טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentShmukler Zvagelsky Inessa
SubjectParameter Estimation of R&D Processes with Uncertain
Implementation
DepartmentDepartment of Industrial Engineering and Management
Supervisors Dr. Benjamin Bental
Professor Dan Peled
Full Thesis text - in Hebrew Full thesis text - Hebrew Version


Abstract

The goal of the study is to improve methods used to estimate the behavior of firms involved in R&D processes, under the assumption that the R&D can be described as a sequential search process over untried technologies.  In such a process, the decision as to whether to continue the R&D is sequential.  “Stopping rules” which are contingent upon the results of R&D to date and on the level of the firm's capital and technology in use are computed.  The stopping  rules have to satisfy several conditions which create problems when the model is empirically estimated. This study attempts to contend with some of these problems.

According to the model, a firm tries to maximize profits by selecting the optimal search strategy of new technologies. The search is conducted under a resource constraint. The available resources may be used either for continued R&D or for production, using the best available technology. The strategy consists of a rule that determines when a firm will stop the serial sampling process, given the remainder of the resources in its possession, the technology it possesses at that moment, and the initial technology that is always accessible.

Empirically, technologies are represented by the firms' total factor productivities. These are assumed to be generated by the R&D process described in the model withing a period of one year. Previous studies encountered the fact that in the data there exist technologies at the end of the search period that are 90% lower than the technologies at the beginning of the period. To cope with this problem, here we develop a model that assumes any technology that is to be implemented is affected by a multiplicative random error, drawn out of a known distribution.

The improved model is used on both simulatated data and on firm level data, stemming from several industries. During the course of the estimating the simulated data, the estimation algorithm earlier suggested by Bental and Peled has not converged properly. As a result, additional estimation methods were developed that seemed to better cope with the data. Using these methods, the estimation with real data is carried out under different assumptions concerning the underlying distributions that generate the technologies and the implementations shocks. The most suitable distributions are selected by comparing simulated data using the empirical distribution to the actual data.