The author studies the impact of social norms and specifically religious attitudes, on the stock returns of “sin firms”: companies which produce products that are conventionally deemed sinful by investors. I expect more religious countries have lower approval of “sin” and sinful industries, resulting in higher societal sin aversion. Therefore, lower demand for sin stocks in these countries depresses prices, evoking higher returns. Abstaining from investment forgoes these abnormal returns. Findings may suggest whether religious factors are worth considering for investors to utilise possible arbitrage opportunities.

I construct panel dataset of 11245 companies, across eight countries in Asia and Europe, from 2005-2014. A country-specific religiosity index is also constructed, studying its impact on the net returns of gambling, tobacco and alcohol companies. Regression analysis finds no evidence of a sin premium (higher return of sin versus non-sin stocks) is found, consistent with literature outside America. Moreover, sin investors are found to incur a penalty in irreligious environments. However, the net return of sin stocks relative to the wider stock universe increases as the level of religiosity goes up. This is considered as evidence in support of the neglected-firm and sin aversion hypotheses. A positive religiosity effect on general stock returns is also found significant, with more religious countries incurring higher returns.

Previous studies focus on Western stock markets. With Asian markets constituting 33.3% of global capitalisation, their inclusion is pertinent. Religion-specific studies assume more sin averse faiths are more religious, which may not hold. Therefore, using religiosity overcomes this and has not been studied.

Cigarette smoking is an important issue for consideration across the world, and governments spend a significant proportion of their resources to reduce smoking. Smoking has detrimental effects on the health of an individual, causing, among other things, cardiovascular and respiratory problems. Grossman (1972) found a positive relationship between health and earnings. Considering this, coupled with the evidence shown by medical literature regarding ill-effects of smoking, it seems interesting to investigate how smoking affects labour market outcomes.

In recent times, limited literature has been published on this topic, especially in the UK. This study takes into account the effect of new anti-smoking laws. Studies such as Van Ours (2004) found ‘bad’ instruments that led to implausibly high estimates. This paper successfully tackles endogeneity by finding relevant and exogenous instruments. The direction of the endogeneity bias in previous researches is not consistent (Auld, 2005; Neumann, 2013). This research provides an explanation for the direction of the bias and instruments used to overcome lacunae in previous literature.

This study investigates the effect of smoking on wages in the UK using the ‘General Household Survey, 2006’ after imposing certain restrictions for a more accurate analysis. Using OLS estimation, we found a wage penalty associated with smoking of 5.4%. This paper showed that a very small proportion of this effect is through health and it is either because of reduced productivity due to smoking breaks or high rate of time preference leading to less investment in human capital. Further, we found a wage differential due to smoking across different occupational class. . Lastly, IV estimates showed the actual wage penalty of 7.6%.

The results of this study suggest a significant negative relationship between smoking and wages. This has important policy implications as governments can increase the awareness of people regarding wage consequences of smoking and target segments of labour market that are more prone to it.

Half of the world’s migrants are female, accounting for 95 million women. Economic improvement is the main reason for female migration, where women who are uneducated and lack resources are left with no choice but to enter the unskilled labour sector, often becoming ‘maids’ in richer countries. The aim of this paper is to examine female migration – specifically in the context of female migrant domestic workers – and whether the feminisation of migration provides these women with greater empowerment. This paper conducts a literature review to analyse cases of migrant domestic workers in Hong Kong and South Africa, including why females migrate to these areas, the governmental legislation in place to protect these workers, and the specific forms of exploitation that domestic workers encounter in these areas. Due to the private nature of domestic work, and the host countries policies, these women are often subjected to exploitation, oppression and abuse. Key findings reveal that domestic work occurs outside of the public eye and, as a result, legislation is difficult to monitor and implement. Furthermore, due to inadequate rights, and the entrenched power status between employer and employee, these women are often abused financially, emotionally and physically. Specifically in the context of South Africa, these migrants are often subjected to xenophobia. Domestics in Hong Kong, however, are often financially abused and are required by law to live in the employers’ home, removing their freedom to move. The overarching conclusion is that female domestic migrants are often exploited rather than empowered.

Despite surrogacy and assisted reproductive technologies (ARTs) forming a large part of reproductive medicine, the law is both underdeveloped and out of line with social opinion. This is particularly obvious with regard to couples who deviate from the societal norm (such as couples of low socio-economic status and same-sex couples). This paper builds upon the theories of Foucault and Spivak in terms of epistemic violence, as well as research papers on same-sex parenting and the effects of surrogacy on family dynamics by Van Gelderen, Spar and Sandel, respectively. This dissertation provides a modern perspective on the issue, taking account of developments such as the legalisation same-sex marriage, in a way other research in the area does not.

Additionally, this dissertation exposes gaps in the current law through close analysis of the legislation, concluding that the law not only ineffectively manages surrogacy and assisted reproduction, but is also unnecessarily complex and makes needless distinctions based on sexual orientation and class. This conclusion challenges the legal concept of equality and seeks to bring the law into line with public opinion – which has changed considerably since the introduction of the legislation in the late 20th century.

This dissertation, therefore, advocates for both legislative reforms (such as adopting an inclusive, standardised framework of eligibility criteria for ARTs and managing surrogacy through licensed fertility clinics) and wider systemic changes. These include reforms to sex education and improving diversity in the law-making process to lessen the disadvantage levied upon intersectional couples in the context of reproductive medicine.

As our lives get busier, it is hard to keep track of one’s tasks and activities as well as how and when to complete them. Therefore, there is a need for some sort of helper in the form of an activity planner to help people, especially students, to keep track of their daily activities and tasks. This project pertains to phase one of the development process of an intelligent planner, developed for NTU Student Wellbeing Centre.. In this phase, the purpose of the project is to develop a web-based activity planner in which the user can manually input information regarding certain tasks or activities, and the system will output them in the user interface in a time-sorted manner. It is also equipped with many other useful features such as a notification feature, a mark task as done feature, integration with NTU timeline, drag and drop function, etc. The application is written in HTML, CSS, JavaScript and PHP, and AngularJS is chosen as the JavaScript framework to write the front-end code of the application. The database chosen to store application data is MySQL.

Growing concern over the environmental damage caused by hydrocarbon-fuelled vehicles has encouraged and accelerated the development of electric alternatives. Despite the increased availability of electric vehicles (EVs) on the market, energy storage and management systems remain a topic of substantial research. This study aims to assist with the provision of current information about the driving range of an EV through an accurate and computationally efficient estimation of the remaining energy, or State-of-Charge (SoC) of the battery pack.Due to the large number of batteries assembled in a pack, proposed method involved parametrisation of a battery pack composed of 6 cells as one Equivalent Circuit Model (ECM). A state-estimator known as the Extended Kalman Filter (EKF), and an alternative version for adaptive noise tuning (B-AEKF) were employed to estimate the battery pack SoC. The accuracy of the pack (8% RMSE) was compared against the single cell (5%RMSE) model to verify the effectiveness of the proposed methodology. Less accurate SoC prediction of the pack EKF suggests that the estimator suffers from initial cell imbalance and an inaccurate pack ECM parametrisation. Therefore, in the EV’s application it may be more beneficial to use a cell-level-EKF on parallel strings of the battery pack to maintain higher SoC accuracy and therefore the driving range. Due to the dynamic operating conditions of an EV, application of the B-AEKF is preferred. Further research could focus on the estimate accuracy improvement via application of the higher order ECM and an Unscented Kalman Filter.

Whilst many initiatives have been developed to tackle social problems, from crime to unemployment, both on a community level and a national one, success rates tend to be low. A lack of feasible social initiatives will have a nation-wide impact on the well-being of society. My research explores how Design Thinking can be used as an approach to adopting successful Social Innovations in my local Community Centre.

A number of Social Innovations already exist in this Community Centre, however the area still suffers from one of the highest unemployment rates in the UK. Previous literature has taken on challenges through Design Thinking, take ‘Innovation and Health’, however this is usually done in the Healthcare sector and only covers the first two stages of Design Thinking. My research aims to tailor such thinking to a new context, given its unique culture and non-technological nature. Using Case Studies such as ‘Social Innovation: Driving Force of Social Change’ my research aims to understand why Social Innovations are yet to achieve their potential, constituting the initial step of Design Thinking, then we explore how Design Thinking can be used as an approach to adopting Social Innovations via the initiation and implementation stages.

This case study research will highlight the causes of failures through Surveys, Observations and Ethnography with Governors, Volunteers, Attendees and non-Attendees. Using this data as well as previous case studies, we’ll ultimately build a framework that can be used across community centres to ensure successful Social Innovations are initiated and sustained.

Chronic Kidney Disease (CKD) corresponding to a progressive loss in kidney functions is a worldwide public health issue which can lead to kidney failure. On that account, it is crucial to detect any potential renal problems at an early stage. Machine learning (ML) is a method of automated model building that is widely applied in biomedical engineering. Such models often require tremendous amount of training; however, the small nature of biomedical data induces sporadic effects on model performances. This work bridges this gap by following a framework comparing models suited for small data in classification of CKD datasets. Consistent model comparisons were primarily developed applying the multiple runs and the surrogate data approach on Random Forests (RF), Decision Trees (DT), Support Vector Machines (SVM) and Naïve Bayes (NB). Then, a trade-off was formulated between the number of features and samples for DT while assessing its robustness accounting for random effects due to small data.

Experiments allowed to define RF along with SVM as best classifier achieving an accuracy of 98.9% on the complete dataset. However, SVM performed poorly on smaller datasets whereas less sophisticated models such as NB and DT achieved substantially better accuracies. The DT model developed from small data was able to be successful in determining the minimal set of parameters that define Chronic Kidney Disease. The work has limitations as it is specific to a given dataset that belongs to one medical centre. An expansion of such analysis to other centres could benefit further model development.

Prehension is a key skill for humans to possess in order to manipulate and handle objects; however, this skill could become impaired due to injury or illness such as spinal injury, stroke or cerebral palsy. Being able to accurately measure the grip force of an individual can be an important part in the assessment and rehabilitation of patients.

Currently, sensors accurate enough to measure grip force for clinical purposes are highly expensive and therefore impractical for use in the areas where they are most needed, such as clinics or schools. A system being developed between the schools of Mechanical Engineering, Psychology and Sports Sciences – FATKAT (Finger and Thumb Kinetic Assessment Tool) – uses a low-cost, single-axis sensor. However, research has been carried out to develop a design methodology and prototype three-axis force sensor that is cheap, reliable and easy to produce. There are a number of parameters involved in producing a sensor of this type and this project focused on designing a sensor to replace the existing single-axis sensor by balancing these parameters for the required purpose. As a result, we were able to develop a sensor capable of measuring human grip forces in three axes for under £40. A number of prototype units have been made to be developed further through the 2017/18 academic year.

Recognition and prediction of movement intention are crucial processes for the development of intelligent wearable robots capable to safely assist humans in their activities of daily living. These processes depend on the systematic data collection in real-time from multiple sensors attached to the human body. For that reason, this work is focused on the development of a wearable sensors network for lower limbs that, together with high-level computational methods, will allow robotic systems to understand human movements and provide accurate and timely assistance. Systematic data collection is proposed by using multiple synchronised inertial measurement units (IMUs) attached to lower limbs. For data analysis, a Raspberry Pi (RasPi), together with Bayesian formulations and Convolutional Neural Networks (CNN), are proposed for recognition of activities such as level-ground walking, ramp ascent and ramp descent. Currently, wireless communication between the RasPi and an IMU sensor has been configured. Furthermore, a CNN has been successfully implemented in the RasPi for recognition of three walking activities in offline mode and using data previously collected from three IMUs attached to the lower limbs of human participants. This process achieved a mean recognition accuracy of 95%. Research is still in progress and requires the following future work: synchronisation of multiple IMUs, implementation of Bayesian approaches and recognition of walking activities in real-time. The results from this work will provide a robust high-level recognition system for control of wearable devices to assist humans in the elderly age or those who have a lower limb impairment.

Air pollution in cities is an important issue that affects millions of people worldwide. It causes many health problems such as headaches and respiratory diseases, which can lead to premature death. Increasingly, around the world, air pollution episodes are happening that exceed safe levels defined by the World Health Organisation. Hence it has become one of the major priorities for the Mayor of London, who is taking tough measures to tackle the rising problem such as with the introduction of the toxicity charge.

Data related to air quality is becoming more readily available. Already in London, there is a network of air pollution sensors that record concentrations of PM10, PM2.5, SO2 and NO2 at high temporal resolutions. While this data is publicly accessible, this project will also make use of a private London land use dataset, provided by the Greater London Authority (GLA).

To build a spatial-temporal model with this data we will use Gaussian Processes (GP). These are a probabilistic method that quantify uncertainty and since we expect a correlation between each of the pollutants we can use a multi-task GP where each pollutant is a task. The aim is for our model to be more accurate than the current one used by the GLA, which is predominately based on the road network. These results can then be used to find areas where adding additional sensors will most improve the quality of the data. This research has the potential to aid the GLA in their efforts to reduce London’s air pollution problem.

This paper presents the process of developing a facial analysis system for household chatbots and related research. The main components of this system are facial identity recognition and facial expression classification. This system will be implemented on the chatbot for real-time visual analysis. The system will provide information extracted from a camera to the linguistic analysis, which will comprehensively give a response based on both natural language and visual information. Since the processor in a household chatbot is lower in performance than one in a desktop computer, we will design a light structure of a deeper model with acceptable performance for the model to meet requirements in terms of speed and accuracy. In this paper, the author discusses the traditional training scheme, which uses a classifier connected with a softmax activation function, and computes the loss using cross entropy equation. Based on the discussion on traditional softmax – cross entropy method – the author made several attempts using different methods to modify the model structure and proposed a modified method in the training procedure. Furthermore, based on statistical and topological analysis, the author also evaluated the availability of cosine similarity, Euclidian distance similarity and the metric learning method of projecting the feature vector to Euclidian distance.