Binaries and Sources Update15 comments
Courses in options trading
Back in December see the original post here. We are going to see if their initial expectations still hold and we are also going to talk a little bit more about why I no longer use them for live trading. I want to take a bit more about the systems first. The strategies all have stoploss and takeprofit values, denominated as percentages of the ATR, and are programmed to risk a fixed percentage of equity on each trading signal. The systems follow a stop-and-reverse logic long signals cause a close and reverse if a short is open and vice versa and signals in the same direction trigger resets of the stoploss and takeprofit levels.
The picture above shows you the 25 year back-testing results for the strategies. It is worth noting that back in I only had access to 1M data, reason why the back-testing performance that we get now using the set shows that returns are not smooth across the whole testing range. Additionally returns cannot be considered very stable at all — even in the to data — as the correlation coefficient for this period is not above 0. Back when I built these systems I was not using stability as a criterion for evaluation as I used the MT4 backtester which does not include this type of statistics in its output so my optimizations never accounted for stability of returns.
Surprisingly the strategies do not collapse on all the data that was not included within the original design phase, a testament to the robustness behind some of the underlying principles like following momentum. You will also easily notice that the risk adjusted returns are not that great at all. The strategies mostly make a 1. This speaks about their compounding efficiency which is quite low due to their limited trading frequency and poor trade expectancy.
It is also evident where the design period is located, the graphs show the smoothest performance and least drawdown for all strategies during the time range, giving us a quite decent example of curve-fitting bias strategies failing to capture an overall general property of the market.
The 6 period was long enough to capture some sense of generality but failed to provide a concrete base. The RSI system plot for example shows how the optimization during this period probability led to an exaggeration of expected performance.
Combining the three in a portfolio increases risk adjusted returns significantly, although results are still too poor.
What is most interesting are the results after , when the strategies were partially traded live. The BB and CCI strategies seem to have performed acceptably during this period — at least generated some return — while the RSI strategy went almost directly into a drawdown period.
Drawing a line for the RSI system during the period — which is the longest most stable performance — we can see that the strategy is probably coming back from a period where performance was much stronger than would be expected for the exploited inefficiency. It is however clear that 6 years was a too short period for the creation of strategies for a single pair — something that I have confirmed repeatedly — and more data is needed to obtain a more general picture for a given trading system.
These strategies are now unacceptable to me from several perspectives, including risk adjusted returns, stability, standard deviation, compounding efficiency, etc. Right now I demand strategies to give highly stable performance during the entire data, with a very strong correlation coefficient on log balance vs time and a much lower historical risk.
The third image shows a comparison between this portfolio and one of the systems I use now a system created one year ago , you can see that this old portfolio is very inferior cumulative return axis is logarithmic!
However it is surprising that these old strategies were able to survive and actually profit, given the low quality that I now assign them according to what I have learned about strategy design and evaluation during the past 5 years. Thanks for the great post. Or is the improvement due to hedging by using the trading logic to other currency pairs or by using systems of multiple strategies?
I would say that the main differences are that the current strategies have much better ways to exit the market plus they exploit simpler price-action based inefficiencies.
These Atinalla systems had very primitive trade management skills. We also now have much more powerful and systematic methods to search for trading strategies while properly measuring data mining bias so all this generates a much better trading strategy output. All these years of experience have given us the tools to radically improve our trading. Glad you like the new posts: Mail will not be published required. Mechanical Forex Trading in the FX market using mechanical trading strategies.
Some algorithmic trading systems from Posted in Articles Tags: Why many academics are doing it all wrong. May 15, at Leave a Reply Click here to cancel reply.