ome from larger fund families, and have higher returns and inflows over the prior 12 months than their peers. They are less likely to charge investors a sales commission(load), but their expense ratios are roughly comparable. Relative to the actual distribution of mutual funds across investment objectives, mentions in the publications we study focus disproportionately on general domestic equity funds Funds receiving positive mentions belong to families that spend a greater percentage of family assets on both print and non-print advertising and, since these families are larger, spend much more than average in absolute terms. Interestingly, the sample of funds recommended by Consumer Reports also come from families that spend an above-average amount on advertising. This suggests that advertising may be corre- lated with characteristics that are unobservable to the econometrician but that the financial media uses to rank funds. Consequently, our tests for advertising bias control for fund families'general level of advertis- ing. Examining the share of print advertising by publication reveals that funds receiving mentions from a oublication tend to come from families with higher than average levels of advertising in that publication II. Does Advertising Influence the Media? A. Motivation and Empirical Framework To motivate our tests for advertising bias, consider the mutual funds that appear on Money magazine's annual Money 100 list during our sample period. In an average year, 83.8 percent of families that spent more than $I million on advertising in Money over the prior 12 months are mentioned on the Money 100 list at least once. In contrast, only 7.2 percent of families that did not advertise in Money over the prior 12 months are mentioned. This difference partially reflects the fact that heavy advertisers tend to manage more mutual funds than non-advertisers. However, an individual fund from a heavy advertiser is more than twice as likely to be included on the Money 100 list as an individual fund from a non-advertiser ( 3.0 percent versus 1.3 percent). This difference is consistent with pro-advertiser bias, but obviously does not control for any of the mutual fund or mutual fund family characteristics that might lead publications to rank one mutual fund over another. In particular, one might worry that"high quality"'mutual funds are both more likely to advertise and more likely to receive positive media mentions [Milgrom and Roberts, 1986. To address this concern, we turn to multivariate tests for advertising bias
come from larger fund families, and have higher returns and inflows over the prior 12 months than their peers. They are less likely to charge investors a sales commission (load), but their expense ratios are roughly comparable. Relative to the actual distribution of mutual funds across investment objectives, mentions in the publications we study focus disproportionately on general domestic equity funds. Funds receiving positive mentions belong to families that spend a greater percentage of family assets on both print and non-print advertising and, since these families are larger, spend much more than average in absolute terms. Interestingly, the sample of funds recommended by Consumer Reports also come from families that spend an above-average amount on advertising. This suggests that advertising may be correlated with characteristics that are unobservable to the econometrician but that the financial media uses to rank funds. Consequently, our tests for advertising bias control for fund families’ general level of advertising. Examining the share of print advertising by publication reveals that funds receiving mentions from a publication tend to come from families with higher than average levels of advertising in that publication. III. Does Advertising Influence the Media? A. Motivation and Empirical Framework To motivate our tests for advertising bias, consider the mutual funds that appear on Money magazine’s annual Money 100 list during our sample period. In an average year, 83.8 percent of families that spent more than $1 million on advertising in Money over the prior 12 months are mentioned on the Money 100 list at least once. In contrast, only 7.2 percent of families that did not advertise in Money over the prior 12 months are mentioned. This difference partially reflects the fact that heavy advertisers tend to manage more mutual funds than non-advertisers. However, an individual fund from a heavy advertiser is more than twice as likely to be included on the Money 100 list as an individual fund from a non-advertiser (3.0 percent versus 1.3 percent). This difference is consistent with pro-advertiser bias, but obviously does not control for any of the mutual fund or mutual fund family characteristics that might lead publications to rank one mutual fund over another. In particular, one might worry that “high quality” mutual funds are both more likely to advertise and more likely to receive positive media mentions [Milgrom and Roberts, 1986]. To address this concern, we turn to multivariate tests for advertising bias. 5
Our general approach is to ask whether lagged publication-level advertising expenditures are correlated with the probability of receiving a media mention, controlling for all of the mutual fund and mutual fund family characteristics that publications might reasonably use to rank funds. Consider predicting positive mentions in a particular publication using the following specificatio Mention, t=a +?(Own-Publication Advertising t-1)+BZi, t-1+8k, t +Ei. here Mentionit equals one if fund i receives a positive mention in the publication in month t and zero otherwise, Own-Publication Advertising i t-1 measures lagged advertising expenditures in the publication by fund i's family, Zi t-1 contains numerous control variables, Sk. t is an investment objective-by-month fixed effect, and Eit is a fund-by-month disturbance term. To test whether advertising and content are related we estimate equation(1)and test whether i is statistically different from zero. The identifying assumption equired to give this test a causal interpretation is that advertising within a publication be uncorrelated with any unobserved fund characteristics that would cause its readers to want the publication to mention the advertisers fund. For products whose quality is partially or totally subjective, the fact that advertising is endogenous would lead us to seriously question this assumption. However, in the context of mutual funds, where er post product quality is objective and easil tified, we believe the assumption may be casona From a financial perspective, mutual fund investors should seek to maximize risk-adjusted returns on an after-expense basis. Therefore, within each investment objective, publications should seek to identify those funds with the highest expected future returns and the lowest expenses. Since Carhart [1997 finds low fund expenses to be a good predictor of future returns, we control for fund is lagged expense ratio. As other potential predictors of future returns, we include fund is log return over the prior twelve months, its lagged log return squared, and its Morningstar rating at the end of the prior calendar year. In addition to predictors of future returns, publications should also focus on the form of distribution S Mutual funds with multiple share classes can earn a different Morningstar rating for each share class. Therefore, to control for Morningstar rating we begin with five dummy variables that indicate whether one or more of fund i's share classes earned a Morningstar rating of one, two, three, four, or five stars. We then scale each dummy variable by the fraction of dollars under nanagement receiving each rating
Our general approach is to ask whether lagged publication-level advertising expenditures are correlated with the probability of receiving a media mention, controlling for all of the mutual fund and mutual fund family characteristics that publications might reasonably use to rank funds. Consider predicting positive mentions in a particular publication using the following specification: Mentioni,t = α + γ(Own-Publication Advertisingi,t−1 ) + βZi,t−1 + δk,t + εi,t, (1) where Mentioni,t equals one if fund i receives a positive mention in the publication in month t and zero otherwise, Own-Publication Advertisingi,t−1 measures lagged advertising expenditures in the publication by fund i’s family, Zi,t−1 contains numerous control variables, δk,t is an investment objective-by-month fixed effect, and εi,t is a fund-by-month disturbance term. To test whether advertising and content are related, we estimate equation (1) and test whether ˆγ is statistically different from zero. The identifying assumption required to give this test a causal interpretation is that advertising within a publication be uncorrelated with any unobserved fund characteristics that would cause its readers to want the publication to mention the advertiser’s fund. For products whose quality is partially or totally subjective, the fact that advertising is endogenous would lead us to seriously question this assumption. However, in the context of mutual funds, where ex post product quality is objective and easily quantified, we believe the assumption may be reasonable. From a financial perspective, mutual fund investors should seek to maximize risk-adjusted returns on an after-expense basis. Therefore, within each investment objective, publications should seek to identify those funds with the highest expected future returns and the lowest expenses. Since Carhart [1997] finds low fund expenses to be a good predictor of future returns, we control for fund i’s lagged expense ratio. As other potential predictors of future returns, we include fund i’s log return over the prior twelve months, its lagged log return squared, and its Morningstar rating at the end of the prior calendar year.8 In addition to predictors of future returns, publications should also focus on the form of distribution 8Mutual funds with multiple share classes can earn a different Morningstar rating for each share class. Therefore, to control for Morningstar rating we begin with five dummy variables that indicate whether one or more of fund i’s share classes earned a Morningstar rating of one, two, three, four, or five stars. We then scale each dummy variable by the fraction of dollars under management receiving each rating. 6
that most appeals to their readers. For example, to the extent that personal finance publications appeal to ivestors who prefer to purchase direct-marketed funds rather than employ a broker and pay a load, these publications should be more likely to recommend no-load funds. Since families of no-load funds should then be more likely to advertise in the personal finance publications, Z includes a dummy variable indicating whether fund i charges a load; it also includes the level of fund is 12b-1(marketing and distribution) fee As additional measures of potential investor interest in fund i, we include log dollars under management within both fund i and the fund family to which it belongs, log net inflows into fund i over the prior twelve months, and the number of mentions in each of the other publications in our sample over the prior twelve months. Since mutual fund families that advertise may differ systematically from those that do not--either because advertisers have systematically higher expected future returns or because investors are more likely to value reviews of funds from families they learned about through advertising--z also includes total print and non-print advertising expenditures by fund is family over the prior 12 months B. Testing for Advertising Bias In Table III, we estimate equation(1)separately for each type of media mention. For example, the dependent ariable in the column titled"SmartMoney Positive"equals one if we coded fund i as receiving a positive mention in SmartMoney in month t and zero otherwise. Estimation is via logit and includes a separate fixed effect for each investment objective-by-month combination. The number of observations in this column reflects the number of mutual funds each month with the same investment objectives as those receiving a positive mention in SmartMoney. The explanatory variable of interest is advertising expenditures by fund i's family within SmartMoney over the prior 12 months, which we refer to as " own-publication advertising expenditures. Standard errors are reported below the coefficients and cluster on mutual fund family Moulto 1990 Looking across the columns in Table Ill, the coefficents on own-publication advertising are positive and re not mentioned in the n, our tests for advertising bias effectively condition on the investment objectives that publications choose to focus on each issue and ask, within these investment objectives, whether advertising expenditures influence which funds are mentioned Since we observe advertising expenditures at the mutual fund family level and many families offer funds that span the set of Ivestment objectives, we have insufficient statistical to test whether the choice of investment objectives favors advertisers 7
that most appeals to their readers. For example, to the extent that personal finance publications appeal to investors who prefer to purchase direct-marketed funds rather than employ a broker and pay a load, these publications should be more likely to recommend no-load funds. Since families of no-load funds should then be more likely to advertise in the personal finance publications, Z includes a dummy variable indicating whether fund i charges a load; it also includes the level of fund i’s 12b-1 (marketing and distribution) fee. As additional measures of potential investor interest in fund i, we include log dollars under management within both fund i and the fund family to which it belongs, log net inflows into fund i over the prior twelve months, and the number of mentions in each of the other publications in our sample over the prior twelve months. Since mutual fund families that advertise may differ systematically from those that do not—either because advertisers have systematically higher expected future returns or because investors are more likely to value reviews of funds from families they learned about through advertising—Z also includes total print and non-print advertising expenditures by fund i’s family over the prior 12 months. B. Testing for Advertising Bias In Table III, we estimate equation (1) separately for each type of media mention. For example, the dependent variable in the column titled “SmartMoney Positive” equals one if we coded fund i as receiving a positive mention in SmartMoney in month t and zero otherwise. Estimation is via logit and includes a separate fixed effect for each investment objective-by-month combination. The number of observations in this column reflects the number of mutual funds each month with the same investment objectives as those receiving a positive mention in SmartMoney. 9 The explanatory variable of interest is advertising expenditures by fund i’s family within SmartMoney over the prior 12 months, which we refer to as “own-publication advertising” expenditures. Standard errors are reported below the coefficients and cluster on mutual fund family [Moulton 1990]. Looking across the columns in Table III, the coefficents on own-publication advertising are positive and 9Because funds with investment objectives that are not mentioned in the publication in month t are excluded from the estimation, our tests for advertising bias effectively condition on the investment objectives that publications choose to focus on each issue and ask, within these investment objectives, whether advertising expenditures influence which funds are mentioned. Since we observe advertising expenditures at the mutual fund family level and many families offer funds that span the set of investment objectives, we have insufficient statistical power to test whether the choice of investment objectives favors advertisers. 7
tatistically significant at the l-percent level for positive mentions in all three personal finance publications. 0 The coefficients are also economically significant. For Money, the marginal effect of Sl million in family advertising expenditure is to increase the probability a of positive mention for each of its funds by 0.2% compared with a predicted probability(at sample means) of 0.5%. For Kiplinger's, those probabilities are 0. 1% and 0.08%, respectively, and for SmartMoney they are 0. 2% and 0. 2%. Put differently, variation own-publication advertising has more explanatory power for positive mentions in each of the personal finance publications than variation in fund expenses, and about the same explanatory power as past returns As another way of gauging the economic significance of our findings, we use the coefficients reporte in Table Ill to predict the set of funds we would expect each publications to mention, first including the influence of own-publication advertising and then excluding it. For example, if SmartMoney mentioned 10 aggressive growth funds favorably in month t, we treat the 10 aggressive growth funds with the highest predicted values based on our estimates of equation (1)as predicted mentions that include the influence of own-publication advertising. We then repeat this exercise, setting the coefficient on own-pu advertising equal to zero For the Money 100 list, the overlap in the two sets of predicted mentions is 91.5% suggesting that 8-9 funds were replaced on the list by advertisers'funds that had otherwise just missed the cutoff. For positive mentions in Kiplinger's and SmartMoney, the overlap is 77.0% and 77.9%, respectively In contrast to the results for the personal finance publications, the coefficient on own-publication adver- tising is a precisely estimated zero for the Wall Street Journal and negative, but statistically indistinguishable from zero, for the New York Times. Since the three personal finance publications receive between a much larger share of their advertising revenues from mutual funds than the newspapers, our findings are consistent with advertising expenditures influencing fund rankings in those publications relatively more dependent on mutual fund advertising. Of course, for Wall Street Journal, the lack of a statistically significant correlation between advertising and mentions could also reflect that mentions in the"Fund Track" column are a mixture of positive and negative, and driven primarily by news with respect to negative mentions, advertising bias predicts that y will be negative, making publica- tions less likely to include advertisers'funds in negative mentions. Here, evidence of bias is weaker. For oThe correlations between advertising and content reported in Tables III and IV are robust to the inclusion of additional fund characteristics, such as fund age, manager turnover, and the standard deviation of fund returns over the prior 36 months
statistically significant at the 1-percent level for positive mentions in all three personal finance publications.10 The coefficients are also economically significant. For Money, the marginal effect of $1 million in family advertising expenditure is to increase the probability a of positive mention for each of its funds by 0.2% compared with a predicted probability (at sample means) of 0.5%. For Kiplinger’s, those probabilities are 0.1% and 0.08%, respectively, and for SmartMoney they are 0.2% and 0.2%. Put differently, variation in own-publication advertising has more explanatory power for positive mentions in each of the personal finance publications than variation in fund expenses, and about the same explanatory power as past returns. As another way of gauging the economic significance of our findings, we use the coefficients reported in Table III to predict the set of funds we would expect each publications to mention, first including the influence of own-publication advertising and then excluding it. For example, if SmartMoney mentioned 10 aggressive growth funds favorably in month t, we treat the 10 aggressive growth funds with the highest predicted values based on our estimates of equation (1) as predicted mentions that include the influence of own-publication advertising. We then repeat this exercise, setting the coefficient on own-publication advertising equal to zero. For the Money 100 list, the overlap in the two sets of predicted mentions is 91.5%, suggesting that 8-9 funds were replaced on the list by advertisers’ funds that had otherwise just missed the cutoff. For positive mentions in Kiplinger’s and SmartMoney, the overlap is 77.0% and 77.9%, respectively. In contrast to the results for the personal finance publications, the coefficient on own-publication advertising is a precisely estimated zero for the Wall Street Journal and negative, but statistically indistinguishable from zero, for the New York Times. Since the three personal finance publications receive between a much larger share of their advertising revenues from mutual funds than the newspapers, our findings are consistent with advertising expenditures influencing fund rankings in those publications relatively more dependent on mutual fund advertising. Of course, for Wall Street Journal, the lack of a statistically significant correlation between advertising and mentions could also reflect that mentions in the “Fund Track” column are a mixture of positive and negative, and driven primarily by news. With respect to negative mentions, advertising bias predicts that γ will be negative, making publications less likely to include advertisers’ funds in negative mentions. Here, evidence of bias is weaker. For 10The correlations between advertising and content reported in Tables III and IV are robust to the inclusion of additional fund characteristics, such as fund age, manager turnover, and the standard deviation of fund returns over the prior 36 months. 8
negative mentions in Kiplinger's, the coefficient on own-publication advertising is negative but statistically indistinguishable from zero: for negative mentions in SmartMoney, the coefficient is also negative and statis- tically insignificant, but quite close to zero. Nevertheless, for both publications, we can reject the hypothesis that the marginal effects of own-publication advertising are equal for positive and negative mentions(at the 5-percent level). This fact casts doubt on one alternative explanation for our findings. Namely, if past advertising in a publication directly increases reader demand for information on advertiser's funds, we would expect advertising to predict more positive mentions and more negative mentions. However in Table Ill,we find evidence that advertising expenditures increase positive mentions more than negative mentions Before exploring the robustness of our main results, several of the coefficients on the control variables deserve mention. First, counter to our expectations, few of the coefficients on the total print and non-print advertising expenditure variables are statistically significant. The fact that the coefficients on total print advertising expenditures are positive for both types of negative mentions, suggests that Kiplinger's and SmartMoney may be responding to subscriber demand for negative reviews on funds they 've seen advertised in general (rather than specifically in Kiplinger's or SmartMoney). Second, the probability of receiving both positive and negative mentions is increasing in the size of fund i and decreasing in the size of its family Third, the probability of receiving both positive and negative mentions is increasing in the level of the fund is expense ratio for every publication except Consumer Reports. Fourth, funds experiencing inflows good returns, and(though not reported)favorable Morningstar ratings over the prior 12 months are more likely to receive positive mentions, while outflows and low returns and ratings are associated with negative mentions. Fifth, with the exception of the New York Times, the probability of receiving a positive mention lower for load funds than for no-load funds 1 I As discussed above, load fund families are less likely to advertise in publications catering to do-it-yourself investors,and ss likely to mention their funds. Including a load dummy variables controls for this effect, but as an additional robustness check, we restrict our sample pad funds and re-estimate equation(1)for mentions in the thre personal finance publications. For positive mentions, the estimated coefficients on own-publication advertising are uniformly larger than those reported in Table Ill, and statistically significant at the l-percent level. For negative mentions, both coefficients remain negative but statistically indistinguishable from zero
negative mentions in Kiplinger’s, the coefficient on own-publication advertising is negative but statistically indistinguishable from zero; for negative mentions in SmartMoney, the coefficient is also negative and statistically insignificant, but quite close to zero. Nevertheless, for both publications, we can reject the hypothesis that the marginal effects of own-publication advertising are equal for positive and negative mentions (at the 5-percent level). This fact casts doubt on one alternative explanation for our findings. Namely, if past advertising in a publication directly increases reader demand for information on advertiser’s funds, we would expect advertising to predict more positive mentions and more negative mentions. However in Table III, we find evidence that advertising expenditures increase positive mentions more than negative mentions. Before exploring the robustness of our main results, several of the coefficients on the control variables deserve mention. First, counter to our expectations, few of the coefficients on the total print and non-print advertising expenditure variables are statistically significant. The fact that the coefficients on total print advertising expenditures are positive for both types of negative mentions, suggests that Kiplinger’s and SmartMoney may be responding to subscriber demand for negative reviews on funds they’ve seen advertised in general (rather than specifically in Kiplinger’s or SmartMoney). Second, the probability of receiving both positive and negative mentions is increasing in the size of fund i and decreasing in the size of its family. Third, the probability of receiving both positive and negative mentions is increasing in the level of the fund i’s expense ratio for every publication except Consumer Reports. Fourth, funds experiencing inflows, good returns, and (though not reported) favorable Morningstar ratings over the prior 12 months are more likely to receive positive mentions, while outflows and low returns and ratings are associated with negative mentions. Fifth, with the exception of the New York Times, the probability of receiving a positive mention is lower for load funds than for no-load funds.11 11As discussed above, load fund families are less likely to advertise in publications catering to do-it-yourself investors, and these publications are less likely to mention their funds. Including a load dummy variables controls for this effect, but as an additional robustness check, we restrict our sample to no-load funds and re-estimate equation (1) for mentions in the three personal finance publications. For positive mentions, the estimated coefficients on own-publication advertising are uniformly larger than those reported in Table III, and statistically significant at the 1-percent level. For negative mentions, both coefficients remain negative but statistically indistinguishable from zero. 9