ce_12_01_numdiffint

Upload: tomcat14d

Post on 02-Jun-2018

223 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/10/2019 CE_12_01_NumDiffInt

    1/10

    Numerical Integration and DifferentiationComputational Economics

    Dietmar Maringer

    WWZ, University of Basel

    Spring 2012

    D Maringer,Computational Economics Numerical Integration and Differentiation (1)

  • 8/10/2019 CE_12_01_NumDiffInt

    2/10

    Finite Differences for Numerical Differentiation

    basic concepts

    difference quotient: yx=

    f(x0+x)f(x0)x

    derivative: dydxf(x) limx0 yx

    numerical approximation offwith centraldifference

    f(x)

    f(x+)f(x)(x+) (x)

    = f(x+)f(x)2

    where 0

    xx

    x+

    x

    f(x)x+

    f(x+)

    multi-dimensional functions

    gradient: list of all first order derivatives

    f= fx1

    fxn

    f(x1+,x2,...,xn)f(x1,x2,...,xn)2

    f(x1,x2,...,xn+)f(x1,x2,...,xn)2

    D Maringer,Computational Economics Numerical Integration and Differentiation (2)

  • 8/10/2019 CE_12_01_NumDiffInt

    3/10

    Finite Differences for Numerical Differentiation

    second derivative for one-dimensional functions

    since second derivative = derivative of first derivative, compute by

    iteratively applying definitions:

    f(x) f(x+)f(x )(x+) (x)

    =

    f(x++)f(x+)2

    f(x+)f(x)2

    (x+) (x)= f(x+2)+f(x2)2f(x)

    (2)2

    = f(x+)+f(x)2f(x)2

    where = 2likewise for higher derivatives

    D Maringer,Computational Economics Numerical Integration and Differentiation (3)

  • 8/10/2019 CE_12_01_NumDiffInt

    4/10

    Finite Differences for Numerical Differentiation

    second derivative for multi-dimensional functions

    Hessian matrix multi-dimensional functions: matrix of all second

    order derivatives

    H=

    2f

    x21

    2fx1xn

    ...

    2f

    xixj

    ...

    2f

    xnx1

    2f

    x2n

    usually, Hessian is symmetric, i.e.,

    2fxixj

    = 2f

    xjxi

    D Maringer,Computational Economics Numerical Integration and Differentiation (4)

  • 8/10/2019 CE_12_01_NumDiffInt

    5/10

    Finite Differences for Numerical Differentiation

    numerical approximation of the Hessian

    vary values in dimensionsiandj(leaving all other values unchanged)

    and evaluate points

    xj xj xj+xi f f f+xi f

    f f+xi+ f+ f+ f++

    diagonal elements of the Hessian matrix

    2f

    x2i

    f++f2f

    2 alternative 1

    12

    f++f2f

    2 + f

    +++f+2f+2

    alternative 2

    off-diagonal elements of the Hessian matrix (cross-derivatives)

    2f

    xixj

    12

    f++fff+

    2 + f

    ++ff+f2

    alternative 1

    12

    f+f++f+f+

    2 + f

    +fff2

    alternative 2

    14

    f+++ff+f+2 average alt.s 1 and 2

    D Maringer,Computational Economics Numerical Integration and Differentiation (5)

  • 8/10/2019 CE_12_01_NumDiffInt

    6/10

    Application Example

    local approximation with Taylor Series Approximation

    approximation for region around a given valuex0

    approximate function value atx= x0+with a polynomial

    f(x)=f(x0+)=f(x0)+f(x0)+

    2

    2 f(x0)+

    3

    6 f(x0)+

    + +

    n

    n!f(n)

    (x0)+Rn

    remainderRn < |xx0|n+1

    D Maringer,Computational Economics Numerical Integration and Differentiation (6)

    i l i i h id l

  • 8/10/2019 CE_12_01_NumDiffInt

    7/10

    Numerical Integration with Trapezoid Rule

    basic idea

    assume function is approximately piecewise linear

    area underneath each line segment is a trapezoidintegral is the sum of all trapezoids areas

    =x1

    =x4x2 x3

    f2

    f3f2+f3

    2

    wa b

    w f2+f32

    D Maringer,Computational Economics Numerical Integration and Differentiation (7)

    N i l I i i h T id R l

  • 8/10/2019 CE_12_01_NumDiffInt

    8/10

    Numerical Integration with Trapezoid Rule

    how solve a problem likeb

    af(x)dx

    split range intonsegments of equal widthw= (ba)/ncompute borders of segments,xi = a+ (i1)wfori= 1..(n+1)approximation for area of one segment:

    xi+1xi

    f(x)dxwf(xi)+f(xi+1)2for the entire range (withfi =f(xi))

    b

    af(x)dx= n

    i=1

    xi+1

    xi

    f(x)dx

    n

    i=1w fi+fi+1

    2 =w

    f12+f2+ +fn+

    fn+12

    more generally (when segments have different width):ba

    f(x)dxn

    i=1(xi+1xi)

    fi+fi+12

    extension: Simpsons rule (piecewise quadratic functions) D Maringer,Computational Economics Numerical Integration and Differentiation (8)

    N i l I t ti ith M t C l Si l ti

  • 8/10/2019 CE_12_01_NumDiffInt

    9/10

    Numerical Integration with Monte Carlo Simulation

    some (very) simple examples

    f(x)= 12

    x=

    2

    0f(x)dx=? f(x)= sin

    |x| 5

    |x|=

    3

    1f(x)dx=?

    Estimating the Value ofvia MCS

    A = (2r)2

    A = r2AA

    = r2

    4r2=

    4

    == 4AA

    y

    x

    r

    1 1

    1

    1

    D Maringer,Computational Economics Numerical Integration and Differentiation (9)

    dditi l lit t

  • 8/10/2019 CE_12_01_NumDiffInt

    10/10

    additional literature

    Gilli, M., Maringer, D., and Schumann, E. (2011).

    Numerical Methods and Optimization in Finance.

    Academic Press.Judd, K. L. (1998).

    Numerical Methods in Economics.

    MIT Press.

    Miranda, M. J. and Fackler, P. L. (2002).

    Applied Computational Economics and Finance.MIT Press.

    D Maringer,Computational Economics Numerical Integration and Differentiation (10)