8 Conclusion

This dissertation introduces a flexible framework for risk-based investing that is a general tool and allows the user to incorporate different estimation techniques into their risk-based portfolio construction. The framework and risk-based investing are empirically sound when tested with South African data, as all risk-based portfolios outperformed the market weight portfolio using the Sharpe ratio measure. The superiority of risk-based portfolios is consistent with similar results for US equities found by DeMiguel, Garlappi, and Uppal (2007) and Kritzman, Page, and Turkington (2010).

Within risk-based portfolios, we found that GMV portfolios performed the best using the same Sharpe ratio measure. Furthermore, broadening the asset universe to include more assets materially improved portfolio performance.

No techniques outperformed the standard sample covariance matrix technique for finding risk-based portfolios across all risk-based portfolios, which is consistent with the findings of Kritzman, Page, and Turkington (2010). However, they propose longer estimation windows (20-years or 10-years) than were used in this study. Due to the similar performance of shrinkage techniques and standard sample estimation techniques, the period of this study (7.69-years) is close to an empirical break-even performance point between sample estimation and lack-of-data adjusted estimation strategies.

Quantile factor modelling with regime switching worked well for the smaller and more concentrated GMV portfolios, where estimating a small number of substantial allocations correctly is essential. Subset resampling performed well for the less concentrated ERC portfolios, where determining the greatest number of allocations is necessary. Ridge regression shrinkage techniques have historically performed well in other studies. Still, the long-only constraint and a restrictive maximum allocation constraint of 10% have the effect of regularising the optimisation in a manner already shown by JM03. Therefore, the impact of ridge regression on performance is diminished.

8.1 Avenues for further research

Avenues for further research include using the flexible investing framework with other risk-based portfolios and penalising the objective function with alternate functions. The structure may also be modified to include different constraints. Further experiments may also be performed with varying universes of assets. Additionally, the general portfolio estimation procedure could be performed in risk factor space instead of the asset space.

References

DeMiguel, Victor, Lorenzo Garlappi, and Raman Uppal. 2007. “Optimal Versus Naive Diversification: How Inefficient Is the 1/N Portfolio Strategy?” The Review of Financial Studies 22 (5): 1915–53.

Kritzman, Mark, Sébastien Page, and David Turkington. 2010. “In Defense of Optimization: The Fallacy of 1/N.” Financial Analysts Journal 66 (2): 31–39.