written by C.P.Wendling

Intellitrade Inc. systems are computer programs that analyze past market dynamics and price patterns to develop inferences about likely future price changes. Price relationships are mapped as probability density functions, relating past price changes to their respective price changes in the future. A Genetic Algorithm (GA) searches the multidimensional space that can be created by considering each of the distibution and trading parameters as a large vector (genotype). The GA searches for Artificial Traders (AT's) which are stable through time (time invariant). The fitness function in the GA is written to encode the concept of generalization (i.e., solutions with low performance variance over different time periods) directly. Our experience has been that finding good generalized solutions is difficult and the key to success. Generalized solutions of financial time series forecasting problems are very elusive. Financial time series are in a class of processes between completely deterministic (sine wave for example), and random (winning lottery ticket numbers, for example.)
Unless one directly encodes the essence of generalization into the search/optimization algorithm, the results will be highly suspect, as it would be impossible to prevent over or under fitting to the training data set. Most "back testing" methodologies fall into this trap, finding solutions that are not general, and usually fail dramatically when run in real time.

IntelliTrade uses "maps" as eyes for the AT's. These maps are closely related to perceptrons seen in neural network literature. A map is a 3-dimensional surface, as shown in the illustration below. The x-axis is time, the y-axis is historical price change, and the z-axis is the future price change. Consider the following question: "If an index has fallen sharply over the last two weeks, is it more likely to rise or fall in the next 5 days?" Maps help answer that question, and others like it. Maps are built by overlaying price "traces" onto the map surface. A price trace is comprised of the historical prices, plotted on the grid surface, with all y-axis values referenced to today's closing price. (The x-axis represents time.) The reference price, or focal point on the map is on the far right side, half way up the grid, as shown with a white tic mark below. All price traces terminate at the same focal point, and all prices are plotted relative to that point. For example, todays "trace" could be the last 60 days plotted with respect to today's closing price, with todays closing price at the focal point. That trace, or graph, would pass through certain cells of the map, depending on its particular price pattern, or signature. The price percent change 5 days into the future for each trace, is then averaged into each of the cells that the trace crosses. For example, if the price 5 days hence was up 1.2%, then 1.2% is averaged into each cell along that trace. This process is repeated for each trace that can be constructed from the time series, building up a 3-dimensional surface on the map, where the altitude (z-axis) is the average outcome (percent price change) of each trace crossing a particular cell. By doing so, one can see at a glance which traces or price paths resulted on average in the largest price appreciation 5 days into the future. This is because those traces will have a ridge or elevated terrain for that path on the map surface. Similarly, low path's into todays reference point will result in larger price depreciations 5 days hence. If the time series is random, there will be no stable ridges or valleys on the map surface, because past prices wouldn't have any relationship with future prices. In the illustration below (which is a map of the S&P 500, 60 day trace lengths, 5 day future look), the red areas are negative, while the green areas are positive cells. One can see a green ridge line approaching the focal point from the left side, slightly below the centerline of the map. This means that price histories, which approached along that path, had on average appreciated in price 5 days later. This confirms what some label as "price momentum", at least in the S&P 500, over the data for which the maps was built, on the time scales represented by that particular map. The very important point is that a map can quantify what was before only a very loosly defined verbal concepts, (in this case: "price momentum") which allows a more rigorous statistical analysis to the forecasting problem at hand.




In the image below, the green "ridge" is circled in yellow. Price "traces" that approach the focal point along that path, went up (on average) in price 5 days later. The blue circled area indicates paths where the price fell (on average) 5 days later, and are seen as low lying, red areas. Each AT's has many "eyes"(maps), each evolved to process the input data streams in usefull ways, by translating, scaling, and mapping raw input price streams (e.g., moving averages, raw price data, slope data, and other input data streams) into transpositions which promote the most [stable through time] representations. We've only shown a single, simple "eye" (map) for illustration purposes. Some AT's have dozens of eyes, which they use to obtain and preprocess data from their environment.


Disclaimer:
Hypothetical or simulated performance results have certain inherent limitations. Unlike an actual performance record, simulated results do not represent actual trading. Also, since the trades have not actually been executed, the results may have under- or over- compensated for the impact, if any, of certain market factors, such as lack of liquidity. Simulated trading programs in general are also subject to the fact that they are designed with the benefit of hindsight (1). No representation is being made that any account will or is likely to achieve profits or losses similar to those shown. The information contained in this site has been compiled in good faith, and in using it, the user agrees that the author and any other entities associated with this site shall not be liable for any direct, indirect, consequential loss arising from this usage, or the use of information and material on the Internet via web links from this site including, but not limited to errors, omissions, defects, interruptions, delays in operation, or transmission, computer viruses, or line failure, to the maximum extent permitted by law. Please note that trading is a risky business. You may lose your entire investment capital and more. Hypothetical performance results have many inherent limitations, some of which are described below. No representation is being made that any account will or is likely to achieve profits or losses similar to those shown. In fact, there are frequently sharp differences between hypothetical performance results and the actual results subsequently achieved by any particular trading program. One of the limitations of hypothetical performance results is that they are generally prepared with the benefit of hindsight (2). In addition, hypothetical trading does not involve financial risk, and no hypothetical trading record can completely account for the impact of financial risk in actual trading. For example, the ability to withstand losses or to adhere to a particular trading program in spite of trading losses are material points which can also adversely affect actual trading results. There are numerous other factors related to the market in general or to the implementation of any specific trading program which can not fully be accounted for in the preparation of hypothetical performance results and all of which can adversely affect actual trading results.

End