These seminars are for Postgraduate Research students in the Department of Statistics and attendance is by invitation only. Seminars in the Michalemas term 2015-16 take place in the Leverhulme Library (COL 6.15), starting at 1pm, except where otherwise stated.
Lent Term 2016
Wednesday 3 February 2016, 3-4pm
Rafal Baranowski
(Time Series and Statistical Learning group)
A quick guide to speeding up R code
Abstract: One thing we often hear about R is how slow it is. That's not necessarily the case! In this short presentation we will discuss the following techniques aiming to speed-up R code.
-
Code profiling.
-
Running R function on multiple cores and machines in parallel.
-
Writing R extensions in C.
-
Replacing standard system libraries to their better optimised variants.
This will be accompanied with an example which demonstrates performance of these techniques in practice.
Wednesday 17 February 2016, 12-1pm
Chao Zhang
(Imperial College Business School)
Hedge fund arbitrage on the US equity market
Abstract: Hedge funds, as well as all institutional investors have shifted from large stocks into small stocks over the last 30 years. We combine hedge fund equity holdings with short interests to revisit hedge funds' preferences for stocks on the U.S. equity market. We use net abnormal trades on a stock to infer the aggregate view of the hedge fund industry. Hedge funds with signi cant net abnormal trades are inferred to be held or short sold for arbitrage purpose and low level abnormal trades indicate that the stock is held for hedging purpose. We hypothesis that, based on net long abnormal trades, hedge funds' preference for certain types of stock characteristics can be explained by their alpha chasing behaviour. For stocks held for hedging purpose, we expect that hedge fund holdings reveal a preference for safe stocks, i.e. large size, low volatility, etc.
--
Mengqiu (Matthew) Cao
(Bartlett School of Planning, University College London)
Investigating Travel Vulnerability in Greater London: Future Changes in Oil Prices and Housing Affordability
Abstract: This presentation discusses the combined problem of high car dependency/car preference and housing affordability – in view of likely continued volatility in oil prices (and hence petrol and diesel prices) and also rising house prices. A travel vulnerability index is developed incorporating measures of car dependence and housing affordability (drawing on Dodson and Sipe, 2007; Lovelace and Philips, 2014; and Dodson et al., 2015). Greater London is used as the case study, with 2001 and 2011 data analysed at the Lower Super Output Area (LSOA), and presented using Geographic Information System (GIS). The results obtained for Greater London reveal that there are high levels of vulnerability in households located in suburban areas, particularly where there are high levels of car dependence, but also those in the exclusive and, to many people, unaffordable inner urban areas, such as Kensington and Chelsea, and Westminster. If oil prices rise and use of petrol or diesel cars becomes much more expensive, and house prices continue to rise relative to average household incomes, then it can be assumed that many parts of Greater London would become more vulnerable, difficult to access for those on average or even relatively high incomes, and that those areas most affected may become much less popular to live in. The seemingly inexorable rise in London’s population is not always guaranteed to continue, and perhaps the affordability issue is one reason mitigating against this. Not considering this possibility represents a major oversight in transport planning in London – there is much too little emphasis in improving public transport accessibility in suburban areas, and alongside housing affordability issues, this is placing quite a stress on household budgets and life in large parts of London – which is only likely to increase in severity in future years.
Wednesday 2 March 2016, 3-4pm
Xinghao Qiao
(Assistant Professor; Time Series and Statistical Learning group)
Introduction to Functional Data Analysis
Abstract: Due to recent advances in high technology, scientific researchers are able to collect and store large complex data. Functional Data Analysis (FDA), a branch of statistics, has attracted increasing attention in recent years. FDA refers to the statistical analysis of data consisting of samples of random curves, where each measured function is treated as the unit of observation. In this talk, several important topics in FDA will be discussed.
-
Functional Principal Component Analysis (FPCA)
-
Functional Regression Models
-
Sparsely Sampled Functional Data (if time allows).
Wednesday 16 March 2016, 12-1pm
Victory Idowu
(Risk and Stochastics group)
Introduction to Capital Requirements for the Banking and Insurance Industry
Abstract: Despite the complexity of regulation developed after the financial crisis, there is still a need for suitable methodology which captures all the material quantifiable risks. The Solvency Capital Requirement (SCR) is a quantitative measure outlined in Solvency II Pillar I regulation. It necessitates that an insurance firm must use a standard formula or provide a model which aggregates the marginal Value at Risk for several key risk drivers. In this talk, recent innovations for aggregation are discussed as well as limitations and possible scope for improvements.
Michaelmas Term 2015
Wednesday 21 October 2015
Evangelia Mitrodima
(LSE Teaching Fellow)
Another way of estimating the asset return distribution
Abstract: In this talk, I will present a dynamic joint quantile model that accounts for both the scale and the shape of the conditional asset return distribution. Inter-quartile range is used as a convenient basis for the estimation of the scale of quantiles that belong to the left and the right side of the distribution. Daily returns on a stock are used for an in-sample and an out-of-sample exercise. Results suggest that the proposed model exhibits a superior performance in terms of better describing the evolution of the right and the left tail in-sample and producing acceptable forecasts out-of-sample. These findings indicate a better trade-off in the modelling of the scale and the shape of the distribution.
Wednesday 11 November 2015
Jose Pasos
(Financial Mathematics group, Department of Mathematics)
Irreversible capacity expansion with possible default
Abstract: We consider an irreversible capacity expansion model in which an economic indicator is driven by a linear diffusion. The investor can increase the project's capital until default or bankruptcy time is reached, which is modeled by the hitting time of zero. The associated optimisation problem takes the form of a singular stochastic control problem which, under suitable assumptions, admits explicit solution
Wednesday 18 November 2015
Haziq Jamil
(Social Statistics group)
Two-stage Bayesian variable selection for linear models using I-priors
Abstract: In a previous work, I showed that the use of I-priors in various linear models can be considered as a solution to the over-fitting problem. In that work, estimation was still done using maximum likelihood, so in a sense it was a kind of frequentist-Bayes approach. Switching over to a fully Bayesian framework, we now look at the problem of variable selection, specifically in an ordinary linear regression setting. The appeal of Bayesian methods are that it reduces the selection problem to one of estimation, rather than a true search of the variable space for the model that optimises a certain criterion. I will talk about several Bayesian variable selection methods out there in the literature, and how we can make use of I-priors to improve on results in the presence of multicollinearity.
Wednesday 25 November 2015
Lorenzo Tanadini
(Visiting Research Student from the University of Zurich)
Neurologists borrow from social statistics: a first progress report
Abstract: The correct modeling of the spatial and temporal components of neurological processes remain a major concern of all neurological disorders.
However, clinical studies with complex ordinal endpoints often resort to
scale collapsing and/or inappropriate statistical methods. Fortunately, the specific structure of the neurological data being analysed in the field of spinal cord injury allows to take full advantage of the most recent advancements in social statistics, which has historically been confronted with complex ordinal outcomes and is a the forefront of methodological advancements for their analysis. This project aims at adopting these latest developments and tailoring them to the modelling of ordinal responses as collected in patients who suffered a spinal cord injury. Our approach is intended to replace scientifically questionable common practice in neurology, provide deep insight into the spatio-temporal process of recovery following spinal cord injury, and fill a major gap hindering the bench-to-bedside translation of promising preclinical results.
Wednesday 9 December 2015 (Start time: 4pm)
Joanna Tumilewicz
Mathematical Institute, University of Wroclaw
Insurance contracts based on drawdown and drawup events of spectrally negative Lévy processes
We consider insurances polices based on drawdown/drawup event where underlying asset is derived by spectrally negative Lévy process. We analyze four contracts.
In the first one the protection buyer will pay constant premium $p$ continuously until the drawdown of size $a$ occurs. In return he receives the insured amount of $\alpha$ at the drawdown time. Another contract provides protection from any specified drawdown with drawup contingency. This contract may expire early if a drawup event occurs prior to drawdown one. The last two contracts are extension of the previous ones by additional cancellable feature, which allows investor to terminate contract earlier. We focus on two problems: calculate fair premium for basic contracts and find the optimal stopping time for polices with cancellable feature. In analysis we use heavily the fluctuation theory of drawdown Lévy processes and general theory of construction of optimal stopping rules
for Markov process.
This is a joint work with Zbigniew Palmowski
This seminar series is co-ordinated by Evangelia Mitrodima.
General enquiries should be addressed to Penny Montague (research administrator).