Real Estate ‘Value’ Stocks and International Diversification

Abstract

In recent years there has been an increased interest in the extent to which managers can improve their property portfolio position through international diversification. Much of this interest has centred on the use of various statistical/econometric tests of time‐varying correlations and long‐run equilibrium positions using whole of country property indices. In this paper, a short‐run tactical asset allocation approach to securitized property is adopted. Using neural network methodology, a neural network model that ‘learns’ well‐established rules of portfolio investment is built. The model uses a set of individual property companies across three of the most highly securitized property markets in the world viz. the US, the UK and Australia. The standpoint of a UK investor is adopted and the model is asked to compare portfolios constructed purely from domestic assets with portfolios constructed from internationally held assets allowing for foreign exchange adjustments. When the foreign exchange risk is actively managed, the outcomes from the analysis suggest that the gains from hedging are conditional on both the return to the unhedged position and the volatility of the underlying currency being hedged.

Acknowledgement

This research was supported by a REGS grant from the University of Technology, Sydney.

Notes

1. This estimate is obtained from UBS, Thompson Financial Datastream via Courtney Darby ‘The Economic Impact of Establishing a Listed Real Estate Industry’ in The REIT Influence, Nov./Dec., 2005, available on the NAREIT Real Estate Portfolio Management website.

2. For example, Liu and Mei (1998) analyse the possible integration of real estate markets and stock markets across a number of countries. The researchers find there are diversification benefits, but these are primarily driven by unanticipated returns that, in turn, are partially driven by changes in exchange rate risk. They find that from a US investor's viewpoint, investing in international real estate securities provided additional diversification benefits over and above that associated with holding international stocks.

3. A Datastream calculated Index, the ‘Real Estate’ series is based on the FTSE classification and includes the following sub‐sectors: Real Estate Development, Property Agencies, and Real Estate Investment Trusts.

4. As provided by Datastream International.

5. Equation (1) in this paper differs from the Datastream model for total return only in that the Datastream model is based on a daily frequency and uses 260, rather than 12, in the denominator.

6. This data and the remainder of the paragraph are a summary of the information available at http://www.asx.com.au/investor/.

7. A REIT‐like structure is now in place in the UK, but the vehicle is outside the study period.

8. As well as binary node models, a set of linear node models are also tested. These showed no significant differences. As opposed to the 0/1 output of the binary model, the linear model develops an output set of discrete numbers between zero and one such that the output may represent the ‘degree of value’.

9. The software used for the analysis is the Braincel Neural Network software version 3.62. Mr Gideon Isaac, Technical Support Unit, Promised Land Technologies (Gideon@micro‐net.com), confirms that the tanh transfer function is used in both input and hidden layers.

10. A large training set was used to ensure stability of outcome. As noted by one referee, changes to learning period (among other factors) can impact predicted outcomes (in much the same way as changes in sample size or period of analysis can impact the estimation of regression coefficients). To assist stability approximately half the data set was used to train the NN model.

11. While the research in this study uses data available from Datastream International at time t‐1 to identify value stocks at time t, the authors acknowledge the comments by an anonymous referee that ‘investors would not have had access to data which in fact were not available for sometime after’ and that this may ‘upwardly bias the observed quality of the trading rule performance’.

12. Source, OECD Main Economic Indicators. UK annual average equals 5.3%; US annual average equals 5.2%; Australian annual average equals 5.9%.

13. For space conservation reasons, the autocorrelations are not shown but are available from the authors by request.

14. Value portfolio z‐scores and p‐values are available from the author's by request.

15. The hedged beginning period value for Australian value portfolios is £0.9452 and for the US £0.9766. These amounts are calculated as the £1 initial investment less the initial option premium expressed in GBP.

16. An anonymous referee suggested that a useful comparison might be between the outcomes from a neural net model and discriminant analysis or a logit/probit model. This represents a potentially useful suggestion for future research to be carried out either by the authors or by other researchers. The referee also noted that lack of comparison actually limits the conclusions that can be drawn. That is, are these results ‘… due to the variables employed or to the way that the neural network non‐linearly combines them together?’

17. As noted above, we are grateful to an anonymous referee for emphasizing this point.