There are lots of ways to filter a strategy (when it is good, when it is bad), google it and numerous methods and ways of evaluating strategy character will come up along with the needed math in terms of dependence. But to answer you, yes, the equity curve of the underlying strategy is indicative of it's near term performance, especially with a strategy with a high dependence. Here's a simple way to look at it. First separate your strategies as long strategies and short strategies. Now take the long component of your strategy and plot it. Clearly it will do better when the underlying market is in an uptrend. If you mix long and short it gets very messy and is nearly impossible to determine switch points. Something else to consider is that a strategy that trades very infrequently (once per week) will need many, many weeks of consistent market behavior to give you good points to switch, and long periods of good performance. Plus you need the underlying market to go into a consistent trend, and hold. So the higher the frequency of your strategy, the more granularity you have to switch it on and off, and the closer your strategy performance will be an indication to the current market conditions. Now, we don't even care if it's a winning strategy or not. The only thing we care about is that it has VERY CLEAR times of doing good, and VERY CLEAR times of doing poorly. We want it to die a death when conditions aren't right. This way we can more easily tell when to turn it off. Here's an example to play with, try a long only moving average crossover and use very small stops, and very large targets. Unless that strategy finds itself a clean, clear uptrend, it's going to die, and clearly die. But when it locks into a trend, it's going to gobble every bit of profit it can. That subsequent equity curve is going to go down, down, down until it finds an uptrend, then it will shoot up. Clear switch points. Have fun, it's my most favorite thing in the world, all this stuff. I like to think we have a collection of the best market direction indicators (our strategies) : - ) Trader 5of7 @ TheCollectiveFX.com
It's all clear, thank you. I will remember these new ideas you shared with me. In the meanwhile the poll has reached 75 votes. What strikes me is that there's a bunch of people investing less than 100k and a bunch of people investing more than 1 million, but very few people in between. It's surprising.
I'm curious, do you still use TS or have you moved to another system... would like to hear if you have moved on as I have been looking at other systems on and off for about a year. Looking for something more robust just am comfortable at TS.
I use TS with NT, using macros to send orders and commands to IB TWS. As far as I can see, TS's live data feed is unfiltered so the price action is pure. I can live with the snapshot data feed that IB provides, but NT does not provide macros within ninjascript so I stick with TS as the data provider. Only on occasion does the TS feed stall or go down. rt
Thanks for that ..... What are your thoughts on Genetic Optimization and Walk Forward Optimizers ... was trolling through the TS Forums and came across an old one on "The Grail" software which from the posts seems to have gone out of business. Do you use them and if so what particular product do you find most useful with the TS platform. Apprec
I have never had a need for a genetic optimizer. The premise behind genetic optimization is to find optimal parameter values when a brute force search is numerically intractable. This is only the case for an optimization over many variables. Genetic or brute force, I have come to the conclusion that the robustness of a system quickly declines as more parameters are introduced. This is the classic problem of curve fitting. The developer is not designing for robustness, but rather is looking for the global optimum of the performance surface using historical data alone. WFO is a different case, however. What is univerally true is that backtest performance should be calculated on out-of-sample data. However, what most people neglect are how in-sample and out-of-sample data lengths influence the equity curve over time. The WFO software tool allows the user to define in and out of sample lengths. In addition, it can perform a brute-force search for the best combination of lengths, based on user-defined optimization criteria. Some combos produce horrible results, some produce very good results. The important aspect is to have a smooth, continuous performance surface that doesn't have pockets or surface spikes when profiling the sample length performances. I use no more than 2 parameters in my own strategies. 3 and above is numerically intractable and risky to trade IMO. You can think of the strategy as not have individual parameters, but rather individual ranges of parameters. The chosen value in each range changes over time, at a frequency dictated by the out-of-sample length. So the strategy itself is adaptive, and the optimization actually is done on the lengths, not the parameter values themselves. rt
Thank you for the prompt response. I have also found that beyond the two parameter approach (which is what I use) the system tends to break down. I had looked at the Genetic Optimizer on the 8.5 platform but have fallen back on the exhaustive as my tests are a relatively small set. I must say its funny (in the ironic sense) to hear someone else saying the same things about testing and optimizing that I have found out vis a vis in sample out of sample data lengths.
Finding a formula for auto trading isn't easy! Thought I had one that backtested great but found out my using the open open parameter should be close close and 1 minute check intervals used and the results are now horrible. May have to go manual in the real world for now. The tech told me about folks who don't call them first and start auto trading to find hundreds of trades executed in a day or two. Sometimes a hundred in an hour. He said the calls are all painful. Seeing how incorrect backtesting parameters can lead to great results and finding out in reality it's the opposite, I feel their pain. I now wonder if this quest is a waste of time or if there really is a formula than can be consistently profitable? Now I'm totally off track...back to the drawing board.
I have created a thread to demonstrate this very thing. I created a small account ($75,000 USD), and started trading with 1 contract to illustrate the progression of my ATS. You might want to take a look before you jump to conclusions that automated trading is not possible. rt
I strongly disagree as well. There are indeed strategies that are profitable. Of course each equity-curve profile will require differing money-management to absorb drawdowns for each profile. But there ARE profitable autotrading strategies. We live and die by a whole slew of them everyday. Trader 5of7 @ TheCollectiveFX.com