Timeseries Forecasting In Financial Services Industry Using Machine Learning Approaches

COMPANY is a share registrar for 30% of S&P500 and 49% of FTSE100 companies, who rely on COMPANY to provide secure, optimized and modern services including analytics and insights that could protect them from market manipulation and market anomalies. In an era of algorithmic insights and trading the best way for COMPANY to understand how it can protect its clients is for it to understand how ML algorithms can generate insights in this area.
This report aims to explore time series forecasting and anomaly detection techniques, which enable creation of predictive model for ‘beating’ the market. Such a model could later be further developed to identify use of “spoofing”, a technique where fake orders are placed to create an illusion of supply and demand, “wash trading”, where same security is bought and sold to create artificial activity, which can lead to price distortions, or “flash crashes”, caused by High-frequency trading algorithms, which can erode investor confidence.
Clustering algorithms can help identify abnormal trading patterns, SARIMA or ARIMA models can identify unusual spikes or drops, and supervised learning models can use historical data to create an understanding of what is “normal” to help detect suspicious activity. Regression models can discover patterns to help identify relationship between various features, such as the impact of trading volumes on share prices, while deep neural networks can tackle even the largest datasets and predict future anomalies, or anomalies in their inception, allowing COMPANY to be proactive rather than reactive. Insights gained from the above can be turned into automated alerts or further predictive analysis.
As this is a POC, publicly available data will be used throughout and will help demonstrate need for preprocessing, feature engineering and ways to tackle real-time data feeds.