WPS7193 Policy Research Working Paper 7193 Empowering Cities: Good for Growth? Evidence from China T. Juni Zhu Megha Mukim Trade and Competitiveness Global Practice Group & Social, Urban, Rural and Resilience Global Practice Group February 2015 Policy Research Working Paper 7193 Abstract This paper utilizes a countrywide, county-to-city upgrade in through state-owned banks as a possible explanation for the the 1990s to identify whether extending the powers of urban improvement in performance. The most important finding local governments leads to better firm outcomes. The paper is that newly-promoted cities with high capacity generally hypothesizes that since local leaders in newly-promoted produce better aggregate firm outcomes compared with cities have an incentive to utilize their new administrative newly-promoted cities with low capacity. The conclusions remit to maximize gross domestic product and employment are twofold. First, in terms of access to credit, the paper growth, there should be improvements in economic out- provides evidence that relaxing credit constraints for firms comes. The analysis finds that aggregate firm-level outcomes could lead to large increases in firm operation and employ- do not necessarily improve after county-to-city graduation. ment. Second, increasing local government’s administrative However, it does find that state-owned enterprises perform remit is not enough to lead to better firm and economic better post-graduation, with increased access to credit outcomes; local capacity is of paramount importance. This paper is a product of the Trade and Competitiveness Global Practice Group and the Social, Urban, Rural and Resilience Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at mmukim@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Empowering Cities: Good for Growth? Evidence from China* T. Juni Zhu and Megha Mukim† Keywords: Urbanization, Decentralization, Credit allocation, Capacity, Firm-level data, China JEL Classification: G21, H81, L11, R11, R51 * The authors are grateful to several World Bank Group staff who contributed their time generously to this paper, providing feedback, comments and suggestions – Stefano Negri, Bob Rijkers, Ana Aguilera, Leo Iacavone and Bilal Zia. Ritam Chaurey worked side-by-side with us and provided constructive and timely comments at every stage of the process. The authors received constructive feedback from seminar participants in Washington DC and in Beijing, including Bert Hofman, Paul Procee, Karlis Smits, Min Zhao, Shahid Yusuf, David Robalino, David Mason, Chandan Deuskar, and Guilherme Lichand. And finally, we are extremely grateful to Xiaobo Zhang (IFPRI) and Lixing Li (Peking University) for providing us access to data, for answering many, many questions and for their feedback. Funding from the Competitive Industries and Innovation Program is very gratefully acknowledged. † T. Juni Zhu is a consultant (Economist) with the Competitive Cities Knowledge Base (CCKB) project. Megha Mukim is an Economist and TTL of the CCKB Project. CCKB is a joint-collaboration between the Trade & Competitiveness GP and the Social, Urban, Rural and Resilience GP. 1 Introduction In an effort to promote urbanization, the Chinese government began to upgrade counties to become county-level cities between the 1980s and up until 1997. Counties were eligible to graduate into county-cities if they met certain minimum requirements (Ministry of Civil Affairs, 1993). An English version of this policy document can be found in Zhang and Zhao (1998). To become cities, counties needed to show that (1) their levels of industrialization were above certain thresholds, (2) the share of population engaged in non-agricultural activities was above a certain threshold, and (3) its fiscal status (as measured by total fiscal revenues and per-capita fiscal revenue) were in good shape – see Table 1 for the full list (taken from Fan, Li, & Zhang, 2012). Table 1: County to county-level-city minimum upgrading requirements Population Density (person/km2) >400 100-400 <100 Percentage of 25% 45% 30% counties in this category Industrialization Industrial output (yuan) 1.5 billon 1.2 billion 0.8 billion level Share of industrial output value 80% 70% 60% in gross value of industrial and agricultural output Population engages Size of non-agricultural 150 K 120 K 100 K in non-agricultural population activities Share of non-agricultural 30% 25% 20% population in total population Fiscal Strength Fiscal revenue (yuan) 60 million 50 million 40 million Per capita fiscal revenue (yuan) 100 80 60 In reality, these three requirements were not strictly enforced, partly because of large regional disparities across the country, wherein even after accounting for factors such as population density, counties in western and inland regions would still have trouble meeting these requirements. Instead, it seems that the decision to upgrade a county to a county-level city was based on rates of economic growth as well as the central government’s discretion (Li, 2011). Thus, this led to a situation where counties that did not meet any of the three requirements were upgraded, while some counties that met all three requirements were not upgraded – see Table 2 (taken from Fan, Li, & Zhang, 2012). 2 Table 2: County to county-level-city upgrade Number of county-year observations by upgrading status and number of requirements satisfied (1994-1997) Number of requirements satisfied Total 0 1 2 3 Non-upgrading cases 6401 4583 1317 465 36 Upgrading cases 99 24 30 39 6 Graduation into city status came with a number of benefits, relating to four main categories: (1) tax and fees, (2) land-related policies, (3) administrative powers, and (4) local government size, rank, salary and reputation - see Table 3 (taken from Fan, Li, & Zhang, 2012). Table 3: Benefits of being a city: An incomplete list Category Benefits Tax and fees - Cities enjoyed a higher urban construction tax (7% compared to 5% for counties). - Could collect the surcharges levied on the issuing of motorcycle registration. - In Liaoning province, cities could get 1-2 million additional subsidies each year after upgrading. Land-related policies - Cities generally convert more land to construction use and retain a larger share of revenue from land sale. Administrative powers - Cities have more authority on foreign trade and exchange management. - Could establish branches of custom and large state-owned banks. - Could approve projects with a higher investment cap. - Gain authority over police recruitment and vehicle administration. - After achieving the status of “line item under province” (Shengji Jihua Dalie), cities could report directly to the provincial administration to ask for investment projects. Government size; Rank and - Cities could establish more branches of government and salary; Reputation have a larger number of government employees. - Sometimes the bureaucratic rank and salary of officials are raised after upgrading. - Cities generally carry greater prestige and are more attractive to investors from outside. In this paper, we investigate if newly-promoted county-level cities in China utilize these new powers to attract new firms and help firms grow. We study macro-level city outcomes and micro- 3 level firm outcomes. We hypothesize that since local government officials wanted to achieve high GDP and employment growth, they would have an incentive to utilize their new administrative remit to promote firm growth and hence we should observe better firm-level outcomes in the newly upgraded cities versus their similar counties. We exploit the ad hoc nature in which the county- upgrading takes place to identify the effect of an increase in city powers on firms’ economic outcomes. The paper adds to the existing literature in two ways – it identifies whether and to what extent firm-level outcomes improve post-upgrading, and importantly, it attempts to identify the channels through which these outcomes take place. In addition, this paper uses two proxies to measure local government capacity and documents the extent to which local government capacity matters for firm and job growth in China. The rest of the paper is organized as follows. Section 2 describes the data sets and how these were prepared. Section 3 presents the empirical strategy used to identify the effects of city-upgrading on firm outcomes. Section 4 reports the econometric results for city-level outcomes and firm-level outcomes and Section 5 concludes. 2 Data Sets and Data Treatment This paper utilizes two panel data sets, one at the level of counties and the other at the level of firms. We use variables from 1993 to 2004 from the annual series of the Public Finance Statistical Materials of Prefectures, Cities, and Counties published by the Ministry of Finance of China to construct a county-level public finance data set (henceforth public finance data set). As of 1998, when the policy of county-to-city upgrading had come to an end, there were 99 counties that were upgraded from counties to county-level cities between 1994 and 1997. The firm-level data set spans from 1998 to 2009 and is compiled from the Annual Survey of Industrial Firms (ASIF) collected by the National Bureau of Statistics of China. This data set is often referred to as the 5 Million RMB data set, since it contains non-SOE firms with main operating revenue (i.e., sales) of more than five million RMB, and all state-owned enterprises (SOEs) regardless of sales volume. Thus, the ASIF survey data set tracks the performance of all SOEs and all big non-SOEs. The ASIF data set details all operational, financial, and managerial facts of firms in three broad categories: (1) mining, (2) manufacturing, and (3) production and distribution of electricity, gas, and water, and classifies each firm to 6-digit sub-industries. Compared with the 2004 China Economic Census, the firms covered by this data set represented 89.5% of the total main operating revenue (sales) from all enterprises in China (Nie, Jiang, & Yang, 2012). The 1998-2009 ASIF data set covers a period when some major structural reforms happened in China, such as the SOE reform in the mid-to-late 1990s. Following the end of the county-to-city upgrading policy in 1997, China carried out a major reform of SOEs between 1998 and 2000. The aim of the reform was to privatize or close down small and unprofitable SOEs (Song and Hsieh, 2013). In 2004, an economic census covering all firms regardless of size was carried out and more detailed information on firms was collected.1 To mitigate the effects of the above SOE changes, and since we are only interested in understanding the effect of county-to-city upgrading, we restrict 1 For example, in this year, breakdown of employment by education level, gender, technical titles was collected (Brandt, Van Biesebroeck, & Zhang, 2014). 4 our analysis to incumbents from 1998 to 2004, hence excluding all new firm entries and exits. We create a balanced panel of 36,778 firms, located in 793 counties and 58 newly promoted county- level cities. Table 4 summarizes the breakdown of firms by ownership. Table 4: Number of firms by ownership type Firms Number of firms As a % of total SOE 15,413 42.12% Non-SOEs (e.g. collective, private, mix) 21,183 57.88% Total 36,596 100% 3 Empirical Strategy To properly identify the effect of a change in city-level powers on economic outcomes, we need to control for selection bias. Simply put, if better performing or better managed counties were more likely to be upgraded to county-level-cities, then a comparison of outcomes across counties and cities would be upwardly biased. Our aim is to evaluate the causal effect of city upgrading on economic outcomes. We exploit the ad hoc nature of county-to-city upgrading to find an appropriate counterfactual. We do this by matching newly-promoted cities with counties that are similar to these cities (and if the upgrading requirements were properly applied, would have been promoted). The counterfactual allows us to analyze how economic outcomes within a city would have evolved if it had not been upgraded to city status. Propensity Score Matching (PSM) Let’s assume that a county is promoted to a county-level city at time s  0 . Let  is be the economic outcome at time s (the outcomes for county i at period s ) following upgrading to city status at s  0 and the variable CITYi takes on the value one if county i becomes a city. The causal effect can be verified by looking at the difference: (  1 is   is ), where  0 the superscript denotes the   is 0  The crucial problem is that promotion.  observable. We follow the micro-econometric is not evaluation literature (Heckman, Ichimura, & Todd, 1997) and define the average effect of   upgrading on economic outcomes as:   E[is 1  is 0 | CITYi  1]  E[is 1 | CITYi  1]  E[is 0 | CITYi  1] (1) The key difficulty is to identify a counterfactual for the last term in Equation (1). This is the economic outcome that a city would have experienced, on average, had it not been promoted to a city. What is mainly of interest is the magnitude of the ‘impact’, labelled in red in Figure 1, and the main problem is the calculation of the counterfactual that is to be deducted from the total change. 5 This counterfactual is estimated by the corresponding average value of counties that remain as counties (and are not upgraded): E[is o | CITYi  0] . An important feature of the construction of the counterfactual is the selection of a valid comparison group. In order to identify this group it is assumed that all the differences in economic outcomes (except that caused by upgrading) between cities and the appropriately selected county comparison group are captured by a vector of observables, including the pre-upgrade county economic outcomes. The intuition behind selecting the appropriate comparison group is to find a group that is as close as possible to the upgraded county in terms of its predicted probability to be upgraded. Following Fan, Li, & Zhang (2012), we control for the selection bias by matching cities that were upgraded with similar counties that could have been upgraded (but were not). Since the county- to-city upgrade policy came to an end in 1997, we use county-level variables from 1993 to 1997 to carry out the matching. We drop cities that were upgraded before 1994 (since the public finance data only start from 1993). We carry out the matching exercise using observable county-level economic outcomes in 1994 – we match cities (i.e. counties that will be promoted to county-level cities) with counties (i.e. counties that were not promoted to cities). Figure 1: Identification of impact of upgrading to city status More formally, we apply the propensity score matching (PSM) method as proposed by Rosenbaum and Rubin (1984). This boils down to estimating a probit model with a dependent variable equal to one if the county is upgraded and zero otherwise, on lagged variables. The probability of being upgraded is modelled as follows. CITY is a dummy variable that equals one if a county is upgraded. The probability of being promoted, i.e. the propensity score, can be represented as follows: Pr(CITYi ,1994  1)  F ( xi ,1993) (2) 6 where F (.) is the normal cumulative distribution function. The variables used in this exercise correspond to the three upgrading requirements mentioned in Table 1. Since these upgrading requirements were not strictly enforced, we used these upgrading requirements variables as a general indicator of the economic development level of a county in 1993 to predict the probability of a county being upgraded to a city in 1994. We use the first and second moments of the three upgrading requirements as our specification to predict propensity scores. If they can pass the common support test: the covariates in each stratified block are balanced, then we will use this specification to predict the effects of city upgrading by comparing outcomes between treatment (cities) and comparison (counties) groups (Rosenbaum and Rubin, 1984). We find that there are no significant differences in covariates between counties and cities within each stratified block using this specification. We hence use this specification to estimate the city upgrading treatment effect on city-level and firm-level outcomes. Since the public finance data set is a panel data set, we obtain one propensity score for each jurisdiction-year pair. Hence each county or city will have multiple propensity scores. To mitigate the potential problem of counties inflating their economic figures right before the upgrading (Li, 2011), we used the earliest data point possible from the public finance data set to conduct the matching exercise. Thus, a county or city’s 1994 predicted propensity score is used to match a city with its similar county. Li (2011) pointed out that the rate of economic growth is one of the key factors in determining which counties can be upgraded to cities. Therefore, in addition to variables corresponding to the three upgrading requirements, ideally we should include the growth rate of GDP as a control in predicting upgrading probability. However, GDP at the county-level is only available starting in 1997 when there was a major statistical standard change in China. Before 1997, we have official statistics on gross value of industrial and agricultural output, and include these variables in our PSM model as an alternative measure of economic development at the county-level.2 The propensity matching score method also assumes that there exists a region of ‘common support’, where the treated and control propensity scores overlap, and over which a robust comparison can be made. Cities that fall outside of the region of common support are disregarded and for these cities the treatment effect cannot be estimated With matching, the proportion of such cities is small (only two city-year observations using the specification that we described earlier fall into this case—see Figure 2). Since the region of common support is vastly improved (see Figure 3 for kernel density distribution comparison) and the balancing test is passed between the treatment and control groups, the estimated effect on the remaining cities can be viewed as representative. 2 Variables that can potentially have extreme values are transformed to logs to minimize distortions. 7 Figure 2: Matched counties and cities by blocks of 1994 propensity scores County City Blocks of PScores Total (comparison) (treatment) 0 2,953 1 2,954 0.003125 470 3 473 0.00625 501 4 505 0.0125 522 16 538 0.025 541 14 555 0.05 532 34 566 0.1 387 73 460 0.2 239 92 331 0.4 64 57 121 0.6 6 19 25 0.8 0 2 2 Total 6,215 315 6,530 Figure 3: Kernel density distribution comparison between cities and counties Kernel Density Estimate 50 40 30 Density 20 10 0 0 .2 .4 .6 .8 1 Estimated propensity score City County kernel = epanechnikov, bandwidth = 0.0552 8 Identification of City-Level and Firm-Level Treatment Outcome We use the treatment and counterfactual groups to carry out two distinct exercises to identify the effect of city-level upgrading. First, we identify changes in city-level outcomes post-upgrading. Fan, Li, & Zhang (2012) used a county and year fixed effects model to identify post-upgrading effects. We use a propensity score model and compare our results with their findings. To identify city-level outcomes, we stratify the county-level data according to their propensity scores, and compare outcomes between treatment and comparison groups within each block and identify the average upgrading effect (Dehejia and Wahba, 1999).3 Second, we identify changes in firm-level outcomes post-upgrading. We compare the performance of firms located in cities versus the performance of firms located in matched counties. Ideally, we would have liked to use difference-in-difference estimation in addition to the PSM, however we are constrained by data limitations (firm-level data begin in 1998, after counties had already been upgraded). We try to minimize the selection bias problem by limiting the results to incumbents, i.e. firms that were already operating in 1997.4 We also control for year, industry, and time- varying industry fixed effects in addition to the propensity score method to compare the robustness of our firm-level outcome. 3 We considered weighting the samples to identify the average treatment effect. However, since there is not a consensus which weighting method is the best in identifying the average treatment effect, we decide not to weight the samples, simply controlling for block and year fixed effects. 4 This assumes that firms in counties and newly-promoted counties were similar previous to the upgrade. 9 4 Econometric Results City-Level Outcomes Table 5 reports the city upgrading effects on city-level outcomes, especially government activities since the upgrading. Compared with a fixed effects model by Fan, Li, & Zhang (2012), the propensity score model with or without year fixed effects yields similar results. Thus, we confirm the findings in Fan, Li, & Zhang (2012) that government spending share in productive activities (as defined by spending in construction and providing support to agricultural production) decreases following upgrading. The share of agricultural tax in total tax also falls – this is intuitive since cities began to shift away from agricultural production. In addition, newly-upgraded cities do not necessarily outperform similar counties in terms of rates of GDP growth in the post-upgrade period of 1998-2004. Similar to Fan, Li, & Zhang’s (2012) fixed effects model, findings using a propensity score matching method suggest that at the county- level, upgrading does not necessarily lead to higher growth. We also try and identify the rate of new firm entrants in newly-promoted counties compared to the counterfactual counties that were not promoted. We find that there is a significant difference between the two, suggesting that new firms favor cities to counties, even if the former are only recently promoted. Correspondingly, tax from business income for cities has increased more significantly than their similar counties. Table 5: City-level outcome of city upgrading using PSM PS Model PS + Fixed Fan, Li, & Effects Zhang 2012 City-level Govt. Activities Outcome # of public employees 414.9** 407.5** 995*** Share of productive expenditure -0.00577** -0.00575** -0.026*** Share of agriculture tax -0.0586*** -0.0588*** -0.053*** Post-upgrade average GDP growth -0.000693 -0.000797 - # of firm births 4.371*** 4.844*** - Log tax from business income 0.515*** 0.517*** - Controls Block FE Block FE County FE Year FE Year FE 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. The pscore model uses variables of the three upgrading requirements and their interactions to generate propensity score. 10 Firm-Level Outcomes We are also keen to understand how upgrading to city status affects firm-level outcomes. Before we employ the propensity score matching method to identify the upgrading effects on firms, please see below a summary of descriptive statistics between firms located in county-level cities versus firms located in counties. At first glance, firms located in newly-upgraded cities on average seemed to significantly outperform firms located in counties. Table 6: Descriptive statistics of mean aggregate firm-level outcomes All Firms Firms in Firms in Cities Counties t-test Firm-level Outcome Log main operating revenue 9.462 9.866 9.380 0.487*** Log main operating cost 9.228 9.649 9.142 0.507*** Log main operating profit 7.330 7.624 7.270 0.355*** Log # of employees 5.153 5.300 5.123 0.177*** Log labor vocational 2.502 2.439 2.514 -0.075 Log wage per employee 8.792 8.871 8.776 0.095*** Log paid-in capital 8.349 8.425 8.334 0.091*** Log export output value 9.489 9.613 9.456 0.157*** 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. In the next sections, we report firm-level outcomes using the propensity score matching method. City-level outcomes suggest that firms tend to favor cities over counties. To avoid distortion of statistical results, we first restrict our firm sample to incumbents (i.e. firms that were established before 1997) and report their firm-level outcomes after the upgrade. To better understand the full extent of firm-level outcomes, we also look at new firms’ performance and examine whether locating in cities helps them perform better. Since there is an inherent selection bias while studying new entrants to cities versus counties, we also study the effect of city upgrading on incumbents. Firm-Level Outcomes (for Incumbents) After matching county-level cities with similar counties using propensity scores, Table 7 shows that, at the aggregate level, firms that are located in upgraded cities do not necessarily perform better than firms located in similar counties. Standard errors are clustered at the county-level. Although Fan, Li, & Zhang (2012) do not use firm level data, our estimates for aggregate firm- level outcomes are in line with their findings that upgrading from county to county-level city does not necessarily generate better city-level economic performance and public service provision. Our results suggest that newly-formed cities are not using their increased powers to help Chinese firms perform better. 11 Table 7: Firm-level outcomes of city upgrading using PSM for incumbents Incumbent Firms (1) (2) (3) Firm-level Outcome Log main operating revenue 0.126 0.109 0.109 Log main operating cost 0.127 0.108 0.108 Log main operating profit 0.0673 0.0275 0.0275 Log # of employees -0.0195 -0.0859 -0.0859 Log labor vocational -0.0780 -0.0531 -0.0531 Log wage per employee -0.0528 -0.0409 -0.0395 Log paid-in capital -0.116 -0.117 -0.117 Log export output value -0.102 -0.118 -0.126 Controls Block FE Block FE Block FE Industry FE Industry × Year Year FE FE 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. When we report the firm-level results after PScore matching, we control for block fixed effects where samples are stratified and compared against each other in each block. 3. Analysis is restricted to firms opened before 1997. 4. Vocational labor data is only available in 2004. To ensure that the above aggregate results are not sensitive to different matching methods, we report below the aggregate firm-level outcomes using three different matching methods commonly used in the propensity score literature. We used the same propensity scores generated above to conduct the matching. We can see that compared with Table 7, similar conclusions can be reached. Upgrading counties into county-level cities alone has very limited effects on aggregate firm level outcomes in terms of firm profits, export, and employment. Table 8: Firm-level outcomes using different propensity score matching methods Local Linear Incumbent Firms Kernel Regression 5-Nearest Neighbor Firm-level Outcome Log main operating revenue 0.116*** 0.340 0.457*** Log main operating cost 0.125*** 0.378 0.552*** Log main operating profit 0.052 0.162 0.077 Log # of employees -0.027 -0.096 -0.308*** Log labor vocational -0.108 -0.114 -0.206 Log wage per employee -0.064*** 0.085 0.206*** Log paid-in capital -0.158*** -0.214 -0.452*** Log export output value -0.062 0.206 0.359 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Average treatment on treated effects are reported, i.e. upgraded cities and their counterfactual counties are compared to obtain the upgrading effects. 12 Are there winners and losers? Next, we try and disaggregate this result using data on firm ownership. Firms are classified as state-owned enterprises (SOEs) and other (which includes collectives, private firms, and firms with mixed ownership). We are interested in understanding if SOEs seem to outperform their non-SOE counterparts after upgrading to city status. We find that although the aggregate results across all firms seem to suggest that cities are not using their powers to help firms, it turns out that SOEs in newly-upgraded cities do significantly better than non-SOEs, compared to those in matched counties. SOEs sell more and employ more (skilled) labor, although they are no more profitable. Table 9: Firm-level outcome of city upgrading by ownership type using PSM for incumbents Log main Log main Log main Log wage Incumbent Log # of Log labor Log operating operating operating per Firms employees vocational export revenue cost profit employee City -0.0828 -0.0827 -0.118 -0.268*** -0.241** -0.0776 -0.160 (0.106) (0.106) (0.125) (0.0812) (0.0978) (0.0519) (0.233) SOE -0.911*** -0.908*** -0.573*** -0.0435 -0.00604 -0.293*** -0.630*** (0.0720) (0.0730) (0.0740) (0.0559) (0.0713) (0.0326) (0.157) City × SOE 0.305** 0.304** 0.249 0.390*** 0.404*** 0.0584 0.332 (0.150) (0.151) (0.163) (0.102) (0.143) (0.0552) (0.368) Constant 11.10*** 10.64*** 9.303*** 6.038*** 2.724*** 8.623*** 9.259*** (0.600) (0.576) (0.690) (0.256) (0.149) (0.563) (0.469) Observations 32,517 32,428 21,283 18,193 3,892 17,496 4,759 R-squared 0.296 0.304 0.186 0.203 0.169 0.197 0.152 Block FE Yes Yes Yes Yes Yes Yes Yes Industry × Year FE Yes Yes Yes Yes Yes Yes Yes 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Standard errors are in parentheses and clustered at county-level. 3. Analysis is restricted to firms opened before 1997. 4. Vocational labor data is only available in 2004. Why do SOEs outperform their private counterparts? In this section, we try to solve the puzzle of why SOEs gain disproportionately compared with other types of firms. Going back to Table 3, one of the main benefits of city upgrading is that a city can now establish branches of state-owned banks (SOBs). According to Wei and Wang (1997), bank loans made from state-owned banks clearly favored SOEs. In the 1990s, many state-owned banks imposed softer budget constraints on SOEs than in the 1980s, such that bank finance and firm productivity were no longer linked (Cull and Xu, 2000). In other words, the literature provides evidence of favorable lending from SOBs to SOEs. We go back to firm-level data and examine whether SOEs receive more credit from SOBs than other firms. Ideally, we would like to identify the source of the increase in credit to SOEs – however, the data do not provide a breakdown of the sources of debt financing. Instead, we use current debt and total debt as proxies to measure credit from SOBs and assume that most of the 13 debt financing in counties and county-level cities comes from SOBs. This assumption is not without foundation. The fact that the establishment of SOBs is one of the major benefits associated with city upgrading indicates that commercial banks or other credit channels are very limited at the county level in China. In Table 10, the first two columns show that compared with non-SOEs, SOEs located in cities saw a big increase in both current debt and total debt. This suggests that part of the SOE performance differential obtained in Table 9 above could be explained by easier access to credit by SOEs (compared to non-SOEs) post-upgrading. As a robustness check of this possible channel, we try to compare the debt profile of negative profit SOEs in cities versus counties. Our hypothesis is that if underperforming SOEs are able to access credit more easily than underperforming non-SOEs, then the debt is probably being financed by SOBs that have branches in cities (since commercial banks or other lending agents would not lend to these underperforming firms). The last two columns of Table 10 report this result. Although the coefficient is positive for cities, it is not significant (even though its p-value is not far from 0.1). Table 10: Debt financing of incumbent SOEs Full Sample Negative profit SOEs Only Incumbent Firms Log current debt Log total debt Log current debt Log total debt City -0.273* -0.207 0.442 0.421 (0.147) (0.142) (0.276) (0.280) SOE 0.261*** 0.248*** (0.0819) (0.0801) City X SOE 0.356** 0.341* (0.172) (0.174) Constant 10.49*** 10.77*** 9.706*** 10.10*** (0.529) (0.513) (0.255) (0.277) Observations 32,240 32,732 1,494 1,515 R-squared 0.201 0.188 0.388 0.350 Block FE Yes Yes Yes Yes Industry × Year FE Yes Yes Yes Yes 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Standard errors are in parentheses and clustered at the county-level. 3. Analysis is restricted to firms opened before 1997. There is a vast literature showing how firms in developing countries are more likely to report access to finance as a major obstacle to their growth, especially for small firms (Bloom, Mahajan, & Roberts, 2010). Results from randomized control trials giving credit to small and medium-sized enterprises in Sri Lanka (De Mel, McKenzie, &Woodruff, 2008) and India (Banerjee and Duflo, 2008) illustrate that access to credit was indeed a big problem for disadvantaged firms in developing countries. There is also cross-country evidence showing if banks are concentrated, with a large share being government-owned banks, then financing obstacles will increase and the chance of smaller firms 14 to access credit will decrease (Beck, Demirguc-Kunt, and Maksimovic, 2004). With the new micro firm-level data from China, it seems this argument is even stronger that compared with SOEs, even large non-SOEs are disadvantaged in accessing credit at the county level. Does capacity matter? Even though the city-level outcome does not show that cities with a bigger scope of power outperform their similar counties, we are worried that the reason we observe this result was because city governments do not have the capacity to carry out these new powers, and not because the new powers themselves are not useful. In urban governance literature, the powers of a city government for managing economic development depend on factors of not only its operational scope, but capacity as well (Davey, 1993). We therefore want to test and see whether newly promoted cities with higher capacity help firms perform better. Regional disparities in human capital are vast in China, and we include province fixed effects to offset partly this regional difference when evaluating firm-level outcomes, in addition to the time and industry fixed effects we have controlled for. Since there is not a unified measure of city capacity, we propose using two proxies to measure capacity to conduct this analysis due to data limitation and availability. We propose using the percentage of public employees supported by public finance out of total city population as a measure of city capacity. Adequate and institutionalized human capital is often cited as one of the key factors in determining city capacity (World Bank, 2009). A recent paper by Acemoglu, García-Jimeno and Robinson (2014) also used a similar measure to measure capacity: the number of municipality-level bureaucrats excluding police officers, judges, all other judicial employees, and public hospital employees. Alternatively, we propose using the total tax revenues collected, excluding land sales revenues, out of total city GDP as a proxy of city capacity.5 Fukuyama (2013) proposed using tax revenues out of GDP as a proxy for state capacity, as the ability to extract tax not only indicates a government’s capability, but also means the government has revenues to carry out public functions and hence tax extraction can be a good proxy for capacity.6 From previous results, we have shown that at the aggregate firm level, simply being located in a newly promoted city does not help firms perform better. However, from Table 11, after taking into account capacity as measured by more institutionalized human capital available to public services, firms located in newly promoted cities with higher capacity seem to perform better in terms of sales, profits and wage rate of employees. 5 The two proxies of city capacity are by no means perfect. In an ideal world, proper understanding and measurement of capacity would require a combination of quantitative proxies supplemented by qualitative data. 6 We are interested in whether city-upgrading itself leads to a jump in capacity post-upgrade. Plotting the two proxies against time period of 1996 to 2004 shows us that city capacity does not change much after the upgrade. Therefore, we are essentially comparing cities with different initial capacity levels here and report how capacity matters. 15 Table 11: Firm-level outcomes of city upgrading and city capacity for incumbents Log main Log main Log main Log Log # of Log labor Log Incumbent Firms operating operating operating wage per employees vocational export revenue cost profit employee City -1.210** -1.216** -1.311** -0.580 -0.252 -0.452* -0.348 (0.504) (0.501) (0.537) (0.390) (0.357) (0.262) (0.925) Capacity -21.56*** -22.01*** -17.69*** -14.79*** -15.43*** -0.127 -7.195 (4.616) (4.775) (4.695) (3.426) (4.028) (1.640) (12.58) City × Capacity 46.61** 46.65** 49.13** 18.57 8.353 17.37* 8.777 (20.33) (20.12) (21.80) (15.02) (13.86) (10.38) (37.08) Constant 11.85*** 11.43*** 10.12*** 6.673*** 3.117*** 8.353*** 9.227*** (0.672) (0.658) (0.752) (0.385) (0.219) (0.509) (0.990) Observations 32,517 32,428 21,283 18,193 3,892 17,496 4,759 R-squared 0.330 0.336 0.228 0.245 0.221 0.251 0.179 Block FE Yes Yes Yes Yes Yes Yes Yes Industry × Year FE Yes Yes Yes Yes Yes Yes Yes Province FE Yes Yes Yes Yes Yes Yes Yes 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Standard errors are in parentheses and clustered at county-level. 3. Analysis is restricted to firms opened before 1997. 4. Vocational labor data is only available in 2004. 16 Table 12: Re-examine incumbent SOE outcomes after taking into account city capacity Log main Log main Log main Log Log # of Log labor Log Incumbent Firms operating operating operating wage per employees vocational export revenue cost profit employee City -0.704 -0.685 -0.848 -0.255 -0.208 -0.209 0.0498 (0.760) (0.767) (0.906) (0.443) (0.444) (0.294) (1.045) Capacity -4.061 -5.576 4.035 8.095 -2.814 0.945 -5.989 (6.818) (7.159) (8.218) (5.278) (6.852) (3.019) (13.56) SOE -0.176 -0.206 0.263 0.721*** 0.425** -0.201** -0.772 (0.218) (0.226) (0.242) (0.172) (0.213) (0.0874) (0.565) City × Capacity 23.63 22.64 29.11 -0.644 -0.324 8.319 -7.217 (31.10) (31.28) (36.52) (16.48) (17.69) (11.62) (42.60) Capacity × SOE -19.84*** -18.57** -25.01*** -26.82*** -15.12** -1.173 10.27 (7.134) (7.401) (8.443) (5.719) (6.890) (2.958) (20.07) City × SOE -0.649 -0.721 -0.521 -0.138 0.173 -0.508 -1.901 (1.010) (0.996) (1.232) (0.640) (0.783) (0.334) (2.101) City × Capacity × 30.70 33.60 22.80 17.64 6.978 18.52 80.49 SOE (39.36) (38.63) (47.66) (24.26) (30.54) (12.77) (80.46) Constant 12.09*** 11.70*** 9.979*** 6.097*** 2.671*** 8.573*** 9.736*** (0.676) (0.667) (0.760) (0.442) (0.265) (0.520) (0.873) Observations 32,517 32,428 21,283 18,193 3,892 17,496 4,759 R-squared 0.351 0.356 0.236 0.252 0.225 0.264 0.190 Block FE Yes Yes Yes Yes Yes Yes Yes Industry × Year FE Yes Yes Yes Yes Yes Yes Yes Province FE Yes Yes Yes Yes Yes Yes Yes 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Standard errors are in parentheses and clustered at county-level. 3. Analysis is restricted to firms opened before 1997. 4. Vocational labor data is only available in 2004. SOEs located in newly promoted cities outperform non-SOEs (see Table 9). However, once we account for city capacity, this result no longer holds – see Table 12. This might suggest that cities with more power but lower capacity are more likely to exploit the extra power to favor certain types of firms, such as SOEs. We re-run the above exercise using another proxy for city capacity, i.e. total tax collection as a proportion of city GDP. We find similar results - only cities with both the scope and capacity to carry out their additional remit are able to help firms grow and help employees increase wages. 17 Table 13: Firm-level outcomes of city upgrading and city capacity for incumbents using an alternative measure of capacity Log main Log main Log main Log wage Incumbent Log # of Log labor Log operating operating operating per Firms employees vocational export revenue cost profit employee City -0.606** -0.560* -0.642*** -0.457** -0.496 -0.340*** -0.734 (0.301) (0.306) (0.234) (0.228) (0.302) (0.114) (0.530) Capacity -0.00236 -0.00171 0.00730 0.00318 -0.00427 0.000172 -0.0136 (0.00674) (0.00672) (0.00841) (0.00387) (0.00519) (0.00173) (0.00899) City × 16.22* 14.70 16.29** 9.629 12.46 9.621*** 16.91 Capacity (8.977) (9.032) (7.336) (6.678) (8.363) (3.185) (16.13) Constant 11.07*** 10.64*** 9.474*** 6.081*** 2.779*** 8.356*** 8.822*** (0.682) (0.658) (0.776) (0.201) (0.190) (0.504) (0.917) Observations 32,517 32,428 21,283 18,193 3,892 17,496 4,759 R-squared 0.325 0.330 0.225 0.240 0.218 0.252 0.180 Block FE Yes Yes Yes Yes Yes Yes Yes Industry × Yes Yes Yes Yes Yes Yes Yes Year FE Province FE Yes Yes Yes Yes Yes Yes Yes 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Standard errors are in parentheses and clustered at county-level. 3. Analysis is restricted to firms opened before 1997. 4. Vocational labor data is only available in 2004. Similarly, SOEs are less likely to outperform private firms in cities with high capacity when an alternative measure of city capacity is used. 18 Table 14: Incumbent SOE outcomes of city upgrading (using an alternative measure of capacity) Log main Log main Log main Log wage Incumbent Log # of Log labor Log operating operating operating per Firms employees vocational export revenue cost profit employee City -0.298 -0.251 -0.302 -0.512** -0.418 -0.338** -0.515 (0.340) (0.339) (0.412) (0.236) (0.284) (0.147) (0.687) Capacity 0.00895** 0.00875* 0.0241*** 0.00706*** 0.00505 0.00402 -0.0114 (0.00401) (0.00446) (0.00579) (0.00252) (0.00508) (0.00701) (0.00845) SOE -0.708*** -0.706*** -0.394*** 0.00481 0.0168 -0.230*** -0.501*** (0.0689) (0.0709) (0.0699) (0.0535) (0.0709) (0.0261) (0.156) City × 4.205 2.606 4.621 6.125 4.677 10.16** 10.44 Capacity (11.15) (11.06) (13.31) (6.679) (8.124) (4.378) (21.20) Capacity × -0.0148*** -0.0137*** -0.0215*** -0.00400 -0.0105 -0.00508 -0.0159*** SOE (0.00379) (0.00343) (0.00309) (0.00243) (0.0107) (0.00826) (0.00380) City × SOE -0.340 -0.339 -0.448 0.222 -0.0435 0.00244 -0.0611 (0.410) (0.403) (0.555) (0.317) (0.413) (0.174) (1.158) City × Capacity × 16.10 16.16 16.56 4.236 13.08 -1.310 5.910 SOE (12.48) (12.15) (16.50) (9.334) (12.06) (4.969) (31.68) Constant 11.74*** 11.30*** 9.851*** 6.061*** 2.650*** 8.602*** 9.209*** (0.666) (0.643) (0.766) (0.206) (0.195) (0.513) (0.944) Observations 32,517 32,428 21,283 18,193 3,892 17,496 4,759 R-squared 0.346 0.350 0.231 0.243 0.221 0.264 0.189 Block FE Yes Yes Yes Yes Yes Yes Yes Industry × Year Yes Yes Yes Yes Yes Yes Yes FE Province FE Yes Yes Yes Yes Yes Yes Yes 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Standard errors are in parentheses and clustered at county-level. 3. Analysis is restricted to firms opened before 1997. 4. Vocational labor data is only available in 2004. Tax collection represents only a subset of total revenues that are available to a local government in China, where land sales and transfers from central government are also major sources of revenues. As a robustness test, we use total government spending as a proportion of city GDP as a proxy to measure the extent to which public services are being delivered using public funding. We obtain similar results - firms in high-capacity cities tend to create more jobs, especially skilled labor jobs, and SOEs are less likely to outperform private firms. 19 Firm-Level Outcomes (for New Entrants) So far, we have shown that for incumbent firms, city status does not necessarily lead to better aggregate firm-level outcomes, unless accompanied by commensurate capacity. Now, we study new firms that are established in newly upgraded cities. After counties upgrade to cities, they tend to attract more firms. We examine whether these new firms also outperform new firms located in similar counties. Admittedly, this exercise is fraught with selection bias. However, we remain interested in knowing if new firms in cities outperform new firms in similar counties – especially, since this finding has policy implications. Cities care about better economic outcomes – these could be generated by better performance by incumbents, or by better-performing entrants. New firms in cities tend to operate on a larger scale than new firms in counties – in terms of operating revenues (sales) and operating costs. However, they are not necessarily more profitable, nor do they generate more or better-paying jobs. Thus, city status does not necessarily attract better and more competitive firms to locate there. We also tried to break down new firms located in high capacity cities versus low capacity cities and see whether there is a difference. Unfortunately, we do not have enough data on entrants to carry out a robustness test. Table 15: Firm-level outcomes of city upgrading using PSM for new firms New Firms (1) (2) (3) Firm-level Outcome Log main operating revenue 0.297 0.344* 0.344* Log main operating cost 0.378* 0.423** 0.427** Log main operating profit 0.0845 0.294 0.322 Log # of employees 0.113 0.154 0.140 Log wage per employee -0.000390 -0.0564 -0.0572 Log paid-in capital -0.111 -0.0189 -0.0180 Controls Block FE Block FE Block FE Industry FE Industry × Year Year FE FE 1. Significance levels of 10%, 5%, 1% are represented by *, **, ***, respectively. 2. Please note that there are not enough observations for new firms’ vocational labor data and export value data. To avoid using an under-representative sample and generalizing results from comparing between new firms located in only a few cities and counties, we do not report the results of these two firm-level outcomes. 3. Analysis is restricted to new firms opened after 1997, i.e. after the upgrading was finished. 20 5 Conclusion This paper utilizes a countrywide county-to-city upgrade in the 1990s in China to explore whether expanding city power leads to better firm performance. When counties are upgraded to cities, their remit expands and they gain additional administrative and fiscal powers. An increase in city powers provides these newly-promoted counties with the ability to provide greater support or at the minimum, a better business environment, for firms, helping to ensure more growth and employment. Unfortunately, in this paper we find that this is not always the case. Increasing the policy space controlled by a city does not necessarily translate into better city and firm performance. This does not suggest that cities could not utilize their new powers effectively – indeed, we find evidence that certain types of firms (state-owned enterprises) begin to outperform their non-SOE counterparts as soon as their credit constraints are relaxed. Newly-established state-owned banks within cities might help explain better access to credit for SOEs, leading to higher levels of employment and increased sales by SOEs. This suggests that if access to finance were a market- based decision, then the gains from the city-upgrading policy may be expanded to all firms, not just SOEs. We also examine the effects of an increased “mayor’s wedge” and take into account not just increased powers, but also city capacity. Governance literature shows that in order for a government to manage well its economic development goals, both its operational scope and capacity matter. Therefore, granting additional powers to newly promoted cities does not necessarily translate into better economic performance unless these cities also have the capacity to utilize the additional remit and benefits. We measure city capacity by local government’s human capital level as well as tax extraction ability. We find that incumbent firms located in newly- promoted cities with high capacity tend to outperform firms in similar counties. Interestingly, SOEs in cities with high capacity do not necessarily outperform non-SOEs, indicating that low- capacity cities are more likely to abuse their additional remit to favor certain groups of firms. The WBG and other development institutions are dealing on an increasing basis with subnational governments to improve economic outcomes. In addition, many developing countries have devolved powers to subnational regions. However, there is a lack of evidence about how changing the powers available to local government policy makers relates to economic outcomes. This paper attempts to address this gap and provides some rigorous evidence in support of administrative decentralization accompanied by commensurate increases in capacity. Governments are making employment their main priority and much of job creation – both in modern sector activities and in the informal sector – is overwhelmingly urban. This paper adds to the empirical evidence linking the ability of city governments to implement pro-active policies to actual economic outcomes. It sheds light on how and under what conditions city leaders can utilize the policy instruments at their disposal to actively target firm growth and job creation in cities. The lessons learned in China point overwhelmingly to the importance of local government capacity. To be an effective strategy for economic development, decentralization should to be accompanied by large and commensurate increases in capacity. 21 References Acemoglu, D., García-Jimeno, C., & Robinson, J. A. (2014). State Capacity and Economic Development: A Network Approach (No. w19813). National Bureau of Economic Research. Banerjee, A. V., & Duflo, E. (2004). Do firms want to borrow more?: testing credit constraints using a directed lending program. Centre for Economic Policy Research. Beck, T., Demirguc-Kunt, A., & Maksimovic, V. (2004). Bank competition and access to finance: International evidence. Journal of Money, Credit, and Banking, 36(3), 627-648. Bloom, N., Mahajan, A., McKenzie, D., & Roberts, J. (2010). Why do firms in developing countries have low productivity?. The American Economic Review, 619-623. Brandt, L., Van Biesebroeck, J., & Zhang, Y. (2014). Challenges of working with the Chinese NBS firm-level data. China Economic Review. Cull, R., & Xu, L. C. (2000). Bureaucrats, state banks, and the efficiency of credit allocation: The experience of Chinese state-owned enterprises. Journal of Comparative Economics, 28(1), 1- 31. Davey, K. J. (1993). Elements of urban management. De Mel, S., McKenzie, D., & Woodruff, C. (2008). Returns to capital in microenterprises: evidence from a field experiment. The Quarterly Journal of Economics, 123(4), 1329-1372. Dehejia, R. H., & Wahba, S. (1999). Causal effects in nonexperimental studies: Reevaluating the evaluation of training programs. Journal of the American statistical Association, 94(448), 1053-1062. Fan, S., Li, L., & Zhang, X. (2012). Challenges of creating cities in China: Lessons from a short- lived county-to-city upgrading policy. Journal of Comparative Economics, 40(3), 476-491. Fukuyama, F. (2013). What is governance?. Governance, 26(3), 347-368. Heckman, J. J., Ichimura, H., & Todd, P. E. (1997). Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme. The review of economic studies, 64(4), 605-654. Li, L. (2011). The incentive role of creating “cities” in China. China Economic Review, 22(1), 172-181. Ministry of Civil Affairs. (1993). Guanyu Tiaozheng Sheshi Biaozhun De Baogao (The report on adjusting the criteria for the designation of new cities). Ministry of Civil Affairs of PRC. Nie, H, Jiang, T., & Yang, R. (2012). Zhongguo Gongye Qiye Shujuku De Shiyong Xianzhuang He Qianzai Wenti (Current usage and potential problems of the Chinese industrial enterprises database). World Economics, 5, 142-158. Rosenbaum, P. R., & Rubin, D. B. (1984). Reducing bias in observational studies using subclassification on the propensity score. Journal of the American Statistical Association, 79(387), 516-524. Song, Z. & Hsieh, C. (2013) Grasp the large, let go of the small: The transformation of the state sector in China. Meeting Papers 1330, Society for Economic Dynamics. Wei, S. J., & Wang, T. (1997). The Siamese twins: Do state-owned banks favor state-owned enterprises in China?. China Economic Review, 8(1), 19-29. World Bank. (2009). Guidance note human resource management actionable governance indicators (HRM AGI) instrument. World Bank. Zhang, L., & Zhao, S. X. (1998). Re-examining China's “urban” concept and the level of urbanization. The China Quarterly, 154, 330-381. 22