The theory categories of your algorithm strategies

Hi guys,

To inspire myself (and you), I would like to list the categories about algorithm strategies I can think of, and I then would like to hear in which category are your strategies, and why. And if you have any other categories to add, please share if you can.

The categories I can think of:

1, Indicator based strategies.
Such as moving average cross over, RSI over bought/sold, or any home-made indicators.

2, Price action.
Such as candle pattern recognization, Turtle break out system, etc.

3, Time series.
Such as ARMA, GARCH, etc.

4, Statistics based.
Such as, after statistics, it's found that Stock Index X likely goes up on Wednesday, then a strategy can buy open on each Wednesday.

5, Arbitrage.
Such as buy XAUUSD, sell XAGUSD, at the same time.

6, HFT.
I don't have interesting in the holy grail HFT because I don't have the condition to do it.

Any other categories?

I work on mostly indicator based strategies. It's easier than other categories. I tried "standard" indicators such as EMA, MACD, etc, and I also modify or create other indicators.
I tried price action but it's too difficult to automate PA because it's too contextual sensitive.
I also tried time series. I can only use existing libraries and I can't understand the background theory.
After all, I didn't find any reliable strategies to put live up to now.

So, what's the categories of your strategies, why, and how successful they are?
Note you don't need to disclose any details.
And any other categories to add to the list?

Thanks

 

HFT should not be a category.....it's just shorter term trading using price action, time series or indicators.

 

Make markets

 

You mean "market making" ?.....
and yes, that's another valid category.

 

I use moving average crossover, breakout rules and measures of contango - perhaps what you'd call 'indicators'.

I'd be interested in anyone using patterns to trade systematically, since it seems to me quite hard to write down a line of computer code to recognise these patterns in a systematic way.

GAT

 

Pattern coding is not that tough....it's just a lot of conditions that must be assigned and then combined:
Cond1=High>High[3] and Low>Low[3]
Cond2=Average((High+Low+Close)/3,4)>(High+Low+Close)/3
If Cond1 and Cond2 then Buy

That being said, I saw one programmer actually show his trades in realtime online using a pattern recognition approach. I think he called it Pattern Trader. How did it perform ? Horribly...it was always losing money.

 

Oh sure it's trivial to write code that recognises a systematic pattern. My suspicion is that most people using patterns aren't doing so systematically though. Probably the most succesfull ones aren't.

I've tried to properly test systematic patterns. So I've coded up things like you said, or to be precise more general versions, with parameters which state whether you have > or <, high or low, and so on. In sample they look great. Then I've tested them on a rolling out of sample basis. They were awful. Too many degrees of freedom.

GAT

 

Yep, that alone kills it.

 

Yeah I just finished refactoring and rewriting the pattern match part in my own algorithm trading system. I did some beginning level tests and it doesn't work.
I really doubt if any pattern match can work because they are just curve fitting.