Originally posted by aphexcoil
There are two main types of systems in my opinion:
a) Generality Systems
b) Specific Systems
What is being suggested here, in my opinion, is a more general system. Suppose we researched historical data and found that, statistically speaking, Wednesdays had a 1.2% greater probability of closing higher than the open. We then statistically find that, between the times of 2:15 and 2:37, statistically the price will be greater at 2:37.
We then say to ourselves, "Hmm, I can make a better system by only buying on Wednesdays at 2:15!"
In effect, we are trying to capitalize on statistical edges from price-data which will, over a very long course of time, generally prove to approach an almost random distribution (Perhaps the past year Wednesday was the best day to buy, but over the course of 20 years, we might find that Wednesday is no better than any other day -- and if it is found to still be so, it may only be a .001% difference).
I've looked over some of Larry Williams systems in his book, "Daytrading Futures Online" and have to say a lot of his systems fall under a general system. There is nothing wrong with this in my opinion (as his performance speaks for itself), but again, over a large sampling of time and data, all generalities will eventually seek a random distribution.
....
However, in a specific system, we wouldn't concern ourselves so much with past events but a system that works directly from the data as it exists in the hear and now. Data from a day ago or today could certainly be used, but this type of system would not be concerned with past performance of variables existing over a long time period.
Simple systems that follow trends, if used correctly and with discipline, can and are quite profitable. This is the part of "keep it simple." Give a 7 year old a chart and he'll probably make the observation by noon that it is a "strong up day." So buy, right? Why not? Why are we going to fight the trend -- why are we building these complex systems that exist off past data that is no longer relevent in our timeframe of the present?
Instead, the best system is simply the one that uses whatever simple tools are necessary to trade into however major market trends exist together at any one point in time, holding that position until profitable, and selling it.
I have never backtested a system, nor will I ever do so with a computer. I look at each graph and I try to observe what things happen before, during and after a trend shift.
These things include:
a) Changes in price action (speed of price change, direction of price change, volatility of price action).
b) Changes in volume levels
c) Changes in volatility
d) Moving averages to understand general directional bias
Running simulations through a computer may find profitable results, but only if you then take the next step to let the computer trade that system and not yourself.
If you want a human to trade a system, you better have a human manually ascertain if that system can be profitable.
If you want a computer to trade a system, then use a computer to backtest a mechanical system, boot it up and go drink coffee.
Either approach could work, but don't mix and match these together or you are setting yourself up for dissappointment.
aphie
