The CBOE volatility index (VIX) is a well-known index that displays the implied volatility of the S&P500. Although the VIX itself is not tradable there are different products closely related to the VIX. This thesis examines intraday lead-lag effects of exchange-traded products (ETPs) that are related to this index, namely VXX, VIXY, TVIX, UVXY and SPY. More precisely, we try to identify one product that reacts to the movements of the other products. For this purpose, we introduce two machine learning approaches that are employed to predict trends of one single product by means of the others investigated in this thesis. To discover possible front running effects we analyze the machine learning models trained in this setting. On the one hand, we employ the boosting method XGBoost (Chen and Guestrin 2016) which is based on decision trees. On the other hand, we suggest an algorithm that relies on the concept of a self-attention mechanism. The structure of the second algorithm is similar to the Transformer network introduced by Vaswani et al. and is used as a reference for the performance of the XGBoost algorithm. Since the analysis of the models reveals irregular effects in the price discovery we can observe the trends of VIXY reacting to movements of the other products and thus, we come to the conclusion that a lead-lag effect occurs, where VIXY is the lagger and VXX is a strong leader. Within our analysis, we establish trading strategies based on the predictions of the machine learning models. Under the assumptions of moderate trading costs, these strategies trading VIXY are profitable.      «

The CBOE volatility index (VIX) is a well-known index that displays the implied volatility of the S&P500.; Although the VIX itself is not tradable there are different products closely related to the VIX. This thesis examines intraday lead-lag effects of exchange-traded products (ETPs) that are related to this index, namely VXX, VIXY, TVIX, UVXY and SPY. More precisely, we try to identify one product that reacts to the movements of the other products. For this purpose, we introduce two machine learning...     »