VoxEU Column COVID-19 Labour Markets

The doughnut effect of COVID-19 on cities

The pandemic has pushed against many of the central forces that create economic agglomeration in cities. This column presents evidence on how US real estate markets have responded to the pandemic and the rise of working from home. The authors find that real estate demand reallocates from high-density regions where many people work from home to low-density regions where fewer people work from home within metropolitan areas for both residential and commercial properties, but do not find much evidence of demand reallocating across metropolitan areas. These changes appear to be limited to highly populated ‘superstar’ cities.

Single family home prices in the US have continued their gains since the Great Recession, rising about 7% over the past year. And commercial office demand has remained relatively flat. But underneath these aggregate trends lies a substantial reallocation of housing and office demand away from dense city centres toward city outskirts and suburbs. This is the ‘doughnut effect’ – the rise of the suburbs and the slump of the city centre, driven by a fear of crowds and the growth of working from home (WFH) (Bloom 2020).

The doughnut effect: Reallocation in real estate demand

The news has been full of stories claiming the end of America’s biggest cities as we know it.1 From New York to Chicago to San Francisco to Washington DC, mobile phone data show2 substantial out-migration after the virus struck – with rich areas taking the biggest hits.3 How have these trends affected real estate markets? 

Our research has examined trends across real estate markets. Figure 1 shows Zillow's rental index for the 12 largest metro areas in the US by population.4 The central business district (CBD) and top 10% of zip codes by population density saw more than a 10% drop in rents confirming that demand for real estate in dense city centres has actually fallen. And other recent research finds that rents are declining in high-density neighbourhoods across the country (Liu and Su 2020).

Figure 1 Normalised Zillow rental index broken down by density group and CBD


Note: 1 February 2020 = 100.

But the outskirts of cities, with cheaper land and more space, have weathered the pandemic without substantial drops with little change in rents since the pandemic.

What is driving the fall in real estate demand? At least four factors are at play, including the economic shock from the virus, the lack of access to city amenities due to lockdowns and social distancing, the aversion to dense areas due to fear of virus spread, and the ability to work from home. The first three factors are likely temporary, which has led some commentators to claim5 that markets could rebound.6 

To test the longer-term effects of the pandemic shock we took a look at real estate transaction price data from Zillow.

Housing prices suggest the trend is longer-term

In theory, property prices should factor in both the short-term and long-term factors that affect the value of property. This is because owning a property is like owning any other asset. You can collect income streams by renting out the property to perpetuity. So just like the stock market is highly forward-looking and weighs future income streams, so do property valuations. And with record-low interest rates, this forward-looking nature is greater than ever.

What do the price data show? Though there is less of an aggregate decrease in prices as compared to rents, there is a similar demand reallocation effect where city CBDs and dense areas experience relative price decreases compared to less dense areas. Interestingly, the doughnut effect for prices appears to be limited to highly populated dense cities. We didn’t observe much of an effect for metro areas outside the largest cities.

Figure 2 Normalised Zillow home value index broken down by density group and CBD for 12 largest metros


Note: 31 January 2020 = 100.

As an example, Battery Park, one of central zip codes of downtown Manhattan, saw a 9.6% decrease in prices from the last three months of 2019 to the last three months of 2020. By contrast, middl- class residential neighbourhoods like suburban Suffolk County in Long Island have boomed. Home prices in Suffolk County are up in every zip code, with an average increase of 7.2% over the same period.

Similarly, the Bay Area of San Francisco has seen moderate decreases in prices. The Presidio, prized for being located directly adjacent to the city centre, has seen price decreases of almost 10%. On the other hand, more remotely located and cheaper Marin County, located right across the Bay, has seen an average increase of over 8%. 

Figure 3 shows heat maps of year over year price changes in both New York City and the Bay Area. In both maps, the city centres, Manhattan and downtown San Francisco, respectively, have taken large hits relative to surrounding areas.

Figure 3 Year-over-year percent change in home price index by zip code for SF Bay Area and New York City


Working from home may be driving price dispersions

Previous research by one of us (Barrero et al., 2020) has shown substantial increases in working from home as a result of the pandemic and this shift will likely be persistent. Indeed, managers expect close to 20% of working days to be done from home post-pandemic. Importantly, though, the types of jobs that can be done from home are not evenly distributed over the different parts of cities. And neither are the people who do them. 

We used data from a recent paper by Dingel and Neiman (2020) that classifies the share of jobs that can be done from home by occupation. Combining this with US Census data on the occupations of people who reside in each zip code, we were able to construct a zip-code level measure of the share of jobs that can be done from home which we term the WFH exposure of a zip code.7

We also found that most of the changes in property prices across different zip codes can be explained by the share of jobs that can be done from home, with density only contributing a small amount. We take this as evidence that people with the ability to work from home are reallocating away from city centres to lower-cost areas on city outskirts because they will not have to commute as frequently. As seen in Figure 4, even after we control for the metro region and the population density of a zip code, the share of jobs that can be done from home exhibits a striking negative relationship with the year over year change in home value index.

Figure 4 Price changes at the zip code level are negatively correlated with the share of jobs that can be done from home after controlling for population density and metro region


Note: Chart bins across zip codes for 12 largest metros into 20 points.

Flight from San Francisco to Austin? Longer-term reallocation is likely within metro areas

So far, we have presented evidence on the reallocation of real estate demand among zip codes within metro areas like the greater New York City area. But what about the stories of people moving away from expensive metro areas like New York or San Francisco to cheaper ones like Austin? To test this, we look at Zillow’s Metro area wide housing price index across different metro areas and plotted the change in the price index over the past year against the mean price level for 2019. If there were a long-term reallocation of demand from expensive metros to cheaper metros we would expect a negative correlation – but we actually observed a relatively flat relationship.

Figure 5 Price changes vs price levels across major US metro areas


We now have two pieces of preliminary evidence that we can put together. We’ve observed within-city reallocation of housing demand from dense areas with a high WFH share to less dense, low-WFH areas. We’ve also observed that housing demand hasn’t reallocated much from expensive cities to less expensive cities.

This suggests that the dominant form of working from home post-pandemic will be several days a week as opposed to full-time. With 1-2 days of working from home a week, employees who previously lived and worked in city centres, may be willing to move further away to the outskirts of cities or nearby suburbs. But since they still have to come to work sometimes, they are not willing to take a flight and move to a cheaper city with more. The survey evidence that has been previously documented is consistent with the real estate price movements we’ve seen here – employers project that the employees will likely work-from-home a couple days a week. 

Long-term commercial office demand may take a hit across cities

So far, we’ve been exploring data on residential properties. But with everyone working from home, one of the most affected markets has been commercial office properties. Interestingly, short-term demand in the aggregate for commercial office space has not changed much8 because the decreased quantity of people going to the office has been counterbalanced by the amount of space needed per employee to ensure social distancing. 

Commercial property transaction data is much sparser than residential property data so we constructed an index ourselves using transaction-level data from Zillow.9 Transactions in CBDs are especially limited, but a similar pattern of divergence between high density tracts and lower density tracts can be seen after the pandemic shock. Overall, we estimate the pandemic led to a 10% drop in commercial office building prices in the densest decile of zip codes relative to other zip codes.

Figure 6 Commercial office buildings have fallen in value in dense areas relative to less dense areas


What policymakers can start thinking about

Though people will still go to work in America’s biggest cities, there will certainly be shifts in urban structure. State and local governments must ease the transition. Though people will commute to work on less days per week, more people will flock to areas further from city centres, which increases the need for spread out public transportation networks. Furthermore, property and sales taxes in city centres will take a hit (Althoff et al. 2020), which may require city governments to make tough choices on services unless states step in to provide support.

Cities will also benefit from more balanced real estate prices across different regions. A best-case scenario is a more affordable city centre with more enjoyable work environments rather than pricy living spaces that are dominated by the rich. 


Althoff, L, F Eckert, S Ganapati and C Walsh (2020), "The City Paradox: Skilled Services and Remote Work", CESifo Working Paper No. 8734

Barrero, J M, N Bloom, and S J Davis (2020), "Why Working From Home Will Stick." University of Chicago, Becker Friedman Institute for Economics Working Paper 2020-174.

Bloom, N (2020), “How working from home works out”, SIEPR Policy Brief.

Dingel, J and B Neiman (2020), “How Many Jobs Can be Done at Home?”, Covid Economics 1: 16-24.

Gindelsky, M, J Moulton, and S A Wentland (2019), "Valuing housing services in the era of big data: A user cost approach leveraging Zillow microdata." Big Data for 21st Century Economic Statistics. University of Chicago Press.

Liu, S, and Y Su (2020), "The impact of the COVID-19 pandemic on the demand for density: Evidence from the US housing market", available at SSRN 3661052.


1 E.g.


3 The hardest hit area in New York was downtown Manhattan, a classic example of a neighbourhood vulnerable to COVID-19. Manhattan has a high-share of residents who can work from home, high population density, expensive rents and prices, and normally benefits from many in-person amenities to attract residents.

4 Our sample of the 12 largest metro areas in the US by population consists of New York, Los Angeles, San Francisco, Chicago, Washington DC, Atlanta, Philadelphia, Boston, Miami, Houston, Phoenix, and Dallas. And our zip code density buckets are high = top 10%, mid = 50-90th percentile, low = 0-50th percentile.



7 Dingel and Neiman (2020) calculate the share of occupations that can be done from home weighted by wage for each industry by determining whether an occupation’s tasks can be done from home or not. Our methodology assumes that this metric is relatively stable across US geographies.

9 We build a hedonic model of commercial property prices to construct a monthly price index across zip code density groups for our sample of MSAs. Our methodology is similar to that of Gindelsky et al. (2019) and aims to control for changes in the composition of properties sold. Data provided by Zillow through the Zillow Transaction and Assessment Dataset (ZTRAX). More information on accessing the data can be found at The results and opinions are those of the author(s) and do not reflect the position of Zillow Group.

6,824 Reads