Use this URL to cite or link to this record in EThOS:
Title: Analysis of key drivers of trading performance
Author: Batrinca, B.
ISNI:       0000 0004 7230 7394
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
This thesis is an applied study for understanding the key factors of trading volume, providing an in-depth investigation of liquidity demand and market impact. This research was conducted in collaboration with Deutsche Bank and presents a series of empirical studies, which examine several underlying factors affecting trading volume when executing orders algorithmically in the European equity markets, which ultimately translates into trading performance and liquidity modelling. This addresses several aspects: the size of the liquidity demand relative to the predicted or actual volumes traded in the market; the choice of execution strategy given the liquidity properties of the stock; and the timing of the trade, given the market circumstances and released or anticipated company news. All of these reflect the investment skill of the portfolio manager and the execution skill of the trader, and to some extent the quality of the execution algorithm being used. The motivation is to investigate various factors that adversely affect the trading performance of algorithms, causing them to have excessive market impact or to under-participate when the market experiences periods of higher volatility. Although measuring the market fairness and efficiency is a crucial component for understanding the execution style and improving trading performance, the research into how to model and decompose trading performance requires further investigation. Trading volumes are a benchmark for determining an appropriate order size so as not to have excessive liquidity demand and therefore it is important to model well and accurately predict the volumes. The problem of sizing an order affects portfolio allocation and trade planning for multi-day trades, as well as intraday slices. To this end, four studies were conducted for time series analysis using machine learning methods based primarily on feature selection and regression. To achieve this, the thesis starts with a broad exploration of the trading volume drivers, followed by in-depth analyses of the effects of notable events, and concludes by proposing a volume prediction modelling framework. The thesis consists of the following four studies: 1. Examining Drivers of Trading Volume. The first study is an in-sample volume analysis, which explores the market dynamics and identifies a series of drivers of trading volume based on the price-volume relation, lagged time series market data, the day-of-the-week effect, and a novel approach on the price-volume asymmetric relation. 2. European Trading Volumes on Cross-Market Holidays. The second study further extends the exploration of the drivers of trading volume and investigates the anecdotal evidence of lower trading volumes when other markets are not trading (i.e. ‘the cross-market holiday effect’). The analysis considers the phenomenon in conjunction with the weekend effect, while indicating that the cross-market holidays are the real driver of the lower volumes on Mondays, and examines other aspects like lagged volumes, market capitalisation or multi-step ahead modelling. 3. Effect Expiry Day Effects on European Trading Volumes. This study examines the impact of sparse periodic events, such as stock index futures expiries and MSCI quarterly rebalances, on trading volume. The analysis explores anticipatory and subsequent effects of the index expiry and review dates. It investigates the main drivers of volume surges by discriminating between the Friday effect and the stock index futures expiries, and between the end-of-month effect and the MSCI quarterly reviews. 4. Developing a Volume Forecasting Model. The final study of this thesis incorporates the findings of the previous in-sample studies and provides an out-of-sample trading volume analysis, exploring the behaviour of time series variables in the context of volume prediction modelling, with seven statistical methods that are fit using the sliding and growing window approaches. The primary objective of the prediction model we propose in this final study is to achieve optimal accuracy in predicting the size of a trade given the market context, by proposing a dynamic model that switches between different models based on the temporal context. Finally, a stock-specific out-of-sample metamodel is constructed based on the recent performance of the initial stock-specific models that are independently fit. The thesis presents the following contributions to science: 1. Detailed exploration of trading volume drivers. The main objective of this thesis is to investigate the causal factors of trading volume. Salient drivers of trading volume include the lagged time series, the price-volume relation and its asymmetry, and temporal factors, e.g. the day-of-the-week effect, holidays and other notable dates. 2. Focus on the volume dimension. The empirical studies in this thesis focus on the trading volumes, unlike the majority of the reviewed literature, which investigates the relation between calendar effects (e.g. day-of-the-week, cross-market holiday etc.) and stock returns; few studies consider the relation between these effects and trading volumes. 3. Comprehensive pan-European stock universe. To the best of our knowledge, this research is conducted on the largest European data set in the relevant literature, covering the daily market data for 2,353 stocks from 21 European countries since 1st January 2000. Most of the relevant literature either employs small data sets or focuses mostly on the US equities market. 4. Accurate trading calendar data. Due to the unavailability of a high-precision trading calendar, we constructed a consolidated and normalised calendar with a comprehensive breadth and depth of events influencing equities across a wide range of European countries. This robust calendar covers the most liquid European exchanges’ trading calendar in 21 countries and the US trading calendar (since the US is the largest financial market and its trading holidays might influence the European liquidity), the stock index futures expiries for seven liquid European indices, and the MSCI quarterly review effective dates, along with the historical evidence of leavers and joiners for each analysed index, since 2000. 5. Established statistical methods applied to a new application domain. Advanced variable selection and machine learning methods have not been typically employed in the analysis of calendar effects and trading volumes; we choose a rather different approach and apply statistics and machine learning to this application domain. 6. Further insights into calendar effects. The field of behavioural finance and its literature on calendar effects contains mixed results that are often inconclusive. This thesis sheds light on a few calendar effects, e.g. the day-of-the-week, end-of-month, holiday and expiry day effects, and examines the extent of their impact on the trading volume. Research validation. The advice and expert validation of a leading investment bank confirm the industry demand and necessity of this research. Deutsche Bank drove the analyses conducted in this thesis, addressing real-world problems, such as the liquidity extraction model, the multi-day trade planning using multi-step ahead forecasting, or the effect quantification of special events on trading volume.
Supervisor: Treleaven, P. ; Hesse, C. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available