The heterogeneous response of real estate prices during the Covid-19 pandemic

Abstract

We estimate the transmission of the pandemic shock in 2020 to the residential and commercial real estate market by causal machine learning, using granular data for Germany. We exploit differences in the incidence of Covid infections and short-time work at the municipal level for the identification of epidemiological and economic effects of the pandemic. We find that (i) a larger incidence of Covid infections temporarily reduced rents for retail real estate; (ii) a larger incidence of short-time work temporarily reduced rents of office real estate; (iii) the pandemic increased prices, particularly in the top price segment of commercial real estate.

affordability crisis, asset prices, Covid pandemic, real estate, rents, short-time work

E21, E22, G12, G51, R21, R31

1 Introduction

Empirical and theoretical research has stressed the importance of the real estate market for the economy (Piazzesi & Schneider, 2016): Housing services constitute an important consumption item with an expenditure share of about 20%, housing is the largest asset in household balance sheets in many developed countries, and real estate is a production input for many firms. Changes in the valuation of real estate affect the net worth of households and firms, as well as the value of collateral in credit contracts, and thus propagate to consumption and investment (Mian et al., 2013; Mian & Sufi, 2011; Schmalz et al., 2017). Although considerable progress has been made in comprehending how real estate markets amplify economic booms and busts (Jorda et al., 2016), several aspects of how macroeconomic shocks transmit to real estate values remain inadequately understood.

We contribute to the literature with new insights into the price responses to the macroeconomic shock of the Covid pandemic in the German real estate market.¹ Germany provides a good setting for our analysis due to its well-developed rental market for both commercial and residential real estate, with the highest proportion of households renting their primary residence among G7 countries (more than 50%). To capture potential heterogeneity in price responses, we use new granular data at the municipal level. We expect non-linear and local price responses at the municipal level because of housing market frictions such as indivisibilities or market segmentation, which restrict arbitrage opportunities (Piazzesi & Schneider, 2016). We thus employ flexible causal machine learning methods to identify these non-linear patterns effectively. These methods allow us to estimate heterogeneous price responses systematically without imposing rigid functional forms that constrain the relationship between responses and municipal characteristics.

To establish causal relationships, we exploit the variation in the epidemiological and economic incidence of the pandemic across municipalities to identify the effect of the pandemic on real estate prices at the municipal level. We uncover heterogeneous effects across different price segments in the real estate market and assess the implications for the premium of urban real estate and its affordability. Methodologically, our analysis adds to the emerging literature that applies recent causal machine learning techniques in the context of asset pricing.

The estimated price responses provide answers to important policy questions: How has the pandemic impacted real estate prices, temporarily or persistently? Has it exacerbated or mitigated the affordability crisis by affecting prices differently across residential market segments?² What are the implications for the collateral value of commercial real estate and the premium on urban real estate? Answers to these questions are not only relevant for asset pricing and assessing the macroeconomic consequences of the pandemic but also for public finances and land use planning in municipalities with varying urbanization and industry employment.

To answer the questions, we use detailed municipal-level data on real estate prices obtained from the German real estate company 21st Real Estate. These data are combined with information on Covid-19 infections and the incidence of short-time work for municipalities with different industry and employment structures. Whereas Covid-19 infections capture the epidemiological incidence of the pandemic, we use short-time work as a measure of the economic incidence of the pandemic. Short-time work has become a popular labour market policy across many developed countries to reduce the labour cost of employers during a recession without resorting to layoffs. It has been widely used by firms in Germany during the pandemic, and we will explain it further in Section 3.3. Our analysis focuses on the period from the last quarter of 2019 (2019Q4) to the first quarter of 2021 (2021Q1), as this captures a key time period during which the transmission of the pandemic to real estate prices occurred, as depicted in Figure 1, Section 3.2.

Figure 1.

The performance of German real estate investment funds during the pandemic. Notes: Residential real estate funds are in blue colour, commercial real estate funds in red colour. See Appendix B.3 of the online supplementary material for further details on the funds. Share prices of traded commercial and residential real estate funds are normalized to 100 on 2 March 2020. Source: YahooFinance, accessed on 24 October 2021.

We distinguish six different outcomes in our analysis: rents and sale prices for residential real estate or commercial real estate, for which we can distinguish retail and office real estate. We find that, on average, the pandemic had stronger effects on prices for commercial than residential real estate in Germany. A high rather than low epidemiological incidence in a municipality decreased rents for retail real estate by $3.5$%. A high rather than low incidence of short-time work in a municipality, capturing the economic incidence of the pandemic, decreased rents for office real estate by $3.2$%. Because we find that the corresponding effects on asset prices have been positive in these real estate segments, standard asset pricing models suggest that the negative demand shock in these segments has been expected to be only temporary. A possible explanation for this finding is that government support programs for households and firms have been generous so that demand for real estate did not fall persistently, and that the pandemic did not trigger persistent regional shifts of the population in Germany.

In terms of the heterogeneity of the price responses, our results indicate that the effects tend to be larger in absolute terms for municipalities that have been in the high-price segment in the last quarter of 2019, i.e. in more urban areas. Concerning the urban premium for commercial real estate associated with the top price segment, short-time work, and the incidence of Covid infections had countervailing effects, with a higher incidence of Covid temporarily reducing the corresponding premium for rents of retail real estate. Overall, we find that the pandemic increased the asset-price premium for real estate units that have been in the top segment of the price distribution prior to the pandemic.

Concerning the affordability crisis for residential real estate, the point estimates imply that a higher incidence of short-time work exacerbated the crisis somewhat by increasing rents in already high-priced municipalities. A higher incidence of Covid infections instead had a countervailing effect. Further analysis of the effect of the pandemic across the price distribution reveals that the pandemic did not have a clear-cut implication for the affordability crisis of residential real estate in Germany.

Real estate prices in Germany have thus shown a less pronounced and persistent response to the pandemic compared to countries like the U.S., where the pandemic reduced price pressure for relatively more expensive commercial and residential real estate in, or close to, city centres (Gupta, Mittal, Peeters, et al., 2022; Gupta, Mittal, Van Nieuwerburgh, 2022; Mondragon & Wieland, 2022; Ramani & Bloom, 2021; Rosenthal et al., 2022; Schwartz & Wachter, 2023). Hence, the implications of the pandemic for the asset prices of real estate, land-use planning, and public finances are quite different in Germany compared to the U.S., on which much of the recent research has focused. We show in Section 7 that the much stronger responses of real estate prices in the U.S. are consistent with the stronger changes in net migration across US regions during the pandemic compared to Germany.

Our analysis is structured as follows. In Section 2, we discuss our contribution in the context of the existing literature. Section 3 provides background on the economic framework for interpreting the results, the real estate market in Germany, the geographic variation contained in real estate rents and sale prices as well as our measures of the epidemiological and economic incidence of the pandemic, and the policy measures implemented in Germany during the pandemic. In Section 4, we describe the data used for our analysis, followed by an explanation of our estimation approach in Section 5. In Section 6, we present the main results, which focus on the heterogeneous price responses of real estate across German municipalities and their implications for the premium of real estate in the top price segment. In Section 7, we discuss the results and then conclude in Section 8. In the Appendix of the online supplementary material, we provide information on the estimation method; additional background and descriptive statistics, including cross-country evidence on the performance of real estate investment funds to put the changes of real estate values in Germany in an international perspective; further information on the data set and the construction of variables, as well as summary statistics; additional estimation results on the price responses and the heterogeneity analysis.

2 Related literature

The Covid crisis triggered a combination of demand and supply shocks across the economy (Baqaee & Farhi, 2022; Guerrieri et al., 2022) with potential consequences for resource allocation (Kozeniauskas et al., 2022). Our analysis contributes to the existing literature on the effects of the pandemic on prices for both residential and commercial real estate, exploiting geographical variation in the incidence of the pandemic across municipalities.

Concerning residential real estate, D’Lima et al. (2022), Gupta, Mittal, Peeters, et al. (2022), and Ramani and Bloom (2021) provide evidence for the U.S. that the premium for residential real estate in centres of US cities has fallen during the pandemic. These findings suggest that the pandemic triggered changes in preferences and behaviour which have been reflected in the valuation of real estate characteristics such as the proximity to the workplace or the number of rooms. Ramani and Bloom (2021) document a 10% decrease in rents in high-density areas of the 10 largest US cities, associated with a fall in prices of commercial real estate in these areas and a higher exposure to working from home.

The impact of remote work on real estate demand during the pandemic has been examined by Delventhal et al. (2022), Mondragon and Wieland (2022), and Van Nieuwerburgh (2023). Mondragon and Wieland (2022) estimate that over half of the increase in house prices in the U.S. since 2019 can be attributed to this factor.

Concerning the effects of the pandemic on commercial real estate, we contribute to the recent literature discussed in Baum-Snow et al. (2022) and Schwartz and Wachter (2023) and the references therein. Rosenthal et al. (2022) estimate a decrease in the premium for locations with better access to cities, employment-dense areas, or transit stations in the U.S. at the beginning of the Covid pandemic. Gupta, Mittal, Van Nieuwerburgh (2022) estimate a large, persistent decline of 40% for the values of office real estate in New York.

We build on the analysis conducted by Ling et al. (2020), who examined the impact of the Covid-19 pandemic on the stock-market valuation of commercial real estate through Real Estate Investment Trusts (REITs) in the US REITs represent only a specific segment of the real estate market, however, and it is unclear to what extent their findings apply to the larger private real estate markets with higher transaction costs and intermediated trade. In Appendix B.3 of the online supplementary material, we provide descriptive evidence on the performance of REITs and other real estate investment funds traded on the stock market, placing the price response in the German real estate market within an international context. Our contribution is to analyse residential and commercial real estate in the private markets of Germany, utilizing new granular data.

Related research has analysed the impact of pandemics on the real estate market from a historical perspective. Ambrus et al. (2020) investigate the effect of the cholera pandemic in London, Francke and Korevaar (2021) analyse the impact of the plague in Amsterdam and the cholera pandemic in Paris, Wong (2008) gauges the effect of the SARS pandemic in Hong Kong, and Lu et al. (2023) compare the effect of the SARS and Covid-19 pandemic. These analyses suggest that the long-run effects of pandemics on house prices and rents depend on policy responses, such as infrastructure investments or changes in housing structures. In the short run, demand effects can lead to significant changes in supply only under specific circumstances where existing real estate units can be quickly repurposed. For instance, in Lisbon, rental housing units shifted from short-term tourist rentals to long-term residential tenancy when the Covid-19 pandemic reduced tourism in Portugal (Batalha et al., 2022).

Our analysis further contributes to the existing research on the German residential real estate market by utilizing disaggregated data which allow us to exploit geographical variation. Kaas et al. (2021) explore the determinants of the lower owner occupation rate in Germany compared to the U.S., whereas Kindermann et al. (2022) investigate learning patterns regarding housing costs in the German real estate market. Kindermann et al. (2022) provide a valuable benchmark for our analysis, as they use proprietary data from Bulwiengesa that report sale prices and rents for residential real estate at the county level. In contrast, our analysis benefits from the more granular data provided by 21st Real Estate, which is based on list (ask) prices derived from primary and secondary data sources. Ardila et al. (2021) and Gupta, Mittal, Peeters, et al. (2022) have shown that list prices and transaction prices are highly positively correlated. The key advantage of our empirical analysis is the utilization of the more extensive and granular dataset from 21st Real Estate, enabling us to capture the heterogeneous characteristics of real estate markets at the municipal level. The larger sample size of $4,162$ observations facilitates our non-parametric analysis of the data.

Breidenbach et al. (2022) also exploit data from ImmoScout24 to examine the impact of airport noise and proximity on rents of residential real estate in Berlin. The dataset provided by 21st Real Estate complements the ImmoScout24 data by incorporating information from two other major real estate online platforms, Immonet and Immowelt. This comprehensive dataset enhances the coverage of the German real estate market, strengthening our analysis.

Methodologically, our analysis contributes to research that uses machine learning techniques to capture the non-linear pricing effects of various real estate attributes. We employ the Modified Causal Forest (Lechner & Mareckova, 2024), a recently developed estimator building on the machine learning literature to estimate causal effects, as surveyed by Hastie et al. (2009). This approach enables us to estimate effects for multiple treatments at different levels of aggregation (individual, group, population average) using a unified estimation and inference framework. Further details on the application of this method for our analysis can be found in Appendix A of the online supplementary material.

3 Background

In this section, we provide background on the standard economic framework that underpins our interpretation of the results. We proceed by describing features of the real estate market in Germany and illustrate the geographic variation contained in real estate rents and sale prices. We then present our measures for the epidemiological and economic incidence of the pandemic, and discuss the economic policies associated with the Covid pandemic.

3.1 Economic mechanisms

In the following, we outline the implications of standard demand-supply and asset-pricing models that we employ to interpret our findings. A significant macroeconomic shock like the Covid-19 pandemic can lead to changes in preferences and behaviour, resulting in shifts in demand for real estate units with varying characteristics. These shifts are reflected in prices, given the relatively slow adjustment of the real estate supply. By estimating the responses of rents and sale prices to the pandemic shock, we evaluate whether the observed demand shifts are temporary or expected to persist over time.

Given the focus of our analysis on the first year of the pandemic, the assumption of a relatively constant supply of real estate appears reasonable. The construction of new real estate in response to the Covid-19 shock is negligible within this period. For instance, new residential real estate buildings constituted less than 1% of the existing housing stock in both 2019 and 2020, based on data from the Federal Statistical Office.³ Additionally, the conversion of office spaces to residential units is a costly and time-consuming process. Overall, the supply side of the real estate market in Germany remained largely unaffected during our sample period. The construction sector continued its operations, and both the number of completed real estate units and issued construction permits remained stable between 2019 and 2020. While there may be regional variation in the number of completed units, these differences remained stable over the two years.

The Covid-19 pandemic changed the demand for real estate across various segments and locations through different channels (Van Nieuwerburgh, 2023). During the lockdowns, households spent more time at home and may have anticipated an increased need for home office space. Consequently, firms may have adjusted their demand for office space, although the overall effect on demand remains ambiguous due to government regulations, such as social distancing requirements and limitations on the number of employees per square meter. The demand for retail real estate may have also been influenced by changes in commuting patterns, leading to shifts in shopping locations, as well as variations in the incidence of online purchases relative to in-store purchases. Although our data do not allow for a distinction between these different sources of demand shifts, it enables us to estimate the combined effect of these shifts on real estate prices.

According to a standard asset-pricing model, the response of rents and asset prices of real estate to the demand shifts triggered by the pandemic may exhibit different qualitative patterns. Since the fundamental value of real estate assets is equal to the present value of future rents, a temporary decrease in rents could be accompanied by an increase in the real estate asset price if future rents are expected to rise sufficiently to compensate for the current temporary decline. Therefore, estimating the effects on both rents and asset prices provides insights into the expected persistence of the demand shifts.

Of course, changes in the expected returns of real estate, used to discount future rent payments, would also influence real estate prices. Such changes may be caused by changes in the risk-free rate or the risk premium of real estate (e.g. Amaral et al., 2021). Inspecting the yield curve provided by the ECB indicates, however, that yields for assets with maturities up to thirty years were already at or close to the (effective) lower bound prior to the pandemic. This implies minimal changes in the yield curve during our sample period. Assuming that the incidence of the pandemic increased risk premia for real estate, the resulting higher expected returns used for discounting future rents should have led to a decrease rather than an increase in asset prices. The observed increase in asset prices, therefore, suggests that future rents were expected to be higher.

3.2 Residential and commercial real estate in Germany

Germany has a relatively low owner-occupation rate of $46.5\mathrm{\%}$⁠, meaning that over half of the households in the country rent their main residence. However, there is significant regional variation in ownership rates, with rates as low as 20% in cities like Berlin and as high as 65% in the Saarland region.⁴ In the period of low interest rates prior to the pandemic (2010–2019), rents for residential real estate increased by an average of 14%, and house prices doubled in many cities.⁵ This growth in prices followed decades of relatively modest price increases in Germany between the 1970s and 2010 (Kaas et al., 2021; Kindermann et al., 2022).

Changes in rents for residential real estate in Germany are constrained by rent control (the Mietpreisbremse specified in the German civil code, 2023). If a regional government denotes the housing market in an area as tight, according to criteria specified by the law, the rent of new rental contracts may not exceed the rent commonly paid for comparable housing units by 10%. This leaves room for the adjustment of rents for listed housing units on the market. We thus expect the effect of rent control on rents to be less relevant for the housing units in the data compiled by 21st Real Estate, which focuses on housing units on the market. In particular, the data do not include rents of housing units with continued tenancy or of units with rents that do not reflect market prices such as social housing.

To assess the potential impact of the Covid-19 shock on real estate valuations in Germany, we examine the stock-market valuation of real estate funds. We use daily data from Yahoo Finance on real estate investment funds for Germany. For further details on the funds, please refer to Table 4 in Appendix B.3 of the online supplementary material. The table includes information on similar real estate funds in France, Italy, Spain, Switzerland, the U.K., and the U.S., allowing us to put the performance of German funds in an international context.

Figure 1 illustrates the heterogeneous performance of German real estate funds, particularly since the onset of the first wave of Covid-19 infections in March 2020. In contrast to the other countries (with the exception of Switzerland), residential real estate funds in Germany recovered from their initial valuation losses in March 2020 and surpassed their initial valuations by September 2020. The disparity in performance between residential and commercial real estate funds is especially pronounced in Germany and Switzerland compared to other countries, as indicated in Table 5 in Appendix B.3 of the online supplementary material. Additionally, the table reveals that certain commercial real estate funds specializing in storage and distribution facilities experienced value increases during 2020, such as Tritax Big Box in the U.K. Figure 1 additionally provides evidence that the transmission of the pandemic to real estate prices in Germany took place until the end of 2020, as indicated by the relatively minor changes observed thereafter. This finding supports our decision to focus on the time period between 2019Q4 and 2021Q1 for estimating the price effects of the pandemic.

Compared to the relatively small changes observed in aggregate price indexes for residential and commercial real estate in Germany, the valuation changes in real estate funds have been more substantial. The aggregate price index for residential real estate in Germany increased by $1.3$% between the first and second quarter of 2020.⁶ The aggregate price index for commercial real estate decreased by $0.3$% and thus only moderately between the first and second quarter of 2020.⁷ These relatively minor changes in the aggregate price indexes likely reflect the limited diversification of the traded real estate funds that specialize in specific types of real estate.

The descriptive results obtained from the publicly traded real estate funds highlight the significant heterogeneity in the performance of real estate during the pandemic in 2020. This underscores the importance of differentiating between residential and commercial real estate in our analysis. By utilizing more granular data, we are also able to conduct a more comprehensive analysis of price changes in the real estate market during the pandemic.

Based on the granular data, Figures 2 and 3 illustrate that rents and sale prices for residential and commercial real estate in 2019Q4, i.e. prior to the pandemic, have been heterogeneous across German municipalities but highly correlated across real estate types within municipalities. This is intuitive because the price of land is a key determinant for both residential and commercial real estate.

Figure 2.

Rents by real estate type, across geographic areas prior to the pandemic in 2019Q4. (a) Residential. (b) Retail and (c) Office. Notes: Rents per month and in units of €/${\mathrm{m}}^2$⁠. The cities indicated on the map are abbreviated as follows. BER = Berlin; COL = Cologne; FRA = Frankfurt; HAM = Hamburg; MUN = Munich. Source: 21st Real Estate.

Figure 3.

Sale prices by real estate type, across geographic areas prior to the pandemic in 2019Q4. (a) Residential. (b) Retail and (c) Office. Notes: Sale price in units of €/${\mathrm{m}}^2$⁠. The cities indicated on the map are abbreviated as follows. BER = Berlin; COL = Cologne; FRA = Frankfurt; HAM = Hamburg; MUN = Munich. Source: 21st Real Estate.

Figures 4 and 5 illustrate that the changes of rents and sale prices during the pandemic between 2019Q4 and 2021Q1 instead are quite different for residential and commercial real estate. Rents for residential real estate increased between 2% and 9% across municipalities. For office and retail real estate, rents changed between $-10$% and $+25$%, thus exhibiting much more heterogeneity, including large decreases in rents in some municipalities.⁸ In our analysis, we combine the different price responses across municipalities and residential and commercial real estate types with different intensities of the pandemic shock at the municipal level, as measured by the incidence of Covid infections or short-time work. We describe these measures further in the next subsection.

Figure 4.

Changes in rents by real estate type, across geographic areas between 2019Q4 and 2021Q1. (a) Residential. (b) Retail and (c) Office. Notes: The cities indicated on the map are abbreviated as follows. BER = Berlin; COL = Cologne; FRA = Frankfurt; HAM = Hamburg; MUN = Munich. Source: 21st Real Estate.

Figure 5.

Changes in prices by real estate type, across geographic areas between 2019Q4 and 2021Q1. (a) Residential. (b) Retail and (c) Office. Notes: The cities indicated on the map are abbreviated as follows. BER = Berlin; COL = Cologne; FRA = Frankfurt; HAM = Hamburg; MUN = Munich. Source: 21st Real Estate.

We examine the relationship between the rent level in 2019Q4 and the subsequent change until 2021Q1 by combining the evidence from Figures 2 and 4. The correlation coefficients between the rent level and the subsequent change are $-0.01$⁠, $-0.24$⁠, and $-0.22$ for residential, retail, and office real estate, respectively. These findings indicate that the distribution of rents for commercial real estate has become more equal after the pandemic, whereas this is not observed for residential real estate. In Section 6.2.1, our analysis explores the extent to which these changes (or lack thereof) are associated with the causal effect of the pandemic.

3.3 The geographic incidence of the pandemic

We employ the incidence of Covid-19 cases and the cumulated incidence of short-time work as measures for the epidemiological and economic impact of the pandemic, respectively. The data for these measures are cumulated from January 2020 when the first infections were recorded in Germany.

Short-time work is a labour market policy. The purpose of short-time work is to mitigate temporary labour demand shocks by allowing firms to temporarily furlough employees, with the government employment agency providing compensation for a portion of their salaries and social security contributions. The replacement rate for short-time work is set at a minimum of 60%. To construct the cumulated incidence of short-time work for each municipality, we combine data on short-time work in 20 industries at the regional level (Bundesland) with the employment shares in these industries in each municipality before the pandemic. Further details on the construction of these pandemic incidence measures can be found in Appendix C.2 of the online supplementary material.

Figure 6 displays the cumulative share of reported Covid-19 cases and short-time work in Germany as of December 2020, during the peak of the second wave. Figure 6a reveals that the majority of reported Covid-19 cases until December 2020 were concentrated in the southern and southeastern regions of Germany, as well as in densely populated urban areas such as Berlin and the Ruhr area. Figure 6b demonstrates that the incidence of short-time work was highest in the southern regions of Germany, particularly in the southwest, which can be attributed to variation in the industry structure across different regions. It is important to note that the available information on Covid incidence is provided at the county level (Landkreise and kreisfreie Städte), resulting in fewer data points on the map. In our analysis, we assign the same value to all municipalities within a county.

Figure 6.

Share of reported Covid cases and short-time work across geographic areas cumulated between January and December 2020. (a) Covid incidence and (b) Short-time work. Notes: The Covid incidence is available at the county level. Our measure for the incidence of short-time work varies at the municipal level. The cities indicated on the map are abbreviated as follows. BER = Berlin; COL = Cologne; FRA = Frankfurt; HAM = Hamburg; MUN = Munich. Sources: Robert Koch Institute, Federal Employment Agency, 21st Real Estate.

In an international context, Germany experienced a similar economic incidence of the pandemic compared to the U.S. in 2020. Germany had a slightly larger contraction of real GDP in 2020Q2 and a somewhat slower recovery compared to the U.S. (The FRED Blog, 2022). The incidence of Covid infections, according to official statistics, was higher in the U.S. than in Germany during 2020. Both countries experienced significant excess mortality towards the end of 2020, however, as illustrated by EuroMOMO (2023).

3.4 Policy measures during the Covid-19 pandemic

During the pandemic, economic policy in Germany supported businesses and households to cope with the economic consequences, by providing additional credit, subsidies, recapitalizations, credit guarantees, and implementing a moratorium on insolvency rules, as documented by Federal Ministry of Economic Affairs and Energy (2023). The majority of these measures were financed by the federal government, with some contributions from regional governments and social insurance schemes. Changes in the public budget between 2019 and 2020 show that the federal government budget accounted for over 62% of the overall change, followed by the regional government budget (21%), social insurance schemes in part due to increased funding for short-time work (15%), and municipalities (2%), as reported by Federal Statistical Office (2021).

Since the economic policy measures helped businesses, which typically operate and have business relationships in more than just one municipality, we cannot construct a separate variable to disentangle the incidence of the policy measures at the municipal level. The effects associated with the economic incidence of the pandemic, as measured by the incidence of short-time work, are captured in our analysis, however. We also account for possible heterogeneous outcomes caused by the policy measures, e.g. by allowing the price effects of the pandemic to vary across municipalities with different industry structures.

The German government also imposed regulatory measures to contain the pandemic. The different timing of these measures across municipalities implied only very short-lived differences in regulation. To the extent that differences in the epidemiological and economic incidence of the pandemic, i.e. the incidence of Covid infections or short-time work, triggered a faster implementation of regulatory measures in some municipalities, the resulting price effects are included in the price responses we estimate if they persist until 2021Q1. Moreover, the timing differences in regulatory measures are more relevant for the price dynamics between 2019Q4 and 2021Q1 rather than for the prices in 2019Q4 and 2021Q1 themselves.

For the pricing in 2021Q1, the differences in the timing of lockdowns across regions during the first and second waves in 2020 are a matter of the past. From April 2021 onwards, the infection protection act (Bundesregierung, 2021) established a common regulatory framework relevant to the fundamental component of real estate prices in 2021 which depends on the present value of expected future rental income or the corresponding utility derived from housing services if the unit is owner-occupied. For the pricing in 2019Q4, the regulatory measures during 2020 have been unexpected so that they have not been priced in the real estate values prior to the pandemic. A qualifier is that the price data provided by 21st Real Estate are one-sided moving averages based on observations in the past year such that price dynamics before 2021Q1 may affect the price in 2021Q1 through this channel.

4 Data

The granular data provided by 21st Real Estate are key for our empirical analysis both from an economic and methodological point of view. The granular data allow us to uncover the heterogeneous responses of real estate prices to the global pandemic shock, exploiting variation at the municipal level. Furthermore, the granularity of the data implies a sample size that allows us to apply causal machine learning so that the heterogeneous responses can be estimated without relying on restrictive functional form assumptions.

The data provided by 21st Real Estate for 2019Q1 to 2021Q1 are based on list (ask) real estate prices, purchased from ImmoScout24 and complemented with web-scraped data. Boelmann and Schaffner (2018) describe the publicly available real-estate data set (RWI-GEO-RED) only based on listings on ImmoScout24. Evidence for other countries, with comprehensive data on both list and transaction prices, reveals that list prices and transaction prices are highly positively correlated (e.g. Ardila et al., 2021; Gupta, Mittal, Peeters, et al., 2022).

The data provided by 21st Real Estate consist of rents and asset prices for real estate units on the market, specifically excluding units subject to tighter rent control measures such as social housing or units with continued tenancy. The data are aggregated at the municipal level for our analysis. The prices are generated by aggregating offers from small geographic tiles, weighted by the inverse of the distance to the tile centre. The time series of prices are smoothed using one-sided moving averages, with greater weight given to prices based on more observations. The data cover residential, retail, and office real estate, and the quarterly data are weighted averages based on the last four quarters. 21st Real Estate focuses on market prices and excludes publicly subsidized real estate. The data undergo rigorous outlier detection procedures to ensure their representativeness and quality. For further information on the dataset, please refer to Appendix C.1 of the online supplementary material.

Our analysis is based on municipalities as unit of observation. In some cases, we also consider associations of municipalities (Gemeindeverbände) if they exist. These associations allow small municipalities to benefit from joint administration while maintaining their autonomy. For simplicity, we refer to both municipalities and associations as municipalities. To ensure an adequate size of the real estate market for each municipality in our sample, we exclude municipalities with less than 2,500 inhabitants. This results in a dataset comprising quarterly data for 4,162 municipalities from 2019Q1 to 2021Q1. Appendix C of the online supplementary material provides additional information, including details and descriptive statistics for the variables capturing the pandemic incidence, as well as the covariates and their respective data sources.

5 Estimation

We estimate the response of rents and sale prices in the real estate market to the Covid pandemic, focusing on the changes between 2019Q4 and 2021Q1. We identify the causal effect of the pandemic using variation in the pandemic incidence across municipalities. This approach allows for more credible identification than straightforward comparisons of outcomes over time. It comes at the cost that effects of the pandemic that are common across municipalities are not captured by our estimates.

We use two alternative measures for the epidemiological and economic incidence of the pandemic as our treatments, which both vary at the municipal level: the cumulated incidence of Covid infections and the cumulated incidence of short-time work between January and December 2020. The limited number of municipalities in the data does not allow us to analyse the interactions between the epidemiological and economic incidence. The sample sizes at the given granularity level of the analysis lead to insufficient common support in an orthogonal analysis.

We have illustrated the variation of both measures in Section 3.3. We discretize both incidences into three categories (low, medium, high), corresponding to the terciles of the respective distribution. We use these categories to estimate the effect of the pandemic on prices in a municipality with a high incidence, for example, relative to a municipality with a low incidence. We take logarithms of rents and sale prices so that the estimations deliver estimates for the changes in rents and sale prices in relative terms. In the benchmark specification, we weight the observation for each municipality by its population. In Section 6.1.2, we also present and discuss the results of the unweighted estimation.

As a parsimonious benchmark specification, we follow the efficient market hypothesis (EMH) and use only the prices in 2019Q4 to control for possible confounders (Fama, 1970). According to the EMH, the prices for residential, retail, or office real estate at the municipal level in 2019Q4 capture all information and, thus, all the heterogeneity relevant to the pricing that is available at this point in time. Hence, any changes to the prices after 2019Q4 are true innovations. In particular, the changes in the prices after 2019Q4 are unpredictable with information available in 2019Q4 and may be explained by the subsequent pandemic as measured by the incidence of the Covid cases or short-time work. Whereas the pure EMH would only require to condition on the respective sale prices in 2019Q4 of the real estate type associated with the outcome variable, we apply a weaker version of the EMH. We condition on the sale prices and rents of residential, retail, and office real estate in the specification for each real estate type. It implies that we estimate the same specification for sale prices and rents of residential, retail, and office real estate.⁹

Given that empirical evidence suggests some frictions in real estate markets (Case & Shiller, 1989; Piazzesi & Schneider, 2016), we also estimate a richer, alternative specification with further control variables in Section 6.1.2 that is more robust to violations of the EMH. These controls account for observable heterogeneity across municipalities on the demand and supply side prior to the pandemic, i.e. in 2019Q4. In Appendix C.3 of the online supplementary material, we describe the set of controls in more detail.

For both specifications, the time at which variables are measured is important to assess possible endogeneity concerns. The prices prior to the treatment and all control variables are measured in 2019Q4 and thus are exogenous. Threats to identification may arise through changes in the observed possible confounders during the treatment if these changes have an effect on the treatment incidence and the real estate prices. The set of control variables we include in our analysis contains characteristics of real estate units, the socio-demographic characteristics of residents of the municipality, and other local indicators. Those possible confounders change only very slowly over time, suggesting that endogeneity may not be a major threat to our identification.

We estimate the price responses to the pandemic non-parametrically applying a Modified Causal Forest (mcf) (Lechner & Mareckova, 2024). The mcf is a specific and improved version of a Causal Forest estimator (Wager & Athey, 2018), which allows for flexible estimators of causal effects at all aggregation levels. Lechner and Mareckova (2024) have proven statistical properties (consistency and asymptotic normality) of the mcf and have compared it to two popular causal machine learning estimators: Double/debiased Machine Learning (dml) developed by Chernozhukov et al. (2018) and Generalized Random Forest (grf) provided by Athey et al. (2019). The comparisons reveal that the mcf is very competitive and more robust in demanding estimation scenarios than dml or grf. The comparisons also include OLS, which turns out to be biased substantially in many situations and, overall, is not competitive at all.

Appendix A of the online supplementary material provides a short description of the mcf. The method allows the construction of estimators for price responses at different levels of aggregation. The average treatment effect (ATE)

$$\text{ATE}(m,l)=\mathbb{E}[Y^m-Y^l]$$

(1)

denotes the average effect between two treatments m and l at the population level. The group average treatment effect (GATE)

$$\text{GATE}(m,l,z)=\mathbb{E}[Y^m-Y^l|Z=z]$$

(2)

denotes the average treatment effect when conditioning only on a subset of covariates z (heterogeneity variables). The individualized average treatment effect (IATE)

$$\text{IATE}(m,l,x)=\mathbb{E}[Y^m-Y^l|X=x]$$

(3)

is the lowest level of aggregation and denotes the treatment effect when conditioning on all covariates x. The GATE and IATE are well suited to uncover heterogeneity in treatment effects across municipalities which differ, for example, in terms of the price segment in the real estate market, the degree of urbanization, the incidence of jobs with home office capability, or average household income. The full set of heterogeneity variables and covariates is listed in Appendix C.3 of the online supplementary material. The set of covariates contains the size of the municipality in square kilometres, the number of households and the average household size so that we account for population density in a flexible way.

6 Price responses to the pandemic shock

We first present the results for the average treatment effect (ATE) of the incidence of Covid infections or short-time work on rents and sale prices. We weight observations with population density in the benchmark specifications to account for changes in rents or prices in densely populated municipalities affecting more persons than in sparsely populated municipalities. We then report the results of robustness checks for specifications without weights or with a richer set of covariates as controls. We also investigate the heterogeneity of the effects by inspecting how the IATEs are associated with covariates of interest.

The estimates of the price effects, which we document below, can be interpreted as the result of demand shifts triggered by the pandemic through the channels discussed in Section 3.1. Estimating the effects on rents and real estate asset prices allows us to obtain insights about the expected persistence of the demand shifts, applying the logic of standard asset pricing. We further discuss the results in Section 7.

6.1 Average effects

6.1.1 Benchmark specification

Figure 7 illustrates the price responses in log point changes after changing from low to high epidemiological or economic incidence of the pandemic. Results for the price responses after changing from low to medium incidence or medium to high incidence are available upon request.

Figure 7.

Average effect on rents and sale prices. Notes: Effect of moving from low to high incidence in log point changes; 90% confidence intervals; the respective p-value is reported below each plotted point estimate; Covid denotes the effect of a higher Covid incidence, and STW denotes the effect of a higher incidence of short-time work.

Figure 7 shows that the average effects on rents and sale prices are smaller (in absolute terms) for residential than for commercial real estate, where the responses for commercial real estate are less precisely estimated. As we explain further below, the results suggest stronger temporary demand shifts for commercial than residential real estate, possibly caused by changes in commuting patterns and shifts in consumer demand discussed in Section 3.1.

Concerning the average effect of a higher Covid incidence, we find that a high rather than low Covid incidence reduced rents for retail real estate by $3.5$%. Although not statistically significant at conventional levels in the benchmark specification, the negative effect on rents of retail real estate turns out to be robust and is more precisely estimated in alternative specifications reported below.

We further find that a high rather than low Covid incidence left residential real estate rents almost unchanged, where the point estimate implies a small reduction of $-0.5$% that is not statistically significant at conventional levels. There is also no clear-cut effect of a higher Covid incidence on office rents. Although the point estimate is positive at $2.7$% in the benchmark specification, our robustness checks below reveal that the sign of the coefficient is not robust in the unweighted estimation, and smaller at $2.4$% and no longer statistically significant at a 90% level if we add further controls to the specification.

Concerning the average effect of short-time work, Figure 7 shows that a high rather than low incidence of short-time work reduced office rents by $3.2$%. The effect on residential and retail real estate has been smaller with point estimates of 1% for residential and 2% for retail real estate, where the estimate for retail is more noisily estimated.

Figure 7 further shows that the effect on sale prices of both a higher epidemiological or economic incidence of the pandemic is either close to zero or positive. This suggests that market participants expected the negative effect on rents (for retail real estate with higher Covid incidence and for office real estate with higher short-time work incidence) to be only temporary because the fundamental value of the sale price depends on current and expected future rents.¹⁰ The positive effect on sale prices in most estimated specifications suggests that there has been no persistent negative demand effect for real estate on average after the Covid shock in Germany.

In terms of the size of the effects, the 1% increase in rents for residential real estate with a higher short-time work incidence, for example, corresponds to roughly a quarter of the annual increase of rents of residential real estate at the aggregate level. Of course, the effect we estimate is identified by variation across municipalities, implying, for the example above, that municipalities with an incidence of short-time work in the top rather than bottom tercile of the distribution of short-time work incidence experienced a slightly stronger growth of rents and no significant change of sale prices for residential real estate.

The finding of sizable, temporary negative effects on rents for commercial rather than residential real estate in some specifications (a higher Covid incidence for retail and a higher short-time work incidence for office real estate) is qualitatively similar to the suggestive evidence on publicly traded real estate, which we described in Section 3.2. Quantitatively, the temporary changes observed for the publicly traded funds are larger than for the prices in private real estate markets. Those differences may be expected for various reasons: The traded funds are typically not well diversified by construction because they focus on specific locations or industries within municipalities. Furthermore, the time series changes for traded funds, illustrated in Figure 1, also contain effects that are common across municipalities, caused by the pandemic or any other event in the considered period.

6.1.2 Robustness

We report robustness results for specifications in which we either do not weight by municipality population or use a rich set of further control variables. For brevity, we focus on the effect on rents for which the benchmark specification revealed more variation across real estate types concerning the impact of the pandemic’s epidemiological and economic incidence.

Weighting

Figure 8 illustrates the robustness of the results when we do not weight municipalities by population in the estimation, with an exception concerning the effect of Covid incidence on office rents which changes sign and becomes negative. Comparing the average effects on rents, the figure indicates that the weighted estimates tend to be slightly larger in absolute terms in most specifications. This suggests a correlation between individualized effects and the population weight, which we will further explore in the subsection on the heterogeneity of the effects.

Figure 8.

Average effect on rents with and without population weighting. Notes: Effect of moving from low to high incidence in log point changes; 90% confidence intervals; the respective p-value is reported below each plotted point estimate; Covid denotes the effect of a higher Covid incidence, and STW denotes the effect of a higher incidence of short-time work.

Further controls

Figure 9 compares the estimates for the effect on rents in the parsimonious benchmark specification (EMH) with the estimates in the specification with the full set of further covariates (FULL) to account for possible confounders, as described in Appendix C.3 of the online supplementary material. The results in Figure 9 indicate that controlling for the additional covariates does not significantly alter the estimates compared to the parsimonious benchmark specification. The estimates in both specifications are not statistically different from each other. However, it is important to note that in the specification with the full set of controls, the point estimates increase substantially in absolute terms and become statistically significant at conventional levels for certain effects, such as (i) the positive effect of higher short-time work incidence on rents for residential and retail real estate and (ii) the negative effect of higher Covid incidence on rents for retail real estate. The effect of a higher incidence of short-time work on office rents becomes smaller and less precise in the specification with additional controls.

Figure 9.

Average effect on rents with further covariates. Notes: Effect of moving from low to high incidence in log point changes; 90% confidence intervals; the respective p-value is reported below each plotted point estimate; EMH denotes the benchmark specification, FULL denotes the specification with the full set of covariates; Covid denotes the effect of a higher Covid incidence, and STW denotes the effect of a higher incidence of short-time work.

6.2 Heterogeneity of the effects

The modified causal forest estimation allows us to examine the variation of the effects across municipalities with different characteristics without relying on parametric assumptions. Figure 10 displays the sorted individualized average treatment effects (IATEs) for a representative combination of outcome and treatment: the effect of short-time work on rents of residential real estate in the benchmark specification. The figure shows that all IATEs are quite close to the average effect. This evidence suggests that effect heterogeneity is unlikely to be identified along the dimension of a single control variable at a common statistical significance level unless this variable exhibits a near-perfect correlation with the IATEs. Inspecting the GATEs defined in (2) for each of the heterogeneity variables listed in Tables 8 or 9 of the online supplementary material, we do not find statistically significant evidence for heterogeneous effects with respect to a single heterogeneity variable.¹¹ The lack of significance may be the result of the limited sample size. We thus adopt an alternative approach to provide additional insight into the relationship between municipality characteristics and variation in the IATEs.

Figure 10.

Sorted individualized effect of short-time work on residential rents. Notes: Effect of moving from low to high incidence in log point changes.

We build on the findings from the robustness analysis, specifically the observation that weights by population in municipalities appear to be correlated with the individualized effects. Additionally, the descriptive analysis in Section 3.2 suggests that price levels in 2019Q4 are associated with subsequent price changes, particularly for certain real estate types. Furthermore, Figure 12 in Appendix B.1 of the online supplementary material reveals that characteristics associated with municipalities, such as the share of employment in jobs with home office capability or access to broadband internet, which help mitigate the economic impact of the pandemic, are positively correlated with more urban areas. Therefore, we jointly investigate heterogeneity across multiple variables that characterize affluent urban municipalities. These municipalities are typically characterized by higher per capita incomes, higher prices in the real estate market prior to the pandemic in 2019Q4, a higher proportion of jobs with home office capability and broadband internet access, classification as highly urbanized, and more densely populated residential areas. For simplicity, we refer to these municipalities as ‘more urban’ throughout the rest of the article.

Figure 11 illustrates that the effects are stronger in municipalities with such urban features. For the effect of short-time work on rents of residential real estate in Figure 11a, which is positive on average, the high IATE cluster is associated with urban municipalities.¹² For the effect of Covid incidence on rents of retail real estate in Figure 11b, which is negative on average, the low IATE cluster is associated with urban features. Quantitatively, the difference between the IATEs in the high and low cluster is sizable. The IATE of short-time work on rents of residential real estate is $2.8$% in the high cluster and $-0.2$% in the low cluster; and the IATE of Covid incidence on rents of retail real estate is $1.5$% in the high cluster and $-11.7$% in the low cluster. Since all urban characteristics discussed earlier exhibit a similar correlation pattern with the IATEs, these findings collectively provide evidence of heterogeneous effects across municipalities that differ across these features.

Figure 11.

Characteristics of covariates by IATE cluster for residential or retail rents and different measures of the pandemic incidence. (a) Residential rents with short-time work incidence and (b) Retail rents with Covid incidence. Notes: Average value of covariate in respective IATE cluster in relation to empirical covariate distribution; Abbreviations for labels - Q4 2019 rent residentiallog = pre-treatment log rent prices for residential real estate; broadband coverage = share of households with broadband access of at least 200 MBit/s; log income pp = log yearly income per person; urbanization = degree of urbanization; HO occ index = share of jobs offering occasional home-office capability; avg apt per res fac = average number of apartments in residential facility.

Table 1 summarizes the sign of the correlation between urban features and the IATEs for all considered real estate types and types of pandemic incidence where the respective graphical illustrations are delegated to Appendix D of the online supplementary material. Table 1 shows that more urban municipalities have experienced a larger effect on sales prices across all real estate types both for a higher Covid or short-time work incidence. For rents, the correlations imply stronger effects in absolute terms in more urban municipalities. For residential and retail real estate, the Covid incidence has had a more negative effect in urban municipalities whereas the short-time work incidence has had a more positive effect. For office real estate, the opposite correlation pattern implies a more negative effect of short-time work and a more positive effect of Covid incidence in more urban municipalities.

Table 1.

Sign of correlation between urban characteristics and the IATEs for all considered real estate types and incidences of the pandemic

		Residential	Retail	Office
Covid incidence	Rent	$\mathbf{-}$	$\mathbf{-}$	$\mathbf{+}$
	Sale	·	$\mathbf{+}$	$\mathbf{+}$
Short-time work	Rent	$\mathbf{+}$	$\mathbf{+}$	$\mathbf{-}$
	Sale	$\mathbf{+}$	$\mathbf{+}$	$\mathbf{+}$

Note. (⁠$+$⁠) and (⁠$-$⁠) define correlations that are consistent across all characteristics. (·) denotes an ambiguous pattern. See Appendix D of the online supplementary material for the corresponding graphical illustrations.

The correlations summarized in Table 1 match with the sign of the ATEs in Figure 7, which is intuitive. Municipalities, which exhibit urban characteristics, receive a larger weight in the estimation of the benchmark specification. As a result, the sign of the ATEs is consistent with the correlation between the IATEs and urban characteristics.

6.2.1 Distributional implications and the urban premium

In order to analyse the effect of the pandemic on the price distribution, we examine the changes of rents and sale prices in a hypothetical scenario where all municipalities have experienced the lowest incidence of Covid or short-time work. By comparing these hypothetical expected changes with the actual changes, we can quantify the extent to which the pandemic has mitigated or exacerbated price pressure in the real estate market and in which price segment. This analysis provides insights into the affordability crisis for residential real estate and the urban premium for commercial real estate, given that real estate in affluent urban municipalities is typically associated with the upper part of the price distribution.

Rents and sale prices increased on average between 2019Q4 and 2021Q1 across all real estate types (see Figures 4 and 5 in Section 3.2). A comparison of the actual and hypothetical expected price changes reveals that the pandemic has triggered different price changes across municipalities with real estate in different price segments prior to the pandemic, which we now discuss in more detail.

Table 2 presents two key statistics: the difference between the actual and hypothetical expected growth rate of real estate prices for the top quintile of the price distribution in 2019Q4, and the difference in prices at the 90th and 50th percentiles (p90/p50 ratio) between the actual price distribution in 2021Q1 and the distribution of expected prices implied by the hypothetical scenario. A negative difference in the growth rate indicates that the pandemic reduced price growth in the top price segment, whereas a negative difference in the p90/p50 ratio suggests a reduction in the premium for real estate in the top price segment relative to the median price.

Table 2.

Differences between price changes in the actual and hypothetical scenario, for all considered real estate types and incidences of the pandemic

		Residential		Retail		Office
		Δ growth top quintile	Δ p90/p50	Δ growth top quintile	Δ p90/p50	Δ growth top quintile	Δ p90/p50
Covid incidence	Rent	$-0.9$	$-1.1$	$-9.3$	$-6.7$	$22.5$	$18.1$
	Sale	$-1.1$	$-0.1$	$13.7$	$1.3$	$26.7$	$3.2$
Short-time work	Rent	$0.7$	$-0.9$	$5.0$	$-0.7$	$5.4$	$6.1$
	Sale	$4.3$	$0.5$	$19.5$	$3.5$	$20.0$	$2.5$

Note. In each cell, the following statistics are displayed: the difference between growth rates in the top quintile in percentage points, and the difference between the p90–p50 ratios in percentage points.

The results in Table 2 show that the effect of a higher incidence of Covid or short-time work on prices in the top quintile of the price distribution of residential real estate is moderate. The incidence of the pandemic does not have a clear-cut effect on the affordability crisis of residential real estate. A higher Covid incidence reduces price growth in the top price quintile, while a higher incidence of short-time work increases it. The differences in the p90/p50 ratio for residential real estate prices are relatively small, indicating that the pandemic did not substantially change the premium for residential real estate in the top price segment.

For retail real estate, a higher Covid incidence reduces the growth rate of rents in the top price segment and decreases the p90/p50 ratio, suggesting a lower premium for retail real estate rents in the top price segment. In contrast, for office real estate, the pandemic increases the premium in the top price segment, especially for a higher incidence of Covid compared to short-time work. Sale prices of commercial real estate also experience increased premiums across all combinations of pandemic incidences and commercial real estate types. The difference in the growth rate of sale prices between the actual and hypothetical scenarios is substantial, illustrating that the pandemic has only temporarily, if at all, reduced rents in the top price segment.

How do these causal results relate to the descriptive negative correlations between the rent level in 2019Q4 and the subsequent rent changes until 2021Q1, which we report for both types of commercial real estate in Section 3.2? Recalling that the estimated causal effects are identified from variation in the incidence of the pandemic across municipalities, our results imply that a higher incidence of Covid in municipalities at the top of the rent distribution in 2019Q4 contributes to the negative descriptive correlation between the rent level and the subsequent rent changes for retail but not for office real estate. Additionally, a larger incidence of short-time work across municipalities does not drive the negative descriptive correlations.

Comparing the patterns observed in Tables 1 and 2, we note the similarity between the sign of the correlation of the individualized average treatment effects (IATEs) with (affluent) urban characteristics and the changes in the top part of the price distribution. This is intuitive because (affluent) urban municipalities typically have rents and sale prices that are in the top part of the price distribution. Thus, a larger premium for real estate in the top price segment is associated with a larger urban premium.

7 Discussion

Our results, which are summarized in Table 3, suggest that the pandemic has triggered changes in demand for real estate through possible channels discussed in Section 3.1, which have been reflected in rents and prices 1 year after the Covid shock. Our evidence is consistent with a higher incidence of short-time work increasing the demand for residential housing and retail space and temporarily reducing the demand for office space. A higher Covid incidence instead temporarily reduced the demand for retail space with less clear-cut effects for residential and office real estate. Our results for the pandemic effect on sale prices imply that the collateral value of commercial real estate increased on average, thus alleviating rather than tightening financial constraints. Government support and the moratorium of insolvency until the end of April 2021 may have prevented large downward price adjustments of real estate associated with times of crisis due to foreclosure and fire sales of real estate.

Table 3.

Summary of key results reported in Section 6

Average effects

$\bullet$ The pandemic had smaller effects on prices of residential real estate in Germany than on prices of commercial real estate, illustrating the relative stability of residential real estate valuations.

$\bullet$ In municipalities with a higher short-time work incidence, office rents temporarily decreased more.

$\bullet$ In municipalities with a higher Covid incidence, rents for retail space faced temporary downward pressure, indicating different effects across sectors.

Heterogeneity analysis

$\bullet$ Urban characteristics have strengthened the effects, highlighting the importance of understanding the context in which pandemic effects unfold.

$\bullet$ The implication of the pandemic for the affordability crisis in the German residential real estate market has been nuanced: whereas short-time work incidence exacerbated the crisis by driving up rents in high-price areas, Covid infections had a contrasting effect.

$\bullet$ The pandemic has increased the premium for commercial real estate in the top segment of the price distribution.

For residential real estate in particular, one may wonder whether the effects resulted from migration or demand shifts of residents. Descriptive evidence suggests the latter because the size and dispersion of net migration flow across municipalities (per resident) are small in Germany. Figure 13 in Appendix B.2 of the online supplementary material shows that net migration flows in German municipalities in 2020 have been at most 2 per mile on average in each decile of the sale price distribution of the residential real estate. The figure further confirms that there has been no significant change to pre-pandemic trends in these net migration flows during the pandemic: net inflows into municipalities with high sale prices have been decreasing, whereas net inflows into municipalities with low sale prices have been increasing, implying convergence of net migration flows over time across municipalities in different price segments of the residential real estate market. The stable evolution of migration flows implies that changes in these flows until 2021Q1 may have been priced in for real estate units, at least partly, already in 2019Q4. Together with the small size and dispersion of net migration flows, this implies that the price changes identified using variation across municipalities were driven by shifts in the demand for real estate services within municipalities, rather than by changes in the number of residents across municipalities.

The pandemic thus triggered much less net migration in Germany than in the U.S. Mondragon and Wieland (2022) report in their Table 1 that the dispersion of net inflows across 895 CBSAs in the U.S. (core-based statistical areas, i.e. economically connected units of counties) doubled during the pandemic implying a standard deviation of $2.4$%. The superstar cities in the U.S. lost sizable parts of their population during the pandemic (Van Nieuwerburgh, 2023). This major difference between the U.S. and Germany is consistent with the more muted price effects in the real estate market in Germany relative to the U.S. (Gupta, Mittal, Peeters, et al., 2022; Gupta, Mittal, Van Nieuwerburgh, 2022; Rosenthal et al., 2022).¹³

Other differences across the two countries that are common across municipalities cannot explain these different findings because the estimated effects are identified by the different incidences of the pandemic across municipalities. Thus, determinants of real estate prices that are common across municipalities are differenced out. Such determinants could provide a challenge for the interpretation if they interacted with characteristics of the municipalities that are not contained in the large set of covariates that we control for in our analysis.

Our estimates do not account for spatial correlations and equilibrium effects (spill-overs) across municipalities. Equilibrium effects may reduce the size of our estimated effects, for example, if households or firms shift their demand to municipalities with relatively lower prices. In this case, our estimates provide a lower bound for the direct effects. The small size of net migration flows into municipalities and the similar migration patterns in 2019 and 2020 discussed above suggest that such effects have been less relevant in Germany than in other countries.

Another concern may be that real estate prices and the incidence of the pandemic evolve jointly in equilibrium and are shifted by other confounders. We measure the price changes in 2021Q1, i.e. after the cumulative incidence of the pandemic that is measured between January until December 2020. Thus, endogeneity may be an issue if the EMH were violated so that some of the price changes would be correlated with characteristics of municipalities that also affect the behaviour of residents relevant for the incidence of the pandemic. The robustness of our findings in the specification including a large set of covariates may mitigate this concern, as it limits the sources of endogeneity which may confound our results.

8 Conclusion

The average response of real estate prices to the pandemic shock was stronger for commercial than residential real estate in Germany. The negative effect of the Covid incidence on rents for retail real estate and the negative effect of short-time work on rents for offices have been expected to be temporary 1 year after the pandemic started. Overall, the effects of the pandemic on rents and asset prices in the real estate market are more muted in Germany than in the U.S., consistent with the much smaller changes in the net migration flows across German municipalities observed in the aftermath of the pandemic.

We find heterogeneous effects across municipalities, with stronger effects in more urban areas that were already in the high-price segment of the real estate market prior to the pandemic. Quantitatively, however, the price changes in residential real estate did not change the housing affordability crisis in Germany.

For commercial real estate, we have found that a higher Covid incidence temporarily reduced rents for retail real estate, particularly in the top price segment, thus reducing the premium for more expensive commercial real estate. The effect has been only temporary, however, because a higher incidence of Covid or short-time work both increased real estate asset prices. We have found that the increase has been stronger in the top price segment thus contributing to a larger premium for commercial real estate in more affluent urban municipalities.

As more data become available for the pandemic episode and beyond, it will be interesting in future research to obtain a clearer picture about which effects of the pandemic on the real estate market have been truly temporary and which effects have been permanent by persistently changing the behaviour of households and firms or by accelerating trends.

Acknowledgments

We thank 21st Real Estate for providing their granular data, the associated support required for organizing the data, and the comments by Alexander Konon. We further thank Raphael Hausheer and Stephan Minger for their research assistance and colleagues as well as participants at workshops for helpful comments.

Funding

We acknowledge support by the Swiss National Science Foundation (SNSF), Switzerland under project 1000018-200917.

Data availability

Replication code is published on the Harvard Dataverse (Heiniger, 2024). Data are proprietary to 21st Real Estate and cannot be shared.

Supplementary material

Supplementary material are available online at Journal of the Royal Statistical Society: Series A.

Footnotes

1

By price responses, we mean the response of the asset price of real estate (referred to as sale price or price) as well as the rental payments made to landlords for the utilization of real estate (referred to as rents).

2

The so-called affordability crisis or housing crisis is relevant in many developed countries. Deutsche Welle (2021) is an example of the discussion on the affordability crisis in the German context.

3

Statistics with IDs 70370 and 70094, retrieved on 16 October 2021

4

Based on data from the last micro census in 2018, retrieved from Federal Statistical Office on 15 October 2021

5

Data retrieved from Federal Statistical Office and empirica-regio on 1 June 2022

6

See the series Q:DE:R:628 provided by the Bank of International Settlement (BIS). Other indices imply similar changes of 1.1% (Pfandbriefbanken), 1.8% (Destatis), and 2.1% (Hypoport).

7

See the series Q:DE:0:C:0:2:6:0 provided by the Bank of International Settlement (BIS) or the index provided by Pfandbriefbanken.

8

Figures 3 and 5 show that the level and the change of sale prices have similar patterns as described for rents. Figure 14 in Appendix C.1 of the online supplementary material further displays the regional patterns in the sale price to annual rent ratio. The population-weighted average of the price-rent ratio is 27 for residential real estate, in line with values reported in Kindermann et al. (2022), Figure 4a.

9

Although the EMH applies to asset prices, the rent and the fundamental asset price map into each other in efficient markets. As is well known, a particularly simple linear mapping obtains between the current rent and the sale price if one assumes constant rent growth and discount rates (Campbell & Shiller, 1988).

10

See the evidence mentioned in Section 3.1 suggesting that changes in the risk-free rate or risk premia do not explain these patterns.

11

Estimation results for the GATEs are not reported for brevity and are available on request.

12

We use a K-Means algorithm to cluster the IATE estimates without a default number of targeted clusters. If more than three clusters are generated, we aggregate them into three clusters of similar size, respecting the ordering imposed by the size of the effect. The resulting clusters by the IATE size, denoted by low, mid, and high and shown for example in Figure 11, should be distinguished from the three categories of the incidence that are also denoted with low, medium, and high.

13

A limitation of these comparisons is that our estimates are identified differently by comparing the effect across municipalities with different incidences of the pandemic and thus abstracting from the effect of the pandemic that is common across all municipalities. The journal issue edited by Yiu et al. (2023) provides further international evidence on the response of housing prices during the pandemic.

Author notes

Conflicts of interest. None declared.