Do US Residential Real Estate Returns Reinforce Wealth Inequality?

Summary:

• Recent studies have examined multiple mechanisms by which wealth inequality is created and sustained. A recent paper using high-quality data on the portfolios of Swedish households has documented higher portfolio returns for wealthy households. The researchers have not yet examined the potential offsetting impact of higher marginal tax rates, however.

• The documented higher returns come primarily from the capacity of wealthy households to take on equity and liquidity risk. However the paper also documents that wealthy households’ “net wealth” returns are partly dampened because they employ less leverage (i.e., mortgage debt) than middle class households.

• I don’t have access to the paper itself, and so can’t tell how they calculated returns on real estate assets. I performed an analysis on realized real estate returns in Lee County, Florida. It suggests pre-tax unleveraged returns (and risk) on real estate assets are higher for wealthy households.

• The Lee County realized return data is also consistent with Zillow's “estimated” house price time series at the US county level. However the Zillow data suggests that the returns on more expensive homes are less volatile than lower priced home returns.

Sources Of Higher Portfolio Returns For Wealthy Households

Bach, Calvet, and Sodini have written a very interesting paper titled “Risk, return, and skill in the portfolios of the wealthy”. I did not have access to the actual paper, but only a summary of it. They use a Swedish data set that is much more detailed and representative of wealthy households than the typical data used to study US households.

Figure 1 is reproduced from their summary. It shows portfolio returns and volatility based on “net wealth” (net assets = assets - debt). They explain the U-shaped pattern:

“Middle-class households have highly leveraged positions in real estate that generate high mean returns on net wealth. Upper-middle-class households have lower leverage and lower average returns. At the very top, households have very little personal debt but achieve high expected returns by bearing high systematic risk.”

Figure 1: Returns And Standard Deviations on Net Wealth From Bach, Calvet, and Sodini

“High systematic risk” in this case simply means that wealthier households hold more stocks in their portfolio (a fact that has been well known for some time). The authors also find that wealthier households benefit from more “idiosyncratic” risk due to their undiversified holdings in private and public equity, hedge funds and other partnerships (these often require high minimum investments or net worth qualifications and thus are unavailable to middle income households).

The authors don’t state it, but some of this “idiosyncratic” return is also due to the illiquidity premium investors demand for private equity/hedge fund/partnership investments because their ability to withdraw funds may be quite limited for some extended period of time. Interestingly they find no evidence that wealthier households make “smarter” investments (i.e., they have no information advantage), they can simply afford to take more risk.

The authors find confirming patterns in return data from US foundations. In many ways these are not surprising findings. Households (and foundations) with less “excess” savings cannot afford to take on as much risk as they are more likely to need to dip into their savings during economic downturns. Nevertheless the paper provides useful detail that documents the specific sources of return advantage that increase and sustain wealth inequality over time. It is important to point out the authors did not look at the offsetting impact of marginal tax rates or propensities to spend.

Realized Real Estate Returns In Lee County, Florida

Residential real estate is not as “homogenous” as publicly traded stocks. A S&P 500 Index fund will provide identical returns and risks to both middle class and wealthy households. It is less clear that their respective real estate holdings are as similar, however. It is quite possible that both the returns and risks of holding “luxury” real estate are different than those from owning a “median” home. So an interesting question is whether wealthy household real estate holdings have higher returns and risk than other households.

I looked at a data set of actual residential real estate transactions from Lee County, Florida. Lee County is one of the few counties I am aware of that makes this information freely available. There is a significant amount of data scrubbing required to limit the data to single family residences and to filter out “bad” data.[1] In addition, homeowners may choose to sell properties to family members at a below fair-market price (although the IRS may deem some portion of the sale to be a gift subject to gift-taxes).

I ended up with a set of approximately 525,000 transactions going back more than 25 years. For every year, sales prices were categorized into the lowest 30% of prices (“Bottom”), middle 40%, and top 30%. Table 1 shows the number of transactions (“N”) and median realized annual returns from each transaction based on sales price category and holding period. Also shown are the 90th percentile minus the 10th percentile realized return spread as an indicator of return variability.[2]

Table 1: Median Annualized Returns And Return Variability From Home Sales In Lee County, Florida

	Bottom 30%			Middle 40%			Top 30%
Holding Period	N	Ann. Return	p90-p10 Ann. Return Range	N	Ann. Return	p90-p10 Ann. Return Range	N	Ann. Return	p90-p10 Ann. Return Range
5	11,572	-3.7%	57.7%	17,479	2.2%	43.5%	13,807	1.3%	74.1%
10	5,520	-1.6%	26.5%	7,558	1.1%	24.7%	5,620	1.6%	37.2%
15	3,084	1.5%	20.1%	3,248	3.3%	18.8%	3,029	5.7%	15.7%
20	1,895	2.4%	13.2%	1,523	3.6%	14.7%	1,561	7.4%	17.9%
25	1,292	3.1%	12.0%	885	4.4%	12.8%	915	11.2%	12.8%
30	769	4.2%	11.3%	379	6.1%	11.4%	229	7.4%	9.8%

While the data is noisy, it suggests more expensive real estate has produced higher returns than lower priced real estate. Consistent with the general findings of Bach, Calvet, and Sodini, the variability of realized returns are also higher for more expensive homes except for the longest time horizon.

Figure 2 shows the annualized return by year for the subset of homes sold that had been owned for 5 years. The data is noisy, but the return advantage of more expensive homes is evident. There is also an indication that less expensive homes lost value for a longer period following the Great Recession of 2008 (perhaps due to the greater incidence of negative equity, foreclosures etc.).

Figure 2: Return By Calendar Year For Homes That Had Been Held For 5 Years

It would be easy to spin causal stories as to why commoditized low priced homes don’t appreciate as much as luxury homes that are often situated in unique settings (e.g., beachfront, waterfront, on Central Park etc.). But it is equally possible that the data shown here is only typical of areas like Florida that are seeing strong population growth. Anecdotally, luxury homes in states that are experiencing population declines are seeing much greater losses than modestly priced homes.

Zillow’s US Historical Price Estimate “Returns”

In an effort to broaden my analysis, I looked at Zillow’s historical time series of estimated prices for top third (by price) and bottom third homes for the entire United States. The series I used was at the US county level with a monthly frequency starting in April, 1996. These are not actual realized returns, but simply average value estimates by month.

Spot checking Lee County, the Zillow data indicates an annualized “return” of 7.7% for top third homes vs. 4.5% for bottom third homes for this roughly 21 year period. Those values are somewhat higher for bottom 30% homes, and somewhat lower for top 30% homes than the ones reported in Table 1 above based on realized transactions.

Figure 3 shows the relationship between return and the starting mean home price in April, 1996 by county for top tier homes.[3] The positive relationship between price is similar and also statistically significant for bottom tier homes by county.

Figure 3: Return vs. April, 1996 Mean Home Price For Zillow's Top Tier Homes

I combined the Zillow top tier and bottom tier data and divided all of the mean home prices into 3 categories of top 25%, middle 50% and bottom 25% (I arbitrarily chose home prices in the year 2000 as there were more missing values for 1996). In this exercise a county with uniformly high priced homes like Marin County, California has both its Zillow Top Tier and Bottom Tier homes in my top 25% category. Table 2 shows returns and return volatility by those categories.

Table 2: Zillow Mean Annualized “Return” And Return Variability By 2000 Home Price Category

Category	2000 Avg. Home Price	Annual "Return" 1996 to 2017	Annualized Standard Deviation
Top 25%	$193,142.09	3.79%	2.76%
Middle 50%	$102,814.79	3.13%	3.13%
Bottom 25%	$49,922.41	2.50%	3.76%

The pattern of higher returns for more expensive homes again is evident, but their lower volatility contradicts the volatility pattern in Table 1.

The national correlation between US home price and return is the dominant pattern. However even if we look cross sectionally within counties, Zillow top tier homes in each county have outperformed their corresponding bottom tier homes by an average of approximately 0.5% per year.

I suspect this pattern of higher residential real estate returns holds for other "developed" economies. I have less intuition about the structure of home appreciation in emerging economies.

[1] The steps I took to filter the raw data are in the R code linked below.

[2] I used median returns and return ranges to guard against spurious data points. However mean returns and standard deviations of returns show similar patterns.

[3] It is important to use starting home prices (rather than ending or mean price over the period) to avoid "look ahead" bias. For example, counties with the highest ending home prices very likely also experienced high home price appreciation.

Transparent and reproducible: The R code used to produce all figures and tables (except Figure 1) is available in “FLrealEstate.r" on github.