How disruptive can Smart Beta be to the South African Active Fund Management Fraternity?

ABSTRACT

This research paper explores the theme of Smart Beta and the current state of inefficiency of active fund fee structures in South African domestic equity unit trusts. The emerging understanding is that many of the latent sources of value-add (or alpha) of active fund managers are currently accessible in cheaper form via Smart Beta products. Smart Beta products use mechanical and automated rules to establish exposures to tradeable instruments that emulate many of the understood and replicable themes in current active asset management. We sample 91 well-known general equity funds along with nine local Smart Beta funds and demonstrate how disruptive Smart Beta products could be to the fee structures of many of these active funds. We do this by mapping the reproducible elements of the active return of these funds to fungible Smart Beta factors. We conclude with five broad predictions around the active fund management industry in South Africa. Additionally, we focus on how active managers might prudently align themselves with an understanding of what aspects of their value-add are not replicable in order to persist and thrive.

KEYWORDS:

• Smart
• beta
• South Africa
• active
• fund
• management

Acknowledgements

The authors would like to thank Ronald Kahn for his invaluable guidance and time. We have omitted fund labels from our figures since our intention here is to point out the disruptive potential of Smart Beta to South African active managers, rather than to name-and-shame. The authors will happily respond to specific fund placements upon request.

ORCID

Jean-Jacques Duyvené de Wit http://orcid.org/0000-0002-9473-4210

Daniel Polakow http://orcid.org/0000-0003-2512-5638

Notes

1. See, for example, Kahn and Lemmon (2016), Melas (2016), Lo (2016), Anson (2015) and Carhart, Cheah, De Santis and Litterman (2014).

2. In an index fund, these are principally related to the cost of the initial purchase, cost of rebalancing and management fees.

3. This is also termed the Security Characteristic Line, where returns are excess returns above the risk-free rate.

4. Internationally, see Pogue and Solnik (1974), Scholes and Williams (1977), Marsh (1979), Black (1993) and Kim (1993). For a South African context, see Bowie (1994), Bowie and Bradfield (1998), van Rensburg and Robertson (2003), Auret and Sinclaire (2006) and Strugnell, Gilbert and Kruger (2011).

5. In a regression context such as the one utilised here, alpha would refer to the historical performance over-and-above the nominated benchmark. It is commonplace, although not often correct, to infer that the performance of active funds will persist going forward. The intention in this research is neither to impute nor test for persistency. Rather, we conduct our study using historical data on an ex post basis alone.

6. For key historical examples, see Arnott, Hsu and Moore's (2005) work on Fundamental Indexation and Hasanhodzic and Lo’s (2007) work on hedge fund replication.

7. In statistics, the coefficient of multiple determination, , is a statistic that is produced in a multiple regression. Its value can range from 0 (where none of the variance of the response variable is explained by the predictor variables), to 1 (where all of the variance of the response variable is explained) (Rice, 2007).

8. 91 Funds have the requisite amount of historical data and known Total Expense Ratios (TERs).

9. In line with the mandates of general equity unit trust funds, we did not allow gearing by restricting the sum of the betas between −1 and 1.

10. Statisticians are aware that while a significant F-statistic will typically be associated with some of the individual coefficients being statistical significant (using a t-test for partial regression coefficients), it is possible to have a significant F-statistic without any significant t's, or even significant t's without a significant F-statistic (Geary and Leser, 1968). However, the latter situations only occur when there is a high degree of correlation between the independent variables. We obviate this concern through our forcing of independent variables (our Smart Beta factors) to be selected only if orthogonal (see Addendum A).

11. It could be argued that the 16 funds that did not display statistical significance, and that were excluded from the reporting, may actually represent true non-replicable returns and should be evident on Figure 2 as low R² funds. We omit the non-significant regression models from the analysis, as they pose an epistemological challenge representing significant models (many of which have low-end R² values) alongside non-significant models. We do note the average TER of non-significant funds is 1.84% with a stdev of 0.64% with a mean R² of 16%. The general patterning in Figure 2 and conclusions based off the same are unaffected by exclusion of such non-significant funds. The only metric that changes is that the aggregate R² would be 39% rather than the 44% reported in the ‘Results and discussion’ section.