Screening rules and portfolio performance

Abstract

We analyze the use of alternative performance measures to rank and select assets. Previous literature centers on the effects of non-normality on rank correlations between orderings. Instead, we select the assets recommended by each performance measure (ordering) and analyze out-ofsample returns of the portfolio that contains them. The overall empirical findings show that performance measures are definitively relevant for subsequent portfolio returns. Assets selected by the Generalized Rachev ratio dominate other selections showing high cumulative returns after the 2008 downturn. The good performance is connected to the fact that these asset returns show high excess kurtosis but positive skewness and are insensitive to the momentum risk factor.

Copula methods for evaluating relative tail forecasting performance

“…Finally, those portfolios based on ETR and SKR exhibit remarkably low correlations respecting the SR portfolio, which enhance the di¤erence between the latter and the former PMs. These results are also in line with those about equity screening by León et al (2019). Because of these …ndings, in the following section we explore the behavior of the upper/lower tail of the bivariate distribution of SR and every other PM portfolio so as to highlight possible di¤erences in simultaneous occurrence of large/small PM portfolio returns.…”

Dynamically Controlled Length of Training Data for Sustainable Portfolio Selection

“…Nowadays, the selection problem becomes very important since the number of trading strategies and stocks listed on stock markets are continuously increasing [20]. A number of algorithms and criteria for stock screening and ranking are proposed in the literature [21][22][23].…”

“…It was demonstrated that with an increase in portfolio complexity and complexity of optimization procedure, m, we can observe the over-fitting phenomena in the selection procedure. For this reason, often one employs simple selection rules such as selection of the individually most effective inputs [20,23,25,26].…”

Outperformance of the pharmaceutical sector during the COVID-19 pandemic: Global time-varying screening rule development

“…Our study makes fundamental contributions to the literature. Specifically, we closely follow the approach of [7] but expands on both the methodological terms and the economic and financial implications as follows: (i) Our overall filtering measure is constructed by considering different sets of performance measures, not merely one. Specifically, our study synthesizes performance measures from two very different families, Kappa and FT, one based primarily on the expected return/downside risk ratio and the other on the upside risk/downside risk ratio; (ii) Unlike [7] , who construct synthesized rankings by performance measure and then select the best performing assets in an in-sample static period, we conduct a time-varying asset allocation from a PCA-dynamic approach based on a rolling window re-estimation procedure.…”

“…Specifically, we closely follow the approach of [7] but expands on both the methodological terms and the economic and financial implications as follows: (i) Our overall filtering measure is constructed by considering different sets of performance measures, not merely one. Specifically, our study synthesizes performance measures from two very different families, Kappa and FT, one based primarily on the expected return/downside risk ratio and the other on the upside risk/downside risk ratio; (ii) Unlike [7] , who construct synthesized rankings by performance measure and then select the best performing assets in an in-sample static period, we conduct a time-varying asset allocation from a PCA-dynamic approach based on a rolling window re-estimation procedure. Accordingly, we have developed a new dynamic screening rule; (iii) While the former authors use the majority of S&P 500 stock components, we focus our study on ETFs that belong to the financial, healthcare and pharmaceutical sectors; (iv) Rather than considering a boom economic period such as the in-sample year 2005 used by [7] , we study a short timeframe that considers only the first waves of the COVID-19 outbreak.…”