Algorithmic Trading: A Practical Tale for Engineers

Re:Algorithmic Trading: A Practical Tale for Engineers by rismay on Tuesday August 27, 2013 @12:43PM (#6284542)

I posted this on topal too, but I want to get your opinions also:

I studied a bit of market theory in college and learned about channel trading. The Efficient Market Hypothesis is just that - a "Hypothesis" - not a law. Plenty of academics have argued against it but they don't teach this in undergrad courses so as not to confuse students (this is seriously the answer I was given when I asked my teachers). Some people like Nassim Taleb are calling for a BAN on teaching entry level finance courses to MBAs who go out and gamble vast amounts of money based on incorrect theories.

I always thought that algo trading would be a good fit for channel strategies since the strategy is recursive in nature.

Does anyone have any pointers on how to implement channel type of strategies (as opposed to Moving Average strategies)? I'd be willing to chat about deeper into this topic.

FYI: Some (old) research shows that Exponential MA strategies make more and even out perform buy and hold strategies without taking into account tax advantages. If you have ever read a "Random Walk Down Wall Street" and don't believe in this "crap," it's actually in the footnotes of the sources the author cites.

[ Parent | on HN ]

Re:Algorithmic Trading: A Practical Tale for Engineers by Choronzon on Tuesday August 27, 2013 @01:41PM (#6284935)

Channels are boundary conditions on an estimation of a price or a price itself,effectively a form of error bar. Your statistically looking at central tendencies and dispersion.

[ Parent | on HN ]

Re:Algorithmic Trading: A Practical Tale for Engineers by rismay on Tuesday August 27, 2013 @02:17PM (#6285158)

No. That's another trading method. I know what you are talking about there too, but that's not how I see the strategy working.

What I found in my research is that the statistical analysis being done on markets is an approximation of the underlying phenomena. Yes - central tendencies and dispersion work 90% of the time, but they hugely under price tail events. This is the basis of Nassim Taleb's trading strategy.

What I see really happening is a recursive channel pattern at different time scales. As the recursion plays out it "zooms in" on the correct price of a security and the central tendencies and dispersion can be analyzed through traditional statistics. However, when a market correction takes place it "zooms out" by stepping back through previous market cycles and then statistical models break down. It's very weird to explain. Essentially, I see the market being a predictable Fractal pattern. Check out Mandelbrot's work on Finance to get some insight. I just disagree with his (and Nassim's) conclusion that these can't be predicted.

I've always been interested in trying to back test my theory but I didn't have the technical chops and didn't know where to start looking.

Thanks for your thoughts.

[ Parent | on HN ]

Re:Algorithmic Trading: A Practical Tale for Engineers by Choronzon on Wednesday August 28, 2013 @02:39AM (#6288352)

Diregard Taleb,the man cant trade (buying OTM puts and hoping for the world to end is not a trading strategy.)You are looking at basically a three regime model,(standard statistical events,macro breakdown,and "dont know").The problem is "fractals" is not a decent way to distinguish the three states,so now you have 3 problems. For what its worth I think Mandelbrot was onto to something but its a long way away from a trading strategy.You could decompose movements over different frequencies ("Fractal Self Similarity) and test for some sort of state wise relationship between recent micro and macro movements.Im largely sure this will exist but extracting a leading indicator out of this may be difficult.

[ Parent | on HN ]

Re:Algorithmic Trading: A Practical Tale for Engineers by rismay on Monday September 09, 2013 @01:05PM (#6354619)

I know this is late: but you get what I'm talking about: You could decompose movements over different frequencies ("Fractal Self Similarity) and test for some sort of state wise relationship between recent micro and macro movements. That's where the channel patterns come in - they are self similar at different time scales. The rules are pretty well known, but getting them coded is a different story. I always find MACD strategies which are only approximations.

[ Parent | on HN ]