| Abstract : | In this thesis, I present the theory behind transaction costs, trade execution and algorithmic trading and empirically examine the transaction costs of SilentSeas Group’s long-short equity hedge fund business. Concerning the theory, we comprehensively report the types of transaction costs, measures and determinants, the way that market participants create and consume liquidity, the orders and the limit order book, the buy-sell asymmetry as well as a well known approach, the implementation shortfall approach, of measuring total execution costs. Furthermore, I analyze popular algorithmic trading strategies, how orders interact in the limit order book, the way that pre- and post-trade equilibrium is established and the rationale behind optimal execution. I also summarize important studies on transaction costs and present the empirical findings. At first, the results suggest that commissions, which are equal to an average 5.1551 bps, and implicit transaction costs, which are equal to an average 39.1533 bps (VWAP cost), do present buy-sell asymmetry, hence sell orders have higher costs than buy orders. For instance, the average of VWAP cost is equal to -72.3893 bps for buys and 147.2049 bps for sells. The results by dividing the sample into high and low stock specific returns are similar concerning the buy-sell asymmetry. What is intriguing is that high movement stocks have lower explicit transaction costs on average, while low movement stocks have lower implicit transaction costs. Additionally, more principal has been traded on low movement stocks. The decomposition of implicit transaction costs into a VWAP cost component and a market movement cost component suggest that, based on separate panel regressions with fixed effects, transaction costs depend largely on VWAP cost and thus these costs can be approximated by VWAP cost. In addition, according to the panel regression analysis of implicit transaction costs, I find that transaction costs are affected by market capitalization, relative volume, inverse prior close, price momentum, VWAP, tap, perimeter and IS strategies, buy indicator and duration. From these factors only the buy indicator and duration are negatively related to transaction costs, while VWAP algorithmic trading strategy presents the highest costs. Also, transaction costs are not driven by return volatility and market index return, whereas the region dummies are omitted because of collinearity. From a forecasting perspective, I test out-of-sample the three models by running stepwise regressions and also include three naïve models which assume that transaction costs are always equal to the mean value of the realized costs. The findings suggest that, based on mean squared errors, the naïve models better predict costs, with the naïve VWAP cost model presenting the lowest respective value and thus predicting with better accuracy. However, these estimates should be interpreted with caution, as the mean values of the realized costs and the respective forecasted costs substantially differ and the standard deviations of the absolute forecast errors are large in magnitude in comparison with the mean values.
|
|---|
Copyright © 2013 Library AUEB. All rights reserved.