kut2k2, I have to say that being an engineer did not help me much for trading the markets, at least initially. I was overconfident and lost money the first 2 years, but I guess this is the price to pay to learn, no way out. Anyway, I think as a systematic trader, one challenging aspect is to gain faith in your methodology/systems and this requires to stick to your plans in difficult times. I found and still find systematic trading very challenging : The learning curve is steep, I had to overcome psychological biases. Much I learned at school except maybe statistics and probabilities did not help me much : With no more then 10 Laws from Physics, you can explain the behavior of any Mechanical system. The financial markets are very different because they are dynamic, the laws evolve all the time because they are driven by people. The "noise" in the market data is huge and it is a real challenge too to discern between real patterns and spurious correlations. I have to say that I do not trade markets for money only, I like the challenges they provide and I find it easier to trade without money pressure. However, no doubt I measure my trading performance by the amount of money I make versus the amount of risk I take. Handle123 : I do not understand exactly what you mean by your message but I would like to hear what you have to say, because of your very long exposure to the markets. From what I understand, you seem to think that long time frames are more profitable than short time frames. Is this correct? This would be contrary to my current belief. I would expect the markets to become more efficient on the long time frames and to display short time inefficiency at higher frequency (ticks or minutes). Although the ultra high frequencies are probably very efficient too because of HFTs . My little experience is that there is some statistical edge to be found between HFTs (operating at tick by tick frequency) and mid/long term trader (> 1 day). ronblack: Thanks for the link to PAL website, interesting reading. Note however that by Price patterns, I do not restrict myself to chart patterns. I want to make a connection between an event and a response. The event could be anything : A technical indicator crossing a threshold, a break in correlation between some contracts, an external event etc.
Oracle's Data Miner has a good implementation and it's easy to use. One thing: The prep work to pull this off is horrendous. To do it right, you have to indicate the buy and sell points. It will then tell you the best price patterns using an SVM-based classification model. You'll need at least a million rows of data to get anywhere near to reliable results.
yup. you need to have some pretty strong infraestructure just to run the ML systems... and you also need to have a good set of tools to analyze the results, and use that analysis to ask the right questions to the AI... i've been working for the last year full time at setting up such a set of tools... right now I have a postgres DB (best option when you need something like Oracle's db, but not their invoices...) that allows me to churn through millions of ticks, but Im still working on the tools to analyze the results...
Understood but I think that price patterns have the least lag. If you are a longer-term trader maybe that is not important and a 50/200 sma cross can do the job as well but if timing matters, then most indicators generate lagged signals. Also note that price patterns are well-defined mathematical relationships of OHLC and are not related to chart patterns that in many cases involve discretion.