Trading systems of the past

[small dog]

I introduced myself in another thread, and here want to share what worked for me in the past. First two systems designed for South African ALSI (All Share) index. The third one - SPI in 2006.

1. End of day data. Entry - if close is higher than open + yesterday's range * 0.7 - buy on close. If close is lower than open - yesterday's close * 1.1 - sell on close. Trailing stop - yesterday's low for longs, high for shorts. You have to second guess the close somehow five - ten minutes before the end of trading, but it generally worked.

2. Larry William's intraday entry. Buy intraday on open + yesterday's range * multiplier. Opposite for shorts. Trailing stop - yesterday's low/high for longs/shorts.

Both of these worked well with overnight positions.

3. Opening range breakout. Buy/sell on high/low (plus couple off points) of the first 30 minutes after opening. Trailing stop on hourly bars. Worked reasonably well on SPI, though you can get whipped on some days, especially if the opening range is narrow. In those cases you could choose, either to keep taking trades - on the third go you would do well most of the time and would at least break even - or get out of the market for a day with a small loss.

 

[InsvestoBoy]

Hey @small dog thanks for posting these, good stuff.

There used to be a website "myforexdot" with a bunch of systems like this listed. You can still find it on archive.org

https://web.archive.org/web/20140207235910/http://myforexdot.org.uk/index-trading-systems.html

I think systems like this actually still work for periods depending on the microstructure of a given market regime. So you can often do things like, track the returns for such a system plus it's inverse system (i.e. short when the other one goes long and long when the other one goes short) and trade the signals for whichever is currently outperforming.

Here are the returns of 1000+ trades for a strategy which trades a single contract of ES long when daily RSI(2) trades below 20 and short above 70.




or check out the paper "MR Swing" which builds on this kind of system into something more refined

https://cssanalytics.wordpress.com/...eversion-and-swing-trading-in-market-regimes/

 

[tech/a]

Excellent nice and simple.

 

[small dog]

The thing is, all mechanical systems work for a while and then don't. The trick is to recognise when it happens. Another, very important point, is to follow the system and not try to second guess it.

I had two significant blowouts of my account. Not annihilations, but giving away lots of profits earned in the previous few months. One, I decided to second guess the system and ignored the spectacular V-reversal recovery after the 9/11 drop. I was bearish, kept selling in spite of my signals - which would work exceptionally well that time. I also kept increasing the size while taking one losing trade after another.

Another one - the system stopped working, but I persevered... And also increase d the size to make sure I give it back faster. Pretty funny now, not so at the time.

 

[IFocus]

The thing is, all mechanical systems work for a while and then don't.

Applies to all methods of trading, maybe one of the most important aspects to trading is to understand not only when your system is working or not but to understand the market you are trading.

Lots of examples on this site of traders using some sort of filter / method to keep them on the right side of the market or out of the market at the right times.

Skate and Perter2 / others discuss at length.

 

[tech/a]

The thing is, all mechanical systems work for a while and then don't. The trick is to recognise when it happens. Another, very important point, is to follow the system and not try to second guess it.

I had two significant blowouts of my account. Not annihilations, but giving away lots of profits earned in the previous few months. One, I decided to second guess the system and ignored the spectacular V-reversal recovery after the 9/11 drop. I was bearish, kept selling in spite of my signals - which would work exceptionally well that time. I also kept increasing the size while taking one losing trade after another.

Another one - the system stopped working, but I persevered... And also increase d the size to make sure I give it back faster. Pretty funny now, not so at the time.

Applies to all methods of trading, maybe one of the most important aspects to trading is to understand not only when your system is working or not but to understand the market you are trading.

Lots of examples on this site of traders using some sort of filter / method to keep them on the right side of the market or out of the market at the right times.

Skate and Perter2 / others discuss at length.

Systems are built on data
If the data is consistent you’ll be able to develop
Systems which perform well against test results

Its when the data alters or suffers outliers that
Systems collapse.
The premises or inputs may still be valid but not
On current data.

 

[MovingAverage]

Systems are built on data
If the data is consistent you’ll be able to develop
Systems which perform well against test results

Its when the data alters or suffers outliers that
Systems collapse.
The premises or inputs may still be valid but not
On current data.

This is exactly why OOS testing on as much uncorrelated data is so important. Unfortunately I don’t think a lot of system traders look at their system performance on proper OOS data, IMHO.

 

[small dog]

There are several key principle of systems testing that can confirm if the system is going to be robust.

First, it should be working over a wide range of parameters. For example, if you are testing moving average crossover and one combination is spectacular, couple are profitable and the rest are unimpressive this is not a good system.

Second, breaking the data into multiple random chunks and testing them separately. So if the results are sort of similar and most periods show profits it's a good sign. It's counterintuitive, but testing one large set of data increase the probability that the system "worked" on this data by pure chance.

 

[MovingAverage]

There are several key principle of systems testing that can confirm if the system is going to be robust.

First, it should be working over a wide range of parameters. For example, if you are testing moving average crossover and one combination is spectacular, couple are profitable and the rest are unimpressive this is not a good system.

Second, breaking the data into multiple random chunks and testing them separately. So if the results are sort of similar and most periods show profits it's a good sign. It's counterintuitive, but testing one large set of data increase the probability that the system "worked" on this data by pure chance.

I prefer to test on completely different markets instead of breaking up say the ASX into chunks as those chucks have a higher degree of correlation than say optimising on ASX then OS testing on NYSE or FTSE.

 

[qldfrog]

I prefer to test on completely different markets instead of breaking up say the ASX into chunks as those chucks have a higher degree of correlation than say optimising on ASX then OS testing on NYSE or FTSE.

Maybe you can find a fit all system, but I doubt;
A system good for asx would NOT be good for NSE, or even XAO vs XJO; behaviours both differ between markets and countries, and will too in time, psychological levels just that, do play a role, outside event be it elections etc
If I build and tune my systems for the XAO, I do not expect them to perform optimally elsewhere..
My uninformed view

 

[qldfrog]

Maybe you can find a fit all system, but I doubt;
A system good for asx would NOT be good for NSE, or even XAO vs XJO; behaviours both differ between markets and countries, and will too in time, psychological levels just that, do play a role, outside event be it elections etc
If I build and tune my systems for the XAO, I do not expect them to perform optimally elsewhere..
My uninformed view

just have to add I work on weekly systems so not intradays jumps etc which could be different after all