Journal Papers

Agency and Institutional Investment

By Feifei Li Michael J. Brennan Xiaolong Cheng

JANUARY 2012 Read Time: 30 min

Executive Summary
In 1952, domestic households owned approximately 90% of US equities; by 2007 that share had dropped to 35%. The expanded proportion of stocks in the hands of institutional investors has potentially important implications for securities pricing dynamics and asset allocation patterns that conventional investment theory is unequipped to recognize. In particular, the capital asset pricing model (CAPM) cannot capture agency effects that arise from the delegation of portfolio management: direct investors are typically concerned only with their portfolios’ absolute risk profile and return characteristics, but institutional investors, like corporate managers, have other concerns, such as the career risk that attends underperformance relative to client-specified benchmarks. In this paper the authors put into operation an extended version of Brennan’s 1993 agency model of institutional investors whose performance is evaluated in comparison with a capital market index. The model enables them to estimate the magnitude of agency effects on equilibrium rates of return and to analyze institutional investors’ equilibrium portfolios.

Unlike the CAPM, Brennan’s agency model distinguishes between institutional and individual investors. The model assumes that competitive, risk-averse agents—portfolio managers employed by banks, insurance companies, investment companies, independent investment advisors, and other institutions—are rewarded in proportion to the difference between the return on the managed portfolio and the benchmark return, and that they have mean-variance preferences over this reward. Accordingly, the authors’ key postulate is that for institutional investors the benchmark portfolio plays the role of the risk-free rate in the CAPM. Incorporating a coefficient of risk aversion, the model allows the researchers to differentiate between the market portfolio, the aggregate benchmark portfolio (comprising all institutionally managed portfolios), and a global minimum-variance portfolio. The assumption of market clearing further enables them to express the vector of equilibrium expected returns and, therefore, to specify the agent’s equilibrium portfolio for any ratio of aggregate institutional investments to the total market.

In estimating equilibrium returns, the authors use the value-weighted index of all stocks in the CRSP universe between December 1925 and December 2009 to represent the market portfolio, and the constituents and returns of the S&P 500 Index to proxy the aggregate benchmark. Analysis of the empirical data reveals that, although (as expected) the returns of the market portfolio return and the S&P 500 Index are closely correlated, the market portfolio is not a perfect substitute as a benchmark for managers whose performance is evaluated relative to the S&P 500 Index: the volatility of the index residual is approximately 80 basis points (bps) a month, and the tracking error ranges as high as 4–5% a month. (The S&P residual is the value e that is unexplained in the regression of the index return on the market return.) In examining the composition of equilibrium portfolios, the authors use quarterly data on institutional holdings, both in aggregate and by manager type, obtained from the CDA Spectrum database for the period from March 1980 to September 2008.

The model predicts that, when the benchmark is highly covariant with the market portfolio, the benchmark effect as measured by the premium per unit of benchmark risk is likely to be small. The authors’ empirical research bears out this forecast. Using the average excess return on the CRSP value-weighted index for the period 1931–2010 yields an estimate of the benchmark risk premium of about 35 bps a year or 3 bps a month. This value is confirmed by directly estimating the benchmark risk premium, an approach which does not have enough power to reject the null hypothesis of no agency effects. The authors conclude that the implied agency effect on expected returns is small (and probably undetectable with standard methods). This is a case, rare in published papers, where the finding of little or no effect when one might have been suspected adds to our understanding of the capital markets.

In order to generate cross-sectional estimates of the agency effect, the authors construct sets of annually rebalanced equal-weighted (EW) and value-weighted (VW) portfolios by market beta quintile, benchmark beta quintile, and size for the period 1931–2009. In most cases, the t-ratios for the two betas prove highly significant. The benchmark beta is highly correlated with the measure of the portfolios’ average firm size: the correlations are 0.83 and 0.90 for the EW and VW portfolios, respectively. This indicates that the returns on large companies, many of which will be contained in the S&P 500 Index, tend to be more responsive to the idiosyncratic component of the index return.

The results of the cross-sectional analysis are generally consistent with the earlier estimates of the agency effect. For the EW portfolios, the agency effect over the entire 1931–2009 period is 3 bps excluding the size parameter, and zero when size is taken into account. For subperiods, excluding size, it ranges from a high of 11 bps from 1931 to 1949 to a low of −2 bps from 1969 to 1987 (the only negative value); including size, the agency effect is zero for the entire period and runs from a low of zero to a high of 8 bps for the subperiods examined. In all cases, however, the standard error is too large to reject the null hypothesis of no benchmark effect. For the VW portfolios, the full-period estimate is −1 bp without the size impact and −2 bps including size in the regression. For subperiods, the estimated agency effect ranges from −4 bps to 5 bps without size, but is never significantly different from zero. Including size, the value of −25 bps registered in the 1969–1987 interval is exceptional: stocks with high benchmark residual betas earned positive abnormal returns. Citing a 1993 study conducted by Chan and Lakonishok, the authors suggest this outperformance might be associated with the growth of index investing in the 1970s and 1980s. For all other subperiods, when size is included in the regression, the agency effect ranges between −5 bps and 6 bps.

Pursuing the possible impact of size, the authors estimate agency effects for small-, medium-, and large-cap portfolios. The results of this analysis do not support the agency model. The large-firm size group yields a significant estimate of 14 bps a month for the period 1969–2009, and the small-firm size group generates an estimate of −30 bps for the same period. When the authors estimate loadings for the Fama–French market, HML, and SMB factors, however, the coefficient of the benchmark residual beta for both the large and the small VW portfolios prove insignificant. This outcome suggests that testing by size alone is inadequate.

With regard to asset allocation, the agency model predicts that institutional portfolios will consist of long positions in the market and benchmark portfolios and a short position in a minimum-variance portfolio expediently constructed on the basis of security-specific Fama–French factor loadings and residual variances. The authors find robust evidence for this pattern of asset allocation across all classes of institutional manager. On average, they report a linear combination of these three portfolios explains 93% of the variance in institutional portfolio weights, with each of the coefficients highly significant and of the expected sign. They also determine that, relative to the market portfolio, institutional holdings are higher for stocks that are included in the benchmark portfolio. In addition, institutional portfolios are tilted toward high-beta stocks with respect to the component of the benchmark return that is orthogonal to the overall market return, defined as the S&P 500 residual beta. Moreover, the weights of institutional holdings decrease in proportion to the stocks’ share in the minimum-variance portfolio. Although the authors do not propose behavioral explanations, their empirical results clearly reflect institutional investors’ decided preferences for stocks with high market betas and high benchmark residual betas.

Summarized by Philip Lawton, CFA

Featured Tags

How to Invest

Select filters Strategies, Asset Classes, Regions


Asset Classes


Feifei Li

Director, Head of Investment Management