High-frequency price patterns and trading behaviour under market stress: determining factors, return models and practical applications
With the fast growth in high-frequency trading, short-term periods of market stress, commonly known as “flash crashes”, have been an increasing phenomenon in financial markets. These periods can be thought of as short-lived malfunctions of capital markets typically involving a substantial price change and a drying-up of liquidity followed by a price reversal. Institutional and retail investors are affected by volatile trading during a flash crash when, for example, stop-losses are triggered and positions are sold at a significantly reduced value. Consequently, it is increasingly necessary to understand what happens in these periods of market stress and how they are triggered. This will help regulators to build mechanisms that aim to reduce future market stress, as well as helping practitioners to deploy quantitative models that optimise their risk and portfolio management in times of market stress. The objective of the study described in this thesis is to extend the current literature’s understanding of short-term market stress including the triggers, the amplification effects and the mathematical modelling of returns. A specific focus of this research on short-term market stress is the behaviour of market participants.
[paragraph redacted]
In the second part of the thesis, the behaviour of market participants and their impact on market liquidity during times of short-term market stress are examined by analysing the flash crash in the pound sterling versus the US dollar (GBP/USD) in October 2016 using proprietary FCA derivatives data reported under the European Market Infrastructure Regulation (EMIR). The analysis leads to the conclusion that dealers (primarily investment banks) contributed most to the fall in liquidity during the GBP/USD flash crash. The behaviour of dealers during the flash crash can be explained by their increased inventory holding costs as they could no longer effectively hedge their client trades in the intra-dealer market. Furthermore, the analysis in these chapters shows that spillover effects between the over-the-counter (OTC) derivatives market and the underlying foreign exchange (FX) market could be a key factor behind the observed illiquidity during this flash crash.
In the third part of the thesis, irregular returns are theoretically modelled in different market regimes including at tranquil times and illiquid times. Based on the earlier findings about the structure of market participants, the underlying price evolutions –Brownian motion (BM) and Ornstein–Uhlenbeck (OU) price processes – are specified. The empirical results of this part of the thesis provide evidence that the theoretical model of irregular returns based on a Brownian motion (BM) price process is mostly appropriate in illiquid times, whereas the theoretical model of irregular returns based on an Ornstein–Uhlenbeck (OU) price process is better suited to tranquil times.
In the final part of the thesis, the practical implications of the study’s results are presented and its applications to portfolio management and optimal execution strategies are shown. Finally, a long short-term recurrent neural network (LSTM RNN) from the machine learning field is developed and tested for use in detection of the switching points of market regimes. Having accurate points for market regime changes is important for the industrialisation of the research presented in this thesis.
