Sectoral Booms and Misallocation of Managerial Talent: Evidence from the Chinese Real Estate Boom1
Author: Ms. Yu Shi

Contributor Notes

Author’s E-Mail Address: YSHI2@imf.org

This paper identifies a new mechanism leading to inefficiency in capital reallocation at the extensive margin when an economy experiences a sectoral boom. I argue that imperfections in the financial market and capital barriers to entry in the booming sector create a misallocation of managerial talent. Using comprehensive firm-level data from China, I first provide evidence that more productive firms reallocate capital to the booming real estate sector, and demonstrate that the pattern is likely driven by fewer financial constraints on these firms. I then use a structural estimation to verify the talent misallocation. Finally, I calibrate a dynamic model and find that the without the misallocation, the TFP growth in the manufacturing sector would have improved by 0.5% per year.

Abstract

This paper identifies a new mechanism leading to inefficiency in capital reallocation at the extensive margin when an economy experiences a sectoral boom. I argue that imperfections in the financial market and capital barriers to entry in the booming sector create a misallocation of managerial talent. Using comprehensive firm-level data from China, I first provide evidence that more productive firms reallocate capital to the booming real estate sector, and demonstrate that the pattern is likely driven by fewer financial constraints on these firms. I then use a structural estimation to verify the talent misallocation. Finally, I calibrate a dynamic model and find that the without the misallocation, the TFP growth in the manufacturing sector would have improved by 0.5% per year.

1 Introduction

Economic cycles are often driven by sectoral shocks (e.g., oil price shocks, housing booms, etc.), accompanied by capital reallocation across sectors. Most of the literature studying capital reallocation focuses on the intensive margin: within a firm, how much capital is reallocated to take advantage of good investment opportunities (Almeida and Wolfenzon, 2005; Eisfeldt and Rampini, 2006, 2008; Cui, 2014). In this paper, I instead study the implication of capital reallocation at the extensive margin: the decision of whether or not to reallocate capital to the booming sector. I use comprehensive firm-level data from China to show that when a sector with a high entry barrier booms, financial market imperfections could lead to suboptimal captial reallocation at the extensive margin and a misallocation of managerial talent across sectors.

What would give rise to the misallocation of managerial talent? In short, it would happen due to a misalignment between firms’ wealth and managers’ comparative advantage in the booming sector. In an economy without any barriers to capital reallocation, managers would choose projects that generate maximum returns to capital. In equilibrium, the decisions to reallocate capital to the booming sector would depend solely on managers’ comparative advantage. However, with financial frictions and barriers to entry to the booming sector, less wealthy managers would be constrained from reallocating capital to the booming sector. A misallocation of managerial talent would occur, as long as the unwealthy and thus constrained managers do not have the comparative advantage in their current sector.

Analyzing the misallocation of managerial talent in real estate booms is particularly pertinent for China, due to the prolonged real estate boom, the highly imperfect financial market, and the massive entry regulations in the land market. Additionally, there is apparent social inefficiency. Since the entry barrier to the real estate sector often varies with land prices, when entering the real estate sector, wealthy firms would bid up land prices, making poor firms even more constrained and thus generating a pecuniary externality.

In this paper, I first provide motivating reduced-form evidence of private firms reallocating capital to the real estate sector. The reduced-form evidence highlights the importance of financial market imperfection in determining capital reallocation at the extensive margin. First, among private firms in the manufacturing sector, the more productive ones are more likely to start real estate businesses. However, this pattern is possibly driven by more productive private firms being less financially constrained. After controlling for initial asset value and credit scores, I show that it is the unproductive firms that are more likely to move to the real estate sector. Second, local real estate booms, but not manufacturing investment opportunities, cause firm-level capital reallocation to the real estate sector. To establish this fact, I use an instrumental variable approach and exploit factors that are only crucial to returns on real estate development: firm connections with the local land bureau interacted with local land supply constraints. I classify the potential links between managers and the head of local land bureau based on their ethnic enclaves1. My identifying assumption is that in cities with constrained land supply, managers with potential connections with the local land bureau do not face worse opportunities in the manufacturing sector. The assumption is reasonable, given that manufacturing firms with better access to the scarce land factor would at least do as well compared to the situation when land supply is adequate.

Second, I structurally estimate the comparative advantage of private-firm managers to formally conclude the existence of misallocation. The model follows the Roy models in the literature (Costinot et al., 2012; Adao 2015; Hurst et al, 2016) to assume that managers can produce at constant returns to scale. Managers are divided into 20 groups based on pre-determined characteristics: education and experience. Talent distribution is homogeneous within groups2 and heterogeneous across groups, with the group-average productivity being non-time-varying. Identification of the comparative advantage of each group comes from comparing the observed returns on capital in the manufacturing sector and the real estate sector for only managers who entered the real estate sector during the boom. To account for the endogeneity in selection into the real estate sector, I proxy for the selection bias using a non-parametric function of manager and firm characteristics, and local real estate factors3 as instruments (Das, Newey, and Vella, (2003)). My estimation implies that managers running more productive manufacturing firms also have a comparative advantage in the manufacturing sector, so they should not have moved to the real estate sector in the boom.

Using a two-period model, I then demonstrate that in addition to the misallocation of managerial talent, financial market imperfection also leads to social inefficiency. The key is the existence of a pecuniary externality given that managers do not internalize the impact of their capital reallocation on land prices. If only managers with ample wealth choose to start developing real estate properties, land prices would rise. This result leaves other managers more constrained by the entry barrier. Financial market imperfections are essential not just to creating distortions at the extensive margin. In the context of China, these imperfections also imply a positive correlation between firm-level wealth and managerial talent4. As a result, talented manufacturing managers who also have a comparative advantage in manufacturing accumulate more wealth than the untalented ones. Therefore, they are more likely to be unconstrained from moving to the real estate sector.

Finally, to quantify the aggregate impact of talent misallocation, I calibrate the Chinese economy in transition to a balanced growth path5. The key innovation of this paper is to jointly estimate the comparative advantage of mnagers and the entry barrier in the real estate sector. In calibrating the model, both factors map to the average productivity of managers who decide to reallocate capital to the real estate sector. Therefore, the standard calibration approach would run out of degrees of freedom. I solve this problem by taking the comparative advantage estimate from the structural estimation and then using the aggregate moment to pin down the entry barrier.

Using the quantitative framework, I analyze three policy tools that are potentially welfare-improving: liberalizing the financial market, reducing the entry barrier in the real estate sector, and taxing the returns from operating in the real estate sector. Liberalizing the financial market improves social welfare but also aggravates the social inefficiency. In a more liberalized financial market, the already wealthy firms can borrow more, which further increases land prices. As a result, more firms with low wealth and the comparative advantage in the real estate sector are constrained by the entry barrier. Lowering the entry barrier in the real estate sector has an ambiguous impact on welfare. It creates more investment opportunities so generally untalented managers continue to run their businesses, resulting in a decline in total credit supply. A taxation on real estate returns can improve both social welfare and efficiency. The gains are economically significant: a 3% tax on real estate returns improves manufacturing productivity by 0.3% after the boom starts.

This paper contributes to the literature on capital reallocation by studying the decisions at the extensive margin. Previous research emphasizes the intensive-margin distortions that prevent firms to take advantage of new investment opportunities (Almeida and Wolfenzon, 2005; Eisfeldt and Rampini, 2008; Cui, 2014) and the cyclicality of within-firm capital reallocation (Ramey and Shapiro, 1998; Eisfeldt and Rampini, 2006). I show that the extensive-margin decision on whether or not to reallocate capital is also of economic significance.

This paper also relates closely to the literature on the allocation of talent. Murphy, Shleifer, and Vishny (1991) propose that some social reward structures may result in more talented specializing in unproductive rent-seeking activities that leads to to stagnation. More related to my paper, Legros and Newman (2002) argue that financial imperfections can lead to the non-monotonic specialization of entrepreneurs, as evidenced in my empirical observations. I take their arguments one step further to show that financial market imperfections can also lead to an inefficient allocation of managerial talent. Moreover, certain policy tools can be used to correct for such inefficiency.

More generally, this paper makes a significant contribution to the literature studying misallocation and economic growth. Most of the existing work is about resource misallocation, which again focuses on the intensive margin. (Hsieh and Klenow, 2009; Song et al. 2011; Buera and Shin, 2013). A few papers document that capital entry barriers can prevent productive entrepreneurs from producing (Buera, Kaboski, and Shin, 2012; Midrigan and Xu, 2014). However, they do not explore the comparative advantage of entrepreneurs, and thus with idiosyncratic productivity shocks alone, there would only exist short-term misallocation of talent. This paper shows that financial imperfections can lead to a longer-run talent misallocation in the absence of idiosyncratic shocks. It also complements the literature by providing substantial empirical evidence on talent misallocation and jointly estimating managers’ comparative advantage and sectoral entry barriers.

Broadly, this paper also relates to the literature on the relationship between house prices and the macroeconomy. Previous research outlines the collateral channel (Hurst and Lusardi, 2004; Chaney, Thesmar, and Sraer, 2012; Mian and Sufi, 2011; Mian, Sufi and Rao, 2013;Schmalz, Sraer and Thesmar, 2015; Kerr, Kerr, and Nanda, 2015), and thus real estate booms are beneficial to the macroeconomy. The paper that is most similar in spirit to mine is Charles, Hurst, and Notowidigdo (2015). They argue that booming labor demand in the construction sector results in a decline in college attendance. This paper, in contrast, emphasizes the distortion on managers’ choice to reallocate capital from their current sector to the booming sector, dampening the impact of a real estate boom on productivity.

Last but not least, this paper fits into the growing literature studying the Chinese real estate market. Fang et al. (2016) and Deng, Gyourko, and Wu (2015) document a rapid real estate price appreciation and substantial variation in city-level real estate prices in China. Chen et al. (2015), Chen and Wen (2014), and Shi et al. (2016) discuss the misallocation of capital and labor in this real estate under the assumption that the real estate boom is a bubble episode. Given that real estate prices in China are still climbing in the first-tier and second-tier cities, this paper suggests a framework in which inefficiency exists without assuming a bubble episode.

The rest of the paper is organized as follows: section 2 describes the institutional background of real estate market reform and the housing boom in China. Additionally, I provide a detailed description of firm-level data sets used in the analyses. Section 3 documents the empirical findings on the allocation of managerial talent in the real estate boom in China. Section 4 provides a structural estimation of the comparative advantage of managers and discusses the social efficiency in the setting of a two-period model. I also discuss several policy tools that can help improve the social efficiency of managerial talent allocation. Section 5 quantifies the impact of the misallocation using a dynamic general equilibrium model. Section 6 concludes the paper.

2 Institutional Background and Data

2.1 Institutional Background

The Chinese real estate sector has gone through a prolonged boom in the last two decades. In the residential property market, state-owned banks started issuing mortgage loans to households in the late 1990s. Local governments, SOEs, and private companies compensated workers on housing purchases made out-of-pocket. These major policy changes contributed to buoyant housing demand, leading to a 10% annual 6 growth in both the price and quantity of houses from 1999 to 2010.

In the land market, reform went hand-in-hand with the residential property market. According to legislation, the state has complete ownership of all urban land. In 2002, the central government started allocating the use rights of land7 through public auctions, which marked the beginning of a fully privatized land market8. The execution of this reform was completed in 2004. City-level land bureaus were established between 2004 to 2006 to hold responsibility for city planning and the sales of urban land use rights. Private agents then actively engaged in residential land development9 in response to the real estate boom.

Several of China’s institutional features provide an ideal environment for examining the distortions of capital reallocation at the extensive margin. First, private firms in China are financially constrained (Allen, Qian, and Qian (2005); Dollar and Wei (2007); Riedel, Jin, and Gao (2007)). Second, local land bureaus require private agents to submit a cash deposit before participating in land auctions, creating a capital barrier to residential land development. No other sources of financing substitutes, such as bank loans or equities, are allowed. The cash deposit requirement varies from 10% to 50% of the reservation land value. With rising land prices, the average cash deposit10 increased to 67.79 million RMB in 2012. Given that the annual median profit in the manufacturing sector is only 0.56 million RMB, the required cash deposit is an almost impossible hurdle for most private manufacturing firms in China. Finally, all residential land developers are required to register a real estate firm and obtain a land development license. The licensing requirement allows me to obtain a complete record of private agents participating in residential land development. Thus I document existing firms reallocating capital to the the real estate sector by looking at the enterprise registration information.

To summarize, I intend to convey that:

  1. The real estate market in China experienced a massive privatization starting in early 2000s, which led to a prolonged boom and the expansion in land development industry. Most developers focus on residential real estate development.

  2. The real estate sector faces a high capital entry barrier. Potential developers need to possess a large amount of capital and cash to enter the land market.

  3. The licensing requirement helps identify capital reallocation of existing firms to the real estate sector.

2.2 Data

Firms are the key units of observation in my analysis. Two firm-level data sets provide information on operation and investment activities: the Annual Industrial Survey (hereafter AIS) conducted by the National Bureau of Statistics (hereafter NBS), and the Enterprise Registration Database (hereafter ERD). The AIS is a panel survey from 1995 to 201011 that covers all state-owned enterprises (SOE) and privately-owned enterprises with revenue above five million RMB12. It provides information on firm balance sheets, income statements, and cash-flow statements. I drop SOEs, publicly-owned firms, and foreign branches from the sample and focus only on privately-owned firms with a single CEO or executive manager. The SOEs are considered to be less financially constrained and more inefficiently operated13, so they would not fit in my model. In the publicly-owned firms and foreign branches , managers have a less dominant role in making firm-level decisions, so I exluce listed firms and foreign branches from my sample. This then leaves me with a sample of 105,298 firms and 1,368,897 firm-year observations.

I rely on the AIS to construct firm-level variables on R&D expenditure, investment expenditure, and manufacturing productivity. The investment expenditure and R&D expenditure are normalized by one-year lagged value of fixed assets14. In this paper, I consider R&D expenditure as a specific type of investment that leads to new product variety, so I normalize it in the same way as I do for investment expenditure. My empirical results, however, are robust to alternative methods of normalizing the R&D expenditure, such as using one-year-lagged total output or one-year-lagged total asset value. Last but not least, I consider two measurements of firm-level productivity in producing manufactured goods: labor productivity and quality-based total factor productivity (TFPQ). The labor productivity is measured as value added per worker; the TFPQ is measured following Hsieh and Klenow (2009) who adjust for quality differences across manufacturing producers.

I control for firm-level heterogeneity with other accounting variables from the AIS. Firm debt dependence is proxied by the ratio of total debt outstanding to the one-year-lagged book value of total assets (the debt-to-asset ratio). The return on asset (ROA) is the ratio of operating income (after depreciation) to the one-year-lagged total asset value. I also include initial book value of asset as a proxy for wealth and the total value of fixed assets to control for capital input. The age of a firm sums up the active years since its registration. Table 1 provides summary statistics of investment and productivity of privately-owned firms in the AIS, as well as the summary of corporate financial variables used in the empirical analysis. The AIS also includes the location of each firm’s headquarters. None of the firms in the sample moved their headquarters between 1995 to 2010. I utilize this property to explore the variation in local real estate markets in predicting business owners’ incentives to enter the real estate sector.

I then pair the AIS with my second data set, the ERD, which covers all enterprises in China. The main purposes of using the ERD are to identify existing private firms moving to the real estate sector and to construct measures of manager characteristics. At the date of registration, all firms are required to disclose their legal representative, shareholders, board members, executives, and other basic information to the China State Administration for Industry and Commerce. Each firm then has a unique record in the ERD. I consider an existing manufacturing firm as entering the real estate sector when the firm becomes the major shareholder of a newly established real estate company. Until 2010, 7,559 privately-owned manufacturing firms in the AIS have become real estate developers. These firms are not a neglectable group as they control for a large fraction of total assets in sample - 22%. They also gain absolute controls of their real estate subsidiaries by holding on average 66.27% of the total shares outstanding. The characteristics of managers include the level of education, birth date, Communist Party membership, home county15, gender, and race. There are a total of 2,854 counties in China, each with an average population of 610,00016, so that a county is comparable to an average U.S. city.

The ERD is matched to the AIS using the unique registered name of each company. The matching rate is more than 80% for the entire data sample. The matching is unlikely to induce a non-random error to my empirical analysis. I drop around 3% of the observations in which the shareholders and their real estate subsidiaries have the same legal representative. This is to ensure the complete separation of the manufacturing firms’ real estate investments and their manufacturing investments. I further restrict the sample to firms active since 1997 to guarantee that pre-period controls exist. This leaves me with a sample of 25,513 firms and 382,695 firm-year observations.

Table 1:

Firm-level Summary Statistics

article image
Notes: This table summarizes the main variables used in the empirical exercises. The data are from the NBS Annual Industrial Survey, the Enterprise Registration Database, and the 2008 National Economic Census. The R&D intensity is measured as R&D expenses normalized by the lagged value of total fixed asset (including properties, plant and equipment), following OECD measurement on relative degree of investment in generating new knowledge. The labor productivity is defined as the logarithm of value added per worker, and the investment rate is capital investment normalized by total fixed assets in the previous year. The age of a firm is the number of years since registration. The debt issuance variable is computed as the ratio of outstanding debt to total asset value. The net cash inflows is defined as the net cash inflows from financing, operation, and investment, normalized by the lagged book value of total fixed asset. Return on asset is measured as operation income (after depreciation) of the total book value of assets. For variables in the manager data, minority, male, and Communist party member are three 0-1 dummy variables, so I only report their mean values.

In addition to firm-level data, I collect city-level demographic data from city yearbooks17 and real estate data from the China Real Estate Index System18 (CREIS). The cities are prefecture-level cities19. There are 333 prefecture-level cities in China, with an average geographical size comparable to the metropolitan statistical areas in the US. The CREIS provides monthly data of the residential property market in 122 cities, including the total floor area and revenue of houses sold and property-specific characteristics, and transaction-level data of the land market in 145 cities since 2005. For each land transaction, the database records its location, total land area, desired floor space area, required cash deposit, reservation land value, and final price. For robustness, I supplement the CREIS with house price indices from Fang et al. (2014), who use mortgage data to construct quality-based price indices for 120 cities. The Chinese cities are officially divided into three tiers based on their economic activities. Table 2 provides a summary of the real estate data based on the regional location and three-tier divisions of the cities.

Table 2:

Summary Statistics of the Real Estate Data

article image
Notes: This table summarizes the real estate data of the four economic regions and three tiers of cities in China. The house price, in thousands of Yuan, is the average price per square meter floor. The city-level average house price is computed by dividing total home value by the total floor area sold using National Bureau of Statistics (NBS) published from 2003 to 2013. The land price, in thousands of Yuan, is the average price per square meters of land area. The land market data is obtained from China Real Estate Index System (CREIS). The land supply elasticity index is constructed manually by the author, following Saiz (2010). The range of the land supply elasticity is from 0 to 1 .

There are several advantages to using the Chinese data. First, the boundary of firms is determined based on legal dependency. Chinese corporate law requires each firm to have one legal representative responsible for all litigation against the firm. Two establishments with different legal representatives is considered as two separate firms. Thus, the return of a manufacturing firm as a real estate developer is reported separately as long as the two businesses are registered with different legal representatives. This appears to be a standard practice. 97.1% of manufacturing companies who entered the real estate sector registered their real estate subsidiaries as separate firms. This feature makes it possible to study the capital reallocation from the manufacturing sector to the real estate sector.

Second, the shareholding data in the ERD provides a near-complete documentation of private manufacturing firms’ capital reallocation to the real estate market. Chinese regulation requires that only real estate development firms with a development license can participate in land auctions, with the exception that industrial companies can rent industrial land for factory buildings. Given that the boom happened mostly in the residential and commercial real estate market, and that industrial land prices did not increase during the housing boom20, I consider the entries observed in the ERD as capturing manufacturing companies’ significant real estate investments21. Such investment is also isolated from the standard structure investment in the manufacturing business. In fact, I do not want to take into account the property or land holdings on the balance sheet of the the manufacturing firms, given that these holdings can be used for manufacturing production later. As noted in Chaney, Thesmar, and Sraer (2012), cost-minimizing firms invest more in structures during real estate price run-ups. The entries into the real estate sector identified from the ERD are separated from the firms’ businesses in the manufacturing sector.

Finally, most real estate developers in China are restricted to operating within their cities, except for those granted an A-class license. These A-class companies, which make up less than 1.3% of the total number of firms in the industry, are often stand-alone firms without a large institutional shareholder. I also verify from the ERD that over 90% of manufacturing firms choose to start up real estate businesses locally. Therefore, I can explore the city-level variation house prices and supply constraints to model the incentive to enter the real estate sector.

3 Capital Reallocation: A Reduced-Form Analysis

In this section, I present reduced-form evidence on existing private firms reallocating capital to the real estate sector in the Chinese real estate boom. Since early 2000s, the Chinese residential house prices persistently increased at an annual rate of more than 10% (Fang et al., 2016; Deng, Gyourko, and Wu, 2015). Meanwhile, a significant number of existing private firms started developing residential real estate properties. Figure 1 shows the decomposition of the entrants22 in the real estate development industry. Until 2013, almost 18% of the newly established real estate firms were controlled by non-real estate and non-financial sectors. I focus on the 3% of the new entrants run by manufacturing firms so I can study their performance in the manufacturing sector before and after entering the real estate sector.

Figure 1:
Figure 1:

The Entrants in the Real Estate Sector

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

Two main findings to be discussed later suggest that the real estate boom causes the more productive manufacturing businesses to reallocate capital to the real estate sector:

  1. Firms that entered the real estate sector had substantially higher measured productivity, investment rate, manager education and manager experience prior to entry. Controlling for firm-specific pre-boom asset value and credit score, I find that the patterns are reversed — among private firms with similar credit market access, it is the ones with inferior performance or more inexperienced managers that are more likely to start up real estate businesses in the boom.

  2. The real estate boom causes within-firm capital reallocation from the manufacturing sector to the real estate sector. Precisely, entering the real estate sector leads to a significant decline in the firm investment and R&D expenditure in the manufacturing sector. I establish the causal evidence using an instrumental variable approach and a propensity-score matching approach.

These two facts together characterize the extensive-margin decisions of private business owners and highlight the importance of financial market imperfections. Being financially constrained, a private manager would face the trade-off between real estate and manufacturing investment opportunities. Since the real estate sector has a substantial entry barrier, a manager’s credit market access becomes a key determinant for starting up real estate businesses. Upon the manager entering the real estate sector, the costly land acquisitions might still result in declines in investment, R&D expenditures, and productivity in the manufacturing sector23.

The reduced-form analysis, however, is inconclusive on the social optimality of the allocation of managerial talent. The entry pattern observed in data could be optimal only if these productive manufacturing firms are even more productive in the real estate sector. In section 4, I develop a structural model to estimate the comparative advantage of managers and draw a conclusion on the social optimality.

3.1 The extensive margin: did the better or worse managers move to real estate?

To find out the determinants of the allocation of managerial talent, I compare private firms that chose to enter the real estate sector in the boom and the ones that did not. Table 3 summarizes their differences in productivity, investment, profit margin, size, and other characteristics prior to the 2002 land privatization. Before entering the real estate sector, the private manufacturing firms that later entered the real estate sector controlled 29.6% of the total fixed assets in the entire manufacturing sector. They were 5.63 times larger in terms of total assets and twice as large in R&D expenditure and capital investment compared to firms that did not enter. These firms also had a 23.4% higher profit margin and a significantly higher credit score. Besides, older firms with larger asset values and larger workforce were more likely to enter the real estate sector.

Table 3:

Comparison Between Real Estate Entrants and Other Firms

article image
Notes: This table compares the manufacturing firms that entered the real estate development industry with firms that did not. The total fixed assets and asset value are presented in millions of Yuan. The fraction of firms is computed as a ratio of the total number of firms in the sample. The fraction of fixed assets (asset value) are also computed as a ratio of the summation of the total fixed assets (total asset value) of all firms in sample.

Additional analysis at the manager level implies that firms with higher levels of manager education or more experienced managers tend to move into the real estate sector. Using the CEO information from the ERD, I divide the CEOs of private manufacturing firms into 20 groups based on their work experience and their education. Table 4 summarizes the fraction of managers who decided to enter the real estate sector in each group.

Table 4:

The Fractions of Managers that Moved to Real Estate

article image

Next, I estimate a firm-specific propensity for entering the real estate sector using a probit model:

Prob(Enteri)=α1ΔPc+α2Δwc+β1Performancei+controlsi(1)

The cross-sectional regression uses the average firm characteristics from 1997 to 2002 as right-hand-side variables. Enteti is a dummy variable indicating whether or not private manufacturing firm i ever entered the real estate sector after 200424; ΔPc is the average house price growth of city c (where the headquarter of firm i is located) from 2004 to 2010; Δwc is the average annual local manufacturing wage growth after 2004; Petfotmancei is a set of measures of firm i’s performance in the manufacturing sector, including labor productivity, revenue-based TFP (TFPR), quality-based TFP (TFPQ25), profit margin, investment rate, and R&D intensity. Firm-level controls include pre-2002 average values of age, employment, debt-to-asset ratios, as well as the initial exporting status and two-digit industry. I exclude about 3% of the companies in sample that already invested in the real estate sector before the land market reform concluded in 2004. The results are summarized in column (1) of Table 526: an average firm’s probability of entering the real estate sector is positively correlated with local real estate price appreciation and firm-specific performance measurements.

Table 5:

Determinants of Entering the Real Estate Sector

article image
Notes: This table presents the determinants of entering the real estate development industry. Columns (1), (3), (5), (7) report the correlations between entry and investment and productivity without controlling for firm-specific financing ability. Columns (2), (4), (6) and (8) add the average pre-period total asset value and credit score as proxies for firm-specific financing ability. Control variables include five quantiles of employment, age, and exporting status. City fixed effects and 2-digit industry fixed effects are also included in all specifications. T-stats are reported in parentheses.*** Significant at the 1 percent level** Significant at the 5 percent level* Significant at the 10 percent level

Despite the somewhat surprising result, it yields no implications to the misallocation of talent, nor to the importance of financial market imperfections. Thus, I further included firm-level initial asset value in 1997 and credit scores in specification (1). The two controls serve as proxies for firm-specific credit market access. All else equal, a company with higher initial asset value and a higher credit score is more likely to be able to borrow more from the banking sector. Columns (2) in Table 5 present the results. While local house price appreciation still significantly predicts the likelihood of manufacturing firms entering the real estate sector, a firm with 1% higher pre-entry TFPQ has a 0.09% lower propensity to enter the real estate sector. Similar patterns can be found when replacing the performance measure with labor productivity and R&D expenditure.

These results suggest that firm credit constraint is an important determinant of private business owners’ decisions to enter the real estate sector. As summarized in Table 1, existing business owners with a higher productivity in the manufacturing sector are larger and have higher credit scores. Therefore, the financial market imperfection increases the probability that more productive private manufacturing businesses move to the real estate sector in a real estate boom. With firm-specific financing ability controlled for, the pattern is reversed: the more productive private manufacturing businesses are more likely to stay in the manufacturing sector.

3.2 The intensive margin: what is the impact of moving to real estate?

This section establishes the causal effect between the real estate boom influencing capital reallocation within private manufacturing businesses. In other words, entering the real estate sector resulted in a significant decline in the investment and R&D expenditure in the manufacturing business. I examine the impact of moving to the real estate sector using the following panel specification:

Yit=αi+δt+βPOSTit+kηktXki+controlsit+it(2)

Yit is the outcome of interest, including the R&D intensity, investment rate, and labor productivity of the manufacturing business of firm i in year t. POSTit = 1 if manufacturing firm i has already entered the real estate sector in year t, and POSTit = 0 otherwise. Xki are dummies indicating the categorization of firm i into group k. The groups are constructed based on initial conditions including exporting status, size class, two-digit industry, and five quantiles of age at 1997. ηkt is a group-year fixed effect, which control for average performance of initial control group k at year t. Local wages, industry-year specific capital-to-labor ratio, and other economic indicators are also added as controls in controlsit. β is the key coefficient of interest, measuring the average change of firm i’s manufacturing production following its entry into the real estate sector. This sample includes the manufacturing firms in the AIS from 1997 to 2010. I identify POSTit through observing manufacturing firms holding shares of a newly established real estate development company in the ERD data27.

The OLS estimate β^ is subject to two omitted variable biases: manufacturing investment opportunities and credit constraints. The unobserved investment opportunities induce a self-selection bias: firms losing good opportunities in the manufacturing sector may reduce manufacturing investments and R&D activities and seek for opportunities in the real estate sector. Thus, the OLS estimate would be biased downward. On the other hand, both investment and R&D expenditures are sensitive to firm credit access, likewise the decision to enter the real estate sector. As is shown in Table 1, firms that entered the real estate sector had a higher asset value, higher investment, more active R&D action, and a higher credit score. The unobserved credit constraint then induces an upward bias on the OLS estimates.

I use an instrumental variable approach and a matching approach (see Appendix B for the matching approach) to deal with the omitted variable biases. Ideally, the instrument should be correlated with firm-specific likelihood of entering the real estate sector, but uncorrelated with either its manufacturing investment opportunities or its credit constraints. While I could not find such an ideal instrument, I explore cross-city and within-city variations to construct an instrument that is arguably orthogonal to firm manufacturing investment opportunities. Therefore, I consider the IV estimates as upper bounds of the actual effect of entering the real estate sector on private firms’ manufacturing production.

The Instrumental Variable Approach

I instrument for firms’ decisions to enter the real estate sector by exploring city-level land supply elasticity and managers’ potential connections with the local land bureaus. Land supply elasticity is related to land appreciation during the construction period such that it predicts the return from developing real estate properties. Managers’ potential connections with the local land bureaus exploit the institutional features of China, given that the land bureaus have absolute control of land sales access to information would be helpful for the firms.

I manually construct a land supply elasticity index28 for 145 cities following Saiz (2010). A city with a land supply elasticity index of 1 means that all areas within 30 kilometers of the city center can be developed into residential or commercial properties. Land supply elasticity matters more through expected house price appreciation, but not the actual land acquisition. I explain in detail the construction of the elasticity index and its validity as part of my instrument in Appendix A.

One caveat of using the city land supply constraint alone is that it can pick up local equilibrium effects that are correlated with manufacturing investment opportunities (Davidoff (2016)). Therefore, I further explore the firm-level heterogeneity in managers’ potential connections with the local land bureau ministers. I define the potential connection between the CEO of a firm and the local land bureau minister based on their ethnic enclaves - that is, whether or not they grew up in the same county. I collect the birthplaces of the CEOs from the ERD and the birthplaces of the local land bureau ministers from their official resumes29. Zhang, Xie (2013) and Edin et al. (2003) justified the significant economic impacts of ethnic enclaves in China. Given that most land bureaus were established after 2004 and that a random sample from the ERD suggested that fewer than 0.5% firms changed their CEOs, the potential connection variable is arguably not affected by the investment opportunities of firms30.

Figure 2 illustrates the significance of the land supply elasticity and the CEO’s potential connection with local land bureaus in predicting entry into the real estate sector. I divide the private manufacturing firms into four groups based on their connections with the local land bureau and whether their home city has an elastic land supply. I then plot the fraction of firms that entered the real estate sector in each group. Over time, the group of firms with local connections and inelastic city-level land supply is significantly more likely to participate in real estate development, as compared to the other three groups.

Figure 2:
Figure 2:

The Predictive Power of the Two Variables

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

In further examinations, the land supply elasticity and the CEO’s potential connection with local land bureaus also significantly predict local land supply. I discuss related evidence in Appendix B.

Identification. In the IV specification, I estimate the first stage as follows:

POSTit=αi+δt+θConnectioni×(1LandElasticityc)×Pt+κ1(1LandElasticityc)×Pt+μ1Connectioni×Pt+kηktXki+controlsit+it(3)

Connectioni is a dummy variable, which is equal to 1 if firm i’s executive in 1997 is from the same county as the head of local land bureau and is equal to 0 otherwise; 1 – LandElasticityc is the land supply inelasticity of city c, in which the headquarter of manufacturing firm i locates31; and Pt is the 3-year house price growth nationwide. Firm fixed effects, year fixed effects, group-year fixed effects, and other controls are also included in the first stage. I do not explore the time series variation of the CEO s connection variable, due to the low frequency of turnovers among local land bureau ministers32. Other controls are the same as in the OLS specification (2).

The instrumental variable is the three-way interaction of the potential connection between the CEO of firm i and the head of local land bureau (Connectioni), the local land supply inelasticity33 (1 – LandElasticityc), and the national house price growth (Pt). The inclusion of Pt is to generate time series variations34. The core of this instrumental variable is triple differences between firms within the same city and the same year. I control for the individual terms Connectioni, LandElasticityc, Pt, and all two-way interaction terms in both the first stage and the second stage. All non-time varying and firm-specific variables are absorbed by the firm fixed effects. (1 – LandElasticityc) × Pt is essentially a proxy for the house price growth in city c in year t. The coefficient θ is expected to be positive because entry is expected to be positively correlated with the CEO’s connection with local land bureau ministers, the inelasticity of land supply, and the growth in house prices.

The second stage is similar to the OLS specification in (2), except that the two-way interaction terms are controlled:

Yit=αi+δt+βPOSTit+κ2(1LandElasticityc)×Pt+μ2Connectioni×Pt+kηktXki+controlsit+it(4)

My identifying assumption for using this instrumental variable is: when the national house price is appreciating, managers who get in touch with the local land bureau more easily should not gain more advantage in the manufacturing sector when located in cities with ample land supply. While I will show several exclusion restriction tests in Appendix B, I now discuss major challenges to my identifying assumption. First, connected managers may face lower production costs in the manufacturing sector and may have easier access to industrial land. However, the unit price of industrial land did not increase in the past decade, and that connection matters for only residential and commercial land auctions (Cai et al. (2013)). Thus, I consider the managers connection with local land bureaus as having little impact on manufacturing production. Second, managers connections with the local land bureau might affect their likelihood of owning residential land, which in turn affects their overall credit constraints. This channel, however, indicates a positive correlation between the instrument and the omitted manufacturing investment opportunities. Therefore, it would not overturn my results, given that I am estimating an upper bound of the effect of entering the real estate sector. Finally, there are two additional concerns about the potential connection variable: first, managers with potential connections with the head of local land bureaus may also have connections to the city mayors; second, the variable could pick up managers and local land bureau ministers who were born locally. To address the first concern, I control for firm-specific political connections with city mayors and vice-mayors, also measured as whether or not they share the same home county. To address the second concern, I divide private firms in my sample into local firms and non-local firms, and added local-firm time fixed effects as controls.

Appendix B provides more discussions on the robustness, the exclusion restrictions, and the external validity of the instrumental variable approach.

Discussion of Results. Table 6 summarizes the empirical results in this subsection. The first stage of the baseline specification is reported in Column (1). For a manufacturing firm that has connections with the local land bureau, a 1% increase in instrumented local house price growth would lead to a 0.15% higher probability of entering the real estate sector.

Column (2) of Table 6 summarizes the OLS estimates using the sample of privately-owned firms that have existed since 1997. After entering the real estate sector, manufacturing firms on average reduce their investment expenditure and R&D expenditure by 4.2% and 0.56% of their assets, respectively; and labor productivity declines by 6.4%. The effects on R&D intensity, investment rates, and labor productivity are equivalent to 7%, 25.4% and 5.2% of their respective standard deviation.

Column (3) shows the results from the instrumental variable approach. The entry decision of manufacturing firms is instrumented by connections with the land officials interacted with local land supply elasticity and national house price growth. Entering the real estate sector results in a reduction of manufacturing R&D by 0.012 and investment by 0.15 out of 1 RMB of total assets. These estimates are equivalent to a 0.15-standard-deviation drop in R&D intensity and a 0.95-standard-deviation drop in the investment rate, almost three times as large as the OLS estimates, respectively. Given that investment and R&D expenditures are both sensitive to credit constraints, unobserved firm credit access could lead to an upward bias of the OLS estimates. Besides, entering the real estate sector could relax firm-level credit constraints, so the IV estimates can still be upward-biased here. Average labor productivity of manufacturing firms declines by 3.98% following entry, which is smaller compared to the OLS estimate. Labor productivity could be more relevant to the unobserved investment opportunity so that the full-sample OLS estimates are likely to be downward biased. Column (4) and column (5) adds controls of the current political connection with city mayors and vice-mayors and local-firm time fixed effects, respectively. The estimates are slightly larger than the ones using the baseline IV approach, but they are statistically indifferent.

Table 6:

The Effect of Entry into Real Estate on Manufacturing Production

article image
Notes: This table tests the effects of entering the real estate development industry on manufacturing production at the firm level. Columns (2) reports the estimates with ordinary least squares using the full sample of the balanced panel of private manufacturing firms. Column (3) reports the estimates using a three-way interaction instrument: local land inelasticity interacts with firm-specific potential connection with the local land bureau and the national house price growth. All two-way interactions are controlled in the regressions. The relevant first stage is reported in column (1). Column (4) and column (5) control for firm-specific connection with city mayor and vice mayors and local firm time fixed effects, respectively. All specifications control for age, local wage, year fixed effects, firm fixed effects, and firm-level initial conditions (exporter dummy, size classified according to employment, two-digit industry, economic region of location and credit rating) interacted with year fixed effects. T-stats are reported in parentheses with standard errors clustered at the firm level.*** Significant at the 1 percent level** Significant at the 5 percent level* Significant at the 10 percent level

Figure 3 provides the event study analyses on entering the real estate sector. The decline in labor productivity does not happen immediately after entering the real estate sector. This is likely due to the internal capital market selection, which cleanses the relatively unproductive projects within the manufacturing sector causing labor productivity to go up temporarily. No pre-trends are detected in either outcome variable of interest.

Figure 3:
Figure 3:

The Event Study Plots

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

4 Misallocation and Social Inefficiency: A Structural Approach

In the previous section, I presented evidence suggesting that firm credit access relates to the decision of moving to the real estate sector. The observed empirical patterns thus may deviate from the first-best scenario in which talent allocation only follows the comparative advantage of managers. To conclude social optimality, this section discusses a theoretical model and the structural estimation of the comparative advantage of managers. The theoretical framework follows the literature of comparative advantage estimation (Costinot et al., 2012; Adao, 2015; Hurst et al., 2018) and the literature studying financial frictions and misallocation (Buera, Kaboski, and Shin, 2012; Buera and Shin, 2013; Midrigan and Xu, 2014).

There are two parts in this section. In the first part, I model the optimization problem of private-manufacturing-firm managers and use firm-level data to structurally estimate their comparative advantage in the real estate sector. The structural estimation verifies the existence of talent misallocation in the Chinese real estate market. In the second part of the section, I move on to a general equilibrium setting by adding the optimization problem of workers, households, and the market clearing conditions. The general equilibrium model then allows us to assess the social efficiency of the allocation of talent. For tractability, I focus on a two-period model in this section. Section 5 provide quantitative results based on an infinite-horizon version of the general equilibrium model.

4.1 Structural Estimation of Comparative Advantage

4.1.1 Theoretical Framework

The theoretical model considers an economy with two sectors: the manufacturing sector and the real estate (hereafter housing) sector. Each manager i produces manufactured goods and housing with different talent levels: zM and zH. Labor is perfectly mobile across sectors. Manufactured goods are produced with capital (k) and labor (l); housing is produced with land (s) and labor (l).

Technology. Manufactured goods and housing are produced at constant returns to scale:

y(ziM,k,l)=[exp(ziM)k]αMl1αM,h(ziH,s,l)=[exp(ziH)s]αHl1αH,

where ziM,ziH>0 and αH, αM ∈ (0,1).35

The structure of comparative advantage. Another important feature is a flexible structure of comparative advantage. Following Hsieh et al. (2018), I divide managers into groups based on their experience and education36. At time t, the talents of manger i in group g is assumed to be:

zitM=z¯gM+ξtM+ωitzitH=z¯gH+ξtH+itz¯itH=c+bz¯gM(5)

z¯gM and z¯gH are constant average productivity of group g in the manufacturing sector and the housing sector; ξtM,ξtH are aggregate productivity shocks to the two sectors; and ωit, vit are idiosyncratic productivity shocks. The structural parameter b governs the correlation between the group-specific comparative advantage and absolute advantage. When b > 1, the groups with an absolute advantage in the manufacturing sector (high z¯gM) should specialize more in the real estate sector in the event of a real estate boom; when b < 1, a larger fraction of business owners with high z¯gM should stay in the manufacturing sector.

Other works studying financial imperfections and misallocation (Burea, Kaboski, and Shin, 2012; Midrigan and Xu, 2014) mostly assume that b equals 1. In such scenario, managers only differ in their idiosyncratic productivity shocks. There would be no misallocation of talent ex-ante, nor would we observe the strong correlation between manufacturing productivity and entry into the real estate sector as in section 3. In ex-post, adding the ex-ante heterogeneity in talents would lead to a negative (b < 1) or a positive (b > 1) correlation between wealth and housing comparative advantage37. As a result, financial imperfections yield a larger (b < 1) or a smaller (b > 1) welfare loss compared to the benchmark case with no ex-ante heterogeneity in talents.

Manager profit-maximizing problem. In each period t, manager i solves a profit-maximizing problem given initial wealth ait and prices: wage wt, house price pt, land price qt, and interest rate Rt. The price of manufactured goods is normalized to 1. Each manager faces a borrowing constraint, which is specified as a maximum leverage ratio of λ38. In the housing sector, there is a minimum operating scale, s_. The profit-maximizing problem of manager i can be written as:

maxsit,kit,litH,litMpth(zitH,sit,litH)+y(zitM,kit,litM)Rt(kit+qtsitait)wt(litH+litM)(6)
s.t.kit+qtsitλait(7)
sits_,sit>0.(8)

The key to modeling the borrowing constraint (7) is to allow the borrowing capacity to increase in wealth and non-decrease in talent, which ensures a positive correlation between wealth and talent (Table 1, Buera and Shin, 2013, Moll, 2014). The assumption can be micro-founded with limited enforcement of borrowing contracts among private managers and banks (Townsend, 1979; Holmstrom and Tirole, 1998). In other words, no one can delegate his or her funding entirely to a more efficient manager. Such financial imperfection implies that self-financing is the primary mechanism for privately-owned firms to accumulate capital, which is well rooted in the literature. Song et al. (2011) and Hsieh and Klenow (2009) show that Chinese private firms rely heavily on self-financing and receive only limited funding from banks and insignificant equity funding. By varying λ, I can trace out all degrees of depth in the financial market: λ = ∞ implies a perfect financial market, while λ = 1 indicates that no firms can finance from external sources. The exact form of the borrowing constraint is not important here as long as the maximum investment is increasing in firm net worth a and non-decreasing in manager talent z. I assume a constant λ across all groups for analytical convenience.

The constraint (8) is the minimum operating scale constraint in the housing sector. This assumption intends to model the inflexible land input and also plays the role of an entry barrier in the housing sector39. If there is no such a constraint, i.e., s_0, the borrowing constraint (7) only determines land and capital input at the intensive margin (the scale) but not at the extensive margin.

Given that the production functions are constant returns to scale, manager i specializes in either manufacturing or housing. The solution to the manager’s problem is:

kit=λait,litM=exp(zitM)(1αMwt)1/αMλait,sit=litH=0;ifait<qts_λ|zitMzitH+ctkit=litM=0,sit=λaitqit,litH=exp(zitH)[pt(1αH)wt]1/αλaitq;ifaitqts_λ&zitM<zitH+ct,

where ct=1αHαHlogpt(1αH)wt+logαH+logptqt1αMαMlog(1αMwt)logαM, which is a constant given market prices.

4.1.2 Structural Estimation

The solution of the manager’s profit-maximizing problem implies the following expressions for return on asset:

logROAitM=zitM+1αMαMlog(1αMwt)+logλαMlogROAitH=zitH+1αHαHlogpt(1αH)wt+logλαHptqt(9)

Using the correspondence between ROA and managerial talents, I estimate the structural parameter b which governs the relationship between group-average comparative advantage and absolute advantage of private business owners. Figure 440 plots the group-average return on asset in the two sectors against the estimated manufacturing talent. Clearly, well-performing manufacturing business owners do not necessarily generate a higher return to capital in the real estate sector.

Figure 4:
Figure 4:

Average Return to Capital in Two Sectors

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

Following the literature of comparative advantage estimation (Eaton and Kortum, 2002; Costinot et al. 2012; Adao, 2015; Hurst et al. 2016), I add two additional assumptions to ensure the identification of b. First, the group-average talent levels (z¯gM,z¯gH) do not vary over time. Second, idiosyncratic productivity shocks ωit and it are independent, and their distribution only depends on z¯gM41.

The firm-level dataset for this structural estimation combines the AIS, the ERD, and the 2004 and 2008 National Economic Census. I first use the ERD to match manufacturing firms in the AIS to their real estate subsidiaries. I then estimate the manufacturing production function and group average manufacturing talent from the AIS. The National Economic Census provides information on the return on asset of the real estate subsidiaries42. The 2004 and 2008 census covered more than 7,099,000 enterprises in China’s secondary and tertiary industries, containing information on production, management, investment, and corporate structure at the firm level. Table 1 provides a summary of the variables from the National Economic Census. 1,299 real estate firms are found to be subsidiaries of manufacturing firms in the AIS.

The ERD further provides managers’ levels of education and work experiences, which are predetermined characteristics used to classify these private firm managers. There are 20 groups based on five levels of education and four quartiles of managers’ work experience. The five education levels include graduate, undergraduate and post-high school, high school and vocational high school, middle school, and elementary school and below. The work experience of managers varies from 1 to 48 years.

Methodology. Re-writing the two equations in (9), we obtain the following equilibrium condition:

logROAitH=zitH+1αHαHlogpt(1αH)wt+logλαHptqt(10)=c+bz¯gM+ξtH+(1αH)log(1αHαHqtRtwt)+it=c+bz¯gM+δtH+it

In the first step, I estimate the group-average talent in the manufacturing sector, z¯gM, using data before 2002. Until the complete privatization in both property market and land market, privately-owned companies had neither the ability nor the incentive to develop real estate properties. Therefore, manufacturing business owners focused only on manufacturing production during the pre-boom era. I estimate the manufacturing talent of private business owners following Basu and Fernald (1997)43 as a residual from the production function:

logyit=αM(zitM+logkit)+(1αM)loglit1αIndiMlogyitlogkit1αIndiMαIndiMloglit=zitM=α0+αg+δt+eit,

where αIndiM is the average capital share of the 4-digit industry to which business owner i belongs to and α^g is an estimate of the group-specific manufacturing talent, z¯gM. The estimation is conducted with the unbalanced sample of firms from 1997 to 2002.

In the second step, I correct for the sample selection bias given that we only observe the ROA of firms that chose to enter the real estate sector. The error term it conditional on self-selection into housing (iHg,t44) is then centered around 0 if and only if b = 145. Therefore, to proxy for the sample selection bias, I estimate firm-specific propensity score for entering the real estate sector non-parametrically, following Das, Newey, and Vella (2003). The approach is similar to the Heckman two-step correction except that it does not require a normality assumption. Due to data limitations, the propensity score is estimated semi-parametrically using the 2004 census and the 2008 census:

Entryi,2008=γ0+γ1logwc,2004+γ2logpc,2008+γ3logqc,2004+F(z¯^gM,ai,2004,CreditScorei,2004)γ4PolConnectioni+γ5(1Elasticityc)+γ6PolConnectioni(1Elasticityc),(11)

where F(z¯^gM,ai,2004,CreditScorei,2004) is a fractional polynomial function of z¯^gM, firm i’s total asset value in 2004 – ai,2004, firm i’s credit score in 2004, and their interaction terms. PolConnectioni and Elasticityc are instruments that ensure the estimated propensity score as a valid proxy. w, p, q are the wage, house price, land price, respectively. The fitted value En^tryi,2008P^i is the estimated propensity score for entry.

In the last step, I plug z¯^gM in equation (10) to estimate the structural parameter b. The identifying assumption is that conditional on z¯^gM, the propensity score P^i, and other observables, the unobserved it has a mean of 046:

E(it|(it,ωit)Hg,t,wt,pt,qt,ait,z¯^gM,P^i)=0

The selection bias in equation (10) is proxied using a non-parametric function of P^i:

logp2008Hhiki=αc+bz¯^gM+Λ(P^i)+i,2008,

where αc is city fixed effect that controls for local prices in city c;Λ(P^i) is a fractional polynomial function of the estimated propensity score P^i, which works as a proxy for the selection bias E(it|(it,ωit) ∈ Hg,t). Under the identifying assumption, b^ is a consistent estimator of b47. To find out if more productive manufacturing managers should have stayed in the manufacturing sector, I test the null hypothesis H0 : b ≥ 1 with a sample bootstrapped at the manager level. In the case that z¯^gM’s are estimated with errors, the third-step estimate b will have a bias towards 0. I include in estimate with the Empirical Bayes approach (the control function approach) to control for the estimation error. Appendix B discusses the details of the empirical Bayes approach.

Discussion of Results.

Table 7 summarizes the estimation results. Columns (1) to (3) report the coefficients estimated using firm-specific talent, group-specific talent, and the group-level talent adjusted with the Empirical Bayes approach, respectively. The coefficients in columns (1) and (2) are subject to attenuation bias due to measurement errors in the first step. The null hypothesis of b ≥ 1 is rejected in all three estimations, indicating that there is likely a negative correlation between the managers’ comparative advantage in real estate and their absolute advantage in the manufacturing sector. Therefore, the fact that more productive manufacturing managers are more likely to move to the real estate sector implies a misallocation of talent.

Table 7:

Quantitative Evaluation - Comparative Advantage on Absolute Advantage

article image
Notes: This table reports the structural estimators of b and the T-test statistics for the sufficient and necessary condition in proposition 2. Columns (1) to (3) report the coefficients estimated using the three-step approach. The entrepreneurial talent in the manufacturing sector is estimated at firm-level, group-level, and by applying an empirical Bayes adjustment at the group level, respectively.

4.2 The Social Efficiency

From the analyses above, there exists a misallocation of talent when b < 1. Managers with a comparative advantage in the real estate sector were forced to stay outside due to financial imperfections. Given that the actual size of the entry barrier varies with land prices, the minimum-operating-scale constraint creates a pecuniary externality, which leads the competitive equilibrium to deviate from the second-best scenario. Next, I illustrate the social inefficiency in the scenario of b < 1 using a two-period model, which emphasizes the inefficient talent allocation at the cross-section. Section 5 generalizes the two-period model into over-lapping generations and provides quantitative estimates for the social inefficiency.

4.2.1 The Two-period Model

I model an economy with L workers and private business owners of a measure of 1. The private business owners have heterogeneous talents, zM ∈ [0,1] and zH = c + bzM (b < 1), in producing manufactured goods and housing respectively. There are five private markets: the labor market, the capital market, the land market, the goods market, and the housing market. The first three markets are used to allocate production factors to business owners; business owners then sell their products through the goods market and the housing market. Without loss of generality, I fix the price of manufactured goods at 1.

Preference. All agents (private business owners and workers) in the economy consume only manufactured goods in period 1, and they consume both manufactured goods and housing in period 2. Given that the housing sector privatization is unexpected, a representative agent maximizes her utility in the first period without taking into account the housing sector. The period-1 utility criterion is thus:

U1=logc1+βlogc2,

where c1 is the consumption in the first period and c2 is the expected consumption in the second period.

In the second period, after the housing sector is privatized, workers and private business owners maximize their total utility from consuming housing and manufactured goods:

U2=σlogc2+(1σ)logh,

where h is consumption of housing and 1 – σ is the housing consumption share in the second period. The elasticity of substitution between manufacturing goods and housing is assumed to be 1.

First period. In the first period, all private business owners only operate in the manufacturing sector, which reflects the years before the real estate market privatization48. The private business owners choose optimal consumption and production input as in (6) except that land s and housing labor input lH are fixed at 0. For simplicity, I further assume that all firms have the same wealth a0 at the beginning of period 1. This assumption then allows the model to emphasize the importance of manufacturing talent on wealth accumulation49.

Lemma 1:

The private business owner’s wealth at the end of the first period, a1 (zM), is non-decreasing in zM.

(Proofs and more details of the optimization problem are provided in Appendix C). Lemma 1 is a direct result of the self-financing mechanism50, given the initial wealth, private business owners with a higher talent in the manufacturing sector accumulate capital at a faster speed.

Second period. In the second period, the economy experiences unexpected privatization in the housing sector, and manufacturing business owners face the option of becoming a real estate developer51. The total land supply S is determined exogenously by the geographical constraint of land development. Denoting pH and q as the house price and land price in period 2, the profit-maximizing problem of the private business owner with talent zM becomes:

maxk(zM),s(zM),lM(zM),lH(zM)[exp(zM)k(zM)]αMlM(zM)1αM+pH[exp(zH)s(zM)]αHlH(zM)1αw2[lM(zM)+lH(zM)]R2[k(zM)+qs(zM)a1(zM)]s.t.k(zM)+qs(zM)λa1(zM)s(zM)s_s(zM)>0,

where k(zM) and lM(zM) refer to the capital and labor input of zM in the manufacturing sector; and s(zM) and lH(zM) refer to the land and labor input of zM in the housing sector.

Appendix C provides a full description of the rest of the competitive equilibrium, including the worker’s problem, the private business owner’s utility maximization problem, and other equilibrium conditions.

Proposition 1:

There exists a unique competitive equilibrium in the economy. When the initial endowment of the representative worker is large enough, the allocation ofmanagerial talent in the second period can be characterized using three cutoffs: z_M,zlM and zhM. Private business owners with talent zM(z_M,zlM](zhM,1] operate in the manufacturing sector; private business owners with talent zM(zlM,zhM] operate in the housing sector; and private business owners with talent zM(0,z_M] exit production and only save in the capital market.

(Proofs are provided in Appendix C).

The planner’s problem. The second-best benchmark in this economy is the social planner’s optimal solution under the same production frictions: the firm-level borrowing constraint and the minimum operating scale constraint in the housing sector. With these constraints, the social planner cannot arbitrarily assign production factors to business owners.

In the first period, the social planner’s solution aligns with the competitive equilibrium, given that the housing market privatization is unanticipated. This result is implied directly by the second welfare theorem. In the second period, the social planner maximizes aggregate utility following the Gorman’s Aggregation Theorem:

U=χlogC+logH,

where C is the total consumption of manufactured goods, and H is the total housing consumption. χ=σ1σ is the Pareto weight that corresponds to the competitive equilibrium.

The crucial difference between the social planner’s solution and the competitive equilibrium is that the social planner internalizes the land price as a function of the allocation of private business owners. There are also private markets for land, capital, labor, and housing. Solving the social planner’s problem is thus equivalent to solving for the three cutoffs described in Proposition 1: one for producing, z_M, and two for operating in the housing sector: zlM,zhM. The social planner’s problem in the second period can be written as:

max{z_M,zlM,zhM,l(zM)}χlogC+logHs.t.C=[z_M,zlM](zhM,1][exp(zM)k(zM)]αMl(zM)1αMdF(zM)(12)
H=zlMzhM[exp(zH)s(zM)]αHl(zM)1αHdF(zM)(13)
λ([z_M,zlM)(zhM,1]a1(zM)dF(zM))=01a1(zM)dF(zM)+a1L(μk)(14)
λa1(zlM)λzlMzhMa1(zM)dF(zM)Ss_(μe)(15)
z_M1l(zM)dF(zM)=L(μl)(16)

Similar to the first case, equations (12), (13), (14) and (16) refer to the market clearing conditions on the manufactured goods market, the housing market, the capital market, and the labor market, respectively. The constraint (15) combines the market clearing condition on the land market and the minimum operating scale constraint.

Proposition 2:

When the least talented manufacturer z_M is constrained from entering the housing sector, the competitive equilibrium is constrained inefficient. Both zlM and zhM are higher in the competitive equilibrium than in the social planner’s problem.

The socially optimal allocation of managerial talent should be solely pinned down by the comparative advantage of private business owners in the housing and the manufacturing sectors. The comparative advantage of housing production is then the marginal social return on capital in the housing sector relative to the marginal return in the manufacturing sector. With a monotonic transformation, it is equivalent to zHzM in this model. In the case of b < 1, the comparative advantage is decreasing in zM, so that the private business owners with low zM’s should operate in housing production in the optimal situation. Figure 5 compares the competitive equilibrium with the social planner’s solution in such scenario. The thresholds labeled with “CE” refer to the competitive equilibrium, and the thresholds labeled with “SP” refer to the planner’s solution (the second-best scenario). The competitive equilibrium inefficiently allocates two groups of private business owners. First, the entry barrier results in a lack of competition in the land market, which extracts monopolistic rents to the landowners. The monopolistic rents attract business owners with talent in (zh,SPM,zh,CEM] to enter the housing sector. Given that the more talented manufacturing business owners also own more wealth, the land price is higher in the competitive equilibrium than in the planner’s solution. Therefore, a larger fraction of private business owners who have a higher comparative advantage in housing is constrained from entry. This group has manufacturing talent in (zl,SPM,zl,CEM]. Endogenizing the land prices allows the social planner to allocate the private business owners efficiently at the extensive margin.

Figure 5:
Figure 5:

The Allocation of Managerial Talent (b < 1)

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

4.2.2 Discussion of Policy Tools

To deal with the social inefficiency, I examine three policy tools that have the potential to improve social welfare: liberalizing the financial market, a more flexible schedule for land supply (or a reduced entry barrier), and taxation on the return on capital in the real estate sector.

The first two policy tools relax the borrowing constraint and the minimum operating scale constraint, respectively. These changes not only affect the competitive equilibrium but also relax the constraints in the planner’s problem. Financial market liberalization and increasing land supply might improve welfare but would not help with alleviating inefficiency. The taxation on returns in the real estate sector, on the other hand, directly reduces the excess return due to the lack of entry, so that it partly solves for the inefficiency in the managerial talent allocation.

In practice, financial market liberalization could affect both firm borrowing counstraint and the entry barrier in the real estate sector. For example, real estate developers can raise funds through REITs and thus their individual wealth is less crucial to entry into the land market. In terms of taxing the returns on real estate businesses, China is now imposing a 1% stamp tax and considering further implementing a recurrent property tax. These tax policies would help correct the incentives of real estate developers, and also prevent overheating from the demand side of the housing market.

Financial market liberalization:

A more liberalized financial market (a higher λ) increases z_M and lowers zlM and zhM, resulting in higher social welfare but larger inefficiency at the extensive margin.

In theory, financial market liberalization can be modeled as an increase in the maximum leverage ratio λ. A higher λ is welfare-improving from two aspects. First, it directly reduces the resource misallocation in the economy, given that more productive private business owners are allowed to borrow more. Second, the constrained manufacturing business owners who have the comparative advantage in the housing sector are more capable of acquiring capital, so that they are more likely to overcome the entry barrier in the housing sector. Given existing constraints in the real estate sector, an economy with a more developed financial market would achieve higher social welfare as compared to a developing economy.

This finding is consistent with the literature, which argues that a more liberalized financial market improves resource misallocation. However, the inefficiency (the discrepancy between the competitive equilibrium and the second-best scenario) at the extensive margin becomes exacerbated. The already wealthy business owners are now able to borrow more, which further increases the land prices. As a result, more manufacturing business owners with low wealth (and therefore comparative advantages in the housing sector) are constrained by the entry barrier in the housing sector. A detailed proof is provided in Appendix C.

A lower cost barrier to entry:

A smaller entry barrier in the real estate sector (a smaller s_S) lowers z_M,zlM,zhM, and lowers inefficiency at the extensive margin. The welfare consequence is ambiguous.

An increase in land supply is equivalent to a decline in the minimum operating scale, which lowers s_S52. All else equal, all private business owners are less constrained by the entry barrier in the housing sector. However, lowering the entry barrier in the housing sector has an ambiguous impact on aggregate welfare. When the minimum operating scale is lower, more firms decide to enter the housing sector. Unproductive manufacturing business owners, who might have become savers when the entry barrier is high, now produce actively. The aggregate productivity is thus lower with a lower entry barrier in the housing sector.

Taxation on land use:

Taxing the land use lowers the return on capital in the real estate sector, and thus lowers z_M,zlM and zhM resulting in higher social welfare and a lower inefficiency at the extensive margin.

This policy tool is equivalent to imposing a wedge on the returns from building and selling houses. Instead of paying a unit price of q for land acquisition, the real estate developers now face cost q(1 + τ). The land tax lowers the return from producing housing so that the more talented business owners in the manufacturing sector lose the incentive to move to the housing sector. Due to the lack of entry for wealthy business owners, the land price is also lower, which alleviates the pecuniary externality. In summary, this policy tool deals with both groups of private business owners who were inefficiently allocated in the competitive equilibrium by correcting their incentives.

5 Quantitative Analysis

In the quantitative analysis, I generalize the two-period model to the infinite horizon, and quantify the productivity loss due to the liberalization of the real estate market. The quantitative exercise generalizes the model in Song et al. (2011) to multiple groups of entrepreneurs with heterogeneous talent and two sectors: the manufacturing sector and the housing sector.

In the infinite horizon model, the economy consists of L workers and N groups of entrepreneurs. Each agent in the economy works for 30 years and saves for 20 years after retiring53. Entrepreneurs work as managers when young; after retirement, they invest either in the capital market or in their family firms. In the latter case, old entrepreneurs hire other young entrepreneurs to manage their family firms. Having entrepreneurs saving in their family firms guarantees that the self-financing mechanism will continue to exist. Therefore, the positive correlation between wealth and talent holds in this infinite horizon model. While young, both workers and entrepreneurs can only save in the capital market. Young workers earn labor compensation, and young entrepreneurs earn manager compensation as a fraction of old entrepreneur’s investment return. In each period, each old entrepreneur (business owner) makes decisions on the capital investment into the family firm, while the young entrepreneur (manager) decides the allocation of capital. Each family firm still faces the borrowing constraint and the minimum operating scale constraint explored in section 4. The technology and preferences are the same as in the two-period model.

The question at the core of this quantitative analysis is to jointly identify the comparative advantage parameter and the entry barrier in the housing sector. In the literature, people have estimated Roy-like models with a flexible comparative advantage structure with no other production frictions, or models with a fixed cost of entry but independent productivity draws. They use the observed selection into different sectors to pin down either the talent distribution or the entry barrier. In this model, given that both b and s_ are unknown, the observed selection does not provide enough degrees of freedom to identify the two parameters together. Therefore, I take the parameter b estimated in section 4 using the micro-level data, and I calibrate the entry barrier s and other parameters of the model by matching empirical moments in the years between 1997 to 2010. Based on the calibration, I conduct policy counterfactual analyses that focus on the three policy tools discussed in section 4.

5.1 Calibration and Counterfactual Policy Analyses

To calibrate the model, I match the equilibrium outcome of the life-cycle model with overlapping generations with data. Given that a series of deregulation led to the real estate boom in China, I model them as permanent shocks. The Chinese economy was transitioning to a balanced growth path for the sample period. Therefore, the value function of each agent depends on the entire path of prices, including wage, interest rate, house price, and land price. The set of state variables contain talent (zitM,zitH), wealth, age, and year. To simplify the computation process, I calibrate the entrepreneurs’ decisions at the group level and assume that both it and ωit are zero. In doing so, I replace the state variable (zitM,zitH) with exogenously fixed average productivity (z¯gM,z¯gH). The talent distribution of group-specific talent, z¯gM, follows a truncated Zeta distribution, which is a discrete approximation to the Pareto distribution54. The probability density function of z¯gM is:

Pr(z¯gM)=1/gζi=1N1/iζ,ζ>1

The old entrepreneurs’ savings decisions determine the total asset value available for each firm. I assume a competitive market for hiring young entrepreneurs as managers. Within the manufacturing sector, the group with a higher z¯gM generates a larger return on capital, and the manager compensation is increasing in g. The expected wage of each group mg,t is determined such that the incentive compatibility constraint for each young entrepreneur is binding. The parameter determining the manager’s compensation is assumed to be ψ, which is the minimum fraction of capital return that managers can steal. The size of ψ determines the overall growth rate of the economy. When ψ is higher, the income dispersion is more significant in younger generations; thus, the group of productive firms in manufacturing accumulates wealth at a faster speed.

To more closely model the Chinese economy, I add the SOEs in the quantitative analysis. Following Song et al. (2011), I assume that the SOEs are neither financially constrained nor managed by any entrepreneurs. Their role in the model is to clear the capital market at the beginning of the transition when entrepreneurs have not accumulated enough wealth. The productivity of SOEs is assumed to be a κ fraction of the average productivity of privately-owned firms.

There are two stages in the calibration. The first stage, the pre-boom stage, corresponds to the first period in section 4. The second stage models liberalization in the real estate market. I consider the real estate boom as driven by a permanent preference shock, such that agents increase the housing consumption share from 0 to ρ. I estimate parameters on manufacturing talent distribution, manager compensation, and initial wealth using in the first stage and parameters that determine average housing talent and entry barrier in the second stage. The rest of the parameters are set exogenously. Table 8 describes the exogenously determined parameters.

Table 8:

Exogenously Determined Parameters

article image

I assume that the first stage starts from 199255 and lasts for ten years until the 2002 land market liberalization. I calibrate the talent distribution using the estimated firm-level productivity between 1997 and 2002. Given the shape parameter ζ=2, the distance parameter is estimated to be dM=0.15. The SOE productivity is set as κ=0.397 to match the relative capital-to-output ratio of SOEs and private companies, 2.65. Last but not least, the initial wealth level of workers and entrepreneurs is calibrated such that the average share of total fixed assets of private companies during 1997-2002 is close to 31.75% as observed in the data. The initial wealth distribution over each agent’s life cycle is similar to a scaled-up version of their life-cycle wealth distribution in the balanced growth path. Entrepreneurs with different talent groups are assumed to start with the same wealth level. The minimum fraction of capital return a manager can steal (ψ) is estimated as 0.2, which matches the growth rate of the average share of total fixed assets owned by private companies. The model-based share in 1992 is 17.36%, which is close to 15.12% in the data. 55The beginning of the first stage follows the assumption in Song et al. (2011).

In the second stage, I take the model-based wealth distribution in 2002 as the initial wealth and calibrate three parameters: s_,c, and housing consumption share ρ. I calibrate the housing consumption share ρ=0.4. Assuming that an average worker works for 30 years, this yields a price-to-income ratio of 10.41, which matches the average price-to-income ratio of 10.2 in the data. Based on the proofs in Appendix C, the entry decisions of entrepreneurs depend jointly on the entry barrier and the comparative advantage of these entrepreneurs. I calibrate the entry barrier using the average difference in log productivity between entrepreneurs who entered the real estate sector and the ones who did not. In the data, the average difference is estimated to be 0.25, which leaves me with an estimate of s_=0.046. I interpret the estimate as follows: in an average city in China, the average startup cost for a real estate project is 4.6% of the total value of land supplied in the city. The parameter c indicates the average log productivity in the housing sector. I estimate c to match the relative firm size in the manufacturing sector and the real estate sector. In the data, an average manufacturing firm hires 7.54 times more employees than an average real estate development firm. The average housing talent c is then estimated to be 0.31. Table 8 summarizes the calibrated parameters and relevant moments in the data.

Table 9:

Calibrated Parameters

article image

Panels A-C of Figure 9 compare several macroeconomic outcomes of the model and data. The most crucial untargeted moment is the relative growth rate of land prices to house prices. At the beginning of the real estate boom, the positive housing demand shock induces a high return on capital in the real estate sector. The unproductive and poor manufacturing entrepreneurs, who have the comparative advantage in the housing sector, are initially constrained from participating in the land market. The initial land price must be low when compared to the house price, so that talented manufacturing entrepreneurs have the incentive to move to the real estate sector. Over time, more entrepreneurs save out of the entry barrier which makes the land market more competitive. Real estate entrepreneurs with the highest manufacturing talents then move back to the manufacturing sector. The increasing competitiveness in the land market results in the faster growth of land prices relative to house prices. I find a similar pattern in the data. In the first three years following the land market liberalization, the land price in China increased by more than 20% per year compared to the increase in house price. After 2009, however, the annual growth rate of land price has remained mostly the same as the annual growth rate of the house price.

Panel B compares the average scale of private firms in the manufacturing sector relative to the real estate sector. Given that land prices increase faster than house prices (and thus total output), more unproductive and poor manufacturing entrepreneurs are constrained from entering the real estate sector. The constraint leads to a larger average scale of real estate firms and a smaller average scale of manufacturing firms. Given data limitations, I am only able to recover the relative scale from the three rounds of National Economic Census in 2004, 2008, and 2013. There is a rough pattern indicating that an average manufacturing firm is becoming smaller compared to an average real estate firm.

Panel C compares the manufacturing TFP growth with and without a real estate boom. An economy experiencing a real estate boom has a smaller manufacturing TFP growth. The land market liberalization changes the composition of entrepreneurs in the manufacturing sector, imposing an adverse effect on the productivity growth in the manufacturing sector.

Figure 6:
Figure 6:

Calibration Results

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

Policy Counterfactual Analyses

In analyzing the effects of the policy tools, I conduct three counterfactual analyses. The manufacturing TFP growth and land price in the model and the counterfactual cases are presented in Figure 7.

I first increase the maximum leverage ratio, λ, from 1.35 to 2 to understand the effect of liberalizing the financial market. A 48% larger borrowing constraint leads to a higher manufacturing TFP growth, as shown in panel A of Figure 7. This policy works indirectly by reallocating more resources to entrepreneurs with more talent in manufacturing, as is also discussed in related works in the literature (Hsieh and Klenow, 2009; Buera and Shin, 2013). Given the composition of entrepreneurs in the manufacturing sector, the productivity growth is higher as productive entrepreneurs accumulate wealth at a faster speed. A more relaxed borrowing constraint, however, also increases land prices so more entrepreneurs are constrained from entering the real estate sector. The larger inefficiency in the allocation of entrepreneurial talent then dampens the effect of liberalizing t he financial market.

Second, I study the effect of a smaller entry barrier. In reality, a lower entry barrier can be achieved by increasing the total land supply or improving corporate laws so several entrepreneurs can jointly operate a real estate firm without additional principle-agent frictions. As discussed in section 4, the welfare consequences are ambiguous. Given prices, a lower entry barrier allows more unproductive entrepreneurs to enter the land market. However, more entries also increase the land price such that a larger fraction of entrepreneurs is needed to clear the land market. As shown in panel B of Figure 7, the land price with a lower entry barrier is strictly higher than the one in the calibrated model. This imposes an adverse effect on manufacturing productivity. Overall, the impact of reducing the entry barrier on manufacturing TFP growth is ambiguous.

My last policy counterfactual focuses on the effect of taxing the return from property sales. In many countries, it is implemented as tax on land use or property stamp tax. When real estate developers sell houses, the buyer or the developers pay a 5% - 15% one-time stamp tax. This policy lowers the ROA from operating in the real estate sector without affecting land prices. Therefore, it prevents productive manufacturing entrepreneurs from operating in the real estate sector. Panel A in Figure 7 shows that a 3% property stamp tax increases manufacturing TFP growth in most years. The land price under this regulation is also always lower than the price in my calibrated model. Without changing the resource constraints, it improves social welfare by lowering land prices.

Regarding the welfare consequences, increasing λ from 1.35 to 2 results in a welfare increase of 6%, while a 3% property stamp tax increases the social welfare by 0.5%. The welfare is higher by less than 0.1% when the entry barrier is reduced from 0.046 to 0.016. Liberalizing the financial market is often costly and requires improving other related regulations. Collecting a property stamp tax, however, can be easily implemented as all properties on sale are already listed publicly. Policy makers then need to take into account the welfare benefits of the policy tools and their costs.

Figure 7:
Figure 7:

Policy Counterfactuals

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

5.2 Limitations and Future Works

The quantitative evaluation in this section is mainly used to evaluate the efficiency of entrepreneurial talent allocation in China and to conduct policy analyses. The calibrated life-cycle model with overlapping generations matches the price dynamics on the land market, as well as the average scale in the manufacturing sector relative to the real estate sector. There are several aspects of this quantitative model that can be improved in future works.

First, a larger efficiency loss could occur if idiosyncratic productivity shocks are added back to the model. In this case, the wealth of entrepreneurs depends not only on their talent group but also on their productivity shocks throughout history. On the other hand, their comparative advantage only depends on the talent group and current productivity shocks. Therefore, adding back the idiosyncratic productivity shocks creates a larger gap between the incentive to entry (the comparative advantage) and the ability to entry (wealth). I am working on estimating a model that includes within-group mean-reverting productivity shocks for additional policy evaluations.

Second, the size and nature of the real estate boom also matter for my inefficiency argument. Imagine that China did not experience a large real estate boom following the deregulation. The talented manufacturing entrepreneurs would not be incentivized to enter the real estate sector. The land price would also remain at a relatively low level, such that unproductive manufacturing entrepreneurs with the comparative advantage in real estate are not constrained from entry into the land market. However, if a real estate bubble exists in addition to the increasing demand for housing, the entry barrier in real estate creates a larger distortion on the allocation of entrepreneurial talent. Moreover, the growing investment in bubble assets yields an additional loss in social welfare, as these investments produce neither more housing nor more manufactured goods.

Finally, I do not model the occupation choice of workers, so that the impact of real estate prices on entrepreneurship via the collateral channel is not considered here. Several empirical works (Hurst and Lusardi, 2004; Schmalz, Sraer and Thesmar, 2015; Kerr, Kerr, and Nanda, 2015) document that rising real estate prices may help alleviate the credit constraints of potential entrepreneurs. While my focus is on the business choices of existing entrepreneurs, I also argue that this is unlikely to matter for the Chinese real estate boom. Prior to the housing market reform, most households in China were not homeowners. Therefore, the increasing house prices do not help households to start up new businesses. The effect of the collateral channel also depends on the correlation between home-ownership and workers’ comparative advantage in operating a firm. If the two factors are independent, the collateral channel should not have a sizable impact on welfare at the aggregate level.

6 Concluding Remarks

In this paper, I have proposed a new channel that links sectoral booms and aggregate inefficiency from the decisions of private business owners. I focus on the recent Chinese real estate boom and show that an inefficient allocation of managerial talent could exist in an economy with ex-ante heterogeneous talent, an imperfect financial market, and a cost barrier to entry in the real estate sector. The model is consistent with new empirical data on Chinese firms; more productive manufacturing firms diverted their resources to the real estate sector, resulting in a decline in R&D, investment, and productivity growth in their manufacturing businesses. The positive correlation between manufacturing productivity and entering the real estate sector can be adequately explained by more productive manufacturing businesses having more wealth. My data also suggests that more productive manufacturing business owners do not have the comparative advantage in the real estate sector. Therefore, the imperfect financial market, together with the entry barrier in real estate, creates an inefficient allocation of managerial talent, resulting in an inefficient loss of productivity.

The new channel highlighted in my paper adds new insights on the role of real estate booms. I argue that productivity loss may exist even without a real estate bubble, which differs from related works (Charles, Hurst, and Notowidigdo, 2015; Chen and Wen, 2014; Mian, Sufi, and Rao, 2013). In addition, the key feature of the real estate sector is that production of properties uses a scarce and inflexible factor - land. As an economy grows, the supply of capital is increasing while the supply of land is fixed. Eventually, the entry barrier only exists in the real estate sector. The real estate sector, however, is not the only sector that produces with scarce factors. The model can be used to study other booming sectors, as long as they also require both a costly barrier to entry and that existing business owners reallocate from other sectors to the booming sectors. For instance, my model can help examine the natural resource booms (the “Dutch disease”) observed when countries with abundant natural resources experience a significant productivity slowdown following a positive shock to commodity prices.

I have also discussed three policy tools that may improve social welfare in real estate booms: relaxing firm borrowing constraint, reducing the entry barrier in real estate, and taxation on real estate returns. These policy tools are specific to alleviating the misallocation of manageral talent. An ideal policy should provide subsidies to managers constrained by the entry barrier in real estate and tax the ones with no comparative advantage in the real estate sector. Taxation on real estate returns appears to play a similar role to the ideal policy without the requirement in identifying the talent of individual managers. Relaxing firm borrowing constraint is also welfare improving, but it changes the planner’s problem without directly targeting at the inefficiency. In fact, it would result in a large efficiency loss. Reducing the entry barrier in the real estate sector yields an ambiguous welfare consequence. A calibrated version of the model has been provided to evaluate the manufaturing TFP gains of the three policy tools.

References

  • [1]

    Alberto Abadie,. Semiparametric difference-in-differences estimators. The Review of Economic Studies, 72(1):119, 2005.

  • [2]

    Alberto Abadie and , Guido W Imbens,. Large sample properties of matching estimators for average treatment effects. Econometrica, 74(1):235267, 2006.

    • Search Google Scholar
    • Export Citation
  • [3]

    Alberto Abadie and , Guido W Imbens,. Bias-corrected matching estimators for average treatment effects. Journal of Business & Economic Statistics, 2012.

    • Search Google Scholar
    • Export Citation
  • [4]

    Daron Acemoglu, Suresh Naidu, Pascual Restrepo, and James A Robinson,. Democracy does cause growth. Technical report, National Bureau of Economic Research, 2014.

    • Search Google Scholar
    • Export Citation
  • [5]

    Rodrigo Adão,. Worker heterogeneity, wage inequality, and international trade: Theory and evidence from brazil. Technical report, Mimeo, 2015.

    • Search Google Scholar
    • Export Citation
  • [6]

    Franklin Allen, Jun Qian, and Meijun Qian,. Law, finance, and economic growth in china. Journal of financial economics, 77(1):57116, 2005.

    • Search Google Scholar
    • Export Citation
  • [7]

    Heitor Almeida and , Daniel Wolfenzon,. The effect of external finance on the equilibrium allocation of capital. Journal of Financial Economics, 75(1):133164, 2005.

    • Search Google Scholar
    • Export Citation
  • [8]

    Diego Anzoategui, Diego Comin, Mark Gertler, and Joseba Martinez,. Endogenous technology adoption and r&d as sources of business cycle persistence. Technical report, National Bureau of Economic Research, 2016.

    • Search Google Scholar
    • Export Citation
  • [9]

    Abhijit V Banerjee and , Andrew F Newman,. Occupational choice and the process of development. Journal ofpolitical economy, pages 274298, 1993.

    • Search Google Scholar
    • Export Citation
  • [10]

    Susanto Basu and , John G Fernald,. Returns to scale in us production: Estimates and implications. Journal of political economy, 105(2):249283,1997.

    • Search Google Scholar
    • Export Citation
  • [11]

    Shai Bernstein,. Does going public affect innovation? The Journal of Finance, 70(4):13651403, 2015.

  • [12]

    Francisco J Buera, Joseph P Kaboski, and Yongseok Shin,. Finance and development: A tale of two sectors. The American Economic Review, 101(5):19642002, 2011.

    • Search Google Scholar
    • Export Citation
  • [13]

    Francisco J Buera and , Yongseok Shin,. Financial frictions and the persistence of history: A quantitative exploration. Journal of Political Economy, 121(2):221272, 2013.

    • Search Google Scholar
    • Export Citation
  • [14]

    Hongbin Cai, J Vernon Henderson, and Qinghua Zhang,. China’s land market auctions: evidence of corruption? The Rand journal of economics, 44(3):488521, 2013.

    • Search Google Scholar
    • Export Citation
  • [15]

    Thomas Chaney, David Sraer, and David Thesmar,. The collateral channel: How real estate shocks affect corporate investment. The American Economic Review, 102(6):23812409, 2012.

    • Search Google Scholar
    • Export Citation
  • [16]

    Kerwin Kofi Charles, Erik Hurst, and Matthew J Notowidigdo,. Housing booms and busts, labor market opportunities, and college attendance. Technical report, National Bureau of Economic Research, 2015.

    • Search Google Scholar
    • Export Citation
  • [17]

    Kaiji Chen and , Yi Wen,. The great housing boom of china. FRB St Louis Paper No. FEDLWP2014-022, 2014.

  • [18]

    Ting Chen, Laura Xiaolei Liu, and Li-An Zhou,. The crowding-out effects of real estate shocks-evidence from china. Available at SSRN 2584302, 2015.

    • Search Google Scholar
    • Export Citation
  • [19]

    RajChetty, John N Friedman, and Jonah E Rockoff,. Measuring the impacts of teachers i: Evaluating bias in teacher value-added estimates. The American Economic Review, 104(9):25932632, 2014.

    • Search Google Scholar
    • Export Citation
  • [20]

    Stefano Corradin and , Alexander Popov,. House prices, home equity borrowing, and en-trepreneurship. Review ofFinancial Studies, page hhv020, 2015.

    • Search Google Scholar
    • Export Citation
  • [21]

    Arnaud Costinot,. An elementary theory of comparative advantage. Technical report, National Bureau of Economic Research, 2009.

  • [22]

    Wei Cui,. Delayed capital reallocation. 2014.

  • [23]

    Mitali Das, Whitney K Newey, and Francis Vella,. Nonparametric estimation of sample selection models. The Review of Economic Studies, 70(1):3358, 2003.

    • Search Google Scholar
    • Export Citation
  • [24]

    Thomas Davidoff,. Supply constraints are not valid instrumental variables for home prices because they are correlated with many demand factors. Available at SSRN 2400833, 2015.

    • Search Google Scholar
    • Export Citation
  • [25]

    Thomas Davidoff, et al. Supply constraints are not valid instrumental variables for home prices because they are correlated with many demand factors. Critical Finance Review, 5(2):xxxxxx, 2016.

    • Search Google Scholar
    • Export Citation
  • [26]

    Chengri Ding and , Gerrit Knaap,. Urban land policy reform in chinaâs transitional economy. Emerging land and housing markets in China, 2:32,2005.

    • Search Google Scholar
    • Export Citation
  • [27]

    David Dollar and , Shang-Jin Wei,. Das (wasted) kapital: firm ownership and investment efficiency in china, 2007.

  • [28]

    Per-Anders Edin, Peter Fredriksson, and Olof Aslund,. Ethnic enclaves and the economic success of immigrantsâevidence from a natural experiment. The quarterly journal of economics, 118(1):329357, 2003.

    • Search Google Scholar
    • Export Citation
  • [29]

    Andrea L Eisfeldt and , Adriano A Rampini,. Capital reallocation and liquidity. Journal of monetary Economics, 53(3):369399, 2006.

  • [30]

    Andrea L Eisfeldt and , Adriano A Rampini,. Managerial incentives, capital reallocation, and the business cycle. Journal of Financial Economics, 87(1):177199, 2008.

    • Search Google Scholar
    • Export Citation
  • [31]

    Bai Chong En, Chang Tai Hsieh, and Yingyi Qian,. the return to capi-tal in china. NBER working paper, 12755, 2006.

  • [32]

    Hanming Fang, Quanlin Gu, Wei Xiong, and Li-An Zhou,. Demystifying the chinese housing boom. In NBER Macroeconomics Annual 2015, Volume 30. University of Chicago Press, 2015.

    • Search Google Scholar
    • Export Citation
  • [33]

    Hanming Fang, Quanlin Gu, Wei Xiong, and Li-An Zhou,. Demystifying the chinese housing boom. NBER macroeconomics annual, 30(1):105166, 2016.

    • Search Google Scholar
    • Export Citation
  • [34]

    Jie Gan,. Collateral, debt capacity, and corporate investment: Evidence from a natural experiment. Journal of Financial Economics, 85(3):709734, 2007.

    • Search Google Scholar
    • Export Citation
  • [35]

    Rainer Haselmann, David Schoenherr, Vikrant Vig,, et al. Rent-seeking in elite networks. Technical report, Research Center SAFE-Sustainable Architecture for Finance in Europe, Goethe University Frankfurt, 2016.

    • Search Google Scholar
    • Export Citation
  • [36]

    Charles Himmelberg, Christopher Mayer, and Todd Sinai,. Assessing high house prices: Bubbles, fundamentals and misperceptions. The Journal of Economic Perspectives, 19(4):6792, 2005.

    • Search Google Scholar
    • Export Citation
  • [37]

    Chang-Tai Hsieh, Erik Hurst, Charles I Jones, and Peter J Klenow,. The allocation of talent and us economic growth. Technical report, National Bureau of Economic Research, 2016.

    • Search Google Scholar
    • Export Citation
  • [38]

    Chang-Tai Hsieh and , Peter J Klenow,. Misallocation and manufacturing tfp in china and india. Technical report, National Bureau of Economic Research, 2007.

    • Search Google Scholar
    • Export Citation
  • [39]

    Chang-Tai Hsieh and , Zheng Michael Song,. Grasp the large, let go of the small: the transformation of the state sector in china. Technical report, National Bureau of Economic Research, 2015.

    • Search Google Scholar
    • Export Citation
  • [40]

    Erik Hurst and , Annamaria Lusardi,. Liquidity constraints, household wealth, and entrepreneur-ship. Journal of political Economy, 112(2):319347, 2004.

    • Search Google Scholar
    • Export Citation
  • [41]

    Thomas J Kane and , Douglas O Staiger,. Are teacher-level value-added estimates biased? an experimental validation of non-experimental estimates. In National Conference on Value-Added Modeling, Madison, WI, 2008.

    • Search Google Scholar
    • Export Citation
  • [42]

    Nobuhiro Kiyotaki, John Moore,, et al. Credit chains. Journal of Political Economy, 105(21):211248, 1997.

  • [43]

    Katharina Knoll, Moritz Schularick, and Thomas Michael Steger,. No price like home: Global house prices, 18702012. 2014.

  • [44]

    Patrick Legros and , Andrew F Newman,. Monotone matching in perfect and imperfect worlds. The Review of Economic Studies, 69(4):925942, 2002.

    • Search Google Scholar
    • Export Citation
  • [45]

    Alberto Martin and , Jaume Ventura,. Economic growth with bubbles. The American Economic Review, 102(6):30333058, 2012.

  • [46]

    Costas Meghir, Steven Rivkin,, et al. Econometric methods for research in education. Handbook of the Economics of Education, 3:187, 2011.

    • Search Google Scholar
    • Export Citation
  • [47]

    Atif Mian and , Amir Sufi,. House prices, home equity-based borrowing, and the us household leverage crisis. The American Economic Review, 101(5):21322156, 2011.

    • Search Google Scholar
    • Export Citation
  • [48]

    Atif R Mian, Kamalesh Rao, and Amir Sufi,. Household balance sheets, consumption, and the economic slump. Chicago Booth Research Paper, (13-42), 2013.

    • Search Google Scholar
    • Export Citation
  • [49]

    Virgiliu Midrigan and , Daniel Yi Xu,. Finance and misallocation: Evidence from plant-level data. The American Economic Review, 104(2):422458, 2014.

    • Search Google Scholar
    • Export Citation
  • [50]

    Benjamin Moll,. Productivity losses from financial frictions: can self-financing undo capital misallocation? The American Economic Review, 104(10):31863221, 2014.

    • Search Google Scholar
    • Export Citation
  • [51]

    Kevin M Murphy, Andrei Shleifer, and Robert W Vishny,. The allocation of talent: Implications for growth. The Quarterly Journal of Economics, 106(2):503530,1991.

    • Search Google Scholar
    • Export Citation
  • [52]

    Emi Nakamura, Jón Steinsson, and Miao Liu,. Are chinese growth and inflation too smooth? evidence from engel curves. Technical report, National Bureau of Economic Research, 2014.

    • Search Google Scholar
    • Export Citation
  • [53]

    Charles G Nathanson and , Eric Zwick,. Arrested development: Theory and evidence of supply-side speculation in the housing market. Technical report, Working paper, 2014.

    • Search Google Scholar
    • Export Citation
  • [54]

    Sari Pekkala Kerr, William Kerr, and Ramana Nanda,. House money and entrepreneurship. Technical report, Harvard Business School, 2015.

  • [55]

    Raghuram G Rajan and , Arvind Subramanian,. Aid, dutch disease, and manufacturing growth. Journal ofDevelopment Economics, 94(1):106118, 2011.

    • Search Google Scholar
    • Export Citation
  • [56]

    James Riedel, Jing Jin, and Jian Gao,. How China grows: Investment, finance, and reform. Princeton University Press, 2007.

  • [57]

    Albert Saiz,. The geographic determinants of housing supply. quarterly Journal of Economics, 125(3), 2010.

  • [58]

    Martin C Schmalz, David A Sraer, and David Thesmar,. Housing collateral and entrepreneurship. Technical report, 2015.

  • [59]

    Jing Shi, Yongxiang Wang, Weixing Wu,, et al. The crowding-out effect of real estate markets on corporate innovation: Evidence from china. In Asian Finance Association (AsianFA) 2016 Conference, 2016.

    • Search Google Scholar
    • Export Citation
  • [60]

    Kelly Shue,. Executive networks and firm policies: Evidence from the random assignment of mba peers. Review of Financial Studies, 26(6):14011442, 2013.

    • Search Google Scholar
    • Export Citation
  • [61]

    Zheng Song, Kjetil Storesletten, and Fabrizio Zilibotti,. Growing like china. The American Economic Review, 101(1):196233, 2011.

  • [62]

    Ilke Van Beveren,. Total factor productivity estimation: A practical review. Journal of economic surveys, 26(1):98128, 2012.

  • [63]

    Jeffrey M Wooldridge,. Inverse probability weighted estimation for general missing data problems. Journal of Econometrics, 141(2):12811301, 2007.

    • Search Google Scholar
    • Export Citation
  • [64]

    Jing Wu, Joseph Gyourko, and Yongheng Deng,. Real estate collateral value and investment: The case of china. Journal of Urban Economics, 86:4353, 2015.

    • Search Google Scholar
    • Export Citation
  • [65]

    Chunni Zhang and , Yu Xie,. Place of origin and labour market outcomes among migrant workers in urban china. Urban Studies, 50(14):30113026, 2013.

    • Search Google Scholar
    • Export Citation

Appendix

Appendix A: The Land Supply Elasticity of Chinese Cities

Employing Geography Information System (GIS), we precisely calculated exogenously undevelopable land within 30 kilometer radii from the central point of each city. Eliminating area lost to steep slope, bodies of water, territory boundaries, and special district boundaries which prohibit urban sprawl, we built a comprehensive measurement of urban land flexibility for all major Chinese cities.

Steep slope significantly constrains residential development. Utilizing GIS software, we calculated areas exhibiting slope over 10% within 30km meter radii of each city’s central point. Based on national contour lines map, we generated a national-wide slope map at a resolution of 1000 by 1000 meters. The data source of the contour lines map is the “1:1 Million Topographic Map of China” compiled by the National Bureau of Surveying and Mapping (NBSM, PRC). Considering that most residential construction projects take more than 1 square kilometer of land and small pieces of flat land within mountain area are expensive to develop, we choose 1000-meter resolution as the grid size for calculating slope.

There are several classifications regarding slope and urban construction suitability, and we set slope over 10% as unsuitable for urban housing development. The most popular standard for slope and urban construction suitability is set by Urban Planning Theory (third edition), which is the national textbook for urban planning majors. In this textbook, slopes between 0.3% -10% are considered to be suitable for residential land use. However, according to the Code for Vertical Planning on Urban Field published in 1999, the maximum slope allowed to be used for residential land use is 25% (P.6). The threshold is increased to cover most land in China56. Considering construction and maintenance costs will increase significantly for residential development on land with slope over 8%, and most Chinese cities are built upon plains, we choose to use the general guideline of slope below 10% instead of the maximum limit of 25% as the threshold for land considered suitable for housing development.

Residential development is also constrained by bodies of water, country territory boundaries, and special district administration boundaries. For example, Shenzhen housing development cannot cross the border with Hong Kong. By intersecting the polygon of coastline, inner water body, country territory boundary and special district administration boundary within the 30 km radii circle of each city, we can eliminate all undevelopable area caused by these factors. Data for coastline, territory boundaries and special district boundaries comes from the “Administrative Division Map of China” compiled by China Cartographic Publishing House and ESRI. Data for inner bodies of water comes from the hydro-graphic map, “1:1 Million Topographic Map of China” compiled by the National Bureau of Surveying and Mapping (NBSM, PRC).

Figure A.1 below illustrates the city-specific land supply elasticity index for 129 Chinese cities.

Figure A.1:
Figure A.1:

The Land Supply Elasticity Index in China

Citation: IMF Working Papers 2018, 221; 10.5089/9781484378465.001.A001

Appendix B: Additional Empirical Analyses

The Matching Approach

To account for the two omitted variable biases, I first select the sample using a semi-parametric matching procedure (Abadie and Imbens, 2006, 2012). I divide my sample into two groups: the treatment group and the control group. The treatment group includes companies that entered the real estate sector together with the comparison sets of all firms that did not. The control group includes companies that did not enter the real estate sector and the comparison sets of companies that entered. The exact matching procedure is as follows: for a given manufacturing firm i entered in the real estate development sector in year t, its matched comparison firms are restricted to operate in the same two-digit industry with the same exporting status, Gi(Ind, E). The set of matched comparison firms is then selected as the closest four matches to firm i from Gi(Ind, E):

J4(i)={1Gi(Ind,E)|Wl=0,XXid4(i)}

Wl and Wi are the treatment indicators of firm l and firm i, where W = 1 indicates the firm entered the real estate in year t; d4(i) is the distance between firm i and its fourth closest match. The matching variables include registered capital, firm age, total asset value, fixed asset value, employment, debt-to-asset ratio, and profit margin at time t – 3. I drop observations whose fourth closest match has a distance larger than 20% of that to the origin.

Firm i’s potential outcome in year t is computed as:

Y^i(0)={Yi</