Analysis of rare events using multidimensional liquidity measures

In this paper we develop a framework to analyze high-frequency (HF) financial transaction data focused on estimating a multidimensional intraday liquidity measure and detecting rare events. Many liquidity measures based on Trade and Quote (TAQ) and Limit Order Book (LOB) datasets are consolidated for this purpose through dimensionality reduction techniques. Several outlier methods based on extreme value theory, distance-based outlier methods, and tree-based algorithms are implemented to identify clusters of rare liquidity events. These methods provide insights into the behavior and occurrence of outliers. The methodology is optimized for HF intraday implementation. The framework is applied to transaction level data covering the beginning of COVID-19 outbreak period. We observe that after peak news activity, high-volume stocks experience extreme low-liquidity events almost immediately, while low-volume stocks have a time delayed reaction. The behavior of a select number of tickers is analyzed in detail over the outbreak period. The framework proposed can detect extreme liquidity events in real time and thus can be used to monitor market activity and provide early warnings about liquidity trends. A new intensity indicator measure is developed to assess and visualize extreme liquidity events.