Applications of data science: policy evaluation and stock market forecasting

In this thesis, I will be exploiting data sciences techniques to evaluate the impact of policy changes and perform forecasting for the stock market. This thesis consists of four chapters. Chapter one is the introductory chapter. Chapter two is titled as "The Winners and Losers of Brexit: UK Manufacturing Sector Analysis" and looks at evaluating the economic impacts of the Brexit vote on the UK manufacturing sector employment. Chapter three is titled as "Land Tax Holiday and House Price: Evidence from the UK" and evaluates the success of the Stamp Duty Land Tax (SDLT) holiday on the UK housing market. In these two chapters, I use the "State of Art" synthetic control method to perform these evaluations. Lastly, chapter four is titled as "Predicting the Volatility of the S&P 500 Stock Index: The Role of News Sentiment" and uses newspaper data to predict the volatility of the S&P 500 index returns.

In chapter two I analyse the effect of the Brexit vote on UK manufacturing divisions using a synthetic control methodology. More specifically, this paper answers the question "What if the Brexit vote never happened?" with the focus on the employment of the manufacturing sector. The findings show that in the three years following the Brexit referendum, the UK total manufacturing sector's employment decreased by an average of 3.6%. When investigating the impact on individual manufacturing divisions, I found that three manufacturing divisions benefited from the Brexit vote, six manufacturing divisions were negatively affected, while seven manufacturing divisions were unaffected. These results are of particular importance to policymakers who will support sectors affected as a result of the Brexit referendum.

In chapter three I investigate the impact that the Stamp Duty Land Tax (SDLT) holiday that was introduced in July 2020 had on the UK housing market. The SDLT was introduced following the ease of the COVID-19 lockdown restrictions in an attempt to boost the UK housing market. I perform a counterfactual analysis using Synthetic Control Method to answer the question "How would the UK housing market perform in the absence of the tax holiday?" The findings of this paper illustrate that the SDLT holiday resulted in an increase of 2.11% and 2.96% on annual rates in the house prices in the two quarters following the tax holiday. On a regional level, I document the heterogeneity of the housing market performance across the 9 Nomenclature of Territorial Units for Statistics 1 regions (henceforth NUTS 1). All the UK large regions exhibited an increase in the house prices except for Greater London. The findings of this paper suggest that the pandemic caused city-dwellers to move out of big cities to the countryside and coastal cities, driving their house prices to increase.

In chapter four I decompose the textual data in the U.S. Newspapers into different news topics using unsupervised machine learning approach. I then measure the tonality of the news topics using dictionary based approach and investigate their abilities in predicting the volatility of stock returns. The results of this paper demonstrate that incorporating topical news sentiment as exogenous regressors in GARCH(1,1) and GJR-GARCH (1,1) models explains the stock returns volatility. More precisely, the news sentiment indices for topics related to currency, monetary policy, manufacturing, trade and trade policy can explain the variation in the volatility of stock returns. When combining the individual forecasts using regression based method, the combined forecasts provide accuracy in predicting the volatility of stock returns.