SlideShare une entreprise Scribd logo
1  sur  44
Télécharger pour lire hors ligne
Introduction     Data     Model           Identification             Estimation             Results




          Estimation of a Dynamic Agricultural Production
           Model with Observed, Subjective Distributions

                                  Brian Dillon

                              Cornell University
                         and Harvard Kennedy School


                             August 30, 2012




                           Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                         Motivation: crop production

       To grow crops, farmers solve a dynamic resource allocation problem

       The problem is not unlike many other dynamic choice problems:
       portfolio management, inventory management, human capital
       investment

       The solution to this problem can involve delay of some choices,
       distribution of activities across time, and updating of expectations
       as new information arrives

       Between-farmer variation in expectations clearly matters (Gin´,
                                                                    e
       Townsend, Vickery 2008)



                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                  What if we measure expectations?


       Early literature in agricultural economics (Bessler and Moore 1979;
       Eales 1990)

       Manski (2004) makes the case for measuring expectations

       Nyarko and Schotter (2002) show that there is a big difference
       between observed and estimated expectations

       Delavande et al (2010) review the recent development literature
       that uses subjective probabilities




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction       Data        Model           Identification             Estimation             Results



               What we get from measuring expectations



       Two contributions to the estimation of dynamic choice models:

         1. Allow us to relax the rational expectations assumptions that
            are standard for these models (Wolpin 1987; Rust 1987, 1997;
            Fafchamps 1993)
         2. There is a lot of information in a subjective distribution over
            an endogenous outcome




                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                Why go through a structural exercise?
       Apart from the pure value of estimating a less restricted
       production function...




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction       Data        Model           Identification             Estimation             Results



                Why go through a structural exercise?
       Apart from the pure value of estimating a less restricted
       production function...
       Production elasticities tell us something about resilience of the
       production process to shocks




                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction          Data       Model           Identification             Estimation             Results



                   Why go through a structural exercise?
       Apart from the pure value of estimating a less restricted
       production function...
       Production elasticities tell us something about resilience of the
       production process to shocks
       What we know about shocks already largely deals with
           • Consumption/asset smoothing (Townsend 1994, Morduch
               1995, Hoddinott 2006, Barrett and Carter 2006, Jacoby and
               Skoufias 1998, Fafchamps et al 1998)
           • Human capital (Hoddinott and Kinsey 2001, Aguilar and
               Vicarelli 2012)
           • Two papers look at how farmers move labor across time,
               within a season: Fafchamps (1993) and Kochar (1999)


                                  Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                Why go through a structural exercise?



       And we can also simulate important, relevant policies:

         1. Insurance
         2. Forward contracting
         3. Improvements in information delivery
         4. Changes in input supply




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction       Data        Model           Identification             Estimation             Results



                          Contributions of this paper
       We use a sequence of observed inputs, price expectations, and
       yield expectations to estimate an agricultural production function
       Methodological contributions:
         1. Develop a general method for estimating dynamic choice
            models with observed subjective distributions
         2. Show how counterfactual choice data (“How much pesticide
            did you want to apply last week?”) can be used in estimation

       Substantive contributions:
         1. Recover estimates of all elasticities of substitution between
            inputs (within and across periods)
         2. Simulate the impact of insurance, forward contracting, and
            information provision policies

                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction        Data       Model           Identification             Estimation             Results



                                Plan of the talk




           • Data set
           • Model basics
           • Identification of shock densities
           • Estimation
           • Results




                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction          Data        Model           Identification             Estimation             Results



                                          Data set


           • 195 cotton farmers in 15 villages in NW Tanzania
           • Face-to-face agriculture and LSMS surveys conducted in
               summer 2009 and summer 2010
           • From September 2009 - June 2010: investment, time use,
               shocks, agricultural input and output, and other data
               gathered every 3 weeks
           • High frequency interviews also gathered subjective probability
               distributions over end-of-season prices and yields, and
               qualitative distributions over pest pressure and rainfall at
               various points throughout the year




                                   Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction         Data         Model           Identification             Estimation                Results



                     Measuring subjective distributions
                                 +,-./0!1,!2,+3,!
           !
                                  4+567879:;<:=>5?!




           #!   $!          %!    &!       "!      !!        *!        )!        (!       '!


                                                                                            !!    !
                                   Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction         Data               Model           Identification             Estimation             Results



               Evolution of subjective price distributions


                       5

                       4

                       3
       Mean number
         of stones     2

                       1

                       0

                            6
                                5
                                    4                                                            7
                                                                                           6
                                          3                                         5
                Survey period                                                 4
                                                2                       3
                                                    1             2                Bin number
                                                            1

                                         Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction                                   Data                      Model                  Identification                      Estimation            Results



                                      Evolution of subjective yield distributions
                                                 ?34+)(*+),-.$                                                     ?34+)(*+),-.%
       ! "!# "$ "$# "% "%#




                                                                                         "%
                                                                                         "$#
                                                                                         "$
                                                                                         "!#
                                                                                         !
                             !#              !               #                   $!            !#               !              #                $!
                                         &'()*+),-./0,1.*2+3*4.5)'46                                       &'()*+),-./0,1.*2+3*4.5)'46
                             7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"#!$>                         7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"#@#$



                                                 ?34+)(*+),-.A                                                        E*0('F+
       ">




                                                                                         "D
       "A




                                                                                         ">
       "%




                                                                                         "%
       "$
       !




                                                                                         !




                             !$!         !#           !           #              $!              !#               !             #               $!
                                         &'()*+),-./0,1.*2+3*4.5)'46                                       &'()*+),-./0,1.*2+3*4.5)'46
                             7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"AB#C                         7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"%A##


                                                                                                                                                     G




                                                                          Brian Dillon         Estimation of a Dynamic Agricultural Production Model with O
Introduction          Data          Model           Identification             Estimation             Results



                             Smoothing of distributions

       Let
           • xi be a response vector
           • d ∈ RN+1 be the interval boundaries
           • z be the random variable in question
           • k be the number of counters

       We fit a four parameter beta CDF, Gi (z | a, b, ρ, κ), by solving:

                                            N       j                                            2
                                                    m=1 xj
         (ai , bi , ρi , κi ) = arg inf                        − G (dj+1 | a, b, ρ, κ)
                                  a,b,ρ,κ             k
                                            j=1




                                     Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction          Data                   Model                     Identification                     Estimation    Results



                                Sample summary statistics
                                                                 Mean            sd            Min           Max
               Household size (people)                            8.33          3.90            2             23
               Dependency ratio*                                  1.31          0.85            0            5.5
               Head age                                          46.85         14.69           20            100
               Head is male (%)                                   85.0            -             -              -
               Years of education (HH head)                       4.19          3.46            0             11
               Radios                                             0.83          0.71            0              4
               Bicycles                                           1.19          1.00            0             10
               Dairy cattle                                       1.33          2.84            0             20
               Non-dairy cattle                                   3.87          7.89            0             60
               Goats                                              5.27          8.05            0             50
               Sheep                                              1.67          3.74            0             30
               Total acres                                        9.67         11.03            1             82
               Number of plots                                    2.71          1.17            1              7
               Number of crops grown                              3.45          1.26            1              8
               Labor expenditure (TSH)                          78,248        139,485           0         1,020,000
               Fertilizer expenditure (TSH)                     21,149         81,359           0          715,000
               Animal labor expenditure (TSH)                   33,497         92,724           0          750,000
               Transport expenditure (TSH)                      10,333         20,049           0          144,000
               Other cultivation expenditure (TSH)               6,929         15,817           0          100,000
               Total cultivation expenditure (TSH)              150,156       254,863           0         1,514,700
               Notes: author's calculation from survey data; cultivation data refers to 2008-2009 cultivation of all
               crops; 1 USD ! 1 400 TSH; *Dependency ratio is number of persons aged < 15 or aged > 65 divided
                             !"1,400
               by number aged between 15 and 65.


                                               Brian Dillon           Estimation of a Dynamic Agricultural Production Model with O
Introduction        Data       Model           Identification             Estimation             Results



                             Model assumptions

       Important:
         1. Farmers are dynamically consistent (will relax, if we have time)
         2. Independence of shocks across time




                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction        Data       Model           Identification             Estimation             Results



                             Model assumptions

       Important:
         1. Farmers are dynamically consistent (will relax, if we have time)
         2. Independence of shocks across time

       Less fundamental:
         1. Separable household model
         2. Risk-neutral maximization of expected plot-level profits
         3. All forms of labor are interchangeable
         4. No binding credit constraints
         5. Functional form choices



                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction     Data             Model            Identification            Estimation             Results



                                    Crop evolution

       Expanding on Fafchamps (1993), crops grow according to:

                        yi0 = φi Ai e θi0
                        yi1 = h1 (yi0 , li1 , pi1 )e θi1
                        yi2 = h2 (yi1 , li2 , pi2 )e θi2
                         yi   = h3 (yi2 , li3 , pi3 )e θi3


                 where             θit ∼ git (θit )          for t = 0, . . . , 3
                                   Ai is acreage
                                   φi is a plot-specific yield shifter
                                   li and pi are labor and pesticides



                                   Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction          Data          Model           Identification             Estimation             Results



                                 Crop evolution cont.

       We use nested CES functions:
                                                                                                1
                                                γ       γ                  γ
         h1 (y0 , l1 , p1 | α1 , α2 , γ) = [α1 y0 + α2 l1 + (1 − α1 − α2 )p1 ] γ
                                                                                               1
                                                    δ       δ                  δ               δ
          h2 (y1 , l2 , p2 | β1 , β2 , δ) =     β1 y1 + β2 l2 + (1 − β1 − β2 )p2
                                                                                                1
                                                ω       ω                  ω
         h3 (y2 , l3 , p3 | κ1 , κ2 , ω) = [κ1 y2 + κ2 l3 + (1 − κ1 − κ2 )p3 ] ω



       Which gives us 9 production parameters to estimate:
           • Share parameters (α1 , α2 , β1 , β2 , κ1 , κ2 )
           • Transformed elasticity parameters (γ, δ, ω)



                                     Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction          Data            Model              Identification              Estimation                Results



                             Farmer’s objective function


       From the viewpoint of the first period:

                                                                                                 δ
                                                γ       γ               γ
           max E [qc ]Eθi1 θi2 θi3    κ1 β1 α1 yi0 +α2 li1 +(1−α1 −α2 )pi1                       γ
                                                                                                     e δθi1
           li1 ,pi1
                                                                   ω
                                                                   δ
                              ∗δ                           ∗δ                       ∗ω
                      +   β2 li2   + (1 − β1 −        β2 )pi2          e ωθi2 + κ2 li3
                                                              1               3
                                                              ω
                                                     ∗ω
                             + (1 − κ1 −        κ2 )pi3           e θi3 −          (ql lit + qp pit )
                                                                             t=1




                                       Brian Dillon      Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                          Identification of gt (θt )


       We need measures of gt (θt ) in order to proceed




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                          Identification of gt (θt )


       We need measures of gt (θt ) in order to proceed

       Nested fixed point method (Rust 1987): iterate between guesses of
       production parameters and gt parameters until convergence




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                          Identification of gt (θt )


       We need measures of gt (θt ) in order to proceed

       Nested fixed point method (Rust 1987): iterate between guesses of
       production parameters and gt parameters until convergence

       But we only observe subjective output distributions
       Ψ0 (y ), . . . , Ψ3 (y )




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction       Data       Model           Identification             Estimation             Results



                          Identification of gt (θt )


       We need measures of gt (θt ) in order to proceed

       Nested fixed point method (Rust 1987): iterate between guesses of
       production parameters and gt parameters until convergence

       But we only observe subjective output distributions
       Ψ0 (y ), . . . , Ψ3 (y )

       We can use those to directly estimate gt (θt ), within the context of
       the model




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data     Model           Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection




                                                                                y reported




                         Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data     Model           Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection



                                                                         !3 realized




                                                                               y reported




                         Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data     Model           Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection
                                                              (l3 , p3) chosen


                                                                         !3 realized




                                                                               y reported




                                                             "3(y) reported
                                                               incl:
                                                               g3(!3)


                         Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data     Model           Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection
                                                              (l3 , p3) chosen


                                                    !2 realized          !3 realized




                                                                                y reported




                                                             "3(y) reported
                                                               incl:
                                                               g3(!3)


                         Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data     Model             Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection
                                        (l2 , p2) chosen        (l3 , p3) chosen


                                                      !2 realized          !3 realized




                                                                                  y reported


                                        "2(y) reported
                                         incl:
                                         g2(!2)                "3(y) reported
                                         g3(!3)                  incl:
                                         (l3* , p3*)             g3(!3)


                         Brian Dillon     Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data     Model              Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection
                                        (l2 , p2) chosen         (l3 , p3) chosen


                                !1 realized           !2 realized           !3 realized




                                                                                  y reported


                                        "2(y) reported
                                         incl:
                                         g2(!2)                "3(y) reported
                                         g3(!3)                  incl:
                                         (l3* , p3*)             g3(!3)


                         Brian Dillon      Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data           Model              Identification             Estimation            Results



     Timeline of decisions, realizations, and data collection
                      (l1 , p1) chosen        (l2 , p2) chosen        (l3 , p3) chosen


                                      !1 realized            !2 realized          !3 realized




                                                                                        y reported
                      "1(y) reported
                       incl:
                       g1(!1)                 "2(y) reported
                       g2(!2)                  incl:
                       g3(!3)                  g2(!2)                "3(y) reported
                       (l2* , p2*)             g3(!3)                  incl:
                       (l3* , p3*)             (l3* , p3*)             g3(!3)


                               Brian Dillon     Estimation of a Dynamic Agricultural Production Model with O
Introduction    Data           Model              Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection
                       (l1 , p1) chosen        (l2 , p2) chosen         (l3 , p3) chosen


               !0 realized             !1 realized           !2 realized           !3 realized




                                                                                         y reported
                       "1(y) reported
                        incl:
                        g1(!1)                 "2(y) reported
                        g2(!2)                  incl:
                        g3(!3)                  g2(!2)                 "3(y) reported
                        (l2* , p2*)             g3(!3)                   incl:
                        (l3* , p3*)             (l3* , p3*)              g3(!3)


                                Brian Dillon      Estimation of a Dynamic Agricultural Production Model with O
Introduction         Data           Model               Identification             Estimation             Results



     Timeline of decisions, realizations, and data collection
                            (l1 , p1) chosen         (l2 , p2) chosen        (l3 , p3) chosen


                    !0 realized             !1 realized            !2 realized          !3 realized




        "0(y) reported
         incl:                                                                                 y reported
         g0(!0)             "1(y) reported
         g1(!1)              incl:
         g2(!2)              g1(!1)                  "2(y) reported
         g3(!3)              g2(!2)                   incl:
         (l1* , p1*)         g3(!3)                   g2(!2)                "3(y) reported
         (l2* , p2*)         (l2* , p2*)              g3(!3)                  incl:
         (l3* , p3*)         (l3* , p3*)              (l3* , p3*)             g3(!3)


                                      Brian Dillon      Estimation of a Dynamic Agricultural Production Model with O
Introduction       Data         Model           Identification             Estimation             Results



                           Identification of g3 (θ3 )


       Taking the normalization E [e θt ] = 1 for all t:

                     Pr[y < Y ] = Pr E [y |Ω3 ]e θ3 < Y

                                                              Y
                                    = Pr θ3 ≤ ln
                                                           E [y |Ω3 ]

       where Ω3 is the period 3 information set

       ⇒ g3 (θ3 ) is constructed by transforming ψ3 (y )




                                 Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction       Data        Model           Identification             Estimation             Results



                      Key proposition (summarized)



       Proposition
       If h = H(θ1 , θ2 ) is a function of two random variables, and
         1. We know densities fh (h) and fθ2 (θ2 )
         2. H is monotonic in θ1
       then we can consistently estimate fθ1 (θ1 ) by taking repeated draws
       from fh and fθ2




                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction       Data           Model             Identification             Estimation             Results



                            Identification of g2 (θ2 )


                  ∗ ∗
       Plugging (l3 , p3 ) into the definition of output allows us to write
       output from the period 2 perspective as:


                y=        H2 φ, α1 , α2 , β1 , β2 , κ1 , κ2 , γ, δ, ω;
                          A, l1 , p1 , l2 , p2 ; ql , qp , E [qc ]; θ0 , θ1 e θ2 e θ3



                            ∞
       And E [y |Ω2 ] =     −∞ y ψ2 (y )dy        = H2 (·)




                                   Brian Dillon     Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data         Model            Identification             Estimation             Results



                         Identification of g2 (θ2 ) cont.



       This gives a method for numerically estimating g1 (θ1 ) using
       repeated draws from ψ1 (y ) and g2 (θ2 )

                                    M
                              1                    ym
               Pr[θ2 < Θ2 ] =            I ln                    − θ3m ≤ Θ1
                              M                 E [y |Ω2 ]
                                  m=1




                                Brian Dillon    Estimation of a Dynamic Agricultural Production Model with O
Introduction        Data       Model           Identification             Estimation             Results



                           Estimation of θt and φ



       Given any guess of the parameters, we find the realized values of
       the shocks:
           • θ0 , θ1 , θ2 come from FOC of the farmer’s decision problem
           • θ3 comes from realized output y and ψ3 (θ3 )

       Lastly
                                        ∞
                               ˆ        −∞ y ψ0 (y )dy
                               φ=
                                               A




                                Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction            Data            Model              Identification             Estimation             Results



                                     Likelihood function


       Then the joint likelihood for the observed inputs, output and
       distributions is:

               L(α1 , α2 , β1 , β2 , κ1 , κ2 , γ, δ, ω |
                                                                 P
                                                                            i        i        i        i
               φ, A, l, p, y , ql , qp , E [qc ], θ0 , θ1 =           gi0 (θ0 )gi1 (θ1 )gi2 (θ2 )gi3 (θ3 )
                                                                i=1

       We maximize the log likelihood over the 9 production parameters
       and: α1 , α2 , β1 , β2 , κ1 , κ2 , γ, δ, ω




                                         Brian Dillon      Estimation of a Dynamic Agricultural Production Model with O
Introduction   Data                Model               Identification             Estimation            Results



                           Results: shock densities

                            Summary statistics for gt (θt )
                      Variable                        Mean               s.d.
                      !0 lower bound                  -2.95              1.86
                      !0 upper bound                  2.43               1.70
                      E[!0]                           -0.14              0.61
                      !1 lower bound                  -2.49              1.84
                      !1 upper bound                  2.01               1.48
                      E[!01]                          -0.01              0.58
                      !2 lower bound                  -4.19            1.4807*
                      !2 upper bound                  3.19             1.4807*
                      E[!2]                           -2.35             1.17
                      N                                212              212
                      *SD of !2 upper and lower bounds is constant by
                      construction, because both reflect variation in acreage




                                    Brian Dillon      Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                                   Conclusion


       Separation of output equation into its dynamic and stochastic
       components is not a necessary condition for this to work

       But monotonicity in θt is necessary

       Observation of shock densities reduces number of parameters to be
       estimated

       But it also increases the pressure on the functional form, because
       the error variance does not adjust to increase the contribution of
       very low probability parameter contributions to the likelihood




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction      Data        Model           Identification             Estimation             Results



                           Where things stand...


       Ongoing work on this paper involves:
         1. Embedding the farmer’s problem in a utility framework
         2. Comparing results with those from the nested fixed point
            method
         3. Interpretation and simulations
         4. Relaxing the dynamic consistency assumption?
            → could use data on counterfactual, optimal pesticide
            application




                               Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O
Introduction     Data   Model           Identification             Estimation             Results




       Thanks!




                         Brian Dillon   Estimation of a Dynamic Agricultural Production Model with O

Contenu connexe

En vedette

09.27.2012 - Santosh Anagol
09.27.2012 - Santosh Anagol09.27.2012 - Santosh Anagol
09.27.2012 - Santosh AnagolAMDSeminarSeries
 
07.12.2012 - Aprajit Mahajan
07.12.2012 - Aprajit Mahajan07.12.2012 - Aprajit Mahajan
07.12.2012 - Aprajit MahajanAMDSeminarSeries
 
08.23.2012 - Karthik Muralidharan
08.23.2012 - Karthik Muralidharan08.23.2012 - Karthik Muralidharan
08.23.2012 - Karthik MuralidharanAMDSeminarSeries
 
قانون تداول المعلومات
قانون تداول المعلوماتقانون تداول المعلومات
قانون تداول المعلوماتMohamed Howity
 
مشروع قانون الصكوك
مشروع قانون الصكوكمشروع قانون الصكوك
مشروع قانون الصكوكMohamed Howity
 
12.06.2012 - Giacomo de Giorgi
12.06.2012 - Giacomo de Giorgi12.06.2012 - Giacomo de Giorgi
12.06.2012 - Giacomo de GiorgiAMDSeminarSeries
 
Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...
Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...
Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...Maciej Malesa
 

En vedette (10)

Boslaa (2)
Boslaa (2)Boslaa (2)
Boslaa (2)
 
09.27.2012 - Santosh Anagol
09.27.2012 - Santosh Anagol09.27.2012 - Santosh Anagol
09.27.2012 - Santosh Anagol
 
07.12.2012 - Aprajit Mahajan
07.12.2012 - Aprajit Mahajan07.12.2012 - Aprajit Mahajan
07.12.2012 - Aprajit Mahajan
 
08.23.2012 - Karthik Muralidharan
08.23.2012 - Karthik Muralidharan08.23.2012 - Karthik Muralidharan
08.23.2012 - Karthik Muralidharan
 
قانون تداول المعلومات
قانون تداول المعلوماتقانون تداول المعلومات
قانون تداول المعلومات
 
مشروع قانون الصكوك
مشروع قانون الصكوكمشروع قانون الصكوك
مشروع قانون الصكوك
 
12.06.2012 - Giacomo de Giorgi
12.06.2012 - Giacomo de Giorgi12.06.2012 - Giacomo de Giorgi
12.06.2012 - Giacomo de Giorgi
 
Boslaa ofst
Boslaa ofstBoslaa ofst
Boslaa ofst
 
Sukuk
SukukSukuk
Sukuk
 
Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...
Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...
Deloitte 2016 - EKF - Digital Customer Journey - conference material - f...
 

Similaire à 08.30.2012 - Brian Dillon

De-Mystefying Predictive Analytics
De-Mystefying Predictive AnalyticsDe-Mystefying Predictive Analytics
De-Mystefying Predictive AnalyticsGalit Shmueli
 
Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...
Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...
Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...Plan4Demand
 
The use of biodata for employee selection: Past research and future implications
The use of biodata for employee selection: Past research and future implicationsThe use of biodata for employee selection: Past research and future implications
The use of biodata for employee selection: Past research and future implicationsAndrea Dvorak
 
Module_1___Analytical_Thinking___Problem_Solving.ppt.pptx
Module_1___Analytical_Thinking___Problem_Solving.ppt.pptxModule_1___Analytical_Thinking___Problem_Solving.ppt.pptx
Module_1___Analytical_Thinking___Problem_Solving.ppt.pptxsolomonvijayanand2
 
Data Quality Considerations for CECL Measurement
Data Quality Considerations for CECL MeasurementData Quality Considerations for CECL Measurement
Data Quality Considerations for CECL MeasurementLibby Bierman
 
ValueBeforeDelivery_Gilb_Final
ValueBeforeDelivery_Gilb_FinalValueBeforeDelivery_Gilb_Final
ValueBeforeDelivery_Gilb_FinalJimmy Chou
 
Literature review: measurement of client outcomes in homelessness services
Literature review: measurement of client outcomes in homelessness servicesLiterature review: measurement of client outcomes in homelessness services
Literature review: measurement of client outcomes in homelessness servicesMark Planigale
 
AIMS presentation_AEA 2016_final
AIMS presentation_AEA 2016_finalAIMS presentation_AEA 2016_final
AIMS presentation_AEA 2016_finalMikhail Paltsyn
 
Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...
Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...
Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...CS, NcState
 
Enablers for Maturing your S&OP Processes, SherTrack
Enablers for Maturing your S&OP Processes, SherTrackEnablers for Maturing your S&OP Processes, SherTrack
Enablers for Maturing your S&OP Processes, SherTrackInnovation Enterprise
 
Governance of Risk in Public Policy - Nigel Gibbens
Governance of Risk in Public Policy - Nigel GibbensGovernance of Risk in Public Policy - Nigel Gibbens
Governance of Risk in Public Policy - Nigel Gibbensmliebenrood
 
Sequence-to-Sequence Modeling for Time Series
Sequence-to-Sequence Modeling for Time SeriesSequence-to-Sequence Modeling for Time Series
Sequence-to-Sequence Modeling for Time SeriesArun Kejariwal
 
Forward-Looking ALLL: Computing Qualitative Adjustments
Forward-Looking ALLL: Computing Qualitative AdjustmentsForward-Looking ALLL: Computing Qualitative Adjustments
Forward-Looking ALLL: Computing Qualitative AdjustmentsLibby Bierman
 

Similaire à 08.30.2012 - Brian Dillon (20)

Final Poster
Final PosterFinal Poster
Final Poster
 
De-Mystefying Predictive Analytics
De-Mystefying Predictive AnalyticsDe-Mystefying Predictive Analytics
De-Mystefying Predictive Analytics
 
Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...
Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...
Demand Planning Leadership Exchange: SAP APO DP Statistical Forecast Optimiza...
 
Earnings forecasting
Earnings forecastingEarnings forecasting
Earnings forecasting
 
Impact evaluation
Impact evaluationImpact evaluation
Impact evaluation
 
The use of biodata for employee selection: Past research and future implications
The use of biodata for employee selection: Past research and future implicationsThe use of biodata for employee selection: Past research and future implications
The use of biodata for employee selection: Past research and future implications
 
Module_1___Analytical_Thinking___Problem_Solving.ppt.pptx
Module_1___Analytical_Thinking___Problem_Solving.ppt.pptxModule_1___Analytical_Thinking___Problem_Solving.ppt.pptx
Module_1___Analytical_Thinking___Problem_Solving.ppt.pptx
 
Data Quality Considerations for CECL Measurement
Data Quality Considerations for CECL MeasurementData Quality Considerations for CECL Measurement
Data Quality Considerations for CECL Measurement
 
ValueBeforeDelivery_Gilb_Final
ValueBeforeDelivery_Gilb_FinalValueBeforeDelivery_Gilb_Final
ValueBeforeDelivery_Gilb_Final
 
Mirage hh dakar_december2011_0
Mirage hh dakar_december2011_0Mirage hh dakar_december2011_0
Mirage hh dakar_december2011_0
 
Literature review: measurement of client outcomes in homelessness services
Literature review: measurement of client outcomes in homelessness servicesLiterature review: measurement of client outcomes in homelessness services
Literature review: measurement of client outcomes in homelessness services
 
AIMS presentation_AEA 2016_final
AIMS presentation_AEA 2016_finalAIMS presentation_AEA 2016_final
AIMS presentation_AEA 2016_final
 
Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...
Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...
Promise 2011: "Local Bias and its Impacts on the Performance of Parametric Es...
 
cas_washington_nov2010_web
cas_washington_nov2010_webcas_washington_nov2010_web
cas_washington_nov2010_web
 
Enablers for Maturing your S&OP Processes, SherTrack
Enablers for Maturing your S&OP Processes, SherTrackEnablers for Maturing your S&OP Processes, SherTrack
Enablers for Maturing your S&OP Processes, SherTrack
 
Governance of Risk in Public Policy - Nigel Gibbens
Governance of Risk in Public Policy - Nigel GibbensGovernance of Risk in Public Policy - Nigel Gibbens
Governance of Risk in Public Policy - Nigel Gibbens
 
Final Six Sigma
Final Six SigmaFinal Six Sigma
Final Six Sigma
 
Sequence-to-Sequence Modeling for Time Series
Sequence-to-Sequence Modeling for Time SeriesSequence-to-Sequence Modeling for Time Series
Sequence-to-Sequence Modeling for Time Series
 
Forward-Looking ALLL: Computing Qualitative Adjustments
Forward-Looking ALLL: Computing Qualitative AdjustmentsForward-Looking ALLL: Computing Qualitative Adjustments
Forward-Looking ALLL: Computing Qualitative Adjustments
 
Kenett on info q and pse
Kenett on info q and pseKenett on info q and pse
Kenett on info q and pse
 

Plus de AMDSeminarSeries

02.16.2014 - Carolina Castilla
02.16.2014 - Carolina Castilla02.16.2014 - Carolina Castilla
02.16.2014 - Carolina CastillaAMDSeminarSeries
 
09.19.2013 - Laura Schechter
09.19.2013 - Laura Schechter09.19.2013 - Laura Schechter
09.19.2013 - Laura SchechterAMDSeminarSeries
 
08.15.2013 - Robert Jensen
08.15.2013 - Robert Jensen08.15.2013 - Robert Jensen
08.15.2013 - Robert JensenAMDSeminarSeries
 
07.18.2013 - Michael Clemens
07.18.2013 - Michael Clemens07.18.2013 - Michael Clemens
07.18.2013 - Michael ClemensAMDSeminarSeries
 
06.20.2013 - Nishith Prakash
06.20.2013 - Nishith Prakash06.20.2013 - Nishith Prakash
06.20.2013 - Nishith PrakashAMDSeminarSeries
 
05.16.2013 - Dilip Mookherjee
05.16.2013 - Dilip Mookherjee05.16.2013 - Dilip Mookherjee
05.16.2013 - Dilip MookherjeeAMDSeminarSeries
 
05.02.2013 - Jonathan Robinson
05.02.2013 - Jonathan Robinson05.02.2013 - Jonathan Robinson
05.02.2013 - Jonathan RobinsonAMDSeminarSeries
 
02.07.2013 - Mark Rosenzweig
02.07.2013 - Mark Rosenzweig02.07.2013 - Mark Rosenzweig
02.07.2013 - Mark RosenzweigAMDSeminarSeries
 
01.17.2013 - Garance Genicot
01.17.2013 - Garance Genicot01.17.2013 - Garance Genicot
01.17.2013 - Garance GenicotAMDSeminarSeries
 
12.10.2012 - Catherine Wolfram
12.10.2012 - Catherine Wolfram12.10.2012 - Catherine Wolfram
12.10.2012 - Catherine WolframAMDSeminarSeries
 
11.29.2012 - Marta Vicarelli
11.29.2012 - Marta Vicarelli11.29.2012 - Marta Vicarelli
11.29.2012 - Marta VicarelliAMDSeminarSeries
 
10.25.2012 - Craig McIntosh
10.25.2012 - Craig McIntosh10.25.2012 - Craig McIntosh
10.25.2012 - Craig McIntoshAMDSeminarSeries
 
09.13.2012 - Seema Jayachandran
09.13.2012 - Seema Jayachandran09.13.2012 - Seema Jayachandran
09.13.2012 - Seema JayachandranAMDSeminarSeries
 
08.09.2012 - Pascaline Dupas
08.09.2012 - Pascaline Dupas08.09.2012 - Pascaline Dupas
08.09.2012 - Pascaline DupasAMDSeminarSeries
 

Plus de AMDSeminarSeries (20)

02.16.2014 - Carolina Castilla
02.16.2014 - Carolina Castilla02.16.2014 - Carolina Castilla
02.16.2014 - Carolina Castilla
 
11.07.2013 - Jenny Aker
11.07.2013 - Jenny Aker11.07.2013 - Jenny Aker
11.07.2013 - Jenny Aker
 
09.19.2013 - Laura Schechter
09.19.2013 - Laura Schechter09.19.2013 - Laura Schechter
09.19.2013 - Laura Schechter
 
09.12.2013 - Esther Duflo
09.12.2013 - Esther Duflo09.12.2013 - Esther Duflo
09.12.2013 - Esther Duflo
 
08.15.2013 - Robert Jensen
08.15.2013 - Robert Jensen08.15.2013 - Robert Jensen
08.15.2013 - Robert Jensen
 
07.18.2013 - Michael Clemens
07.18.2013 - Michael Clemens07.18.2013 - Michael Clemens
07.18.2013 - Michael Clemens
 
06.20.2013 - Nishith Prakash
06.20.2013 - Nishith Prakash06.20.2013 - Nishith Prakash
06.20.2013 - Nishith Prakash
 
05.16.2013 - Dilip Mookherjee
05.16.2013 - Dilip Mookherjee05.16.2013 - Dilip Mookherjee
05.16.2013 - Dilip Mookherjee
 
05.02.2013 - Jonathan Robinson
05.02.2013 - Jonathan Robinson05.02.2013 - Jonathan Robinson
05.02.2013 - Jonathan Robinson
 
02.21.2013 - Petra Todd
02.21.2013 - Petra Todd02.21.2013 - Petra Todd
02.21.2013 - Petra Todd
 
02.07.2013 - Mark Rosenzweig
02.07.2013 - Mark Rosenzweig02.07.2013 - Mark Rosenzweig
02.07.2013 - Mark Rosenzweig
 
01.17.2013 - Garance Genicot
01.17.2013 - Garance Genicot01.17.2013 - Garance Genicot
01.17.2013 - Garance Genicot
 
12.10.2012 - Catherine Wolfram
12.10.2012 - Catherine Wolfram12.10.2012 - Catherine Wolfram
12.10.2012 - Catherine Wolfram
 
11.29.2012 - Marta Vicarelli
11.29.2012 - Marta Vicarelli11.29.2012 - Marta Vicarelli
11.29.2012 - Marta Vicarelli
 
11.08.2012 - Lori Beaman
11.08.2012 - Lori Beaman11.08.2012 - Lori Beaman
11.08.2012 - Lori Beaman
 
10.25.2012 - Craig McIntosh
10.25.2012 - Craig McIntosh10.25.2012 - Craig McIntosh
10.25.2012 - Craig McIntosh
 
10.11.2012 - Saumitra Jha
10.11.2012 - Saumitra Jha10.11.2012 - Saumitra Jha
10.11.2012 - Saumitra Jha
 
09.13.2012 - Seema Jayachandran
09.13.2012 - Seema Jayachandran09.13.2012 - Seema Jayachandran
09.13.2012 - Seema Jayachandran
 
08.09.2012 - Pascaline Dupas
08.09.2012 - Pascaline Dupas08.09.2012 - Pascaline Dupas
08.09.2012 - Pascaline Dupas
 
08.02.2012 - Shawn Cole
08.02.2012 - Shawn Cole08.02.2012 - Shawn Cole
08.02.2012 - Shawn Cole
 

08.30.2012 - Brian Dillon

  • 1. Introduction Data Model Identification Estimation Results Estimation of a Dynamic Agricultural Production Model with Observed, Subjective Distributions Brian Dillon Cornell University and Harvard Kennedy School August 30, 2012 Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 2. Introduction Data Model Identification Estimation Results Motivation: crop production To grow crops, farmers solve a dynamic resource allocation problem The problem is not unlike many other dynamic choice problems: portfolio management, inventory management, human capital investment The solution to this problem can involve delay of some choices, distribution of activities across time, and updating of expectations as new information arrives Between-farmer variation in expectations clearly matters (Gin´, e Townsend, Vickery 2008) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 3. Introduction Data Model Identification Estimation Results What if we measure expectations? Early literature in agricultural economics (Bessler and Moore 1979; Eales 1990) Manski (2004) makes the case for measuring expectations Nyarko and Schotter (2002) show that there is a big difference between observed and estimated expectations Delavande et al (2010) review the recent development literature that uses subjective probabilities Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 4. Introduction Data Model Identification Estimation Results What we get from measuring expectations Two contributions to the estimation of dynamic choice models: 1. Allow us to relax the rational expectations assumptions that are standard for these models (Wolpin 1987; Rust 1987, 1997; Fafchamps 1993) 2. There is a lot of information in a subjective distribution over an endogenous outcome Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 5. Introduction Data Model Identification Estimation Results Why go through a structural exercise? Apart from the pure value of estimating a less restricted production function... Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 6. Introduction Data Model Identification Estimation Results Why go through a structural exercise? Apart from the pure value of estimating a less restricted production function... Production elasticities tell us something about resilience of the production process to shocks Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 7. Introduction Data Model Identification Estimation Results Why go through a structural exercise? Apart from the pure value of estimating a less restricted production function... Production elasticities tell us something about resilience of the production process to shocks What we know about shocks already largely deals with • Consumption/asset smoothing (Townsend 1994, Morduch 1995, Hoddinott 2006, Barrett and Carter 2006, Jacoby and Skoufias 1998, Fafchamps et al 1998) • Human capital (Hoddinott and Kinsey 2001, Aguilar and Vicarelli 2012) • Two papers look at how farmers move labor across time, within a season: Fafchamps (1993) and Kochar (1999) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 8. Introduction Data Model Identification Estimation Results Why go through a structural exercise? And we can also simulate important, relevant policies: 1. Insurance 2. Forward contracting 3. Improvements in information delivery 4. Changes in input supply Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 9. Introduction Data Model Identification Estimation Results Contributions of this paper We use a sequence of observed inputs, price expectations, and yield expectations to estimate an agricultural production function Methodological contributions: 1. Develop a general method for estimating dynamic choice models with observed subjective distributions 2. Show how counterfactual choice data (“How much pesticide did you want to apply last week?”) can be used in estimation Substantive contributions: 1. Recover estimates of all elasticities of substitution between inputs (within and across periods) 2. Simulate the impact of insurance, forward contracting, and information provision policies Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 10. Introduction Data Model Identification Estimation Results Plan of the talk • Data set • Model basics • Identification of shock densities • Estimation • Results Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 11. Introduction Data Model Identification Estimation Results Data set • 195 cotton farmers in 15 villages in NW Tanzania • Face-to-face agriculture and LSMS surveys conducted in summer 2009 and summer 2010 • From September 2009 - June 2010: investment, time use, shocks, agricultural input and output, and other data gathered every 3 weeks • High frequency interviews also gathered subjective probability distributions over end-of-season prices and yields, and qualitative distributions over pest pressure and rainfall at various points throughout the year Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 12. Introduction Data Model Identification Estimation Results Measuring subjective distributions +,-./0!1,!2,+3,! ! 4+567879:;<:=>5?! #! $! %! &! "! !! *! )! (! '! !! ! Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 13. Introduction Data Model Identification Estimation Results Evolution of subjective price distributions 5 4 3 Mean number of stones 2 1 0 6 5 4 7 6 3 5 Survey period 4 2 3 1 2 Bin number 1 Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 14. Introduction Data Model Identification Estimation Results Evolution of subjective yield distributions ?34+)(*+),-.$ ?34+)(*+),-.% ! "!# "$ "$# "% "%# "% "$# "$ "!# ! !# ! # $! !# ! # $! &'()*+),-./0,1.*2+3*4.5)'46 &'()*+),-./0,1.*2+3*4.5)'46 7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"#!$> 7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"#@#$ ?34+)(*+),-.A E*0('F+ "> "D "A "> "% "% "$ ! ! !$! !# ! # $! !# ! # $! &'()*+),-./0,1.*2+3*4.5)'46 &'()*+),-./0,1.*2+3*4.5)'46 7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"AB#C 7'0-'4.8.'9*-'2:-)7,(;.<*-6=)6+:.8.!"%A## G Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 15. Introduction Data Model Identification Estimation Results Smoothing of distributions Let • xi be a response vector • d ∈ RN+1 be the interval boundaries • z be the random variable in question • k be the number of counters We fit a four parameter beta CDF, Gi (z | a, b, ρ, κ), by solving: N j 2 m=1 xj (ai , bi , ρi , κi ) = arg inf − G (dj+1 | a, b, ρ, κ) a,b,ρ,κ k j=1 Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 16. Introduction Data Model Identification Estimation Results Sample summary statistics Mean sd Min Max Household size (people) 8.33 3.90 2 23 Dependency ratio* 1.31 0.85 0 5.5 Head age 46.85 14.69 20 100 Head is male (%) 85.0 - - - Years of education (HH head) 4.19 3.46 0 11 Radios 0.83 0.71 0 4 Bicycles 1.19 1.00 0 10 Dairy cattle 1.33 2.84 0 20 Non-dairy cattle 3.87 7.89 0 60 Goats 5.27 8.05 0 50 Sheep 1.67 3.74 0 30 Total acres 9.67 11.03 1 82 Number of plots 2.71 1.17 1 7 Number of crops grown 3.45 1.26 1 8 Labor expenditure (TSH) 78,248 139,485 0 1,020,000 Fertilizer expenditure (TSH) 21,149 81,359 0 715,000 Animal labor expenditure (TSH) 33,497 92,724 0 750,000 Transport expenditure (TSH) 10,333 20,049 0 144,000 Other cultivation expenditure (TSH) 6,929 15,817 0 100,000 Total cultivation expenditure (TSH) 150,156 254,863 0 1,514,700 Notes: author's calculation from survey data; cultivation data refers to 2008-2009 cultivation of all crops; 1 USD ! 1 400 TSH; *Dependency ratio is number of persons aged < 15 or aged > 65 divided !"1,400 by number aged between 15 and 65. Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 17. Introduction Data Model Identification Estimation Results Model assumptions Important: 1. Farmers are dynamically consistent (will relax, if we have time) 2. Independence of shocks across time Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 18. Introduction Data Model Identification Estimation Results Model assumptions Important: 1. Farmers are dynamically consistent (will relax, if we have time) 2. Independence of shocks across time Less fundamental: 1. Separable household model 2. Risk-neutral maximization of expected plot-level profits 3. All forms of labor are interchangeable 4. No binding credit constraints 5. Functional form choices Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 19. Introduction Data Model Identification Estimation Results Crop evolution Expanding on Fafchamps (1993), crops grow according to: yi0 = φi Ai e θi0 yi1 = h1 (yi0 , li1 , pi1 )e θi1 yi2 = h2 (yi1 , li2 , pi2 )e θi2 yi = h3 (yi2 , li3 , pi3 )e θi3 where θit ∼ git (θit ) for t = 0, . . . , 3 Ai is acreage φi is a plot-specific yield shifter li and pi are labor and pesticides Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 20. Introduction Data Model Identification Estimation Results Crop evolution cont. We use nested CES functions: 1 γ γ γ h1 (y0 , l1 , p1 | α1 , α2 , γ) = [α1 y0 + α2 l1 + (1 − α1 − α2 )p1 ] γ 1 δ δ δ δ h2 (y1 , l2 , p2 | β1 , β2 , δ) = β1 y1 + β2 l2 + (1 − β1 − β2 )p2 1 ω ω ω h3 (y2 , l3 , p3 | κ1 , κ2 , ω) = [κ1 y2 + κ2 l3 + (1 − κ1 − κ2 )p3 ] ω Which gives us 9 production parameters to estimate: • Share parameters (α1 , α2 , β1 , β2 , κ1 , κ2 ) • Transformed elasticity parameters (γ, δ, ω) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 21. Introduction Data Model Identification Estimation Results Farmer’s objective function From the viewpoint of the first period: δ γ γ γ max E [qc ]Eθi1 θi2 θi3 κ1 β1 α1 yi0 +α2 li1 +(1−α1 −α2 )pi1 γ e δθi1 li1 ,pi1 ω δ ∗δ ∗δ ∗ω + β2 li2 + (1 − β1 − β2 )pi2 e ωθi2 + κ2 li3 1 3 ω ∗ω + (1 − κ1 − κ2 )pi3 e θi3 − (ql lit + qp pit ) t=1 Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 22. Introduction Data Model Identification Estimation Results Identification of gt (θt ) We need measures of gt (θt ) in order to proceed Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 23. Introduction Data Model Identification Estimation Results Identification of gt (θt ) We need measures of gt (θt ) in order to proceed Nested fixed point method (Rust 1987): iterate between guesses of production parameters and gt parameters until convergence Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 24. Introduction Data Model Identification Estimation Results Identification of gt (θt ) We need measures of gt (θt ) in order to proceed Nested fixed point method (Rust 1987): iterate between guesses of production parameters and gt parameters until convergence But we only observe subjective output distributions Ψ0 (y ), . . . , Ψ3 (y ) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 25. Introduction Data Model Identification Estimation Results Identification of gt (θt ) We need measures of gt (θt ) in order to proceed Nested fixed point method (Rust 1987): iterate between guesses of production parameters and gt parameters until convergence But we only observe subjective output distributions Ψ0 (y ), . . . , Ψ3 (y ) We can use those to directly estimate gt (θt ), within the context of the model Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 26. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection y reported Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 27. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection !3 realized y reported Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 28. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection (l3 , p3) chosen !3 realized y reported "3(y) reported incl: g3(!3) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 29. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection (l3 , p3) chosen !2 realized !3 realized y reported "3(y) reported incl: g3(!3) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 30. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection (l2 , p2) chosen (l3 , p3) chosen !2 realized !3 realized y reported "2(y) reported incl: g2(!2) "3(y) reported g3(!3) incl: (l3* , p3*) g3(!3) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 31. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection (l2 , p2) chosen (l3 , p3) chosen !1 realized !2 realized !3 realized y reported "2(y) reported incl: g2(!2) "3(y) reported g3(!3) incl: (l3* , p3*) g3(!3) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 32. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection (l1 , p1) chosen (l2 , p2) chosen (l3 , p3) chosen !1 realized !2 realized !3 realized y reported "1(y) reported incl: g1(!1) "2(y) reported g2(!2) incl: g3(!3) g2(!2) "3(y) reported (l2* , p2*) g3(!3) incl: (l3* , p3*) (l3* , p3*) g3(!3) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 33. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection (l1 , p1) chosen (l2 , p2) chosen (l3 , p3) chosen !0 realized !1 realized !2 realized !3 realized y reported "1(y) reported incl: g1(!1) "2(y) reported g2(!2) incl: g3(!3) g2(!2) "3(y) reported (l2* , p2*) g3(!3) incl: (l3* , p3*) (l3* , p3*) g3(!3) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 34. Introduction Data Model Identification Estimation Results Timeline of decisions, realizations, and data collection (l1 , p1) chosen (l2 , p2) chosen (l3 , p3) chosen !0 realized !1 realized !2 realized !3 realized "0(y) reported incl: y reported g0(!0) "1(y) reported g1(!1) incl: g2(!2) g1(!1) "2(y) reported g3(!3) g2(!2) incl: (l1* , p1*) g3(!3) g2(!2) "3(y) reported (l2* , p2*) (l2* , p2*) g3(!3) incl: (l3* , p3*) (l3* , p3*) (l3* , p3*) g3(!3) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 35. Introduction Data Model Identification Estimation Results Identification of g3 (θ3 ) Taking the normalization E [e θt ] = 1 for all t: Pr[y < Y ] = Pr E [y |Ω3 ]e θ3 < Y Y = Pr θ3 ≤ ln E [y |Ω3 ] where Ω3 is the period 3 information set ⇒ g3 (θ3 ) is constructed by transforming ψ3 (y ) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 36. Introduction Data Model Identification Estimation Results Key proposition (summarized) Proposition If h = H(θ1 , θ2 ) is a function of two random variables, and 1. We know densities fh (h) and fθ2 (θ2 ) 2. H is monotonic in θ1 then we can consistently estimate fθ1 (θ1 ) by taking repeated draws from fh and fθ2 Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 37. Introduction Data Model Identification Estimation Results Identification of g2 (θ2 ) ∗ ∗ Plugging (l3 , p3 ) into the definition of output allows us to write output from the period 2 perspective as: y= H2 φ, α1 , α2 , β1 , β2 , κ1 , κ2 , γ, δ, ω; A, l1 , p1 , l2 , p2 ; ql , qp , E [qc ]; θ0 , θ1 e θ2 e θ3 ∞ And E [y |Ω2 ] = −∞ y ψ2 (y )dy = H2 (·) Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 38. Introduction Data Model Identification Estimation Results Identification of g2 (θ2 ) cont. This gives a method for numerically estimating g1 (θ1 ) using repeated draws from ψ1 (y ) and g2 (θ2 ) M 1 ym Pr[θ2 < Θ2 ] = I ln − θ3m ≤ Θ1 M E [y |Ω2 ] m=1 Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 39. Introduction Data Model Identification Estimation Results Estimation of θt and φ Given any guess of the parameters, we find the realized values of the shocks: • θ0 , θ1 , θ2 come from FOC of the farmer’s decision problem • θ3 comes from realized output y and ψ3 (θ3 ) Lastly ∞ ˆ −∞ y ψ0 (y )dy φ= A Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 40. Introduction Data Model Identification Estimation Results Likelihood function Then the joint likelihood for the observed inputs, output and distributions is: L(α1 , α2 , β1 , β2 , κ1 , κ2 , γ, δ, ω | P i i i i φ, A, l, p, y , ql , qp , E [qc ], θ0 , θ1 = gi0 (θ0 )gi1 (θ1 )gi2 (θ2 )gi3 (θ3 ) i=1 We maximize the log likelihood over the 9 production parameters and: α1 , α2 , β1 , β2 , κ1 , κ2 , γ, δ, ω Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 41. Introduction Data Model Identification Estimation Results Results: shock densities Summary statistics for gt (θt ) Variable Mean s.d. !0 lower bound -2.95 1.86 !0 upper bound 2.43 1.70 E[!0] -0.14 0.61 !1 lower bound -2.49 1.84 !1 upper bound 2.01 1.48 E[!01] -0.01 0.58 !2 lower bound -4.19 1.4807* !2 upper bound 3.19 1.4807* E[!2] -2.35 1.17 N 212 212 *SD of !2 upper and lower bounds is constant by construction, because both reflect variation in acreage Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 42. Introduction Data Model Identification Estimation Results Conclusion Separation of output equation into its dynamic and stochastic components is not a necessary condition for this to work But monotonicity in θt is necessary Observation of shock densities reduces number of parameters to be estimated But it also increases the pressure on the functional form, because the error variance does not adjust to increase the contribution of very low probability parameter contributions to the likelihood Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 43. Introduction Data Model Identification Estimation Results Where things stand... Ongoing work on this paper involves: 1. Embedding the farmer’s problem in a utility framework 2. Comparing results with those from the nested fixed point method 3. Interpretation and simulations 4. Relaxing the dynamic consistency assumption? → could use data on counterfactual, optimal pesticide application Brian Dillon Estimation of a Dynamic Agricultural Production Model with O
  • 44. Introduction Data Model Identification Estimation Results Thanks! Brian Dillon Estimation of a Dynamic Agricultural Production Model with O