Optimal Trend Labeling in Financial Time Series

Predicting asset price trends is often posed as a classification problem, where are classified positive or negative. Since series noisy and volatile, it difficult to distinguish true from short-term fluctuations. To this end, several trend definitions have been proposed in the literature, but yet be known how these affect performance of algorithms designed learn such labels historical data. In paper, we define robustness labeling algorithm measure well classifier can withstand change cumulative return considering classifier’s generalization error. Moreover, propose noise model simulate desired accuracy score, which allows us evaluate without need train an actual consequently choose optimal terms robustness. Experimental results confirm adequacy show that perform better when trained with labels.