Filter Matching

The filter Matching feature is a new and experimental feature that applies only to the rebalance test. It is useful for making sure that a rebalance test only compares against simulated alternative portfolios with the same filter scores.

When executing a rebalance test, the tracking error (relative variance) of the portfolio being tested against an optimal frontier portfolio is compared with the tracking error (relative variance) of portfolios from simulated alternative Michaud frontiers, created with perturbed alternative outputs. Normally these alternative efficient frontier portfolios are selected by associating portfolio rank, e. g. if the optimal portfolio point was portfolio 31 by arc length on the Michaud frontier, the simulated alternative frontier points for comparison will also be portfolio 31 by arc length on their corresponding efficient frontiers.

However, when filter matching is enabled, the alternative frontier portfolio is selected by matching the filter score. This means that in the rebalance test, the tracking error between the portfolio being tested and an optimal portfolio will be compared to tracking errors between that optimal portfolio and simulated alternative portfolios with the same filter scores, or the nearest available if no portfolio exists on the alternative frontier which exactly matches the filter score. This can be useful when the filter score is used to select the optimal portfolio for a particular strategy, and may help tame the alternative tracking errors and boost the rebalance signal at the top of the frontier where sometimes the rebalance signal is weak, especially in cases with non-homogenous assets such as multi-asset-class models.

This feature can be useful for certain cases to calibrate the relative sensitivities of rebalance tests across multiple strategies on the same frontier. Many cases may show little difference in the rebalance probabilities generated with or without this feature, but in certain cases, the normally more sensitive lower risk strategies can be made less sensitive relative to the upper strategies with this feature on. In these cases all strategies may be less sensitive in absolute terms, which can be corrected by adjusting the typical rebalance periods to a lower number, thereby lowering the relative variances between simulated alternative frontiers and the main optimal frontier, and increasing the sensitivities of all rebalance tests across the frontier. Clearly a lot of calibration is necessary to determine the operating characteristics of a rebalance test and the correct settings.