Comparing parametric and nonparametric regression methods for panel data: the optimal size of Polish crop farms

Tomasz Gerard Czekaj, Arne Henningsen

    Abstract

    We investigate and compare the suitability of parametric and non-parametric stochastic regression methods for analysing production technologies and the optimal firm size. Our theoretical analysis shows that the most commonly used functional forms in empirical production analysis, Cobb-Douglas and Translog, are unsuitable for analysing the optimal firm size. We show that the Translog functional form implies an implausible linear relationship between the (logarithmic) firm size and the elasticity of scale, where the slope is artificially related to the substitutability between the inputs. The practical applicability of the parametric and non-parametric regression methods is scrutinised and compared by an empirical example: we analyse the production technology and investigate the optimal size of Polish crop farms based on a firm-level balanced panel data set. A nonparametric specification test rejects both the Cobb-Douglas and the Translog functional form, while a recently developed nonparametric kernel regression method with a fully nonparametric panel data specification delivers plausible results. On average, the nonparametric regression results are similar to results that are obtained from the parametric estimates, although many individual results differ considerably. Moreover, the results from the parametric estimations even lead to incorrect conclusions regarding the technology and the optimal firm size.
    Original languageEnglish
    PublisherInstitute of Food and Resource Economics, University of Copenhagen
    Pages1-30
    Number of pages30
    Publication statusPublished - 2012
    SeriesFOI Working Paper
    Number2012/12

    Fingerprint

    Dive into the research topics of 'Comparing parametric and nonparametric regression methods for panel data: the optimal size of Polish crop farms'. Together they form a unique fingerprint.

    Cite this