When I first heard of systems trading, I thought that it wouldn't be viable because of the dynamics of the market. When a system's optimal parameters change to a large enough extent, the system will yield different results, equal to or less than the maximum profitability of the system, which is why all systems have periods of sustained profits and sustained losses. The sustainability of profits and losses tells me that the changes in a system's parameters are directional (changes are in the same direction for a given set of bars), and therefore can be patternized or correlated. But then something occured to me: "In order to develop a system that will be effective all the time, adapt it's parameters to the current market conditions" Thus, designing a system to change with the market conditions will always present you with the most profitable trades to take. How is this done? First, you must realize that new data is being presented every business day, whether it is in minute, hourly... etc. This in itself gives the neural net a chance to 'learn' from its mistakes. Note: this 'learning' should be restricted to a rolling range, such as a rolling 2 month database. Forward step through this range to ensure that the system only uses parameter values that are relavent in today's market. But, some events will happen less frequently, so for those, a longer learning range should be used. The optimal values for these ranges can be defined by the neural net. Decide on what time frame to use, and gather some previous data to build the basis of your neural network (eventually, all of the main time frames 1,2,5,10,15,30... will be correlated to each other. Develop a database and algorithm that will mathematically define how any 3 consecutive bars relate to each other. I chose '3' for this value because it was suggested previously on this thread, but I will eventually have the neural net decide what is optimal. Develop a second database that will define what will happen in the next 'X' amount of bars after a given set of 3 consecutive bars are encountered. 'X' will be defined by the neural network as this parameter will adapt itself to find the optimum amount of data to use. Note: Both databases can be further broken down in a tree fashion so data can be retrieved faster. If the program encounters a new variety of a 3 bar combination, then add new entries to the first and second databases. You also want to add another indicator in the neural net that determines the correlation of price movements and patterns with which side of resistance you are on at many time frames. Now, with all patterns and results mathematically defined, set up the neural net to define what the best current price target and stop are. These parameters also change over time. Finally, the program can calculate the absolute probabilities that a certian event will happen given a known set of 3 bars. You've hit GOLD! Not all combinations will predict certian events with satisfactory probability. In fact, most will be split among hundreds, if not thousands of different probabilities, each taking it's piece of the 100% pie. But every once in a while, you will come across a particular pattern that will produce a given result with high confidence and probability. I believe that the degree of accuracy is related to the method one chooses to define the 3-bar set. There is a trade off when designing your method because the more loosely you define your sets, the more garbage will get sucked up into the overall probability. Anyways, this is something i've been researching for the past 4 months, and I plan on building a prototype of this system within the next 3-8 months. The very thought of making a system that adapts to current market conditions and makes optimal trades makes my mouth water. On top of that, once the neural net is designed, there is no need to tinker with it because it is designed to find the best parameters that work in current market conditions. This, to me, is the way to live life - using your intellect to make you money.