Previous versions of this work parted from joint losses distributions, considering market and credit losses.
Once getting the joint distribution (using a copula) simulations were made to generate initial losses, which were used as starting point to analyze the contagion process. Nevertheless, to make a more accurate assessment of systemic risk, it has to be established a link between the real sector and banks’ losses.
So as far as the previous work had gone, underlying the simulation process is the idea that “something” happens, but what is that “something”? To attain this goal a Vector Autoregressive Model (VAR) approach has been followed. It is widely known that VAR models allow taking into account multiple variables and their interactions in a parsimonious way so despite its limitations16 this was the methodology followed to model the economic background underlying banks’ losses. Market losses are directly determined by some financial variables such as yield curves or stock indexes, but in their determination, these variables interact with other “real economy” variables, and they are also affected by external shocks either directly (exposition to a foreign stock index, for example) or indirectly by the effect that these external shocks may have in the domestic variables (for example, the domestic stock index follows closely movements of the Dow Jones index).The availability of highly detailed data about banks’ positions allows us to consider very specific conditions for each bank and to realistically assess the effect of each one of these shocks in the current17 period. At any given time, given that time’s financial positions and a relatively small set of variables and “risk factors”18 the losses distribution is computed.
The scenarios generation process consists of three steps: the estimation of the VAR model, the estimation of other complementary models to complete the scenarios to make the valuation of the position, and finally the scenarios generation process.
1.