Analysis and Prediction of High Frequency Foreign Exchange Data
This thesis investigates the stochastic properties of high frequency foreign exchange data. We study the exchange rate as a process driven by Brownian motion, paying particular attention to its sampled total variation, along with the variance and distribution of its increments. The normality of its increments is tested using the Khmaladze transformation-2, which we show is straightforward to implement for the case of testing centred normality. We found that while the process exhibits properties characteristic of Brownian motion, increments are non-Gaussian and instead come from mixture distributions. We also introduce a technical analysis trading strategy for predicting price movements, and employ it using the exchange rate dataset. This strategy is shown to offer a statistically significant advantage, and provides evidence that exchanges rates are predictable to a greater extent than current mathematical models suggest.