Greetings all -- Thanks to all who are making the R package accessible from within AmiBroker. I plan on using it, and I'm sure others will as well.
This posting is a word of caution about the use of statistics in trading system development. I see two major areas of application: 1. Monitoring the results of real trades and using statistics to determine whether the model (the AFL) of the system is out of sync with the data. This is certainly a legitimate application. 2. Using statistical methods to analyze the price series, searching for potentially profitable patterns. Be aware that the results of the search are in-sample. Data mining results are always in-sample. And data mining results are very prone to being over-fit to the data. It is extremely important that enough data be reserved for out-of-sample testing to determine whether the patterns found represent a usable signal, or whether they are either lucky or fit to the noise. Those of you who have read Quantitative Trading Systems will recall two examples where data mining found what appeared to be reasonable trading signals. In both cases, there were over one million closed trades in the in-sample period. And in both cases, the system was not profitable out-of-sample. In any event, no matter how promising the statistics look, in order to trade the signals, they must be written into AFL and run as a trading system. That gives an opportunity to look at the signals and their results -- equity growth, holding periods, drawdowns, portfolio applications, and so forth. If there has been no hold-out period when performing the data mining, then there is no out-of-sample period for the trading system. Please be careful. It is very easy to fool ourselves when data mining. Thanks for listening, Howard
