Optimisation Pro's Con's

[NEILM16]

I would like to know everyones views on optimisation!?I have been sent emails from updata recently saying how amazing there ta2 programme is as it can optimize allsorts of technical indicators from new highs to macd!Now i have to say i have heard many things over the years like the market breaths and moves in waves cycles and you need to get in touch with the market to understand what is likely to happen next. I say likely as we all know you cant be right everytime!
However optimisation by its very nature designs a set of rules around a share/index and creates the most profitable set of figures it can!
So we then move into backtesting !How can we learn to trade without the past being relevant to future moves!Traders who have made a fortune have strategies that have worked and many still do.Is optimisation the solution?Well my feeling is im not sure but what it does do is give you alot to ponder!There will be many who will immediately dismiss it and say no way!! But would i would say is have they really honestly looked at it?Im still learning and want to learn more.If i believe the idea that every market share index is different and moves on its own cycle wave!Then if theres a way to measure that maybe optimisation is it!

 

[Grey1]

Few Points :--

The updata team are junk of amateurs.. .. I have known them since 1999..

When they talk about optimisation what are they talking about ? Optimisation of what ? TA indicators settings, over all strategy, exit, overall return , there are many many parameters which are bad to optimise such as Return on account based on exit ..

Send them an email and ask them what they mean and we shall discuss the theoretical as well as practical pro's and cons...

 

[barjon]

Neilm16

optimisation of ma isn't bad but often it's just a way of
rationalising the past movement of the particular instrument.
when backtesting split the data in two - optimise the first half
and see if it works on the second.

 

[Trader333]

If you ever intend to use optimisation of a strategy to determine future trades it should be the last thing that is done. In other words your strategy should be profitable without any optimisation first and then only optimised as a final step.

Too many of these "too good to be true" methods and systems are just curve fitted optimisations to make past trading appear profitable and then fail when tested in real time.

I know this because I have conducted numerous backtests on many different strategies and found that even hopeless systems could be made to appear profitable using optimisation techniques.



Paul

 

[NEILM16]

Hi Trader 333
Thanks for your comments!What you have said u see interests me lets take an example lets say for arguments sake you like a ma crossover system and you choose to trade the FTSE using a 25 and 12 crossover system but it makes a loss.Do u now give up on any ma crossover system or do u optimise?Lets say u do and u find the most profitable combination is say 18 and 9 do u trade that? Or do u say well the past doesnt mean that it will happen in the future and therefore not bother?

 

[DaxTrader]

Optimisation

Agree with you T 333

Jack Schwager devotes a chapter to the myth in his TA book.

If you mix and match numerous strategies over a given time frame (easy with Tradestation), you can come out with the best system (to fit the past) in the world....

Also, some of the variables could be a coincidence.

EG you could find that fitted in with some system that if you went long on the first day of the month, or when it rained more than an inch, for the last 2 years, it made you a ton of cash.

But would you want to trade off it going forward ?

This was discussed in a Market Wizards interview - having to discard data that has no real bearing

 

[Grey1]

Trader333,

Well said ,, This is an overall optimization of strategies.. Now a days we have another claim to deceive traders ... They know their curve fitting trick has no listeners so they go on by saying we are optimizing the best parameters for your TA indicators... In another word our software is going to tell you what parameter to use in your MA or RSI or CCI for optimum performance..

There are other performance optimizations based on pure manipulation of money and risk managment parameters and it works great on paper and in universities but could you see the face of a trader who has just made twice more profit than his stop loss level and the system tells him you must not close or you are violating the system rules .. Jesus

This is why I thought one should not just read and assume these guys know or want to tell you what they are optimizing .. specially the Updata team ..

Pick up the phone and speak to updata them .. you soon understand what I am talking about

 

[wysinawyg]

lets say for arguments sake you like a ma crossover system and you choose to trade the FTSE using a 25 and 12 crossover system but it makes a loss.Do u now give up on any ma crossover system or do u optimise?Lets say u do and u find the most profitable combination is say 18 and 9 do u trade that? Or do u say well the past doesnt mean that it will happen in the future and therefore not bother?

I think one of the general suggestions is that as a minimum you should make sure that the 18/9 in the example is actually optimal rather than the single profitable. In the above example if you found that 17/9, 19/9, 18/8 and 18/10 crossovers all made losses and 18/9 made a good profit, I personally would have no confidence whatsoever in the 18/9 results being anything other than a statistical fluke.

wysi

 

[Trader333]

This is a difficult question to answer as I know others will have different views on this so I will give mine but accept that it is just my view.

The past "MAY" have some influence on the future but others have also said that approaches that proved profitable in the past are no longer so. If you are trying a MA crossover approach and it doesnt work in its raw form then optimising it is not a good idea.

You will also find research from the likes of Sripophilist saying that the shorter the timeframe the more random the price movement. Now I knew this a while ago and so decided to develop a method that used weekly charts as the timeframe. I posted the results elsewhere on the board showing that profits of £2.8M could have been made over 17 years.

Would this work today ? well I wouldnt trade it but I could always market myself as a new VS or DW with this backfitted system which would probably generate me more than £2.8M if I used the historical techniques used by VS and DW but then optimised to suit my style. Do you think it would work as well for me as it has for them ? Just joking of course


Paul

 

[NEILM16]

Ok people so to sum up so far optimisation should not be attempted as the results are fitted around past performance!
The question i now raise if this is the case where do u go from here?Do we avoid TA indicators altogether?I see many traders talk about using patterns and looking for the same patterns repeating themselves.I also here people rubbishing Moving averages then i see ma crossover system with a few tweaks here and there!!If u just use price action what do u base your trades on?There must be some sort of methodology or is everyone coming down on the side of random markets in which case they are not tradeable!!??

 

[turtle trader]

As T333 says - if you're gonna optimize, leave it until you're sure you've got a system that works anyway. Then optimize it over a certain period (say 3 years for daily data), then run the same opt for the previous 3 yrs & then the 3 yrs before that. If all 3 tests give similar results then you might be onto something, if not then you've demonstrated that you may as well use random MAs in future.

If you do get similar results over the 3 tests then you'll probably get a cluster of of similar MAs giving good (not necessarily the best) results each time. You may as well pick whichever you feel most comfortable with from the cluster, since they are likely to swap around every time you test a different period.

I wonder whether optimizing is of any tangible benefit to system writing, but if it helps you feel more comfortable with your system then that's probably enough to make it worth while.

Hope that makes sense
Turtle

 

[mmillar]

I don't think it is that you shouldn't optimise, it is that you should be aware of the pitfalls of optimisation. Using the example from above, if ...

17/9 loses money
18/9 loses money
19/9 makes money
20/9 loses money
21/9 loses money

...then clearly optimising your results to use 19/9 is a bad way to go.

However, if...

17/9 makes £1000 profit
18/9 makes £2000 profit
19/9 makes £3000 profit
20/9 makes £2000 profit
21/9 makes £1000 profit

...then optimising to 19/9 is a good idea (probably)

 

[Rognvald]

Hi

I am still an Updata customer until my sub runs out (although I am no longer using the software) and I have the TA Optimisation program.

I have to say that there may be some merit in some of the arguments put forward for optimisation but I really don't have the time or interest to run optimisations on bundles of shares in an attempt to discover the secrets of the universe. I think the concept probably has limited value in the context of intraday futures trading which is my present area of interest

I have attended 2 Updata TA seminars. In a conversation at one of thse Jeremy du Plessis - Updata's TA man - told me that he had spent 2 yrs of his life some time ago on optimisation and found it was not the holy grail!.

Updata are good on marketing and eventually will probably have an excellent product if they dont run out of money and customers first. They presumably think that there are people prepared to pay for the optimising facility. If optimising is your fancy then fair enough. I don't want to spend another 2/3 years beta testing it.I would rather Updata spent time/money on other aspects but that is not what they are about just now.

Sorry if this is more of an Updata than an optimising post

 

[NEILM16]

As T333 says - if you're gonna optimize, leave it until you're sure you've got a system that works anyway. Then optimize it over a certain period (say 3 years for daily data), then run the same opt for the previous 3 yrs & then the 3 yrs before that. If all 3 tests give similar results then you might be onto something, if not then you've demonstrated that you may as well use random MAs in future.

Thanks for all your comments this really seems to be causing some good discussion although noone has yet said how they trade themselves are you day traders? end of day?If so what sort of method do u use or recommend?
If you listen to many they will say that ba market changes over a period of time so im interested that if it works back over 3 years you expect it to work back even further?Now of course if that is the case and it works back over many years theres no reason why it wont work in the future?!

 

[turtle trader]

I like the idea of system trading (takes the emotion out of decision making & frees up your time) and hope that that's what i'll end up relying on, although I suspect i'll always spend a certain amount of time looking at charts, trying to find trends & patterns to trade coz I enjoy that challenge. Upto now though, i've not managed to write a system that i'm completely happy with. I started trying to write systems for trading stocks, but decided that you've then got to decide which type of stock is best for your system (I couldn't find a system that worked across the board). This defeats the purpose for me since I want my system to make all the decisions, so I moved onto indeces.
I managed to write a couple of systems that work well on several different indexes but felt uncomfortable with the stop loss that was required to make them work. They were also EOD systems that held positions for upto several weeks & as I became more active as a trader I became less happy with those timeframes.

Most of my trading now is intraday FX trading, as it works well with basic TA. Unfortunately I don't yet have the facility to system test this type of data, so for now at least i'll continue to stare at charts all day, wondering whether i'm going to make the right decisions.

cheers
turtle

 

[Grey1]

IMHO The best indicator to tackle the market is Intuition .. PAST IS PAST , GONE , FINITO .. NO MORE EXIST..
( since this is a TA board and other traders love their MA then I will try contribute every so often on these indicators but my feelings are very clear on past/Future correlation)

We trade the market as it happens and as it un folds ,... Look at some of my posts on VWAP .. look into L2 , as well as risk management …

Saying that if you are not still convinced then choose a strategy , back test it over 5 DAYS ( yes DAYS AND NOT YEARS ) REMEMBER YOU WANT THE SYSTEM TO PEFORM INTRA DAY AND NOT OVER THE NEXT 10 YEARS OR SO.. YOU ARE A DAY TRADER ) ago , then add a bit more data say 12 DAYS ago then add more data say 18 DAY to it and look for the performance.
TRY TO BREAK THE STRATEGY DOWN by adding out of sample data..

Few other tips:--

1) Back testing is not an easy task…. some knowledge is needed to back test a strategy correctly .. for example how to set the exit parameters as well as the entry to have a tolerable risk level .. NO GOOD TO HAVE A RANDOM ENTRY and OPTIMSE THE EXIT for strategy to perform …

2) Try to design a system to have high win/ loss ratio strategy ( the number of wins are much higher than losses but smaller amount ) than a one BIG WIN STRATEGY .. The latter strategy is very damaging psychologically..

 

[NEILM16]

Thanks for your comments.It seems that those who are posting are day traders what about End of day traders or medium term traders would any like to comment on systems or optimisation?