Mirghaemi, M.;
(2012)
Bayesian learning in financial markets: economic news and high-frequency European sovereign bond markets.
Doctoral thesis , UCL (University College London).
Abstract
This thesis investigates algorithmic trading in fixed-income markets, with specific reference to pre-trade analysis of high-frequency European sovereign government bonds and investigates the response of the prices of European government bonds to public announcements of economic news. To achieve this, the thesis is separated into three different experiments: 1. The first experiment investigates the effects of macroeconomic news on prices, trading volume and bid-ask spread, by using data on scheduled economic announcements and consensus forecasts to calculate the surprise component in the announcements. These data together with the tick-by-tick price information and trading volume helps to determine which announcements significantly affect prices as well as trading activity. 2. The second experiment investigates price changes as well as the size and sign of the price changes, and how quickly the public macroeconomic news is incorporated into European government bonds, by using vector autoregression (VAR) models: one of the most successful models for the analysis of multivariate time series. 3. The third experiment shows a way to forecast European government bonds prices, which change due to unanticipated information. Trading decisions are dependent on the relative precision of traders' prior and posterior beliefs. This is done by employing Bayesian models with Bayesian learning to measure the quality of information by its relative precision. Two sets of data have been used in this thesis: trading data that has been provided by Euro MTS (Mercato dei Titoli de Stato) and Dow Jones Elementized News Feed for macroeconomic announcements. The Euro MTS database is very rich in information, being composed of all bonds traded on the European platform. The original data set is recorded from 2003 with time stamps to the millisecond. Fixed-income securities can be traded on a domestic platform (like MTS France, MTS Germany and MTS Italy) as well as on a pan-European platform called the Euro MTS. I have selected only the data from January 2007 until end of December 2008, which includes 10 years of "on-the-run" (the most recent issued bonds) government bonds issued by the following countries: Germany, France and Italy. These countries and bonds were selected for their high liquidity. Selected bonds are traded on both the Euro MIS platform and also a domestic platform, like Germany MTS. Economic announcements and expectations data have been provided by Dow Jones. Dow Jones Elementized news feed places discrete pieces of news which that contain keywords, timestamps, symbols and other crucial data which introduces a faster and richer news data feed. The Elementized news feed includes economic and corporate news, with more than 230 indicators critical to market direction and pricing. I have selected only macroeconomic and federal reserve announcements starting from January 2007 until end of December 2008 from: Germany, France, Italy, EU, United Kingdom and United States. The intuition behind the selection of the news data is to have a broad set of news data that are aligned with the trading data in terms of both countries under research as well as the period under investigation [8]. The contribution of this thesis to science is to implement the forecasting model in two distinctive ways: an econometric model as well as statistical model, both of which are able to describe the effect of domestic and international macroeconomic news announcements on German, French and Italian 10-year, on-the-run government bonds. This thesis contributes to the existing literature in a number of ways. First, this research examines the high-frequency responses of German, French and Italian long-term bond prices to macroeconomic and U.S. federal reserve announcements, a topic which has received little attention in empirical literature. Second, the combined and broad set of German, French, Italian, Euro area, UK and US macroeconomic announcements, which cover more than 148 domestic and international news releases, is novel. Third, this research identifies the most important news based on its impact on price changes, bid-ask spread, volatility and trading volume. Fourth, this research provides an analysis of the responses of bond prices to news surprises by studying both return and order flow. Thus, my analysis corroborates not only earlier results on the relevance of both macroeconomic news and order flows for high-frequency bond price changes but also ties in with recent lower-frequency analysis, which indicates that at least part of the response of price changes to news comes via order flow. Fifth, this thesis presents a model known as "mixture of normal distributions" to model the price changes, from opening prices to mid-day prices, and from mid-day prices to closing prices. Finally, this thesis presents state-space models as well as Hidden-Markov model, to forecast state changed within a trading day. These results will be consistent with a variety of stylized facts regarding the yield curve. The effects that are documented in this research have relevant implications for yield curve modeling and for the microstructure of bond markets.
Type: | Thesis (Doctoral) |
---|---|
Title: | Bayesian learning in financial markets: economic news and high-frequency European sovereign bond markets |
Language: | English |
UCL classification: | UCL > Provost and Vice Provost Offices > UCL BEAMS > Faculty of Engineering Science > Dept of Computer Science |
URI: | https://discovery.ucl.ac.uk/id/eprint/1344061 |
Archive Staff Only
View Item |