Colin brought up the following point:
Christian, it is extremely interesting to read your thoughts on neural nets. I’ve been looking at Neural Nets in a piece of software called Merchant of Venice – http://mov.sourceforge.net/ . Beyond that, what’s interesting is that as humans, we can look at current day situations and have an innate sense that history is repeating itself, yet there are these outlying situations that break those rules – like Amazon’s recent performance.
How to put that “sense” into code and avoid getting burned by the outliers is the real rub
I have looked at MOV and have to say that is the direction that I am moving in. I have used Genetic Programming (not Genetic Alogrithms) previously in another project and had some interesting results. From reading, “Fooled by Randomness”, and other sources the reason we have a sense of history repeating itself is because of the way we store information.
A computer stores a picture in terms of pixels, which makes for great pictures. Though pixels require an awful lot of room. Thus to save some brain space we convert pixels in vector graphics. The downside is that we loose details and two “similar” pictures give us the sense of history repeating itself.
Regarding “sense” into code, let me tell you about a funny situation that I am discovering with my AI, and indicator model. Let’s say I have two indicators A, and B.
- A: over timespan X is profitable, but over timespan Y is semi profitable, but over timespan Z not profitable.
- B: over timespan X is semi profitable, over timespan Y is not profitable, and over timespan Z is profitable.
My problem is that my indicators are not always profitable at all times. They have their moments of unprofitability. A simple answer would be to look at the situations where each indicator is unprofitable, and not use them. Ah, but if it only were so simple. Remember my previous blog entries about stochastics? Well I have found I cannot predict when an indicator will be profitable or unprofitable. And to put salt into the wound I cannot say if indicator A is unprofitable for timespan Z1 to Z2 then it will be unprofitable for timepsan Z3 to Z10.
So I am caught between a rock and a hard place. Of course some of you are saying, “well duh, that is what I am running into now.” The solution would be to use AI, and I have found out some very interesting things. The AI is always profitable, but not always more profitable than the indicators. For example I have found in one situation that a single indicator squeezed 50% out of a time period, whereas the AI could only squeeze 25%. Though the AI squeezed consistently a profit.
What I found interesting about my AI is that it seems to be no better than a good trader. The AI did not provide with me a magic wand to make millions. Here are some of my findings.
- AI most likely will not make you more money than a good trader.
- AI automates the process so that you don’t have to watch the tick screen.
- AI allows you to play multiple equities and lets you focus on the strategy.
- The indicators play a huge role. For example I used the indicator [price – MA] and even though I could make money the AI was unable to squeeze money out of the system. You need to rethink (as I have done) indicators that can be used for an automated trading system.
Now to the question of getting burnt. Colin, I don’t ever get burnt (ok once, but that made me smarten up). And it is not because I use stop limits, but because I delta neutral trade. I wish I was in the market with Amazon as I would have made a killing. For Yahoo I was in the market and made a nice chunk of change. But I think the point is that you can’t predict the market and trying to do so will work most of the time, until you get burnt with a stock like Amazon.
My advice is that you never ever directional trade because these “outliers” of sudden movement happen all too often. Delta neutral trading is the process of entering a trade using a combination of equities, and/or options. If you don’t know about this type of trading or the term delta neutral then get one of the two books; Options, Futures and Other Derivatives, and Option, Volatility, and Pricing. The maths are going to scare you, unless you happen to be a maths geek, and I advise a step by step approach, but if you are serious about trading these books are not optional.