Logistics, valuing sectors in listed markets using ML
We illustrate how machine learning techniques can enhance valuation methodologies of sectors in listed real estate markets and how investors can incorporate changing market conditions when considering sector tilts or underwriting of individual stocks. Sector tilts are an important source of alpha for investors and predicting future performance prospects is a critical part of the investment process. We illustrate how ML and econometric models can be utilised for both spot valuations and to form data driven expectations about future performance.
Given the value of real estate, including listed real estate, is closely tied to economic activity (demand and supply factors such as employment and new construction), we are primarily interested to evaluate sector dynamics from a top-down macroeconomic perspective rather than bottom-up stock information such as company valuation multiples. Also, we are interested in establishing structural dynamics and drivers for a sector within the investable universe of other reit sectors or more generally as compared to the overall reit market such as a benchmark. We use monthly data for US to model industrial/logistics reits as defined by FTSE EPRA NAREIT US Industrial index and FTSE EPRA NAREIT US Index as benchmark since December 2009.
Figure 1 A illustrates the performance of industrials/logistics and market in absolute terms (indexed) and B the relative performance of industrials/logistics
Figure 1A Industrials (blue) and benchmark (white)
Figure 1B Industrials relative to benchmark
Industrials/logistics have of course seen strong growth over the past decade as consumption and retail activity has migrated structurally towards online sales creating a need to increase real estate and distribution capacity and expand employment in the sector etc to meet this shift. As figure 1B illustrates, in listed markets this shift accelerated significantly around 2015/16, with half a decade of consistent growth and strong relative performance (slope) until COVID. Post initial COVID-related external shock (spike) the sector has stagnated to some extent in terms of relative trend and has also seen sharp market oscillation within a much wider range than before.
These changing dynamics, change in trend and higher volatility, make sector valuation more challenging. Constructing econometric models that reflect structural demand and supply conditions within the sector compared to demand and supply conditions within the general economy (impacting overall reit market/benchmark) can provide valuable information to assess if the sector is priced rationally as expected or if market pricing is driven by behavioural aspects or other overreactions beyond what macroeconomic conditions would suggest.
We use ML to select variables and to fit an econometric model incorporating a range of sector and economy wide aspects and asset pricing conditions. As figure 2 illustrates the model fit is very strong, R-sqr is above 0.95, this is partly due to the trending nature of the data but as illustrated the model specification correctly identifies key periods;
the change in trend around 2015/16 when industrials/logistics relative performance accelerated; and
correctly captures the post-COVID stagnation and importantly sector fluctuations during this period.
The lower part of Figure 2 illustrates the difference between actual and predicted performance or “fair value range” with the upper (red) and lower (green) boundaries (1 st.dev) identifying times when the sector has traded outside of what macroeconomic conditions suggest, i.e. identifies time periods of relative sector deviations and potential opportunities to capitalise on those typically shorter-term dislocations. In addition, the model includes lagged variables which means that it allows us to form expectations of future sector performance ca 6-12 months out (not shown, see below).
Currently, industrials/logistics trade at the lower boundary of “fair economic value” from a spot valuation perspective.
Changing impact of performance drivers
As the time frame used includes meaningful changes in market conditions, we split the sample into two time periods to measure any changes in market pricing dynamics, P1. pre-COVID, strongly trending market and P2. the more range bound and oscillating post-COVID period. Modelling the two periods separately but keeping the model structure the same identifies changes in variable impact and coefficient sensitivities that might be useful for both listed and private markets real estate investors to form expectations going forward, those are:
sensitivity to demand conditions such as increases in online retail activity and employment has fallen by ca 30% across our demand variables, implying that further growth in e.g. share of online retail sales is likely to have lower impact on the sector going forward
sensitivity to supply conditions has increased by ca 50%
sensitivity to asset pricing conditions such as interest rates has increased by over 70%, nearly doubled
Figure 2 Actual (black) and model (blue) values with difference confidence interval ranges
These changing market dynamics and pricing conditions of publicly traded logistics assets will also impact private markets asset performance and valuations.
If you are interested to discuss our 6m forward-looking view on the sector and other sectors, please contact info@kaniaadvisors.com
About Kania Advisors
Kania Advisors is an independent research and advisory firm focused exclusively on institutional real assets allocations and investment programmes. We provide advice and solutions to improve outcomes in real assets investment programmes. We conduct detailed industry research and custom studies typically focused on quantitative analysis and provide insights which form a critical part of a client's decision process.