using genetic marker selection to reduce asymmetric
TRANSCRIPT
Using Genetic Marker Selection to ReduceAsymmetric Information in the U.S. Beef
Supply Chain
Alex Kappes
27 April 2020
1 / 27
Overview
Background and Objective
Theoretical Environment
Data
Empirical Methodology
Results and Discussion
2 / 27
Background
Bovine Respiratory Disease (BRD)
- Multi-factorial disease
- Largest cause of morbidity and mortality
- $54.12 million estimated feedlot treatment costs
→ Underestimates total costs from BRD incidence
- Metaphylaxis treatment programs not significantly effective
- Calf stress and environment are important factors
3 / 27
Background
Genetic Marker Selection
- Neibergs, H. et al. (2014) have shown that calf geneticsdetermine BRD susceptibility
- Select breeding stock with BRD resistant markers
- Effectively reduces BRD susceptibility compared to othertreatment/prevention programs
- Calf stress and environment still important factors
4 / 27
Background
Production Environment
- Begins with cow-calf producers
→ calves raised alongside birthing cows→ weaned and pre-weaned separation occurs
- Feedlots ultimately source calves from cow-calf producers
- Livestock health risk shifts to feedlots after purchase
5 / 27
Background
Agricultural Asymmetric Information
- Literature focused on crop production
→ input control and scheduled planting decisionsassumed by crop contractors
→ commodity cooperatives→ crop insurance
U.S. Beef Supply Setting
- Feedlots are uninformed about calf health during lotpurchase
- Cow-calf producers hold private information about calfhealth
6 / 27
Objective
1. Show there exists a theoretical price premium mechanismthat provides incentive for cow-calf producers to reveal calfhealth type.
2. Show empirical support for theoretical results.
How are objectives accomplished?
- Develop feedlot’s problem in an adverse selectionenvironment
- Evaluate profit-maximizing model outcomes
- Specify functional forms and empirically analyze modeloutcomes
- Use a stochastic data generating process to test empiricalresults
7 / 27
Contribution
New Markets
- Call attention to beef supply chains
Theoretical
- Show profit-maximizing conditions that support pricepremiums
Empirical
- Provide representative price premium values
8 / 27
Theoretical EnvironmentFeedlot costs
Feedlot cost function C[q(θi), r, t]
Where
q(θi): calf q of health type θ for i = {MS,N}r: per unit production price
t: weekly time period following calf intake
Assumptions
θMS > θN
Ct[q(θN ), r, t] > 0 and Ctt[q(θN ), r, t] > 0
Ct[q(θMS), r, t] > 0 and Ctt[q(θMS), r, t] < 0
limt→t̄
Ct[q(θMS), r, t] = 0
9 / 27
Theoretical EnvironmentFeedlot costs
Figure 1: Feedlot calf health type costs over time
t
C[q(θi), r, t]
C[q(θMS), r, t]
C[q(θN ), r, t]
10 / 27
Theoretical EnvironmentFeedlot problem
Feedlot provides price schedule (TMS , TN ) to cow-calf producers
Expected profit maximization provides optimal (T ∗MS , T∗N )
Let
γ ∈ (0, 1) be the probability of calf type q(θN )
f [q(θi)] be the weight production function for i = {MS,N}p be dressed beef price received
β be the discount factor
11 / 27
Theoretical EnvironmentFeedlot problem
maxq(θMS),q(θN )
T∑t=0
βt(1− γ) {pf [q(θMS)]− C[q(θMS), r, t]− TMS}
+ βtγ {pf [q(θN )]− C[q(θN ), r, t]− TN}
s.t. TMS − c[q(θMS), ZMS ] ≥ 0 (PCMS)
TN − c[q(θN ), ZN ] ≥ 0 (PCN )
TMS − c[q(θMS), ZMS ] ≥ TN − c[q(θMS), ZN ] (ICMS)
TN − c[q(θN ), ZN ] ≥ TMS − c[q(θN ), ZMS ] (ICN )
12 / 27
Theoretical EnvironmentFeedlot problem
Recall θMS > θN
Participation Constraints
(PCMS) binds at the optimum
(PCN ) nonbinding at the optimum
Incentive Compatibility
(ICMS) nonbinding at the optimum
(ICN ) binds at the optimum
13 / 27
Theoretical EnvironmentFeedlot problem
Constraints become
s.t. TMS − c[q(θMS), ZMS ] = 0 (PCMS)
TN − c[q(θN ), ZN ] = TMS − c[q(θN ), ZMS ] (ICN )
With substitution (PCMS)→ (ICN )
TMS = c[q(θMS), ZMS ] (PCMS)
TN = c[q(θMS), ZMS ] + {c[q(θN ), ZN ]− c[q(θN ), ZMS ]} (ICN )
14 / 27
Theoretical EnvironmentFeedlot problem
maxq(θMS),q(θN )
T∑t=0
βt(1− γ) {pf [q(θMS)]− C[q(θMS), r, t]− c[q(θMS), ZMS ]}
+ βtγ {pf [q(θN )]− C[q(θN ), r, t]}
− βtγ {c[q(θMS), ZMS ] + (c[q(θN ), ZN ]− c[q(θN ), ZMS ])}
q∗MS and q∗N solve FOC optimality conditions
pf [q∗θMS] = C[q∗θMS
, r, t] +1
(1− γ)c[q∗θMS
, ZMS ] (1∗)
pf [q∗θN ] = C[q∗θN , r, t] + c[q∗θN , ZN ]− c[q∗θN , ZMS ] (2∗)
15 / 27
Theoretical EnvironmentFeedlot problem
Constructing optimal price schedule (T ∗MS , T∗N ) from
TMS = c[q(θMS), ZMS ]
TN = c[q(θMS), ZMS ] + {c[q(θN ), ZN ]− c[q(θN ), ZMS ]}
From (1∗) for T ∗MS
T ∗MS = c[q∗θMS, ZMS ] = (1− γ)
(pf [q∗θMS
]− C[q∗θMS, r, t]
)(3∗)
From (2∗) and (3∗) for T ∗N
T ∗N =(1− γ){pf [q∗θMS
]− C[q∗θMS, r, t]
}(4∗)
−{C[q∗θN , r, t]− pf [q∗θN ]
}≥ 0
16 / 27
Theoretical EnvironmentFeedlot problem
(3∗) and (4∗) admit price premium expression for q(θMS)
T ∗MS − T ∗
N = (1− γ){pf [q∗θMS
]− C[q∗θMS, r, t]
}−[(1− γ)
{pf [q∗θMS
]− C[q∗θMS, r, t]
}−{C[q∗θN , r, t]− pf [q∗θN ]
}]= C[q∗θN , r, t]− pf [q∗θN ] ≥ 0 (5∗)
17 / 27
Data
Sourced from a large-scale U.S. commercial feedlot in Colorado
Sample includes 952 observations on
- Cattle health and treatment
- Arrival and processing dates
- Processing weight
Representative price and production cost data sourced fromUSDA AMS and LMIC
18 / 27
DataBRD categorization
BRD categorization for feedlot data (Shaefer et al., 2012)
0-5 categorical scale for calf presentation
→ 0: normal respiratory function→ 3: ≥ 103◦F rectal temp, not tachypnic or dyspnic,
with nasal discharge→ 5: ≥ 103◦F rectal temp and tachypnic or dyspnic
Figure 2: Proportion Statistics for Calf Respiratory HealthCharacterization (n = 927)
Normal Respiratory Morbidity Morbidity BRDHealth Scale 3 Scale 5
Proportion 0.525 0.263 0.183 0.446
Note: Each value is a proportional representation of the calf respiratory healthcharacterization based on morbidity scale rankings within the data set
19 / 27
Empirical MethodologyCalf level
Representative calf profit (loss) for i = {MS,N}- Data
E[π(q(θi))] = 1N
∑Nn=1
(pt,nfn[q(θi)]− Cn[q(θi), r, t]
)- Simulation
E[S(M){π(q(θi))}] = 1M
∑Mm=1
(E[s(l){π(q(θi))}]
)for
M sampling iterationsl number of random π(q(θi)) draws
20 / 27
Empirical MethodologyAggregate level
Simulated aggregate profit (loss)
Π(S)(q(θMS), q(θN )) = E[S(M){π(q(θMS))}] + E[S(M){π(q(θN ))}]
Probability of non-genetic marker selection calf q(θN )
γ =1
M
M∑m=1
(∑Ll=1 I[s(l)(q(θN ))]
L
)
21 / 27
ResultsData - calf level
Figure 3: Representative Per-Calf Revenue, Cost, andProfit Outcomes ($/head)
Healthy Calf BRD Calf
Mean Per-Calf Revenue 1786.820 1789.904Mean Per-Calf Cost 1755.227 1799.619Mean Per-Calf Profit 31.039 -11.359
Note: Representative revenue, cost, and profit for a normal res-piratory health and BRD calf are computed as the means ofper-calf costs, revenues, and profits. The means for mean cost,mean revenue, and mean profit of morbidity scale three and fivecalves are computed to represent a BRD calf.
22 / 27
ResultsSimulated - calf level
Figure 4: Summary Statistics for MonteCarlo Representative Per-Calf Profit(-Loss) Outcomes ($/head)
Healthy Calf BRD
Mean 31.654 -15.719Std. Dev 19.553 20.127
Min -34.673 -75.951Max 95.816 57.106
Note: Per-calf profit (-loss) summary statis-tics are computed for Monte Carlo simula-tion profit outcomes of normal respiratoryhealth calves and calves considered to havecontracted BRD
23 / 27
ResultsSimulated - aggregate level
Figure 5: Monte Carlo simulation profit ($) outcomes
Healthy Calf BRD Total Profits
Mean 1718.38 -714.461 1003.922Std. Dev 1066.25 905.943 1416.553Min -1698.99 -3949.47 -3353.120Max 5142.72 2284.23 5674.469
Pr(loss) 0.236γ 0.456
Note: Total profits (losses) are computed as the sum of per-calfprofit outcomes of normal respiratory health calves and calvesconsidered to have contracted BRD. The probability of feedlottotal loss, where total loss from BRD calves is greater than totalprofits from normal respiratory health calves, is the numberof simulated observations resulting in loss over the simulationsample space
24 / 27
Discussion
How do we link BRD categorization to genetic markerselection?
Interpret data as
q(θMS) ≡ morbidity scale 0 ranking
q(θN ) ≡ morbidity scale 3 and 5 ranking
Genetic marker selection fee, price premium, and profits
T ∗MS = (1− γ)(pf [q∗θMS]− C[q∗θMS
, r, t]) = $17.22
T ∗MS − T ∗N = $15.72
π(q(θMS)) > 0
25 / 27
Discussion
Asymmetric information
- Optimal price premium mechanism exists
- Empirical results support theoretical outcomes
- Simulation validates mechanism
Limitations
- Production data
- Cost assumptions
- BRD binary categorization in data
26 / 27