Impact of Residential Real Estate on Portfolio Allocations: An Application of AI

Download Prospectus


In the past 18 months, two major themes have been elevated in the public discourse in Australia: house price appreciation and Artificial intelligence(AI). In the area of housing, we have observed the establishment of the Housing Affordability Future Fund[1] and the National Housing Supply and Affordability Council (NHSAC)[2]. These two initiatives aim to shed light on the pressing issue of housing affordability, prompting a revaluation of residential real estate as a viable investment asset class. On the other hand, AI has extended the spectrum of being espoused as a tool that will unlock unrivalled productivity gains[3] to a tool that will result in the demise of human kind[4].

In this note I undertake a thought experiment to examine how one can use tools such as ChatGPT[5] to explore the impact of the inclusion of residential real estate in portfolio allocation choices, a common exercise undertaken by anyone seeking to balance the effects of risk and return. Extant literature (eg Ciochetti et al, 1999) highlights for the US, many private and public pension funds underweight real estate relative to theoretical allocations based on modern portfolio theory. A 2023 survey by Asian Association for Investors in Non-Listed Real Estate Vehicles, highlights that institutional investors allocate 12% and 22% of their real estate portfolio to residential real estate in the Asia Pacific and Europe, respectively[6].

Table 1 reports that Australian residential real estate is worth more than $10.7 trillion or 3.4 times the value of all securities listed on the ASX with over 503,729 sales per annum. Based on these facts alone, one would expect that Australia would have a mature industry built around housing investment. However, this is not the case. Australia real estate market is predominantly characterised as a private market with individual landlords. In Australia the NHSAC[7] identified a number of barriers and recommendations to improve institutional investment including improving data quality, product design and scale of investment.

Table 1 Value of Assets

Residential Real Estate $10.7T
Australian Superannuation $3.9T
Australian Listed Stocks $3.1T
Australian Commercial Real Estate $1.3T
Number of Dwellings 11.1
Outstanding Mortgage Debt $2.3T
Total Sales P.A 503,729
Source: CoreLogic

Data and Methodology

This study utilizes 15 years of monthly historical data, specifically the period extending January 2009 to June 2024. The periods are selected considering two factors. First CoreLogic Total Return Index (ie capital plus yield) data is available from 2005 onwards. Second, we avoid the significant impact of the Global Financial Crisis of 2007/07 which heavily impacted equities during the period. Data is sourced from Bloomberg and CoreLogic, is summarised in Table 2.

Table 2: Asset Classes and Sources

Name Ticker Returns Source
S&P/ASX 200 A-REIT Total ASA5POP Total Bloomberg
All Ordinaries Accumulation ASA30 Total Bloomberg
MSCI World Equities ex-Australia (Net) M1W00 Net Bloomberg
FTSE Australian Government Bond Index CFIIADL Total Bloomberg
S&P Global REIT Total SREITTGL Total Bloomberg
CoreLogic Hedonic Value Greater Sydney Capital CoreLogic
CoreLogic Hedonic Total Return Greater Sydney Total CoreLogic
CoreLogic Hedonic Value Greater 8 Capital Cities Capital CoreLogic
CoreLogic Hedonic Total Return 8 Capital Cities Total CoreLogic

Following modern portfolio methods (Markowitz, 1952) we construct historical logarithmic returns and volatility estimates for each of the assets and report correlations amongst the asset benchmark indices. Sharpe ratios are measured, relative to the average of the FTSE Australian Government Bond Index over the period.

In order to produce efficient frontiers of various portfolio mixes, we make use of OpenAI’s large language ChatGPT[8]. ChatGPT is able to ingest and source data such a stock prices, and then be promoted to undertake various instructions. In the current instance the prompt was: “You are a portfolio manager. Take the attached file which contains monthly index data for several securities. Calculate the monthly logarithmic returns and summarize the annualised arithmetic returns and standard deviation. Prepare a correlation matrix. Prepare an efficient frontier for the following portfolio mixes.[9] Several portfolio mixes were considered to evaluate the impact of the addition of each asset class on the efficient frontier. Specifically, the portfolios created are reported in Table 3

Table 3 Portfolio Mix

Portfolio Benchmarks
1 Australian Equities and Bonds
2 Australian Equities, Bonds and REITS
3 Australian Equities, Bonds, REITS and Global Equities and REITS
4 Australian Equities, Bonds, REITS and Global Equities and REITS
5 Australian Equities, Bonds, REITS, Global Equities, REITS and Australian 8 Capital City Residential Real Estate
6 Australian Equities, Bonds, REITS, Global Equities, REITS and Greater Sydney Real Estate

Findings: Risk and Return Statistics

Table 4, reports average risk and return statistics for each of the indices covering equities, bonds, REITS and residential real estate. Table 2 highlights while in absolute terms equites generates the highest returns, on a per unit of risk measure ie Sharpe ratios Australian residential real estate (either the 8 capital cities, or Sydney) delivers enhanced benefits. Moreover, Australian residential real estate generates favourable capital and total returns[10], especially in light of the leverage effects inherent in the equites and REIT benchmark measures.

Table 4: Risk and Return January 2009- June 2024

S&P/ASX 200 A-REIT All Ords Accum MSCI World Equities FTSE Aus Govt Bond S&P Global REIT Greater Sydney Capital Greater Sydney Total Capital Cities Capital Cities Total
Ann return 9.64% 9.56% 11.24% 2.90% 9.44% 6.31% 9.69% 5.26% 8.95%
Ann. Volatility 19.96% 13.90% 15.33% 5.10% 18.42% 3.56% 3.60% 2.69% 2.68%
Sharpe 0.36 0.51 0.58 0.38 1.00 2.01 0.92 2.40

Table 5 reports the correlation matrix. Unsurprisingly we observe a high degree of correlation amongst equities and REITS, in addition to the defensiveness of Government bonds. Table 5, however also highlights the defensive nature of residential real estate, which exhibits low correlation coefficients with the other asset classes. Interestingly, we also observe a very high correlation between Sydney and the 8 capital cities index.  On average the Greater Sydney index represents between 40-45% in any given month of the value weighted 8 Capital City Index.

Table 5 Asset Class Correlation Matrix

  S&P/ASX 200 A-REIT All Ords Accum  MSCI World Equities FTSE Aus Govt Bond S&P Global REIT Greater Sydney Capital Greater Sydney Total Capital Cities Capital Cities Total
S&P/ASX 200 A-REIT 1.000
All Ords Accum 0.777 1.000
 MSCI World Equities 0.585 0.776 1.000
FTSE Aus Govt Bond 0.317 0.049 -0.014 1.000
S&P Global REIT 0.761 0.767 0.821 0.161 1.000
Greater Sydney Capital -0.011 0.009 0.086 -0.090 0.043 1.000
Greater Sydney Total -0.004 0.015 0.088 -0.074 0.049 0.997 1.000
Capital Cities 0.014 0.028 0.093 -0.134 0.070 0.962 0.951 1.000
Capital Cities Total 0.019 0.033 0.096 -0.121 0.075 0.968 0.962 0.997 1.000

Findings: Efficient Frontiers

Figure 1 plots the efficient frontier for the various portfolios reported in Table 3. Common to Panels A-D are the impact of constructing a portfolio of Australian bonds, equities and REITS and later introducing their global equivalents. We report a shift in the efficient frontier north-west, highlighting the benefits on portfolio performance relative to risk by introducing assets exposed to a more diverse range of market factors. Correspondingly, Figure 1, highlights the inclusion of residential real estate has similar, albeit it larger, directional effects.

Figure 1

Panel A: Equities, Bonds, REITS + Australia 8 Capital City Real Estate (Capital)

Panel B: Equities, Bonds, REITS + Australia 8 Capital City Real Estate (Total)

Panel C: Equities, Bonds, REITS + Sydney Real Estate (Capital)

Panel D: Equities, Bonds, REITS + Sydney Real Estate (Total)


The paper leverages artificial intelligence to demonstrate how industry can now easily analyze data and uncover insights into various asset classes. Through such analysis, we identify the significant role residential real estate can play in expanding the efficient frontiers of investment portfolios. Unanswered questions include what investment structures and financial engineering must be developed to provide an investible residential real estate index, vis-à-vis equites and bonds which have an established exchange traded fund product suite. Further is there a role for active based management of residential real estate. Anecdotally one could envisage influencing factors such as location, dwelling characteristics and tenancy. Given the ease and interactive ability of AI, to apply established portfolio construction techniques developed in the academic literature, this author sees a role for the use of AI in the investment decision making process.


Ciochetti, B., Sa-Aadu, J. & Shilling, J.(1999) Determinants of Real Estate Asset Allocations in Private and Public Pension Plans. The Journal of Real Estate Finance and Economics 19, 193–210

Markowitz, H. (1952). Portfolio Selection. The Journal of Finance7(1), 77–91.


The information on this website is for general information only. It should not be taken as constituting professional advice. You should consider seeking independent legal, financial, taxation or other advice to check how the website information relation to your unique circumstances. The author is not liable for any loss caused, whether due to negligence or otherwise arising from the use of, or reliance on, the information provided directly or indirectly, by use of this website.

[1] A $10 billion fund dedicated to support social, affordable an acute housing needs of Australians

[2] Appointed in January 2024 includes members from industry, government and academia



[5] Other common Ai tools such as Claude and Perplexity, can also assist.



[8] The author has validated the outputs of the OpenAI tool in excel to triangulate conclusions.

[9] ChatGPT was able to produce the analysis in the space of 2-minutes.

[10] Note the total return for residential real estate assumes that daily rental income is reinvested into the hedonic index daily. Transaction, taxation and ownership costs associated with purchasing and owning residential property are not considered (along with individual tax concessions such as negative gearing).


Dr Vito Mollica,
Macquarie University