Should a system work across all stocks?

[oscar74]

I've been playing with metastock for a few months now - and - having a slight knack for programming - haven't found it too difficult to come up with a handful of systems that appear to have some success.

I'm well aware of the pitfalls of backtesting, but: should a system that works well on a few stocks, be thrown out just because they don't work on all stocks and therefore would lose money if applied to the market as a whole?

If a system works on stocks A, B and C, but not very well on X, Y and Z be used to trade just A, B and C? or am I kidding myself?

any thoughts appreciated
Oscar74

 

[Pippppin]

a good trading system will work on all financial instruments, not just stocks. The more a trader has to narrow down the number of markets he or she thinks the system applies to, the weaker the system

 

[dcraig1]

a good trading system will work on all financial instruments, not just stocks. The more a trader has to narrow down the number of markets he or she thinks the system applies to, the weaker the system

Why should this be so ? I very much doubt it. For example why should currency futures trade the same as stock index futures ? In fact they do not.

A successfull system for trading stock index futures may use a number of inputs that are not even available in the forex market - eg DOM, things like VIX, TRIN, TICK, BID/ASK ratios, perhaps the behaviour of related options etc etc.

 

[blackcab]

If a system works on stocks A, B and C, but not very well on X, Y and Z be used to trade just A, B and C? or am I kidding myself?

It depends what underlying behaviour and/or inter-stock or inter-market correlations the system has identified. Maybe A/B/C are seasonal, or influenced by a commodity price, or are tracked by long-only funds, or are disproportionately program traded, and X/Y/Z aren't. Or maybe they reacted to 'important' S/R in your testing period that X/Y/Z lacked. Or maybe you've just over-fitted your system to A/B/C. Whatever, if it works in backtesting and out-of-sample testing and, optionally, forward testing, then it works, regardless of what else it works on.

 

[garethb]

If a system works on A but not on X can you identify, in generic terms, and without the benefit of hindsight which markets it works in and which it doesn't? If not, what is to stop A trading in the future like X did in the past?

I think it is best to assume that the system's future performance will be, at best, its average performance accross a broad selection of stocks rather than match the best ones.

Suppose you took 100 coins and tested them by tossing each 10 times. If one coin came up heads 8 times out of 10 in that test would you expect it to reproduce that performance in the future?

Good luck in your quest

Gareth

 

[pssonice]

yes, as long as the free float is larger enuf

.

 

[Pippppin]

dcraig1

currencies, stocks and commodities exhibit the same patterns of supply and demand. It has nothing to do with "inputs"

 

[dcraig1]

dcraig1

currencies, stocks and commodities exhibit the same patterns of supply and demand. It has nothing to do with "inputs"

Yes, but supply and demand are abstractions, you cannot directly measure them. A trading system needs measurable inputs, from which you can reach conclusions about supply and demand if that is the framework from which you view the market.

 

[Pippppin]

i agree

 

[the blades]

I've been playing with metastock for a few months now - and - having a slight knack for programming - haven't found it too difficult to come up with a handful of systems that appear to have some success.

I'm well aware of the pitfalls of backtesting, but: should a system that works well on a few stocks, be thrown out just because they don't work on all stocks and therefore would lose money if applied to the market as a whole?

If a system works on stocks A, B and C, but not very well on X, Y and Z be used to trade just A, B and C? or am I kidding myself?

any thoughts appreciated
Oscar74

Hello Oscar,

I believe the more markets (or stocks) the system works in, the less chance that you've "curve fitted" your system.

I don't believe it's a pre-requisite that it works in all markets, for the reasons suggested above.

I operate a system, relatively successfully, that works in limited markets.

Cheers,
UTB

 

[damianoakley]

I've been playing with metastock for a few months now - and - having a slight knack for programming - haven't found it too difficult to come up with a handful of systems that appear to have some success.

I'm well aware of the pitfalls of backtesting, but: should a system that works well on a few stocks, be thrown out just because they don't work on all stocks and therefore would lose money if applied to the market as a whole?

If a system works on stocks A, B and C, but not very well on X, Y and Z be used to trade just A, B and C? or am I kidding myself?

any thoughts appreciated
Oscar74

Hi Oscar,

Being a trader of stocks, I would say that no system will work across all stocks at a given time. The trick is to try to determine when your method is best traded on a particular stock at a paticular time.


Thanks

Damian

 

[Pippppin]

I have a system I have developed over ten years.......it works across all stocks, commodities. In short, all financial instruments

 

[oscar74]

Nice to see a discussion going re my question. Thanks for all the replies!

 

[the blades]

I have a system I have developed over ten years.......it works across all stocks, commodities. In short, all financial instruments

Good for you - care to share it?

Seriously, I don't think this scenario was ever in debate - of course the most robust systems will work in more markets. But the question was - does it necessarily have to? - and I believe the answer is no.

Cheers,
UTB