Park, 2009, 2010). However, their promise remains largely unfulfilled: uptake amongst practitioners has been low to nil, likely because the bounds tend to be wide and the underlying assumptions are often neither testable nor economically meaningful. The estimator proposed by (Frandsen and Lefgren, 2018) potentially overcomes these limitations. However, their nonparametric plug-in estimator is non-differentiable and could be outperformed in finite samples by a flexible parametric form. I use simulations to examine the possible advantages of instead plugging-in a finite mixture model, namely: tighter bounds, parameters with economic meaning, and better small sample performance. I then apply the estimator to a number of published randomized controlled trials in order to illustrate the potential benefits of the proposed approach to other practitioners.


A distributional synthetic control estimator: Application to measuring the economic impacts of biotechnology.

In progress.

While synthetic control estimators present a promising avenue for studying policy questions for which a controlled experiment may be impractical or impossible (Abadie, Diamond and Hainmueller, 2010; 2015) to date their focus has been limited to estimating an average "treatment effect." However, for many practical problems higher-order moments are equally (if not more) important: for example, risk is an important economic consideration in evaluating biotechnologies adopted by agricultural producers. I propose a distributional synthetic control estimator, where aggregated data is replaced by panel data and the weighted average synthetic control estimator is replaced by a finite mixture model (essentially a weighted average of distributions). While Lusk, Tack and Hendricks (2017) examine the impact of genetically engineered (GE) seeds on US corn yields, their study design (another application of the panel approach) is limited by the potential selection bias induced by possibly non-random adoption of the biotechnology. Instead, I use yields from European regions (where GE seeds are prohibited), to construct a synthetic Iowa. The distributional yield impact of GE is then simply the difference between observed and synthetic Iowa yields at any point in their respective distributions, which allows me to consider heterogeneity and, in particular, effects at the higher-order moments which are important for agricultural producers (Tack, Harri and Coble, 2012).