Information and sentiment in financial markets

Zhang, Qi

Information and sentiment in financial markets

Zhang, Qi

Permalink

Publication Type:: Thesis
Issue Date:: 2024

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download thesisAdobe PDF (3.91 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Zhang, Qi
dc.date.accessioned	2024-12-02T22:18:09Z
dc.date.available	2024-12-02T22:18:09Z
dc.date.issued	2024
dc.identifier.uri	http://hdl.handle.net/10453/182311
dc.description	University of Technology Sydney. Faculty of Business.	en_US.UTF-8
dc.description.abstract	The first paper introduces a model that estimates the common and market-specific information flows when an asset is traded in multiple markets. The model does not require ultra-high frequency data and avoids the ‘who moves first’ interpretation of price discovery. We show that at sampling intervals from 0.01 to 2 seconds, the common information across all markets accounts for 67% to 94% of the total information flow, and the listing exchange accounts for half of the remaining information. The common information between quotes and trade prices ranges from 58% to 85% of the total information flow, with the remaining information coming mostly from quotes. Trade prices have very little information. The second paper measure news sentiment using BERT and explores return predictability based on a new database, Refinitiv Machine Readable News (MRN). The resulting portfolio achieves an annualized Sharpe ratio of 3.96, significantly higher than that of a passive investment (as proxied by S&P 500 index) and dictionary method, which achieves a Sharpe ratio of 0.32 and 2.94 respectively. We find that dictionary methods struggle to extract information from complicated texts compared with BERT. An interesting finding is that news of positive sentiment is tailored to fewer audiences, contain fewer topics, and are generally shorter. We further show that seasonalities and holiday effects do not appear to explain sentiment portfolio returns while article complexity does. The third paper uses XGBoost, a powerful and state-of-the-art machine learning algorithm, to predict next-day volatility jumps and then form portfolio for the next day. We show that controlling for news sentiment and volatility measures and using over 1,400 news topics, next-day RSJ, a recent development of volatility jump measure, is reasonably predictable, and using the predicted RSJ to form one-day portfolio achieves outstanding portfolio performance where annualized Sharpe ratio is 2.06 with only 44 stocks available . The performance is significantly higher than portfolio selection based on today’s RSJ (0.619) or today’s news sentiment (1.94). The improved Sharpe ratio is mainly from higher return. The portfolio earns a significant alpha relative to Fama-French 5-factor model benchmark. We show that investor attention and media coverage potentially explain the portfolio return.	en_US.UTF-8
dc.format	Thesis (PhD)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/182311/1/thesis.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2024 Qi Zhang
dc.rights	au.edu.uts.lib/nph
dc.title	Information and sentiment in financial markets	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

The first paper introduces a model that estimates the common and market-specific information flows when an asset is traded in multiple markets. The model does not require ultra-high frequency data and avoids the ‘who moves first’ interpretation of price discovery. We show that at sampling intervals from 0.01 to 2 seconds, the common information across all markets accounts for 67% to 94% of the total information flow, and the listing exchange accounts for half of the remaining information. The common information between quotes and trade prices ranges from 58% to 85% of the total information flow, with the remaining information coming mostly from quotes. Trade prices have very little information. The second paper measure news sentiment using BERT and explores return predictability based on a new database, Refinitiv Machine Readable News (MRN). The resulting portfolio achieves an annualized Sharpe ratio of 3.96, significantly higher than that of a passive investment (as proxied by S&P 500 index) and dictionary method, which achieves a Sharpe ratio of 0.32 and 2.94 respectively. We find that dictionary methods struggle to extract information from complicated texts compared with BERT. An interesting finding is that news of positive sentiment is tailored to fewer audiences, contain fewer topics, and are generally shorter. We further show that seasonalities and holiday effects do not appear to explain sentiment portfolio returns while article complexity does. The third paper uses XGBoost, a powerful and state-of-the-art machine learning algorithm, to predict next-day volatility jumps and then form portfolio for the next day. We show that controlling for news sentiment and volatility measures and using over 1,400 news topics, next-day RSJ, a recent development of volatility jump measure, is reasonably predictable, and using the predicted RSJ to form one-day portfolio achieves outstanding portfolio performance where annualized Sharpe ratio is 2.06 with only 44 stocks available . The performance is significantly higher than portfolio selection based on today’s RSJ (0.619) or today’s news sentiment (1.94). The improved Sharpe ratio is mainly from higher return. The portfolio earns a significant alpha relative to Fama-French 5-factor model benchmark. We show that investor attention and media coverage potentially explain the portfolio return.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/182311