1. Conceptual Model

Producers face uncertain outcomes when utilizing risk management practices. They will choose to implement a practice if their expected utility of profits when using the practice exceeds their expected utility of profits without the practice. Following Moschini and Hennessy (Reference Moschini, Hennessy, Gardner and Rausser2001), we assume feedlot operators make decisions on output price and animal health risk management by comparing the expected utility of profit from different scenarios. Assume feedlot operator $i$ will make decisions to maximize their expected utility:

(1) $$E U_i = E \left[ U_i \left( w_{0,i} + \tilde{\pi _i} \right) \right]$$

where $E{U_{i}}$ is the expected utility of feedlot operator i, ${w_{0,i}}$ is initial wealth, and $\tilde{\pi _i}$ is profit from the cattle feeding enterprise which is a random variable (i subscripts are hereafter omitted for convenience). Profit for the cattle feeding enterprise is the sum of profit per pen ( $b$ pens):

(2) $$\tilde{\pi} =\sum \limits_b \tilde{\pi _b}.$$

Profit per pen of cattle is a function of input and output prices and quantities. However, when feedlot operators place cattle there is uncertainty about prices and quantities—making profit a random variable. Following Moschini and Hennessy (Reference Moschini, Hennessy, Gardner and Rausser2001), profit can be rewritten as:

(3) $$\tilde{\pi} = PG (x;\ \tilde{e})- rx - K$$

where P is output price, $G ( x; \tilde{e})$ is a stochastic production function where realized output depends on the input vector $x$ and a random variable $\tilde{e}$ , $r\;$ is a vector of input prices, and $K\;$ is fixed costs. This framework can be adapted to feedlot operators’ decision-making under price and animal health risk, holding all else equal. In online supplementary Appendix A, we consider two demonstrative scenarios, allowing one risk type to vary while holding the other fixed.

One link between mitigating output price risk and utilizing animal health production practices could be the expectation of total pounds of finished cattle versus the actual pounds produced. Expected and actual pounds produced can vary from weather, animal disease, and management, among other factors. These production risks can result in reductions in ADG per animal or death loss. A large variance in pounds produced per pen could alter producers’ risk mitigation strategies. For example, feedlot operators may be less likely to establish an expected selling price because they cannot properly assess the number of futures contracts needed or specifications they should agree to in a forward contract. However, if animal health production practices decrease finishing weight variability and death loss, then operators may make more informed output price risk management decisions.

Substitute, complementary, or no relationship could exist between output price risk and animal health risk mitigation strategies. Risk mitigation strategies are not free and feedlot operations have a limited budget. A feedlot operator could decide the feedlot should only invest in animal health mitigation strategies instead of also managing output price risk—an example of substitution. Alternatively, operators could view output price and animal health risk mitigation strategies as complements. Instead there could be no relationship between feedlot operators’ decisions regarding price risk and animal health risk mitigation strategies. Determining this relationship is a core component of our analysis. We hypothesize there is some relationship between output price risk and animal health risk mitigation strategies. However, to investigate this hypothesis, we need to analyze individual feedlot operators’ decision-making.

2. Data Collection

Primary data were collected from feedlot operators, see online supplementary Appendix B for the survey instrument. The survey was programmed for Web application using Qualtrics software (Qualtrics Provo, UT, USA). Feedlots in Colorado, Iowa, Kansas, Nebraska, and Texas were surveyed. These states comprise five of the eight states in the widely cited, “5-market” average price reported by the USDA. Furthermore, these states house nearly 31% of U.S. feedlots with sales for slaughter and 76% of feedlot sales according to the 2017 Census of Agriculture (USDA-NASS, 2019a). The Colorado Livestock Association, Iowa Cattlemen’s Association, Kansas Livestock Association, Nebraska Cattleman, and Texas Cattle Feeders Association distributed a uniform resource locator through an email list of members. To increase survey response and expand distribution, Feedlot Magazine also distributed the survey web address to its subscribers.Footnote ²

After answering several introductory questions on the survey, respondents were asked to participate in a choice experiment. The respondent’s past use of risk management and attitudes concerning risk were also obtained in the survey.

The survey was live from January 19, 2017 to February 14, 2017.Footnote ³ There were 588 responses.Footnote ⁴ However, 232 participants who did not have a feedlot enterprise and/or did not make price or animal health risk management decisions were dismissed from the survey after the qualification questions. Additionally, 75 participants who qualified to continue but did not answer the choice experiment questions were excluded—reducing the useable sample to 281.

Table 1 reports selected survey respondent characteristics. The sample is representative of U.S. feedlot producers. Feedlot operators from Iowa comprise 50% of the sample, Nebraska 19%, Texas 10%, Kansas 6%, and Colorado 5%. According to the 2017 Census of Agriculture, there were 9,309 feedlot operations with sales for slaughter in these five states: 4% from Colorado, 59% from Iowa, 11% from Kansas, 22% from Nebraska, and 4% from Texas (USDA-NASS, 2019a). Fifty-eight percent of respondents are from operations with capacity over 1,000 head. With respect to December 1, 2017, the Census of Agriculture reports operations with 1,000 or more head of cattle on feed comprised 12% of feedlots from these five states but 86% of the cattle on feed inventory (USDA-NASS, 2019a). Thus, the operations within our sample are larger than the census average but do represent the majority of feedlot inventories.Footnote ⁵ Just over 20% of survey participants are considered custom feeders owning less than 40% of cattle in their feedlot.Footnote ⁶ According to the 2017 Census of Agriculture, in these five states 7% of farms custom fed cattle shipped directly for slaughter, accounting for 42% of the sales for slaughter (USDA-NASS, 2019a).

Table 1.

Summary statistics

The average respondent age is 49 years old, with a minimum and maximum age of 23 and 85 years. In the 2017 Census of Agriculture, the simple average age of cattle feedlot producers was 55 for the five surveyed states (USDA-NASS, 2019b).Footnote ⁷ Given that our survey was administered online, the younger average age is expected. Nearly half of the participants have at least a Bachelor’s degree. This educational attainment is similar to other studies of beef producers. In McKendree, Tonsor and Wolf (Reference McKendree, Tonsor and Wolf2018), 51% of cow–calf producers surveyed had earned at least a Bachelor’s degree.

Participants were asked questions to gauge their risk aversion following the Global Risk-Attitude Construct (GRAC) defined in Pennings and Garcia (Reference Pennings and Garcia2001). It was determined that factor variables were not needed as only one GRAC question captures risk attitudes. Therefore, participants are considered risk averse (nearly 57%) if they somewhat agree, agree, or strongly agree with the statement, “I usually like ‘playing it safe’ (for instance, ‘locking in a price’) instead of taking risks for market prices for fed cattle.”

Since animal health and price risk management are of key interest, participants were asked about their past price determination methods and past feeder cattle sourcing. Participants were considered to actively manage animal health risk if they purchased feeder cattle from a single source. Nearly 65% of participants have purchased single source calves before. This finding is consistent with NAHMS beef feedlot 2011 study which found that 56.4% of feedlots had purchased feeder cattle through direct sales. However, direct sales accounted for less than 30% of feeder cattle purchased (USDA-APHIS-VS-NAHMS, 2013). Participants were also asked how they believed calves from a single source perform compared to calves sourced with unknown backgrounds (Table 2). Over 85% of producers stated single source calves performed somewhat or much better than calves from unknown backgrounds.Footnote ⁸

Table 2.

Participants’ response to “Compared to calves sourced from auctions with unknown backgrounds, how do you believe calves from a single source ranch perform (i.e. average daily gain, feed conversion, morbidity) in the feedlot?”

There was variability in futures hedging and forward contracting behavior, with the percentage use of each ranging from 0 to 100% (Table 3). On average, participants hedged 19% of finished cattle using futures contracts and 18% utilized forward contracts (Tables 1 and 3). Spot cash market was the most frequently used with over 50% of producers selling at least 50% of their cattle this way. LRP insurance and LGM insurance were rarely used.

Table 3.

Participants’ response to “In the past 12 months, what percentage of the following pricing methods did your operation use for marketing finished cattle (should sum to 100%)”

3. Research Methodology: Past Behavior

The survey contained two questions regarding past risk management behavior which serves as the first test of whether a relationship exists between price and animal health risk management. The first question was designed to identify feeder cattle sources. Respondents were asked to allocate the percentage (summing to 100%) for each source including traditional auction; satellite/video auction; purchased direct from seller (ranch); home raised from own cow-herd; custom fed, so I did not buy or own animals; and other. We choose to look at participants’ use of purchasing direct from seller (ranch) as the health risk mitigation strategy of interest. Participants were considered to mitigate animal health risk if they purchased feeder cattle directly from the seller (ranch). The second question was designed to identify pricing methods for marketing finished cattle. Respondents were asked to allocate the percentage (summing to 100%) for each method including spot cash market; forward contract or marketing agreement; futures hedge; options hedge; LRP insurance; LGM insurance; and other. Cattle marketed using the spot price only were considered to not be mitigating price risk.

Tobit models were utilized to estimate the relationship between past behavior of purchasing feeder animals direct from seller and output price risk management. The two latent variables of interest (indicated with a * subscript), the percent of feeder cattle purchased direct from seller ( $directseller_i^*$ ), and the percent of finished cattle marketed on the spot cash market ( $spot_i^*$ ) were modeled as:

(4) $$directseller_i^{*} = {\delta_1}spo{t_i} + {\bf\it X}_{direct,i}^{\rm{'}} {\it\bf \beta} _{direct} + {\varepsilon _{direct,i}}$$

(5) $$spot_i^* = {\delta_2}directselle{r_i} + {\bf\it X}_{spot,i}' {\it\bf \beta} _{spot} + {\varepsilon _{spot,i}}$$

where the relationships between the latent variables and the observed variables are

(6) $$directselle{r_i} = \left\{ {\matrix{ {directseller_i^{\rm{*}}} \cr {0} \cr {100} \cr } {\rm{}}\matrix{ {if\;{\rm{}}0 \le directseller_i^{\rm{*}} \le 100} \cr {if\;{\rm{}}directseller_i^{\rm{*}} \lt 0} \cr {if{\rm{}}\;directseller_i^{\rm{*}} \gt 100} \cr } } \right.$$

(7) $$spo{t_i} = \left\{ {\matrix{ {spot_i^*} \cr {0} \cr {100} \cr } \matrix{ {if\;0 \le spot_i^* \le 100} \cr {if\;spot_i^* \lt 0} \cr {if\;spot_i^* \gt 100.} \cr } } \right.$$

In equations (4) and (5), ${\delta _1}$ and ${\delta _2}$ are the coefficients of interest. ${\bf\it X}_{S,i}'$ (where $\;S = direct,\;spot$ ) is a vector of explanatory variables for each individual $i$ and an intercept, ${\it\bf \beta} _S$ are coefficient estimate vectors, and ${\varepsilon _{S,i}} \sim N\left( {0,\sigma _S^2} \right)$ . Equations (4) and (5) are estimated with maximum likelihood. Models were estimated using the cmp command in Stata (Roodman, Reference Roodman2011).

5. Research Methodology: Choice Experiment

Each respondent completed a choice experiment, designed to not be overly complex, which resembled a realistic turn of cattle in a feedlot and decisions regarding either feeder cattle procurement or live cattle marketing. To assess individual feedlot operators’ decision-making process, operators were placed in a realistic decision-making mindset where they were making decisions and forming expectations around events that will happen in the future. They were asked to make decisions as if it were February 15, 2017 for feeder animals being placed in March 2017 with an expected August 2017 closeout.

A seven-treatment design (Table 6) was utilized to test if a relationship exists between animal health and output price risk management. Comparing results across scenarios isolates differences of central interest, similar to Tonsor, Schroeder, and Lusk (Reference Tonsor, Schroeder and Lusk2013). The animal health, feeder cattle procurement practice of interest was known single source feeder steers versus feeder steers of unknown background. The live cattle output price risk management strategies were futures hedge, forward contract, other, or none (accept cash price at the time of sale). An additional difference across designs is how the expected futures basis was presented. The futures hedge basis was presented two ways: unambiguous (e.g., −$1.00/cwt) or ambiguous (e.g., 35% chance of being less than −$1.00/cwt and 65% chance of being greater than −$1.00/cwt) (Di Mauro and Maffioletti, Reference Di Mauro and Maffioletti2004). Basis ambiguity was included to understand how producers form their price expectations and how basis uncertainty might alter risk mitigation decisions.

Table 6.

Split-sample design

Each participant was randomly assigned to one of the seven treatments (Table 6). Treatments fall into two broad categories consistent with the initial assessment of past behavior: feeder cattle procurement (treatments 1–3) or live cattle marketing (treatments 4–7). Treatments 1–3 consisted of two choice scenarios about procuring a lot of feeder steers, see Figure 1 for an example of treatment 2. Participants were given the following information:

Figure 1.

Treatment 2 example.

Note: The two questions were presented on successive screens and not simultaneously.

Then they were asked,

In this first question, no output pricing information is given and the exact same initial question is given in treatments 1–3. However, in the second question additional potential output pricing information is provided as an information shock. In treatments 1 and 3, participants are provided information needed for a futures hedge, including the August CME live cattle futures contract price and expected local basis. In treatment 1, the futures basis is unambiguous, but is ambiguous in treatment 3. In treatment 2, information for a forward contract, including the August CME live cattle futures contract price and offered basis, is provided. By comparing responses across the two questions, we can test our core hypothesis as it relates to feeder cattle procurement.

Treatments 4–7 each include one scenario where the participant was told they just purchased 150 head of feeder steers for March placement which they expected to sell in August (Figure 2). A random August CME live cattle futures contract price was also provided. Treatments 4 and 5 are the base treatments where no feeder cattle source information was given. In treatments 6 and 7, participants were also told the steers were purchased from a single source and given a random premium paid (information shock). After this introductory information, participants were asked how many head they would place in each of the four output pricing strategies provided—futures hedge, forward contract, other output price strategy, or accept local cash price at the time of sale. In treatments 5 and 7, an ambiguous live cattle basis for futures hedges was presented while basis was unambiguous in treatments 4 and 6. By comparing responses across treatments, we can understand if/how producers alter decisions when animal health and price risks are individually versus jointly examined. In particular, treatments 4 and 6 (non-ambiguous basis) can be compared, and treatments 5 and 7 (ambiguous basis) can be compared. To keep the manuscript concise, methods and results for treatments 4–7 can be found in online supplementary Appendix C.

Figure 2.

Treatment 7 example.

Values of key variables in the choice design were randomly drawn for each participant from a range selected to match current market conditions. The source premium shown ranged from $1.00 to $10.00/cwt (Blank, Saitone, and Sexton, Reference Blank, Saitone and Sexton2016), the August CME live cattle futures contract price ranged from $95.00 to $110.00/cwt (consistent with the market as of January 9, 2017), all basis numbers ranged from −$5.00 to $5.00/cwt (consistent with historical basis numbers from the Livestock Marketing Information Center [LMIC] [2016]), and the random ambiguous basis percent ranged from 1 to 99%.

The choice experiments were hypothetical; however, our instructions specifically stated, “[…]. It is important that you make your selection as if you were actually facing these choices in operation of your feed yard.” Cheap talk scripts, such as the one provided, have been shown to reduce hypothetical bias in choice experiment research (Cummings and Taylor, Reference Cummings and Taylor1999; Lusk, Reference Lusk2003; Tonsor and Shupp, Reference Tonsor and Shupp2011). Furthermore, Lusk and Schroeder (Reference Lusk and Schroeder2004) found that although total willingness to pay was overstated in hypothetical choice experiments, marginal willingness to pay was not statistically different across hypothetical and actual payment scenarios. Thus, hypothetical bias concerns are mitigated since our core hypotheses tests depend on net differences across treatments (Tonsor, Reference Tonsor2011).

Econometrically, systems of Tobit models are utilized because the dependent variables (either feeder cattle purchased or head placed in each output price risk strategy) are continuous but censored between 0 and 150. Using these methods, marginal effects can be calculated and compared across designs to identify if relationships exist between animal health risk mitigation and output price risk mitigation.

5.1. Feeder Cattle Placement Scenarios (Treatments 1–3)

For treatments 1–3, the two latent variables of interest (indicated with a * subscript) are the total head purchased when output pricing information is not shown ( $feederheadA_i^*$ ) and total head purchased when output price information is shown ( $feederheadB_i^*$ ). These variables can be modeled as:

(8) $$feederheadA_i^* = {\bf\it X'}_{A,i}{\it\bf\beta} _A + {\varepsilon _{A,i}}$$

(9) $$feederheadB_i^* = {\bf\it X'}_{B,i}{\it\bf\beta} _B + {\varepsilon _{B,i}}$$

where the relationships between the latent variables and the observed variables are

(10) $$feederheadA_i^{\rm{}} = \left\{ {\matrix{ {feederheadA_i^{\rm{*}}} \cr {0} \cr {150} \cr } {\rm{}}\matrix{ {if\;{\rm{}}0 \le feederheadA_i^{\rm{*}} \le 150} \cr {if\;{\rm{}}feederheadA_i^{\rm{*}} \lt 0} \cr {if{\rm{}}\;feederheadA_i^{\rm{*}} \gt 150} \cr } } \right.$$

$$feederheadB_i^{\rm{}} = \left\{ {\matrix{ {feederheadB_i^{\rm{*}}} \cr {0} \cr {150} \cr } {\rm{}}\matrix{ {if\;{\rm{}}0 \le feederheadB_i^{\rm{*}} \le 150} \cr {if{\rm{}}\;feederheadB_i^{\rm{*}} \lt 0} \cr {if\;{\rm{}}feederheadB_i^{\rm{*}} \gt 150.} \cr } } \right.$$

In equations (8) and (9), ${\bf\it X}_{k,i}'$ (where $k = A,\;B$ ) is a vector of information given in the question (e.g., source premium, CME price, basis) and explanatory variables for each individual $i$ , ${\it\bf \beta} _k$ are coefficient estimate vectors, and ${\varepsilon _{k,i}} \sim N\left( {0,\sigma _k^2} \right)$ . Equations (8) and (9) are modeled jointly with maximum likelihood. The error terms ${\varepsilon _{A,i}}$ and ${\varepsilon _{B,i}}$ are specified following a bivariate normal distribution with zero mean, standard deviations $\sigma _A^2$ and $\sigma _B^2$ , and correlation $\rho $ . By estimating these equations jointly, we can test if unobservable factors are impacting total head purchased in each question. If $\rho $ is zero, then the equations can be estimated independently (Cornick, Cox, and Gould, Reference Cornick, Cox and Gould1994).