MARKOV SWITCHING MODELS OF POPULAR FOREIGN EXCHANGE CARRY TRADE STRATEGIES

The nature of the carry trade produces periods of steady profitability and periods of extreme terror. The 1980s proved to be a particularly profitable time period. However, during market crashes in the either equity or bond market, the carry trade is marked with short periods of substantial losses (Menkhoff, Saro, Schmeling, Scrimpf 2012). The global financial crisis of 2007 – 2008 was associated with large losses to carry trades. History repeatedly suggests these two bull and bear states in the economic environment (Fabozzi, Francis 1977). An additional state of market neutrality or stability could also be considered. The purpose of this dissertation is to develop a model of the carry trade with multiple states using the Markov switching methodology. To accomplish this, we use two different popular carry trade strategies: (a) logistic regression and (b) mean-variance optimization. As a benchmark, we include an equally weighted portfolio of long positions in foreign currencies against the dollar. We develop a single state model as well as a normal mixture model for each of the two carry trade strategies. The mixture models assume a static probability of the economy being in either state. However, the financial markets are not static. Applying a Markov chain allows us to build a dynamic model, which allows for new information to determine the probability of the next state. We applied a Markov chain to determine the probability of the current state and the next state to improve trading results. We found the application of a Markov chain did not improve trading performance. The portfolio consists of 12 different currencies including both mature and emerging markets. The training period for determining the weights is 1998 through 2002. Using daily data from 2002 through 2015, we evaluate the performance of each strategy using cumulative returns. These results demonstrate the periods of profitability followed by short periods of terror. Next we evaluate the performance of each strategy with an applied mixture-model. The mixture-model improves the results of each strategy. Applying a Markov chain allows for better determination of both the bear and bull states. We use only the two state environment as the three state environment was unstable.Ph.D. in Management Science, July 201