ARE MERGERS BENEFICIAL OR DETRIMENTAL? EVIDENCE FROM ADVERTISING AGENCIES1 Julian L. Simon, Manouchehr Mokhtari, and Daniel H. Simon INTRODUCTION Mergers have always been a burr under the economist's saddle. On the one hand, the economic literature has provided little or no evidence that mergers are profitable, and considerable evidence that they are not profitable. On the other hand, mergers continue to occur. This does not square with the vision of a rational market. This situation suggests that 1) the economic research is not sound, or 2) those who effect mergers are unaware of the research findings and of the underlying facts that mergers are unprofitable, or 3) those who effect mergers have motives for their actions other than the long-run profitability of the merged entity, or 4) there are other forces at work that are not observable in the firms' accounting data that make mergers desirable, such as the impending retirement or death of an official or a particularly talented person in one of the firms, whose loss would be expected to reduce profitability2. The first possibility would do the field of economics no credit, and the second and third possibilities call into question the extent of profit-seeking rationality in the economic system; the fourth possibility is consistent with the concept of a rational market and soundly-performing owners and agents. Research on mergers helps us discriminate among these possibilities. It also throws light on the extent of economies of scope. And the topic is of interest - for better or for worse - in various policy discussions. This article tackles the subject by examining a single industry - advertising agencies.3 The advantage of this research device is that there is a considerable body of neatly comparable data free of most of the complications that bedevil merger research. The disadvantage is that the results directly pertain to only a small segment of the economy, and may not be representative. THEORY The theoretical concepts given as justifications for mergers are: 1) Substantial economies of production scale or scope. But the physical capital used in making advertising consists mainly of word processors, telephones and other communications devices, and artists' tools; there is no basis here for production economies. 2) The creation of a position of substantial market power. There are so many advertising agencies, and the market shares of even the largest are so small, that this motive is not likely to be at work here. 3) The workings of the factor known in organization theory as "synergy". This motive is often spoken of when mergers are bruited. To the extent that this factor operates, one would expect to see mergers be successful in the advertising-agency industry. But we will not be able to test whether mergers produce cost savings due to synergy or other organizational changes. 4) Economies of the scarcest factor - the creative powers of the top executives. Again, if this factor is important, one would expect to see mergers be successful in our study. 5. Substantial economies of scale or scope deriving from non-physical factors of production, including mode of organization, ability to relate to large clients in a fashion that they like, financial factors such as the cost of borrowing, pooling of knowledge, and the like. These elements can only be observed in their totality, in terms of the overall success or failure of firms. We must therefore briefly discuss the evidence on the matter. Even for industries in which physical capital is a potential source of economies of scale, the economy-wide data on operating performance give no reason to expect substantial economies of scale as a general matter. It has long been well-established that larger firms are not on balance more profitable than small firms (Blair, 1965; Stigler, 1952). And bigger firms do not grow at a faster rate than do smaller; "no significant correlation between growth rates and size could be found", as Scherer summarized the literature in 1970 (p. 128); a later summary by Scherer and Ross (1990, p. 144) even suggested that the correlation between growth and size is negative for the U. S. - "The most careful recent studies for the United States reveal that growth rates tend to decline with greater size"; Montgomery's review (1994) concurs with the negative assessment. And this general appraisal (to the extent of no correlation) apparently is true in specific of advertising agencies (J. Simon, no date). These findings that there is no a priori reason to believe that large size confers advantage undercut the theoretical reasons why merged firms might be more profitable than their smaller predecessors. Considering the industry from the point of view of the survivorship principle, there also seems to be no warrant for the belief that being bigger than any particular size is better; the industry is well-populated with advertising agencies of all sizes. This squares with the rate of growth of agencies being much the same in all size categories, as mentioned above. 6. For firms whose customers' business is limited to only some regions of the country, there could be a market-broadening motive in merging, in connection with economies of scale. But most or all of the firms dealt with here are national in scope. There are also theoretical suggestions that mergers are not likely to be profitable. In addition to the general list found in the literature - costs of reorganization, communications difficulties, conflicts of corporate "cultures", and such - there is also the problem with advertising agencies that clients prefer not to have the agency provide services to other firms advertising the same product. Theory obviously does not add up to a firm conclusion either way. Hence the burden falls upon empirical work. RELEVANT PREVIOUS STUDIES Mergers have been evaluated in several ways. One method used has been the study of stock prices. The advantage of this approach is that it avoids all complexities that stem from accounting assessments of profits. The price of the stock may be considered to be the variable of ultimate interest, or it may be considered to be a proxy for profits and "true" performance. If the latter, the method has the disadvantage that it assumes that the market accurately evaluates the worth of a stock at each moment. This method also has difficulty in choosing an appropriate period of analysis because of the possibility of non- public flows of information about the merger. Another method of evaluating mergers is by studying accounting profits (see especially Mueller, 1980). But accounting methods are not uniform, especially when the sample includes firms from different industries, as they almost always do. Still another method is by measuring sales revenue or market share as a proxy for profits (see for example Mueller, 1985). But as with stock prices, the proxies may be faulty. Much of the research on mergers has considered together a wide variety of firms - very broadly-defined product markets, or even firms in many industries. And many large firms themselves span several industries. This heterogeneity causes many difficulties in interpreting the results, though it provides considerable generality. The case-study work of Ravenscraft and Scherer (1987) avoids many of the problems due to accounting methods and to heterogeneity, but convincing statistical conclusions are hard to come by in case-study work. Scherer and Ross (1990) summarized the literature on the effect of mergers, and hence it is not necessary to do so here. Their general conclusion is as follows: To sum up, statistical evidence supporting the hypothe- sis that profitability and efficiency increase follow- ing mergers is at best weak. Indeed, the weight of the evidence points in the opposite direction: efficiency is reduced on average following merger, especially when relatively small firms are absorbed into much larger and more bureaucratic enterprises lacking experience in the targets' specialized lines of business. To be sure, the statistical averages are just that; there is considerable variation from the central tendencies. Individual cases can be found to substantiate virtually all of the efficiency gain hypotheses identifiable in principle Yet the overall historical record is far from reassuring (p. 174). One finding of particular interest here is that pre-merger profitability was above average for acquired companies in the Ravenscraft and Scherer study (Scherer and Ross, 1990, p. 170), though the same need not hold for acquiring companies. Yet another method is to examine the rates of mergers and divestments. Lichtenberg (1990) found that between January, 1985 and November, 1989, "there was a substantial reduction in the degree of industrial diversification", which he interprets as a sign of negative results on balance of the earlier wave of mergers. Srinivasan and Wall (1992) studied all bank mergers between 1982 and 1986 in which the firms had total assets of $100 million or more. This study had the advantage of a homogeneous population. But the design did not include control firms, or make before-and-after comparisons, but only examined whether the expense ratio changed after the merger. "We have not found any evidence that mergers significantly lower expenses", they say. All the aforementioned evidence has been for U. S. firms. Generalizing from studies in seven countries, Mueller concludes: If a generalization is to be drawn, it would have to be that mergers have but modest effects, up or down, on the profitability of the merging firms in the three to five years following merger (1980, p. 306) The conclusion that mergers increase economic efficiency, as evidenced by profit and growth increases, was rejected in every one of our seven countries (p. 312). Johnson and Simon (1969) carried out a study on a sample of advertising agencies using the same method described below, but it was then possible to find only seven usable cases of mergers. Based on those thin data, the finding was that mergers tended to have poor results. DATA AND METHOD Mergers among advertising agencies have several useful properties for the study of mergers. (1) The firms in the sample are homogeneous. (2) The firms are not conglomerate; their product is strictly the making of advertisements. (3) In this industry (as we shall see) the crucial data pertaining to the mergers cover a time span of only three or four years, thereby not being subject to secular changes in conditions that could create noise; this is unlike many other industries, where the effects of a merger take place much more slowly. (4) Our sample of mergers includes observations covering more than three decades, thereby allowing us to check that there are no differences in results over this span of time (though our use of a nominal-dollar floor for inclusion of firms somewhat muddies comparison of mergers in earlier and later years). (5) No advertising agency has more than a small share of the market. This means that the scale effects of the merger can be assessed knowing that the merger will not result in even a short-run monopoly effect, which may not be the case in other industries. There is also an important peculiarity of mergers among advertising agencies, mentioned earlier, which may lessen the generality of the results. Many advertiser-clients do not wish to have their agency work on another brand of the same product. And a fairly small number of products account for important chunks of advertising-agency business: autos, cigarettes, beers, etc. This would seem to constitute a built-in force against profitable mergers. Yet mergers do occur anyway, and we shall therefore try to assess their effects. Our data are taken from the yearly lists of billings of major advertising agencies in Advertising Age, plus the news stories about mergers therein. The data are self-reports by the agencies rather than outside measurements or reports to the government whose validity is required by law. They are, however, scrutinized for reasonability by the staff of the newspaper and by a large number of readers, in light of general industry knowledge and of such other data as trade publications that report the expenditures of the various clients in major media. The data have a good reputation in the industry. We have identified mergers of 33 pairs of advertising agencies who merged between 1947 and 1985 and had the following characteristics: 1) combined billings exceeded $10 million (in nominal dollars, which implies that smaller mergers were included as the years went on (and we recommend that future research use a constant-dollar cutoff) ), 2) the smaller agency had billings at least 15 percent as large as those of the larger agency, 3) did not engage in other mergers within three years before or after the merger studied, and 4) data exist for at least the year of the merger and the four years following. For lesser numbers of cases we also have data for one or more years before the merger, and for more years following the merger. To assess the effect of a merger we compared the post-merger performance of the merged agencies against two control-group standards: (1) for each merging unit (called MERGEi, a pair of two agencies (called MATCH2i), whose merger-date t=0 billings totaled within +18 per cent of the merged units, were designated as matches for each merge agency. (2) For each merged unit, a single agency MATCH1i, whose pre-merger-date billings were +25 percent of the combined billings of the merged agencies. There were, then, a total of 66 matches for the 33 mergers. Where there was more than one candidate (or sets of candidates) to be matches, we took as matches those that came closest in billings to those of the combined pre-merged firms, and hence the selection criteria may be considered objective rather than subject to arbitrary choice. We use agencies' gross revenues as the dependent variable both because revenue is a meaningful indicator in itself and because data on profits are not available. Revenue may not be the ideal variable, but there probably is not, and cannot be, an ideal empirical dependent variable. The appropriate theoretical concept is an increment to the firm's economic worth during a given period (a year, in this case), which implies an increment to the firm's present value (assuming a closely-held firm, as most agencies are). But present value depends upon the discount factor of a particular firm, and upon future events that one could only hope to measure after they have occurred, which could be decades later, and there inevitably would be great variability in that stream of income due to many exogenous events. Even accounting profit (if data were available, which they are not) would not be an ideal measure, even though it may be more appropriate than revenue. We argue that revenue is a meaningful indicator of the vaguer concept of how well an agency is doing; a greater increase in revenues means a firm has done better, compared to another advertising agency. Revenue certainly is not the only such indicator, and a judgment about the effect of mergers can best be made by consulting research using a variety of indicators, including the indicator of revenues, none of which are ideal. Here are our reasons for believing that revenue is a valuable indicator: 1. Revenues probably are more closely correlated with earnings and accounting profit (another relevant measure) in advertising agencies than in most other lines of business. Until the 1960s, the prices charged by almost all advertising agencies were the same by industry self-regulation - 15 percent of the price paid to the media; price-cutting apparently was rare or non-existent among major advertising agencies. And though in more recent years agencies sometimes discount this rate, much business is transacted at the standard 15 percent. Second, agencies usually proceed by hiring people to service accounts they have already obtained, and personnel tend to be hired on a rule-of-thumb fixed basis - say, four people per million dollars of billings - in such manner as to provide an anticipated (and easily anticipatable) profit margin. Third, new accounts are usually gained on the basis of the agency's supposed success with its existing accounts. Fourth, occasional surveys in the trade publication Advertising Age indicated that the ratio of profit to billing shows no connection to agency size. If the data showed that mergers led to higher revenue than otherwise expected, one might wonder whether profit was reduced or made negative thereby, i. e. whether increased revenue was purchased at the expense of profit, in which case increased revenue would be a false signal of betterment (at least in terms of the profit indicator). But given that the opposite result is observed, this possible loose link in the analysis is not of concern here. 2. There is wide agreement among advertising people that the test of an agency's own success is the number of new accounts it gains (less the ones it loses). 3. Because of lagged effects due to continuing relationships between sellers and buyers (related to the accountants "good will" category), current revenue usually has a positive relationship to future profit. If one assumes equal efficiency in the future (even if not now) for two firms, the larger has the better basis for future earnings; that is why a firm can usually "sell" its list customers (as in the case of a magazine going defunct selling its ongoing subscription list). True, a firm can have large sales and be bankrupt, but even then a corporation tends to have positive value to its owners; compare the situation of any of the celebrated major bankruptcies in recent years to a flourishing local bar-and-grill. In brief, if two agencies A and B have billings of Y in year t, and billings in t+1 are Y and (Y+K) in t+1 respectively, and given an equal proportion of profit, there seems much reason to say that agency B has "done better" over the period in question. There isat least one shortcoming of revenues as an index that seems worth mentioning, however: Equal total billings for both merged and unmerged combinations could be compatible with greater profitability of the former if costs decline as a result of merging. This is not observable with our data. RESULTS Table 1 compares each MERGEi unit separately against the two comparison units ordinally - whether MERGEi did better, worse, or about the same compared to MATCH1i and to MATCH2i - for each year before and after the merger year. The result for each year is a rank, "3" being best and "1" being worst; that is, if MERGEi did better than MATCH1i but worse than MATCH2i in year t, it receives a "2" for that year, and a "3" if it did better than both the comparison groups.. Table 1 Panel A of Table 1 shows the proportions (and numbers) of these growth-rate ranks for MERGE firms in each year. (The total in any year depends upon the amount of data available.) The MERGE units do badly in years -1, 0, and +1; this may be seen in their having the smallest proportion of top ranks in years -1 and +1 in any year, and the most bottom ranks in year t=0; than does any unit in any other year and still more so in year t+1. These are the years when the poorest results for MERGE would be expected. Panel B of Table 1 shows the mean rank results for MERGE and MATCH units across the units in each group. If (say) MERGE units achieved average results in a given period, the mean rank would be 2.0. The (top) line for the MERGE units in Table 1 shows very poor results in year t+1, and also poor result in t=0 and t-1. To test these qualitative results for statistical signifi- cance, we use the criterion of the chi-square test.4 For the rank data for year t=1, the probability that a mean rank as low as actually observed would occur by chance is less than p=.0001. For the years t-1 and t, the hypothesis that results as poor as observed would occur by chance may be rejected at the 5 percent level. These results are not subject to the danger of data dredging, because there is a strong theoretical reason to expect the poor results in these years and not in others. As a check upon the research design, the ranks for MATCH2i and MATCH1i may be compared against each other. No systematic differences are found. This implies that the merged entities are being compared against two fair benchmarks, providing more infor- mation than if a single benchmark were chosen. Now we turn to cardinal rather than ordinal estimates of merger effects. Table 2 shows the mean growth rate in each year for the MERGE and the MATCH groups considered as three aggregates (SUMMERGEt, SUMMATCH1t, AND SUMMATCH2t respectively), as well as the difference between the former and the mean of the latter two (ALLMATCHt). That is, instead of the individual comparisons of each MERGE unit against its matched units, Table 2 shows the results of pooling within the MERGE, MATCH1, and MATCH2 groups in each year. This forgoes the specificity of the separate matches, but gains in statistical simplicity and in the information contained in the quantitative data. Table2 We can assess statistical significance for the growth-rate data with a bootstrap resampling test5. According to these simulations, it is exceedingly unlikely that the results for year t=0 and t+1 occurred by chance; the cardinal-data results (the prob-values in Table 2 Panel B) show even more unambiguously than do the ordinal data that merging does not produce more growth than achieved by agencies that are comparable in size but that do not merge. As to the time-span of the negative effect of merging (disregarding for now the possibility that pre-merger experience might have foreshadowed the post-merger result even if the firms had not merged): The MATCHes' advantage line in Table 2 Panel A shows that if there is an effect, it surely has ceased to exist by year 5, and in years 2 and 3 it is insignificantly small or non-existent. In light of all the various measures in Tables 1 and 2, we may conclude with considerable surety that the differences between merged and unmerged firms are limited to the three years centered on the merger year. Whenever there are many possible comparisons, one must inquire whether statistically significant results might be the result of hunting-and-picking among many other results that are not reported or that are examined and found not to be significant statistically. But in this case, the obvious ex ante expectation is that if there is an effect it would occur in the years in which the effect was indeed observed. And in none of the other years does the effect as large as in the years from t-1 to t+1, taking all the measures together. So the fact that an effect is not observed in the other years only strengthens the observed results by highlighting them. We also checked whether the merged firms in the smaller half of the sample (measured immediately after merger) performed differently than did the larger half of the merged firms. We found that there was no meaningful difference. (The larger half did much better in year t+1 while the smaller half did about as much better in year t+2, suggesting that the offsetting differences were both artifacts.) Were the Merging Units Already Doing Worse Than the Others? The evidence for year t-1 suggests that either the merging agencies already were not doing well or that the impending merg- ers had bad effects. The evidence for year t-2 in Table 2 has a prob-value of 8 percent, not statistically significant by conventional criteria but lower than the result for any MATCH group in any year (other than those years just before and after the merger). This suggests that the merging agencies may well have already been in trouble, a finding not consistent with the Ravenscraft and Scherer finding mentioned above (though these results include not just the "acquired" company). Concerning the possibility that the merging units were already doing worse before the mergers actually took place: Scherer and Ross note that the price of a merging firm's stock, used as a measure of managerial performance, tends to fall in the 18 months prior to a merger, and "This is often interpreted as evidence that the target firm's management has strayed from the path of profit maximization or has otherwise run into difficulties" (1990, p. 168). But this stock-market phenomenon may also be related to the diffusion of information that the firm is investigating a merger and may therefore not be fully stable. Indeed, stock price tends to move upwards in the weeks prior to the merger, testifying to the diffusion of merger information. It is reasonable, then, to assume that information of pending mergers leaks out. And information that an advertising agency is exploring mergers could be detrimental to client rela- tions, because a client may worry that the situation will change in various ways that are not desirable; uncertainty is to be avoided in business, as the premium for risk in securities mar- kets testifies. Therefore, it may well be that the merger itself, rather than poor operating performance, causes poorer results in the year prior to the merger. Even if the MERGE units did worse prior to the merger for reasons not connected with the merger, they did even worse after- wards - a mean relative performance of -10 percent in year t+1 compared to a relative performance of -7 percent in year t-1 (Table 2 Panel A). If it is true that mergers detract from performance in year t-1, then this negative effect should also be charged against the merger, in which case it is reasonable to compare post-merger performance to performance prior to year t-1. We are not able to determine the extent to which this is true, however, and there- fore we actually measure the performance in year t=0 relative to t-1 rather than relative to a prior year. The Cost to the Merging Firms of Merging If it is reasonable to believe that mergers cause diminished performance, we should try to quantify the damage. If we assume that by the end of t+1 the merging firms again have average prospects, we may simply sum the relative losses in t=0 and t+1; these equal about 16 percent of firm revenues and hence of firm profit. (Properly, one computes the loss directly from beginning to end points, but that calculation will not differ from this one.) Assuming the same subsequent rate of growth for these firms as for non-merging firms, the results imply a loss of 16 percent of firm value due to merging. If one believes that the poor results in t-1 also were due to the merger, there is an additional 7 percent loss in value, for a total of 23 percent. On the other hand, if one believes that the merging firms would have continued to do worse than their matches even if they had not merged, the loss is less than 16 percent or even non-existent. But we know of no evidence in other studies substantiating that poor performance in one or two years predicts poor performance in future years, within a given industry. DISCUSSION The data imply that mergers are not beneficial but rather are costly in the short run, according to the indicator of revenue - so much so that there is no reason for prospective merging firms not to know that the prospects are indeed poor, rather than just sufficiently murky to make the prospects unclear. This conclusion is subject to the qualification that some mergers may have been caused by factors such as death or retirement of important persons, which might have caused reverses for the firm even without merger. A fuller assessment of mergers would include effects in the longer future in a present-value framework. But the results within perhaps five years suggest that at any reasonable rate of discount there is little reason to expect that opposite-direction longer-run results would dominate the short-run negative results. The likely fate of mergers is particularly clearly written in the advertising business, both because of the account-conflict problem and because the results of previous mergers are relatively easy to assess. Certainly this result gives the lie to the frequently-heard argument that "Mergers must be profitable or they would not take place." In light of these results, the original question comes back with renewed vigor: If mergers are not profitable, and if igno- rance is not the explanation, why do mergers take place? In our view, the likeliest explanation is the agency problem, which is the private equivalent of public choice theory: decision-makers have interests other than, or in addition to, the interests of the firms. One of the two merging firms' heads becomes the head of a larger entity, with a salary increase. And there are benefits to promoters, law firms, and investment brokers. Financial news starting at the end of the 1980s has provided much corroboration that takeover mergers often fail, and that the cause of the mergers is something other than profitability. As one investment banker put it when describing the Campeau bank- ruptcy following a series of merger that went bad, "The deal was put together by a group of people - investment bankers, commer- cial bankers and investors - who were fundamentally fee-driven guys as opposed to long-term [stock] appreciation guys" (The Washington Post, January 16, 1990, p. C1). We do not completely discount lack of knowledge of the expected results as a reason why mergers take place despite their poor prospects. (Indeed, if the results were entirely obvious and conclusive, we would not need to do studies like this one). And the intellectual snare of the theory of economies of scale is enormously powerful; the success in the 20th century of communism - which relied on the theory of economies of scale - in winning so many adherents is proof positive of its seductiveness even when unwarranted. But results as strong as those seen here weaken these arguments. The most important possible weakness of this study is that the fact of a merger may itself be a signal that the merging agencies' futures are relatively not good, as suggested by this quote by the president of a merging agency: Because mergers tend to be looked upon as ways of overcoming weaknesses, and because ours is a business in which assumptions and suppositions are given the widest circulation, I want to emphasize that this merger represents the coming together of two strong and financially sound companies. [Weir, 1968, p. 8.] Future research might study the data on advertising-agency mergers using tools other than our matching technique. Regression analysis might be employed to good effect to investigate firm and environmental characteristics that might affect results of mergers. In our view, the value of the paper is not in whether it provides a conclusive answer to long-standing questions about the causes of mergers; it does not do so, nor is any single study likely to. Rather, it should be considered as one more piece of evidence, of a different sort than most, that the thoughtful observer can combine in a mosaic of professional and scientific judgment. REFERENCES Blair, John M., "Statement", Hearings Before the Subcommittee on Anti-Trust and Monopoly, in Economic Concentration, Part V, Concentration and Efficiency, 89th Congress, 1965, p. 1538 Johnson, Harold W., and Julian. L. Simon, "The Success of Mergers: The Case of Advertising Agencies," in The Bulletin of the Oxford Institute, Vol. 31, No. 2, 1969, pp. 139-144. Lichtenberg, Frank R., "Want More Productivity? Kill That Conglomerate", Wall Street Journal, January 16, 1990, p. A 16. Montgomery, Cynthia A., "Corporate Diversification", Journal of Economic Perspectives, Vol 8, Summer, 1994, 163-178, Mueller, Dennis C., "A Cross-National Comparison of the Results", pp. 299-314 in Mueller (ed.), The Determinants and Effects of Mergers (Cambridge: Oelgeschlager, Gunn and Hain, 1980) Mueller, Dennis C.,"Mergers and Market Share", Review of Economics and Statistics, 67, May, 1985, 259-267. Ravenscraft, David, and Frederic M. Scherer, Mergers, Sell- Offs, and Economic Efficiency (Washington: Brookings, 1987) Scherer, Frederic M., Industrial Market Structure and Economic Performance, (Boston: Houghton Mifflin, 1970). Scherer, Frederic M., and David Ross, Industrial Market Structure and Economic Performance, 3rd ed., (Boston: Houghton Mifflin, 1990). Simon, Julian. L., How to Start and Operate a Mail-Order Business (New York: McGraw Hill, 1965), second edition, 1976; third edition, 1980; fourth edition, 1986; fifth edition, 1993; abridged paperback edition, 1984. Simon, Julian. L., Issues in the Economics of Advertising (Urbana: University of Illinois Press, 1970). Simon, Julian. L., The Management of Advertising (Englewood Cliffs: Prentice-Hall, 1971). Simon, Julian. L., "Relationship Between Advertising Agency Size and Growth Rate", unpublished article, circa 1966. Srinivasan, Aruna, and Larry D. Wall, "Cost Savings Associated with Bank Mergers", Federal Reserve Bank of Atlanta Working Paper, February, 1992. Stigler, George J., "The Case Against Big Business", Fortune, May, 1952. Weir, Walter, quoted in Advertising Age, April 15, 1968. page 1 /article2 mergers/April 10, 1995 NOTES 1. We have copied some paragraphs verbatim from Johnson and Simon (1969), and adapted several others, without the use of quotation marks which would only tax the reader's attention. We appreciate unusually helpful comments from the referees and the editor. 2. This possibility was suggested by H. E. Frech III. We have made a cursory inspection of the text in Advertising Age that accompanied a few of the yearly presentations of data used here to look for such accounts; we found hardly any such cases, but this is hardly conclusive evidence against this possibility; a better test would require much deeper digging. In our judgment, however, one should not rely on this possibility as support for the concept of a rational market until there is positive evidence for the presence of this explanatory factor. 3. A few words about the industry itself may be welcome to some readers. This description emerges from J. Simon's experience working for an agency in the 1950s and reading the main trade paper until the 1970s, plus his continuing relationship with the mail-order business (J. Simon, 1965, 5th edition, 1993) , as well as his research on the industry (see J. Simon, 1970 for more data). But some of his information may be obsolete. There are a great many advertising agencies, so many that the yearly compilation of the one-hundred largest leaves out perhaps thousands of local agencies which may have one or more national accounts. An agency may have dozens of clients or only one; in the latter case the difference between an independent agency and a "house agency" (an advertiser's subsidiary) can be very thin. Agencies can specialize - in prescription drugs, for example - or they can handle a wide range of goods. Agencies often begin with the talent of a single copywriter, or the connection of a single client representative (the "account executive"), and that person's work may continue to be the center of the agency, even after the agency has grown to dozens of employees and partners. The industry is very volatile because the loss of a single large account - which happens frequently - can halve an agency's revenue and cause mass firings; copywriters and artists change jobs frequently for this and other reasons. From time to time during the past few decades there have been forecasts of consolidation among agencies, but there are no obvious signs of increased concentration. Law firms are perhaps the closest analogy, but they are seldom as dependent on a single client's billing, if only because in many cases law work is sporadic whereas advertising is ongoing. For that reason, and because the presence or absence of a single good idea often affects a large fraction of an advertising agency's business, employment at law firms surely is vastly more stable than at advertising agencies. 4. The calculation is carried out most clearly with the resampling method rather than the chi-square table (which is, after all, a subtle approximation to a multinominal formula). The test is performed as follows: For the given number of cases in each particular year, the computer equivalent of a three-sided dice is rolled n times, and a mean rank is computed for the n trials. More precisely, we a) Constitute a set of the numbers 1, 2, 3. b) For each MERGEi firm, select a number at random from the set, then for each MATCH2i entity select another number without replacement; and MATCH1i gets the remaining number. c) Compute the mean ranks for the three randomly-selected groups. d) Compare each actually observed mean rank with the simulated mean ranks. The observed mean rank is then compared against (say) 10,000 of these trials, and the proportion of times that the actual exceeds the random trial estimates the prob-value. The test is stated most precisely by the computer code used to carry it out; the few lines of code in the language RESAMPLING STATS are available upon request. 5. The test is as follows: a) Compute the growth-rate difference between i) MERGEi and MATCH2i, and ii) between MERGEi and MATCH1i. b) Compute the two means for the two sets of comparisons. c) Give each MERGEi a growth rate randomly selected from the set of 99. d) Select randomly a growth rate for MATCH2i. e) Compute the mean differences for the pairs from the data in (c) and (d). f) Select a growth rate for ALLMATCH (see definition below Table 2 Panel B), and compute the mean differences between it and the growth rates originally selected for MERGE firms, parallel to step e). (This avoids problems of lack of independence.) g) Compare the pair of actual mean differences in (b) with the distribution of pairs of trial mean differences. page 2 /article2 mergers/April 10, 1995