* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Mutual fund intermediation, equity issues, and the real
Survey
Document related concepts
Fundraising wikipedia , lookup
Land banking wikipedia , lookup
Financialization wikipedia , lookup
Stock trader wikipedia , lookup
Private equity in the 1980s wikipedia , lookup
Syndicated loan wikipedia , lookup
Public finance wikipedia , lookup
Private equity wikipedia , lookup
Stock selection criterion wikipedia , lookup
Private equity in the 2000s wikipedia , lookup
Early history of private equity wikipedia , lookup
Fund governance wikipedia , lookup
Private equity secondary market wikipedia , lookup
Transcript
Mutual fund intermediation, equity issues, and the real economy∗ Bill Yiqing Zu† University of Melbourne August 2014 Abstract I investigate the real economic effect of mutual funds. As informed financiers, mutual funds have superior ability to identify productive firms, and can allocate capital more efficiently in equity issues. A simple model of mutual fund intermediation demonstrates a positive relation between mutual fund investment in equity issues and subsequent output growth. Empirically, at both the aggregate and industry level, I find that new equity issues lead to higher real output, only when there is high mutual fund participation in these issues. In addition, the participation-output sensitivity is stronger when fund flows are low. At the firm-level, mutual fund investment in seasoned offerings positively predicts subsequent firm productivity growth. Further tests suggest that the results are unlikely to be driven by reverse causality explanations. ∗ I am greatly indebted to Neal Galpin, Bruce Grundy, Spencer Martin and Juan Sotes-Paladino for their extensive help and support, and would also like to thank Antonio Gargano, Simon Gervais, Vincent Gregoire, Patrick Kelly, Robert Korajczyk, Tong Yob Nam (discussant), Selim Topaloglu, and seminar participants at the University of Melbourne Brown Bag and the 26th Australasian Finance and Banking Conference 2013 for their valuable comments. I thank the Investment Company Institute (ICI) for providing the aggregate fund flows dataset. All errors are my own. † Ph.D. Candidate, Department of Finance, University of Melbourne, Australia. Email: [email protected] 1 Introduction The mutual fund industry has grown immensely over the past 20 years. As the largest group of institutional investors, the industry’s total assets under management has grown from only $134 billion in 1980 to about $12 trillion at the end of 2007, accounting for approximately 33% of the total value of the U.S. stock market. The vast majority of academic studies on mutual funds have focused on their secondary market activities, and in particular, their ability to deliver superior returns above benchmarks. However, the social benefit of active management does not come from the ability to earn excess returns – the average fund has to hold the market portfolio. Instead, the real economic benefit of asset managers should be in the forms of increased firm value and productivity growth (Greenwood and Scharfstein, 2013). Despite their importance, the real economic effects of mutual funds have received much less attention in the academic literature. In this paper, I investigate this issue by examining the relation between mutual fund investment and real economic activities, through the channel of primary market equity issues. As informed financiers, mutual funds have superior screening abilities in the primary market (Gibson, Safieddine, and Sonti, 2004; Chemmanur, He, and Hu, 2009). I argue that with this screening ability, mutual funds can allocate capital more efficiently, and on average would finance the more productive firms in the economy. As a result, mutual fund investment conveys information about the economy’s productivity, and real output is higher when more equity issues are financed by mutual funds. In this regard, mutual funds have a positive role in the real economy, in their capacity as informed financial intermediaries. The question of whether mutual funds have real effects is important to examine, but the research in this area remains scarce. As the largest group of institutional investors, mutual funds generate large trading volume and earn high fees from asset management. There is much debate regarding their stock-selection or market-timing skills (see, for example, Elton and Gruber (2013) for an extensive review). However, little is known about the real economic benefit of these activities. Mutual funds may exert influence on the real economy, either through their primary market investment or through their secondary market activities. Arguably, informed trading in the secondary market helps to incorporate information into prices, thereby enhancing price efficiency and hence capital allocation efficiency (Fama, 1970; 2 Grossman and Stiglitz, 1980).1 However, this is a second-order effect, and the more direct influence of mutual funds through their primary market investment as informed financiers has received much less attention, especially regarding their real effect on firms and at the aggregate level. This paper contributes to the literature by demonstrating the positive real effects of mutual funds through this primary market channel. I first present a simple one-period, two-sector model of mutual fund intermediation. In the model, firms of differing productivities raise equity to finance their investment projects. Individual investors cannot differentiate between firms’ productivities, nor do they know the true productivity level. On the other hand, mutual funds have the ability to ex-ante differentiate between firms at a cost, and they also know the true productivity. Individuals choose to allocate their wealth between mutual funds and direct investment in firms. More allocation to mutual funds would entail better investment returns, but individuals would also incur the convex information cost. In equilibrium, fund flows increase with individuals’ expected productivity, but the realization of output is determined by mutual funds’ investment (because funds are assumed to have perfect information about productivity). Higher investment by mutual funds is associated with higher subsequent real output, because mutual funds only invest when they know that productivity is high. Furthermore, if individuals’ incorrectly perceive productivity to be low and thus allocate less fund flows, there would be an even higher return to mutual funds’ investment, because funds are constrained to only the best projects. The model therefore helps to establish a relation between real output, mutual fund investment and fund flows that is empirically testable. Empirically, I find evidence that is largely consistent with the model’s predictions. Using U.S. data during the period from 1984 through 2011, I find that mutual fund participation2 in new equity issues (seasoned equity offerings and initial public offerings) positively predicts subsequent issue-output sensitivity, after controlling for the total amount of issues and other macroeconomic variables. When mutual funds are actively participating in new issues, the issues are able to generate more real output, reflecting the higher productivity of the firms that receive funding from mutual funds. Moreover, when fund flows are low, the fund participationgrowth relation is stronger. This is also consistent with the model’s prediction. When there is low fund flows but high fund participation, it suggests that individuals and fund managers 1 2 Recently, Piacentino (2013) presents a theoretical model in this spirit. Participation is measured as change in mutual fund holdings. 3 disagree about productivity. If fund managers have better information, their investment based on the constrained capital would produce higher returns, generating the “flight-to-quality” effect (Holmstrom and Tirole, 1997). Since fund flows and equity issues are cyclical in nature, I test the model separately on the overall growth in output as well as the cyclical component of output. The results predominantly show up in the cyclical component of output, rather than the long-run trend.3 Further analysis at the industry-level confirm the aggregate results. Industries that receive higher mutual fund participation in their equity issues have higher subsequent issue-output sensitivity. In addition, the “flight-to-quality” effect is also observed at the industry level, where high fund participation leads to higher cyclical growth especially when fund inflows are lower. At the micro level, firms with highly participated issues by mutual funds exhibit higher subsequent total factor productivity growth, after controlling for firm-specific characteristics. This complements and offers further support for the aggregate and industry-level evidence, and is consistent with the model assumption where mutual funds identify the most productive firms to invest. Taken together, the evidence is consistent with the hypothesis that mutual funds have superior ability to identify productive equity-issuing firms, and have positive roles in the economy by investing in these firms. Given these results, one may have concerns regarding the potential reverse causality explanations. Specifically, one may argue that mutual funds do not have superior skills. Instead, the results may be driven by the fact that mutual funds simply invest more when they have more fund flows, and that more fund flows occur when anticipated growth is high. This anticipation effect could induce a positive correlation between fund participation and future productivity, even when mutual funds do not possess superior skills. There are two observations that may alleviate this concern. First, if this story is true, it would imply that fund flows and mutual fund participation are highly correlated. This is not the case in the data. The contemporaneous correlation between aggregate fund flows and mutual fund participation is only about 0.16. Thus, although fund flows contain some information about future productivity, mutual fund participation contains additional information independent of fund flows. Furthermore, in the regressions, I control for fund flows to capture this passive effect. 3 The empirical tests here do not have implications for cross-sectional long-run growth effects, since there is only one country studied. 4 Secondly, results show that when fund flows are low, higher participation leads to higher subsequent growth. This is inconsistent with the reverse causality story above. If mutual funds are simply “riding the wave”, lower fund flows would lead to lower participation, and future growth would be lower because of low productivity. This would imply either a positive relation or no relation between future growth and the interaction term of fund flows and participation. Instead, the results show a strong negative relation, which suggests that mutual funds have different information than individuals, and their investment contains superior information about future productivity, especially during low fund flow times. In order to further examine this alternative explanation, I separate the mutual fund participation into two groups – participation by active funds versus participation by passive funds. If the positive relation between mutual fund participation and firm productivity is indeed due to funds’ screening ability, one would expect the effect to predominantly come from active funds rather than passive funds. Indeed, the results show that most of the effects are driven by active funds’ participation, and passive funds’ participation have little predictive power of future productivity. These results lend further support to the model’s hypotheses, and suggest that the results are unlikely to be driven by reverse causality explanations. The rest of the paper is structured as follows. I firstly review the related literature in section 2. I then present the simple model of mutual fund intermediation in section 3, and provide the empirically testable implications. Section 4 describes the data and variables used. In section 5, I present and discuss the empirical results, along with some robustness tests and discussions about alternative explanations. Section 6 concludes. 2 Related Literature The role of mutual funds in the real economy can be largely categorized into primary market investment and secondary market trading. In the primary market, a large body of empirical literature examines the role of financial institutions as active monitors, and find that the stock market reacts positively towards pension fund activism announcements (Gillan and Starks, 2000), but there is little evidence of long-run improvement in stock returns or operating performance (Smith, 1996; Del-Guercio and Hawkins, 1999). In contrast, Brav, Jiang, Partnoy, and Thomas (2008) find an average abnormal return of 7% around the announcement of hedge fund activism activities, and they attribute the results as due to differences 5 in the governance and incentive structure of hedge funds versus pension funds. Although the evidence on institutions’ monitoring capacity is mixed, there is strong evidence of their screening ability. Several papers have shown, using different datasets, that institutions have superior information and are able to identify above-average seasoned equity offering firms with long-run abnormal returns (Gibson, Safieddine, and Sonti, 2004; Chemmanur, He, and Hu, 2009; Demiralp, D’Mello, Schlingemann, and Subramaniam, 2011). The empirical evidence of institutional monitoring and screening roles are perfectly consistent with the theory of financial intermediation. In order to overcome information problems created by differently informed traders, one of the major services provided by financial intermediaries is to monitor, screen, and certify firm investment (Bhattacharya and Thakor, 1993). By having intermediaries that specialize in information acquisition and allowing them to capture a return on the information (by holding the valuable assets), one can mitigate the public good problem, and thus enable the funding of some positive net present value projects which would otherwise be unfunded due to high information costs (Leland and Pyle, 1977; Diamond, 1984). Incorporating this idea, Holmstrom and Tirole (1997) examine the role of intermediaries in the real economy. In their model, intermediaries perform the monitoring role and commit their own capital as part of the monitoring process. In times of intermediary capital shocks, the economy faces a reduction in the amount of valuable projects funded, and this creates a flightto-quality effect where intermediary capital is rationed to higher-quality firms. Schumpeterian models of finance and growth also argue that financial intermediaries and markets help to alleviate information problems and allocate capital more efficiently, thus promoting long-run growth (Greenwood and Jovanovic, 1990; King and Levine, 1993b). Empirical studies have found strong support for the causal relation between financial development and economic growth. At the country level, most of the empirical tests examine the banking sector for measures of financial development (King and Levine, 1993a; Beck and Levine, 2002), and find that financial development leads to higher economic growth in crosscountry regressions. Levine (2005) provides an extensive survey of the empirical results along this line of research. For firm-level evidence in the U.S., Chava and Purnanandam (2011) use the Russian crisis in 1998 as an exogenous shock to the U.S. banking sector, and find that bank-dependent firms experience larger valuation losses and reductions in real investment. Lemmon and Roberts (2010) examine regulatory changes in the banking sector that adversely 6 affect below-investment-grade firms’ borrowing ability, and find that reductions in firms’ bank borrowing lead to an almost one-to-one reduction in net investment. Ivashina and Scharfstein (2010) and Duchin, Ozbas, and Sensoy (2010) provide evidence of reduced bank lending and real investment after the financial crisis in 2008, suggesting that capital supply constraints have real effects. In addition to the banking sector, Bhattacharya and Thakor (1993) suggest that ratings agencies also act as intermediaries in providing screening and certification services. In this spirit, Faulkender and Petersen (2006) find that firms that have bond ratings have significantly higher leverage. Similarly, Sufi (2009) find that the introduction of syndicate loan ratings leads to an increase in firms’ subsequent asset growth, cash acquisitions, and investment. These results suggest that ratings agencies provide valuable certification to the bond market to reduce information problems. Despite the large body of literature on the relation between credit intermediaries and the real economy, limited attention has been paid to the real economic role of other financial intermediaries such as mutual funds, albeit them being one of the largest group of equity investors.4 Most of the studies on mutual funds have focused on their secondary market effects. As mentioned previously, one of the main activities of mutual funds is to enhance price efficiency. In this spirit, Piacentino (2013) presents a market microstructure model where career-concerned delegated portfolio managers acquire more information and trade more than individual investors, due to their additional incentives to signal through their trades. This induces more private information being impounded into prices, thereby improving price efficiency and capital allocation efficiency. This result is consistent with the empirical evidence on institutional investors and price efficiency (Boehmer and Kelley, 2009), and suggest that active portfolio managers may have positive externalities on real economic activities. Although there is little research that links the primary equity market to mutual funds, this is a relevant question to examine. From a theoretical perspective, equity market intermediaries are similar to banks in their capacity as monitors and alleviating the public-good problem (Shleifer and Vishny, 1986). Moreover, if intermediation helps to reduce information problems, one should expect them to be most useful in equity markets, where information problems are more severe than debt markets. Although there is some empirical evidence of 4 Lewellen (2011) shows that financial institutions together hold around 68% of the U.S. equity market value in 2007, and approximately half are held by mutual funds. 7 superior institutional monitoring and screening abilities through their investment returns, one cannot conclude that this directly leads to improvements in economic growth, for the following reasons. First, there is little evidence of how institutional behavior affects firms’ real investment decisions and productivity, which is central to the link between finance and growth. As argued by Levine (2005), in order for financial development to influence growth, institutions need to influence resource allocation decisions that foster productivity growth, rather than simply promoting physical capital accumulation. Brav, Jiang, and Kim (2011) is one of the first papers to examine these real effects, and they find significant increases in total factor productivity of target firms within two years after hedge fund activism. The authors also find that systematic risks of target firms decrease within three years post-activism event. The paper offers important insights in demonstrating the role of institutions through the productivity channel, in addition to the effect on stock returns and operating performance. However, increases in the productivity of individual firms do not imply growth at the aggregate level, since competition could result in a loss of output by other unproductive firms. Moreover, studies on shareholder activism can only capture the monitoring aspect of institutions expost, but not the screening or certification effects ex-ante. To provide evidence for the latter, one needs to examine the extent of institutional participation in primary markets, both in aggregate and at a micro level. Second, theoretical models such as Holmstrom and Tirole (1997), Chen (2001) and Bernanke and Gertler (1989) suggest that in a general equilibrium setting, it is the capital shocks to institutions or productivity shocks that generate the macroeconomic dynamics and not just their participation (although participation by institutions certainly improves output by facilitating better investment). In Bernanke and Gertler (1989), productivity shocks reduce firms’ collaterals, and thus prevent them from raising capital for further investment, especially during economic downturns when information asymmetry problems are more severe. In Holmstrom and Tirole (1997), capital shocks to financial intermediaries hamper their ability to successfully monitor projects, thus reducing the amount of real investment in the economy. In addition, real business cycle models have also examined the impact of financial constraints/shocks on firm investment and the real economy, showing that financial shocks or disintermediation can have large effects in amplifying productivity shocks and generating business cycles (Jermann and Quadrini, 2012). Therefore, when one looks at equity intermediation, capital constraint 8 of intermediaries is also important. Empirically, a natural proxy for mutual funds’ capital constraints is fund flows. Large fund outflows would reduce the amount of discretionary cash that funds have available to invest in equity issues, and when they are unable to fund all the positive NPV investment, they are forced to choose only the most profitable ones. On the other hand, large fund inflows may force funds to purchase stocks even when trading opportunities are scarce (Khan, Kogan, and Serafeim, 2012), and this could potentially reduce the efficiency of fund investment. Thus, in order to examine the real effects, it is important to take into consideration funds’ capital availability, both theoretically and empirically. I now present a simple model that illustrates how mutual funds can enhance real activities through their informational advantages, and derive some empirical implications. 3 A simple model The model economy has one period, and consists of two production sectors, a representative household, and a representative mutual fund. Each sector is described in detail below. 3.1 Production technology There are two production sectors in equal mass, denoted as sector H and sector L. Firms in each sector raise equity capital to fund their investment projects, which pay off at the end of the period. The two sectors differ in their productivities. The L sector has productivity AL , while the H sector has productivity AL + θ, with θ > 0. Both sectors have constantreturns-to-scale production technologies that take capital (which I denote as I) as the only input: YL = AL I, YH = (AL + θ)I (1) Since these two sectors produce the consumption good, I refer to them as the “primary market”. In addition to the primary market, there is also a secondary market that trades shares in existing investment projects. One could think of the secondary market as an index fund in the stock market. These existing projects will deliver some fixed level of output, regardless of who is owning the shares. I denote the amount of capital stock in the secondary market as S. I assume that the secondary market has an expected return that is equal to the average of the two primary sectors, so that without information advantages, the risk neutral 9 investors are indifferent between investing in the primary market and secondary market. This assumption implies that the secondary market return Rm is given by Rm = AL + θ/2, and the existing projects will produce Rm S of output at period end. Further, I assume that the average investment project is profitable, i.e. AL + θ 2 > 1, therefore all investors are willing to participate in either the primary market or the secondary market. 3.2 Mutual fund sector There is one representative mutual fund in the economy that can invest on behalf of the representative household. The mutual fund cannot take leverage, and thus can only invest up to the amount of fund flows it receives, which I denote as F . The mutual fund has access to a technology that can identify the more productive sector H, and therefore allows them to only invest in that sector and avoid sector L. To obtain the information, the mutual fund incurs cost, and I assume that the cost is convex in the amount of investment, in the spirit of Berk and Green (2004). One can think of the convex cost structure as giving the fund investor diminishing returns in their mutual fund investment. Specifically, for an investment of f into the H sector, the mutual fund incurs 21 φf 2 of information cost, where φ is a cost parameter. Ultimately this cost is borne by the investor. I also assume that the mutual fund knows the true value of θ. In addition to investing in the primary market, the mutual fund can also invest some fund flows into the secondary market. I denote the primary market investment by mutual funds as f , and thus the secondary market investment is given by F − f . Since there is no screening in the secondary market, no information cost is incurred on these investments.5 I assume that there is no agency problem between the mutual fund and the household. The fund will seek to maximize the return to households. Given their fund flows F , the mutual fund solves the following problem: max 0≤f ≤F 1 (AL + θ)f − φf 2 + (F − f )Rm | {z } | {z 2 } Primary market where Rm (2) Secondary market θ = AL + 2 5 I do not explicitly model the supply of the secondary market shares, and assume that there are liquidity traders who supply shares to the mutual fund. The liquidity traders do not participate in the primary market, and hence do not affect real output. 10 The fund flows F is determined endogenously by the household, which I describe below. 3.3 Households There is one representative household that maximizes period-end consumption. The household is risk-neutral, and has initial wealth W , which it can invest in the primary market to generate future consumption.6 The household is uninformed, in the sense that it cannot differentiate between the two production sectors, nor does the household know the true value of θ. Instead, the household has its own information about θ, which I denote as θh , and all of the household’s investment decisions are based on θh .7 Further, the household does not try to infer the true θ from mutual funds (they firmly believe in their own θh ), therefore the household only expects mutual funds to invest in the H sector, given the fund flows they provide. The household can choose to invest through the mutual fund sector by giving F to the mutual fund, or invest directly in the primary market. The household perceives that any investment through the mutual fund is guaranteed to provide an expected return of AL + θh , less the information cost. On the other hand, investment made directly in the primary market will earn an expected return of AL + θh /2. Given their initial wealth W , the household solves the following problem: max 0≤F ≤W E(RM F (F )) + D × E(RD ) (3) s.t. W = F + D where E(RM F (F )) represents the expected return from mutual fund investment, given the fund flows, and E(RD ) is the expected return from direct investment. Since the household only expects the mutual fund to invest in the H sector, with the convex cost structure and 6 I assume that the household cannot invest in the secondary market. By assumption, the primary market and the secondary market delivers the same expected return to the household. These assumptions are to simplify the model, and do not change the model implications. 7 Here I do not explicitly model the household’s information acquisition problem, nor do I specify why the household has different information from the mutual fund. If households and mutual funds have the same information, households would be able to allocate the perfect amount of capital to funds, and fund flows would not have an impact on funds’ real effects. 11 with E(RD ) = AL + θh /2, the above problem becomes: max 0≤F ≤W s.t. 1 2 θh (AL + θ )F − φF + D AL + 2 2 h (4) W =F +D where D denotes the direct investment, and F denotes the household’s fund flows. 3.4 Equilibrium analysis and model implications I now solve for the mutual fund and the household’s optimization problems. The household’s problem in equation (4) along with the budget constraint imply the first order conditions: F∗ = θh 2φ D∗ = W − (5) θh 2φ (6) For the fund manager, the optimization problem in equation (2) implies: f= θ 2φ However, since the mutual fund cannot take leverage, the fund’s investment in primary market is bounded by the amount of fund flow it receives, F , which in turn is determined by the household’s perception of investment opportunities, θh . Therefore, the actual level of mutual fund investment in the primary market is given by: ∗ f = min θ θh , 2φ 2φ (7) A simple comparison between (7) and (5) shows that if the household has perfect information about investment opportunities, that is, if θh = θ, then the household will provide the fund flows that match with the optimal investment strategy of the mutual fund, i.e. f = F ∗ . However, if the household has different (and incorrect) information about investment opportunities, then the supply of funds to the mutual fund sector may not coincide with the fund manager’s investment strategy. Specifically, if the household perceives productivity to be low, and allocates too little capital to the mutual fund relative to what the mutual fund desires, this may carry negative consequences for the aggregate economy, since some of the capital 12 would be unnecessarily invested in the unproductive sector. The model shows that the extent to which the fund manager invests in the primary market will provide some information about investment opportunities, as long as the mutual fund has superior information relative to households. To examine the aggregate effect of mutual funds’ information advantages, I define aggregate output in the economy by: 1 2 θ Y = (AL + θ)f − φf + (W − F ) AL + + Rm S 2 2 (8) where the first two terms represent output attributed to mutual fund investment less information costs, the third term represents output produced by direct investment by the household, and the last term represents output from the existing projects (the secondary market). As noted above, mutual fund investment f contains some information about investment opportunities θ. However, the extent to which this is true may depend on the size of θh versus θ. In particular, (7) shows that if θf ≤ θ, mutual fund investment f would reflect the household’s perceived productivity, but if θf > θ, f would reflect the true productivity. This suggests that aggregate output may take two possible forms in terms of the observables, which I analyze separately. Case 1: θh ≤ θ and F = f . In this case, one can substitute F = f into equation (8) and obtain 1 Y = (AL + θ)f − φf 2 + W 2 θ 1 2 = f − φf + W AL + 2 2 θ θ AL + − f AL + + Rm S 2 2 θ + Rm S 2 (9) Aggregate output is increasing in the linear term of mutual fund investment f , and is decreasing in the square term of fund investment f 2 , due to the convex information cost. Further, since output is determined by the actual productivity θ and not the household’s perceived θh , fund flows do not play a part in determining output here. Case 2: θh > θ and F > f . In this case, mutual fund investment f is determined by the fund’s information about true productivity θ. Therefore, one can rewrite the aggregate 13 output in equation (8) in terms of the observable f , given that θ = 2φf from equation (7): 1 2 2φf + Rm S Y = (AL + 2φf )f − φf + (W − F ) AL + 2 2 3 = (AL + W φ)f + φf 2 − φF f + W AL − F AL + Rm S 2 (10) Consistent with the first case, equation (10) shows that mutual fund investment in primary issues conveys information about the true productivity θ. In particular, aggregate output is positively related to the linear term of f . This is due to fund’s information advantage and its ability to invest in the more productive sector, so that its investment decision reveals future productivity and hence output. This is the key prediction of the model. As shown above, this result does not depend on the relative size of θh and θ, i.e. the extent to which the household overestimates or underestimates productivity. Another interesting prediction is in the third term in equation (10), where the interaction term between fund flows F and fund primary investment f has a negative effect on aggregate output. As fund flow F decreases, the sensitivity of aggregate output to mutual fund investment f becomes stronger. Note that equations (10) and (9) provide similar predictions about the relation between mutual fund investment and output, albeit different in magnitude. In addition, equation (10) provides additional predictions on the interaction effect between fund flows and mutual fund investment. Hence, in the empirical tests I will adopt (10) as the testing equation.8 Although the variables f, Rm , and F in equation (10) are all empirically observable, one problem with estimating the equation is that the variables are in levels. Therefore I scale equation (10) by initial wealth W to obtain a relation with stationary variables: Y f 3 f2 φF f F S = AL + (AL + W φ) + φ − − AL + Rm W W 2 W W W W (11) Equation (11) can then be tested using empirical counterparts of f /W, F/W , and Rm . This relation should hold at both the aggregate and industry level, since in the model, the sectors could either be within an industry, or different industries within the whole economy. I summarize these predictions in the hypotheses below: 8 One may argue that the two equations do not give the same predictions, because the coefficient on the square term of f is positive in (9), while it is negative in (10). This difference is due to the functional form adopted for the convex information cost, and is a caveat of the model. Intuitively, the square term of f should load negatively on aggregate output due to the assumption of convex costs. 14 Hypothesis 1: At the aggregate level, mutual fund investment in the primary market (f ) is positively related to subsequent output growth. Furthermore, the interaction between fund flows and mutual fund investment negatively predicts future growth. Hypothesis 2: At the industry level, mutual fund investment in the primary market positively predicts subsequent industry output. The interaction between fund flows and mutual fund investment negatively predicts industry output. An important assumption of the model is that mutual funds have superior screening skills. If this is indeed the case, the equity issuing firms that receive more mutual fund investment should have higher productivity growth in subsequent periods. This is a testable assumption which will also be explored. 4 Data and methodology In this section I describe the data and variable constructions, along with the empirical specification that maps the model to the data. 4.1 Macroeconomic and equity issuance data I use real GDP and real industry value-added to measure aggregate and industry output, respectively. I obtain quarterly real GDP and annual real industry value-added from the U.S. Bureau of Economic Analysis (BEA), and calculate growth rates as the log-differences of levels. In addition to overall output effects, I also examine the effect of mutual fund investment (and the lack of it) on the cyclical fluctuations in output. Although the model does not differentiate between the long-run trend and the cyclical component, there is substantial motivation for this differentiation from an empirical perspective. Both equity issues and aggregate fund flows are highly cyclical in nature (Covas and Den Haan, 2011; Jank, 2012), and in addition, Peress (2014) shows that information aggregation in the stock market has an impact on the transitory component of growth. Therefore it is worthwhile to examine the long-run trend and the cyclical component separately. I follow Covas and Den Haan (2011) and use the Hodrick-Prescott filter (Hodrick and Prescott, 1997) with a smoothing coefficient 15 of 1600 for quarterly data and 6.25 for annual data to extract the cyclical component of output. Data on aggregate fund flows for domestic equity funds in the U.S. is obtained from the Investment Company Institute (ICI). ICI collects monthly aggregate data on fund sales, redemptions, and exchanges, as well as total net asset values (TNA).9 I follow Warther (1995) and Jank (2012), and calculate quarterly net fund flows as new sales minus redemptions plus exchanges-in minus exchanges-out.10 I aggregate monthly flows to the quarterly/annual level in order to match up with the frequency of macroeconomic variables. The sample period spans from 1984Q1 to 2011Q4. Equity issuance data is obtained from the Securities Data Corporation’s (SDC) Global New Issues database, which provides information on issuer profile, issuer SIC code, issue price, issue size, issue date, filing date, among other details, of all seasoned equity offerings and initial public offerings issued on the NYSE, AMEX, and NASDAQ. I exclude units, ADRs, closed-end funds issues, and exclude financial firms (SIC codes 6000-6999) and utilities (SIC codes 4900-4949) from the sample. The firms’ accounting information and stock market information are obtained from COMPUSTAT and CRSP, respectively. I use the firms’ NAICS (North American Industry Classification System) codes from COMPUSTAT to match with the industry value-added data. If the NAICS code is missing from COMPUSTAT, I use the NAICS-SIC matching table provided by the U.S. Census.11 The final sample yields a total of 15,038 firm-issue observations. Other variables used include the aggregate stock market returns Rm , which is the valueweighted market return of the CRSP universe. For industry-level tests, I construct industrylevel value-weighted returns. To measure wealth available for investment W at the aggregate level, I use the total assets of non-financial business sectors (BusAssets, obtained from the Federal Reserve) as a proxy, while at the industry level, I use the market value of equity for the industry. I also control for total credit issuance in the economy (Credit), using quarterly/annual total credit borrowings in non-financial business sectors, obtained from the Fed9 According to ICI, the database covers about 98% of assets in the mutual fund industry. The advantage of using ICI data for aggregate flows over other mutual fund databases such as the CRSP Survivorship-bias-free database is that ICI reports actual flows, whereas fund-level databases do not have such information, and thus fund flows would have to be estimated. 10 Exchanges-ins and exchanges-outs are transfers into (out of) equity funds from/to other fund groups. These differ from sales/redemptions in that there is no cashflow between the funds and the investor. Nevertheless, they represent funds’ reduced ability to invest in equity. 11 http://www.census.gov/eos/www/naics/concordances/2002_NAICS_to_1987_SIC.xls 16 eral Reserve. This is an important control variable, as the amount of credit issuance influence firms’ ability to invest, in the same way as equity issues. In addition, I use quarterly/annual consensus GDP growth forecasts as a control variable. The one period ahead forecasts are obtained from the Survey of Professional Forecasters, compiled by the Federal Reserve of Philadelphia. For firm-level tests, control variables include firm size (defined as market value of assets) and prior returns (defined as the cumulative stock returns from months t − 12 to t − 1 prior to the month of the issue). I deflate all aggregate variables that are in levels by the GDP deflator, and deflate all industry variables in levels by the industry value-added price index. 4.2 Mutual fund investment Due to data limitations, it is very difficult to obtain information on allocations to institutions in new equity issues (Chemmanur, He, and Hu, 2009). Therefore I follow the approach of Gibson, Safieddine, and Sonti (2004) and Demiralp, D’Mello, Schlingemann, and Subramaniam (2011), and use the change in mutual fund holdings as a proxy to gauge the extent of fund investment in these issues. I calculate a measure of mutual fund participation for each new issue (SEO or IPO). I obtain information on mutual fund holdings at the quarter-end prior to and the quarterend immediately after each issue, from the Thomson Reuters Mutual Fund Holdings (s12) database. I define mutual fund participation for issue i at time t to be the increase in the number of shares held by all funds over the issuing quarter, scaled by the number of shares in the issue. Average participation is then defined as the mean participation rate over all issues during the quarter/year, and represents the proportion of all new issues that is contributed by mutual funds. Since this measure is based on holdings and is noisy, I create a dummy variable P ARTt that equals to one for quarters where the average participation is higher than the sample median, and zero otherwise (for the industry tests, the dummy equals one when the average participation is higher than the cross-sectional median across all industries for each year). The variable f in the regression can be obtained by multiplying P ARTt with total equity issues Issuet (total issues multiplied by the proportion coming from mutual funds). Table 1 shows the aggregate level summary statistics. Average quarterly GDP growth rates are about 0.68%, while market returns are about 2.8% per quarter. Total proceeds from primary issues are similar to net fund flows as a ratio of total assets of non-financial 17 business sectors, however fund flows display larger variation than equity issues. The average participation rate in primary issues by mutual funds is about 24%, which is approximately in line with the average holdings of the stock market by mutual funds. This participation rate also exhibits a fair amount of variation, suggesting that mutual funds actively move in and out of primary issues, rather than passively investing in the issues. However, this average participation rate is very noisy since the measure is constructed from change in holdings (which contains trading activities in the secondary market), therefore I use the dummy variable P ART for high/low participation in the regressions.12 4.3 Model specification In order to specify the regression in terms of stationary variables, I use the total assets of non-financial business sectors as a proxy for W , and scale all aggregate variables in equation (11) by it.13 For industry tests, I scale the variables by the lagged industry market value of equity (for all firms in that industry in CRSP). For firm-level tests, I scale by the lagged market value of assets (the market value of equity plus the book value of total liabilities). The full specification for the growth rate regression and cyclical component regression is as follows (the only difference being the dependent variable): Ln(Cycl)t 1 2 =α+ β1 × ft−1 + β2 × ft−1 + β3 × F lowt−1 BusAssetst−2 BusAssetst−2 ! 1 + β4 × (F lowt−1 ft−1 ) + γ1 × Issuet−1 + γ2 × (F lowt−1 P ARTt−1 ) BusAssetst−2 ! Grtht or + γ3 × (F lowt−1 Issuet−1 ) + γ4 × P ARTt−1 + γ5 × Rm,t + γ6 × Rm,t−1 + δ × Controlst−1 + εt (12) where ft−1 ≡ Issuet−1 × P ARTt−1 , and the controls include credit issuance Creditt−1 BusAssetst−2 , consensus forecast of GDP growth F orecastt−1 , total assets of non-financial business sectors BusAssetst−2 (to control for the scale effect), and four lags of the dependent variable to account for autocorrelation in output 12 Using the actual participation rate yields similar estimates. For robustness I also use total market capitalization (of CRSP stocks) as an alternative proxy, and all the results are similar. 13 18 growth or the cyclical component of output. The first four variables along with Rm,t in equation (12) correspond to those in equation (11) implied by the model. The main coefficients of interest are β1 (positive) and β4 (negative). The second set of variables with coefficients γ1 through γ3 are included in the regression to control for the base effects arising from the interaction terms in the main variables. The empirical specification above corresponds to the model in specifying a lead-lag relation between equity issues / mutual fund investment and subsequent output growth. The only period t independent variable is the stock market return Rm , and this is because in the model, the output from new issues Y and existing projects Rm are realized at the same time. I scale the model by BusAssetst−2 , because this corresponds to the beginning-of-period assets when agents make their issue/investment decisions, which is consistent with the model. Empirically, this choice does not affect the results, and all the results remain consistent when the model is scaled by BusAssetst−1 . 5 Empirical results 5.1 Aggregate results To test the model at the aggregate level, I perform OLS regressions on equation (12) using various specifications. All regressions include four lags of the dependent variable, and I use Newey-West standard errors with four lags to calculate the t-statistics. Table 2 reports the predictive regression results for GDP growth.14 Panel 1 shows that fund flows are strongly correlated with subsequent economic growth. The predictive power of fund flows subsumes the predictive power of market returns in panel 2, which is consistent with Jank (2012). This result also suggests that although individual investors may not know the true productivity, they do have some noisy signals and are able to predict growth to some extent.15 Panel 3 shows that current period market returns is weakly and positively related to output. Panel 4 shows that equity issues by themselves do not have predictive power for growth. Indeed, models of finance and growth (Levine, 2005) suggest 14 I scale the regression by BusAssetst−2 as in equation (12), however for brevity I leave it out in tabulating the tables. 15 Note that although equation (12) shows that fund flows should negatively predict output, the empirical findings of a positive prediction does not invalidate the model. In the model, the household’s information θh is not correlated with the true productivity θ, and the negative prediction in the model is due to the budget constraint D = W − F . In reality, when households have some information about the true productivity, their fund flows would positively predict output. This is the story in Jank (2012). 19 that the accumulation of physical capital does not contribute strongly to growth, but rather it is the advances in productivity that drive growth. Panels 5 and 6 show the results for two specifications of the main model, with panel 6 being the full regression in equation (12). The main variable of interest is f here. Although not statistically significant in any of the regressions, the coefficients on f are all positive, and have opposite signs than Issue with similar magnitudes. This provides some suggestive evidence in support of the model, albeit weak. The empirical specification of f as Issue × P ART gives an additional interpretation. Since P ART is a dummy variable, the coefficient on Issue itself measures the impact of new equity issues on subsequent growth when mutual funds have low participation in these issues, and the coefficient on f is the effect of new issues on subsequent growth when mutual funds participate more in the issues. The opposite signs on these coefficients suggest that when mutual funds do not participate in the issues, the issues do not generate growth (or are even associated with lower subsequent growth16 ), but when mutual funds participate the issues are more productive. This is consistent with the model and the notion that mutual funds are informed. However, given the weak statistical significance (and lack of significance on the other variables), the results on aggregate growth is not conclusive. Next I examine whether mutual fund intermediation affects the cyclical component of growth. Erel, Julio, Kim, and Weisbach (2012) and Covas and Den Haan (2011) have shown that equity issues vary greatly over the business cycle. Further, Peress (2014) theoretically show that informed trading in the stock market can enhance capital allocation efficiency and thus promote transitory TFP growth, even in the absence of technological growth. In this sense, mutual fund intermediation may also affect the cyclical component of output. Table 3 shows the predictive regression results for the log cyclical component of real GDP. The variables are as defined in table 2. As opposed to the aggregate growth results, fund flows do not predict cyclical growth unconditionally (panels 1 and 2). On the other hand, equity issues predict lower subsequent cyclical growth unconditionally in panel 4, and this result is similar to the aggregate growth results. In the main regressions in panels 5 and 6, the key variable f is significantly positive. Consistent with the model, more equity issues invested by mutual funds predict higher output. 16 It may seem strange that increasing equity issues lowers subsequent growth. However this may be consistent with market-timing by firms to issue new securities when prices are high. If high prices are followed by slowing growth, one may observe high issues followed by lower growth. 20 The empirical specification of Issue×P ART can also be interpreted in a similar fashion. When mutual funds participate less in equity issues (P ART = 0), new issues are associated with lower subsequent growth than when mutual funds participate more in those issues, holding the level of issues constant. The result are consistent with the hypothesis that mutual funds are able to identify the productive issues and invest when their private signals are good. This gives them a positive role in the economy. Note that while mutual fund participation P ART negatively predicts growth unconditionally, the magnitude is small compared to the interaction term with Issue (-0.0003 versus 0.5377). Other significant predictors of cyclical growth include the consensus professional forecast and fund flows (conditional on the other variables). In addition, the interaction variable F low × P ART negatively predicts subsequent cyclical growth. This variable is included as a base for F low × f . Although the variable in the model F low × f is insignificant in the regression, the negative coefficient on F low × P ART suggests that when fund flows are low, higher mutual fund participation can generate higher subsequent output, over and above the high fund participation effect from P ART . Although the model does not predict an interaction effect between F low and P ART , the regression result suggests that mutual funds may be particularly effective when they are capital constrained. Overall the aggregate-level evidence is largely in support of the model, especially in explaining the cyclical component of output. The results suggest that mutual funds may indeed have superior ability to identify productive equity issues, and hence have a positive role in the real economy through efficient capital allocation. However, the aggregate results have weak statistical power due to the small sample. This problem can be partially alleviated by extending the tests to the industry-level, and exploiting both the time-series and cross-sectional variations. 5.2 Industry-level results In this section, I present the industry-level results. Specifically, I test whether the relation in equation (12) is able to explain cross-industry variations in value-added growth and the cyclical component of value-added. I run OLS panel regressions, using annual industry value-added growth and the log cyclical component of value-added as the dependent variables. For the independent variables, only the variables F low, Credit and F orecast are aggregate-level variables, and all other variables 21 are defined at the industry level. Issuei,t is the new equity issues in industry i in year t, and P ARTi,t is the mutual fund participation dummy for industry i in year t, and equals to one if industry i has higher mutual fund participation in new issues relative to the cross-sectional median in year t, zero otherwise. As in the aggregate tests, fi,t is defined as Issuei,t × P ARTi,t . The market return Ri,t is defined as the industry’s value-weighted return. To scale the regression, I replace BusAssetst−2 in equation (12) with the industry market value of equity IndM Vt−2 . Table 4 reports the pooled panel regression results. All regressions include four lags of the dependent variable, and standard errors are clustered at the industry level. I also include industry fixed-effects to capture unobservable sources of value-added specific to each industry. For exposition purposes, I only present the results for the full model here. Panel 1 in table 4 is the predictive regression results for value-added growth. The control variables credit issuance, consensus forecast, lagged industry market capitalization and lagged market return are significant predictors of value-added growth. The main variables of interest, f is positive and weakly significant (t-stat 1.74). These results are similar to the aggregate evidence in suggesting that mutual fund intermediation has some impact on the overall growth of industry value-added, albeit weak. Panel 2 presents the predictive regression results for the log cyclical component of valueadded. In line with the aggregate evidence, the main variables of interest have strong effects on the cyclical component of output. Unconditionally, new equity issues are associated with lower subsequent output (coefficient -1.1039), while the variable f is significantly positive, and the magnitude is larger than the coefficient on Issue, suggesting that the equity issues invested by mutual funds lead to higher cyclical output. When mutual funds do not participate in the new issues (P ART = 0), the issues are not productive, and when there is high mutual fund participation (P ART = 1), the issues are associated with much higher subsequent output, and the net effect is positive (1.7154 − 1.1039 − 0.0014 = 0.6101). The interaction term F low × P ART is significantly negative, which is also consistent with the aggregate evidence. In times of low fund flows, higher mutual fund participation leads to even higher cyclical output, which suggests that funds are particularly effective at identifying good projects in low fund flow times (which are often associated with bad economic conditions). In addition, the squared term of f is significantly negative, which is consistent with the convex information cost in the model. 22 In panels 3 and 4, I include time fixed effects in addition to industry fixed effects, in order to account for potential cross-sectional correlations among industry growth rates due to common macroeconomic shocks, which cannot be captured by within-industry clustered standard errors. The aggregate independent variables F lowt , Creditt and F orecastt are dropped due to collinearity with the time dummies. Overall, the results are still very similar to those reported in panels 1 and 2, both in coefficient estimates and significance levels, with Issuei,t negatively predicting cyclical output, and f positively predicting cyclical output. 5.2.1 Robustness: dynamic panel regressions Although the industry results presented above support the model’s main hypotheses, there is one econometric problem in these panel regressions. The unobservable industry fixed effects in the panel regressions are captured by the error terms, and due to the lagged dependent variables in the specification, this creates a positive correlation between the dependent variables and the error term, leading to biased coefficients. One way to address this problem is to apply the dynamic panel regression technique proposed by Arellano and Bond (1991) as a Generalized Method of Moments (GMM) estimator. This estimator takes first-difference of the regression equation to eliminate the fixed-effect, and then uses lagged values of the dependent variable in levels as instruments for the first-differences. Asymptotically, this approach produces consistent and efficient estimates even when the industry fixed effect is correlated with the lagged dependent variable. However, as pointed out by Beck, Levine, and Loayza (2000), there are several shortcomings with the first-difference estimator, the main one being that in finite samples, the difference estimator has a large bias and poor precision. To address this problem, Arellano and Bover (1995) propose a system GMM estimator that estimates the difference equation jointly with the regression equation in levels. In addition to having better precision in finite samples, the system estimator maintains the cross-industry variation by including the regression in levels, and also improves the strength of the instruments (Blundell and Bond, 1998). The consistency of the dynamic panel GMM estimators requires two critical assumptions, one being that the error term does not exhibit serial correlation, and second being that the instruments are valid. Arellano and Bond (1991) propose two tests for these assumptions. The serial correlation test examines the first- and second-order serial correlation in the error term, under the null hypothesis of no serial correlation (with a normal distribution). The Hansen’s 23 J-test of overidentifying restrictions tests the overall validity of the instruments, and has a χ2 distribution with J − K degrees of freedom, where J is the number of instruments and K is the number of regressors. Under the null hypothesis, the instruments are valid and exogenous, and failure to reject the null hypothesis lends support to the model. At this point, it is important to note that although the GMM estimators produce consistent and efficient estimates asymptotically, one should be careful with their performances in finite samples, due to the potential over-fitting problem when instrument numbers are large. In particular, when the lagged dependent variable is persistent, lagged levels of these variables may be weak instruments for the difference equations (Blundell and Bond, 1998). Thus, the advantage of using the GMM estimators to achieve asymptotic consistency needs to be balanced with the potential biases in finite samples and the potential poor performance of using internal instruments. Therefore, as Beck and Levine (2004) point out, it is important to apply different estimators (GMM and OLS), and to use these results as complements to draw inferences. With this in mind, I now proceed to the regression results. I only report results for the cyclical component of industry value-added, as the previous results suggest that the effect of mutual fund intermediation is primarily on the cyclical component of growth. Table 5 reports the GMM results for the main regression in equation (12), using lags six and further of the dependent variable Ln(Cycl)t /BusAssetst−2 as instruments for the four lags of dependent variable for the difference equation, and lags five and further as instruments in the levels equation. All other independent variables are treated as predetermined variables, and are instrumented with the original variable and all lags. Panels 1 and 2 apply the one-step estimator with robust standard errors, while panels 3 and 4 apply the two-step estimator. Panels 2 and 4 also include time fixed-effects to capture cross-sectional correlation across industries. Overall, results are very similar across the four regressions, and are similar to those reported in table 4. In times of low fund participation, new equity issues lead to lower cyclical growth, but become productive when they are actively invested in by mutual funds. In addition, f 2 is significantly negative in all specifications, suggesting that mutual fund intermediation may have diminishing returns. The interaction term F low × P ART is negative at the 10% level in all specifications. Also note that all the coefficient estimates are similar in magnitude and significance across the four regressions, suggesting that the estimator used does not affect the results. 24 The autocorrelation tests fail to reject the null hypothesis of no first and second order autocorrelation for the two-step estimators, and there is only weak evidence of first-order autocorrelation for the one-step estimators. The Hansen test fails to reject the null hypothesis of valid instruments in all four specifications. However, it is important to note that in the case of large instrument count in finite samples, the Hansen test could be weakened to generate implausibly good p-values (Roodman, 2009), and should be interpreted with caution. 5.3 Firm-level productivity In this section, I test the key assumption of the model, which is that mutual funds have superior information relative to uninformed investors and can invest only in the productive firms. A direct test of this assumption is to examine the relation between mutual fund investment and the firms’ future productivity growth. If mutual funds indeed have superior information, their investment in firms should convey information about the firms’ future performance and productivity. In order to examine productivity, I restrict the sample to SEO-issuing firms only, due to data requirements for variables prior to the issue, which are unavailable for IPO firms. I employ several measures of productivity. First, I follow King and Levine (1993a) and assume a Cobb-Douglas production function of the form Y = AK α Lβ , where Y is output or revenue (COMPUSTAT item REV), K is capital (property, plant and equipment, net of depreciation, COMPUSTAT item PPENT), and L is labor (number of employees, COMPUSTAT item EMP). A denotes total factor productivity (TFP). Taking logarithms and differencing over five years yields GY = α(GK) + β(GL) + GA, where GY is the growth rate of output, GK is the growth rate of capital, GL is the growth rate of labor, and GA is the growth rate of total factor productivity, which captures the growth rate of all other inputs than capital and labor that contributes to output growth. I denote this growth rate as KLT F Pi,t+5 for firm i over the five years post-SEO. I follow King and Levine (1993a) and use α = 0.3 and β = 0.7 in the estimation. The shortcoming of KLT F P is that capital and labor shares are exogenously specified rather than estimated. As Olley and Pakes (1996) demonstrate, such estimates obtained from log-linear Cobb-Douglas production functions may suffer from simultaneity and selection biases. Simultaneity arises because productivity is known to the profit-maximizing firms when they choose their input levels. Since productivity is observed by the firms but not the econo25 metrician, failing to account for increases in inputs due to increased productivity will result in biased OLS estimates. On the other hand, selection biases result from the correlation between productivity shocks and exit decisions. Firms with larger capital stocks are more likely to stay in the market despite a negative productivity shock than firms with smaller capital stocks, because more capital stocks can be expected to produce greater future profits. Therefore, the exit decision and capital stock for a given productivity shock needs to be controlled for in estimating α. Olley and Pakes (1996) propose a three-step semi-parametric estimator to address these problems. They use investment as a proxy for unobserved time-varying productivity shocks, and thus controlling for simultaneity problems. The selection biases are addressed by estimating survival probabilities. The resulting estimator gives consistent estimates of α and β, from which estimates of time-varying firm-level productivity can be obtained17 . Applying the estimator to the production function Y = AK α Lβ , I obtain estimates of α = 0.1934 and β = 0.6397. From these I calculate annual total factor productivity in the same way as King and Levine (1993a), and calculate productivity growth for five years post-SEO issue for each firm-issue. I denote this growth rate as OP T F Pi,t+5 . Complementary to the TFP measures above, I also employ two additional measures, revenue growth and five-year average return on assets (ROA). Revenue growth (RevGrthi,t+5 ) is calculated as the five-year growth-rate in revenue for firm i post-SEO issue. Average ROA is defined as operating income before depreciation (COMPUSTAT item OIBDP) divided by beginning-of-period book value of assets (COMPUSTAT item AT), and taking five-year averages post-SEO issue. The advantage of these two measures is that they do not rely on structural assumptions about production functions. However, this is also a drawback at the same time, because it is unclear how these measures map to TFP in a theoretical framework. Since the test here is on the effect of mutual fund investment on firm-level productivity growth, the full regression implied by the model does not apply here. Instead, I run OLS regressions of the various productivity measures on equity issue, fund participation and other control variables. If mutual funds indeed have superior information, their investment should lead to higher productivity growth. Therefore, I only include a subset of the variables in equation (12) that are consistent with this prediction. For control variables, I include firm size (LnSizet , defined as log of market value of total assets in the year before the SEO issue), 17 For detailed algorithms on the estimation method, see Yasar, Raciborski, and Poi (2008) and Keller and Yeaple (2009) Appendix A. 26 prior 12 month stock returns (P riorrett ), aggregate fund flows during the quarter of issue scaled by total assets of non-financial business sectors (F lowt ), and aggregate credit issuance during the quarter of the issue scaled by total assets of non-financial business sectors (Creditt ). I define firm-level equity issue EqIssuei,t as issue proceeds scaled by market value of assets prior to the issue. The mutual fund participation dummy P ARTi,t is equal to one if fund participation for issue i is above the median across all issues during the same year (to capture cross-sectional variation in participation). In all regressions, I include industry and year fixed effects, and cluster the standard errors at the industry level. Table 6 reports the regression results. Panels 1 and 2 report results on TFP measures, and panels 3 and 4 report results on operating performance measures. For both TFP measures, equity issue in low participation times have little impact on TFP growth, but when fund participation is high, more issues lead to significant higher subsequent TFP growth, given by the coefficient on EqIssue × HighP art. With the exception of F lowt , most of the coefficients are similar in magnitudes between the two TFP measures. For the operating performance measures, results are a little mixed. Equity issues in low participation times lead to higher subsequent revenue growth, and the effect is even stronger during high participation times. However for ROA this is not the case. High equity issues lead to no changes in subsequent ROA (the coefficient is negative but statistically insignificant), and in high participation times the effect is marginally negative. However, high participation unconditionally leads to higher subsequent ROA, which is in contrary to the results on TFP measures and revenue growth, and the magnitude on HighP art is smaller than the magnitude on EqIssue × HighP art, suggesting that the overall effect of high participation on ROA is still negative. Hence, ROA may be capturing information other than TFP, and may not be a good measure for firm-level productivity. Moreover, Brav, Jiang, and Kim (2011) find that hedge fund activism has a much larger effect on plant and firm-level productivity (TFP) than on operating margins. This is similar to the findings here, in that operating performance measures and TFP measures are not substitutable. One other result worth noting is that the coefficient on F low × HighP art is insignificant across all four measures of productivity. This is in contrary to both the aggregate and industry-level results, and suggests that aggregate fund flows have little impact on the firm-level participation-growth relation, albeit being significant at the macro level. Overall, these firm-level results suggest that mutual fund participation is positively related 27 to subsequent firm productivity growth, measured by TFP. This confirms the assumption that mutual funds can identify productive firms and invest in those firms only, and provides complementary evidence to the aggregate and industry-level results. 5.4 Alternative explanations and robustness Although the results above are in support of a positive role of mutual funds in the economy, one cannot fully reject alternative explanations of the findings. Specifically, one may argue that the association between mutual fund investment and subsequent growth is not driven by mutual funds’ active investment, but is instead driven by fund flows. When investors perceive future productivity to be high (and rightly so), they would invest more into mutual funds. Given their mandates, fund managers are then forced to invest some of the additional fund flows into the primary market, driving up their participation and hence f . This would naturally create a positive relation between mutual fund investment in new issues and subsequent growth, even when fund managers do not possess superior information. In fact, it is the individual investors that are doing the work, rather than mutual funds. This is the explanation offered by Jank (2012) in explaining the predictive power of fund flows on GDP growth. In order to alleviate this reverse causality concern, I make two observations. First, if this alternative story is true, there should be high correlations between fund flows F and mutual fund participation in equity issues P ART , at both the aggregate and industry level. However, the data shows that at the aggregate level, the contemporaneous correlation between fund flows and participation is only 0.16 (p-value = 0.08). At the industry level, this correlation is even lower, at only 0.002 (insignificant with a p-value of 0.94). This suggests that on average, mutual funds do not passively invest upon receiving fund flows, but instead make active decisions to move in and out from the primary market over time. Furthermore, the passive effect of fund flows is already partially taken into account by including fund flows as an independent variable in the regressions. Second, the reverse causality story implies that when fund flows are high, mutual funds would participate more in equity issues, and future output would increase (although not due to mutual fund investment). This implies a positive relation between the interaction term F × P ART and subsequent output, or at least no relation. Instead, the regression results at both the aggregate and industry level show a strong negative relation. This result cannot be explained by the reverse causality story, where individuals and mutual funds invest more at 28 the same time. To further examine this alternative explanation, I separate mutual fund investment into participation by active funds and passive (index) funds, and examine their effects on output. If the reverse causality explanation is true, there should be no distinction between active and passive funds, since all mutual funds are passive in their primary investment. However, in the information story, the effect should be predominantly coming from active funds, while investment by passive funds should have little real effects. I use the fund names in the s12 fund holdings database to identify passive funds. I follow Schwarz (2012), and classify funds with names containing “index”, “indx”, or “idx” as index or passive funds. All other funds are considered as active. I perform regressions at both the aggregate and industry level, and include participation by active (ActP ART ) and passive funds (P assP ART ) separately in the regression. In addition, I interact Issue with both the active and passive participation, to obtain investment by active funds (fact ) and passive funds (fpass ). The other variables f 2 and F low × f are constructed using only the active funds’ participation. Table 7 presents the results. Since the previous results suggest that mutual funds mainly affect the cyclical component of output, I only present the cyclical results here. Panel 1 shows the aggregate regression results. When the overall mutual fund investment is separated into active and passive, neither has a significant relation with subsequent output. Nevertheless, the coefficients on fact and fpass have opposite signs, and fact is positively related to subsequent output, which is in line with the previous results. On the other hand, fpass has a negative sign. However since none of the coefficients are significant, one cannot infer much from this regression. The lack of significance may be due to the weakened statistical power from separating participation into two groups, since the aggregate regression only has 107 observations. Panel 2 presents the industry-level results. New issues invested by active funds (fact ) positively predicts subsequent growth, while the coefficient on issues invested by passive funds (fpass ) is insignificant. This is consistent with the information explanation, where active funds have superior information, but is inconsistent with the passive investment explanation. In addition, F low × fact is significantly negative, and this is consistent with the model’s prediction. When fund flows are lower, more equity investment by active funds faciliates higher subsequent growth. In table 4, this coefficient is insignificant, and the results here 29 suggest that this may be due to the weakened power when including passive funds in the sample. Although the tests here have weak statistical power due to the split sample, they offer some evidence in support of the information story, and suggests that the main results are unlikely to be due to the reverse causality story. 6 Conclusion This paper investigates the role of mutual funds in the real economy, by examining the impact of mutual fund participation in primary equity issues on economic growth. Prior literature on mutual funds have largely focused on their secondary market activities, with little attention to their intermediation role in the primary market. However, consistent with intermediation theory, mutual funds should have informational advantages, and thus be able to exert screening and monitoring efforts more effectively to reduce information problems and facilitate efficient allocation of capital. Although there is evidence that institutional shareholders exert monitoring and screening effort to add value, the role of these participants at the macroeconomic level is still unclear. This is one of the first papers to examine this issue, and results suggest that intermediaries play an important role in equity markets in reducing information problems and facilitating quality investment. Consistent with a simple model of mutual fund intermediation, I find that high mutual fund investment in equity issues lead to significantly higher subsequent economic growth, at both the aggregate and industry level. Moreover, the participation-growth effect is stronger when fund flows are low and funds are constrained. These effects predominantly appear the cyclical component of output, and there is little effect on long-run trend growth. At the firm-level, high mutual fund investment in SEOs also leads to significantly higher subsequent productivity growth (measured by TFP). These results are largely consistent with the notion that intermediaries’ advantage in screening may facilitate efficient capital allocation and enhance output. Future research in this area still needs to address potential alternative explanations. Although some suggested evidence has been offered here, by examining separately the active versus passive funds, the results are somewhat weakened by a small sample size. In addition, the model takes equity issues as exogenous, and does not consider the strategic decision of 30 firms in issuing new securities. These features could be incorporated in future research. The findings in this paper have additional implications for future research in the economic growth literature. Levine (2005) suggests that financial intermediaries and financial development can foster growth, and a large body of empirical literature has found support in cross-country evidence. However, the literature has largely ignored the role of intermediation outside of private credit and the banking sector, and uses measures such as the size and liquidity of the stock market as proxies for stock market development. Results in this paper suggests that these may not be accurate proxies, because they ignore the difference in levels of stock market intermediation across markets. 31 References Arellano, M., and S. Bond, 1991, Some tests of specification for panel data: Monte carlo evidence and an application to employment equations, The Review of Economic Studies 58, 277–297. Arellano, Manuel, and Olympia Bover, 1995, Another look at the instrumental variable estimation of error-components models, Journal of Econometrics 68, 29 – 51. Beck, Thorsten, and Ross Levine, 2002, Industry growth and capital allocation: does having a marketor bank-based system matter?, Journal of Financial Economics 64, 147 – 180. , 2004, Stock markets, banks, and growth: Panel evidence, Journal of Banking & Finance 28, 423 – 442. , and Norman Loayza, 2000, Finance and the sources of growth, Journal of Financial Economics 58, 261 – 300. Berk, Jonathan B., and Richard C. Green, 2004, Mutual fund flows and performance in rational markets, Journal of Political Economy 112, pp. 1269–1295. Bernanke, Ben, and Mark Gertler, 1989, Agency costs, net worth, and business fluctuations, American Economic Review 79, 14–31. Bhattacharya, Sudipto, and Anjan V. Thakor, 1993, Contemporary banking theory, Journal of Financial Intermediation 3, 2 – 50. Blundell, Richard, and Stephen Bond, 1998, Initial conditions and moment restrictions in dynamic panel data models, Journal of Econometrics 87, 115 – 143. Boehmer, Ekkehart, and Eric K. Kelley, 2009, Institutional investors and the informational efficiency of prices, Review of Financial Studies 22, 3563–3594. Brav, Alon, Wei Jiang, and Hyunseob Kim, 2011, The real effects of hedge fund activism: Productivity, risk, and product market competition, Working Paper 17517 National Bureau of Economic Research. Brav, Alon, Wei Jiang, Frank Partnoy, and Randall Thomas, 2008, Hedge fund activism, corporate governance, and firm performance, The Journal of Finance 63, 1729–1775. Chava, Sudheer, and Amiyatosh Purnanandam, 2011, The effect of banking crisis on bank-dependent borrowers, Journal of Financial Economics 99, 116 – 135. Chemmanur, Thomas J., Shan He, and Gang Hu, 2009, The role of institutional investors in seasoned equity offerings, Journal of Financial Economics 94, 384 – 411. Chen, Nan-Kuang, 2001, Bank net worth, asset prices and economic activity, Journal of Monetary Economics 48, 415 – 436. Covas, Francisco, and Wouter J. Den Haan, 2011, The cyclical behavior of debt and equity finance, The American Economic Review 101, 877–899. Del-Guercio, Diane, and Jennifer Hawkins, 1999, The motivation and impact of pension fund activism, Journal of Financial Economics 52, 293 – 340. Demiralp, Ilhan, Ranjan D’Mello, Frederik P. Schlingemann, and Venkat Subramaniam, 2011, Are there monitoring benefits to institutional ownership? evidence from seasoned equity offerings, Journal of Corporate Finance 17, 1340 – 1359. Diamond, Douglas W., 1984, Financial intermediation and delegated monitoring, The Review of Economic Studies 51, 393–414. Duchin, Ran, Oguzhan Ozbas, and Berk A. Sensoy, 2010, Costly external finance, corporate investment, and the subprime mortgage credit crisis, Journal of Financial Economics 97, 418 – 435. 32 Elton, Edwin J., and Martin J. Gruber, 2013, Chapter 15 - mutual funds, in Milton Harris George M. Constantinides, and Rene M. Stulz, ed.: Handbook of the Economics of Finance SET vol. 2, Part B of Handbook of the Economics of Finance . pp. 1011 – 1061 (Elsevier). Erel, Isil, Brandon Julio, Woojin Kim, and Michael S. Weisbach, 2012, Macroeconomic conditions and capital raising, Review of Financial Studies 25, 341–376. Fama, Eugene F, 1970, Efficient capital markets: A review of theory and empirical work, The Journal of Finance 25, 383–417. Faulkender, Michael, and Mitchell A. Petersen, 2006, Does the source of capital affect capital structure?, Review of Financial Studies 19, 45–79. Gibson, Scott, Assem Safieddine, and Ramana Sonti, 2004, Smart investments by smart money: Evidence from seasoned equity offerings, Journal of Financial Economics 72, 581 – 604. Gillan, Stuart L., and Laura T. Starks, 2000, Corporate governance proposals and shareholder activism: the role of institutional investors, Journal of Financial Economics 57, 275 – 305. Greenwood, Jeremy, and Boyan Jovanovic, 1990, Financial development, growth, and the distribution of income, Journal of Political Economy 98, pp. 1076–1107. Greenwood, Robin, and David Scharfstein, 2013, The growth of finance, Journal of Economic Perspectives 27, 3–28. Grossman, Sanford J, and Joseph E Stiglitz, 1980, On the impossibility of informationally efficient markets, The American economic review 70, 393–408. Hodrick, Robert James, and Edward C. Prescott, 1997, Postwar u.s. business cycles: An empirical investigation, Journal of Money, Credit and Banking 29, 1–16. Holmstrom, Bengt, and Jean Tirole, 1997, Financial intermediation, loanable funds, and the real sector, Quarterly Journal of Economics 112, 663 – 691. Ivashina, Victoria, and David Scharfstein, 2010, Bank lending during the financial crisis of 2008, Journal of Financial Economics 97, 319 – 338. Jank, Stephan, 2012, Mutual fund flows, expected returns, and the real economy, Journal of Banking and Finance Forthcoming. Jermann, U., and V. Quadrini, 2012, Macroeconomic effects of financial shocks, American Economic Review 102, 238 – 271. Keller, W., and S.R. Yeaple, 2009, Multinational enterprises, international trade, and productivity growth: firm-level evidence from the united states, The Review of Economics and Statistics 91, 821–831. Khan, M., L. Kogan, and G. Serafeim, 2012, Mutual fund trading pressure: Firm-level stock price impact and timing of seos, The Journal of Finance 67, 1371–1395. King, Robert G., and Ross Levine, 1993a, Finance and growth: Schumpeter might be right, The Quarterly Journal of Economics 108, pp. 717–737. King, Robert G, and Ross Levine, 1993b, Finance, entrepreneurship and growth, Journal of Monetary Economics 32, 513 – 542. Leland, Hayne E, and David H Pyle, 1977, Informational asymmetries, financial structure, and financial intermediation, Journal of Finance 32, 371–87. Lemmon, Michael, and Michael R. Roberts, 2010, The response of corporate financing and investment to changes in the supply of credit, Journal of Financial and Quantitative Analysis 45, 555–587. 33 Levine, Ross, 2005, Chapter 12 finance and growth: Theory and evidence, vol. 1, Part A of Handbook of Economic Growth . pp. 865 – 934 (Elsevier). Lewellen, Jonathan, 2011, Institutional investors and the limits of arbitrage, Journal of Financial Economics 102, 62–80. Olley, G. Steven, and Ariel Pakes, 1996, The dynamics of productivity in the telecommunications equipment industry, Econometrica 64, pp. 1263–1297. Peress, Joel, 2014, Learning from stock prices and economic growth, Review of Financial Studies p. Forthcoming. Piacentino, Giorgia, 2013, Do institutional investors improve capital allocation?, Working paper, London School of Economics. Roodman, D., 2009, How to do xtabond2: An introduction to difference and system gmm in stata, Stata Journal 9, 86–136(51). Schwarz, Christopher G, 2012, Mutual fund tournaments: The sorting bias and new evidence, Review of Financial Studies 25, 913–936. Shleifer, Andrei, and Robert W. Vishny, 1986, Large shareholders and corporate control, Journal of Political Economy 94, 461–88. Smith, Michael P, 1996, Shareholder activism by institutional investors: Evidence for calpers, Journal of Finance 51, 227–52. Sufi, Amir, 2009, The real effects of debt certification: Evidence from the introduction of bank loan ratings, Review of Financial Studies 22, 1659–1691. Warther, Vincent A., 1995, Aggregate mutual fund flows and security returns, Journal of Financial Economics 39, 209 – 235. Yasar, Mahmut, Rafal Raciborski, and Brian Poi, 2008, Production function estimation in stata using the olley and pakes method, Stata Journal 8, 221–231. 34 Table 1: Summary Statistics Reported are summary statistics of variables. Grtht is the quarterly log-differences of real GDP levels (in 2009 dollars). Ln(Cycl)t /BusAssetst−2 is the log of the cyclical component of real GDP from the HP-filter, scaled by two-quarter lagged total assets of the non-financial business sectors. F lowt /BusAssetst−1 is the net cash flow into domestic equity funds during the quarter, scaled by beginning-of-quarter total assets of the non-financial business sectors. Creditt /BusAssetst−1 is the total borrowing in non-financial business sectors during the quarter, scaled by beginning total assets of the non-financial business sectors. Rm,t is the quarterly value-weighted return of the CRSP universe. Issuet /BusAssetst−1 is the sum of proceeds from all SEOs and IPOs during the quarter, scaled by beginning total assets of the non-financial business sectors. AvgP artt is the average participation rate by all mutual funds in the SEOs and IPOs, measured by the change of holdings over the quarter scaled by issue proceeds. F orecastt is the one-quarter ahead real GDP growth rate forecast by the Survey of Professional Forecasters. Variable N Mean Std. Dev. Grtht 111 0.006841 0.006198 Ln(Cycl)t /BusAssetst−2 110 0.000045 0.000589 Flowt / BusAssetst−1 111 0.000938 0.001669 Creditt / BusAssetst−1 111 0.072867 0.024595 Rm,t 112 0.027925 0.088196 Issuet / BusAssetst−1 111 0.000821 0.000546 AvgPartt 112 0.237586 0.230758 Forecastt 112 0.006496 0.002589 Table 2: GDP growth rates regressions Reported are predictive OLS regression results for GDP growth rates. The dependent variable is log-differences of real GDP. Rm,t is the value-weighted market return of the CRSP universe of stocks. Creditt is the total borrowing in non-financial business sectors during the quarter, scaled by beginning total assets of the non-financial business sectors. F lowt is the net cash flow into domestic equity funds, scaled by beginning total assets of the non-financial business sectors. Issuet is the sum of proceeds from all SEOs and IPOs during the quarter, scaled by beginning total assets of the non-financial business sectors. P ARTt is a dummy variable that equals one if the quarter has higher average mutual fund participation in new issues than the median over the sample period, zero otherwise. ft is defined as Issuet × P ARTt . F orecastt is the one-quarter ahead real GDP growth rate forecast by the Survey of Professional Forecasters. BusAssetst is the total assets of the non-financial business sectors. All regression specifications include four lags of the dependent variable (unreported for brevity). t-statistics are reported in parentheses and are calculated using Newey-West standard errors with four lags. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. Dependent Variable: Grtht Flowt Creditt−1 Forecastt−1 BusAssetst−2 (1) (2) (3) (4) (5) (6) 0.9811*** 0.7850** 0.9227*** 0.9471** 1.1157** 2.8991 (3.09) (2.38) (2.67) (2.05) (2.27) (1.11) 0.0214 0.0155 0.0119 0.0119 0.0108 0.0090 (0.90) (0.70) (0.52) (0.52) (0.47) (0.39) 0.1716** 0.1779** 0.2057** 0.2057** 0.2044** 0.2121** (2.19) (2.36) (2.53) (2.52) (2.48) (2.59) -0.0000** -0.0000** -0.0000* -0.0000* -0.0000 -0.0000 (-2.07) (-2.10) (-1.95) (-1.93) (-1.20) (-1.07) 0.0093 0.0094* 0.0094* 0.0104 0.0142** (1.49) (1.73) (1.67) (1.60) (2.45) 0.0172* 0.0172* 0.0165* 0.0161 (1.77) (1.73) (1.72) (1.59) -0.0893 -4.3261 -8.4938 (-0.06) (-1.17) (-1.49) -0.0035* -0.0042* (-1.71) (-1.73) 2.5840 6.6818 (1.31) (1.41) Rm,t−1 Rm,t Issuet−1 PARTt−1 ft−1 f2t−1 -0.0044 (-1.03) Flowt−1 × ft−1 0.0749 (0.71) Flowt−1 × PARTt−1 -2.1787 (-0.95) Flowt−1 ×Issuet−1 -0.0000 (-0.26) Intercept N Adj. R 2 0.0039 0.0039 0.0024 0.0024 0.0066* 0.0077 (1.56) (1.55) (0.89) (0.84) (1.67) (1.63) 107 107 107 107 107 107 32.29% 33.16% 38.59% 37.94% 38.44% 38.23% Table 3: Cyclical component of GDP regressions Reported are predictive OLS regression results for the cyclical component of GDP. The dependent variable is the log of the cyclical component of real GDP, extracted using the HP-filter, scaled by two-quarter lagged total assets of the non-financial business sectors. Rm,t is the value-weighted market return of the CRSP universe of stocks. Creditt is the total borrowing in non-financial business sectors during the quarter, scaled by beginning total assets of the non-financial business sectors. F lowt is the net cash flow into domestic equity funds, scaled by beginning total assets of the non-financial business sectors. Issuet is the sum of proceeds from all SEOs and IPOs during the quarter, scaled by beginning total assets of the non-financial business sectors. P ARTt is a dummy variable that equals one if the quarter has higher average mutual fund participation in new issues than the median over the sample period, zero otherwise. ft is defined as Issuet × P ARTt . F orecastt is the one-quarter ahead real GDP growth rate forecast by the Survey of Professional Forecasters. BusAssetst is the total assets of the non-financial business sectors. All regression specifications include four lags of the dependent variable (unreported for brevity). t-statistics are reported in parentheses and are calculated using Newey-West standard errors with four lags. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. Dependent Variable: Ln(Cycl)t /BusAssetst−2 Flowt−1 Creditt−1 Forecastt−1 BusAssetst−2 (1) (2) (3) (4) (5) (6) 0.0202 0.0121 0.0129 0.0231 0.0379* 0.2999** (1.17) (0.66) (0.68) (1.08) (1.76) (2.53) 0.0018 0.0016 0.0015 0.0015 0.0017 0.0019* (1.63) (1.51) (1.33) (1.31) (1.59) (1.85) 0.0056* 0.0059* 0.0061* 0.0061* 0.0061* 0.0068** (1.68) (1.89) (1.85) (1.89) (1.87) (2.27) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (0.37) (0.41) (0.49) (0.62) (1.59) (1.47) 0.0004 0.0004 0.0004 0.0005 0.0005* (1.34) (1.36) (1.37) (1.35) (1.89) 0.0001 0.0001 0.0001 0.0000 (0.34) (0.30) (0.14) (0.09) -0.0379 -0.4296** -0.8798*** (-0.47) (-2.10) (-2.64) -0.0003*** -0.0004*** (-2.74) (-2.96) 0.2267* 0.5377** (1.99) (2.47) Rm,t−1 Rm,t Issuet−1 PARTt−1 ft−1 f2t−1 -0.0002 (-1.26) Flowt−1 × ft−1 0.0058 (1.30) Flowt−1 × PARTt−1 -0.2025** (-2.39) Flowt−1 ×Issuet−1 -0.0000 (-0.83) Intercept N Adj. R 2 -0.0003* -0.0003* -0.0003** -0.0003* -0.0000 0.0001 (-1.97) (-1.98) (-1.99) (-1.81) (-0.11) (0.51) 106 106 106 106 106 106 82.27% 82.41% 82.27% 82.13% 82.88% 83.71% Table 4: Industry Panel Regressions Reported are predictive OLS regressions for annual industry value-added growth and the cyclical component. Grthi,t+1 is the log-differences in real industry value-added for industry i in year t. Ln(Cycli,t )/IndM Vt−2 is the log cyclical component of industry value-added, scaled by two-period lagged industry market value of equity in CRSP. Ri,t is the value-weighted industry return of all CRSP stocks within industry i in year t. Creditt , F lowt and F orecastt are as defined in Table 2 and 3. Issuei,t is the sum of proceeds from all SEOs and IPOs during the quarter in industry i, scaled by beginning industry market value of equity. P ARTi,t is a dummy variable that equals one if the industry has higher average mutual fund participation in new issues than the median industry in year t, zero otherwise. fi,t is defined as Issuei,t × P ARTi,t . IndM Vi,t is the industry market value of equity in CRSP. All regressions include four lags of the dependent variable (unreported for brevity). t-statistics are reported in parentheses, and standard errors are clustered at the industry level. The first two panels include industry fixed effects, and the last two panels include both industry and year fixed effects. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. (1) (2) (3) (4) Dependent variable Grthi,t Ln(Cycli,t )/IndM Vt−2 Grthi,t Ln(Cycli,t )/IndM Vt−2 Flowt−1 1.1770 0.0026 (1.13) (0.07) 0.1641** 0.0047* (2.38) (1.95) 1.4222** -0.0004 (2.53) (-0.01) -0.0000* 0.0000* 0.0000 0.0000 (-1.74) (1.83) (0.41) (1.11) 0.0322** 0.0056* 0.0264* 0.0060* (2.20) (1.93) (1.68) (1.81) 0.0165 0.0008 0.0106 0.0006 (1.23) (1.60) (0.64) (0.85) 1.2864 -1.1039*** 1.3089 -1.1028*** (1.17) (-3.98) (1.27) (-3.95) -0.0120 -0.0014** 0.0005 -0.0016** (-1.07) (-2.15) (0.04) (-2.42) 1.3525* 1.7154*** 0.9890 1.7351*** (1.74) (3.14) (1.25) (3.16) 0.4318 -2.7251** 1.6898 -2.8640** (0.03) (-2.61) (0.13) (-2.55) 0.1107 0.0160 0.1198 0.0149 (0.38) (0.84) (0.40) (0.80) 0.5053 -0.1588** -0.3580 -0.1457* (0.40) (-2.14) (-0.27) (-1.71) -0.0000* 0.0000 -0.0000 0.0000 (-1.83) (1.44) (-1.63) (1.38) -0.0405 -0.0018** 0.0903*** 0.0003 (-1.25) (-2.02) (4.82) (0.22) 1101 1043 1101 1043 9.56% 82.49% 11.83% 82.77% Industry FE Y Y Y Y Time FE N N Y Y Creditt−1 Forecastt−1 IndMVi,t−2 Ri,t−1 Ri,t Issuei,t−1 PARTi,t−1 fi,t−1 f2i,t−1 Flowt−1 × fi,t−1 Flowt−1 × PARTi,t−1 Flowt−1 × Issuei,t−1 Intercept N Adj. R 2 Table 5: Industry Dynamic Panel Regressions Reported are dynamic panel GMM regressions for cyclical component of annual industry value-added, using the Arellano and Bover (1995) system GMM estimator. Ln(Cycli,t )/IndM Vt−2 is the log cyclical component of industry value-added, scaled by two-period lagged industry market value of equity in CRSP. Independent variables are defined as in Table 4. All regressions include four lags of the dependent variable (unreported for brevity), instrumented with lags 6 and further. t-statistics are reported in parentheses, and standard errors are robust for both the one-step and two-step estimators. Panels 1 and 3 include industry fixed effects, and Panels 2 and 4 include both industry and year fixed effects. The Wald test is for joint significance of all independent variables. The autocorrelation test tests for first and second order autocorrelation in the residuals according to Arellano and Bond (1991). The Hansen’s J-test tests for validity of instruments. *, **, and *** denote statistical significance at the 10%, 5%, and 1% levels, respectively. Depdendent Variable: Ln(Cycli,t )/IndM Vt−2 One-step (1) Flowt−1 Creditt−1 Forecastt−1 IndMVi,t−2 Ri,t−1 Ri,t Issuei,t−1 PARTi,t−1 fi,t−1 f2i,t−1 Flowt−1 × fi,t−1 Flowt−1 ×PARTi,t−1 Flowt−1 × Issuei,t−1 Intercept N Time FE Wald test (p-value) Two-step (2) (3) 0.0597 0.0614 (1.37) (1.35) 0.0073** 0.0064 (2.03) (1.51) 0.0015 0.0064 (0.04) (0.15) (4) 0.0000** 0.0000*** 0.0000* 0.0000** (2.53) (2.65) (1.96) (2.23) 0.0046* 0.0049 0.0044 0.0035 (1.74) (1.60) (1.62) (1.43) 0.0009* 0.0007 0.0010* 0.0007 (1.70) (1.02) (1.79) (0.78) -0.3784** -0.3428** -0.3645* -0.3300** (-2.11) (-2.00) (-1.95) (-2.22) -0.0018*** -0.0020*** -0.0017*** -0.0018*** (-2.83) (-2.87) (-2.81) (-3.15) 1.8439*** 1.8452*** 1.8568*** 1.9603*** (3.41) (3.46) (3.43) (3.53) -3.3307*** -3.4829*** -3.2353*** -3.5798** (-3.34) (-3.25) (-3.03) (-2.56) 0.0300 0.0266 0.0272 0.0392* (1.59) (1.44) (1.24) (1.96) -0.1487* -0.1439* -0.1471* -0.1493** (-1.94) (-1.71) (-1.81) (-1.96) -0.0000 -0.0000 -0.0000 -0.0000 (-0.82) (-1.06) (-0.89) (-1.15) -0.0032*** -0.0002 -0.0031*** -0.0003 (-3.03) (-0.13) (-2.83) (-0.33) 1043 1043 1043 1043 N Y N Y 0.000 0.000 0.000 0.000 Autocorrelation test 1st order 0.075 0.066 0.175 0.183 (p-value) 2nd order 0.768 0.744 0.790 0.827 Chi-sq 42.02 23.99 42.02 30.25 p-value 1.000 1.000 1.000 1.000 Hansen test Table 6: Firm-level Productivity Regressions: SEO firms Reported are OLS regression results for firm-level productivity measures on equity issues and mutual fund participation, for SEO issuing firms. KLTFPi,t+5 is the five-year growth rate in productivity, with the productivity measured using the method in King and Levine (1993a). OPTFPi,t+5 is the five-year growth rate in productivity post-SEO, with the productivity measured using the method in Olley and Pakes (1996). RevGrthi,t+5 is the five-year revenue growth rate post-SEO. AvgROAi,t+5 is the average return on assets over the five years post SEO issuance. LnSizei,t is the natural log of market value of assets ending the year before the issue. Priorreti,t is the cumulative stock returns from month t-12 to t-1. EqIssuei,t is issue proceeds scaled by market value of assets prior to the issue. PARTi,t is a dummy equal to 1 if fund participation is above the median across all stocks during the issuing year. Other variables are defined as in Table 4. All regressions include industry and year fixed effects. t-statistics are reported in parenthesis, and standard errors are clustered by industry. *, **, and *** denote statistical significance at the 10%, 5%, and 1%, respectively. Dependent Variable LnSizei,t Priorreti,t EqIssuei,t PARTi,t EqIssuei,t × PARTi,t Flowt CreditIssuet Flowt × PARTi,t Intercept Adj. R2 (1) (2) (3) (4) KLTFPi,t+5 OPTFPi,t+5 RevGrthi,t+5 AvgROAi,t+5 -0.0002 -0.0009 -0.0024 0.0004 (-0.40) (-1.13) (-1.58) (1.26) -0.0157** -0.0042 0.0174 0.0013 (-2.05) (-0.40) (1.42) (0.87) 0.0364 0.1380 0.3448*** -0.0641 (0.48) (1.22) (2.63) (-1.59) -0.1066*** -0.1001*** -0.0411 0.0271** (-3.04) (-2.82) (-1.05) (2.14) 0.2543** 0.2537** 0.3434*** -0.0532* (2.29) (2.37) (2.81) (-1.69) 0.9737 -21.5279* -30.3687* -10.4809*** (0.08) (-1.78) (-1.79) (-2.94) 1.2556 -0.8136 -3.2203** 0.1015 (1.40) (-0.66) (-2.42) (0.26) 14.8408 10.0109 5.3618 3.4190 (1.27) (0.87) (0.31) (1.08) 0.5328*** 0.8996*** 1.4001*** 0.1451*** (4.52) (6.29) (8.87) (3.93) 0.057 0.080 0.071 0.294 Table 7: Robustness: Active versus passive funds Reported are OLS regression results for active versus passive funds. Panel (1) shows the aggregate results, the dependent variable being the log of the cyclical component of real GDP, extracted using the HP-filter, scaled by two-quarter lagged total assets of the non-financial business sectors. ActP ARTt (P assP ARTt ) is a dummy variable that equals one if the average participation in new equity issues by all active (passive) funds are above the sample median, zero otherwise. fact,t is defined as Issuet × ActP ARTt . fpass,t is defined as Issuet × P assP ARTt . The other variables are defined as in Table 3. Panel (2) shows the industry level results, with the dependent variable being the log cyclical component of industry value-added, scaled by two-period lagged industry market value of equity in CRSP. ActP ARTi,t (P assP ARTi,t ) is a dummy variable that equals one if the average participation in new equity issues in industry i by all active (passive) funds are above the cross-sectional sample median, zero otherwise. fact,i,t is defined as Issuei,t × ActP ARTi,t , and fpass,i,t is defined as Issuei,t × P assP ARTi,t . The other variables are defined as in Table 4. All regressions include four lags of the dependent variable. t-statistics are reported in parenthesis, and standard errors are clustered by industry in Panel (2). *, **, and *** denote statistical significance at the 10%, 5%, and 1%, respectively. Panel (1): Aggregate Panel (2): Industry Ln(Cycl)t /BusAssetst−2 Ln(Cycli,t )/IndM Vt−2 F lowt−1 0.0694 F lowt−1 (1.24) Creditt−1 0.0015 (-1.38) Creditt−1 (1.45) F orecastt−1 0.0073** -0.0000 F orecastt−1 0.0004 IndM Vi,t−2 -0.0001 Ri,t−1 0.0054* (1.88) Ri,t 0.0010* (-0.23) Issuet−1 -0.1159 (1.97) Issuei,t−1 (-0.38) ActP ARTt−1 -0.0000 0.0001 ActP ARTi,t−1 0.0504 P assP ARTi,t−1 -0.0736 fact,i,t−1 -0.0002 fpass,i,t−1 0.0046 -0.1351 F lowt−1 × fact,i,t−1 -0.0000 F lowt−1 × ActP ARTi,t−1 -0.0002 F lowt−1 × Issuei,t−1 Adj. R 82.64% 0.0000** (2.33) Intercept (-1.11) 2 0.0642 (0.54) (-0.20) Intercept -0.0114* (-2.00) (-1.45) F lowt−1 × Issuet−1 -1.5779 (-1.35) (0.71) F lowt−1 × ActP ARTt−1 0.2299 (0.42) 2 fact,i,t−1 (-0.32) F lowt−1 × fact,t−1 2.6880*** (2.76) (-0.74) 2 fact,t−1 -0.0006* (-1.80) (0.17) fpass,t−1 -0.0026** (-2.52) (0.62) fact,t−1 -1.9571*** (-3.66) (-0.01) P assP ARTt−1 0.0000* (1.75) (1.35) Rm,t 0.0135 (0.42) (-0.14) Rm,t−1 0.0001 (0.15) (2.40) BusAssetst−2 -0.1794 0.0003 (0.48) Adj. R 2 83.06%