Performance persistence and optimal asset allocation strategies

Abstract

This study explores whether optimal asset allocation strategies, defined by permutations and combinations of different predictor variables, produce consistently superior performance for investors. We extend the literature by exploring whether such strategies benefit investors over the entire investment period or whether investors are forced to switch among alternative strategies over time. As benchmarks, we employ the 1/N (equally weighted) and the myopic (no predictability) strategies. Persistence tests suggest that no single optimal strategy outperforms the remaining optimal and benchmark strategies over the entire sample. However, in two out of three subsample periods, some optimal strategies persistently outperform the benchmarks.

KEYWORDS:

• Optimal strategies
• asset allocation
• persistence
• models

JEL CLASSIFICATIONS:

• G11
• G12

Acknowledgements

We would like to thank the editor Chris Adcock and two anonymous referees for their helpful comments and suggestions. We are grateful to Ian Garrett, Bjorn Jorgensen, David McMillan, and conference participants at the Financial Management Association Annual Meeting (Orlando) for suggestions and discussions. Any remaining errors are our own.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Studies such as Aggarwal and Jorion (2010), Bollen and Busse (2005), Brown and Goetzmann (1995), Brown et al. (1992), and Droms and Walker (2006) in the mutual-fund and hedge-fund literatures test for persistence in fund performance.

2 In this study, we use the term ‘superior’ in relation to a performance evaluation measure, mainly certainty equivalent return (CER). CER is the risk-free return that would make investors indifferent between following a strategy or accepting this risk-free return. Being a utility-based measure, CER reflects an investor’s welfare after accounting for the higher order moments of portfolio returns captured by the investor’s power utility. In this sense, a superior performance implies that it outperforms other strategies in terms of CER. Hence ‘superior’ and ‘outperformance’ carry the same meaning for our purposes. In particular, we aim to identify strategies that deliver superior performance given by CER consistently through time.

3 In our main analysis, we ignore transaction costs and short-sales constraints. But we account for them in the robustness tests and find that our overall results remain unaffected. This is further discussed in ‘Robustness Checks’ section.

4 Our paper is also closely related to Jurek and Viceira (2011), who develop a closed-form analytical solution to the dynamic portfolio-choice problem. Their framework involves a finite horizon investor characterized by a power utility function defined over wealth and facing a time-varying opportunity set parametrized by a VAR model. Their recursive solution is based on the Campbell-Viceira approximation to the log-portfolio return. However, the main focus of their study is on analyzing value and growth tilts in equity allocation, whereas we examine persistence in the performance of asset allocation strategies.

5 The REITs in this study are equity REITs (eREITs). For simplicity, we label them ‘REITs’. Also note that we use the terms ‘model’ and ‘strategy’ interchangeably, since for our purposes they have the same meaning in that they are defined by different permutations and combinations of predictor variables for the specific asset menu under consideration.

6 The initial date of the sample is determined by the availability of prices and realized total returns on REITs.

7 The choice of these predictor variables is explained in relation to the predictability literature in the Methodology section of this paper. In our main analysis we include six predictors while considering predictability for T-bills, stocks, and bonds, for both the alternative asset menus, with and without REITs. However, in our robustness tests we also consider predictability for REITs by including two additional predictors, thus amounting to eight predictors in total when the asset menu involves T-bills, stocks, bonds, and REITs.

8 Some of the studies in the asset allocation literature which use similar values for the coefficient of relative risk aversion are Umar, Shehzad, and Samitas (2019), Brandt (1999), Fugazza, Guidolin, and Nicodano (2007, 2009, 2015), and Campbell, Chan, and Viceira (2003).

9 Most studies implementing an expanding window analysis for obtaining portfolio weights such as Diris, Palm, and Schotman (2015) and Fugazza, Guidolin, and Nicodano (2015) report a downside to this method in the form of few non-overlapping observations, also for long horizons. This also applies to our study. Although this could be partially mitigated by extending the dataset further back prior to January 1972, this can only be done for the asset menu excluding REITs as the data on REITs is available only starting from January 1972. This would render the comparative analysis across the two asset menus inconsistent. Hence, we stick to our sample period which spans from January 1972 to December 2019 thus allowing comparative analysis across both asset menus, excluding and including REITs.

10 We have experimented with VAR (2) models but with a shorter list of predictors essentially finding qualitatively identical empirical results. Because adopting a VAR (2) considerably increases the computation burden, we have refrained from reporting full results here. However, these are available upon request from the authors.

11 See Cogneau and Hubner (2009) for a survey of the various portfolio performance measures used in the asset allocation literature.

12 Since it is not feasible to use realized utility in order to compare strategies with different underlying assumptions on the investor’s risk aversion levels and differing investment horizons, we take into consideration another measure of portfolio performance evaluation, CER. Being a utility-based measure, it reflects an investor’s welfare after taking into account the higher order moments of portfolio returns (in our case captured by the investor’s power utility function) with which a long horizon investor in particular is more concerned.

13 Ranking the strategies by CER provides an indication of how each one benefits an investor in terms of the utility she derives from terminal wealth obtained by implementing the strategy. The higher the CER, the more desirable the strategy.

14 The identities of the top 10 models ranked by CER in terms of the combinations of predictors involved are reported in Tables BI (A) and (B) for the asset menu excluding REITs and including REITs respectively in the Online Appendix.

15 For the sake of brevity, we do not present all the persistence test tables for the Bayesian strategies. However, to provide a flavor of the related findings, the tables for ranking by CER test are presented in the Online Appendix [Tables BII (A) and (B)]. We do however provide a brief overview of all the Bayesian persistence test results in the main text in order to compare and contrast these results with those from the frequentist method.

16 Section 5.6 specifically considers the role of the presence of short-sale constraints as a robustness check.

Additional information

Notes on contributors

Prajakta Desai

Prajakta Desai is currently an assistant professor at Bocconi University. She was previously a post-doctoral fellow at the London School of Economics and Political Science. She holds a PhD in Accounting and Finance from Manchester Business School, University of Manchester. Her current research focuses on the capital market implications of accounting choices. Specifically, her on-going research projects are in the areas of accounting-based valuation models, earnings management and accounting fraud, credit risk models and their association with accounting-based firm attributes around U.S. regulations such as Sarbanes-Oxley Act (SOX), sovereign credit risk, corporate social responsibility (CSR), the association between corporate credit risk and syndicated loan market, bond pricing and political affiliation, insider trading, and impact of CFO turnover on firm performance.

Massimo Guidolin

Massimo Guidolin holds a PhD from University of California, San Diego. His previous employment includes the University of Virginia as an assistant professor in financial economics, the Federal Reserve Bank of St. Louis first as a senior economist and then as an Assistant Vice-President (Financial Markets), and Manchester Business School as a chair full professor in Finance. Massimo has published in top outlets such as the American Economic Review, the Journal of Financial Economics, the Journal of Econometrics, the Review of Financial Studies, and the Economic Journal. He serves on the editorial board of the Journal of Economic Dynamics and Control, the Journal of Financial Econometrics, and the International Journal of Forecasting. Massimo's research spans across non-linear time series models in finance and macroeconomics, methods and models in forecasting, applied dynamic portfolio choice in the presence of predictable asset returns, empirical option pricing, and asset pricing models with learning and belief dynamics.