There are many ways to find a price trend. Some may think that the methods for finding trends have been exhausted, but as seen with some new research that is not true. In the paper, "(Re-) Image(in)ing Price Trends" in the Journal of Finance, the researchers have used focused pattern recognition techniques to make price trend predictions. They have found that pattern recognition, through convolutional neutral networks (CNN), which uses images to make predictions different from traditional time series analysis, can be effective. Behavior creates patters that can serve as the foundation for finding price trends. Humans digest a lot of information in pictures which are used to make decision about trends.
In this research, the authors focus on the images created from open, high, low, and close (OHLC). The set of information from this range data can be create images that can be standardized and compare with the past to form predictions. When these CNN models are compared with classic trend, momentum, and reversal models, it is found that imaged based models will have significantly higher Sharpe ratios and do especially well in the extreme deciles. Now, calibrated this type of model takes time and effort, but the size of the gain suggests that trying to offset the barrier to entry is worth it. Perhaps the technicians who focus on charts are not the crazy alchemists portrayed by finance professors. A technical or visual approach to investment decision-making may have merit, albeit its success may be due to diligent systematic measurement of patterns.
No comments:
Post a Comment