On Causal Networks of Financial Firms: Structural Identification via Non-parametric Heteroskedasticity

Various business interactions of banks create a network of hidden relationships, which cannot be directly inferred from the correlation of bank stock returns. When stocks of bank A and bank B co-move, is it because of variations in A or in B? This question about the order of causal responses is vital for regulators. Without causality, it remains unclear how policy interventions change the network. Thus, this paper aims to find the causal network as anticipated by investors.

For this purpose, I introduce a new method to identify hidden relationships of financial time series. The identification is based on the belief that individual risks change faster than the network of relationships. I apply this method to the return volatilities of US financial firms.

The estimated network is more in line with narratives of the 2007–09 global financial crisis than alternative methods. For instance, I show that the liquidity stress of the insurer AIG was a major source of spillovers. I also confirm a decrease in risk stemming from banks over time. This finding may be a result of the implementation of the Basel accords. Eventually, I can easily monitor and rank banks conditional on their spillover of risk in the network.