Working Papers
Presented at: QRFE one-day workshop on Market Microstructure, Fintech and AI, IFABS meetings at Oxford University, seminars at Commodity Futures Trading Commission, Paderbon University and Leicester University.
We propose two novel metrics inspired by the theory of price impact to show that intraday momentum trading at illiquid times exacerbates market fragility in high frequency setting. We provide evidence from Covid-19 selloffs in March 2020 and an out-of-sample test on the May 2010 flash crash using intraday US futures electronic market data. Our results suggest that machine learning models outperform logistic regression in identifying the persistence in patterns that our metrics capture in the onset of systemic liquidity crises. Placebo test further shows that these patterns do not emerge in good liquidity conditions, consistent with our hypothesis.
We propose two novel metrics inspired by the theory of price impact to show that intraday momentum trading at illiquid times exacerbates market fragility in high frequency setting. We provide evidence from Covid-19 selloffs in March 2020 and an out-of-sample test on the May 2010 flash crash using intraday US futures electronic market data. Our results suggest that machine learning models outperform logistic regression in identifying the persistence in patterns that our metrics capture in the onset of systemic liquidity crises. Placebo test further shows that these patterns do not emerge in good liquidity conditions, consistent with our hypothesis.
Quantum Walk Model of a Flash Crash, (2024), with Stuart Adams, Christopher McCarty and Jack Waller
Quantum Walk Model of a Flash Crash, (2024), with Stuart Adams, Christopher McCarty and Jack Waller
Financial asset prices may exhibit ’fat tails’ where large fluctuations around the mean occur more frequently than would be expected from a classical random walk. The frequency and scale of fat tails have been exaggerated in modern markets populated by algorithmic trading that relies on fast execution speeds and low latency. In this paper, we explore whether quantum walks provide a more versatile tool to model asset price fluctuations, particularly extreme price moves such as flash crashes. These events are increasing in occurrence in modern electronic markets wherein crowded exit by algorithmic traders rapidly amplifies price declines to the point of collapse. Our results suggest that a quantum walk model is indeed a more suitable choice for price simulation of these extreme price events.
Financial asset prices may exhibit ’fat tails’ where large fluctuations around the mean occur more frequently than would be expected from a classical random walk. The frequency and scale of fat tails have been exaggerated in modern markets populated by algorithmic trading that relies on fast execution speeds and low latency. In this paper, we explore whether quantum walks provide a more versatile tool to model asset price fluctuations, particularly extreme price moves such as flash crashes. These events are increasing in occurrence in modern electronic markets wherein crowded exit by algorithmic traders rapidly amplifies price declines to the point of collapse. Our results suggest that a quantum walk model is indeed a more suitable choice for price simulation of these extreme price events.
Ticker, Ticker, Boom: Algorithms Defuse Ticker Confusion, (2024), with Olga Balakina, Claes Backman and Anastasiia Parakhoniak
Ticker, Ticker, Boom: Algorithms Defuse Ticker Confusion, (2024), with Olga Balakina, Claes Backman and Anastasiia Parakhoniak
Presented at finance seminars at SAFE Frankfurt University and Durham University Business School.
This paper examines the relationship between short-term non-fundamental pricing and algorithmic trading. We study trading in two firms that share similar tickers across a large sample of US stocks for over two decades. We show that events that trigger trading in the first ticker trigger spill over into trading for the second ticker, leading to non-fundamental pricing errors. We find that these pricing errors attract algorithmic trading, which contribute to more efficiency by lowering the spread. The results suggest that algorithmic trading arbitrage away pricing errors and stabilise the market around non-fundamental trading.
This paper examines the relationship between short-term non-fundamental pricing and algorithmic trading. We study trading in two firms that share similar tickers across a large sample of US stocks for over two decades. We show that events that trigger trading in the first ticker trigger spill over into trading for the second ticker, leading to non-fundamental pricing errors. We find that these pricing errors attract algorithmic trading, which contribute to more efficiency by lowering the spread. The results suggest that algorithmic trading arbitrage away pricing errors and stabilise the market around non-fundamental trading.
Urgency in Action: Quantifying Wallet Behaviour using Ethereum Blockchain Data, (2024), with Dimitar Bogoev and Jiexiuhui Chen (Preliminary: May 2024)
Urgency in Action: Quantifying Wallet Behaviour using Ethereum Blockchain Data, (2024), with Dimitar Bogoev and Jiexiuhui Chen (Preliminary: May 2024)
We analyse a unique data set of individual wallets on Ethereum Network. Using the blockchain transaction data, we identify the wallets active for at least 30 days and quantify their behaviour by proposing a novel metric namely ”urgency metric”. Wallets are collated by block and matched with high frequency liquidity metrics from Binance and Coinbase using block time. On average, we document high frequency price momentum and lower market depth on cryptocurrency exchanges when urgency metric develops and increases, particularly at times of crises such as around ETH price crash in January 2022 and around FTX collapse. Using XGBoost, we show that urgency score metrics have substantial predictive power for the subsequent impact of wallet behaviour on risk and liquidity outcomes. Our work is the first to propose a blockchain based metric. It also provides a comprehensive framework for policy debate on the use of AI and machine learning for crypto market stability.
We analyse a unique data set of individual wallets on Ethereum Network. Using the blockchain transaction data, we identify the wallets active for at least 30 days and quantify their behaviour by proposing a novel metric namely ”urgency metric”. Wallets are collated by block and matched with high frequency liquidity metrics from Binance and Coinbase using block time. On average, we document high frequency price momentum and lower market depth on cryptocurrency exchanges when urgency metric develops and increases, particularly at times of crises such as around ETH price crash in January 2022 and around FTX collapse. Using XGBoost, we show that urgency score metrics have substantial predictive power for the subsequent impact of wallet behaviour on risk and liquidity outcomes. Our work is the first to propose a blockchain based metric. It also provides a comprehensive framework for policy debate on the use of AI and machine learning for crypto market stability.
Contagion in high-frequency (il)liquidity networks, (2024), with Kumushoy Abduraimova (Preliminary: July 2024)
Contagion in high-frequency (il)liquidity networks, (2024), with Kumushoy Abduraimova (Preliminary: July 2024)
Presented at: 12th Bachelier World Finance Society in Rio De Janeiro.
Financial market liquidity is a complex phenomenon, and so is its measurement. Market liquidity has various, and often interrelated, aspects such as transaction costs, breadth, depth, price impact and so on, and no single liquidity indicator accounts for all of those. The task of measuring liquidity is further complicated by the interconnectedness of financial markets which could lead to propagation of illiquidity across (not necessarily related) assets (for instance due to funding constraints) thereby amplifying the initial shock. We propose a new liquidity measure that accounts for that cross-asset amplification of shocks to liquidity: centrality measure that is based on the theories of copula and networks to capture the heavy-tailedness property of the considered variables and the network effects. In a directed network we further differentiate between amplification based on outgoing and on incoming links. The former indicates contagiousness of a given asset in terms of its impact on the liquidity in the rest of the network. The latter reflects its vulnerability to the liquidity shocks to other assets. The introduced liquidity measures can be used as indicators of systemic importance of individual assets in the network and as early warning signal of large-scale liquidity dry-ups.
Financial market liquidity is a complex phenomenon, and so is its measurement. Market liquidity has various, and often interrelated, aspects such as transaction costs, breadth, depth, price impact and so on, and no single liquidity indicator accounts for all of those. The task of measuring liquidity is further complicated by the interconnectedness of financial markets which could lead to propagation of illiquidity across (not necessarily related) assets (for instance due to funding constraints) thereby amplifying the initial shock. We propose a new liquidity measure that accounts for that cross-asset amplification of shocks to liquidity: centrality measure that is based on the theories of copula and networks to capture the heavy-tailedness property of the considered variables and the network effects. In a directed network we further differentiate between amplification based on outgoing and on incoming links. The former indicates contagiousness of a given asset in terms of its impact on the liquidity in the rest of the network. The latter reflects its vulnerability to the liquidity shocks to other assets. The introduced liquidity measures can be used as indicators of systemic importance of individual assets in the network and as early warning signal of large-scale liquidity dry-ups.