Nuevas Metodologías para la evaluación del impacto en las empresas del sector TIC de las Políticas de Innovación Tecnológica

  1. José María Insenser
Supervised by:
  1. José Molero Zayas Director

Defence university: Universidad Complutense de Madrid

Fecha de defensa: 03 March 2022

  1. Rafael Myro Sánchez Chair
  2. Ana Fernández Zubieta Secretary
  3. Gonzalo León Serrano Committee member
  4. Santiago Manuel López García Committee member
  5. Juan Carlos Salazar Elena Committee member

Type: Thesis


The objective of this Thesis is to demonstrate the usefulness and effectiveness of new methodologies based on Machine Learning, for evaluating the impact of Innovation Policy instruments in firms. The interest of this research stems from the verification of the limitations of some of the current impact evaluation methodologies, including patent analysis and surveys. Among the most obvious, we can mention the updating time of the surveys, the cost to implement the surveys, the non-inclusion of data such as the management and accumulation of tacit and explicit knowledge, and, especially, the possibility of continuous monitoring. With the objective to develop these impact evaluation methodologies, the Thesis follows a development in which a review of the literature and the theoretical bases of business innovation is done. In chapter 1, the background and current state of the subject are exposed, the objectives that we intend to achieve, as well as the structure of the work, the methodology and the sources. Chapter 2 is dedicated to reviewing the literature that lays the theoretical foundations and empirical evidence of technical change and innovation as the engine of economic growth and competitiveness. The seminal works of Sollow, Rommer and Lucas, among others, are explored with the aim to expose the theoretical foundation of the importance of innovation for the productive fabric. Schumpeterian "creative destruction" constitutes a clear paradigm of the capitalist economic system. These theoretical bases are the foundation of the common interest of companies and public administrations in supporting innovation as a key to the prosperity of States and regions. From the need to stimulate companies to invest in R &D & I, since due to the issue of the appropriability of knowledge and asymmetric information, the risk was not negligible, many companies did not have a sufficient stimulus to invest in R & D & I, so that the public administrations of many states decided to make up this investment financing deficit through specific incentives, such as subsidies, loans under special conditions, etc. From there several public policies arise such as science and technology policy, Industrial Policy and Innovation Policy. The different types of Innovation Policies are mentioned and developed, to verify that the impact evaluation methodologies will also need different indicators of the companies to evaluate specific aspects, however, if it is intended to evaluate the degree of innovation, in principle it is valid for any Innovation Policy. Chapter 3 tries to justify the need to evaluate the impact of innovation policies in companies to try to adapt them to the desired objectives, so that policymakers can monitor their impact with the aim to modify them in order to try to produce a greater impact on the degree of innovation of companies. The current methods most used in the impact evaluation of Innovation Policies are reviewed, as well as the current limitations of these methods. Chapter 4 addresses the problem of business R & D & I Management, since the synthetic indicators, that will be used in the new impact assessment methodologies, it is essential that they are based on those aspects of R&D management that are more relevant for firms and that are more aligned with their strategies. As one of the most important barriers that companies face when approaching R&D projects and technological innovation is usually the associated uncertainty (Technological, market, competition, etc.), concepts to face this problem are briefly exposed, such as the use of real options, as well as the optimization in the use of resources such as adequate planning of R & D &I, support for R & D & I, dynamic and complementary capabilities, etc. The first four chapters focus on the importance of innovation and its policies for companies and public administrations, and on the foundations on which the parameters to be used in the impact assessment are based. It can be seen that the first part of the Thesis focuses on the innovation and management capacity of the R & D & I of the company, especially the technology-based company, where the author of the Thesis has carried out his entire professional career. It is normal that the common thread is the company and the management of innovation in it, since the central agent of innovation and economic growth is the company. Once the problem of the need for the most efficient evaluation lies in sharing and using the same information that companies use in their R & D & I management has been addressed, the Thesis focuses, in Chapter 5, on analyzing the particularities of the ICT sector, for several reasons: The analysis of indicators must have a certain homogeneity (for example, use of intellectual property protection and average project execution time), the ICT sector has a very important penetration in many other sectors and is a lever for innovation in many of them, has a very high intensity of innovation and the author knows better than other sectors for having exercised his business experience in it. In order to know the trends in innovation and its management in the sector, a survey was designed and carried out (see Annex A) so that microdata that we consider essential could be collected, especially the assessment of the different innovation management indicators framed on the Balanced Scorecard.This information is collected in the body of the Thesis and the rest of the microdata, collected in the survey, can be consulted in Annex B. The rest of the microdata has been collected to be able to compare the R&D&I management microdata in subsequent studies with other similar data from other surveys, since collecting data is expensive and once the decision to do so is made, it is useful to collect as much useful information as possible. Chapter 6 presents the theoretical and practical bases for the construction of the composite or synthetic index, FIGI (Index of the degree of business innovation). It is a weighted index based on the opinions collected in the ICT sector survey. The weighting is carried out from the BW decision-making method. Since the ICT sector survey is not statistically significant, since the sample obtained is not a sample representation of the population, the ESEE survey has been used, only for methodological purposes, which is a significant sample of the industrial business population from Spain. An attempt has been made to approximate FIGI to the data available (variables) in the ESEE. In Chapter 7, Machine Learning methodologies are exposed and the design cycle of an ML system is discussed. An essential point is that of the engineering of characteristics, since if the predictors (characteristics) are not correctly chosen, problems of overfitting and lack of precision in the prediction can occur. On the other hand, it is important that FIGI, synthetic indicator, can be predicted for other companies knowing their associated characteristics, which should be indicators shared by the companies and the Public Administrations and, if possible, easily obtainable from public databases such as ORBIS (for example, R&D expenses and personnel dedicated to R&D), as well as the new innovative products that they bring to the market identified through web scraping and classified as innovative or not enough innovative by deep supervised learning algorithms (neural networks). Each type of classifier is studied using the data of each company, FIGI as the "label" and some R &D &I indicators as characteristics. In Chapter 8 the impact evaluation of the degree of business innovation (FIGI) is carried out. First, comparisons are made between the different Machine Learning Algorithms used to predict the FIGI of companies in which it is unknown and in those that are known to see its evolution over time. An assessment of the characteristics is carried out, that is, to analyze which ones have the greatest relevance in the prediction. Predictive models for impact evaluation are also used by econometric analysis of the FIGIs in the treatment group, that is, those that have participated in the program that is the object of the Innovation Policies instrument and those that appear in the control group (have not participated in the program). Finally, a Hypothesis contrast is made to reinforce the influence of certain R & D & I indicators, taken from the ESEE. Said hypotheses are: H1: Companies that participate in open innovation and receive public funding to enhance their R &D &I, improve their productivity. H2: Companies that receive public funding to enhance their R & D &I increase their relational capital. H3: Companies that increase their technological effort, participate in open innovation and dedicate resources to training their employees, enhance their R & D & I, improve their human capital. In Chapter 9 the results of the investigation, the conclusions and the future work are presented. The results of this Thesis show that the methodologies for evaluating the impact of the Innovation Policy instruments, developed in this work, especially those referring to the evolution over time of the degree of innovation (FIGI) of firms, more concretely those of the ICT sector, based on a panel of indicators (predictors) used by firms in their daily R & D & I management, allow an analysis of the situation and decision-making with the proposed system ISCOM (Innovation Scoreboard COntinous Monitoring). This decision-making allows suspending, stopping, delaying or accelerating a project that is linked to a specific instrument of Innovation Policies. Likewise, an important conclusion is that feature engineering is essential in Machine Learning, since these are key for obtaining better precision in classification and, consequently, for prediction.