What is the optimal offshore allocation for South African investors?

ABSTRACT

In 2022, the South African Pensions Fund Act was changed to allow funds to allocate up to 45% of their portfolio to offshore investments. This is a material change to fund regulations and naturally prompts the question: what is the optimal offshore allocation? In this research, we implement two allocation frameworks commonly used in practice and show that such a question has no universally optimal answer and that any solution will be determined principally by five investment factors, with the most important of these being an investor’s return objectives or risk limits, and the investor’s ability to hedge currency risk. Based on these factors, we suggest practical guidelines to aid in setting strategic offshore allocation policies.

KEYWORDS:

• Offshore investment
• asset allocation
• portfolio optimisation
• currency hedging

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 See the South African Reserve Bank Exchange Control Circular No. 10-2022 for more detail.

2 There was an additional allowance of up to 10% in African assets, although this remained largely unutilised.

3 In contrast to the plethora of global studies, South African-specific research on the topics of international diversification and offshore investing remains scarce. Some notable early work includes the studies of Barr (1986), Swart (2004), Polakow and Gebbie (2008), and van Heerden and Koegelenberg (2013). See Bradfield and Munro (2015) for further references. The common theme in both local and global studies though is that the long-run case for international diversification remains compelling, even in the event of strong local market performance (see e.g., Asness et al., 2023).

4 We fit a factor-selection augmented best subsets regression model (FS-BSR) to rolling 60-month windows, with the four factors including global equity, bond, and commodity returns in USD, USDZAR currency returns, and all interaction terms. See Polakow and Flint (2015) for full details.

5 Appendix A details the data series utilised in our analysis.

6 Estimation enhancements such as shrinkage (Jorion, 1985; Ledoit & Wolf, 2004) may also be applied here.

7 Log, or compound, returns are additive over time and thus projection is done by simple linear scaling. Convertion to linear return inputs is then achieved by employing standard equations to transform joint lognormal moments into joint normal moments. See Appendix A in Meucci (2001) for full mathematical detail.

8 Appendix A details the data series utilised in our analysis.

9 The pseudo-random algorithm is based on the work of Shaw (2010) and ensures sufficient coverage of the possible portfolio space, while the rejection sampling step ensures only Reg28-compliant portfolios (excluding the offshore limit) are retained.

10 Block bootstrapped returns are preferred over rolling historical returns as it significantly increases the number of returns available for analysis and also removes any overlapping data issues (Boudoukh et al., 2019).

11 The true adjustment is based on the geometric rather than arithmetic differential in rates, although the difference is generally insignificant when considering short hedging periods.