The Effect of the China Connect

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Effect of the China Connect Chang Ma, John Rogers, and Sili Zhou 2019-087 Please cite this paper as: Ma,Chang,JohnRogers,andSiliZhou(2019). “TheEffectoftheChinaConnect,”Finance and Economics Discussion Series 2019-087. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2019.087. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Effect of the China Connect∗ Chang Ma† John Rogers‡ Sili Zhou§ December 2019 Abstract WedocumenttheeffectonChinesefirmsoftheShanghai(Shenzhen)-HongKong Stock Connect. The Connect was an important capital account liberalization introducedinthemid-2010s. Itcreatedachannelforcross-borderequityinvestmentsinto aselectedsetofChinesestockswhileChina’soverallcapitalcontrolspolicyremained inplace. Usingadifference-in-differenceapproach,andwithcarefulattentiontosampleselectionissues,wefindthatmainlandChinesefirm-levelinvestmentisnegatively affectedbycontractionaryU.S.monetarypolicyshocksandthatfirmsintheConnect are more adversely affected than those outside of it. These effects are stronger for firmswhosestockreturnhasahighercovariancewiththeworldmarketreturnandfor firms relying more on external financing. We also find that firms in the Connect enjoylowerfinancingcosts,investmore,andhavehigherprofitabilitythanunconnected firms. Wediscusstheimplicationsofourresultsforthedebateoncapitalcontrolsand independenceofChinesemonetarypolicy. Keywords: Capital Controls; Global Financial Cycle; Foreign Spillovers; FOMC Shocks;ChinaConnect;CorporateInvestment JELClassification: F38;E40;E52;G15 ∗For comments and discussions, we are grateful to Anusha Chari, Kaiji Chen, Huancheng Du, Andre´s Ferna´ndez,HuashengGao,KindaHachem,HongLiu,LiangJiang,YangJiao,RonKaniel,VictoriaNuguer, QiushaPeng,JunQian,DonghuiShi,SungbinSohn,ShangjinWei,WeiXiong,DanielXu,TongbinZhang andseminarparticipantsatthe8thAnnualWestCoastWorkshopinInternationalFinance,FederalReserve Board, Five-Star Finance Workshop, Fudan University (FISF), Graduate Institute of Geneva, GWU 12th Conference on China’s Economic Development and US-China Relations, Huazhong University of Science andTechnology,IMF,NanchangUniversity,the6thInternationalconferenceontheChineseeconomy: past, presentandfuture,andtheSecondFISF-SSEFinanceWorkshop. ChangMagratefullyacknowledgesfinancial support from the Shanghai Pujiang Program. This project was begun while John Rogers was visiting the Fanhai International School of Finance, Fudan University, whose support and hospitality he is highly appreciativeof. TheMSCIdatainthispaperwasusedunderlicensebyFudanandwasnotprovidedtothe FederalReserveBoard. Theviewsinthispaperaresolelytheresponsibilityoftheauthorsandshouldnotbe interpretedasreflectingtheviewsoftheBoardofGovernorsoftheFederalReserveSystemorofanyother personassociatedwiththeFederalReserveSystem. †FanhaiInternationalSchoolofFinance,FudanUniversity(changma@fudan.edu.cn). ‡InternationalFinanceDivision,FederalReserveBoard(John.Rogers@frb.gov). §FanhaiInternationalSchoolofFinance,FudanUniversity(silizhou@fudan.edu.cn).

1 Introduction The Shanghai (Shenzhen)-Hong Kong “Stock Connect” program, the China Connect, allows investors in mainland China, Hong Kong residents and foreign investors to trade eligible stocks listed on the other market, through the exchange and clearing houses in their home markets. This program, announced in April 2014 and begun in November 2014, is a major step toward internationalizing China’s security markets. In 2016, the program was extended to the Shenzhen exchange. The China Connect is a natural experiment in equity market liberalization that took place amid an overall capital controls policy that remained tight and unchanged. Importantly, the program allows only a set of Chinese firms to be tradedbyforeigninvestors,whiletheremainingfirmsareleftout. TheChinaConnectthus providesauniquesettingtotestthewide-rangingeffectsofstockmarketliberalization. Existing literature has studied the short-run effect of stock market liberalization on the domestic economy (see Chari and Henry (2004, 2008), Bekaert, Harvey, and Lundblad (2005) for example). Yet, few papers investigate its long-run implications and effects on the real economy. One difficulty is identification: when a country liberalizes its stock market,ittypicallyallowsforeignerstotradeallstocks. Aswedescribeindetailbelow,the China Connect liberalization was much different, and in a way that enhances econometric identification.1 Thatsaid,evenacarefullydesignedpolicyexperimentliketheConnectcan still expose the domestic economy to the global financial cycle in the long run, consistent with Rey (2015) and as we document below in the case of China, given the hole it creates inthe“wall”ofcapitalcontrolspolicy. In this paper, we systematically study the effect of the Connect on Chinese firms. We document both long-run exposure to the global financial cycle and short-run benefits such as stock price reevaluation, lower financing costs, and expanded investment. Because the Connectshockcreatedtwogroupsoffirms,wedifferentiatebetweenthecontrolgroupthat remainedprotectedbycapitalcontrolsandtheconnectedtreatmentgroupthatbecamemore opentoforeigninfluences.2 Amajormethodologicalconcern,however,stemsfromthefact that connected firms were not chosen randomly and that choice may not be orthogonal to unobservedfactorsthatalsoaffectfirmequityreturns,financingcosts,andinvestment. We 1Twofeaturesareimportantfortheidentification. First,theChinaConnectonlyincludesselectdomestic stocks. Second, the government liberalizes gradually and does not change the overall tight capital controls policy(seeSongandXiong(2018)andBrunnermeier,Sockin,andXiong(2018)). 2TheConnectisdifferentfromChina’spartialopeningtoforeigninvestmentexaminedbyFernaldand Rogers (2002): the A-share, B-share market, in which different classes of shares in the same firm were allowedtobeheldonlybydomesticandforeigninvestors,respectively. 1

addressthisconcerninanumberofwaysthatleadustobelievethatthelinkbetweenbeing intheConnectandtheresultingfirm-leveloutcomeswedocumentiscausal. The first hypothesis we investigate is that investment by firms in the Connect, with less protection from inland capital controls, will be more sensitive to external shocks than investment by unconnected firms. Our proxy for external shocks is the U.S. monetary policy shocks series of Rogers et al. (2018). We use this along with quarterly firm-level investmentoflistedcompaniesinChina. WefindthatfirmsintheConnectareindeedmore sensitivetoFedmonetarypolicyshocksthanthosenotintheConnect,afterinclusion. The investmentratebytreatedfirmsdeclinesbyanetaverageof2.8%followingaunitincrease in the shock, controlling for firm-level investment opportunity, cash flow, size, and local economicconditions. Thisresultisrobusttoabatteryoftests. OursecondhypothesisconcernsthechannelthroughwhichU.S.monetarypolicyshocks affectdomesticinvestment. Wefindthatfirmswhosestockreturnco-movesmorewiththe global market return are affected more by these shocks after inclusion in the Connect. Furthermore, firms relying more on external financing are more sensitive. These findings are consistent with a risk-premium channel. Given that the China Connect permanently changes the cost of funding for connected firms, firms whose stock returns have a higher covariance with the global market return are more responsive when U.S. monetary policy changes the risk-free rate and global risk-aversion. These effects transmit to the real economybyalteringfirminvestmentdecisions. Forfirmswhosereturnco-movesmorewiththe global market return, their investment is more responsive because their risk-premiums are moresensitive. Ourlasthypothesisisthat,ifthisincreasedsensitivityofChinesecorporateinvestment to external shocks were the only effect of the Connect, firms would act to remain out of the Connect, something we do not observe. We establish that firms in the Connect have a higher stock price reevaluation and thus invest more than unconnected firms in the short run. Furthermore, they also enjoy lower financing costs, and earn higher net income on equity (ROE) and assets (ROA), relative to firms outside of the Connect. These findings areconsistentwithpreviouspaperssuchasChariandHenry(2004,2008). Literature Review We contribute to several strands of literature. First, the literature on global financial cycles. For example, Rey (2015) and Miranda-Agrippino and Rey (2019) providecompellingevidencethataglobalfinancialcyclemightleadassetpricesandfinancialvariablestoco-moveacrosstheglobe. Moreover,theyarguethatU.S.monetarypolicy 2

is the driving force. Meanwhile, many papers have focused on the channel through which the global financial cycle can affect the local economy (see di Giovanni et al. (2017)). Cerutti et al. (2019) challenge the importance of the global financial cycle in explaining variations in capital flows, however. We also study the spillover effects of U.S. monetary policyshocksinthepresenceofcapitalcontrols. Second, our paper is related to the literature on capital controls. One conclusion from theGlobalFinancialCyclesliteratureisthatcapitalcontrolscancreateausefulwallagainst external shocks (see IMF (2012), Jeanne et al. (2012), Rey (2015) and Miranda-Agrippino and Rey(2019)). The empiricalevidence forthe effectiveness ofcapital controlsis mixed, however (Magud et al. (2018), Rebucci and Ma (forthcoming) and Erten et al. (forthcoming)). One difficulty is that the policy is usually endogenous and sticky: many countries put capital controls in place simultaneously with adverse events and do not change them frequently.3 For example, Forbes et al. (2015) find that most capital flows management measuresdonotsignificantlyachievestatedobjectivesofexchangeratemanagement,capital flows management, monetary policy independence, and taming volatility. However, Miniane and Rogers (2007) and Han and Wei (2018) do find evidence that capital controls buffer the spillover effects from U.S. monetary policy to emerging market exchange rates and interest rates, while Ostry et al. (2012) and Bruno et al. (2017) find some supporting evidence for the effectiveness of capital controls on bank credit.4 One key difference in our paper lies in the identification of the policy shock. The Connect program is unlike nationwide capital control reforms documented in other papers (Henry (2000a,b, 2003) and Bekaert et al. (2005) for example), and is thus a cleaner policy experiment from which we canestablishcausalrelationships. Third, our paper is related to the literature on corporate investment and macro uncertainty. For example, Ottonello and Winberry (2018) document an investment channel of U.S. monetary policy and find that firms with low default risk are the most responsive to monetary policy shocks. Husted et al. (forthcoming) find that monetary policy uncertainty significantly delays U.S. firm investment in ways that are in line with both real options theory and a financial frictions channel. Consistently, we also find that Chinese corporate investment is negatively affected by contractionary U.S. monetary policy shocks. Differ- 3AnexceptionisBrazil(seeAlfaroetal.(2017)whostudytheeffectofcapitalcontrolsinBrazil). 4Arelativelynewliteraturejustifiestheuseofcapitalcontrolstoaddresspecuniaryexternalitiesoraggregatedemandexternalities. Forpecuniaryexternalities,seeLorenzoni(2008),JeanneandKorinek(2018, 2010a), Bianchi (2011), Korinek (2018), Benigno et al.(2013) andMa (forthcoming). For papers withaggregatedemandexternalities,seeKorinekandSimsek(2016)andFarhiandWerning(2016). 3

ently, we document a reduction in corporate investment for connected firms relative to unconnectedonesfollowingacontractionaryFOMCshock. Ourresultsprovideadditional evidence,derivedfromacleanidentification,ontheeffectsofcapitalaccountpolicies. Fourth,ourpaperisrelatedtotheliteratureestablishingpositiveeffectsofstockmarket liberalizations. For example, Henry (2000a,b, 2003), Chari and Henry (2004, 2008) and Bekaert et al. (2005) document positive long-run growth effects for liberalizing countries. Consistent with this, we find a positive effect of China’s stock market liberalization on Chinese corporate investment. Differently, we provide a more comprehensive analysis of theliberalizationonthecorporatesectorunderourpolicyexperiment. Finally, our paper belongs to the literature using the China Connect as a natural experiment to test theoretical predictions. For example, Xing et al. (2018) use the Connect to test the impact of capital market openness on high frequency market quality. Similarly, Liu et al. (2018) use the policy shock to test the speculative nature of beta and the multipliereffectofspeculationondemandshocks. Differentfromthosepapers,wealsoanalyze macroeconomictransmissionandstudybothrealandfinancialeffectsoftheConnect. Policy Implications As is well known, China has imposed very strict capital controls (see Figure1). Despitethis,ChinesepolicymakersinitiatedtheConnect. Tradingunderthisprogram is subject to a maximum cross-border investment quota together with a daily quota. IthasbeenarguedthattheConnectisawell-designedcontrolledcapitalaccountliberalization (Prasad (2017)), which presumably should minimize the impact of external shocks to domestic Chinese sectors. Our results indicate that even such a carefully designed policy experiment can expose domestic listed firms to external shocks. The findings in this sense support the use of capital controls in curbing external shocks. However, our results also point to many positive effects that firms enjoy from inclusion in the Connect. Overall, this suggests that firms are able to hedge the negative consequences from increased sensitivity toforeignshocksunderthiscarefullycalibratedliberalization. In the next section, we describe the institutional background of the Connect. Section 3 developsourmainhypothesesthroughasimpletheoreticalframework. Section4describes our data and key variables construction. Section 5 discusses estimation strategy, including how we address sample selection issues, and presents the baseline empirical findings on firm investment. Sections 6 and 7, respectively, present results from firm heterogeneity on thebaselineandresultsonthe“positive”effectsoftheConnect. Section8concludes. 4

Figure1ChineseCapitalAccountRestrictions PanelA:DejureMeasure PanelB:DefactoMeasure 1 1 Chinn-Ito (left) 110 FKRSU (right) 0.95 0.5 100 0.9 90 0 0.85 80 -0.5 0.8 2005 2010 2015 2005 2010 2015 NOTE:PanelAplotsdejuremeasuresofcapitalcontrolsfromChinnandIto(2006)andFerna´ndez,Rebucci, and Uribe (2015). A higher value for the former (latter) means a higher (lower) degree of capital account openness. Panel B plots the de facto measure, the sum of gross stocks of foreign assets and liabilities as a ratiotoGDP,fromLaneandMilesi-Ferretti(2007). 2 Institutional Background China’s two domestic stock exchanges, the Shanghai Stock Exchange (SHSE) and ShenzhenStockExchange(SZSE),wereestablishedinDecember1990andApril1991,respectively. Their A share markets combined are the second largest in the world in total market capitalization, trailing only the US. The number of listed firms has been growing since marketinception,withmorethan3,500firmslistedandtradedattheendof2018. Foreign investors were traditionally restricted from trading in the A-share market. After the Asian financial crisis, the China Securities Regulatory Commission (CSRC) has taken a gradual and prudential approach to opening the financial markets (see Prasad and Wei (2005) and Song and Xiong (2018)). The CSRC first introduced a B-share market exclusively to foreign investors in 2001. One year later, the Qualified Foreign Institutional Investor (QFII) program was initiated to certain overseas institutional investors, which allowed limited access to A-share stocks. However, getting QFII licences was extremely difficult, requiring applicants to meet certain standards for financial stability concerns. In thefirstyear,only12qualifiedforeigninvestorswereapprovedandapprovalceasedduring 5

2006-2007.5 There are also restrictions on domestic residents purchasing overseas stocks. However,beginningin2006,domesticinstitutionalinvestorshavebeenallowedtopurchase foreignstocksundertheQualifiedDomesticInstitutionalInvestor(QDII)program. The Shanghai (Shenzhen)-Hong Kong Stock Connect was first proposed in 2007 by the Binhai New Area of Tianjin and the Bank of China. However, regulators postponed the program until on April 10, 2014, the CSRC and Hong Kong Securities and Futures Commission (SFC) made a joint announcement to start the program. The plan was to includeallforeigninvestorsaswellasanymainlandinvestorswhohaveastockaccountwith balances no less than 500,000 RMB (approximately 72,000 USD), regarded as a relatively low barrier to enter both markets.6 The Connect was officially launched on November 17, 2014. Unlike both QFII and QDII, which have a relatively small size and only focus on institutional investors, the China Connect is larger and includes both institutional and retail investors. In December 2016, the Shenzhen Stock Exchange was also opened to the Hong Kong Stock Connect. The Shenzhen Exchange includes both growth and high-tech startupfirmslikeChiNext. Overall,morethanonethousandstocksfromthemainlandhave becomeconnectedtooverseasinvestors,includingbothlarge-capandmid-capstocks. AlthoughtheConnectisalooseningofcapitalaccountrestrictions,tradingthroughthe program is nevertheless subject to aggregate quotas. The daily quota of trading capitalization is 13 billion RMB for the Shanghai Exchange and 10.5 billion RMB for the Hong Kong Exchange. On April 11, 2018, the daily quota increased four-fold to 42 billion and 52billion,respectively. Moreover,shortsellingthroughtheConnectisbanned. There were two big waves of the Connect program. For the Shanghai-Hong Kong Connect, eligible stocks include all the constituent stocks of the SSE 180 Index, SSE 380 Index, and all the SSE-listed A shares that are not included as constituent stocks of the relevant indices but which have corresponding H shares listed on SEHK (so called “A- H” dual listed stocks), except for SSE-listed shares which are not traded in RMB and SSE-listed shares which are under risk alert (including shares of “ST companies”, “*ST companies companies” and shares subject to the delisting process under the SSE rules). Similarly, for Shenzhen-Hong Kong, eligible stocks include all constituent stocks of the SZSE Component Index, SZSE Small/Mid Cap Innovation Index which have a market 5Detailed comparison between the QFII/QDII and Stock Connect can be found at: http://english. sse.com.cn/investors/shhkconnect/introduction/comparing. 6Detailed information can be found on the website of the Hong Kong Stock Exchange. https: //www.hkex.com.hk/-/media/HKEX-Market/Mutual-Market/Stock-Connect/Getting-Started/ Information-Booklet-and-FAQ/Information-Book-for-Investors/Investor_Book_En.pdf 6

capitalizationofnotlessthanRMB6billionandalltheSZSEA-Hduallistedstocks,except for SZSE-listed shares which are not traded in RMB and for SZSE-listed shares which are under risk alert (including shares of “ST companies”, “*ST companies companies” and sharessubjecttothedelistingprocessundertheSSErules)orunderdelistingarrangement.7 Eligiblesecuritiesareincludedandexcludedbasedonadjustmentsmadetotheindexesand thetimingatwhichrelevantAsharesareplacedunderriskalertorreleasedfromriskalert. Theauthoritymakesadjustmentssemi-annually,basedonthesecriteria. Table A.1 shows the timeline of the Connect program. On November 17, 2014, the Shanghai-Hong Kong Stock Connect was made effective, with 416 constituent stocks in the SSE 180 index, SSE 380 index, and A-H dual listed stocks eligible for the Program. The list was revised slightly due to adjustment of the 180 and 380 index. On December 5, 2016,theprogramwasexpandedtoShenzhen,with676stocksfromtheSZSEComponent Indexonadesignatedlisteligibleforoverseasinvestors.8 3 Theoretical Motivation and Hypothesis Development 3.1 A Simple Conceptual Framework Our framework combines insights from both the literature on financial liberalization and the global financial cycle. Following the standard neoclassical production framework, e.g. Chari and Henry (2004, 2008), the optimal investment decision for firm i equates the marginal benefit of production to the cost of funding. Stock market liberalization has no impact on the marginal benefit of production since it is determined by production technology. However,fundingcostschangewithliberalization. Asaresult,theglobalfinancial cyclecanhaveanimpactoninvestmentthroughitsimpactonthefundingcostafterliberalization. In a world with efficient markets, the funding cost for firm i should equal its stock return. Specifically,thefirst-orderconditionafterliberalizationcanbewrittenas E[f(cid:48)(k∗)]=r∗+γ∗cov(r,r ) (1) i i i W 7Detailed information is found from the Hong Kong Stock Exchange at https://www.hkex.com.hk/ Mutual-Market/Stock-Connect/Getting-Started/Information-Booklet-and-FAQ?sc_lang=en. 8Originally,therewere537(856)connectedstocksfromShanghai(Shenzhen). Followingtheliterature, wedropsomefirms(asdetailedinSection4),suchthatwehave416(676)firmsintheend. 7

where f (·) is a concave production function like Cobb-Douglas, k∗ is capital per unit of i i effectivelabor(totalcapitalstockdividedbytotaleffectivelabor),r∗ istheglobalrisk-free rate, γ∗ is the risk-aversion for the marginal investor, and cov(r,r ) is the covariance bei W tweentheequityreturnr forfirmiandtheglobalmarketreturnr (ignoringdepreciation). i W U.S. monetary policy, the crucial source of transmission emphasized by the global financialcycleliterature,canaffecttheriskaversionofglobalinvestorsandthushaveanimpactontheglobalmarket(seeKalemli-Ozcan(2019),Miranda-AgrippinoandRey(2019)). Therefore, according to our simple framework, there should be two effects through which the global financial cycle can affect domestic investment after liberalization. When U.S. interest rates rise, (1) the global risk free rate r∗ increases and (2) the global risk-aversion coefficientγ∗ becomeshigher. Bothleadtoareductionindomesticinvestment. Asaresult, Chinese firm investment should be differently affected by U.S. monetary policy after the Connect, depending on inclusion. Other implications emerge from this framework. The riskfreeratechannelreflectsacommonshocktoallstocksaftertheConnect. Ontheother hand, the risk-aversion channel is firm-specific and depends on (1) whether the firms can be traded by overseas investors and (2) how sensitive the stock returns are to the global systematicriskfactor,measuredbycov(r,r ).9 i W Furthermore, and importantly, those effects should be absent / weaker before the Connect since the cost of funding is unaffected (less affected) by U.S. monetary policy. One canseethisfromtheinvestmentdecisionbeforetheConnectasfollows E[f(cid:48)(k)]=r+γcov(r,r ) (2) i i i M where k is capital per unit of effective labor, r is the domestic risk-free rate, γ is the riski aversion for the domestic marginal investor, and cov(r,r ) is the covariance term of the i M equityreturnr forfirmiandthemarketreturnr forthedomesticmarket. i M Two implications follow. First, the domestic risk-free rate should be less sensitive to U.S. monetary policy because China has imposed very tight capital controls, as shown in Figure 1 (see Han and Wei (2018)). Second, it is less likely that domestic investors’ risk aversionwillbeaffectedbytheglobalfinancialcyclebeforetheConnectsincecapitalcontrols policy prevents them from accessing international financial markets. As a result, one shouldnotexpectanysignificantimpactfromU.S.monetarypolicytodomesticinvestment beforetheConnect(barringleakagesinoverallcapitalcontrols). 9Thislogicissimilartotherisk-sharingchannelidentifiedinChariandHenry(2004,2008). 8

Capital controls thus play the role of a “wall” between the domestic economy and the international market, reducing the impact of the global financial cycle on the domestic economy. With the introduction of the China Connect, domestic investment is more sensitive to the cycle due to the funding cost channel: for connected firms, their investment shouldbemoresensitivetotheU.S.monetarypolicyshockcomparedwithboththeunconncetedfirmsandthemselvespriortoinclusionintheConnect. In addition to increasing the sensitivity of domestic investment to the global financial cycle, liberalization can bring benefits in the short run through the risk-sharing channel, as in Chari and Henry (2004, 2008). To see this, assuming that γ =γ∗, one can write the impact of stock market liberalization on investment as follows. Subtracting equation (1) fromequation(2), ∆E[f(cid:48)(k∗)]≡E[f(cid:48)(k)]−E[f(cid:48)(k∗)]=r−r∗+γDIFCOV (3) i i i i i i i where DIFCOV = cov(r,r )−cov(r,r ) is a measure of risk-sharing as in Chari and i i M i W Henry (2004, 2008). Testable predictions for investment and equity prices emerge from equation (3). The Connect enables international investors to trade domestic stocks, which ultimately leads to stock price revaluation and thus investment. Specifically, there are two factors that change with the liberalization: one is a common factor, i.e. the risk-free rate r−r∗, and the other is a firm-specific risk premium component, measured by γDIFCOV. i GiventhattheConnectchangestherisk-freeratepermanently,itcanaffectbothconnected and unconnected firms. As for the risk premium, however, it affects the connected and unconnected firms differently. Presumably, firms in the connect are more affected than the unconnected ones because those firms are now priced by a new systematic risk factor, the global market return, while unconnceted firms are still priced by the domestic systematic riskfactor,i.e.,thedomesticmarketreturn. Furthermore,firmswithahighDIFCOVshould experienceagreaterrepricingafterliberalization,otherthingsequal. 3.2 Hypothesis Development We form our hypotheses based on the simple conceptual framework above. As seen in the Chinn-Ito index of countries capital account restrictions (see Figure 1), China has imposed a very tight and persistent capital controls policy. Capital controls measures from Ferna´ndez et al. (2015) confirm this characterization of policy, albeit with a small relaxation after 2014. De facto capital account restrictions, as measured by the sum of gross 9

stocksofforeignassetsandliabilitiesasaratioofGDPindicateanupwardtrendforChina starting from the early 2000s, with fluctuations around 100 after 2010. That China’s overall capital controls policy has not changed significantly in recent decades implies that the impactoftheglobalfinancialcycleonthedomesticeconomybeforetheConnectshouldbe minimal. However,theConnectcreatedachannelthroughwhichtheglobalfinancialcycle can affect the domestic economy, via the cost of funding channel. The absence of a sharp changeintheabovedefactomeasureofcapitalcontrols,despitetheConnect,isconsistent withtheinitialintentionofthepolicy: reducingexcessivecapitalflowsandopeningpartof the stock market to foreign investors. To the extent that controls are effective, there should besmallerexternalspillovereffectsonfirmsthatarenotintheConnectandhencefunction more under the protection of capital controls. If controls are not effective, there should notbesignificantdifferencesbetweenconnectedandunconnectedfirmsintheirinvestment responsestoexternalshocksaftertheconnection. Thus,ourfirsthypothesis: Hypothesis 1. Firms included in the Connect program become more sensitive to external shocksthanunconnectedfirms,aftertheConnect. We further investigate which types of firms are more sensitive to external shocks after the Connect. According to our conceptual framework, firms’ investment should be more sensitivetoexternalshockswhentheyhavegreaterrisk-sharingwiththeglobalmarket,i.e. ahighercov(r,r ). Thisleadstothecorollarytoourfirsthypothesis: i W Hypothesis 2. Firms with relatively higher sensitivity to the global market (i.e., higher cov(r,r )) in the Connect program have more sensitive investment expenditures to exteri W nalshocksaftertheConnect. Finally, we hypothesize that if the only effect of the Connect were that Chinese firms’ investment became more sensitive to external shocks, firms would behave so as to remain outoftheConnect. Weareunawareofanysuchbehavior,andthusconjecturethat: Hypothesis 3. Firms included in the Connect experience positive effects, such as a higher stock price response and higher investment boom, after the Connect. Moreover, these effectsarestrongerforfirmswithahigherrisk-sharingmeasure(i.e. ahigherDIFCOV). i As noted above, we test these hypotheses with a detailed data set and difference-indifferenceestimationandcarefulconsiderationofsampleselection. Wefindstrongsupport forallthreehypotheses. 10

4 Data We combine data from two main sources. The first is the U.S. monetary policy shock of Rogers et al. (2018). The second is firm-level data from the China Stock Market and AccountingResearch(CSMAR)Database. 4.1 U.S. Monetary Policy Shock Rogers et al. (2018) construct a Fed monetary policy shock series (MPSUS) that is a combination of three surprises: Target Fed Funds rate surprises, which were zero between December 2008 and December 2015; Forward Guidance surprises; and Large Scale Asset Purchasesurprises(zerobeforetheQE1program). Thisisahigh-frequencysurpriseseries, measuring changes in yields from 15 minutes before the Federal Open Market Committee (FOMC) announcement to 30 minutes afterward.10 The MPSUS series begins in January 1990 and ends in December 2017.11 During this period, the 250 shocks have a mean of −0.022 and standard deviation of 0.119. To match the US monetary policy shock with our quarterly firm data, we aggregate the MPSUS within each quarter in two ways, as in Ottonello and Winberry (2018). One is a simple sum of the (typically two) surprises that occureachquarter. Theideaistocapturethecumulativeamountofmonetarypolicyshocks in a given quarter. Recognizing the slow adjustment of corporate investment decisions, we alsouseavalueweightedsumtoconstructthequarterlyMPSUS,wheretheweightisgiven by the number of days remaining in the quarter after FOMC announcement day. We estimate all of our regressions using both shock series. Because results are highly robust to thealternativedefinitions,wefeaturesimpleaggregationofFOMCsurprisesinourtable.12 ThesummarystatisticsofthemonetarypolicyshockseriesarereportedinTableA.2. 4.2 Firm-level Variables We collect firm-level data from the China Stock Market and Accounting Research (CS- MAR) Database. Our sample starts at the time all A-share stocks were traded on the Exchanges. B-sharestocksareexcludedbecausetheycanonlybetradedbyforeigninvestors. 10Theseriesalsoincludesahandfulofinter-meetingannouncements. Seetheoriginalpaper(orWright’s website)fortheunderlyingdataanddetailsonconstructionofthesurprises. 11We use the Eastern U.S. time zone, a half-day behind the Chinese time zone. This is not an issue for ouranalysisofquarterlydata. 12Resultsusingvalueweightedsurprisesthatwedonotdisplayhereareavailableuponrequest. 11

As is conventional, we drop financial and utility firms since they share different disclosure regulations and their liquidity positions are special compared with firms in other sectors. Following the literature, we require firms to have at least two years of historical data as in Fama and French (1993). We exclude firms listed after year 2014 to get rid of the effect of new IPOs. Our sample period runs from 2002 to 2017, with the beginning date chosen to reflect when the CSRC required all listed firms to file quarterly financial reports.13 We drop observations with missing key values for investment, Tobin’s Q or cash flow. The finalsamplecomprises87,740firm-quarterobservations,covering2,174uniquefirms. The detaileddistributionbyindustryandyearcanbefoundinTableA.4oftheAppendix. Our main measure of firm-level investment is defined as capital expenditures divided by beginning-of-quarter book value of total assets (lagged total assets), where the capital expenditures are calculated as cash payments for the acquisition of fixed assets, intangible assetsandlong-termassets(fromthecashflowstatement)minuscashreceiptsfromselling those assets, plus cash paid for operating lease.14 We control for an array of firm-level characteristics that might affect corporate investment (see Julio and Yook (2012) and Cao et al. (2016) for example). The key control variables include Tobin’s Q, calculated as the book value of total assets minus the book value of equity plus the total market value of equity (close price at quarter end multiple by share outstanding) scaled by book value of total assets; size, the natrual logarithm of the book value of total assets; cash flow, measured by earnings before interest and taxes (EBIT) plus depreciation and amortization minus interest expenses and taxes scaled by lagged total assets; and sales growth, defined as the growth rate of revenue. We winsorize our sample at the top and bottom 1% of all continuous variables to guard against outliers. The details of variable construction are reportedinTableA.5oftheAppendix. Table A.3 reports summary statistics for the firm characteristics. Quarterly capital expenditure is 3.5% on average, with a standard deviation of 4.5%, slightly higher than for U.S.listedfirms(seeJens(2017)). Tobin’sQis2.624onaveragewithastandarddeviation of 1.94. Size is 21.781 on average with a standard deviation of 1.275. The mean of cash flow is 0.036 with a standard deviation of 0.046. Sales growth is 0.413 on average with a standard deviation of 0.8. All statistics are consistent with previous studies on China (see Caoetal.(2016)forexample). 13The announcement date is April 6, 2001 and became effective in 2002. Detailed information can be foundat: http://www.gov.cn/gongbao/content/2002/content_61983.htm. 14Ourmeasureofinvestmenttoassetratioisequivalenttocapitalexpenditure(Compustatdataitem#128 CAPX)whichiscommonlyusedinU.S.basedstudies. 12

5 Estimation Strategy and Firm-Level Investment Results Our objective is to identify the average effect of the Connect on outcomes such as investment, equity returns, and financing costs for Chinese firms that were included in the program,i.e.,theaverageimpactoftreatmentonthetreated. Specifically,weareinterested incomparing,e.g.,investmentofconnectedfirmstothecounterfactual—investmentofunconnected firms at the same point in time. Conceptually, we would like firms to have been randomly assigned to the Connect and compare the average outcomes of the two groups. Absentthat,weuseadifference-in-differencesmethodthatmimicsarandomselectionhypothetical under reasonable conditions.15 This compares the change in outcomes in the treatment group before and after the Connect announcement to the change in outcomes in the control group. By comparing changes, we control for observed and unobserved firm characteristics that might be correlated with the Connect decision and with the outcomes. Thechangeinthecontrolgroupisanestimateofthetruecounterfactual: whatwouldhave happenedtothetreatmentgroupiftherehadbeennoConnect. 5.1 Estimation Strategy: Difference-in-Differences Weutilizethefollowingaugmentedversionofthestandardinvestment-Qspecification. Y =α +α +β Connect +β MPSUS+β MPSUS×Connect +ΓZ +ε (4) it i s 1 it 2 t 3 t it it it whereiindexesthefirmandt isatimeindex(quarterlyfrequency). Thedependentvariable is corporate investment Y , defined as quarterly capital expenditure scaled by book value it of total assets at beginning of the quarter. α is a firm fixed effect and α is a year fixed i s effect. The explanatory variables of interest are MPSUS, Connect , and their interaction. t it Weconsiderbothequalweightedandvalue(date)weightedquarterlyMPSUS asdescribed t above. In our regressions, Connect is a dummy variable indicating whether firm i is init cludedintheConnectprogramatquartert. Firmscanbeincludedorexcludedperiodically, as explained above, and there is often a lag between the announcement date and effective 15A major concern is that firms that were chosen to be connected could be different from those that remainedoutside,andthatthesedifferencesarecorrelatedwithoutcomeslikefinancingcostsandinvestment sensitivitytoforeignshocks. Forexample,politicallyconnectedfirmsforwhichfinancingcostsarealready low(er) may have been the ones that lobbied for inclusion. In principle, many of the (unobservable) characteristics that may confound identification are those that vary across firms but are fixed over time. Our difference-in-differencesmethodofcontrollingforthisunobservedheterogeneityisconventional. 13

date for a firm to be included (see Table A.1). Thus, we make the dummy 1 (0) for all quarters of the year in which the firm is first included in (removed from) the Connect.16 The controls Z include both firm-level and macro-level variables that could potentially it affect corporate investment decisions. Following the literature, we use lagged Tobin’s Q, cashflows,salesgrowthandfirmsizeatthefirmleveltocontrolforfirmheterogeneity. We also use the quarterly change of nominal GDP at the provincial level to control for local economic cycles, with the firm’s headquarter address identifying its location.17 We add both firm and year fixed effects to control for unobserved individual and year effects, and quarterly dummies to adjust for seasonality. Standard errors are clustered at both firm and year level (see Petersen (2009)). To control for regional time-variation, we also include interaction terms between regions and year indicators as an alternative specification and findthatourresultsarerobust.18 Thoseresultsarehighlyrobustandavailableonrequest. 5.2 Validity of Empirical Strategy Webeginbyevaluatingthevalidityofourdifference-in-differencesregressionframework. Tothisend,weevaluatesampleselectionandconductaparalleltrendstest. 5.2.1 SampleSelection Unsurprisingly, firms in the Connect were not chosen randomly, as would be ideal for our econometric objectives. Instead, firms were selected based on whether they belong to the constituent indexes, as described above. Table 1 provides a comparison of ex-ante observable differences between connected and unconnected firms for the two big waves of the Connect, for twelve variables: Investment, Size, Tobin’s Q, Cash Flow, Sales Growth, MarkettoBookratio,Cashholdings,Age,SalesGrowth,GlobalCov(thehistoricalcovariance of firm i’s stock return with the MSCI world market return), DIFCOV (the difference 16Our results still hold if we don’t make this adjustment. We prefer the adjustment for an additional reason. The periodic in-and-out of the Connect is due to adjustment of the stock indices that are typically doneinJuneorDecembereachyear(selectioncriteriacanbefoundattheofficialwebsiteoftheShanghai andShenzhenStockExchanges). Theannouncementofinclusionandexclusioncanhappenseveralmonths beforeimplementation. OuradjustmenttotheConnectdummycapturesthisannouncementeffect. SeeTable S.3 and S.4 in the Online Appendix when we 1) do not do this adjustment; 2) eliminate all the periodic changestotheindexesandonlykeepthetwobigwavesin2014Q4and2016Q4. 17In Table S.5 of the Online Appendix, we also include lagged year-over-year M2 growth rate and the 7-dayReporateinChinatocontrolforChinesemonetarypolicy. Ourmainresultsarerobust. 18Geographic regions in China can be classified into six areas based on the National Census Bureau: Bohai,Central,Northeast,Northwest,Southeast,Southwest. Weusefirmheadquarterstoidentifyregion. 14

Table1SummaryStatistics: Connectedvs. UnconnectedFirms (1) (2) (3) (4) (5) (6) (7) (8) Connected(a) Unconnected(b) Difference(a)-(b) Mean Median S.D. Mean Median S.D. MeanDiff T-test PanelA:OneQuarterbeforeShanghai-HongKongConnect(2014Q3) Investment 0.035 0.026 0.032 0.023 0.013 0.030 0.012 *** 5.09 Size 23.104 22.969 1.337 21.750 21.763 1.180 1.354 *** 14.26 Tobin’sQ 1.756 1.463 1.049 2.332 1.594 2.045 -0.575 ** -4.96 CashFlow 0.036 0.030 0.032 0.007 0.007 0.038 0.029 *** 11.43 M/B 2.848 2.248 2.168 4.995 2.887 6.082 -2.147 *** -6.60 Cash 0.152 0.124 0.099 0.143 0.110 0.117 0.009 1.15 Age 12.821 13.000 5.498 14.467 15.000 4.773 -1.646 *** -4.24 Salesgrowth 0.538 0.522 0.145 0.546 0.510 0.220 -0.008 -0.59 GlobalCov% 0.068 0.069 0.057 0.069 0.065 0.060 -0.001 -0.15 DIFCOV% 0.317 0.313 0.108 0.349 0.346 0.104 -0.032 *** -4.03 ReturnVolatility 0.020 0.019 0.007 0.021 0.020 0.006 -0.001 *** -2.96 MarketCap 23.171 23.011 0.856 22.189 22.057 0.616 0.981 *** 16.87 PanelB:OneQuarterbeforeShenzhen-HongKongConnect(2016Q3) Investment 0.032 0.021 0.033 0.025 0.016 0.029 0.007 *** 4.04 Size 22.476 22.342 0.991 21.545 21.525 0.870 0.932 *** 17.65 Tobin’sQ 3.724 3.048 2.548 3.692 3.049 2.617 0.032 0.22 CashFlow 0.039 0.034 0.038 0.020 0.019 0.034 0.019 *** 9.38 M/B 4.788 4.033 3.106 5.158 4.215 3.962 -0.370 * -1.84 Cash 0.181 0.134 0.148 0.165 0.133 0.130 0.017 ** 2.15 Age 9.881 7.000 5.808 9.567 6.000 6.088 0.314 0.94 Salesgrowth 0.576 0.552 0.172 0.589 0.554 0.203 -0.013 1.25 GlobalCov% 0.130 0.131 0.078 0.127 0.131 0.083 0.004 0.82 DIFCOV% 1.180 1.071 0.513 1.220 1.129 0.501 -0.040 -1.40 ReturnVolatility 0.020 0.019 0.005 0.022 0.022 0.006 -0.002 *** -7.49 MarketCap 23.324 23.217 0.598 22.484 22.430 0.391 0.840 *** 28.39 NOTE:summarystatisticsofkeyvariablesforconnectedandunconnectedfirmsusedinoursample.Detailed definitions can be found in Appendix A.5. Panel A includes firms only listed on the Shanghai Exchange in 2014 Q3. Panel B includes firms listed on Shenzhen Stock Exchange in 2016 Q3. All variables are winsorizedatthetopandbottom1%. ∗, ∗∗ and∗∗∗ indicatestatisticalsignificanceatthe10%, 5%, and1% level,respectively. between the historical covariance of firm i’s return with local market and its covariance with the MSCI world market), Return Volatility and Market Cap. See Appendix A.5 for datasources. As seen in Table 1, connected firms invest more, are larger, younger, and have lower stockreturnvolatility. ThisheterogeneitymotivatesustoinvestigateaHeckmanTwo-Stage estimationandpropensityscorematchingmethodtocontrolforobservedfirmheterogeneity. Private conversations with a governor at the SSE suggest that the authorities select stocks into the Connect primarily based on the composite indexes. However, there is no simpleruleforconstructingsuchindexesthatwecouldmechanicallyplugintoanempirical 15

selection model. From our reading of the public information concerning index construction and the ex-ante firm differences in Table 1, we include stock return volatility, market cap, leverage, age and dividend payout decision in our first stage Probit model regression, controllingforindustry,provinceandexchangefixedeffects. 5.2.2 ParallelTrendsAssumption The validity of difference-in-difference estimation relies on the parallel trends assumption: before the Connect, treated firms exhibit a similar pattern of investment sensitivity to MPSUS as control firms. To test this, we introduce seven dummies, Connect (-3), Connect (-2), Connect (-1), Connect (0) (the year when Connect Program was effective), Connect (1), Connect (2) and Connect (3+), to flag the years relative to the effective year. For example, Connect (3+) refers to years beyond three years after the connection. We then re-estimate our baseline regression by replacing the Connect dummy with these seven indicatorsandinteractthemwithMPSUS shocks. Iftheparalleltrendsassumptionholds,we should expect that interaction terms with Connect (-3), Connect (-2), Connect (-1) have a relativelysmallermagnitudeandlesssignificancethantheotherinteractionterms. Table2reportstheregressionresultsandFigure2displaysthecoefficientsfromcolumn (2).19 The coefficients on the interaction term between pre-trend dummies (i.e. Connect (- 3),Connect(-2),Connect(-1))andMPSUSareclosetozeroandnotstatisticallysignificant, satisfying the parallel trends assumption. These results have three implications. First, the Shanghai (Shenzhen)-Hong Kong Connect could not be anticipated by the treated firms. Furthermore, even though some firms might be able to anticipate the possible outcome after the Connect, they cannot react before the Connect actually took place. Second, the negative response of corporate investment to the MPSUS only shows up after the Connect. Furthermore,thecoefficientsontheinteractionbetweenMPSUS andConnect(0)(Connect (1)) are statistically significant. The coefficients on the interaction term between Connect (3+)andMPSUSaretwicelargerthantheinteractiontermbetweenConnect(1)andMPSUS, suggestingthattheeffectofMPSUS oncorporateinvestmenttakestimetomaterialize. Our findings indicate that the effect of U.S. monetary policy shocks on corporate investment is bothnegativeandlonglastingforconnectedfirms. 19Becauseresultsthroughoutarerobusttothecalculationofthemonetarypolicyshock,wedisplayresults usingonlytheequalweightedmeasureofMPSUS. ResultsusingvalueweightedMPSUS areavailableupon request. 16

Table2ParallelTrendsAssumption Investment (1) (2) MPSUS*Connect(-3) -0.001 0.003 (0.004) (0.006) MPSUS*Connect(-2) -0.002 -0.005 (0.006) (0.007) MPSUS*Connect(-1) -0.004 -0.001 (0.005) (0.005) MPSUS*Connect(0) -0.024*** -0.019** (0.009) (0.008) MPSUS*Connect(1) -0.016** -0.016** (0.007) (0.007) MPSUS*Connect(2) -0.050*** -0.055*** (0.012) (0.013) MPSUS*Connect(3+) -0.034*** -0.030*** (0.009) (0.009) MPSUS -0.007* -0.011** (0.004) (0.004) Connect 0.002* 0.001 (0.001) (0.001) Size 0.005*** (0.001) LagTobin’sQ 0.002*** (0.000) CashFlow 0.173*** (0.011) SalesGrowth 0.001*** (0.000) GDPGrowth 0.024 (0.015) FirmFE Yes Yes YearFE Yes Yes QuarterDummy Yes Yes Observations 87740 87740 AdjustedR2 0.387 0.410 NOTE. The dependent variable is corporate investment, defined as quarterly capital expenditure scaled by thebeginning-of-quarterbookvalueoftotalassets. WeusesevenConnectdummiestointeractwithMPSUS, Connect(-3),Connect(-2),Connect(-1),Connect(0)(theyearwhenConnectProgramwaseffective),Connect (1), Connect (2) and Connect (3+), to flag the years relative to the effective year. Other firm level controlscanbefoundatA.5. Allstandarderrorsareclusteredatbothfirmandyearlevelandreportedinthe parentheses. ∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. 17

Figure2CorporateInvestmentSensitivitytoMPSUS: ParallelTrendsAssumption 20. 0 20.- 40.- 60.- 80.- -3 -2 -1 0 1 2 3+ NOTE.ThefigureplotscorporateinvestmentsensitivitytoMPSUSofconnectedfirmsrelativetounconnected firms—coefficientestimatesand95%confidenceintervalfromcolumn(2)inTable2. 5.3 Benchmark Results for Chinese Investment Table 3 reports the regression results that form the backbone of our paper. We begin with estimates of the Probit model of Connect selection, as motivated above. What follows are estimates from three different approaches: panel OLS in columns (1)-(3), Heckman Second-Stage regressions in columns (4)-(6), and Propensity Score Matching (PSM) in columns (7)-(8). These establish robustness of our evidence to different attempts to tackle sample selection issues. As discussed above, the selection of connected firms is not random.20 We follow the literature in employing the Heckman and PSM correction methods, but because of recent critiques of both, we put equal stock in the OLS results (see Tucker (2010)andWolfoldsandSiegel(2019)). Resultsarehighlyrobust. Panel OLS Columns (1)-(3) present the panel OLS regression results. The first column, which excludes the foreign spillover terms, shows the positive effect of the Connect on 20TheChinaSecurityIndexCompanyisresponsibleforcompositionoftheSSE180andSSE380. Accordingtotheirdisclosure,stocksareselectedintoSSE180basedontheirmarketcap,undertheconditions thattheyshowgoodperformanceandhavenoseriousfinancialproblemsorlargepricevolatility.FortheSSE 380,authoritiesselectstocksbasedonrevenuegrowth,returnonnetassets,turnover,andtotalmarketvalue. Detailedinformationcanbefoundathttp://www.csindex.cn/en/indices/index-detail/000010and http://www.csindex.cn/en/indices/index-detail/000009. 18

Table3BaselineResults: U.S.MonetaryPolicy,ChineseCorporateInvestment,andtheConnect PanelA:FirstStageProbitModel PanelB:Investment ConnectDummy PanelOLSRegression HeckmanTwo-Stage PropensityScoreMatching (1) (2) (3) (4) (5) (6) (7) (8) StockVolatility -15.396*** Connect 0.001* 0.002* 0.001 0.017*** 0.034*** 0.017*** 0.031*** 0.013*** (0.594) (0.001) (0.001) (0.001) (0.005) (0.005) (0.004) (0.004) (0.003) MarketCap 0.758*** MPSUS*Connect -0.020** -0.019** -0.028** -0.024** -0.020*** -0.013** (0.007) (0.010) (0.009) (0.012) (0.011) (0.006) (0.005) Leverage -0.357*** MPSUS -0.008* -0.011** -0.009** -0.011*** -0.010*** -0.009*** (0.045) (0.004) (0.004) (0.004) (0.004) (0.003) (0.003) Age 0.058*** IMR -0.009*** -0.020*** -0.010*** -0.018*** -0.007*** (0.001) (0.003) (0.003) (0.003) (0.002) (0.002) DividendDummy -0.019 Size 0.004*** 0.004*** 0.003*** 0.003*** 0.005*** (0.019) (0.001) (0.001) (0.001) (0.001) (0.001) LagTobin’sQ 0.001*** 0.002*** 0.001*** 0.001*** 0.002*** (0.000) (0.000) (0.000) (0.000) (0.000) CashFlow 0.172*** 0.173*** 0.169*** 0.170*** 0.154*** (0.011) (0.011) (0.012) (0.012) (0.015) SalesGrowth 0.001*** 0.001*** 0.001*** 0.001*** 0.005*** (0.000) (0.000) (0.000) (0.000) (0.001) GDPGrowth 0.018 0.024 0.020 0.026 -0.007 (0.016) (0.015) (0.017) (0.016) (0.030) IndustryFE Yes IMR No No No Yes Yes Yes Yes Yes ProvinceFE Yes FirmFE Yes Yes Yes Yes Yes Yes Yes Yes ExchangeFE Yes YearFE Yes Yes Yes Yes Yes Yes Yes Yes QuarterDummy Yes QuarterDummy Yes Yes Yes Yes Yes Yes Yes Yes Observations 85486 Observations 87740 87740 87740 85486 85486 85486 20003 20003 PseudoR2 0.313 AdjustedR2 0.409 0.387 0.410 0.411 0.393 0.413 0.521 0.543 NOTE. The dependent variable is corporate investment, defined as quarterly capital expenditure scaled by the beginning-of-quarter book value of total assets. Other firm level controls can be found at A.5. For column (1)-(6) (column (7)-(8)), the standard errors are clustered at both firm and year (region-year)levelandreportedinparentheses. ∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. 19

Chinese corporate investment (row 1). This is consistent with previous literature on stock market liberalization (Henry (2000b) and Chari and Henry (2008)). Our results suggest that average quarterly corporate investment increased by 2.86% once a firm was included in the Connect, which is statistically significant and economically large.21 We explore the channels of this effect in the following sections, while noting here that this result is evidenceinfavorofourhypothesis3,whichwereturntobelowindetail. Columns (2)-(3) present our baseline results for testing hypothesis 1. We use simple aggregation of MPSUS across months in a quarter as our benchmark while noting that our results are robust to using the value (date) weighted sum to construct the U.S. monetary policy shock. In column (2), we report the regression of corporate investment on MPSUS, t Connect and the interaction term, with firm and year fixed effects, and quarter dummies, it while column (3) adds firm characteristics: Tobin’s Q, cash flow, sales growth, size, and provincial GDP growth. Consistent with our hypothesis, the interaction term is negative and both economically and statistically significant. The reduction in conditional investment is around 0.02, meaning that a 1 percent unexpected increase in the US monetary policyshockreducescorporateinvestmentby0.02percentonaverageforfirmsincludedin the Connect compared to firms not in the Connect, after controlling for investment opportunities and economic conditions.22 In terms of economic magnitudes, these coefficients translateintoareductionof2.80%basedontheaverageinvestmentrateandMPSUS.23 t Heckman-TwoStageResultsFortheHeckmanTwo-Stageapproach,wefirstestimatethe Probit model that gives us determinants of the Connect dummy: stock volatility (standard deviation of the daily stock return in each quarter), market cap measured as the natural logarithm of market capitalization, leverage, firm age, and an indicator for whether a firm pays cash dividends. We also include industry, province, and exchange fixed effects in the first-stage. The results suggest that firms more likely to be selected into the Connect are those with: lower stock volatility, larger size, lower leverage, older, and a non-dividend payer. These are consistent with the objectives of the index selection procedure. We then re-estimate our baseline regression (4) adding as an explanatory variable the inverse Mills ratio(IMR):theProbitmodel’sprobabilitydensityfunctiondividedbythecumulativedis- 21Thecalculationofeconomicmagnitudeisasfollows: 0.001/0.035=2.86%. 22Thecoefficientontheinteractiontermislargerwhenusingvalueweightedseriesthanequalweighted. Thisisbecausethevalueweightedseriestakesintoaccountthefactthatinvestmentmightbeslowtorespond toexternalshocksandthusgivesmoreweighttoshockshappeningearlierinthequarter. 23Thecalculationisasfollows: 0.020*0.049/0.035=2.80%. 20

tribution function. Columns (4)-(6) present the results of re-estimating the baseline model aftercorrectingforselectioninthisway. Clearly,theimpactoftheConnectonfirminvestment is stronger both economically and statistically. Estimates of the interaction term are consistentwiththepanelOLSresults. Propensity Score Matching Another concern is that the effect of the Connect may not be homogeneous across firms, but may vary as a function of firm characteristics. Simple difference-in-differences estimates may be biased if there are some firms which were connected but there are no comparable firms which were left unconnected, and vice-versa. MatchingmethodseliminatethispotentialsourceofbiasbypairingConnect(treated)with unconnected(control)firmsthathavesimilarobservedattributes. Usingobservationsinthe treatmentandcontrolgroupsoverthe“regionofcommonsupport”eliminatesthissourceof bias. In general, conventional matching methods assume that, conditional on the observed variables, the counterfactual outcome distribution of the treated firms is the same as the observedoutcomedistributionoffirmsinthecontrolgroup(seeHeckmanetal.(1997)). The strategy is thus to construct a new group by finding unconnected firms with observables similar to those of connected firms. We then examine robustness of the baseline estimates to those estimated only on the observations that lie on the common support. We firstusealogitregressiontoestimatetheprobabilityofafirmbeingconnected,byincluding setsofvariablesandindustry,province,exchangemarketfixedeffects. Wethenexclude(1) unconnected-firm observations whose propensity scores are less than the propensity score of the connected stocks at the first percentile of the treatment propensity score distribution and (2) all treatment observations whose propensity score is greater than the propensity scoreofthecontrolobservationattheninety-ninthpercentileoftheun-treateddistribution. Re-estimating the difference-in-differences model with these “nearest neighbors” on the common support region allows us to analyze the extent of this source of bias. As seen in columns(7)and(8),ourresultsarerobust: theinteractiontermbetweentheU.S.monetary policy shock and the Connect dummy remains significantly negative. Because the PSM exercise substantially reduces the sample size, from over 85,000 to 20,000, we revert back tothefullsampleofobservationsintheremainderoftheregressions. 21

Table4CorporateInvestmentandFOMCShocks: GlobalFinancialCycles Investment (1) (2) (3) (4) (5) (6) PanelA:VIXIndexfromCBOES&P500 PanelE:News-basedEconomicUncertaintyIndexfromBBD MPSUS*Connect -0.023** -0.022** -0.025** MPSUS*Connect -0.019* -0.019* -0.025** (0.010) (0.009) (0.011) (0.011) (0.010) (0.011) Log(VIX)*Connect -0.006*** -0.005** -0.002 EPU*Connect 0.001 0.000 -0.001 (0.002) (0.002) (0.003) (0.003) (0.003) (0.003) Connect 0.019*** 0.014*** 0.023*** Connect 0.000 0.000 0.019*** (0.005) (0.005) (0.006) (0.004) (0.004) (0.005) Observations 87740 87740 85486 Observations 87740 87740 85486 AdjustedR2 0.387 0.410 0.413 AdjustedR2 0.387 0.410 0.413 PanelB:DollarIndexReturn PanelF:GlobalEconomicPolicyUncertaintyIndexfromBBD MPSUS*Connect -0.016** -0.016** -0.021*** MPSUS*Connect -0.020* -0.020* -0.025** (0.008) (0.008) (0.008) (0.011) (0.011) (0.011) DollarReturn*Connect -0.038 -0.036 -0.042** GEPU*Connect -0.000 -0.001 -0.003 (0.025) (0.023) (0.020) (0.002) (0.002) (0.002) Connect 0.002** 0.001 0.018*** Connect 0.002 0.002 0.022*** (0.001) (0.001) (0.004) (0.003) (0.004) (0.006) Observations 87740 87740 85486 Observations 87740 87740 85486 AdjustedR2 0.387 0.410 0.413 AdjustedR2 0.387 0.410 0.413 PanelC:ExchangeRateReturnofRMB-USD PanelG:WorldUncertaintyIndexfromABF MPSUS*Connect -0.016* -0.017* -0.023** MPSUS*Connect -0.013** -0.012** -0.013** (0.008) (0.009) (0.010) (0.006) (0.005) (0.005) RMBUSD*Connect -0.005 -0.014 -0.014 WUI*Connect 0.004 0.002 -0.002 (0.049) (0.045) (0.039) (0.005) (0.006) (0.006) Connect 0.002** 0.001 0.017*** Connect 0.001 0.000 0.019*** (0.001) (0.001) (0.004) (0.002) (0.002) (0.005) Observations 87740 87740 85486 Observations 87740 87740 85486 AdjustedR2 0.387 0.410 0.413 AdjustedR2 0.387 0.411 0.414 PanelD:MonetaryPolicyUncertaintyIndexfromHRS PanelH:TEDrate MPSUS*Connect -0.025*** -0.024** -0.029** MPSUS*Connect -0.020** -0.020** -0.023** (0.009) (0.010) (0.012) (0.010) (0.010) (0.010) MPU*Connect -0.002 -0.001 -0.002 TEDrate*Connect 0.010 0.010 0.002 (0.002) (0.003) (0.002) (0.012) (0.012) (0.013) Connect 0.004 0.003 0.019*** Connect -0.001 -0.002 0.017*** (0.004) (0.004) (0.006) (0.003) (0.004) (0.005) Observations 85797 85797 83611 Observations 87740 87740 85486 AdjustedR2 0.391 0.415 0.417 AdjustedR2 0.387 0.410 0.413 IMR No No Yes IMR No No Yes FirmControls No Yes Yes FirmControls No Yes Yes FirmFE Yes Yes Yes FirmFE Yes Yes Yes YearFE Yes Yes Yes YearFE Yes Yes Yes QuarterDummy Yes Yes Yes QuarterDummy Yes Yes Yes NOTE: thedependentvariableiscorporateinvestment. PanelAaddstheVIXindexanditsinteractionwith Connect. Panel B adds a dollar index return and its interaction with Connect. Panel C adds the bilateral exchangeratereturnbetweenthedollarandRMBanditsinteractionwithConnect. PanelDaddsamonetary policy uncertainty index (MPU) identified by Husted et al. (forthcoming) and its interaction with Connect. PanelEaddsanews-basedeconomicpolicyuncertaintyindex(EPU)fromBakeretal.(2016)anditsinteraction with Connect. Panel F adds a GDP-weighted average of national EPU indices for 16 countries that account for two-thirds of global output (GEPU) and its interaction with Connect (see Davis (2016) for details).PanelGaddsaworlduncertaintyindexfromAhiretal.(2018).PanelHusestheTEDspreadmeasured asthedifferencebetweeninterestratesoninterbankloansandshort-termU.S.governmentdebt. Allstandard errorsareclusteredatbothfirmandyearlevelandreportedinparentheses. ∗, ∗∗ and∗∗∗ indicatestatistical significanceatthe10%,5%,and1%level,respectively. 22

5.4 Additional Robustness Checks Potpourri In Panel A of Appendix Table S.1, we replace firm fixed effect with industry fixed effects. Panel B drops the dual-listed stocks, including A-B dual listed and A-H dual listed, in order to see whether these already-opened firm shares are driving our baseline results. Panel C adds the interaction term of firm size and U.S. monetary policy shock to alleviate the concern that firm size affects the investment sensitivity to U.S. monetary policy shock. In all three robustness exercises, the coefficients on the interaction term are quantitatively similar to our baseline results. Panel D uses the alternative measure of U.S. monetary policy shocks estimated by Bu et al. (2019).24 Results are consistent with our baseline,butwitharelativelysmallermagnitudeandlesssignificance. PanelEaddslagged investment to the baseline specification. The new coefficient is insignificantly positive, suggesting that investment is persistent, while the interaction term remains statistically significant. Panel F introduces a lag of MPSUS and its interaction with Connect, to see if investment responds slowly to external shocks. The coefficients on the lagged interaction termareinsignificant,however. OthermeasuresofexternalshocksWealsoincludedifferentmeasuresofexternalshocks to examine whether our results relying on MPSUS are robust. Table 4 presents the results. Panel A adds the VIX index and its interaction with Connect. Panel B adds a U.S. dollar index return and its interaction with Connect. Panel C adds the bilateral exchange rate change between dollar and RMB and its interaction with Connect. Panel D adds the monetary policy uncertainty index of Husted et al. (forthcoming) and its interaction with Connect. PanelEaddsthenews-basedeconomicpolicyuncertaintyindexfromBakeretal. (2016)anditsinteractionwithConnect. PanelFaddsaGDP-weightedaverageofnational EPUindicesfor16countriesthataccountfortwo-thirdsofglobaloutputanditsinteraction with Connect (see Davis (2016)). Panel G adds a world uncertainty index from Ahir et al. (2018) and its interaction with Connect, and Panel H does the same with the TED spread. In all cases, the interaction between the MPSUS shock and Connect remains statistically significantandsimilarinmagnitudetoourbaselineresults. (Placebo)effectofChinesemonetarypolicyOurbaselineresultssuggestthatbeingconnected makes corporate investment more sensitive to external shocks. However, because both connected and unconnected firms are exposed to Chinese monetary policy shocks, 24ThismeasureappliesaFama-MacBethproceduretotheresponseofthefullmaturityspectrumofinterestratestoFOMCannouncements. Themeasurecomparesfavorablytoalternativesintheliterature. 23

there should be no different responses to these domestic policy shocks. To formally test this, we use the Chinese monetary policy shock estimated by Chen et al. (2018) and reestimate our baseline regression.25 The results in Appendix Table S.2 show that there is no significantdifference bythese twotypes of firmsin theirresponse toChinese monetary policy shocks. Furthermore, we also “horse race” the Chinese monetary policy shock with theU.S.monetarypolicyshock. Ourmainresultsstillhold. 6 Firm Heterogeneity: Hypothesis 2 6.1 Risk-sharing (Risk-premium) Channel Ourconceptualframeworkimpliesthattheglobalfinancialcyclecanaffectallfirmsthrough acostoffundingchannel,butalsothatconnectedandunconnectedfirmsareaffecteddifferently. Inaddition,firmsontheconnect—withhighercovariancewiththeglobalmarket— should be more sensitive to U.S. monetary policy shocks because their risk-premiums are more responsive to the global financial cycle. To formally test firm heterogeneity through the risk-sharing (risk premium) channel, we multiply our connect dummy, Connect by a it firm-level variable Global Cov, i.e. cov(r,r ), the historical covariance of firm i’s stock i W returnr withtheglobalmarketreturnr . Thisproducesacontinuousmeasurewhichcapi W tures both the extensive and intensive margin of the risk-sharing channel: it equals zero whenthefirmcannotbetradedbyforeigninvestorsbutvarieswiththefirm’ssensitivityto the global market for a firm in the Connect program. We then replace the connect dummy inourbaselineregressionwithGlobalCov∗Connect toassessthisheterogeneity. it Table 5 presents the results. Consistent with our baseline results, the coefficient on the GlobalCov∗Connect is significantly positive. Furthermore, the interaction term between GlobalCov∗Connectand theU.S. monetarypolicyshock issignificantlynegative, implying that the spillover effects from the global financial cycle are stronger after the Connect. Thus, firms with higher covariance with the global market enjoy higher benefits after the Connectalongwithgreatersensitivitytotheglobalfinancialcycle,afterinclusion. 25We are grateful for the data shared by Chen et al. (2018). The Chinese monetary policy shock is an estimatedshocktoChineseM2growthrate,aquantitymeasureofmonetarypolicyandthusdifferentfrom ourU.S.monetarypolicyshock,whichisapricemeasure. 24

Table5CorporateInvestmentandFOMCShocks: Risk-Sharing(Risk-premium)Channel Investment (1) (2) (3) (GlobalCov*Connect) 0.019*** 0.014*** 0.031*** (0.006) (0.005) (0.010) (GlobalCov*Connect)*MPSUS -0.131** -0.106* -0.106* (0.057) (0.056) (0.063) MPSUS -0.009* -0.012*** -0.012*** (0.004) (0.004) (0.004) Size 0.005*** 0.004*** (0.001) (0.001) LagTobin’sQ 0.002*** 0.001*** (0.000) (0.000) CashFlow 0.171*** 0.172*** (0.012) (0.012) RevenueGrowth 0.001*** 0.002*** (0.000) (0.000) GDPGrowth 0.025 0.026* (0.015) (0.015) IMR No No Yes FirmFE Yes Yes Yes YearFE Yes Yes Yes QuarterDummy Yes Yes Yes Observations 86447 86447 84202 AdjustedR2 0.389 0.413 0.415 NOTE: the dependent variable is corporate investment. Global Cov is the historical covariance of an individualstockreturnwiththeMSCIworldmarketreturn(exchangerateadjusted),estimatedusinga36-month rollingwindow. DetailedinformationcanbefoundatA.5. Allstandarderrorsareclusteredatbothfirmand yearlevelandreportedintheparentheses. ∗,∗∗ and∗∗∗ indicatestatisticalsignificanceatthe10%,5%,and 1%level,respectively. 6.2 External Financing Channel WeexplorewhetherfirmsrelyingonmoreexternalfinancingforinvestmentaremoresensitivetoU.S.monetarypolicyshocks.26 Tothisend,weimplementsub-sampletestsexploring firm heterogeneity in the treatment group. For example, we divide our full sample into two groups in each quarter based on measures of external financing. We then re-estimate our baseline regression on the two sub-samples separately. To the extent that the global financing cycle affects domestic investment through the cost of funding, one should expectthatfirmswithdifferentexternalfinancialconditionsresponddifferently. Weformally test this by dividing firms according to their equity dependence to investment or long-term debt to investment in Table 6. Firms with greater reliance on external financing, equity or long-termdebt,aremoresensitivetoUSmonetarypolicyshocksafterinclusion. 26WeexploreothertypesoffirmheterogeneityinonlineAppendixS.6, forexamplewhetherfirmswith moreexposuretotheexternalsector,asmeasuredbytradablevs. non-tradablesectorortheshareofforeign 25

Table6ExternalFinancingChannel Investment PanelA:EquityDependencetoInvestment PanelB:Long-termDebttoInvestment High Low High Low High Low High Low MPSUS*Connect -0.030** -0.012 -0.033** -0.018* MPSUS*Connect -0.027*** -0.011 -0.032*** -0.015 (0.014) (0.008) (0.014) (0.009) (0.010) (0.009) (0.011) (0.010) MPSUS -0.011** -0.010** -0.011** -0.011** MPSUS -0.009** -0.012*** -0.009** -0.012*** (0.005) (0.004) (0.005) (0.004) (0.004) (0.004) (0.004) (0.004) Connect 0.003 0.001 0.015*** 0.021*** Connect 0.001 0.002* 0.013*** 0.020*** (0.002) (0.001) (0.005) (0.005) (0.001) (0.001) (0.005) (0.005) Observations 39870 47769 39008 46389 Observations 42778 44861 41758 43639 AdjustedR2 0.484 0.400 0.486 0.403 AdjustedR2 0.457 0.461 0.459 0.465 FirmControls Yes Yes Yes Yes Yes Yes Yes Yes IMR No No Yes Yes No No Yes Yes FirmFE Yes Yes Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes Yes Yes QuarterFE Yes Yes Yes Yes Yes Yes Yes Yes H0:βH=βL χ2Test 8.19*** 5.57** 6.26** 6.15** P-value 0.004 0.018 0.012 0.013 NOTE: thedependentvariableiscorporateinvestment. PanelAdividesfirmsintotwogroupsaccordingto themedianlevelofequitydependencetoinvestmentineachquarter. PanelBdividesfirmsintotwogroups accordingtothemedianleveloftheirlong-termdebttoinvestmentineachquarter. Allstandarderrorsare clusteredatbothfirmandyearlevelandreportedinparentheses. ∗,∗∗and∗∗∗indicatestatisticalsignificance atthe10%,5%,and1%level,respectively. 7 Positive Effects of the China Connect: Hypothesis 3 If the primary effect of the Connect were that connected firms’ investment becomes more sensitive to U.S. monetary policy, we would expect that firms would prefer to remain unconnected; this suggests that connected firms enjoy many positive effects from inclusion. In addition, our conceptual framework motivates us to explore considerations like the predictedpositiveshort-runeffectsoftheConnect. Ceterus paribus, the effect of being in the Connect is to boost firm investment, as seen from the positive coefficient on the Connect dummy in our baseline results of Table 3. Furthermore, those effects are very persistent, lasting for 7-8 quarters after the Connect according to Appendix Figure S.1. Consistent with the previous literature, we also find thatthoseeffectsoccurthrougharisk-sharingchannelinTable7,asmeasuredbyourDIF- COV term (see Chari and Henry (2008)). Another positive effect of the Connect can be seen through event study analysis, which indicates that connected stocks experience a significant value appreciation, compared with unconnected ones, upon announcement of the sales in total sales, respond differently. We do not find large differences along these dimensions, perhaps becauseChinesefirmsissueverylittledollardenominateddebt. 26

Figure3CumulativeAbnormalReturnsAroundAnnouncementDay: ConnectedFirmsrelativetoUnconnectedFirms 10% 8% 6% 4% 2% 0% -15 -10 -5 -11 5 10 15 -2% NOTE. Thefigureplotsthedifferenceincumulativeabnormalreturnsbetweenconnectedandunconnected stocksaroundtheannouncementwindow(days-15,20)intheShanghai-HongKongStockConnectprogram. The 95% confidence interval is plotted in the dashed lines. The vertical line marks the announcement date forthelistofeligiblestockstobeincludedintheConnect,November10,2014. program. Figure 3 shows the cumulative abnormal returns difference between connected and unconnected stocks surrounding the event date.27 The rising, positive effect on stock returnsforconnectedfirmsrelativetounconnectedfirmsisstatisticallysignificantandeconomically large. Furthermore, we find a risk-sharing channel in Table 8, consistent with ourtheoreticalmotivationandpreviousliterature(ChariandHenry(2004)).28 Moreover,inTable9,wepresenttheeffectsoftheConnectonmeasuresoffirmperformanceandfinancingcosts. Asseenincolumns(1)-(4),returnsonassets(ROA)andequity (ROE) are significantly higher for those in the Connect than those outside. Furthermore, financing costs such as the cost of debt and dividend to price ratio are lower for connected 27We only consider stocks listed on the Shanghai Stock Exchange since the first Connect is between ShanghaiandHongKong,whichisregardedasanunexpectedeventtoinvestors. WechooseNov. 10,2014 (ratherthanNov. 17,2014)asourannouncementdaybecausethelistofeligiblestocks(tobeincludedinthe ConnectfromNov17)wasannouncedonNov. 10. 28Weusethemarketmodeltocalculatethecumulativeabnormalreturn. A250-dayestimationwindow isusedtoestimatedtheβcoefficientbetweenthemarketreturnandstockreturn. A30-daygapbetweenthe estimationwindowandeventwindowisrequired. Moreover,werequireatleast100daysreturndatainthe estimation window. We also estimate a Fama-French three-factor and Carhart four-factor model, and find robustresults. 27

Table7Investment,RiskSharing,andtheChinaConnect Investment (1) (2) (3) DIFCOV*Connect 0.006** 0.007** 0.004* (0.003) (0.003) (0.002) DIFCOV -0.002 0.000 -0.001 (0.003) (0.003) (0.003) Connect 0.000 -0.005** -0.023*** (0.003) (0.003) (0.007) Size 0.002*** 0.004*** (0.001) (0.001) LagTobin’sQ 0.000 0.001** (0.000) (0.000) CashFlow 0.223*** 0.232*** (0.019) (0.018) SalesGrowth 0.002*** 0.002*** (0.001) (0.001) GDPGrowth 0.033* 0.035** (0.017) (0.017) IMR No No Yes FirmFE Yes Yes Yes YearFE Yes Yes Yes QuarterDummy Yes Yes Yes Observations 86447 86447 84202 AdjustedR2 0.190 0.236 0.239 NOTE;thedependentvariableiscorporateinvestment.DIFCOVismeasuredbydefinedascov(r i ,r M )-cov(r i , r ),wherer isthestockreturnforfirmi,r isthedomesticstockreturnandr istheglobalmarketreturn. W i M W All standard errors are clustered at both industry and year level and reported in parentheses. ∗, ∗∗ and ∗∗∗ indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. 28

Table8StockPriceRevaluationsforConnectedandUnconnectedFirms Month[0]Window Month[0,+1]Window (1) (2) (3) (4) (5) (6) DIFCOV*Connect 0.242*** 0.276*** 0.273*** 0.431*** 0.574*** 0.573*** (0.046) (0.046) (0.047) (0.095) (0.091) (0.091) DIFCOV 0.074*** 0.085*** 0.096*** -0.099*** -0.043 -0.040 (0.017) (0.017) (0.019) (0.025) (0.026) (0.029) Connect -0.038* -0.053** -0.057*** 0.005 -0.057 -0.058 (0.016) (0.016) (0.016) (0.034) (0.033) (0.033) MarketCap 0.000** 0.000** 0.001** 0.001** (0.000) (0.000) (0.000) (0.000) Turnover -0.003*** -0.004*** -0.017*** -0.017*** (0.001) (0.001) (0.002) (0.002) SHSE 0.011 0.003 (0.006) (0.012) Constant 0.008 0.012 0.006 0.025 0.046** 0.045** (0.009) (0.009) (0.010) (0.014) (0.014) (0.016) Observations 2309 2309 2309 2309 2309 2309 AdjustedR2 0.054 0.063 0.064 0.127 0.179 0.179 NOTE: thedependentvariableisthelogstockreturn. Columns(1)–(3)usethestockreturnintheconnection month while Columns (4)–(6) use the stock return in both the connection month and the following month. DIFCOVisdefinedascov(r,r )-cov(r,r ),wherer isthestockreturnforfirmi,r isthedomesticstock i M i W i M returnandr istheglobalmarketreturn. Marketcapisthenaturallogarithmofthemarketcapitalizationof W total assets. Turnover is the average individual turnover rate within a month. SHSE is a dummy variable indicating that a firm listed on the Shanghai Stock Exchange, and zero if listed on the Shenzhen Stock Exchange.∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. firms (columns (5)-(8)).29Thus, the Connect firms exhibit sizable stock price revaluations, increased growth rate of capital stock, and better firm performance, consistent with Chari andHenry(2004,2008),whichcoincideswithareductioninfinancingcosts. Finally,weconsiderimplicationsforaneconomywhosecorporateinvestmentexpenditures are more sensitive to external shocks. Table S.7 shows that firms in the Connect hold less cash after the connection, consistent with our baseline results on the investment since cash and investment are substitute. But firms in the Connect are more sensitive to U.S. monetary policy shocks and increase their cash holdings following a contractionary U.S. monetary policy shock (row (2)). This reinforces the notion that U.S. monetary policy has large spillover effects, especially considering China’s tight capital controls (see Kalemli- Ozcan (2019)). One potential downside of the extra sensitivity to U.S. monetary policy relates to the independence of Chinese monetary policy. In light of the (additional) foreign spillover effects working through the Connect, Chinese monetary policy might have 29ThecoefficientonConnectisnegativebutnotsignificantinthecostofdebtregression. Thismayoccur becausewehaveanaggregatemeasureforcostofdebt,ratherthanfirm-specific. WeiandZhou(2019)use loanleveldatatomeasurecostofdebtandfindthatstockmarketliberalizationreducesfirmscostofdebt. 29

Table9FirmPerformance,FinancingCosts,andtheChinaConnect ROA ROE CostofDebt(%) Changeofln(D/P)(%) (1) (2) (3) (4) (5) (6) (7) (8) Connect 0.001*** 0.005*** 0.005*** 0.023*** -0.041 -0.016 -0.024*** -0.354*** (0.000) (0.001) (0.001) (0.007) (0.031) (0.100) (0.006) (0.023) LagTobin’sQ -0.001*** -0.001*** -0.000 -0.000 0.012 0.012 0.019*** 0.024*** (0.000) (0.000) (0.000) (0.000) (0.008) (0.009) (0.001) (0.001) CashFlow 0.915*** 0.914*** 1.652*** 1.653*** 0.050 0.057 -0.256*** -0.153*** (0.006) (0.006) (0.054) (0.054) (0.187) (0.189) (0.052) (0.047) SalesGrowth -0.003*** -0.003*** -0.002* -0.002** 0.022* 0.019 0.060*** 0.062*** (0.000) (0.000) (0.001) (0.001) (0.012) (0.012) (0.009) (0.009) GDPgrowth -0.008** -0.008** 0.048* 0.043 0.912* 0.954* 0.456*** 0.434*** (0.004) (0.004) (0.026) (0.027) (0.528) (0.526) (0.136) (0.142) Size -0.000 -0.001* -0.005** -0.007*** 0.275*** 0.267*** 0.015*** 0.047*** (0.000) (0.000) (0.002) (0.002) (0.026) (0.027) (0.003) (0.004) Leverage -0.010*** -0.009*** 0.046*** 0.052*** 1.845*** 1.845*** 0.082*** 0.003 (0.001) (0.001) (0.009) (0.009) (0.106) (0.108) (0.016) (0.016) IMR No Yes No Yes No Yes No Yes FirmFE Yes Yes Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes Yes Yes QuarterDummy Yes Yes Yes Yes Yes Yes Yes Yes Observations 87740 85486 87737 85484 87740 85486 80271 78161 AdjustedR2 0.929 0.928 0.437 0.438 0.522 0.523 0.014 0.018 NOTE: the dependent variable is return on assets (ROA) in columns (1)-(2), return on equity (ROE) in columns (3)-(4), cost of debt (%) measured by borrowing cost in columns (5)-(6), and change of dividend to price ratio, ln(D/P) (%) in columns (7)-(8). All standard errors are clustered at firm level and reported in parentheses. Detailed information on the controls can be found in Appendix A.5. ∗, ∗∗ and ∗∗∗ indicate statisticalsignificanceatthe10%,5%,and1%level,respectively. torespondtoU.S.monetarypolicyinawaythatdeviatesfromitsdomesticmandate.30 8 Conclusion We exploit an important and unique capital account liberalization in China, the Shanghai (Shenzhen)-Hong Kong stock Connect, to jointly test hypotheses concerning spillover effects from external shocks and the efficacy of capital controls. The Connect allows certain stocks to be eligible for foreign investors while restricting other shares to remain available only to domestic investors, and is a natural experiment to study transmission of external shocks. We devote considerable attention to sample selection issues concerning connected firms,issuesthatareimportantaboveandbeyondthe“natural-ness”oftheexperiment. Wefindtwomainresults. First,ChinesefirmsaremorenegativelyaffectedbycontractionaryU.S.monetarypolicyshocksaftertradingintheirsharesbecameopentoforeigners 30Forexample,duringeventslikethe2013TaperTantrum,Chinesemonetarypolicywouldhavetoease inordertostabilizethedomesticeconomy. 30

than are unconnected firms. Firms whose stock returns have a higher covariance with the global market return are affected more. Our results indicate that firms relying more on external financing are important in driving our results. If these were the only effects of the Connect, we expect that Chinese firms would act to remain outside of it. Furthermore, to the extent that Chinese monetary policy transmission and independence are diminished by increasedsensitivitytoUSshocks,wewouldexpectChineseauthoritiestopullbackonthe Connect. Investigating further leads to our second main finding: firms in the Connect had higher investment expenditures, enjoyed lower financing costs, and earned higher returns on equity (ROE) and assets (ROA) than firms outside of the Connect. This suggests that connected firms are able to hedge the negative consequences concerning increased sensitivity to external shocks. Our findings have strong policy implications. U.S. monetary policy shocks, the crucial driver in the literature on Global Financial Cycles, have importantspillovereffectsworkingthroughthepartialopeningoftheChinesestockmarket,even with tight overall capital controls. Nevertheless, our results indicate that capital controls arestilleffectiveincurbingthenegativespilloversontoChinesefirminvestment,thuspreservingadegreeofmonetarypolicyindependencerelativetofullyopencapitalmarkets. References AHIR, H., N. BLOOM, AND D. FURCERI (2018): “The world uncertainty index,” Manuscript. ALFARO, L., A. CHARI, AND F. KANCZUK (2017): “The real effects of capital controls: Firm-level evidence from a policy experiment,” Journal of International Economics, 108,191–210. BAKER, S. R., N. BLOOM, AND S. J. DAVIS (2016): “Measuringeconomicpolicyuncertainty,”QuarterlyJournalofEconomics,131,1593–1636. BEKAERT, G., C. R. HARVEY, AND C. LUNDBLAD(2005): “Doesfinancialliberalization spurgrowth?” JournalofFinancialEconomics,77,3–55. BENIGNO, G., H. CHEN, C. OTROK, A. REBUCCI, AND E. R. YOUNG (2013): “Financial crises and macro-prudential policies,” Journal of International Economics, 89, 453–470. 31

BIANCHI, J. (2011): “Overborrowing and systemic externalities in the business cycle,” AmericanEconomicReview,101,3400. BRUNNERMEIER, M. K., M. SOCKIN, AND W. XIONG (2018): “China’s model of managingthefinancialsystem,”Manuscript. BRUNO, V., I. SHIM, AND H. S. SHIN(2017): “Comparativeassessmentofmacroprudentialpolicies,”JournalofFinancialStability,28,183–202. BU, C., J. ROGERS, AND W. WU (2019): “A unified measure of Fed monetary policy shocks,”FEDSworkingpaper,2019-043. CAO, J., B. JULIO, T. LENG, AND S. ZHOU (2016): “Political turnover, ownership, and corporateinvestment,”Manuscript. CERUTTI, E., S. CLAESSENS, AND A. K. ROSE (2019): “How important is the global financialcycle? Evidencefromcapitalflows,”IMFEconomicReview,67,24–60. CHANG, C., K. CHEN, D. F. WAGGONER, AND T. ZHA (2016): “Trends and cycles in China’smacroeconomy,”NBERMacroeconomicsAnnual,30,1–84. CHARI, A. AND P. B. HENRY (2004): “Risk sharing and asset prices: evidence from a naturalexperiment,”JournalofFinance,59,1295–1324. ——— (2008): “Firm-specific information and the efficiency of investment,” Journal of FinancialEconomics,87,636–655. CHEN, K., J. REN, AND T. ZHA (2018): “The nexus of monetary policy and shadow bankinginChina,”AmericanEconomicReview,108,3891–3936. CHINN, M. D. AND H. ITO (2006): “What matters for financial development? Capital controls,institutions,andinteractions,”JournalofDevelopmentEconomics,163–192. DAVIS, S. J. (2016): “An index of global economic policy uncertainty,” NBER Working PaperNo.22740. DI GIOVANNI, J., S. KALEMLI-OZCAN, M. F. ULU, AND Y. S. BASKAYA (2017): “Internationalspilloversandlocalcreditcycles,”NBERWorkingPaperNo.23149. ERTEN, B., A. KORINEK, AND J. A. OCAMPO (forthcoming): “Capital controls: Theory andevidence,”JournalofEconomicLiterature. 32

FAMA, E. F. AND K. R. FRENCH (1993): “Common risk factors in the returns on stocks andbonds,”JournalofFinancialEconomics,33,3–56. FARHI, E. AND I. WERNING(2016): “Atheoryofmacroprudentialpoliciesinthepresence ofnominalrigidities,”Econometrica,84,1645–1704. FERNALD, J. AND J. H. ROGERS (2002): “Puzzles in the Chinese stock market,” Review ofEconomicsandStatistics,84,416–432. FERNA´NDEZ, A., A. REBUCCI, AND M. URIBE (2015): “Are capital controls countercyclical?” JournalofMonetaryEconomics,76,1–14. FORBES, K., M. FRATZSCHER, AND R. STRAUB (2015): “Capital-flow management measures: Whataretheygoodfor?” JournalofInternationalEconomics,96,S76–S97. HAN, X. AND S.-J. WEI (2018): “International transmissions of monetary shocks: Betweenatrilemmaandadilemma,”JournalofInternationalEconomics,110,205–219. HECKMAN, J. J., H. ICHIMURA, AND P. TODD (1997): “Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme,” Review of EconomicStudies,64,605–54. HENRY, P. B. (2000a): “Do stock market liberalizations cause investment booms?” JournalofFinancialEconomics,58,301–334. ——— (2000b): “Stock market liberalization, economic reform, and emerging market equityprices,”JournalofFinance,55,529–564. ———(2003): “Capital-accountliberalization,thecostofcapital,andeconomicgrowth,” AmericanEconomicReview,93,91–96. HUSTED, L., J. H. ROGERS, AND B. SUN (forthcoming): “Monetarypolicyuncertainty,” JournalofMonetaryEconomics. IMF (2012): “Theliberalizationandmanagementofcapitalflows: aninstitutionalview,”. JEANNE, O. AND A. KORINEK (2010a): “Excessive volatility in capital flows: A Pigouviantaxationapproach,”AmericanEconomicReview,403–407. ———(2018): “Managingcreditboomsandbusts: APigouviantaxationapproach,”JournalofMonetaryEconomics. 33

JEANNE, O., A. SUBRAMANIAN, AND J. WILLIAMSON (2012): Who needs to open the capitalaccount,PetersonInstitute. JENS, C. E. (2017): “Political uncertainty and investment: Causal evidence from US gubernatorialelections,”JournalofFinancialEconomics,124,563–579. JORDA`, O`. (2005): “Estimation and inference of impulse responses by local projections,” AmericanEconomicReview,95,161–182. JULIO, B. AND Y. YOOK (2012): “Political uncertainty and corporate investment cycles,” JournalofFinance,67,45–83. KALEMLI-OZCAN, S. (2019): “U.S. monetary policy and international risk spillovers,” NBERWorkingPaperNo.26297. KORINEK, A. (2018): “Regulating capital flows to emerging markets: An externality view,”JournalofInternationalEconomics,111,61–80. KORINEK, A. AND A. SIMSEK (2016): “Liquiditytrapandexcessiveleverage,”American EconomicReview,106,699–738. LANE, P. R. AND G. M. MILESI-FERRETTI (2007): “Theexternalwealthofnationsmark II:Revisedandextendedestimatesofforeignassetsandliabilities,1970–2004,”Journal ofInternationalEconomics,73,223–250. LIU, C., S. WANG, AND K. WEI (2018): “Demand shock, speculative beta, and asset prices: EvidencefromtheShanghai-HongKongstockconnectprogram,”Manuscript. LORENZONI, G.(2008): “Inefficientcreditbooms,”ReviewofEconomicStudies,809–833. MA, C. (forthcoming): “Financial stability, growth and macroprudential policy,” Journal ofInternationalEconomics. MAGUD, N. E., C. M. REINHART, AND K. S. ROGOFF (2018): “Capital controls: Myth andreality,”AnnalsofEconomicsandFinance,19,1–47. MINIANE, J. AND J. H. ROGERS (2007): “CapitalcontrolsandtheinternationaltransmissionofUSmoneyshocks,”JournalofMoney,CreditandBanking,39,1003–1035. MIRANDA-AGRIPPINO, S. AND H. REY (2019): “US monetary policy and the global financialcycle,”Manuscript. 34

OSTRY, J. D., A. R. GHOSH, M. CHAMON, AND M. S. QURESHI (2012): “Tools for managing financial-stability risks from capital inflows,” Journal of International Economics,88,407–421. OTTONELLO, P. AND T. WINBERRY (2018): “Financial heterogeneity and the investment channelofmonetarypolicy,”NBERWorkingPaperNo.24221. PETERSEN, M. A. (2009): “Estimating standard errors in finance panel data sets: Comparingapproaches,”ReviewofFinancialStudies,22,435–480. PRASAD, E.(2017): Gainingcurrency: TheriseoftheRenminbi,OxfordUniversityPress. PRASAD, E. AND S.-J. WEI (2005): “The Chinese approach to capital inflows: patterns andpossibleexplanations,”NBERWorkingPaperNo.11306. REBUCCI, A. AND C. MA (forthcoming): “Capital controls: A survey of the new literature,”OxfordResearchEncyclopediaofEconomicsandFinance. REY, H. (2015): “Dilemma not trilemma: The global financial cycle and monetary policy independence,”NBERWorkingPaperNo.21162. ROGERS, J. H., C. SCOTTI, AND J. H. WRIGHT (2018): “Unconventional monetary policyandinternationalriskpremia,”JournalofMoney,CreditandBanking. SONG, Z. AND W. XIONG (2018): “Risks in China’s financial system,” Annual Review of FinancialEconomics,10,261–286. TUCKER, J. W. (2010): “Selection bias and econometric remedies in accounting and financeresearch,”JournalofAccountingLiterature. WEI, S.-J. AND Y. ZHOU (2019): “The liberalization spillover: From equities to loans,” Work-in-progress. WOLFOLDS, S. E. AND J. SIEGEL (2019): “Misaccounting for endogeneity: The peril of relying on the Heckman two-step method without a valid instrument,” Strategic ManagementJournal,40,432–462. XING, L., K. XU, X. ZHENG, AND X. ZHANG (2018): “Stock market openness and market quality: evidence from the Shanghai-Hong Kong stock connect program,” Manuscript. 35

Appendix: Data Description Tables TableA.1Shanghai(Shenzhen)-HongKongStockConnectProgramOverview EffectiveDate AnnouncementDate Numberofstocksadded Numberofstocksonlist Nov17,2014 Apr10,2014 416 416 Dec5,2016 Aug16,2016 676 1092 NOTE: NumberofstocksincludedintheShanghai(Shenzhen)-HongKongConnectprogram inoursample. TableA.2U.S.MonetaryPolicyShock: SummaryStatistics Daily QuarterlySum QuarterlyValue-weighted Mean -0.022 -0.049 -0.026 Median -0.005 -0.018 -0.003 Std 0.119 0.164 0.105 Min -0.582 -0.571 -0.555 Max 0.295 0.326 0.196 Num 250 112 112 NOTE. TheoriginaldataisfromRogersetal.(2018). ThequarterlysumcolumntakesthesimplesumoftheirFOMCannouncement daymeasurewithinaquartertoconstructthequarterlyfrequencyseries.Thequarterlyvalue-weightedcolumntakesthevalueweighted sumwithinaquarterwheretheweightisgivenbythenumberofdaysleftinthequarter. TableA.3Firm-levelVariables: SummaryStatistics Obs Mean Std.Dev. Min Max Investment 87740 0.035 0.045 -0.069 0.426 Size 87740 21.781 1.275 11.911 28.526 Tobin’sQ 87740 2.624 1.94 0.741 26.39 CashFlow 87740 0.036 0.046 -0.331 0.315 SalesGrowth 87740 0.413 0.8 -0.978 6.173 NOTE. Thistablereportsdescriptivestatisticsforkeyvariablesusedinoursamplefrom2002to2017. Investmentdenotesthecapital expendituredividedbythebookvalueoftotalassets. Sizeisthenaturallogarithmoftotalassets. Tobin’sQistheratioofbookvalue oftotalassetsminusthebookvalueofequityplusthemarketvalueofequitybybookvalueoftotalassets. Cashflowismeasuredas earningsbeforeinterestandtaxes(EBIT)plusdepreciationandtaxesscaledbylaggedtotalassets.Salesgrowthisdefinedasthegrowth rateofsales.Allvariablesarewinsorizedatthetopandbottom1%toruleoutoutliers. A.1

TableA.4DataSample: IndustryandYearDistribution PanelA:IndustryDistribution PanelB:YearDistribution Industry #Obs #Firm Percentage Year #Obs #Firm Percentage Automobiles&Components 4523 107 4.9% 2002 1293 755 3.1% CapitalGoods 17683 467 21.5% 2003 2495 843 3.4% CommercialServices&Supplies 3051 63 2.9% 2004 2929 946 3.8% CommunicationsEquipment 2020 54 2.5% 2005 3012 951 3.8% Computer&ElectronicEquipment 5562 161 7.4% 2006 2975 959 3.9% ComputerApplication 3836 118 5.4% 2007 4397 1195 4.8% ConsumerDurables&Apparel 5499 144 6.6% 2008 4810 1289 5.2% ConsumerServices 1645 34 1.6% 2009 5031 1322 5.3% Energy 2988 70 3.2% 2010 5918 1644 6.6% Food&StaplesRetailing 319 8 0.4% 2011 7197 1953 7.9% Food,Beverage&Tobacco 5547 128 5.9% 2012 8168 2151 8.7% HealthCareEquipment&Services 773 24 1.1% 2013 8520 2172 8.8% Household&PersonalProducts 470 10 0.5% 2014 8350 2172 8.8% Materials 17394 416 19.1% 2015 7936 2169 8.8% Media 2096 56 2.6% 2016 8022 2173 8.8% MedicalBiology 7031 162 7.5% 2017 6687 2055 8.3% Retailing 2902 57 2.6% Semiconductors 456 9 0.4% TelecommunicationServices 175 4 0.2% Transportation 3770 82 3.8% Total 87740 2174 100% Total 87740 100% A.2

TableA.5VariableConstructionandDataSources Variable Definition Source PanelA:Firm-levelVariables Connect Adummyvariableindicatingwhetherthefirmiisincludedin HongKongStockExchange theConnectprograminquartert. Investment Capitalexpendituredividedbythebookvalueoftotalassets CSMAR measuredattheendofquartert-1(laggedtotalassets). Size The natrual logarithm of the book value of total assets mea- CSMAR suredattheendofquartert. MarketCap Thenaturallogarithmoftheclosepriceatquarterendmulti- CSMAR pliedbytheshareoutstandingattheendofquartert. Tobin’sQ Thebookvalueoftotalassetsminusthebookvalueofequity CSMAR plusthemarketvalueofequityscaledbythebookvalueoftotal assetsattheendofquartert. CashFlow Theincomebeforeextraordinaryitemsplusdepreciationand CSMAR amortizationdividedbythebookvalueofassets,measuredat theendofquartert. SalesGrowth Afirm’squarterlysalesgrowthrate CSMAR Leverage Thebookvalueofdebtdividedbythebookvalueoftotalassets CSMAR measuredattheendofquartert. ROA Netincomedividedbythebookvalueoftotalassetsmeasured CSMAR attheendofquartert-1(laggedtotalassets) ROE Netincomedividedbythebookvalueofshareholders’equity CSMAR measuredattheendofquartert-1(laggedtotalassets) DividendDummy Adummyvariableequalstooneifafirmpaycashdividendon CSMAR commonstockatquartert,andzerootherwise. Cash Cashandcashequivalentsdividedbythebookvalueoftotal CSMAR assetsmeasuredattheendofquartert-1(laggedtotalassets). CostofDebt The sum of short-term market borrowing rate multiple by CSMAR short-term corporate leverage ratio and long-term borrowing ratemultiplebylong-termcorporatedebtratio. Changeofln(D/P)(%) Thechangeofaggregatedyieldforeachfirmwithinquarter. CSMAR,Henry(2003) DIFCOV The difference between the historical covariance of firm i’s CSMAR,MSCI,WIND stock return with local market index and its covariance with theMSCIworldstockmarketindex(Weconvertallthereturns toRMB).Weuse36-monthrollingwindowtoconstructDIF- COVateachquarterend. GlobalCov Thehistoricalcovarianceoffirmi’sstockreturnwiththeMSCI MSCI,WIND worldstockmarketindex(WeconvertallthereturnstoRMB). Weuse36-monthrollingwindowtoconstructglobalcovarianceateachquarterend. StockVolatility Thestandarddeviationofdailystockreturnwithinaquarter. CSMAR Notethatwerequireatleast20tradingdaystoconstructthis variable. M/B Theratioofmarketvalueofassetsdividedbybookvalueof CSMAR netassets. Turnover Averageindividualturnoverratewithinamonth. CSMAR Age ThenumberofyearssinceIPO. CSMAR PanelB:MacroVariables MPSUS AmeasureforunexpectedU.S.MonetaryPolicySurpriseson Rogersetal.(2018) eachFOMCannouncement. MPSChina AmeasureforunexpectedChineseM2growthrate Chenetal.(2018) RepoRate 7-dayReporateinChina. Changetal.(2016) M2Growth Year-over-yearM2growthrate. Changetal.(2016) LocalGDPGrowth QuarterlyprovincialnominalGDPgrowthrate CEIC A.3

Online Supplement to ‘The Effect of the China Connect’ byC.Ma,andJ.Rogers,andS.Zhou December2019

FigureS.1DynamicImpactofConnectonInvestment tnecreP 50. 40. 30. 20. 10. 0 0 1 2 3 4 5 6 7 8 9 10 Quarters NOTE. TheimpulseresponsefunctionisestimatedthroughalocalprojectionmethodasinJorda` (2005). H ∑Y =α +βHConnect +ΓZ +ε it+h i it it it h=0 whereH=1,2,··· isthehorizonandZ isthefirm-levelcontrolforinvestmentequation.Thebluelineisthe it estimationparameterforβH. Thegrayareaisthe90%confidenceinterval. Allstandarderrorsareclustered atbothfirmandyearlevel. S.1

TableS.1CorporateInvestmentandFOMCShocks: Robustness Investment (1) (2) (3) (4) (5) (6) PanelA:IndustryFixedEffect PanelD:AlternativeMeasureofMonetarySurprise MPSUS*Connect -0.022* -0.024* -0.019* BRW*Connect -0.015* -0.015* -0.018* (0.013) (0.013) (0.010) (0.008) (0.008) (0.010) MPSUS -0.008** -0.011** -0.011*** BRW -0.006* -0.008** -0.008** (0.004) (0.004) (0.004) (0.003) (0.003) (0.003) Connect 0.004** -0.001 -0.024*** Connect 0.003** 0.002* 0.018*** (0.002) (0.002) (0.006) (0.001) (0.001) (0.004) Observations 87740 87740 85486 Observations 87740 87740 85486 AdjustedR2 0.168 0.219 0.224 AdjustedR2 0.386 0.410 0.412 PanelB:DropDual-listedStocks PanelE:IncludingLaggedDependentVariable MPSUS*Connect -0.020** -0.019** -0.025** MPSUS*Connect -0.019** -0.019** -0.024** (0.010) (0.009) (0.011) (0.010) (0.009) (0.010) MPSUS -0.008* -0.011** -0.011** MPSUS -0.008* -0.010*** -0.010*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Connect 0.002** 0.001 0.019*** Connect 0.002** 0.001 0.018*** (0.001) (0.001) (0.004) (0.001) (0.001) (0.004) LagDV 0.007 0.007 0.007 (0.006) (0.006) (0.006) Observations 81151 81151 79006 Observations 82532 82532 80347 AdjustedR2 0.382 0.405 0.408 AdjustedR2 0.399 0.422 0.425 PanelC:Size PanelF:IncludingLaggedMonetaryPolicyShock MPSUS*Connect -0.017* -0.016* -0.021** MPSUS*Connect -0.021** -0.022** -0.026** (0.010) (0.009) (0.010) (0.010) (0.009) (0.010) MPSUS*Size -0.002 -0.002 -0.002 MPSUS -0.009* -0.012** -0.012** (0.003) (0.002) (0.003) (0.005) (0.005) (0.005) MPSUS 0.040 0.029 0.034 LagMPSUS*Connect -0.002 -0.006 -0.002 (0.067) (0.054) (0.057) (0.004) (0.004) (0.006) Connect 0.001 0.001 0.017*** LagMPSUS -0.004 -0.004 -0.004 (0.001) (0.001) (0.004) (0.005) (0.005) (0.005) Size 0.004*** 0.004*** 0.003*** Connect 0.002* 0.001 0.017*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.004) Observations 87740 87740 85486 Observations 87740 87740 85486 AdjustedR2 0.388 0.410 0.413 AdjustedR2 0.387 0.410 0.413 FirmControls No Yes Yes FirmControls No Yes Yes IMR No No Yes IMR No No Yes FirmFE Yes Yes Yes FirmFE Yes Yes Yes YearFE Yes Yes Yes YearFE Yes Yes Yes QuarterDummy Yes Yes Yes QuarterDummy Yes Yes Yes NOTE: thedependentvariableiscorporateinvestment,definedasquarterlycapitalexpenditurescaledbythe beginning-of-quarterbookvalueoftotalassets.PanelAuseindustryfixedeffectsinsteadoffirmfixedeffects. Panel B drops A-H and A-B dual listed stocks. Panel C controls for the size on the investment sensitivity toU.S.monetarypolicyshock. PanelDusesanalternativemonetarypolicyshock(BRW)identifiedbyBu etal.(2019). PanelEcontrolsforlaggedcorporateinvestment. PanelFcontrolsforalaggedmonetarypolicy shock. Allstandarderrorsareclusteredatbothfirm(industry)andyearlevelandreportedinparentheses. ∗, ∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. S.2

TableS.2CorporateInvestmentandChineseMonetaryPolicyShocks Investment (1) (2) (3) (4) (5) (6) MPSChina*Connect -0.130 -0.170 -0.146 -0.129 -0.160 -0.137 (0.192) (0.212) (0.225) (0.159) (0.177) (0.172) MPSChina 0.091 0.121* 0.134* 0.056 0.073 0.088 (0.070) (0.073) (0.073) (0.078) (0.083) (0.081) Connect 0.002** 0.001 0.017*** 0.001 0.000 0.017*** (0.001) (0.001) (0.005) (0.001) (0.001) (0.004) MPSUS*Connect -0.021** -0.021** -0.026** (0.010) (0.009) (0.011) MPSUS -0.007 -0.010** -0.010** (0.005) (0.005) (0.005) Size 0.004*** 0.003*** 0.005*** 0.003*** (0.001) (0.001) (0.001) (0.001) LagTobin’sQ 0.002*** 0.001*** 0.002*** 0.001*** (0.000) (0.000) (0.000) (0.000) CashFlow 0.172*** 0.169*** 0.173*** 0.170*** (0.011) (0.012) (0.011) (0.012) SalesGrowth 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) GDPGrowth 0.023 0.026 0.026* 0.029* (0.017) (0.017) (0.016) (0.016) IMR No No Yes No No Yes FirmFE Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes QuarterDummy Yes Yes Yes Yes Yes Yes Observations 87740 87740 85486 87740 87740 85486 AdjustedR2 0.386 0.409 0.412 0.387 0.410 0.413 NOTE: the dependent variable is corporate investment. The Chinese monetary policy shock MPSChina is the quarter-over-quarter (QoQ) change of M2 growth rate shock identified by Chen et al. (2018). Detailed informationonthecontrolscanbefoundinAppendixA.5. Allstandarderrorsareclusteredatbothfirmand yearlevelandreportedinparentheses. ∗,∗∗ and∗∗∗ indicatestatisticalsignificanceatthe10%,5%,and1% level,respectively. S.3

TableS.3CorporateInvestment,U.S.MonetaryPolicyShocks,andtheChinaConnect: AlternativeDefinitionofConnectDummy Investment (1) (2) (3) (4) (5) (6) Connect 0.003** 0.003*** 0.004*** 0.002* 0.002** 0.004*** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) MPSUS*Connect -0.017* -0.018* -0.019* (0.009) (0.010) (0.010) MPSUS -0.009* -0.011** -0.011*** (0.004) (0.004) (0.004) Size 0.004*** 0.004*** 0.005*** 0.004*** (0.001) (0.001) (0.001) (0.001) LagTobin’sQ 0.002*** 0.001*** 0.002*** 0.002*** (0.000) (0.000) (0.000) (0.000) CashFlow 0.172*** 0.173*** 0.173*** 0.174*** (0.011) (0.011) (0.011) (0.011) SalesGrowth 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) GDPGrowth 0.019 0.020 0.024 0.026* (0.016) (0.016) (0.015) (0.015) IMR No No Yes No No Yes FirmFE Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes QuarterDummy Yes Yes Yes Yes Yes Yes Observations 87740 87740 85486 87740 87740 85486 AdjustedR-squared 0.386 0.409 0.411 0.386 0.410 0.412 NOTE: thedependentvariableiscorporateinvestment,definedasquarterlycapitalexpenditurescaledbythe beginning-of-quarterbookvalueoftotalassets. Connect equals1ifafirmsiisintheConnectforquartert it and0otherwise. DetailedinformationonthecontrolscanbefoundinAppendixA.5. Allstandarderrorsare clusteredatbothfirmandyearlevelandreportedinparentheses. ∗,∗∗and∗∗∗indicatestatisticalsignificance atthe10%,5%,and1%level,respectively. S.4

TableS.4CorporateInvestment,U.S.MonetaryPolicyShocks,andtheChinaConnect: EliminatePeriodicAdjustmenttoIndexes Investment (1) (2) (3) (4) (5) (6) Connect 0.002* 0.003** 0.004** 0.002 0.002* 0.004** (0.001) (0.001) (0.002) (0.002) (0.001) (0.002) MPSUS*Connect -0.015* -0.017* -0.017* (0.009) (0.010) (0.009) MPSUS -0.009* -0.011*** -0.011*** (0.004) (0.004) (0.004) Size 0.004*** 0.004*** 0.004*** 0.004*** (0.001) (0.001) (0.001) (0.001) LagTobin’sQ 0.001*** 0.001*** 0.002*** 0.001*** (0.000) (0.000) (0.000) (0.000) CashFlow 0.165*** 0.166*** 0.166*** 0.167*** (0.011) (0.011) (0.011) (0.012) SalesGrowth 0.001** 0.001** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) GDPGrowth 0.029 0.031* 0.035** 0.037** (0.018) (0.018) (0.017) (0.016) IMR No No Yes No No Yes FirmFE Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes QuarterDummy Yes Yes Yes Yes Yes Yes Observations 76309 76309 74363 76309 76309 74363 AdjustedR-squared 0.392 0.413 0.414 0.392 0.414 0.415 NOTE: thedependentvariableiscorporateinvestment. WekeeponlystocksthatareaddedtotheConnectin 2014Q4and2016Q4andstocksthatareneveraddedtotheConnect. Detailedinformationonthecontrols canbefoundinAppendixA.5. Allstandarderrorsareclusteredatbothfirmandyearlevelandreportedin parentheses. ∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. S.5

TableS.5CorporateInvestment,U.S.MonetaryPolicyShocks,andtheChinaConnect: withMacroControls Investment (1) (2) (3) (4) (5) (6) Connect 0.002** 0.001* 0.016*** 0.002* 0.001 0.017*** (0.001) (0.001) (0.005) (0.001) (0.001) (0.004) MPSUS*Connect -0.021** -0.020** -0.025** (0.009) (0.009) (0.010) MPSUS -0.007* -0.010** -0.010** (0.004) (0.004) (0.004) LagRepoRate 0.206* 0.189 0.170 0.166 0.129 0.106 (0.118) (0.124) (0.124) (0.117) (0.127) (0.128) LagM2Growth 0.013 0.005 0.000 0.017 0.011 0.005 (0.024) (0.028) (0.029) (0.024) (0.027) (0.026) Size 0.004*** 0.003*** 0.005*** 0.003*** (0.001) (0.001) (0.001) (0.001) LagTobin’sQ 0.002*** 0.001*** 0.002*** 0.001*** (0.000) (0.000) (0.000) (0.000) CashFlow 0.171*** 0.168*** 0.173*** 0.169*** (0.011) (0.012) (0.011) (0.012) SalesGrowth 0.001*** 0.001*** 0.001*** 0.001*** (0.000) (0.000) (0.000) (0.000) GDPGrowth 0.022 0.025 0.023 0.026 (0.019) (0.019) (0.018) (0.019) IMR No No Yes No No Yes FirmFE Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes QuarterDummy Yes Yes Yes Yes Yes Yes Observations 87740 87740 85486 87740 87740 85486 AdjustedR-squared 0.386 0.409 0.412 0.387 0.410 0.413 NOTE: thedependentvariableiscorporateinvestment. MacrocontrolsincludetheM2growthrateand7-day Repo rate in addition to the local GDP growth rate. Detailed information on the controls can be found in AppendixA.5. Allstandarderrorsareclusteredatbothfirmandyearlevelandreportedinparentheses. ∗,∗∗ and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. S.6

TableS.6CorporateInvestmentandFOMCShocks: FirmHeterogeneity Investment PanelA:Tradable(High)v.s.Non-tradable(Low) PanelB:ForeignSales>25%(High) High Low High Low High Low High Low MPSUS*Connect -0.021** -0.016* -0.026** -0.020* MPSUS*Connect -0.008 -0.018* -0.010 -0.024** (0.009) (0.009) (0.011) (0.010) (0.010) (0.010) (0.013) (0.011) MPSUS -0.012** -0.009** -0.012** -0.009** MPSUS -0.011 -0.011*** -0.011 -0.011*** (0.005) (0.004) (0.005) (0.004) (0.007) (0.004) (0.007) (0.004) Connect 0.001 0.002 0.019*** 0.014*** Connect 0.004** 0.001 0.027*** 0.018*** (0.001) (0.001) (0.005) (0.005) (0.002) (0.001) (0.007) (0.004) Observations 58466 29274 56929 28557 Observations 6826 79584 6644 77524 AdjustedR2 0.410 0.414 0.413 0.417 AdjustedR2 0.435 0.415 0.439 0.418 FirmControls No No Yes Yes No No Yes Yes IMR No No Yes Yes No No Yes Yes FirmFE Yes Yes Yes Yes Yes Yes Yes Yes YearFE Yes Yes Yes Yes Yes Yes Yes Yes QuarterFE Yes Yes Yes Yes Yes Yes Yes Yes H0:βH=βL χ2 0.49 0.91 0.93 1.59 P-value 0.484 0.339 0.336 0.208 NOTE. The dependent variable is corporate investment. Panel A divides the firms into tradable and nontradable sectors. Panel B divides the firms into two groups according to the median level of foreign sales share,definedastheshareofforeignsalestototalsales,ateachquarter. Allstandarderrorsareclusteredat bothfirmandyearlevelandreportedintheparentheses. ∗, ∗∗ and∗∗∗ indicatestatisticalsignificanceatthe 10%,5%,and1%level,respectively. S.7

TableS.7CashHoldingsandFOMCShocks ∆CashHoldings (1) (2) (3) Connect -0.003*** -0.002** -0.046*** (0.001) (0.001) (0.004) MPSUS*Connect 0.036*** 0.038*** 0.048*** (0.011) (0.011) (0.011) MPSUS -0.001 -0.007*** -0.007*** (0.002) (0.002) (0.002) Size 0.005*** 0.010*** (0.001) (0.001) CashFlow 0.288*** 0.293*** (0.015) (0.015) LagTobin’sQ 0.005*** 0.006*** (0.000) (0.000) Leverage 0.060*** 0.051*** (0.003) (0.003) Invest -0.111*** -0.102*** (0.012) (0.012) Dividend 0.002 0.001 (0.001) (0.001) IMR No No Yes FirmFE Yes Yes Yes YearFE Yes Yes Yes QuarterDummy Yes Yes Yes Observations 80337 80337 78225 AdjustedR-squared 0.006 0.030 0.032 NOTE: Cashholdingsaredefinedasquarterlycashholdingsscaledbythebeginning-of-quarterbookvalue oftotalassets. Thedependentvariableisthequarterlychangeofcashholdings. Detailedinformationonthe controls can be found in Appendix A.5. All standard errors are clustered at firm level and reported in the parentheses. ∗,∗∗and∗∗∗indicatestatisticalsignificanceatthe10%,5%,and1%level,respectively. S.8