The paper proposes a new method based on threshold conditional correlation (TCC) that allows for regime changes in the correlations between financial assets. This method allows the dynamics of the correlations to change from one state (or regime) to another as a function of observable transition variable(s). The TCC is similar in spirit to the smooth transition conditional correlation (STCC) in Silvennoinen and Terasvirta (2009) but with the appealing feature that is easier to estimate, even in large dimensional problems. Estimation of the parameters of the TCC involves a grid search-MLE method in which the correlation matrix is positive definite by construction. The methodology is illustrated by evaluating the behaviour of international equities and government bonds, first separately and then jointly. We further generalize our approach by allowing for different parts in the correlation matrix to have their own transition mechanism, while at the same time guaranteeing that the resulting correlation matrix is positive definite. Finally, we evaluate the out-of-sample economic performance of the TCC model against the popular dynamic conditional correlation (DCC) model by using the Engle-Colacito (2006) test and by comparing their density forecasts. The results show that threshold model with four regimes outperforms the DCC, mainly in the recent global financial crisis a period with significant shifts in the level of correlations.

Modeling cross correlation across major financial markets: a threshold approach

Thu 1 Oct 2015 12:00pm2:00pm

Venue

Room 629, Colin Clark Building (#39)