Mortgage and Other Loan Risk Modeling
The financial crisis in the US that took place in the first decade of the XXI Century has exposed may distortions at all levels of our financial system, beginning from the excessive supply of "easy money", continuing to the top management practices at the leading financial institutions, continuing to the modeling and assessment of risk at banks, GSEs, and other entities, and ending with the insufficient understanding and appreciation of the underlying risks by the average investors and their sometimes unmoderated appetite for abnormally high returns.
These conditions created an environment of excessively high volume of lending at interest rates that were too low to compensate for the risks underlying the loans. Additionally, the benefits of risk sharing and securitization of loans were overestimated and the risks of highly correlated defaults across different loan classes were considerably underestimated. Credit spreads had become very thin, and it was becoming increasingly clear that the situation was unsustainable in the long-term, but it was quite difficult to pinpoint the precise time at which it would all come apart. As we now know, the unraveling began in the early 2008, lasted for several years, and required creative actions by the Federal Government and the Federal Reserve.
One of the important factors that contributed to the riskiness of the situation is the sometimes inadequate modeling of loan risks, including the probability of default and the severity of loss. At many financial institutions, the models were benchmarked against and calibrated to the prices of risk that were implicit in the market, and since the market was "flush with cash", the market prices of risk were too low. Thus, the very purpose of loan risk models, which is to impart to the market adequate prices of loan risks, was bypassed.
No model is precise, by the very definition of a model. Nevertheless, we strive to build models based on logic, economic fundamentals, and intuition, using econometric techniques as needed, and we avoid going down the easier path of building models mostly based on statistical fit and data mining. We help clients with modeling and model validation in the areas of: