Stock time series analysis

With 1.5 billion new messages added every day and a yearly data growth rate estimated at 30%-50%, both historical and real time financial data processing becomes increasingly challenging. Market forecasts and risk assessments with thousands of portfolio members, resulting in millions of measured dependencies, need to be processed periodically. The SmartDataLake Financial use case, led by the SME partner Spring Techno, targets exactly these needs of banks and hedge funds as well as producers and traders of agricultural goods.

The SmartDataLake toolkit includes a REST API for analyzing time series data, including functions for performing similarity joins, correlations, change point detection, seasonality detection and forecasting. These functionalities are currently being integrated in three products of Spring Techno, namely SeasonalBull, SysRiskPro and Maran Trader, supporting their target clients in their daily process of identifying investment opportunities while estimating and mitigating risk.

In this context, we have also addressed the problem of twin subsequence search in time series, as presented in our paper published at the EDBT 2021 conference. Specifically, our approach addresses the problem of subsequence search in time series using Chebyshev distance. First, we show how two state-of-the-art time series indices, namely KV-Index and iSAX can be adapted for this task. Then, we introduce a novel index, called TS-Index, which is tailored to this problem. TS-Index is a tree structure that summarizes the subsequences contained within each node using Minimum Bounding Time Series (MBTS). An MBTS consists of an upper and lower bounding sequence that enclose a set of time series. We present an extensive experimental evaluation showing that executing twin subsequence search using TS-Index is significantly faster compared to adapting the query execution over other indices.