Algorithm of Assessing Dynamic Correlation between Time Series Connected by TVP-Regression Model

Abstract

The present research proposes algorithm of assessing dynamic correlation of time series connected by TVP-regression model. Topicality of this task is stipulated by the fact that this model often describes asset behavior on finance markets, while modeling of their correlation link over time could help take into account risks, which is an integral part of building strategy of shaping the investment portfolio. This methodology can also be used to study the effect of shock proliferation on finance markets in time of crises. The goal of the research is to assess efficiency of the algorithm described in the work in comparison with the classic algorithm DCC GARCH. Comparison of the present algorithm with DCC GARCH method was carried out on synthetic data with several values of process error dispersion. As a result with all considered values of dispersion of the process error the advanced algorithm showed best figures in terms of mean-square error of assessed and real correlation. However, it was noticed that for higher values of process error the difference in result obtained by advanced algorithm and DCC GARCH method drops. In conclusion certain drawbacks of the algorithm were shown.