non-linear models.docVIP

Linearization of non-linear regression models Consider the following model: yt = ( 0 + (1+ ut yt = ( 0 + ut Non-linear model. Can not use OLS method to estimate the parameters The model can be transformed into linear model Several typical non-linear models which can be linearized: Exponential function model yt = figure: Taking natural logarithm for both sides of the exponential function, we obtain: Lnyt = Lna + b xt + ut Let Lnyt = yt*, Lna = a*, then yt* = a* + bxt + ut logarithmic function model yt = a + b Ln xt + ut figure let xt* = Lnxt, then yt = a + b xt* + ut (3) power function model: yt = a xt b figure: Taking the natural logarithm for both sides of power function, we obtain: Lnyt = Lna + b Lnxt + ut let yt* = Lnyt, a* = Lna, xt* = Lnxt, then above equation is expressed as: yt* = a* + b xt* + ut (4) hyperbolic function model: 1/yt = a + b/xt + ut or yt = 1/ (a + b/xt + ut) figure: yt = 1/ (a + b/xt ), (b 0) yt = a + b/xt , (b 0) let yt* = 1/yt, xt* = 1/xt , then yt* = a + b xt* + ut (5) Polynomial function model: yt = b0 +b1 xt + b2 xt2 + b3 xt3 + u figure let xt 1 = xt，xt 2 = xt2，xt 3 = xt3, above equation becomes yt = b0 +b1 xt 1 + b2 xt 2 + b3 xt 3 + ut Figure is like the total cost curve in microeconomics There is another expression of the polynomial function yt = b0 + b1 xt + b2 xt2 + ut Let xt 1 = xt，x t 2 = xt 2, then above model is linearized as: yt = b0 + b1 xt1 + b2 xt2 + ut figure: Above figures are like the marginal cost curve or the average cost curve (6) Logistic model: yt = - f(t) = a0 + a1 t + a2 t 2 + … + an t n. Usually, f(t) = a0 - a t so yt = = Figure: K and 0 are the growth upper band and lower band respectively. a and b are the estimated parameters The model is used to describe the growth process of the organism When k is given, we make the following transformation: k/yt = 1 + take the natural logarithm for above equation