Complexity approach to the dynamics of financial systems

In order to understand complex systems, quite often a very diverse set of tools is required. In this thesis a combined approach of an agent-based modelling and linguistic textual analysis has been taken in order to try and gain a better insight into the connection between crises that happen in our economy and the words that people use as an indication of their economic behaviour The model used for this work has a crucial evolutionary adaptation mechanism inbuilt which enables the characteristics of banks and investors to change with time. It was chosen because it reproduces the endogenous crashes in the banking sector which are purely driven by the changes that occur among banks and investors which are adapting and readjusting their behaviour as a function of comparing their performance and strategy relative to others. Firstly, using this modelling framework, we investigate the effect of the distinct levels of interest rates on the stability of the financial network. We demonstrate that under sustained high interest rates banking failures are likely to emerge early, whereas under sustained low rates they happen significantly later and with higher probability. Under low rates lack of liquidity becomes the primary reason for failures and high rates lead predominantly to bankruptcies due to credit losses. We show that by applying interest rate shocks it is possible to change the timing of crises, but it is impossible to prevent them completely. This follows from the fact that banking strategies are continuously converging, making the system less diverse and thus less resilient. Building on this observation of the convergence of bank business strategies, we used a bag-of-words method to analyse annual reports. We show that companies within the same industrial sector are becoming progressively similar, especially since the 2008 crisis crisis, which reinforces the hypothesis that with time, business strategies are converging. Finally we show that it is possible to find a direct connection between word counts and stock prices. We thus create a basic filtering procedure to identify the minimal set of words that could potentially be used to characterise different sectors.Open Acces