Modeling information ratio variability across time and across quantiles

The body of theoretical and empirical academic work in the area of active portfolio management has increased substantially in recent years. This development has coincided with the growth of active portfolio management in practice and the emergence of new financial institutions, such as hedge funds. The size of the industry, according to latest figures, stands at close to 3 trillion dollars. An important metric of active management is the so-called information ratio (IR), which is calculated as the excess or alpha return per unit of idiosyncratic risk. Almost all applications involving the IR use unconditional measures assuming that IR is stable.  

This paper investigates the dynamic properties of the IR in the time domain as well as across the quantiles of the conditional distribution of hedge fund returns. The advantage of this approach is that it allows active portfolio managers to rebalance their portfolios in accordance with the changing pattern of the IR.

More specifically this study aims to provide answers to the following questions:

1)  Are the IRs of hedge fund returns constant across time and across the range of the distribution of returns?

2) To which extent are alpha returns contributing to IR variability?

 3) Is idiosyncratic risk responsible for IR variability?

The source of IR variability is an important aspect of this investigation, because different sources of variation will dictate different approaches to portfolio management and rebalancing. For example, if variability is due to changes in risk, it would imply that modeling risk and detecting predictable patterns is an important activity of portfolio management. If variability is due to alpha returns, it is important to empirically investigate the nature of such changes.

The empirical investigation of IR variability will be based on a pricing model of the Fama-French type with idiosyncratic risk following a conditionally heteroscedastic process (ARCH-type). This model will allow detection of IR variability due to changes in idiosyncratic risk.

 The IR variability across the conditional distribution of returns will be based on the quantile regression model of Koenker and Basset (1978).  

We expect the conditional IR to be changing across time  based on the evidence regrading the variability of risk over time. Furthermore, we expect the IR to exhibit significant variation across the distribution. The findings could be useful for portfolio strategies, rebalancing techniques and performance evaluation