Search results
(1 - 1 of 1)
- Title
- BAYESIAN MOMENTUM STRATEGY OF EXCHANGE RATES
- Creator
- Lee, Namhoon
- Date
- 2011-11, 2011-12
- Description
-
A disagreement has existed between the foreign currency trading community and academic researchers relating to the time series properties of...
Show moreA disagreement has existed between the foreign currency trading community and academic researchers relating to the time series properties of exchange rates. Traders typically view exchange rates as strongly trending prices and suggest that simple rules, based solely on past prices, have generated predictable profits with acceptable risk over most of the floating-rate period. However, many surveys presenting controversial results. This research identifies the non-linear trend momentum in monthly exchange rate and examines the profitability of momentum trading model within exchange rate returns in the context of Bayesian econometrics. A development of Bayesian momentum trading strategy based on trend component of the spot exchange rate is established. First, parameters of momentum model for each main currency are estimated. The momentum is defined as a simple nonlinear function of return series and the model is designed to estimate the expected conditional mean and associated conditional volatilities simultaneously. The empirical results reported several notable confirmation and findings; first, predictability of momentum model with Bayesian approach show better accuracy than model with maximum likelihood estimation or moving average rule in terms of directionality and model fitting. Second, parameters are restricted to be same across the currencies with the assumption that currencies share some degree of commonality within the system. The result confirms that the restricted model work as well as the unrestricted model within the currency model in terms of model fitting and directional accuracy. Third, principal component analysis is used to analyze the exchange rate movements. PCA found that the first principal component shows parallel shift of all currencies and second principal component tilt shift where high yield currencies move down and low yield currencies move up. Fourth, the parameter estimates from the models are used for portfolio allocation ix applying Bayesian Principal Component(PC) GARCH(1,1) model and the portfolio performance is compared with the performance with classical maximum likelihood approach and other benchmarks. The results show that the Bayesian PC-GARCH(1,1) performs better than classical PC-GARCH(1,1) in terms of Sharpe ratio, Value at Risk, Expected shortfall, maximum drawdown and other statistical criteria. Sixth, the GARCH parameter space is found to be non-symmetric confirming that maximum likelihood estimation would have over or under estimated the parameter causing misspecifying the model parameters. The result from this research confirms simple nonlinear momentum model combined with Bayesian approach can be a good forecasting tool, and restricted model can simplify the complexity of parameter space of exchange rate movement. In addition, by correctly detecting the parameter space, Bayesian approach outperforms the classical maximum likelihood approach. Keywords : Bayesian framework, Momentum, Moving Average rules, Carry trade strategy, Mean-variance Optimization, Trading strategy, Metropolis-Hastings Algorithm, Gibbs Sampler
Ph.D. in Management Science, December 2011
Show less