Not true IMO. If a system works only for one ticker it is probably due to selection bias.
You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
"Universality" of your trading system?
- Thread starter gmailer
- Start date
Not true IMO. If a system works only for one ticker it is probably due to selection bias.
Quote from gmailer:
Hello guys,
Do you think system should be good at all markets(or at least as many as possible) or can it be good only at one?
Must work at least on 2 -3 different markets. Systems that work in only one market can be the outcome of
Curve-fitting
Data mining bias
Survivorship bias
Data snooping bias
Quote from gmailer:
I have few trading systems designed for forex. But I have following problem - some of them look great on EUR/USD but completely suck on GBP/USD.
See above.
Quote from gmailer:
Should I worry about that? My intuition tells me that system should work good on many markets - this means it is not "tailored" for one market/curve. But my experience shows me that my systems tend to look different on different pairs. I use 5000 periods - intra.
You should worry. It is a sign of some problem. See above.
Quote from gmailer:
What is your opinion on that?
I had similar questions when I purchased PAL. I asked the developer to provide an option for the program to search for patterns that test profitable on several markets. They provided that and after running again my previous searches I found out that the results tested better in OOS. Now I only use patterns that are profitable on at least 2 different markets. This doesn't completely fix all problems but it is a step forward. Here are some examples for fx and etfs from their website so you can get an idea.
|Quote from jcl:
The Forex markets are rather similar. In my limited experience with trading systems so far, if a system works reliably with one currency, it should also work with most others, but might require different parameters.
]
We are talking about system that did not have parameters changed. It was like "break above HighestHigh of 100 periods". Tested on many markets and works great on EUR/USD...a little bit on USD/JPY and completely sucks.
I use very long periods although it is on 1h charts. This way during 8 years it had like 80 trades only. My assumption is that fewer trades system makes the less likely it is to be dependent on optimization.
Ok. So we have conflicting comments here? Some claim I should not worry, some claim I should worry that system does not work on all markets(as I wrote - with the same parameters for all, in each case. I did not check if it would work on other pairs if I change anything)
Here's a simple trick to check if the EUR/USD system indeed has an edge that only works with this currency pair. Trade the same system with the inverse price curve (USD/EUR). It should then also be profitable. If you instead get a very different result, you should probably worry.Quote from gmailer:
We are talking about system that did not have parameters changed. It was like "break above HighestHigh of 100 periods". Tested on many markets and works great on EUR/USD...a little bit on USD/JPY and completely sucks.
I use very long periods although it is on 1h charts. This way during 8 years it had like 80 trades only. My assumption is that fewer trades system makes the less likely it is to be dependent on optimization.
Ok. So we have conflicting comments here? Some claim I should not worry, some claim I should worry that system does not work on all markets(as I wrote - with the same parameters for all, in each case. I did not check if it would work on other pairs if I change anything)
Quote from jcl:
Here's a simple trick to check if the EUR/USD system indeed has an edge that only works with this currency pair. Trade the same system with the inverse price curve (USD/EUR). It should then also be profitable. If you instead get a very different result, you should probably worry.
Interesting test you propose.
My question is - if a system works on price series P[t], why should it work on 1/P[t] ?
What is the logic behind your assertion ? Can you prove it mathematically/logically/rationally/conceptually ?
Thanks a lot.
Quote from jcl:
Consider a symmetrical system, with an algo such as you described:
"break above HighestHigh of 100 periods"
"break below LowestLow of 100 periods"
This system would break both on p and 1/p.
That was example....in reality it makes EXIT on different condition - although as simple as in example. So for example Enter when break above HighestHigh100 and Exit below LowestLow100(from X periods ago, Ref function). So it is not true and complete REVERSAL as I sometimes have periods of "out of market" condition.
It's not necessary for a symmetric system to be always in the market, but it needs to trade long and short under the same conditions.
Strategies exploit market ineffectivities, such as trends or cycles. Trends and cycles are invariant under a 1/p transformation. Therefore a profitable strategy, if it is symmetrical in long and short trades and does not prefer a certain trend direction, should be still profitable on the inverse currency.
Another method, aside from inversing the prices, would be resampling them to generate slightly different bars. A profitable strategy must then also stay profitable. This also works when your system only places long trades. Such methods are not a sufficient, but normally a necessary condition for profitability of a strategy.
Strategies exploit market ineffectivities, such as trends or cycles. Trends and cycles are invariant under a 1/p transformation. Therefore a profitable strategy, if it is symmetrical in long and short trades and does not prefer a certain trend direction, should be still profitable on the inverse currency.
Another method, aside from inversing the prices, would be resampling them to generate slightly different bars. A profitable strategy must then also stay profitable. This also works when your system only places long trades. Such methods are not a sufficient, but normally a necessary condition for profitability of a strategy.
Quote from jcl:
Strategies exploit market ineffectivities, such as trends or cycles.
ATTENTION ATTENTION INVESTORS IN AAPL:
JCL JUST DECLARED THE TREND IN AAPL A MARKET INEFFECTIVITY.
It is not the ipad, the iphone and the millions of users worldwide. It is a market ineffectivity that causes the trend.
This must be the most hillarious and idiotic statement that was ever posted in ET.
Market ineffectivity? What the hell is that?
All technical trade systems are based on market inefficiencies. They are the very reason why a strategy can beat the market. There's an ongoing discussion among economists about in which degree the markets are efficient or inefficient. For some basics, look here: http://en.wikipedia.org/wiki/Efficient-market_hypothesis