The MSc project is a substantial component of the programme, occupying around 4 months. It is a piece of original work undertaken by the students under the supervision of an academic researcher and, in most cases, also with an external supervisor. Most projects are carried out in association with a bank, hedge fund, consultancy, or systems provider in the finance industry, and we endeavour to arrange suitable placements. 

You will find below a large sample of past theses, covering wide range of topics.

-

19 November 2020- Goldman Sachs-Imperial MSc Flash Talk Series

 We are proud to partner with Goldman Sachs for a Flash Talk series of 2019-2020 MSc Math Finance Theses.

Event type : Zoom Webinar
Date: Thursday, 19 November 
Time: 4:30-6pm GMT
 
TITLESPEAKERABSTRACT
A Signature Approach to Regime Classification Conor McIndoe

We provide a data-driven approach to regime  classification. Regimes are identified with distributions of path signatures; a distance between these distributions is described and a data-driven clustering algorithm is applied to propose natural classifications of market regimes with as little manual specification as possible.

Truncated order for linear functional Signature regression Chenhao Jin

The Signature transform can be used as a feature extraction method for time-sequential data. In practice, a key problem for this method is to decide its truncated order. We will give a brief introduction for Signature and discuss how truncated order can be increased as a function of the observation size under linear functionalSignature regression framework.

Distributional reinforcement learning for optimal execution Toby Weston

Distributional Reinforcement learning is one of the most promising advances to deep Q reinforcement learning in recent years. We show that by adapting it to the trading environment we can outperform the previous cutting-edge reinforcement learning tools for optimal execution, in environments using both simulated and historical data.

Joint S&P 500/VIX smile calibration problem with rough volatility models Pierre-Alexis Corpechot

Rough volatility models are able to capture and explain stylised facts observed from historical market data in volatility time series and in the implied volatility of option prices. Since the fits to implied volatility surfaces are promising, rough volatility models open the door to further calibration problems, including the challenging issue of joint calibration of S&P 500 and VIX implied volatility surfaces.

Offline calibration of the SABR model with neural networks Hugues Thorin

This last decade, markets have become more automatized and trading frequency has made calibration of models, such as the SABR model, more time-consuming. This time constraint is critical and will become more pronounced in the future. Neural Networks offer the possibility to offline the calibration and to obtain updated parameters instantaneously.

Increasing venture capital investment success rates through machine learning Thomas Hengstberger

Machine learning models are developed to predict which start-up companies are successful investment opportunities for Venture Capital investors, with models trained on a time series database of start-up company snapshots. We show that this approach increases the investment success rate by at least 50% when compared with the success rate of the average Venture Capital investor.

 
Summary of the table's contents

Past and Current Project Partners

Alcazar Bainbridge Partners
Bank of America Barclays
BNP Paribas Beekin
Blackrock Citigroup
Credit Suisse Deutsche Bank
EBRD Ernst & Young
Goldman Sachs HSBC
IHS Market IXIS-CIB
JP Morgan Janus Henderson
Jetstone Asset Management Lloyds TSB
Marney Capital Mazars
Mitsubishi Group Morgan Stanley
Natwest Markets Norges Bank
Quod Financial Rogge Global Partners
Santander Swiss Re
Synergis Toronto Dominion Securities
UBS Velador Associates
XSOR Capital  

2019-20

2017-18