I promised that I would do another post on my findings. And here it is.
In my last post regarding my experiment about long vs Long Short, I found that Long only breakout trend following system show systematically higher Sharpe for the intermediate run. Giving it a second thought, I went on to do more sensitivity analysis to check if such phenomena was true for other trend following strategies. As if such was true, it may be justified to long the futures market outright.
In the secondary experiment, I will use the Dual Moving average crossover system and the Bollinger band Breakout system.
Simulation parameters are the same compared to the previous test. 56 instrument across all sectors and 1% of equity.
The Dual Moving Average system has two parameters, Short term MA and the Long term MA. I classify the short term MA any where between 10-40 days while long term to be anywhere between 50-200. In the following test, I fixed the short term MA and stepped through the Long term MA in increments of 10 as I found through stepping that the performance were nearly all the same, regardless of how I incorporated the parameters. Again, my fitness function will be the Annual Sharpe.
From this simple test, we can see that the incorporating the short side into a MA strategy always improved Sharpe. I did additional test to allow only short trades and found that all (not typo) the stepped tests that traded short lost money. Initially, I found that this to be counter-intuitive. If the short side was a losing proposition, how can it improved the result when combined together? One explanation I felt was acceptable was just the underlying nature of the entire system. From the smoothing nature of the moving average, the number of bars and the magnitude of each bar required to reverse a signal is greater compared to a breakout system. Given this, we can also infer that the number of false breakout of MA strategy is less than a pure breakout strategy. Therefore the MA strategy with lower false signals and less trades (due to less whipsaws) will have a short side that is less volatile. Aggregating long short together will nevertheless improve performance. (In a follow up post, I would like to confirm this with additional tests as currently I am not near my personal laboratory !)
In this graph, I varied the days from 20 to 300 in increments of 10. I found the results to be confirming the last one. Nothing much can be said as a picture yields a thousand words. The conclusion is same.
All in all, it seems as if trading long vs long/short wholly depends on the underlying system. This is an acceptable conclusion since now I have statistics that confirms this. I hope the reader enjoyed this short experiment.