Don’t Stop Me Now! Clustering and Classification of Wanted and Unwanted Volatility Interruptions

Don’t Stop Me Now! Clustering and Classification of Wanted and Unwanted Volatility Interruptions

Market operators and regulators worldwide implement safeguards to pause or slow down trading in case of large price changes to ensure price continuity and the proper functioning of fully electronic securities markets. However, critics argue that these safeguards can negatively impact market efficiency by delaying price discovery, especially when price changes are driven by substantial new information rather than erroneous orders or misconfigured trading algorithms. This paper introduces a cluster-based approach to differentiate between wanted and unwanted volatility interruptions, a common type of market safeguard. By employing a combination of convolutional neural networks (CNNs) and long short-term memory (LSTM) layers, we are able to predict unwanted volatility interruptions using ex-ante order book, trade, and news information. Our analysis shows that unwanted volatility interruptions are more likely to occur when liquidity and volatility are high. Our findings can enhance market safeguard mechanisms by reducing unnecessary interruptions, thereby improving market efficiency.