FastVar : Alpha capture without the handcuffs

Nachiket Garde

Linear factor models typically focus on medium-term risk estimation while the need for short time horizon VaR numbers is usually met using an implementation of exponentially weighted moving average (EWMA) approach based on some sample of historic data.

Post the GFC, due to a recognition that both approaches were reporting low levels of risk prior to the crisis, a greater interest was evinced in being able to estimate short term risk – up to 1 month ahead – and this trend was further encouraged by regulatory developments such as the EU’s UCITS rules which required managers to calculate a near-term risk estimate and confirm its performance against outcomes.

One of the obvious dilemmas in any risk forecasting system is the trade-off between accounting for the risk of extreme event scenarios ("Black Swans") and overestimating the typical daily risk of a portfolio ("Cry Wolf"). While underestimating the risk has obvious implications, over estimating the near-term horizon risk of an investment strategy can result in lost opportunities if a risk-model repeatedly indicates a greater level of risk in near-term projections than tends to be borne-out over time, and this can lead to undesirable effects on portfolio return, in effect handcuffing the manager.

Sapiat’s near term volatility forecasts have several advantages in addressing this challenge. The forecast consists of two components. The first component is based on an estimation of a latent factor model, where each factor is sample orthogonal to the others. Then for each factor, a tracking portfolio is constructed by utilizing only securities with sufficient trading frequency and volume. By applying a GARCH process to these daily returns a near-term variance forecast for each factor is then generated.. The use of GARCH has important advantages over exponentially weighted moving average (EWMA) as it allows for forecasts at different time horizons – say 1 day or 20 days – that are not simple scaled values but which take account of the tendency of volatility to revert to trend over time. An additional advantage of this approach is that it is possible to estimate the daily performance of the estimated factors out of sample by deduction from the price performance of a large number of liquid assets.

The second component, a residual variance scale (RVS), is the incorporation of an innovative residual scaling factor into the forecasts to take account of the tendency for residuals to increase across the board during high volatility periods. There are many investment strategies in which this ( diversifi able) risk can still have a major impact on the overall variability of portfolio returns. For example, "Stock-picking" strategies seek to minimise, relative to a benchmark, systematic exposures to, for example, country and sector risk. Absolute return strategies seek to match systematic exposures on the long and short side and manage leverage appropriately in response to changes in residual risk levels. In such cases most of the relative performance risk will reside in proper management of the residual risk of the stocks. As the residual variance is around three times systematic variance for a typical equity, it is critical for such strategies to manage this risk to achieve desired outcomes.

The adjustment of residual risk vs. systematic risk is a key aspect of the short term model methodology. As we all know .

(a) there are periods when (active) returns are driven mainly by systematic factors,

(b) periods when systematic factors go to sleep and returns are driven by stock specific exposures, and

(c) periods in between, when certain factors are moving while others are quiet

Hence, when Sapiat’s models forecast what each factor volatility would be, they also make a forecast/adjustment of the residual risk (across the board of all residuals). As the Sapiat model offers a true forward looking set of factor variances and covariances, it is valuable in both risk analysis and efficient portfolio construction and is an innovative compromise between the sometimes conflicting goals of alpha capture and risk management. In effect, it allows you to capture alpha without the handcuffs.

Shota Ishii