The recent global economic crisis illustrates how financial leverage can generate couplings between different sectors of the financial system and how problems in the financial sector can affect the real economy. This research team will develop new types of agent-based quantitative models to enable a deeper understanding of the causes of extreme or systemic risks in the financial system, focusing in particular on the roles of leverage, network structure, and institutional rules.
The goals of the project include determining optimal leverage levels; understanding the effects of network structure on systemic risks; predicting default rates; understanding the role of leverage in contagion between assets; understanding the role of time granularity in risk monitoring; finding improved risk control schemes for banks; and exploring intervention policies for preventing financial failure. The Federal Reserve Bank of New York and the Bank of England will provide access to data for model calibration purposes.
The models developed should provide guidance about policies that can foster stability in the financial system. Insight into the effects of various leverage levels, repayment schedules with less systemic impact, or a better understanding of the properties of networks of banks and lenders might be helpful in mitigating future financial crises.
The aim of the project was understanding how interactions between different components of the financial system can generate instabilities such as asset bubbles, crashes or domino effects in the banking system, and to explore policies aimed at increasing the resilience of the financial sector and the economy as a whole. We have addressed these questions both from the theoretical and empirical point of view, and we have taken an interdisciplinary approach in our research, where methods and concepts borrowed from physics, ecology and computer science have been used as a complement and enhancement of standard approaches of finance and economics. We highlight in the following the main outcomes of this project. We have developed a network model to study how stress propagates between banks with similar portfolios. The idea is very simple: when an investor liquidates its investment portfolio, the liquidation process causes a devaluation of the assets that are being sold, and a loss for other banks investing in these assets. We have found that, because of this mechanism, a stressed investor can trigger domino effects that result in liquidity spirals and cascades of defaults when leverage increases beyond a critical value, and we have studied how the parameters that characterize the instability interact with each other, as shown in Figure 1. Thus upshot is that, depending on how crowded the space of assets is and how diversified banks are, modest values of leverage are fine, but above a critical threshold the system becomes unstable. Surprisingly, while diversification is good for individual banks, from a systemic point of view it is destabilizing because it makes banks too connected. We have proposed a new accounting method that explicitly incorporates into risk management tools the idea that selling has market-impact. The amount of cash that can be raised by liquidating a portfolio is in fact smaller than the portfolio mark-to-market value because selling depresses prices.We have shown that our method results in better risk control than mark-to-market valuation (see figure 2). We have developed an agent-based model of the housing market in Washington D.C. We carefully calibrated the model to empirical data and showed that it can match historical time series of quantities such as the housing price index, days on market, number of houses sold, and many other quantities. This represents an important breakthrough as it demonstrates that agent-based models can be calibrated to match time series data and used to analyze the consequences of policies. In this case we are able to deduce that the likely cause of the bubble is excessively liberal lending policies (see figure 3).
