The dynamic trading strategies of Gârleanu and Pedersen (2013) are being evaluated by their insample and out-of-sample performance for respectively 75% and 25% of the data. We adopt two different investor perspectives using two different datasets: Commodity futures and ETFs tracking commodities. Gârleanu and Pedersen (2013) focus on the implementation of transaction cost, why predictable returns must be taken into account. We test three Gârleanu & Pedersen (2013) strategies 1) the Dynamic Markowitz, which is optimal in the absence of transaction costs, 2) the optimal strategy when including transaction costs, and 3) the Static portfolios where we trade at a fixed rate towards the Dynamic Markowitz portfolio. We include two single-period portfolios as benchmarks for the portfolios proposed by Gârleanu & Pedersen. We measure the performance of the strategies, by applying the Sharpe ratio, maximum drawdown and Calmar ratio before and after transaction costs. We find that the Dynamic Markowitz is the optimal trading strategy in absence of transaction costs. It performs well in terms of gross Sharpe ratio both in-sample and out-of-sample, but it incurs substantial drawdowns which affect the Calmar ratio negatively. Including transaction costs, the Dynamic Markowitz performs poorly, since it trades fully towards the optimum, thereby incurring a substantial amount of transaction costs. Assessing the optimal portfolio when accounting for transaction costs, it performs well both insample and out-of-sample. However, it has a lower performance compared to the Dynamic Markowitz in terms of gross Sharpe ratio, and does not have as pronounced drawdowns. This is due to the portfolio only trading partially towards the aim and scaling down predictions by meanreversion speeds. Taking transaction costs into account, we find that the portfolio out-performs the Dynamic Markowitz on all performance measures, both in-sample and out-of-sample. We find that the Static portfolios are not optimal and, in some instances, suffer from non-linear effects. Finally, our analysis lack statistically significant return predicting coefficients, and there is little evidence of predictability in ACF plots. However, the strategies perform well, which indicates that they do not only hunt white noise.
|Educations||MSc in Applied Economics and Finance, (Graduate Programme) Final Thesis|
|Number of pages||153|