Investment decision-making under economic policy uncertainty

ABSTRACT

It is widely established that economic policy uncertainty (EPU) affects investment decisions and performance, yet research in this area has overlooked the direct property investment market. This article seeks to rectify this and proposes a multistage multilevel analytical framework to offer new insights and a richness of findings. Using a news-based measure of EPU in the United Kingdom, and controlling for economic conditions, a national-level analysis reveals some evidence of Granger-Causality between EPU and total returns, indicating that pricing is responsive to uncertainty. These findings suggest that EPU is an important risk factor for direct property investments, with pricing implications. Differences in data and performance measure are important, however, with income returns unresponsive. A micro-level investigation begins to reveal some of the asset-pricing decisions underpinning the national results, indicating investors’ concerns for income streams are consistently high, regardless of varying EPU. Pricing can also cause changes in EPU, such as in the retail and industrial markets (increasingly linked through logistics) reflecting sector-specific stakeholder groups and newsworthy issues. This evidence highlights how important it is for policy-makers to understand the complex and bi-directional relationship, that indecision can undermine investment confidence and cause investment market volatility, in turn raising EPU.

KEYWORDS:

• Economic policy uncertainty
• commercial real estate
• investor decision-making
• investor behaviour
• conjoint analysis

Acknowledgments

The authors would like to thank the respondents who took part in the survey, without whom such a study would not be possible.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental data

Supplemental data can be accessed here.

Notes

1. Johansen cointegration trace and Max-eigenvalue tests indicated no cointegration at the 0.05 level, implying that it was suitable to employ a VAR modelling framework for the differenced data.

2. The dummy Econ_t variable enabled time points when the market was contracting to be categorised and avoided potential orthogonal matrix issues associated with the inclusion of EPU and a continuous variable measuring economic conditions.

3. Lizieri et al. (2012) found that their TAR-TAR model outperformed conventional AR smoothing techniques which underestimate the variance of the true underlying series. However, for fullness and comparison, the valuation-based indices have also been desmoothed using the first order autoregressive (AR(1)) process proposed by Geltner (1993). The results are highly similar, although it is interesting to note that, where there are differences, the results using the AR smoothing technique are consistent with the results obtained using the raw data.

4. Selection of these regime determinants was based on the minimisation of the sum of errors and Aikake Information Criterion as advised in Lizieri et al. (2012).

5. In the case of retail income returns, they had to be second differenced before becoming stationary.

6. However, a shift in credit risk perceptions in the general economy was evident during this study period as the TED spread had started to rise in April/May.

7. Software developments allowed the CBC questionnaire to be delivered online, and this gave rise to some possible differences in completions. By using an online tool, it was possible for the respondents to start but not complete the survey, whereas this was encountered only once in the first period. In both survey periods, respondents worked through the tasks alone, although in 2006 the researcher was in the room. By contrast, in 2016, the respondents may have rushed or, indeed, have taken greater care when completing the survey. No obvious patterns were found when interview and question completion times were reviewed across the respondents of both surveys.

8. In 2007, the sample was developed and contacted in two stages, increasing the number of respondents iteratively, until resources were exhausted and sufficient responses achieved. In 2016, due to the use of the online survey method of data collection, a larger sample was needed, and thus, in total, a sample of 377 was drawn, giving an achieved sample of 336 after allowing for fund managers moving company, funds or position, on maternity/paternity leave and erroneous contact details. From this, a response rate of 15.5% was achieved.

9. Multinomial logit (MNL) analysis, which examines the relative importance of attributes by considering the difference each attribute could make to the total utility of a real estate asset, is used in a preliminary analysis to check the robustness of the results and thus the appropriateness of proceeding with the HB analysis.

10. The top-level model, a multivariate logit model with random effects to allow parameters to vary across individuals, derives sample averages from 51 respondents, each performing 20 tasks. This prior and posterior information from the between- and within-group estimations informs the likelihood provided by the lower level model as specified in Equation (7). A total of 1020 observations were collected at each survey point enabling segmentation and analysis, and as recommended with relatively small samples design, efficiency tests were undertaken before and after the fieldwork to ensure the survey design was efficient.

11. The standard MNL model is specified as $p_k=\frac{e^{(U_k)}}{e^{U_1}+e^{U_2}+e^{U_3}}$ where the probability of selecting a specific investment is proportional to the total utility for that concept ($U_k$), estimated by adding the utility associated with each attribute level, relative to the total utility for the three options available (Sawtooth Software, 2009b).

12. The fund strategy was collected from respondents, but the results cannot be disaggregated further by this additional variable due to the resulting small sample sizes.

13. The results in the table were generated using Equations (5) and (6) and fully align with those generated by testing the initial version of the model represented by Equations (3) and (4). The market rental growth index was also tested but found to generate no significant Granger-Causality results.

Additional information

Funding

The second stage of this study was supported by the RICS Research Trust [495].

Notes on contributors

Cath Jackson

Cath Jackson is a senior lecturer in the Department of Urban Studies and Planning, University of Sheffield. Her research explores investment decision-making at stock and local market levels, with particular interest in retailing markets.

Allison Orr

Allison Orr joined Urban Studies at the University of Glasgow in June 2006 as a senior lecturer. She has extensive experience modelling the pricing of commercial and residential property markets, and researching urban change and decision-making in the direct investment market.