# Samuel kaski thesis

Due to the conditional dependencies between states at different time points, calculation of the likelihood of time series data is somewhat tedious, which illustrates the motivation to use ABC. A computational issue for the basic ABC is the large dimensionality of the data in an application like this. This can be reduced using the summary statistic S, which is the frequency of switches between the two states. As a distance measure ρ ( ⋅ , ⋅ ) {\displaystyle \rho (\cdot ,\cdot )} , the absolute difference is used, combined with a tolerance ϵ = 2 {\displaystyle \epsilon =2} . The posterior inference about the parameter θ {\displaystyle \theta } can be done following the five steps presented in Figure 1 :