1 (edited by Minch 2020-06-15 00:15:11)

Topic: Meaningful Strategies

My tendency so far in my EA journey has been to let the software do all the work and to squeeze all the strategies into a portfolio without checking each individual one. After all they work in backtest, why would I need to check them? However, after a random check of some strategies, I found that some strategies don't make sense.

As an example,

Long Entry : Fast MA Crosses above Slow MA & ADX is above level line 25
Long Exit   : MFI Crosses the Level Line upwards.

I realise that the software is simply looking for a correlation between the optimum settings/parametrisation over the backtest period and that, in this individual case, the entry and exit results work as intended. But you will likely never see an exit strategy like this in a live strategy - surely the MFI should be going lower to reflect the lower volume.

Sure, the MFI level isnt going to always behave in a predictable way and it may oscillate when the price is falling (which is the reason of using multiple confirmation indicators - not one of them is "perfect") - but the general rule of thumb for something like the MFI indicator is that in a falling market, the price, volume, trend etc will be decreasing and so will the indicator. Im not trying to open a discussing on the MFI indicator - Im seeing some contradictory signals being generated and Im wondering whether I need to do anything about it before putting these into a portfolio. Maybe its normal, but Im a newbie so I reserve the right to ask dumb questions.

Entry : Fast MA Cross Slow MA downward
Exit : Fast MA Cross Slow MA downward

Entry : Fast MA crossed above slow MA and;
Exit : Fast MA was lower than the slow MA

Although the first example with the MFI is not outside the bounds of reality, can someone please explain how the Fast MA can be both above and below the Slow MA at the same time?

There seem to be therefore 2 types of problem: 1) where the technical indicators are contradicting the assumed behaviour of that technical indictor and 2) where the indicators selected are contradicting each other. And since there doesnt look to be anything built in to stop these strategies being created, they are just as likely to crop up as not. There are three ways around this that spring to mind.

1) manually check the strategies for inconsistencies before putting them in a collection.
2) just ignore them or
3) how would it be possible to build in an integrity check which could avoid this situation, e.g. some additional rules for the software to follow, e.g.

- Entry and Exit rules cannot be the same
- Entry and Exit rules should be opposite if the same indicator is used (ie fast crosses above for entry, crosses below for exit)
- Entry rules should make sense, e.g. long entry should be MA rising (not MA falling)

Would appreciate some feedback, do I need to be worried or just checking my strategies more often?


Post's attachments

MA Cross.jpg 796.43 kb, file has never been downloaded. 

You don't have the permssions to download the attachments of this post.

2 (edited by ats118765 2020-06-15 06:47:16)

Re: Meaningful Strategies

Minch wrote:

Would appreciate some feedback, do I need to be worried or just checking my strategies more often?

I was just having a similar discussion about this with Zikka.

Does the fact that the result defies intuitive sense mean that the solution is invalid?

IMO it is a possible symptom arising simply from over-fitting to noise....but it does not necessarily mean that this is the case. In fact these unique weird solutions might be really handy for those looking for a unique edge in the market that is unknown to competitors.

The problem lies however in developing confidence in your strategy that doesn't make intuitive sense in accordance with how you understand the market to behave. When that non-intuitive solution enters a very natural drawdown in the future, due to the lack of confidence you have in understanding these unusual rules....you might be inclined to assume it is broke and bin it.

On the other hand, if I understand my model and how it behaves, I tend to be much more lenient when it enters drawdowns and put it down to 'a very natural result' arising from the way I understand markets to behave from time to time.

There are billions of possible solutions that can be generated by data mining methods so quickly these days.....so  I tend to reduce the subset of possible solutions to those that:
1) Make intuitive sense so I can see and understand the possible 'causative' reasons for why they pass or fail the uncertain future test. I also like to ensure they all have divergent properties and positive skew to provide robustness at the portfolio level;
2) Are very simple in design allowing for more degrees of freedom when applied to an uncertain future. I far prefer a linear equation of simplicity than a polynomial function of complexity. A complex model may be able to more precisely define a historic market condition but less capable of navigating a future condition that varies from this exact historic condition; and
3) Simply assist the greater portfolio. There is more edge made available in the portfolio design phase than focusing on any single system.  A system is simply a means to an end of a more comprehensive robust solution of performers that can navigate the Law of Large numbers.

I therefore would bin the unusual non-intuitive result.....but I could very well be wrong.  I tend to be wrong on many things but occasionally right when it really matters. That is the problem when being a trend trader at heart  :-)

Diversification and risk-weighted returns is what this game is about

Re: Meaningful Strategies

Thanks, Rich - I am glad I am not the only one who has noticed this.

I agree with what you say - when they say that 99% of traders lose money over the long term, if we were to look at the strategies they are using, you'd probably see the usual suspects which are being used straight out of the box, e.g. MACD on standard settings; MA on standard settings etc (of course, other factors like emotion, position sizing, risk management lead to their losses as well) but - having the edge is one advantage in assembling a collection of successful strategies. Randomising and taking on the risk of having a true random selection of strategies means there will be, as you say, billions of probabilities in there. If one was to limit the selection process, you might also be discounting some decent / alternative approaches.

And I must remember that we are looking at the performance of the collection and not the individual strategies too. So I will indeed go through the collection and remove the obviously contradictory ones but over time, they might prove to be interesting as well.