A Black-Litterman portfolio allocation model combined with a Markov switching framework

This is a M.Sc. thesis investigating the compatibility and performance of a regime switching framework as a complement to the Black-Litterman portfolio allocation model. Conclusively, it is considered to be a compatible match of models in terms of practical implementation and the results indicate that the model is performing well.Portfolio allocation using short term predictions of business cycles The Black-Litterman model is an asset allocation model developed by Fischer Black andRobert Litterman in the early 90’s. It was published in the internal Goldman Sachs Fixed Income Research Note, Black and Litterman (1990). The model is an advanced Mean Variance Optimization framework, with an objective to maximize the return of a portfolio in relation to its risk. The primary reason for using the Black-Litterman model instead of a traditional Mean Variance Optimization model is to overcome problems such as unintuitive, highly concentrated portfolios, input sensitivity, and estimation er-ror maximization. This is achieved by incorporating subjective investor opinions, called views, in the model, which usually works fine if the views are appropriately formulated. The problem with this concept is that it is not very easy to achieve consistency and operational efficiency when specifying the views. This thesis examines a model using historical data to find trends in the business cycle as a tool to extract the views. The model used is a regime switching model and more specifically a Markov switching model. There are two reasons for choosing a model using historical data when creating views. Firstly, it is less time consuming to implement a model that automatically turns public information into views. Secondly, it formulates every view from a precise framework and is therefore more consistent in its allocation. However, the Black-Litterman model is successful because of its ability to use additional information not found in the historical exchange databases. This raises the question whether a method that is generating views from market data will enhance the performance or not. Someone once said investing this way is like “driving a car looking in the rear mirror”. The response to people distrusting the concept is that a regime switching model can spot trends which are assumed to be related to market consensus and indirectly to the investor opinions. In financial modelling a portfolio return series can traditionally be assumed to have a constant mean and a constant variance over a certain period of time. A two state Markov switching model instead calculates the probability for the portfolio to be in certain states during different stages of the same period. The model does this by assuming that the states can be described by a hidden Markov chain. The parameters of the hidden Markov chain cannot be observed and must be estimated from the market data. Each state has different statistical parameters, such as mean and variance. A high variance combined with a low mean could indicate that bear market conditions are dominating and vice versa could indicate a bull market. This market insight will be used to assign views. In this thesis a model combining the Markov switching and Black-Litterman models is used to repeatedly reallocate a portfolio with 10 stocks listed on the OMXS30 during the period August 2014 to October 2017. Four reallocations are made and the result is compared to a market weighted portfolio. A positive conclusion is that the two frame-works are practically compatible and that the new portfolio outperforms the traditional market portfolio in terms of absolute return and Sharpe ratio. The challenge lies in refining the Markov Switching model in order to let it handle larger data sets. It would be beneficial to test the entire model with different market conditions and other assets before it can be considered a reliable investment tool