Risk analysis of stocks markets by a merged unsupervised model, time evolution comparison, and optimization.

The objective of this comparative study was to determine which algorithm of the self-organizing maps (SOM) type is best suited when grouping companies from the SPLatinAmerica40 financial index. This work sought a point of convergence between artificial intelligence (AI) and economy because AI has only been applied in finance in recent years, despite the AI exponential use increase. For any nation, stock markets represent a potential factor for its economic growth as they are financial engines, which generate income from the money produced by the industrial strength of the countries. First, the investigation was focused on establishing the best SOM architecture for the treatment of stock markets through the literary review. Afterward, the historical data of the selected financial index from 2014 to 2019 were extracted, using the Yahoo Finance platform. Then, pre-processing of data was carried out using a cohesion algorithm. The name of the SOM method proposed in this work is ISOMSP40, and uses a suitable combination of hexagonal architecture and neighborhood function based on Manhattan distance. Two other similar methods were tested under the same conditions to compare their metrics. These measures determined the best algorithm for the SP Latin America 40 data set. The study used as reference the nine companies with the highest profits in the S&PLATAM 40 stock index. There were mainly analyzed the metrics of density by profit, industrial area, and geographic correlation detected with a success rate of 80%. The correct clustering was also verified in a time-frequency analysis developed with the top six companies during the same data period. The execution time in the proposed ISOMSP40 algorithm also improved by two decimal places. The minimum execution time was 5, 79E − 01(s) against the 9, 01E + 00 average in the other two models. Thus, it was established that the proposed ISOMSP40 algorithm showed a better performance for the S&P LATAM 40 stock index over two other existing methods. The comparative experiments demonstrated by the metrics an efficient adaptation for the chosen index achieving the main objective of this study.

• Hovedforfatter: Pilliza Chicaiza, Gissela Estefania (author)
• Format: bachelorThesis
• Sprog: eng
• Udgivet: 2020
• Fag: Self Organized Maps; Stock Market; Stock Index; S&P Latin America 40; IBEX35; NYSE; NASDAQ; Investment risk; Mapas Auto Organizados; Bolsas de valores; Índices financieros; Riesgo; Inversión
• Online adgang: http://repositorio.yachaytech.edu.ec/handle/123456789/273