77 Pages Posted: 30 Sep 2013
Date Written: August 7, 2013
This thesis investigates models of market risk assessment based on genetic algorithms, with specific reference to asset portfolio choice under volatile market conditions. It does so by developing computational simulations of asset portfolios, which are then subjected to stressful price events. A genetic algorithm functions as an optimising process, allowing portfolios to evolve towards a structure that is – on average – less fragile against asset shocks. The importance of this research is dictated by the grave outcomes of, for instance, the 2008 financial crisis: 371 commercial banks failed between 1/1/2008 and 1/7/2011 in the United States alone. Such events highlighted the need for the renovation of the financial risk framework supposed to at least partially shield banks from unexpected adverse events.
A synthetic, computational model is constructed, where asset portfolios are structured so as to invest in four main asset categories: sovereign bonds, financial institutions bonds, corporate bonds and real estate. Each of the categories has an underlying non-normal probability distribution of prices, empirically derived by the literature. These are used to simulate volatile and adverse scenarios affecting the value of the portfolios. A genetic algorithm is designed to select, crossover and mutate, at each generation, the portfolios that best perform under the simulated conditions. After a number of generations, it is expected that one or more portfolios structures will be highlighted as the ones that best perform under adverse scenarios.
The model is run three times with different sets of optimization constraints, each specifying the minimum relative proportion of portfolios to be dedicated to each asset category. All versions of the model indicate that the best performing portfolios structures under volatile conditions are the ones that are mainly composed by the asset category featuring less fat tails. The results of the model are checked for their robustness, by running versions with different sets of simulated scenarios and additional numbers of synthetic asset categories.
Limitations of the design of the study are identified. The model lacks a simulation of the liability side of financial institutions, and its results are not tested on a systemic level, thus not shedding light on what consequences the indicated portfolio strategy for a single bank would have on the network of banks. Such issues will be addressed in future research.
Keywords: financial crisis, resilience, antifragility, fragility, finance, portfolio, genetic algorithms, banks, commercial banks, Monte Carlo Methods, evolutionary economics, evolutionary optimization
JEL Classification: C15, C19, D81, D83, G11, G18
Suggested Citation: Suggested Citation
Gurciullo, Stefano, Strengthening Banks' Portfolio Against Asset Shocks: A Genetic Computational Approach (August 7, 2013). Available at SSRN: https://ssrn.com/abstract=2333471 or http://dx.doi.org/10.2139/ssrn.2333471