The optimal allocation to Australian direct real estate: A dissertation submitted in partial fulfillment of the requirements for the degree of Master of Property Studies at Lincoln University

Abstract

The impact of uncertainty on the strategic allocation to Australian direct real estate is examined for a mixed-asset portfolio consisting of asset classes most commonly held by Australian superannuation funds. The asset opportunity set comprises world equities, world government bonds, Australian industrial and resource equities, listed property trusts, Australian fixed interest, and direct real estate. Assuming the maintenance of price stability, an ex post timeframe is used to derive expected risk and inter-asset correlations for each asset class. Three sources of return data are used for real estate: appraisal-based, listed property trust, and unlisted property trust returns. In addition, three desmoothing procedures are used to derive corrected returns for each source of real estate data. Since all three sources of real estate data are based on large property samples and because fund managers can rarely match the level of economic diversification that exists in these indices, it is important to consider the impact of small real estate portfolios on the allocation decision to real estate. To that end, real estate portfolios comprising one, five, ten, twenty-five, and fifty properties are considered. The use of Bootstrapping and Latin Hypercube resampling results in a dynamic portfolio selection model where a range of portfolio solutions are optimal versus point-estimate allocations. The incorporation of the risk-free asset in the decision space results in a simplified Mean-Variance (MV) asset selection model where one risky-asset portfolio solution is optimal for all investors, regardless of their risk aversity level. The main findings of this study include: confidence intervals for direct real estate and various other asset weight vectors were so wide that point-estimate allocations are rendered effectively useless; the high level of fuzziness surrounding optimal solutions may limit the practical application of this technique to the mixed-asset allocation problem; a mean simulated allocation of 52.58% can be justified for direct real estate when raw appraisal-based, composite, data is used to derive the risk and pair-wise correlation parameters for this asset class; at a composite level, the potential over-allocation to direct real estate from using appraisal-based data to derive risk and pair-wise correlation inputs for this asset may be in the order of 5.59 to 38.62% based on mean simulated allocations; point-estimate allocations were subject to both upward and downward bias vis a vis mean and median simulated weight vectors; mean simulated allocations for less than perfectly diversified real estate portfolios ranged between 1.09 and 34.59% for portfolios comprising one, five, ten, twenty five, and fifty properties respectively; and the key variables affecting the allocation to direct real estate included the risk estimate for real estate, real estate's risk premium, the risk premium for Australian fixed income, and the risk estimate for Australian fixed income.