Arman Glodjo
Duke University, Durham, NC 27708, USA
Campbell R. Harvey
Duke University, Durham, NC 27708, USA
National Bureau of Economic Research, Cambridge, MA 02138, USA
Abstract
This paper bridges the important recent work in computer science on information theory and data quantization to the forecasting of high frequency financial data. The technique of vector quantization has found its primary application in data compression algorithms. We argue that this technique is ideally suited for forecasting. Indeed, our paper shows that popular forecasting techniques such as, neural nets, are sub-classes of the more general vector quantization. Importantly, the neural nets are not just subclasses but are surely suboptimal. The vector quantization provides much more flexibility and a framework for efficient algorithmic approximation. In addition, vector quantization provides a way to incorporate conditioning information into the forecasting exercises. While much of our proposal details the theoretical motivation for entropy (amount of information) based coding, our empirical work is designed to implement forecasting models on intraday exchange rate data.