Strategic Performance Allocation in Institutional Asset Management Firms: Behold the Power of Stars and Dominant Clients *

Transcription

1 Preliminary Please do not cite without permission Strategic Performance Allocation in Institutional Asset Management Firms: Behold the Power of Stars and Dominant Clients * RANADEB CHAUDHURI Oakland University ZORAN IVKOVIĆ Michigan State University CHARLES TRZCINKA Indiana University ABSTRACT We identify strong and very robust evidence of strategic performance allocation in the institutional money management industry, directed toward strong recent performers, the money management firms highvalue products. The extent of cross-subsidization varies with demand the size of the portfolio held by the largest client invested in the product. We identify three channels through which strategic performance allocation likely operates, and show that its extent is related positively to all three: performance of the styles pursued by bigger products in the same firm, availability of IPO opportunities, and illiquidity of the products investment styles. * We thank Stephen Dimmock, Joshua Pollet, Clemens Sialm, and Andrei Simonov for comments and suggestions.

2 Institutional asset management firms are a major player in the arena of financial intermediation. The sheer volume of assets under management, comparable to that of the mutual fund industry, is in the range of several trillion dollars. 1 The gargantuan size of the institutional asset management industry notwithstanding, relatively little is known about a host of issues surrounding it. A key reason is relative obscurity. Compared with the mutual fund industry, institutional asset management firms are required to disclose very little. Put differently, a detailed database containing a host of information about performance, fees, and fund holdings, akin to, for example, the CRSP Open-End Mutual Fund Database, is not available. Moreover, additional data-collection efforts to ascertain, for example, inflows and outflows of investment may be extracted from mutual funds N-SAR SEC filings (available also through EDGAR) and can shed further light on the mutual fund industry (e.g., Bergstresser and Poterba, 2002). Comparable opportunities do not abound in the study of institutional management. 2 There is a certain superficial similarity between the way investment options are structured in the institutional asset management and mutual fund arenas. In both industries, the prevalent organizational form is that of a firm (alternatively called complex or family) offering multiple funds or products that cover various investment objectives. In the domain of domestic equity, the focus of this study, this translates into both industries offering funds or products covering a range of investment objectives, spanned primarily along the dimensions of the underlying stock capitalization (small, mid-cap, and large) and growth-value gradation (growth, blend, value). Some of the unique features of the institutional asset management firms make its study particularly germane for a broader and deeper understanding of agency issues. 1 According to Standard & Poor s (2007), at the end of 2006, more than 51,000 plan sponsors allocated more than seven trillion dollars in assets to about 1,200 institutional money managers. 2 A notable exception is Goyal and Wahal (2008), who compile a unique database of 8,755 hiring decisions by 3,417 plan sponsors that delegate $627 billion in mandates between 1994 and 2003, as well as 869 firing decisions by 482 plan sponsors that withdraw almost $105 billion in mandates between 1996 and Their analyses proceed to address investors (plan sponsors ) decisions as a function of their characteristics and the characteristics of investment products. In that sense, their work is most similar to the study of individuals mutual fund trades through a discount broker from Ivković and Weisbenner (2009). 2

3 Indeed, although the structure of investment options may resemble that offered by the mutual fund industry, the two industries are quite different. Aside from a vast difference in the extent of transparency, as well as differences in institutional framework, 3 a pivotal difference is the structure of their respective investors. Mutual funds, investment vehicles inaugurated historically in pursuit of the goal of providing small investors access to diversified investment, have a broad investor base. A mutual fund is held by thousands, sometimes even tens of thousands of investors or more, each of whom typically holds a miniscule fraction of fund shares. In the institutional money management industry, on the other hand, each product typically has relatively few investors. Indeed, the largest client invested in a product holds at least one-half of the assets managed by the product in nearly one-half (46 percent) of all investment products. This extent of concentration makes many investors in the institutional arena very powerful by comparison. For example, though not desirable, a mutual fund investor s decision to leave the fund will have a fairly limited effect on the assets under management, the primary determinant of mutual fund managers compensation. It would take a strongly correlated action of many mutual fund investors to create a visible dent. By contrast, an investor s (plan sponsor s) decision in the institutional asset management industry to fire the manager often effectively shrinks the assets under management by a large percentage, perhaps even up to one-half of total assets or more. Whereas it is clear that not all client departures can be prevented (especially those not undertaken for reasons related to performance), this implicit threat of asset base depletion wields considerable power. Lack of transparency and the nature of this industry create a fertile soil for the emergence of agency issues. In their seminal work on institutional money management firms, Lakonishok, Shleifer, and Vishny (1992) have identified and described some 3 For excellent reviews of the institutional framework in which the institutional money management industry operates and many of the related issues see Lakonishkok, Shleifer, and Vishny (1992) and Goyal and Wahal (2008) in the academic literature, as well as Fabozzi (1997), Logue and Rader (1998), and Travers (2004) in the practitioneroriented literature. 3

4 agency issues, primarily related to the delicate interplay among plan sponsors, consultants, and money management firms, and their respective incentives. This paper looks into another facet of agency issues in the institutional money management industry, associated with the potential for strategic allocation of performance through cross-subsidization across investment products within the firm. In the mutual fund arena, Gaspar, Massa, and Matos (2006) present evidence consistent with strategic cross-subsidization of those mutual funds in the family that have higher value to the family (by virtue, for example, of their track record) by the funds in the family that have lower value. Availability of quarterly fund holdings enables Gaspar, Massa, and Matos (2006) further to explore mechanisms of strategic allocation of performance and uncover evidence of strategic IPO allocations and opposite trades across funds within the family. 4 The primary sources of data for this study are the 58 quarterly surveys from June 1993 to December 2007 from the Mobius Group (toward the end of the sample period, Mobius was subsumed by Informa PSN). Mobius was one of the primary vendors of institutional money management data, used by most large pension fund sponsors and endowment funds to identify money managers, study their track records, and consider a range of other variables relevant for the investment decision-making process. We find strong and robust evidence of strategic cross-subsidization in the institutional money management industry across products managed by the same firm. We begin by developing a proxy for the availability of resources to cross-subsidize a product. We consider all products in the same firm with substantially larger assets under management than the product under consideration itself and calculate the 4 Similarly, in the domain of large mutual fund families, Guedj and Papastaikoudi (2008) find that persistent excess performance is related to the number of funds in the family, a measure of the latitude the family has in allocating resources unevenly between its funds. Relating their work to Elton, Gruber, and Green (2007), who find that mutual fund returns are more closely correlated within than between fund families, Guedj and Papastaikoudi (2008) further show that the average pairwise correlations between funds in the same family are lower in larger families and proceed to interpret these results not as evidence that bigger families maintain superior performance persistence across all of their funds, but, instead, that they do so selectively because there are resource limitations and good investment ideas and strategies at the family level are not fully scalable and hence cannot be applied to the entire universe of funds within one family. The net effect is that there is selective allocation of these ideas, presumably guided by the effects of that allocation on the flow-performance relation of the funds involved, as well as on the aggregate inflows to the family. 4

5 product s BTRatio, that is, the ratio between the sum of the assets of all those substantially larger products and the assets of the entire firm (less the assets under management of the product under consideration itself 5 ). The cornerstone of this approach is the asymmetry embedded in the effect of taking away relatively minor extent of performance from a substantially larger product and applying it toward the performance of a smaller product, thereby enhancing the performance of the latter quite substantially. 6 In our empirical specifications, substantially larger translates into twice as large or larger. Our robustness checks show that the results are not sensitive to the choice of this threshold. 7 Finally, we define for each product an indicator variable BTHigh. It is set to one if the product s BTRatio is in the top third of the distribution of BTRatio values, and to zero otherwise. 8 BTHigh thus captures the presence of resources for strategic performance allocation. We acknowledge readily that allocating performance away from larger products toward smaller products need not be the only channel that accomplishes cross-subsidization. In that sense, BTHigh is a noisy proxy that likely underestimates overall opportunities for cross-subsidization. Thus, in a very real sense, cross-subsidization through reallocating performance from larger to smaller products represents a lower bound on the overall extent of strategic cross-subsidization. We further define two indicator variables, Top and Dominant. Top characterizes the products that have high value for the firm because of their historical performance record. As pointed out in the mutual fund literature, mutual fund families favor historically strong performers (Gaspar, Massa, and Matos, 2006), as continued strong 5 Excluding the product s assets from the denominator of BTRatio is inessential. The alternative that encompasses the product assets results in a measure of BTRatio that is extremely highly correlated with the one we use throughout (the correlation is ) and the results are indistinguishable from those we report in the paper. 6 To pick an illustrative example, a one-million dollar position in an asset that experiences a 100% return would contribute ten basis points to the performance of a one-billion dollar portfolio. At the same time, it could increase the performance of a one-hundred million dollar portfolio much more substantially, by a whole percent. 7 For example, focusing on products that are at least four times as large as the product under consideration does not affect the results reported in the paper. 8 The value of BTHigh at the th percentile of distribution is , indicating that products for which more than 90.24% of the assets managed by the firm are available for cross-subsidization are regarded as those with high level opportunity for cross-subsidization. The use of alternative cutoffs, for example, at the 50 th percentile of BTRatio (0.7392), though predictably decreasing the magnitude of the effect, preserves its strong statistical significance and still has a large economic magnitude. These robustness checks are reported in Section VII. 5

6 performance leads both to the larger inflows of investment to the fund (e.g., Ippolito, 1992; Chevalier and Ellison, 1997; Sirri and Tufano, 1998; Ivković and Weisbenner, 2009, among others) and to flow spillover effects (e.g., Nanda, Wang, and Zheng, 2004) that benefit other products managed by the same mutual fund family. Indeed, we show in Section III the existence of these phenomena in the institutional money management industry too. Namely, we find that high performance affects flows into products both directly (a results consistent with Del Guercio and Tkac, 2002, and Heisler, Knittel, Neumann, and Stewart, 2007) and indirectly, through flow spillover, a finding previously undocumented in the institutional investment management literature. Top is set to one if the product s annual returns have been ranked in the top quintile of returns of all products in the sample pursuing the same investment objective, and to zero otherwise. 9 Dominant characterizes the products that have a very concentrated client base. Such concentrated ownership, as discussed earlier, is a radically different feature of the institutional money management industry relative to the mutual fund industry. Because the departure of a dominant client significantly alters total assets under management, whether a product has a dominant client may be an important consideration in the process of strategically allocating performance across a firm s products. Moreover, ceteris paribus, highly concentrated ownership of the assets in a product indicates that the firm may be keenly interested in cultivating the relationship with the (few) client(s) invested in the product, and thus may be particularly inclined to allocate performance strategically toward such a product. Dominant is set to one if the ratio of the assets held in the largest portfolio in the product and the product s total assets under management is in the top third of the distribution of this ratio in the sample, and to zero otherwise Consideration of top decile as the cutoff for strong historical performance, predictably, makes the results stronger (see Section VII for results of the related robustness checks)., 10 The value of Dominant at the th percentile of distribution is , indicating that products for which more than 82.86% of the assets managed by the product are held by its largest client (approximately five sixths) are regarded as those with a dominant client. Once again, the use of alternative cutoffs, for example, requiring that the largest portfolio in a product accounts for one-half or more of the product s overall assets under management (a feature shared by 46% observations in the sample), does not alter the results reported in the paper. 6

7 Our canonical specification relates a product s annual performance in excess of its investment benchmark 11 to these three indicator variables and all of their interactions, thus implementing the difference-in-difference-in-difference (DDD) approach. The covariates in the specification also include lagged product s annual performance in excess of the investment benchmarks, 12 controls for the product s and the firm s assets under management (both linear and quadratic terms, to capture potential nonlinearities), as well as firm, investment objective, and year effects. Our panel estimations incorporate adjustment of standard errors by clustering that accounts for heteroskedasticity and dependence of observations across the same firm. We find very robust evidence of strategic performance allocation toward high-value products in the firm, to the extent of around 1.6 percentage points per year. Because of our reliance on the DDD approach, as well as the inclusion of numerous controls and effects, this estimate indeed reflects cross-subsidization. We next consider the loss of performance that plagues the products that are likely candidates to have performance taken away from them in favor of smaller, highvalue products from the firm. That performance loss is estimated at around 73 basis points per year. However, the overall effects on assets under management are more intricate, as they also depend on the flow-performance relation (and a decrease in future flows because of a 73-basis point performance reduction), as well as a flow spillover that accrues to the fund otherwise slated for provision of cross-subsidization. Regression analyses reported in Table III and Table IV enable us to estimate that the overall effect in year t+1 is a net gain in flows of 1.1 to 3.2 percentage points (obtained by subtracting 73 basis points from the percentage-point star effect). The effect in year t+2 is a percentage-point decline prompted by the response of the flow to a 73-basis-point decrease during year t We have calculated adjustments relative to investment benchmarks both relative to the median performance among all products pursuing the same objective and relative to the returns to a broadly diversified style index provided by Russell or Standard and Poor s. The results are consistent across all of these methods of benchmark adjustments. 12 Exclusion of lagged returns from the specification does not alter the results. In fact, all the directions and statistical significances of key coefficients are preserved, and their magnitudes are generally larger by one-third. 7

8 Despite lack of detailed transaction-level data of periodic snapshots of holdings, we can characterize the strategic performance allocation from several perspectives. First, we find that the extent of cross-subsidization is substantially larger around 2.6 percentage points per year if the recipient is a high-value product with a concentrated client base (with Dominant =1). The effect is much smaller and statistically indistinguishable from zero in cases of high-value products with a diffuse client base. In that sense, our results show that the extent of strategic performance allocation varies with demand (interpreted as the dominance of the largest client in the portfolio). We also show that the extent of cross-subsidization varies with measures of supply of performance. We identify and study three channels and ascertain the importance of all three: performance of the styles of bigger products in the firm, availability of IPO opportunities, and illiquidity of the products investment styles. We find sizeable and, in many cases, statistically significant effects. We also show that there is further differentiation along each of the channels by the clients demand (as characterized by the dominance of the largest client). Next, we proceed to formulate and test two hypotheses that relate firm size to the extent of strategic performance allocation. First, when contrasting large and small firms, a natural hypothesis is that small firms may not have the depth of resources (nor the requisite heterogeneity) to be as effective at cross-subsidization as large firms are. The second hypothesis contrasts large, medium, and small firms, and states that the extent of strategic performance allocation will be the largest among medium firms, that there will be strategic performance allocation in large firms, albeit some less pronounced than it is for medium firms, and that small firms, given their scarcity of resources, will feature the least extent of strategic performance allocation. We find strong support for both hypotheses. Our robustness checks indicate that our results are very robust. We explore different classification criteria within our econometric approach, a matching product technique (employed in Gaspar, Massa, and Matos, 2006), and a portfolio formation 8

9 technique. Our baseline estimates of strategic performance allocation hold up remarkably well. The remainder of the paper is organized into eight sections. Section I reviews data sources and the sample. In Section II, we establish the baseline result indicative of strategic performance allocation. Section III analyzes the other perspective, the losses to the larger products that provide the cross-subsidization, and all the ensuing effects that stem from a careful consideration of the flow-performance relation. Section IV analyzes the role of demand for strategic performance allocation, defined as the concentrated ownership of assets in a product. Section V considers three channels for supply and finds strong evidence that all three are at play (performance of the styles of bigger products in the firm, availability of IPO opportunities, and illiquidity of the products investment styles). In Section VI we study the role of firm size. In Section VII, we report three robustness checks, each of which confirms our baseline results with remarkable consistency. Section VIII considers the wealth transfers that result from strategic performance allocation and provides concluding remarks. I. Data Sources and Sample Overview We compile data from several sources. We have obtained 58 quarterly surveys from June 1993 to December 2007 from the Mobius Group (toward the end of the sample period, Mobius was subsumed by Informa PSN). Mobius was one of the primary vendors of institutional money management data, used by most large pension fund sponsors and endowment funds to identify money managers, study their track records, and consider a range of other variables relevant for the investment decisionmaking process. The data are self-reported. As a gauge of data accuracy, we have crosschecked the returns of the lowest return decile of managers with a private source of return data, and have found the Mobius data to be of high quality. 13 Aside from quarterly product returns, the Mobius data contain a range of firm and product 13 Ten managers were in Mobius and hired by the Virginia Retirement System. We had access to the internal records of the VRS from The data are identical to those in Mobius. 9

10 characteristics, including products firm affiliation, assets under managements, total number of portfolios, the assets of the largest portfolio in the product, and so on. For some of our analyses, we use investment style benchmarks from Russell (see Table AI in the Appendix). Finally, in some of our robustness checks we use standard performance benchmarks and risk factors available from the Kenneth French data library (RmRf, SMB, HML, and UMD). 14 Our panel analyses extend over annual data. Most of the variables are available with annual frequency. We compound quarterly returns into annual product returns. Our sample consists of all product-year observations that have the requisite variables for our analyses. We exclude mutual funds reported in Mobius by screening out all product-year observations from our sample that have 100 or more clients). Table I presents summary statistics. TABLE I ABOUT HERE II. Baseline Results Our baseline specification relates products objective-adjusted annual returns to indicator variables BTHigh, Top, and their interaction, as well as a number of controls and effects: OAR i,t+1 = β 0 BTHigh i,t + β 1 Top i,t + β 2 BTHigh i,t x Top i,t + controls + effects + ε i,t+1. (1) Regression coefficient β 0 reflects the performance differential between past nontop performers with high presence of bigger products (with more opportunities for strategic performance allocation) and past non-top performers with low presence of bigger products (with fewer opportunities for strategic performance allocation). The sum of regression coefficients β 0 and β 2, β 0 + β 2, reflects the extent to which the performance of past top performers will be higher in the settings with more resources for strategic performance allocation (with high presence of bigger products) than in the 14 The data for these (and numerous other) time series are available at the following URL: 10

11 setting with fewer resources for strategic performance allocation (with low presence of bigger products). The differential, the difference-in-difference estimator β 2, is pivotal. It reflects the extent to which differential between returns on products with high presence of bigger products and low presence of bigger products is higher for past top performers than it is for past non-top performers. A positive and statistically significant coefficient β 2, therefore, would reveal evidence of strategic performance allocation. Regression coefficient β 1 reflects the performance differential between the products with fewer opportunities to receive strategic performance allocation that had been top performers and such products that had not been top performers. β 1 + β 2 reflects the extent to which the performance of products with high presence of bigger products (with more opportunities for strategic performance allocation) will be higher if they had been top performers than if they had not been top performers. Finally, once again, the differential, that is, the difference in difference estimator β 2 reflects the extent to which differential between returns on products with high presence of bigger products and low presence of bigger products is higher for past top performers than it is for past non-top performers. 15 Controls are lagged objective-adjusted annual product returns (expressed in percentage points), product assets and firm assets (both in logarithmic form), as well as their squares (to control for size effects very carefully). Effects include year effects, investment objective effects, and firm effects, thus ensuring that any variable that varies only by time, objective, or firm is absorbed and cannot explain any of our regression findings. Moreover, we calculate standard errors by clustering in a way that allows for heteroskedasticity as well as correlation across observations associated with the same firm. Annual product returns are objective-adjusted in three different ways. First, by subtracting from product annual returns the contemporaneous return to the style 15 The interpretation of regression coefficients from Equation (1) is also summarized in tabular form in Table A.II in the Appendix. 11

12 benchmark defined by the appropriate Russell index. 16 Second, by subtracting from product annual returns the contemporaneous median return among all the products pursuing the same investment objective. The third adjustment method addresses the potential concern that the covariates that control for product size (that is, its asset under management) may not suffice to control for size effects adequately. Accordingly, the benchmarks adjustment is done more stringently, by grouping all products within each objective and within each year into quartiles according to their assets under management and subtracting from product annual returns the contemporaneous median return among all the products pursuing the same investment objective and belonging to the same size quartile. The results of estimating all these regressions are displayed in Table II. It features three panels, in accordance with the approaches to return benchmark adjustment. In each panel, the first column presents the results of fitting a simpler specification, featuring BTHigh only (as well as all the other controls). Regardless of the specification, that is, across all three panels, the products with a high fraction of bigger products outperform those without it by 62 to 66 basis points per year. Of course, it is difficult to ascertain to what extent this result stems from strategic performance allocation. Among other alternative explanations, it could also be that a high percentage of much bigger products in the same firm is indicative of managerial skill present in the firm. The second column in each panel presents the results of fitting the specification from Equation (1). Moreover, the two bottom rows of Table II in the second column of each panel feature estimates of β 0 + β 2 (labeled in the table as BTHigh + BTHigh x Top for readability) and β 1 + β 2 (labeled in the table as Top + BTHigh x Top for readability). Coefficient estimates of β 1 (labeled in the table as Top) show that, among products with fewer opportunities to receive strategic performance allocation, the performance differential between the products that had been top performers and had not been top performers is between 71 basis point per year (Panels B and C) and See table AI in the Appendix for the indexes used for style adjustment. Using corresponding Standard and Poor s style benchmarks yielded very similar results. 12

13 basis points (Panel A), with high levels of statistical significance across all three. Estimates of β 1 + β 2 (the bottom row of Table II, labeled as Top + BTHigh x Top) show substantially larger performance differentials between products that had and had not been top performers among products with more opportunities to receive strategic performance allocation (that is, with high values of BTRatio, as characterized by the value of BTHigh equal to 1); across the three panels, these estimates are very similar and are all highly statistically significant; they range from 2.3 percent (Panel B), 2.33 percent (Panel C), to percent (Panel A). These differentials may partly reflect momentum in portfolio returns, but there also is a very large component that is consistent with strategic performance allocation within the firm. That component is captured directly by the difference-in-difference coefficient estimate of β 2 (labeled in the table as Top + BTHigh x Top). Indeed, β 2 reflects the extent to which the difference between returns on products that had been and had not been top quintile performers last year is higher for products with more opportunities for strategic performance allocation than for the products with fewer such opportunities. Once again, the estimates are very similar across the three panels once again, and are all highly statistically significant; they range from 1.54 percent (Panel A), 1.60 percent (Panel B), to percent (Panel C). TABLE II ABOUT HERE III. Strategic Performance Allocation Away from Products The analyses from the previous section uncover firms strong and robust tendency to allocate performance strategically, notably in the circumstances in which the products that have more opportunities to receive it (that is, high values of BRatio, as captured by BTHigh = 1) and are regarded as high-value products by virtue of their strong recent performance. This reallocation, estimated at approximately 1.6 percent per year, needs to come from some other product(s) in the firm. We asses this, the other side of strategic performance allocation, by computing for each product STRatio, the ratio between the sum of all the assets of the products managed by the same firm that are 13

14 each at least twice as small or smaller than the assets of the product, and the assets of the entire firm (less the assets under management of the product under consideration itself). Consistent with the cross-subsidization hypothesis, the performance of the product should be weaker if there is a broader set of products in the firm to which performance needs to be allocated strategically. We define the indicator variable STHigh for each product to characterize whether the product may extent cross-subsidization toward other, smaller products in the firm. We set STHigh to one for the values of STRatio in the top third of the STRatio distribution, and to zero otherwise. The results from the previous section show that firms tend to support strong historical performance of their products over the next time period. 17 Looking at this from the perspective of the products that, presumably, would provide that support, the cross-subsidization hypothesis further suggests that there should be strategic performance allocation away from the products, and toward the smaller products with strong historical performance. To capture this, we define the indicator variable Top Among Smaller in Firm i,t and set it to one if the one-year performance of at least one of the other smaller products (those with one-half of the assets of product i or smaller) in the same firm has been in the top quintile in its investment objective for the year. Top Among Smaller in Firm i,t is set to zero otherwise. We employ STHigh, Top Among Smaller in Firm, and their interaction, to test these predications: OAR i,t+1 = β 0 STHigh i,t + β 1 Top Among Smaller in Firm i,t + (2) β 2 STHigh i,t x Top Among Smaller in Firm i,t + controls + effects + ε i,t+1. The controls and effects are the same as they were in the preceding analyses. Similar to the analyses reported in the previous section, the key estimator is the differential, the difference-in-difference estimator β 2. As shown in Table III, the estimate of strategic performance allocation away from larger products managed in firms with 17 The issues of star creation in the mutual fund industry have been studied in, among, others, Nanda, Wang, and Zheng (2004). 14

15 many smaller products in the firm ranges between 62 and 89 basis points per year, depending on the method of benchmark-adjusting annual product returns. TABLE III ABOUT HERE The results reported in Table II and Table III are very consistent across all three respective panels, suggesting that they are not sensitive at all to the method of benchmark-adjusting annual product returns. This holds for all subsequent analyses too. Thus, to avoid repetition, we henceforth report only the results based upon the most stringent of the three methods, based on grouping all products within each objective and within each year into quartiles according to their assets under management and subtracting from product annual returns the contemporaneous median return among all the products pursuing the same investment objective and belonging to the same size quartile (Panel C in Table II and Table III). The products that are well-suited for strategic performance allocation away from them (products with STHigh = 1) may experience both direct and indirect effects of such performance reallocation. According to the estimates from Table III, once there is a top-quintile performer among the products smaller than product i in the firm at the end of year t, if product i has STHigh = 1, product i will keep on supporting such top performers, incurring the direct effect of a 73 basis-point reduction in the assets it manages. This reduction follows mechanically, from a 73 basis-point reduction in product i s own performance in year t+1. Indirect effects stem from the flow-performance relation. The first indirect effect, in light of the positive relation between flows to a products in a year and the product s performance during the previous year, is a decrease in flows to product i in year t + 2 because the performance of a STHigh = 1 product had decreased over the year t + 1 by around 73 basis points (Panel C, Table III). The second effect is akin to the star phenomenon has potential to create a gain for the product. The existence of a top performer among smaller products in the firm at the end of year t (in this context, 15

16 captured as a top quintile performer), may lend itself to a star phenomenon similar to that documented in the mutual fund literature (Nanda, Wang, and Zheng, 2004), whereby there will be a flow spillover to product i in year t+1, prompted by a strong performance of stars among smaller products in the firm. We assess these indirect effects by estimating the following flow-performance relation: Flow i,t+1 = β 0 Quintile_5 i,t + β 1 Quintile_4 + β 2 Quintile_2 + β 3 Quintile_1 + (3) β 4 Top Among Smaller in Firm i,t + controls + effects + ε i,t+1. We measure the flows during year t+1 in two ways. The first measure, based on assets, is defined as the change in assets from year t to t+1 divided by assets at the end of year t: Flow i,t+1 = (Assets i,t+1 (1+R i,t+1 ) x Assets i,t )/Assets i,t. The second measure, based on the change of the number of portfolios in the product, is defined analogously: Flow i,t+1 = (NumberOfPortfolios i,t+1 NumberOfPortfolios i,t )/NumberOfPortfolios i,t. Product performance is captured through four indicator variables, Quintile_5 i,t, Quintile_4 i,t, Quintile_2 i,t, and Quintile_1 i,t, where Quintile_k i,t is set to one if the product s one-year performance is in quintile k (5 th quintile denotes top performers) of all product returns during the year in the same objective and the same size quartile (as defined by portfolio assets). The middle quintile is omitted. Product performance is expressed through these indicator variables to allow for potential nonlinearity of the flow-performance relation. 18 The indicator variable Top Among Smaller in Firm i,t is set to one if the one-year performance of at least one of the other smaller products (those with one-half of the assets of product i or smaller) in the same firm has been in the top quintile in its investment objective for the year, and to zero otherwise. The remaining covariates are lagged flows, as well as size controls: product assets and firm assets (both in logarithmic form), as well as their squares (to control for potential nonlinearity of size effects). 18 The nonlinearity in the flow-performance relation, though prevalent in the mutual fund industry (e.g., Chevalier and Ellison, 1997; Ippolito, 1992; Sirri and Tufano, 1998), is not as pronounced in the institutional money management industry (e.g., DelGuercio and Tkac, 2002). 16

17 The results are presented in Table IV. Panel A reports results based on the assetbased measure of flows, whereas Panel B reports results based on the measure of flows obtained on the basis of the number of portfolios in the product). The flow-performance relation is consistent with previous literature (e.g., DelGuercio and Tkac, 2002), featuring a steep flow-performance relation that appears to feature less pronounced nonlinearities in this industry than in the mutual fund industry. The slope of the flowperformance relation is smaller for the products with STHigh = 1 than it is for the products with STHigh = 0. This is not surprising because the latter tend to be smaller products. The coefficient of central interest in this context is associated with the flow spillover effect, captured by Top Among Smaller in Firm i,t. The presence of a top-quintile performer in the firm among the products with assets that are one-half the size of assets of product i or smaller is associated with a 1.56 (1.50) percentage-point increase in assetbased (portfolio-based) measure of flows to product i in year t+1, statistically significant at the 10-percent level. Estimated over the subsample of products tapped for strategic cross-subsidization (those with STHigh = 1), the corresponding increase in flow in year t+1 is somewhat larger (1.84 percentage points, significant at the 10-percent levels) for asset-based flow measures and about twice as large (3.91 percentage points, significant at the 1-percent level) for portfolio-based flow measures. TABLE IV ABOUT HERE The estimates presented in Table III and Table IV enable simple back-of-the envelope calculations that gauge the indirect effects of strategic cross-subsidization of a high-value product in the firm (captured in our analyses as a top-quintile performer within its objective in year t) during year t+1. The first indirect effect is associated with a 73 basis-point performance decrease over year t+1. Based on the distribution of benchmark-adjusted product returns, we estimate that a 73 basis-point decline corresponds roughly to a two-percentile decline in performance ranking. The difference 17

18 in coefficient magnitudes associated with two successive quintiles for products with STHigh = 1, very consistent across successive quintiles within the two panels, is around 6.3 percentage points for asset-based flows and 7.6 percentage points for portfolio-based flows. It captures the decline in flow associated with a twenty-percentile decline in performance. One-tenth of that magnitude, corresponding to the two-percentile decline estimated just above, corresponds to the crudely estimated first indirect effect of a percentage-point decline in flows to the product over the next year, year t+2. The second indirect effect is a percentage-point increase in flows in year t+1 (depending on the way of expressing flows across the two panels) if there had been a top-quintile performer in the remainder of the firm at the end of year t. In sum, the overall effect in year t+1 is a net gain in flows of 1.1 to 3.2 percentage points (obtained by subtracting 73 basis points from the percentage-point star effect). The effect in year t+2 is a percentage-point decline prompted by the response of the flow to a 73-basis-point decrease during year t IV. Demand for Strategic Performance Allocation The key result in the preceding section is evidence of strategic crosssubsidization at the level of around 1.6 percent per year (Table II). This section builds upon that result by relating the extent of strategic cross-subsidization to a measure of demand for it and to three measures of supply of it. We take advantage of a particular feature of the institutional money management industry, a high extent of concentration of client portfolios in products. As discussed in the introductory section, although the structure of investment options 19 These figure are conservative lower bounds. In an alternative specification, featured in Table A.III in the Appendix, performance of product i in year t is captured by its objective-adjusted (OAR i,t ). The coefficient associated with OAR i,t ranges from 0.59 (asset-based flows for STHigh = 1 products) to 0.79 (number of portfolio-based flows for STHigh = 1 products) across specifications that parallel those reported in Table V, leading to about 43 to 58 basis points decline in flows in year t+2 (only around 40% of the 1.2 percentage-point effect estimated on the basis of the specification from Table V). The star coefficient, associated with Top Among Smaller in Firm, ranges in these alternative specifications from 3.0 percentage points (asset-based flows for STHigh = 1 products) to 4.9 percentage points (number of portfoliobased flows for STHigh = 1 products), that is, it is about 50% stronger. These calculations would suggest that the support by a STHigh = 1 product of a highly ranked product elsewhere in the firm is associated with a net gain of 2.3 to 4.2 percentage points in year t+1 and only a one-half percentage point net loss in year t+2. 18

19 in the institutional money management industry may resemble that offered by the mutual fund industry, the two industries are quite different. Aside from a vast difference in the extent of transparency, as well as differences in institutional framework, 20 a pivotal difference is the structure of their respective investors. Mutual funds, investment vehicles inaugurated historically in pursuit of the goal of providing small investors access to diversified investment, have a broad investor base. A mutual fund shares is held by thousands, sometimes even tens of thousands of investors or more, each of whom typically holds a miniscule fraction of fund shares. In the institutional money management industry, on the other hand, each product typically has relatively few investors. Indeed, the largest client invested in a product holds at least one-half of the assets managed by the product in nearly one-half (46 percent) of all investment products. This extent of concentration makes many investors in the institutional arena very powerful by comparison. For example, though not desirable, a mutual fund investor s decision to leave the fund will have a fairly limited effect on the assets under management, the primary determinant of mutual fund managers compensation. It would take a strongly correlated action of many mutual fund investors to create a visible dent. By contrast, an investor s (plan sponsor s) decision in the institutional asset management industry to fire the manager and take their portfolio elsewhere often shrinks the assets under management by a large percentage, perhaps even up to onehalf of total assets or more. Whereas it is clear that not all client departures can be prevented (especially those not undertaken for reasons related to product performance), this implicit threat of asset base depletion wields considerable power. Because the departure of a dominant client significantly alters total assets under management, whether a product has a dominant client may be an important consideration in the process of strategically allocating performance across a firm s products. In that sense, on 20 As noted earlier, for excellent reviews of the institutional framework in which the institutional money management industry operates and many of the related issues see Lakonishkok, Shleifer, and Vishny (1992) and Goyal and Wahal (2008) in the academic literature, as well as Fabozzi (1997), Logue and Rader (1998), and Travers (2004) in the practitioner-oriented literature. 19

20 the margin, the firm may wish to cater to the demand for returns from products with dominant clients to a higher extent than to the comparable demand from products without dominant clients. This tendency may also be related to the firms propensity to build relationships with powerful clients by managing their portfolios in arrangements in which there are fewer clients (perhaps none!) in the same product, thus providing (nearly) exclusive attention to such clients. Ceteris paribus, highly concentrated ownership of the assets in a product indicates that the firm may be keenly interested in cultivating the relationship with the (few) client(s) invested in the product, and thus may be particularly inclined to allocate performance strategically toward such a product. These considerations produce a testable implication that the extent of crosssubsidization should be larger when directed toward products with a more concentrated client base. We proceed to test this implication. First, we define an indicator variable Dominant i,t to capture products that have a very concentrated client base. Dominant i,t is set to one if the ratio of the assets held in the largest portfolio in the product and the product s total assets under management is in the top third of the distribution of this ratio in the sample, and to zero otherwise. 21 Next, we estimate the regression from Equation (1) separately for products that feature dominant clients and for those that do not. Panel A of Table V features the difference-in-difference estimates (as in the previous section, regression coefficients associated with BTHigh x Top) of strategic performance allocation for the subsample of products with a dominant client (the first column) and the subsample of products without a dominant client (the second column). The effect is substantially stronger among products with dominant clients it is as large as 2.6 percent per year, that is, 62% stronger than the baseline effect estimated over the full sample. The extent of cross-subsidization among products without a dominant 21 The value of Dominant at the th percentile of distribution is , indicating that products for which more than 82.86% of the assets managed by the product are held by its largest client (approximately five sixths) are regarded as those with a dominant client. Once again, the use of alternative cutoffs, for example, requiring that the largest portfolio in a product accounts for one-half or more of the product s overall assets under management (a feature shared by 46% observations in the sample), does not alter the results reported in the paper.. 20

21 client is not statistically distinguishable from zero (the point estimate is 46 basis points per year, but the standard error is 56 basis points). To evaluate the statistical significance of the difference between the two, we estimate the following specification over the full sample: OAR i,t+1 = β 0 BTHigh i,t + β 1 Top i,t + β 2 BTHigh i,t x Top i,t + controls + effects + (4) (β 0 BTHigh i,t + β 1 Top i,t + β 2 BTHigh i,t x Top i,t + controls + effects) x Dominant i,t + ε i,t+1. That is, every covariate from the baseline regression specification (Equation (1)) is also multiplied by the indicator variable Dominant i,t. As before, controls include lagged objective-adjusted annual returns, product assets and firm assets (both in logarithmic form), as well as their squares (to control for size effects very carefully). Effects include year effects, investment objective effects, and firm effects, thus ensuring that any variable that varies only by time, objective, or firm is absorbed and cannot explain any of our regression findings. The coefficient β 2, associated with the triple interaction BTHigh x Top x Dominant, captures the desired difference-in-difference-in-difference. 22 Its point estimate, by construction, is equal to the difference between the two difference-indifference estimates reported for each subsample). Its magnitude is 2.14 percent, and it is highly statistically significant at the one-percent level. TABLE V ABOUT HERE V. Supply of Strategic Performance Allocation As discussed in the introductory section, one of the key characteristics of the institutional money management industry is its relative obscurity. Compared with the 22 The number of observations in this regression is 25,618, a 12% decrease relative to the number of observations reported in Panel C of Table III. This is a reflection of the fact that the assets of the largest portfolio are reported for 88% of the observations. 21