Predicting the performance of various sectors of industry in the United Kingdom (UK) has a range of benefits. Whether you are a member of the government working to guide economic policy or a private investor performing due diligence before making a decision on purchasing shares in a company; having an idea of how various companies will perform in the future provides essential insight to any decision-making process. In this report, the authors investigate how various economic indicators can be used to predict the performance of the UK luxury fashion sector.
A dataset of macroeconomic indicators is created with a monthly sampling frequency. It includes a “Sentiment” feature to gauge public opinion towards the UK government. The goal is to use these indicators to predict the UK’s economic health and, in turn, the willingness of consumers to spend on luxury goods.
During the analysis of the dataset, the authors chose the price-earnings ratio (PE) as the target variable. This is because it reflects the market’s opinion of the luxury fashion industry by considering both the share price and the company’s actual performance through its earnings.
The datasets features are prepared by de-meaning and normalizing the data to address variations in magnitude, and checking for stationarity using the Augmented Dickey-Fuller (ADF) test with differencing or log-norm transformations applied where needed.
Subsequently, EDA is conducted to identify significant features by examining their correlation with the target feature. The features with the strongest correlation are further investigated to determine the lag that yields the highest correlation. This comprehensive process involves normalization, stationarity testing, cyclic embedding, lag addition, and correlation analysis to ensure that the features are properly prepared for subsequent modeling and analysis.
Once the data preperation was completed, various time-series models were trained and tested using this dataset. These models were then compared against each other. The trained time-series models include:
- Last Observation Carried Forward (LOCF)
- Simple Moving Average (SMA)
- Exponential Moving Average (EMA)
- XGBoost
- Monte Carlo Simulation
- Prophet
Prophet was found to be the best performing model, achieving a MAPE of
The performance of each model is shown in the table below.
| Method | MAPE (%) |
|---|---|
| Last Observation Carried Forward (LOCF) | 104.64 |
| Simple Moving Average (SMA) | 95.93 |
| Exponential Moving Average (EMA) | 97.28 |
| XGBoost | 157 |
| Monte Carlo Simulation | 22.6 |
| Prophet | 11.36 |
The methodology, results and analysis are discussed in the final report.
Joel Edgar
Benjamin Sanati
Sayedur Khan
Adam Ali
Adwaith Kakkadath Palat