| Abstract [eng] |
Stocks is an often seen subject in academic research trying to investigate relationships with other indicators and so have better odds at beating the market without growing the risk. Despite its undeniable influence on global economy, bond market linked variables are not so often met among those indicators. When bond market is involved, in such work papers as D’Arcy, Poole (2010) or Kang (2007), the whole complexity and diversity of bond market may not necessarily be showcased. What is more, despite growing success and diverse use of machine learning algorithms, these algorithms still do not have established best practices and are not so often accepted in subjects concerning financial time series. Even counting in such examples as Zankova (2016), Sheta, Ahmed, Faris (2015). In this research, what is sought is adding and uniting these 2 niche subjects — stocks and bonds relationship analysis, utilizing machine learning algorithms. A variety of indicators, representing both private and government sector, including USA bond yields and related factors were analyzed with multiple stock market metrics. To even better understand the situation, more traditional statistical methods, like correlation coefficients and various visualizing techniques were used in preparation for the appliance of 4 machine learning algorithms. To summarize, it was confirmed that chosen bond variables do have influence on the stock market, however it is very unstable and requires disciplined model calibrating to maintain effectiveness for a long time horizon. Neural networks showcased best results no matter the circumstances. Factors related to more risky corporate bonds, were most influential to all stock indexes, while indexes trading volumes seemed to be unaffected by selected bond dataset at all. |
