Featured in PREA Quarterly Spring 2022 issue:
“When full employment is reached, any attempt to increase investment still further will set up a tendency in money-prices to rise without limit…i.e. we shall have reached a state of true inflation.” — John Maynard Keynes[1]
With an unemployment rate of 3.6%, year-over-year job openings growing at 7.1%, and average monthly nonfarm payroll gains exceeding 500,000 in the past year, the US economy arguably has reached a state of Keynesian full employment and true inflation. Signs of inflation are pervasive in year-over-year statistics: the headline Consumer Price Index (CPI) rose by 8.3% in April, the GDP deflator increased 8.0% in 1Q22, and West Texas Intermediate crude oil prices breached $100 per barrel earlier this year for the first time since 2014. Real estate costs have not been invulnerable to these macroeconomic shifts. According to one prominent index, construction costs increased by 7.1% year-over-year as of 1Q22.[2] Additionally, the Bureau of Labor Statistics reported that total compensation for individuals in the real estate, rental, and leasing sector increased by 5.2% in 1Q22 relative to 1Q21.
Conditions not seen since the early 1980s now confront real estate investors, who want to know how higher inflation impacts cash flows, valuations, and returns. The preponderance of empirical studies Heitman has reviewed on private real estate and inflation shows that real estate has been at least a partial inflation hedge. However, incorporating that conclusion into a broader risk, return, and asset allocation framework adds another layer of complexity. Uncertainty abounds in the current environment with respect to the direction of the economy, inflation, and real estate returns, and investors need a method that incorporates both their informed views and the dispersion surrounding their assumptions. Heitman has developed a framework to address these issues.
Heitman views the contour of macroeconomic consensus forecast distributions as a useful starting point for thinking about market expectations. Further, real estate market participants can use these distributions—with proprietary adjustments—in a forward-looking simulation analysis to understand more thoroughly inflation risks and their impact on real estate.
Real Estate as a Partial Inflation Hedge
When inflation was raging in the late 1970s, Eugene Fama and William Schwert set out to determine the extent to which stocks, bonds, and real estate were inflation hedges.[3] In the pre-NCREIF era, the authors relied on the housing component of the CPI to conduct tests for residential real estate and found that it was a “complete” inflation hedge (i.e., it served as a hedge against both expected and unexpected inflation). New data and econometric techniques have flourished over the past 45 years since the article was written, but its methods are still applicable.
In that 1977 paper, Fama and Schwert disaggregated inflation into expected and unexpected components, drawing inspiration from Irving Fisher’s famous equation: nominal interest rate = expected real return + expected inflation rate. Empirically, the proxy for expected inflation was the Treasury bill yield, and the stand-in for unexpected inflation was the inflation rate minus the Treasury bill yield. If the regression coefficient for expected inflation and that for unexpected inflation were indistinguishable from 1.0, then Fama and Schwert deemed the asset to be a “complete hedge against inflation.”
Taking inspiration from these assertions, Heitman reproduced a fragment of the study using NCREIF data and slightly different methods. [4] Exhibit 1 stacks NCREIF Property Index (NPI) return sensitivities to expected and unexpected inflation on a rolling, trailing five-year basis over the past few decades. The exhibit highlights that real estate is a partial inflation hedge, on average, and that its inflation hedging abilities are time varying. It may be that the results from this chart (and many modern academic studies) reach a more modest conclusion relative to that of Fama and Schwert (real estate as a partial instead of a complete inflation hedge) in part because of the underlying indices used. [5] But it is important to note that real estate also seems to be a good hedge against inflation when inflation levels are high, although data is limited. Given estimation error and other limitations, investors should also consider commonsense or theoretical models to supplement these types of calculations. For example, the presence of explicit CPI provisions, percentage rent, and other features of leases and real estate itself should create at least some inflation protection. In addition, real estate investors may wish to incorporate these kinds of foundational assumptions into a broader risk, return, and asset allocation framework. The remainder of this article progresses toward a simplified illustration of one such synthesis.
A Non-Normal Environment for Inflation
Many economic and financial variables exhibit non-normal return distributions, and inflation is no exception. Since 1948, US CPI inflation on a year-over-year basis has hovered in the 0% to 5% range in just under 75% of months. That means more than 25% of the time, inflation might plausibly be characterized as uncomfortably high or low (outside the 0% to 5% range), as demonstrated in Exhibit 2. The chart shows heavy tails and a conspicuous long right tail of the distribution, which explains why mean inflation since 1948 is 3.5% while the median is only 2.8%. For comparison purposes, a simulated normal distribution with a mean of 2.8% is also included in the graph.
For purposes of forward-looking decision-making, perhaps just as interesting is the fact that macroeconomic consensus forecast distributions may exhibit unexpected shapes. Often, the configuration of consensus forecasts for GDP, inflation, unemployment, the federal funds rate, the ten-year note, and other variables are surprising. For example, given the long right tail in historical CPI percentage changes shown in the exhibit and the uncertainty about the path of future inflation, a nontrivial number of forecasts might be expected to differ significantly from the middle of the pack. But for 4Q22, inflation forecasts in the tails are not dramatically different from that implied by a normal distribution. In part, this phenomenon occurs because consensus forecast distributions represent measures of central tendency without variance around the forecasts. Even so, the amount of certainty rooted in consensus forecasts might be surprising to some market participants.
Why does the contour of consensus forecast distributions matter? Because the underlying assumptions represent the views of some of the largest financial institutions with extensive sway in markets. Rather than starting from scratch, real estate investors can use the distributions to begin a conversation about initial capital market expectations. For example, if a real estate investor believes that the consensus distribution for CPI does not sufficiently capture inflation risk, then the analyst might create macroeconomic suppositions with an inflation distribution that is positively skewed (long right tail) relative to that of consensus. Subsequently, the investor could take additional steps toward a broader framework.
Steps Toward a Quantitative Procedure
An approachable quantitative procedure could advance along these five steps:
- Gather macroeconomic consensus forecast distributions for the applicable variables and other relevant data.
- Fit the individual distributions along with any adjustments for a “house view.”[6]
- Specify how the variables are related to one another.
- Simulate all variables forward over the given horizon.
- Analyze risk, return, and portfolio construction.
Even though these steps entail some complexity, the procedure is approachable in the important sense that the analysis can be performed in a spreadsheet without the need for more advanced econometric software.[7]
Example: Inflation Risk and Real Estate Price Changes in a Dynamic Framework
Gather
Returning to the Keynes quote at the beginning of the article can help frame a model for purposes of illustration.[8] Namely, the much-discussed time-varying, short-term trade-off between the inflation rate and the unemployment rate is used as the basis for a rudimentary macroeconomic framework. Inflation and unemployment may be combined with changes in real estate prices to observe how the three variables evolve over time and influence one another. Specifically, the headline CPI, the unemployment rate, and commercial property prices are used and information over the historical period December 2000 to February 2022 is gathered.[9] Moreover, the macroeconomic consensus forecast distributions for the period 4Q21 to 4Q22 are analyzed because the analysis of a one-year period is straightforward. On their own, the mean and median for inflation and unemployment forecasts are fairly close to each other, suggesting somewhat symmetrical distributions.[10] In this example, the shape of the distribution for real estate is based on the historical contour of price changes, but PREA’s Consensus Forecast Survey can be used in future research.
Fit
A real estate investor either may adopt the general shape and central position of the forecast distribution or may adjust it. If a market participant has no special insight into inflation or unemployment forecasts, then the general form of the consensus distribution can be assumed. This option may be appropriate for many real estate observers who do not generate proprietary forecasts of macroeconomic variables. Because the forecast distributions represent individual estimates of the central outcome, they may understate the true dispersion in the forecast. Therefore, even if the shape of the forecast distribution is adopted, there is room to increase the weight in the tails to account for more risk.[11]
Some institutions have their own forecasts or prefer to weight specific forecasts more heavily (perhaps in a way that changes the configuration of the distribution). In such a case, the institution should use its own house view instead of embracing the more passive approach described. But investors should be mindful of overconfidence bias when doing so. For purposes of this example, the general shape of consensus forecast distributions is considered as a starting point, but based on the shape of the CPI historically and a hypothesis, just for illustration, that inflation surprises will be on the upside, subjective distributions are created by variable and by month that assign upside risk to inflation and the unemployment rate. The historical distribution for real estate prices is negatively skewed (long left tail), and this general shape is applied for the forecast.
Specify
Correlation analysis is one useful method for understanding the relationship among variables. But static correlation matrices should be used with caution for a variety of reasons.[12] For purposes of this example, a couple of basic adjustments are included to increase the likelihood that the correlation estimates are useful, but more sophisticated adjustments should be contemplated. Specifically, all three variables are normalized before calculating correlations, and correlations from different time periods are applied (50% probability of the correlation matrix from April 2020 to the present, 30% from 2011 to March 2020, and 20% from 2001 to 2010; only one correlation matrix is used for each simulation run over the one-year forecast horizon). Finally, each variable is related to its past values through a regression model.[13] In summary, real estate prices, inflation, and unemployment are related to their own past values as a baseline forecast; forecast distributions around this baseline are applied by variable and by month, as described in the “Gather” section; and the normalized variables are related to one another through correlation matrices (more recent correlations are given a higher weight).
Simulate
Changes in real estate prices and inflation and differences in the unemployment rate are all simulated forward over a one-year period.[14] Importantly, correlated variables are simulated forward; the model does not use indices independent of one another. In other words, uncertainty is integrated into the framework but in such a way that allows for dynamic relationships among the variables.
Analyze
The results convey that the average real estate price appreciation over the forecast horizon is 3.3%, CPI inflation is 7.5%, and the increase in the unemployment rate is 260 basis points (bps). Inflation and unemployment risk are on the upside, but the average and median for real estate are quite close (a more symmetrical distribution). If these results are to be believed, portfolios could be tilted to favor assets with shorter lease terms, downside protection, and related attributes. However, pricing is key in the portfolio construction process because savvy investors with a similar view may have already caused market pricing to reflect these valued attributes.
The most likely outcome is that the average of the simulated forecasts in Exhibit 4 will be wrong. But even though these forecasts are quite likely to be wrong, the method outlined can be used to appreciate inflation and real estate risk more thoroughly and identify other areas in which underappreciated risks should be acknowledged. Wrestling with the assumptions and mechanics of the model can be just as important or more important than observation of the forecasts themselves.
A discrete probability table is another way to display the results. One advantage of this approach is its concreteness and intuitiveness that may reinforce insights from the continuous probability distributions. The views of an investor and the set of possible outcomes are captured holistically in the continuous distributions, but discrete scenarios may be useful for purposes of communication. Exhibit 5 shows one way of presenting the information, but the number of scenarios and other features can be adjusted based on personal preferences. In the table, for Scenario #2, a 50% probability is assigned to the median values derived from the continuous distributions, and the remaining values are implied based on the probabilities given and the relationships described previously. Scenario #1 shows lower inflation but at the cost of higher unemployment and no real estate price appreciation. Conversely, Scenario #3 depicts higher inflation, continued low unemployment, and higher real estate price appreciation.
Conclusion
The contour of macroeconomic consensus forecast distributions can be combined with insights from historical data along with proprietary assumptions and integrated into a dynamic framework. This framework may be used to assess the impact of inflation on real estate more thoroughly and to consider how portfolio construction might need to adapt to the current environment. It is well established that real estate offers at least a partial inflation hedge. The five-step procedure recommended in this article starts with that well-established foundation and translates it into a broader set of forecast distributions. At a time when the US economy may be in a state of Keynesian “true inflation,” investors need tools to address an uncertain future with few domestic historical parallels. The methods outlined here may represent one component of a more comprehensive approach to appreciating how inflation impacts real estate.
Footnotes
1. John Maynard Keynes, 1936, The General Theory of Employment, Interest and Money, reprint, Cambridge University Press, 2014.
2. Turner Building Cost Index.
3. Eugene F. Fama and G. William Schwert, “Asset Returns and Inflation,” Journal of Financial Economics, Vol. 5, No. 2, 1977, pp. 115–146.
4. NCREIF Property Index (NPI) data is used instead of the housing component of the CPI. A regression is also run with NPI returns as the dependent variable and four quarterly lags of NPI returns, the three-month Treasury bill yield, and the CPI minus the three-month Treasury bill yield as independent variables (instead of using lags of the CPI minus the Treasury bill yield, as Fama and Schwert proposed in the original article).
5. The housing component of the CPI should be highly correlated with the headline CPI.
6. The initial consensus forecast distributions along with adjustments may be thought of as the “prior” in Bayesian analysis.
7. Roger B. Myerson’s SIMTOOLS Excel Add-In is used for the calculations, but other simulation software can produce similar results.
8. The example in this section is not Heitman’s official house view; it is intended solely as an illustration.
9. The time period selected is based on data available from Real Capital Analytics; inflation (monthly log percentage changes; source: St. Louis Fed); unemployment rate (first difference of unemployment rate; winsorized at the 5th and 95th percentiles; source: St. Louis Fed); commercial property prices (Real Capital Analytics CPPI National All-Property Commercial Property Price Index).
10. Bloomberg is the source of the macroeconomic forecast distributions. 11. A slightly more complex version of this first approach could include combining a heavy-tailed variation of consensus forecast distributions with the shape of the distribution based on historical data.
12. For example, nonlinear relationships, non-normal distributions, and time-varying correlations can render static correlation matrices less meaningful.
13. An autoregressive model with two lags is used for simplicity—AR(2).
14. 10,000 simulations are created for each variable at the end of the horizon (4Q22).
