Swedish Business Market Cycles as a Hidden Markov Model : An analysis of the usefulness of Hidden Markov Models for modeling the business market cycles in Sweden

Sammanfattning

The state of business cycles affects the economy greatly. During economic expansion the unemployment rate is most likely low and the inflation rate high, whereas the opposite holds during a recession. Recessions can have a long lasting negative impact on a country's ability to produce. Because of this, it is important to be able to model business cycles. Hidden Markov models have been used to model stock prices through its emissions. There are several emissions from a business cycle, such as inflation and unemployment. This study intends to investigate whether or not it is possible to model Swedish business cycle states as a hidden Markov model. Hidden Markov models have previously been used to predict turning points in business cycle states, and this study expands on this to also model the severity of the expansions and recessions. Similar studies have been done on Poland, France and Germany, but the present study explores whether the findings are applicable on Sweden's business cycle. The groupings of variables also differ in the present study. Additionally, the present study uses observations of inflation and unemployment only. The model used is trained on transitions of output gap states in Sweden. These output gaps were determined using a Hodrick-Precott filter. The model was then fed with data on unemployment and inflation. The resulting models' performance was compared with one another and a naive model, where the most common output gap is constant. The results show that a model based on inflation and unemployment, with a trigram approach, performs significantly better than a naive model. However, the performance is mostly limited to a descriptive model which is trained on the data to be modeled. A predictive model is not considered in this study. The study concludes that hidden Markov models may be used in modeling business market states through output gaps. However, in order to improve performance, alterations to the model presented in this paper are needed.