How does time variation in global integration affect hedge fund flows, fees, and performance? Abstract

Transcription

1 How does time variation in global integration affect hedge fund flows, fees, and performance? October 2011 Ethan Namvar, Blake Phillips, Kuntara Pukthuanghong, and P. Raghavendra Rau Abstract We document the determinants of time variation in the level of integration of hedge funds with global equity market risk factors. We find that integration increases significantly during periods of stress in financial markets, suggesting that hedge fund exposure to systematic risk peaks precisely when investors most need diversification. We document manager and fund characteristics that reliably predict integration levels and trends across financial market states. Specifically, better educated and more experienced managers have greater skill in managing integration and this ability is more pronounced during down markets. Younger and smaller funds typically have lower integration while distressed funds are characterized by higher integration levels. Funds with higher and increasing integration realize significantly lower flows. The relation between flow and integration is markedly asymmetric, with investors showing little sensitivity to integration decreases regardless of market conditions. Though investors pay high fees to invest in low integration funds, these higher costs are more than offset by the superior performance of the fund. We also find that funds appear to preferentially time fee increases to coincide with decreases in integration. JEL Classification: G11; G14; G23; G32 Keywords: Hedge Funds; Integration; Alternative Investments * The authors are from the Haas School of Business, University of California, School of Accounting and Finance, University of Waterloo, College of Business, San Diego State University and the University of Cambridge. Corresponding author: Blake Phillips, School of Accounting and Finance, University of Waterloo, 200 University Avenue West, Waterloo, Ontario, Canada, N2L 3G1, brphilli@uwaterloo.ca.

2 How does time variation in global integration affect hedge fund flow, fees, and performance? Abstract We document the determinants of time variation in the level of integration of hedge funds with global equity market risk factors. We find that integration increases significantly during periods of stress in financial markets, suggesting that hedge fund exposure to systematic risk peaks precisely when investors most need diversification. We document manager and fund characteristics that reliably predict integration levels and trends across financial market states. Specifically, better educated and more experienced managers have greater skill in managing integration and this ability is more pronounced during down markets. Younger and smaller funds typically have lower integration while distressed funds are characterized by higher integration levels. Funds with higher and increasing integration realize significantly lower flows. The relation between flow and integration is markedly asymmetric, with investors showing little sensitivity to integration decreases regardless of market conditions. Though investors pay high fees to invest in low integration funds, these higher costs are more than offset by the superior performance of the fund. We also find that funds appear to preferentially time fee increases to coincide with decreases in integration.

3 1. Introduction Over the past decade, hedge funds have become increasingly popular as alternative investment strategies that have low correlations with other funds and popular asset benchmarks. Hedge fund managers typically market their funds as being relatively uncorrelated with broad market movements, thus providing downside protection for investors in times of crisis. 1 Titman and Tui (2011) argue that superior hedge funds distinguish themselves by choosing lower levels of exposure to systematic risk factors. Specifically, they show that funds with lower sensitivity to systematic risk factors have higher information ratios, Sharpe ratios, and alphas. In addition, controlling for past performance, funds with low systematic risk attract greater investment inflows and are able to charge higher fees. However, this leaves a conundrum. If maintaining a low systematic risk exposure leads to higher performance and high management revenue, why don t all fund managers just reduce the level of systematic risk in the funds they manage, perhaps by varying the amount of funds they hold in the form of risk free assets? One possible answer is that choosing a consistently low level of exposure to systematic risk is not necessarily the best strategy for a hedge fund. Over the past two years, the popular press reports that significant numbers of hedge funds have performed poorly during periods of extreme volatility. 2 However Brown, Hwang, In, and Kim (2011) show that while hedge funds with a high systematic risk contribution underperform relative to funds with a low systematic risk contribution during crisis periods, on average they significantly outperform over the entire period It is plausible therefore, if investors are unable to distinguish between the alpha of the fund and its exposure to risk factors, funds with high risk exposures may earn higher fund inflows and consequently earn significantly higher fees in up markets, which compensate them for the lower fees earned during down markets. Alternatively, timing integration to peak during periods of financial market strength and reducing integration when markets weaken may jointly optimize hedge fund fees and fund performance. It is 1 For example, Daniel Och, chairman and chief executive of Och-Ziff Capital Management, stated Investors continue to actively seek access to investment managers that generate risk-adjusted returns which have a low correlation to the equity markets and consistently protect capital. We believe that this focus has led to increased allocations to the hedge fund industry in the first half of this year, and that this acceleration will continue, quoted in Ahmed, Azam, Och-Ziff Quarterly Earnings rise 19%, New York Times, August 2, See for example, Ahmed, Azam, Hedge funds get unfamiliar taste of losing, New York Times, August 11, 2011, p. B1. -1-

4 therefore important to understand the determinants of time variation in systematic risk exposure of hedge funds and its effects on investor fund preferences and hedge fund performance. We examine these issues in this paper. To measure systematic risk exposure, we draw on the market integration analysis in Pukthuanthong and Roll (2009). Specifically, we calculate the matrix eigenvectors from the equity indexes of 38 countries and use these values to estimate principal components (PCs) in the subsequent year. This process is repeated for each calendar year, generating a time-series of out-of-sample global risk proxies which capture commonalities across international equity markets. We then regress hedge fund returns on these proxies and use the R 2 from this regression as our measure of systematic risk exposure. We use this approach in preference to correlation or factor model analysis that have previously been utilized to measure hedge fund integration across styles or with equity markets (see for example, Boyson, Stahel, and Stulz, 2010 or Titman and Tiu, 2011). Correlation analysis is not appropriate for measuring hedge fund integration as hedge fund returns are not normally distributed. While smoothing returns can potentially mitigate bias when using correlation analysis, our measure of integration is not dependent on any assumptions of normality in the distribution of returns. Our estimate is also free of any assumptions regarding factor selection, eliminating the potential for omitted variable bias, which is a significant consideration in factor models. In our approach, we directly measure commonalities in the returns of global equity markets without requiring any underlying knowledge of the source of these commonalities and variation in source significance over time. Measuring integration relative to these risk factors yields estimates of integration that are 60% and 30% higher than suggested by correlation or factor model approaches, respectively. We find that the level of integration varies significantly across hedge fund styles. For example, 64% of the variation in the average return to the Convertible Arbitrage style is explained by global risk factors. In contrast, only 24% of average return variability is explained by the same factors for the Fixed Income Arbitrage style. While the relative integration across hedge fund styles is reasonably time invariant, there is a considerable degree of variation in aggregate integration which coincides with market conditions. Partitioning our sample based on the TED spread, the default spread (DEF) and the VIX, we find that hedge fund integration increases markedly during periods of financial stress. -2-

5 Contrasting high and low terciles of the conditioning variables, the average proportion of variability in hedge fund style returns explained by the global risk proxies increases by approximately 14% in stress periods, reflecting a 57% increase in integration. In other words, hedge fund integration is highest precisely when diversification from equity market systematic risk is most needed by investors. In a multivariate framework, we next explore fund-level determinants of integration. Focusing first on the fund manager, we find that while education and experience are significant predictors of integration, the influence of education is around four times higher than experience. Funds managed by better educated managers (proxied by SAT score) have lower integration while, perhaps surprisingly, funds managed by more experienced managers typically have higher integration. Turning to fund characteristics, smaller funds and funds with higher initial investment requirements typically have lower integration. Distressed funds, as indicated by low (negative) investor flows and fund returns and high return standard deviation, also typically have higher integration levels. Of the above factors, integration is most sensitive to fund size and distress indicators (returns and flow), which typically have predictive coefficients in our models twice the size of those for the other significant variables. As an extension of these tests, we seek to differentiate between luck and skill in managing integration. To make this distinction we examine persistence in integration and compare the determinants of integration to the determinants of the standard deviation of integration over the prior three years. The contrasts are striking. While funds managed by better educated managers and more experienced managers typically have lower integration volatility, funds managed by better educated managers typically have lower levels of integration in addition. Larger funds typically have more stable integration levels while younger funds typically have both lower integration and lower integration volatility. Distressed funds also typically have both higher and more volatile integration. Partitioning these models by TED, DEF, and VIX, we find the relation between either integration or integration volatility and the determinant variables is typically constant across financial market states. As exceptions, funds managed by better educated managers typically have notably lower integration and integration volatility during periods of stress in financial markets (top tercile of TED, DEF and VIX). Integration and integration volatility is similarly significantly more sensitive to fund distress factors during the same periods of market weakness. As previously discussed, low integration -3-

6 may not always be optimal for the fund or investors. It is noteworthy that we do not find any variables which are predictive of higher integration solely during strong financial markets. Investors appear to recognize and value hedge fund integration. Funds with higher integration realize significantly lower flows, and funds where integration increases also lose flows. Investors appear to recognize the incremental significance of integration during periods of stress in financial markets. Controlling for performance and other variables that have been shown to affect hedge fund flows, high integration funds and funds with increasing integration both realize incrementally greater outflows during periods of high TED, DEF and VIX. We also find a striking asymmetric investor reaction to levels and changes in integration. While investors punish funds with large relative increases in integration in the prior period by withdrawing investments, changes in inflows to funds with low intergration and decreasing integration are insignificant. The asymmetry is persistent across financial market states, suggesting that investors either do not recognize the benefit of increased systematic risk exposure during periods of strong market performance, use other investment vehicles to vary their exposure or rely on the fund manager to make such adjustments. The high sensitivity of investment flow to integration increases suggests investors are aware of and manage the integration of their portfolio, which makes the asymmetry in their response particularly puzzling. To resolve the puzzle, we analyze the relation between fund performance and integration. Regressing the alpha for each fund on its integration level and changes in integration in the previous year, we find that funds with lower integration, with decreasing integration, and lower integration volatility typically earn significantly higher alphas. Similar results are noted when we examine changes in alpha where funds with lower integration, with decreasing integration and lower integration volatility typically earn larger increases in alpha. Partitioning the performance by financial market states does not change our conclusions. Although the negative impact of integration on fund performance is reduced in strong financial market states, high integration remains negatively related to performance in both states. Overall, our results suggest that investing in high integration funds is a losing proposition regardless of market conditions, and that investors recognize this. Finally fund managers also appear to recognize the value of integration while setting fees. Funds with skill in managing integration charge significantly higher incentive fees, even after controlling for other factors that have been shown to affect fees in the prior literature. Similarly, -4-

7 increases in both management and incentive fees typically follow periods of decreases in integration and integration volatility. Although investors pay high fees to invest in low integration funds, we find that these higher costs are more than offset by the superior performance of the fund. Investment in low integration funds yields significantly greater benefits through its effect on hedge fund performance and diversification. Investing in funds with persistently low integration, suggesting skill as opposed to luck in managing integration, yields the greatest benefits. To provide a sense of scale, a joint decrease of one standard deviation in integration and integration volatility relates to a 1.9% increase in the annual incentive fee and a 1% increase in the annual alpha of the fund. Our paper contributes to the growing hedge fund literature that documents the relation between performance and flows for hedge funds. As in Titman and Tiu (2011), we find that funds with low levels of integration tend to attract more inflows, though the magnitude of our results is strikingly larger than those documented by Titman and Tiu. Our contribution lies in analyzing the fund level determinants of systematic risk exposure and the determinants of skill in managing systematic risk. We also analyze why funds choose persistently low levels of integration as opposed to changing integration depending on market conditions, why investors value these funds and the consequent effects on the fees charged by the funds. The rest of the paper is organized as follows. Section 2 describes the data, while section 3 discusses the construction of our measure of integration and contrasts our measure to those used in prior literature. Section 4 documents the time varying patterns in hedge fund integration that is the focus of our study and section 5 analyzes the determinants of fund integration. In sections 6-7, we document the sensitivity of investors to hedge fund integration, the relation between fund integration and fees set by hedge funds, and the relation between integration and performance respectively. Section 9 concludes. 2. Data We obtain hedge fund data from the Center for International Securities and Derivatives s (CISDM) and TASS Lipper. Although there is a substantial overlap between the two data providers, they do track different funds, and so we utilize the combined datasets after excluding duplicates. Our sample covers January 1990 to December 2010 for a total of 252 monthly observations and consists of -5-

8 nine single strategy indices or hedge fund investing styles: Convertible Arbitrage, Managed Futures, Event Driven, Equity Hedge, Equity Neutral, Fixed Income Arbitrage, Global Macro, Multi Strategy, and Emerging s. 3 We exclude fund of funds, multi-strategy funds, funds with non-us dollar returns, and funds which report only gross returns, resulting in a final sample of 3,405 funds. 4 Our samples include both the graveyard and live portions of each dataset, mitigating potential survivorship bias. 5 As hedge fund reporting to data providers is voluntary, the potential exists for several biases in our data. For example, funds do not typically start reporting performance until after the fund inception date and managers may elect to back-fill data once reporting commences. Aggarwal and Jorion (2010) report that the median back-fill period in the TASS dataset (the difference between the inception and reporting commencement dates) is 480 days and 37% of funds have a back-fill period in excess of two years. Some delay between inception and reporting is natural as the fund may be acquiring investment capital after inception. The concern is that managers may preferentially commence reporting for funds with positive performance and this performance is then back-filled. To control for this potential bias, we follow Aggarwal and Jorion (2010) and exclude funds with a back-fill period greater than 180 days. 6 Despite this adjustment for back-fill bias, the potential still exists that poorly performing funds could elect to not report at all. Similarly, a very good fund may quickly close to new investment and hence have no reason to advertise by reporting its performance. Fung and Hsieh (2000) conclude however, that these two effects cancel each other and do not create a self-selection bias. Agarwal, Fos, and Jiang (2010) report that the performance of reporting and non-reporting funds is similar. Drawing on these empirical findings, we do not control for self-selection bias. 3 See Appendix A for a description of each strategy. 4 A range of hedge fund data providers exist, each of which significantly overlap. We utilize two of the primary data providers available resulting in a sample size comparable to Titman and Tui (2011) who utilize a sample of 3,642 funds in their analysis. If funds of funds and multi-strategy funds are likewise excluded from the Titman and Tui sample, their sample then includes 2,641 funds. 5 Survivorship bias in hedge fund data was a concern prior to 1994 when data vendors generally discarded funds that ceased reporting due to, for example, liquidation. The graveyard sample includes data on funds that ceased reporting. Our results are robust to exclusion of the 1990 to 1993 period from our sample which is not covered in the graveyard sample. 6 We obtain similar results and reach the same conclusions if funds with back-fill periods greater than 180 days are included in the sample. -6-

9 Table I presents summary statistics for the hedge funds included in our sample, reported by hedge fund style. Nearly half the funds in our sample (1,376) are classified into the Equity Hedge strategy, illustrating market demand for diversification from equity market exposure. Between 132 and 482 funds are classified as belonging to each of the other eight investment strategy categories. On average, funds in our sample manage 200 million USD in assets, charge management fees of 1.3% and incentive fees of 16%. The life of a fund in our sample is, on average, 62 months or 5 years, which reflects the rapid growth in the industry in the later portion of our sample, rather than a high failure rate of funds. 3. Measuring Integration To measure hedge fund integration, we draw on Pukthuanthong and Roll (2009) who analyze levels of world integration across equity indexes for a broad sample of developed, emerging, and frontier countries. Specifically, they calculate the principal components of a global sample of country indexes and then regress daily country index returns on these global PCs, using the R 2 of that model to measure integration. For hedge funds, we measure integration with global equity markets in the following manner. We obtain country index data from Datastream. A total of 82 country equity indexes are available, from which we exclude country indexes with an incomplete time-series of monthly returns during our sample period of 1990 to This filter results in a sample of 38 country indexes, which we summarize in Appendix B. 7 For each calendar year, we calculate the covariance matrix of USD monthly returns to the 38 country equity indexes. From this matrix, we compute eigenvectors and sort them from the smallest to the largest eigenvalue. The PCs are then estimated from returns in the subsequent calendar year. This process is repeated in each calendar year, ultimately resulting in 21 calendar years of monthly frequency, out-of-sample PCs. Figure 1 plots the average cumulative percentage of variance explained within the estimation year by the principal components. The first principal component explains up to 65% of the variance, 7 All indexes are converted to a common currency, the US dollar. For countries with multiple indexes, we utilize the same index as in Pukthuanthong and Roll (2009). Our sample of country indexes for the PC analysis is larger than Pukthuanthong and Roll, as they were constrained to using 17 country indexes with data available starting in

10 while the next two PCs explain approximately 26%. In other words, there seems to be clear evidence supporting the existence of a dominant global risk factor. While the next two PCs are less dominant than the first PC, they are significant and explain the majority of remaining equity index return variation. As proxies for global factors, we follow Pukthuanthong and Roll (2009) and retain the first 10 PCs from this analysis, which on average, explain approximately 98% of the variation in the global equity index returns. In the remainder of the analysis, we refer to these 10 PCs as global risk proxies. To measure hedge fund integration with global equity markets, we regress the monthly return to each hedge fund or style index on the 10 global factors. The adjusted R 2 of this regression measures the proportion of hedge fund returns which are explained by global risk factors, while the coefficient on the first PC is comparable to the loading on a global market factor. As the PCs are mutually orthogonal, interpretation of the remaining coefficient estimates is not straightforward. This approach has several advantages over correlation or factor model analysis which has previously been utilized to measure hedge fund integration with equity markets. 8 First, since hedge funds usually hold illiquid assets (Aragon, 2007), their returns are not normally distributed. They typically exhibit high serial correlation, low skewness and high kurtosis. 9 Our measure of integration is not dependent on any assumptions of normality in the distribution of hedge fund returns, a key assumption in correlation estimates. Second, our estimate is free of any assumptions regarding factor selection, eliminating the potential for omitted variable bias, which is a significant consideration in factor models. 10 Our method directly measures commonalities in the returns of global equity markets without requiring any underlying knowledge of the source of these commonalities and variation in source significance across time. 8 For example, Boyson, Stahel and Stulz (2010) use the correlation in hedge fund index returns to measure contagion in hedge funds and Titman and Tui (2011) use factors models to estimate systematic risk in hedge funds. 9 See for example, Getmansky et al. (2004) and Brooks and Kat (2001). 10 As discussed in Titman and Tui (2011), the problem of identifying the factors to use in factor models is non-trivial due to the large number of known and unknown potential factors and the limited degrees of freedom in each regression. Estimating models with 24 monthly observations, factor models are limited to 7 factors. -8-

11 4. Hedge Fund Integration with Global s We start our empirical analysis by analyzing aggregate hedge fund integration. We partition this analysis by hedge fund style to capture the relative impact of investment strategies on systematic risk exposure. Table II reports the adjusted R 2 and the coefficient for the first PC, obtained by regressing the value-weighted index return to each hedge fund strategy on the first 10 principal components calculated from 38 country indexes. 11 To allow a comparison of integration measures, we also report: 1) the simple correlation between the hedge fund style return and the S&P 500 index return, 2) the correlation between each hedge fund style return and the equal-weighted global index return calculated from the 38 country indexes, and 3) the adjusted R 2 from the Fung and Hsieh (2004) seven factor model, where the hedge fund style return is regressed on the excess return to the S&P 500 index, the Wilshire small cap minus large cap return, change in the constant maturity yield of the ten-year Treasury, change in the spread between Moody s Baa yield and the ten-year Treasury and three Prime Trend Following Strategies for bonds, currencies and commodities. 12 Focusing on the variation of integration across hedge fund styles, we find that strategies which focus on credit quality and the volatility of underlying instruments (convertible arbitrage and managed futures) typically have relatively high integration with global equity markets, with approximately 60% of the variability in returns to these styles explained by the global risk proxies. Predictably, strategies that utilize both long and short equity positions and which aim to be market neutral (equity hedge, equity market neutral and global macro) or strategies which target low return volatility (fixed income arbitrage), typically have lower systematic exposure. For these hedge fund styles, on average only 40% of variability in returns is explained by the global risk proxies. We also find that sensitivity to global systematic risk follows the same general trend. The convertible arbitrage and managed futures styles tend to be the most sensitive to global systematic factors (coefficient on the first PC and in the full sample, respectively). Likewise, fixed income arbitrage style funds are the least sensitive to systematic risk (coefficient on the first PC in the full sample). 11 We obtain similar results and reach the same conclusions with equal-weighted indexes. 12 The Prime Trend Following Strategy data is obtained from David Hsieh s website, while the remaining variables are obtained from Datastream. -9-

12 We next examine the relative integration over different time frames. Relative integration across hedge fund styles appears reasonably time invariant. For example, the convertible arbitrage style is consistently most integrated with global equity markets across all three timeframe sub-samples, while the fixed income arbitrage style is consistently the least integrated. However, integration for each hedge fund style varies significantly over time. Integration tends to be higher in the decade than during the decade, consistent with globalization trends in equity markets. These time-series trends may transcend globalization effects and could also be reflective of market characteristics in each period. It is well understood that equity markets tend to be more highly correlated during periods of financial stress. 13 We note that this appears to also be the case with hedge funds. Examining the financial crisis period, integration across all nine hedge fund styles is notably higher than over the entire sample period (on average 45% higher across all the styles). We examine the determinants of time-series variation in integration in more detail below. Finally, contrasting the three integration measures, we find that the level of integration suggested by the Pukthuanthong and Roll (2009) approach yields integration estimates notably higher than suggested by correlation or factor model R 2 values. Focusing on the full sample, on average, integration estimates are 61% and 32% higher than suggested by correlation with the equal weighted index and the factor model R 2, respectively. This proportionality is generally consistent across the time-series partitions, suggesting that our model is structurally superior in detecting integration with global equity markets Determinants of Hedge Fund Integration In the previous section, we noted significant time-series variation in hedge fund integration with global equity market conditions. We now explore the determinants of integration both in aggregate across styles and at the individual fund level. 13 See for example, Longin and Solnik (2001) and Ang and Bekaert (2004). Similarly, Boyson, Stahel and Stulz (2010) find that hedge fund style indexes are more highly correlated during liquidity shocks. 14 It should be noted that Titman and Tui (2011) use a stepwise regression model drawing from over 30 potential risk factors which yields a R 2 value of 43% across their full sample, which is comparable to the average R 2 across all nine styles from the Pukthuanthong and Roll (2009) approach of 47%. The R 2 values are not fully comparable due to variation in timeframe and data providors. Titman and Tui (2011) also include funds of funds and multiprocess funds which we exclude. -10-

13 5.1. Aggregate Analysis We start our analysis of time-series variation in hedge fund integration at the aggregate level across hedge fund styles. The focus of this analysis is to evaluate the influence of broad market factors on integration. As proxies for market conditions we include the TED spread (difference between 3 month Libor and US T-Bill rates), the default spread (difference between AA and BB corporate bond index yields), the VIX index, and the return to the S&P 500 index. These commonly used measures capture tightness in capital availability, aggregate default likelihood perceptions across industries and equity market performance and volatility expectations, respectively. Using the R 2 measures from the regression of hedge fund index return on the 10 global factors (as described in section 3), we calculate the average R 2 to each tercile of the conditioning variables. Each tercile contains 84 monthly observations which contribute to seven annual observations that form each mean, with tercile 1 being the lowest value of each partition variable. The results are reported in Table III. Consistent with trends noted in equity markets, we find that a greater proportion of hedge fund returns are explained by the global risk proxies during times of financial stress. In other words, hedge funds become more integrated with global equity markets precisely when diversification effects are most needed by investors. Specifically, average R 2 values across the nine hedge fund styles increases by 13%, 12% and 16%, respectively, between the 1 st and 3 rd terciles of TED, DEF and VIX, respectively. This reflects an average increase in integration of 57% between the two terciles. Conversely, equity market performance (proxied by the S&P 500 index return) typically has a concave relation with hedge fund integration. We find that hedge funds tend to be more integrated with global equity markets during extremes in performance, both negative and positive. The exceptions are the equity hedge and equity market neutral strategies, which are more integrated during periods of poor market performance. Turning to variations across styles, surprisingly, the convertible arbitrage style tends to have low sensitivity to variations in market conditions. Despite having one of the largest first PC coefficient values in the full sample of Table II (1.646), suggesting high systematic risk exposure, the increase in integration for this style is consistently below the average across styles for the TED, DEF and VIX partitions. Across the sample partition variables, the increase in integration is consistently higher than -11-

14 average for the distressed securities and event driven styles while the fixed income arbitrage and global macro styles are consistently at or below the average Fund-Level Analysis We now focus our analysis on fund-level determinants of hedge fund global integration. To measure fund-level integration with global equity markets, for each calendar year, we regress the monthly return to fund i on the previously calculated global risk factors. This results in an annual timeseries of adjusted R 2 values which capture the level of integration with global equity markets. To explain time-varying integration at the fund level, we consider both manager and fund characteristics. As a proxy for either intelligence or education of the fund manager, we include the composite SAT score from the U.S. News and World Report and Princeton Review of 2003 of the undergraduate college that the fund manager attended. We also include the number of years the fund manager has worked as a proxy for management experience or career concern (Work Experience). Both variables are reported in TASS. Following Li et al. (2011), we exclude from our analysis manager age and tenure (which are also reported by TASS) as age is missing for over half of the dataset and tenure is highly correlated (0.95) with work experience. 15 The fund-level determinants we consider are: fund age in years, fund size measured by total net assets, annual return, the standard deviation of monthly returns over the prior year and fund flow calculated as: Flow i,t = (TNA i,t TNA i,t-1 ) (1+R i,t )/TNA i,t-1 ) (1) where TNA is total net assets to fund i at the end of year t and R is fund return. We also consider the natural logarithm of the required minimum initial investment, an indicator variable equal to one if the fund uses leverage in its investment strategy and the length of the lockup, redemption and subscription periods (in days). The redemption period is defined as the sum of the redemption and advance notice periods required for investors to redeem shares and the subscription period is defined as the delay between investing in the fund and the purchase being processed. Finally, we include fund fee attributes, 15 In further support for including work experience as opposed to tenure in our models, Chevalier and Ellison (1999) show that the number of years worked is a better proxy for manager experience. -12-

15 including the management and incentive fee and an indicator variable equal to one if the fund has a high water mark provision. We include average style return and flow to control for macro effects on integration. We also include indicator variables equal to 1 if the fund is active or open to new investment at the end of Since TASS reports static fee data in each data release, to build the time-series of fees for each fund, we access multiple end-of-year releases of the database. As multiple database releases were not available for the CISDM dataset, funds unique to that data provider are excluded from the fund-level analysis. Although TASS commenced reporting lockup and high water mark data in 2002, this data is available in CISDM from the start of the dataset. Therefore, we use the CISDM dataset to fill lockup and high water mark values missing in the TASS dataset for overlapping funds. Funds in the TASS dataset which closed prior to 2002 and which do not overlap with the CISDM dataset are excluded. We obtain the same results using the full TASS dataset, excluding the lockup and high water mark controls. We cluster by fund family to control for flow variation explained by family-level characteristics and include time and style fixed effects. Results of the pooled, time-series regressions are reported in Table IV. To differentiate between skill and luck in integration management, we analyze both integration and the standard deviation of integration over the prior three years, to allow examination of persistence in integration. To control for potential serial correlation in integration, due to overlapping observations used in its calculation, we include two lags of integration standard deviation in that model. Focusing first on manager characteristics, we find that funds managed by better educated managers (proxied by SAT score) tend to have lower integration and less volatile integration. Perhaps surprisingly, funds managed by less experienced managers tend to have lower integration levels while more experienced managers are more adept at minimizing integration volatility. These results suggest that better educated and more experienced managers have greater skill in managing integration. Turning to the fund characteristics, we find that older funds tend to have higher integration volatility though fund age is not a significant predictor of the level of integration. Larger funds tend to have higher integration but lower integration volatility. Periods of low returns, investment outflows and high return volatility tend to be followed by higher integration and integration volatility. These results suggest that distressed or struggling funds tend to be less focused on integration management. Funds -13-

16 with larger initial investment requirements tend to have lower integration and integration volatility. Similarly, funds with high water mark provisions and higher incentive fees also tend to have lower integration and integration volatility. Focusing briefly on the control variables, consistent with the results in Table III, periods of high investor flows and high returns at the style level tend precede periods of low integration and integration volatility. Funds that are active as of the end of 2010 tend to lower integration and integration volatility Symmetry in Determinant Variables As previously discussed, management of integration is of primary importance during periods of stress in financial markets, at which times diversification from equity markets is most desirable. To capture the influence of market conditions on the fund-level determinants of integration, we partition our sample into terciles, sorting by TED, DEF and VIX as previously described in section 4. We then calculate the same pooled, time-series regressions on time sub-samples when each conditioning variable is in its top and bottom tercile. We term the periods when each of the conditioning variables is in its top tercile as the weak market state. Correspondingly, the strong market state corresponds to periods when each of the conditioning variables is in its bottom tercile (i.e. when counterparty risk (TED), default risk in the corporate bond market (DEF) and market uncertainty (VIX) is low). By contrasting the coefficient size in the top and bottom tercile of each conditioning variable, we are able to make inferences regarding relative importance of each determinant variable when diversification effects are most desired. The results are reported in panel B of Table IV. Size and significance of the regression coefficients tend to be similar in the high and low state of the conditioning variables, but a few exceptions are noteworthy. Better educated managers are more adept at minimizing integration during periods of stress in financial markets proxied by TED and VIX (coefficients are on average 2.5 times more negative in the high TED and VIX states). The same trend is noted in relation to integration volatility when conditioning on TED but the coefficients are of similar size for DEF and VIX. We also find that funds with lower relative returns and fund flows tend to have higher integration in high TED and DEF states, suggesting that poorer performing funds realize incrementally greater increases in integration during periods of financial stress. Similar trends are noted for integration standard deviation -14-

17 during periods of high TED, DEF and VIX in relation to fund flows. It is also noteworthy that none of our predictive variables are associated with increases in integration during strong market conditions. As previously discussed, high integration may actually be desirable during period of strong market performance. Based on the factors we examine, there is little differentiation in the ability (or perhaps desire) to increase integration in up markets across fund types. 6. Investor Sensitivity to Fund Integration Next, we examine the extent to which investors recognize and value hedge fund integration. To proxy for investor fund preferences, we use fund flow which we relate to integration and control variables shown previously to influence hedge fund flows. 16 All variables are as previously described in Section 5. As in previous models, we cluster by fund family and include year and fund style fixed effects. Coefficient values for the control variables are reported in panel A but are suppressed in panels B and C in the interest of brevity (full results are available upon request). The results are reported in Table V. Panel A relates flow to the integration level, Panel B examines asymmetry in investor response to integration, and in Panel C we examine investor sensitivity to integration standard deviation. For robustness, in unreported results, we replicate the models in Panel A and B using the change in integration as the dependent variable and find the same results. Focusing first on model 1 in Panel A, consistent with Titman and Tui (2011), we find an inverse and significant relation between flow and lagged integration. Funds with higher integration realize significantly lower flows (t-statistic 2.67), suggesting investors seek low systematic risk exposure funds. Next we augment the models to include market condition proxies and examine their joint influence with integration. Investors should be particularly sensitive to integration when it matters most, i.e. during periods of uncertainty or financial market stress, when poor performance due to systematic exposure is more likely. Consistent with findings in the mutual fund literature (see Chalmers, Kaul and Phillips, 2011), we find significant outflows from hedge funds during periods of weak financial markets (negative and significant coefficient values on VIX, TED and DEF in models 2 to 4 of Panel A). Given the non-linear relation between integration and equity market performance, in 16 See for example, Ding et al. (2009) who examine the effect of share restrictions on investor flows in the hedge fund industry and Li et al. (2011) who examine the performance-flow relation in hedge funds. -15-

18 place of a continuous market return variable we include two indicator variables equal to one if the return to the S&P 500 index was in its bottom or top tercile in the previous period. Also consistent with the equity mutual fund literature, we find significant inflows into hedge funds following periods of strong equity market performance (t-statistic 2.57 in model 5 of Panel A) but the relation between flow and poor market performance is negative but insignificant (t-statistic 1.33). 17 Turning briefly to the control variables, we find evidence of return chasing both at the fund and style level. Without exception, the coefficients on fund and style return are positive and significant in all five models. Larger funds typically realize lower relative flow, consistent with gravity and exposure effects in the hedge fund industry. We also find a significant and positive coefficient on average style flow, consistent with differentiation at the style level influencing investor allocations. Investors tend to favor funds with higher high water marks, higher initial investment requirements, lower incentive fees and lower subscription periods. Funds in operation at year-end 2010 also typically receive higher flows, perhaps reflecting greater hedge fund participation in the later portion of our sample. The remaining control variables are typically insignificant. To capture incremental sensitivity to integration during periods of financial stress, we interact integration with each of the market condition proxies. In panel A, for VIX, TED and DEF we find negative and significant coefficients for each of the interaction variables (each t-statistic greater than 2.24) suggesting that high integration funds realize incrementally greater outflows during periods of financial stress. Conversely, the interaction variables based on the S&P 500 indicator variables are not significant. In other words, investors recognize the significance of integration to the risk profile of their hedge fund investments and react most markedly when the implications of higher integration potentially have the most negative effect. Turning to panel B of Table V, we explore potential asymmetry in the response of investors to integration. In the context of performance in equity mutual funds, Sirri and Tufano (1998) document that investors invest disproportionately in funds that performed well in the prior period. Following their approach, we form fractional ranks based on fund integration. The fractional rank represents a fund s percentile integration relative to funds in the same style in the same period, and ranges from 0 to 1. The coefficient estimates reported in Panel B are estimated using a piecewise linear regression framework 17 Wang and Zheng (2008) report a similar result in relation to aggregate hedge fund flow and past equity market returns. -16-

19 over five quintiles. For example, the bottom quintile is defined as min(rank t-1, 0.2) and the 4 th quintile is defined as min(0.2, RANK t-1 bottom quintile t-1 ) and so forth. Similar to investor sensitivity to prior performance, we find that investor reaction to integration is also asymmetric. Focusing first on model 1 in Panel B, the top quintile coefficient is significant and negative (t-statistic 4.69) suggesting that investors punish funds with large relative integration in the prior period by withdrawing investments. While funds in the bottom quartile realize investment inflows, the relation is only marginally statistically significant. Most striking, the standardized coefficient for the top quartile is relative to 0.02 for the bottom quartile suggesting an almost 25- fold higher investor sensitivity to high relative to low integration. As previously mentioned, in unreported results, we find nearly identical effects if the fractional ranks are instead formed from changes in integration. Similar results are found in model 2 where we aggregate funds in the 2 nd to 4 th quintile. We find that funds in the aggregated mid quartiles and the top quartile realize significant outflows while funds in the bottom quintile realize marginally significant inflows (t-statistic 1.70). We next partition the integration rank regressions by TED, DEF and VIX using the same process described in section We seek to understand if the same asymmetry in response exists in good and bad economic states. Consistent with the interaction coefficient results in panel A, we observe two-fold greater average outflows from funds with the largest integration during times of financial stress (proxied by TED and DEF). Without exception, the asymmetry in response to integration noted in the full sample (models 1 and 2) is also present in each of the sub-sample models. In both financial market states, investors punish hedge funds with higher integration via investment outflows. Conversely, there is no significant relation between hedge fund flows and low integration (bottom integration quintile). Our final analysis of investor preferences and integration examines luck versus skill in managing integration, differentiated by persistence in integration. We seek to understand if investors differentiate between luck and skill and the extent to which they place importance on skill across economic states. To measure persistence in integration management, we use the standard deviation of integration over the prior three years. We then relate flow to integration standard deviation, including 18 As noted for the full sample, results are similar when the 4 th to 2 nd quintiles are considered separately or combined into one partition. In the interest of brevity we report results with the 4 th to 2 nd quintiles combined for the market condition sorts. -17-

20 interaction effects. The results are reported in panel C of Table V. We find a negative and significant relation between fund flow and integration standard deviation (t-statistic 3.02). Investor sensitivity to integration volatility is typically greater than the integration level (-0.11 vs ) suggesting, to an extent, that investors pay greater attention to skill in managing integration than the current integration level, but both are clearly important. Investor differentiation between integration level and persistence in integration becomes significantly more marked considered jointly with market conditions. Comparing the sum of the base and interaction coefficients between the integration level and standard deviation models for VIX, TED and DEF, the coefficient sum is on average 47% larger in relation to integration standard deviation than for the integration level. Similar trends are noted in relation to the S&P 500 tercile indicator variable coefficients. As mentioned above, we also obtain similar results when we analyze changes in integration in place of integration levels. 7. Integration and Hedge Fund Performance Our results so far pose a puzzle. Investors seem to consistently prefer funds with low levels of integration and withdraw flows from funds with high or increasing levels of integration, regardless of market states. If hedge fund managers have a timing ability in choosing levels of integration, they may earn higher returns and fees by varying the level of integration between market states. Brown, Hwang, In, and Kim (2011) show that while hedge funds with a high systematic risk contribution underperform relative to funds with a low systematic risk contribution during crisis periods, on average they significantly outperform over the entire period Funds that increase systematic risk exposures in up markets may earn higher fund inflows and consequently earn significantly higher fees, which may either compensate them for the lower fees earned during down markets, or if they have the ability to vary integration levels, can reduce the level of integration preferentially in down markets. We therefore next examine the relation between fund performance and integration, focusing on differentiating between luck and skill in integration management across different market states. To measure performance sensitivity to integration, we relate the alpha for fund i in year t to integration for fund i in year t-1 plus control variables. Fund alphas are calculated using the Fund and Hsieh (2004) seven factor model. The factors include: the S&P 500 return minus the risk-free rate, Wilshire small cap minus large cap return, change in the constant maturity yield of the 10-year Treasury, change in the -18-