Three essays on the use of neural networks for financial prediction

Resumen

The number of studies trying to explain the causes and consequences of the economic and financial crises usually rises considerably after a banking crisis occurs. The dramatic effects of the most recent financial crisis on the real economy around the world call for a better comprehension of previous crises as a way to anticipate future crisis episodes. It is precisely this objective, preventing future crises, the main motivation of this PhD dissertation. We identify two important mechanisms that have failed during the latest years and that are closely related to the onset of the financial crisis: The assessment of the solvency of banks along with the systemic risk over the time, and the detection of the macroeconomic imbalances in some countries, especially in Europe, which made the financial crisis evolve through a sovereign crisis. Our dissertation is made up of three different essays, trying to go a step ahead in the knowledge of these mechanisms. In the first essay, we develop a model of neural networks to study the bankruptcy of U.S. banks, considering the specific features of the current financial crisis. We first reach a more in-depth understanding of the causes of the crisis and how the financial statements of banks deteriorated over the time and resulted in the burst of the financial crisis. We combine multilayer perceptrons and self-organizing maps, two different techniques of neural networks, to provide a tool that displays the probability of distress up to three years before bankruptcy occurs. Coherently with the motivation of this dissertation, we posit that the failure of the detection of the macroeconomic imbalances in some countries is another factor related to the onset of the financial crisis. Accordingly, in our second essay we use self-organizing maps and a set of the usual macroeconomic variables to compare the financial situation of the European countries in 2009. Our results show the existence of several groups of countries, each one of them with specific characteristics. We also find that Government expenditure and the saving rate are the most influential variables on the macroeconomic financial imbalances. We also study the influence of the macroeconomic situation of each Spanish AA.CC. and German state on the national situation. We find that the macroeconomic situation of the regional entities is a key determinant of the country financial (im)balance. The two first essays are closely related to the financial crisis and the possible impact on the real economy. Nevertheless, we also are concerned about other factor that could affect the recovery of the economy after the crisis. Corruption is clearly one of these factors. Thus, in the third paper we develop a neural networks model to predict public corruption based on economic and political factors. We apply this model to the Spanish provinces in which corrupt cases have been uncovered by the media or have gone to trial. The output of our model is a set of SOMs, which allows us to predict corruption in different time scenarios before corruption cases are detected. Our model provides two main insights. First, we identify some underlying economic and political factors that can result in public corruption. Taxation of real estate, economic growth, and an increase in real estate prices, in the number of deposit institutions, and the same party remaining in office for a long time seem to induce public corruption. Second, our model provides different time frameworks to predict corruption. In some regions, we are able to detect latent corruption long before it emerges (up to three years), and in other regions our model provides short-term alerts, and suggests the need to take urgent preventive or corrective measures. Our model can be helpful to improve the efficiency of the measures aimed at fighting corruption. Resources available to combat corruption are limited, and authorities can use the early corruption warning system, which categorizes each province according to its corruption profile, in order to narrow their focus and better implement preventive and corrective policies. Taken together, our research can provide interesting insights to a wide number of potentially interested agents. First, our procedure can be a useful tool for bank supervisors and other stakeholders to delineate the risk profile of each bank. As far as the supervisory authority is concerned, the preventive measures to correct imbalances can be different in the short, medium, or long term depending on the probability of banking failure—that is, on the group to which the bank belongs. Investors, depositors, and other participants in capital markets can assess the risk profile of their investment and, consequently, define their optimal risk-return combination. Our model outperforms most of the previous ones in terms of predicting ability, which was compared against a wider set of alternative methods than previous papers do. Moreover, our model is simpler and, at the same time, provides a clearer visualization of the complex temporal behaviors. Our research also provides interesting insights for policymakers. We provide a complementary method to analyze the international macroeconomic financial situation. Yet, the identification of groups of similar countries can allow discovering possible channels of financial contagion and financial turbulences propagation across countries. Thus, our model could serve as an early warning system for some countries when the counterpart countries get into financial troubles. In the same vein, the enlargement of the EU can be eased through diagnosing and forecasting the future financial situation of the candidate countries to avoid any destabilization effect. In addition, the identification of the regional disparities within European countries can lead to focus on the countries and regions in the most need of receiving European financial help. From a microeconomic point of view, our research is also useful for banks and other institutional investors since the international risk map can be a relevant input to assess the risk exposure of each institution. This tool would be complementary to the stress tests and other analyses of sovereign risk carried out by national and international financial supervisors in recent months. The study of new methodologies based on neural networks is a fertile field to be applied to a number of legal and economic issues. In this dissertation, we use these methodologies to explain the origins of the financial crisis and to avoid or limit the effects of future crisis. This dissertation is aimed to be the starting point of future research in this field and to keep on contributing to the literature. As most of the research, this dissertation is not exempt of some weaknesses or limitations. Neural networks have usually been seen like black boxes, being difficult to explain how the predictions are issued. Moreover, the results can suffer from difficulties to be generalized because of model overfitting, which results in needing a lot of time to train the models and to obtain the most adequate configuration. However strongly the explanation ability of the models is tested, the comparability of the models remains as a critical issue and the extension to other samples or environments can be troublesome. Overfitting risk always remains in every study. There are other more specific limitations in this dissertation. When trying to predict the bankruptcy of banks, we do not control for all the macroeconomic factors potentially affecting the banks propensity to fail. Furthermore, since our study focuses on commercial banks, some concerns may arise about whether our results can be applied to large investment banks. In addition, when trying to understand and predict the public corruption, we base on local or country specific variables, in this case Spanish macroeconomic variables Although the variables we used are commonly available for each country, especially in Europe, we have not applied the model to other countries, so the validations of the model in other institutional frameworks remains a concern. Far from discouraging us, the above mentioned limitations motivate us to go on with our research. In this sense, the next step will be to replicate the results of the European banks stress test using only the financial statements of banks. The underlying intuition is that most of the European stress test results can be predicted, which could cast some doubts on this supervisory tool. Our aim is to suggest different ways to improve the existing early warning systems in line with the content of this dissertation. Another avenue for future research is the role of the external rating agencies in the financial crisis. These agencies play an outstanding role and have been under severe criticisms. During some crisis episodes such as the Asian or Russian financial crises in the late 1990s and in the recent global crisis, the rating agencies have failed to predict the financial turbulences. Concerns on the quality of ratings have called for new research, mainly based on qualitative rather than quantitative information. In turn, we plan to show the importance of the qualitative information by replicating the credit ratings using the country reports issued by the European Commission for the European Member States. It will be another tool to improve the international financial stability.