Alexandre Pauli, UC Berkeley: Forecasting a Drawdown-based Risk Measure

Goldberg and Mahmoud (2014) have defined Conditional Expected Drawdown (CED) as the tail-mean of the maximum drawdown distribution and abstractly showed the attractive properties of the risk measure for risk management and portfolio construction. The purpose of this project is to empirically investigate how CED can be employed in practice. The major challenge is to find accurate estimators and to retrieve enough information about maximum drawdowns, which vary on both magnitude and duration, to produce a forecast of CED with minimum uncertainty.

We compare CED to the well-known Value-at-Risk (VaR) and Expected Shortfall (ES) measures and analyze different methodologies to obtain the maximum drawdown distribution from a time series. Then, we select backtesting methodologies, proposed for ES, that can be applied to CED. In this framework, we undertake an in-sample analysis that allows us to evaluate the precision of different estimators, resulting in regime-switching models as best parametric estimators. These results are an important first step before undertaking an out-of-sample analysis, and hence toward the assessment of uncertainty in CED forecasting.