Determining Optimal Dimension of Financial Prices Time Series

Since the beginning of the 21st century the application of non-linear time series analysis can be found across the broad variety of science fields like engineering, biology, humanities and beyond. However, few academic works on the topic of state-space reconstruction in financial time-series can be found due to the strict data requirements like the size of sampling period, equally spaced observations and etc. This research aims to apply embedding coordinates techniques developed and widely used in other branches of science to exchange rates time-series using basic machine learning algorithms as the forecasting model. The research is based on high-quality data provided by Interactive Brokers database. Hence, this study might shed a powerful light on applicability of state-space reconstruction models for practitioners in the field of automated foreign currencies trading.