CONTAGION AND COBREAKING

In this project we attempt to make two contributions: one financial and one statistical. The financial contribution is centred around modelling and testing contagion defined in a new way as co-breaking in stock market indices during a crisis or crash in one market. In order to make this contribution the statistical problem of developing a new test for co-breaking must first be solved. Since the break in the first difference (defined as returns) of a non-stationary time series (defined as stock market index) is cumulated to a broken trend via Granger's representation theorem, it opens up a new way to testing for co-breaking, which at the same time dispenses with the unrealistic assumption that the series are stationary. This approach also avoids the use of dummy variables, so that a limiting distribution can be derived, thus avoiding to resort to the assumption of normally distributed errors. In this formulation a test for contagion becomes a test for the absence of a trend break in a linear combinations of non-stationary time series. A first starting point would be to see if existing tests for trend breaks. Solving the research problem will require co-operation between statistics and financial economics. The plan is to start with a review of some recent literature on contagion and model real data using existing tests, but modify them to better suite testing for contagion. The final aim is that after gaining some experience with real data develop a new test for co-breaking which can be applied to testing for contagion. We expect to produce 1 statistical paper and 2--3 papers in applied econometrics and finance.