Firm complexity and post-earnings announcement drift

Pseudo-conglomerates emulate real conglomerates by using information available about single-segment firms. First, Cohen and Lou (2012) calculate industry-level returns using only returns of single-segment firms operating in each industry and then compute a composite pseudo-conglomerate return assigning to each segment its industry return and taking the value-weighted average of those returns using as the weight the fraction of sales each segment generates.

We estimate information intermediation via analyst following and forecast error while we use institutional ownership and relative short interest to proxy for investor sophistication.

Unsophisticated investors, on the other hand, invest for savings/liquidity reasons and do not attempt to process firm-specific information. This leads to unsophisticated investors inadvertently holding relatively more shares of conglomerates, compared to sophisticated investors who avoid investing in more difficult to understand multi-segment companies. This, however, does not imply that unsophisticated investors will improve their investment performance by picking up excess returns through trading more complicated firms that are ignored by institutions. Such investment improvement does not occur because unsophisticated investors passively hold both winners and losers, the alphas of which cancel out.

Gleason and Lee (2003) document that analysts play a significant role in mitigating market inefficiency. We use the number of analysts (# Analysts) as a proxy for this effect and control for both (# Analysts) and its interaction with SUE (SUE* # Analysts) in all our main analyses to account for this. The exceptions are Tables 6 and 8, as using our usual complete set of controls would severely restrict the sample size.

We investigate whether alternative explanations of PEAD, which could be tied to other dimensions of firm complexity, can explain our results explicitly in Tables 7, 8, 9 and 10.

The same argument can apply to analysts who can choose the segment of the former conglomerate to follow according to their industry expertise after the conglomerate is disbanded.

In Panel B of Table 3, calculating SUE as the deviation from consensus analyst forecasts, we find results that are qualitatively and quantitatively consistent with our main findings.

In untabulated results, we find that 27% of firms in the sample are conglomerates. This number varies from 47% in the late 1970s to 17% in the late 1990s back to 25% in the 2000s.

The number of firms in quarterly Compustat files is larger than the number of firms reported in Compustat segment files, because single-segment firms and firms with relatively small segments do not have to report segment data. In our main analysis, we do not use firms covered by Compustat quarterly that are not on Compustat segment files, because we cannot exclude the possibility that such firms have small unreported segments. However, we confirm that our main results remain qualitatively intact if we assume that all firms that are on Compustat quarterly but not on Compustat segment files are single-segment firms.

In unreported tables, we also investigate the impact of organizational complexity on forecast dispersion, analyst quality proxied by analysts’ industry specialization as well as accounting disclosure quality using segment disclosure quality, following Franco et al. (2016). Those analyses yield results consistent with our hypotheses. Results are available upon request.

Institutional ownership (relative short interest) is the number of shares held by institutions (number of shares shorted) divided by number of shares outstanding.

The turnover regression uses the control variables from Chordia et al. (2007), the institutional ownership regression follows Gompers and Metrick (2001), determinants of forecast errors are from Thomas (2002), the analyst coverage regression is from Hong, Lim, and Stein (2000), and the short interest regression follows Barinov and Wu (2014).

For example, Duru and Reeb (2002) study the impact of international diversification on analyst accuracy and report that prior to year 2000 analyst forecast accuracy is lower for firms with internationally more diverse operations. In unreported results, we replicate and extend the analysis of Duru and Reeb (2002) and find no evidence that international diversification reduces analyst forecast accuracy in the post-2000 period. Results are available upon request.

In untabulated results, we find that simply controlling for the confounding effect of size shows that conglomerates have larger forecast errors (18% higher), lower analyst coverage (1 to 2 fewer), lower turnover (1.4% less), and lower short interest (0.5% less), compared to single-segment firms of comparable size.

Short interest can also reflect a directional bet, but this consideration works against our finding that short sellers avoid conglomerates, like institutions and analysts do. A long literature on the conglomerate discount, starting with the work of Lang and Stulz (1994) and Berger and Ofek (1995), finds that conglomeration is, on average, value-destroying and leads to conglomerates having worse operating performance and lower price multiples. Barinov (2019) further shows that conglomerates, on average, underperform by 3%–6% per annum on a risk-adjusted basis. Hence conglomerates should be attractive shorting targets everything else fixed, and the fact that we find the opposite result strongly indicates that organizational complexity influences sophisticated investors’ trading choices.

It is interesting that the momentum control is insignificant; in untabulated results, we verified that this insignificance is due to the momentum crash of 2009. If 2009 and later years are dropped from the sample, the momentum control becomes significant but still does not impact our main result (the slope on the SUE-Conglo interaction).

Complexity of 0.368 or HHI equal to 0.632 roughly corresponds to a two-segment firm with one segment taking slightly over 76% of sales, or to a three-segment firm with one segment taking 78% of sales and the other two taking 12% and 10% respectively.

In a related paper, Kang et al. (2017) investigate the impact of international diversification on PEAD. Using an international diversification measure that resembles the GeoMulti measure we use in Table 2, Kang et al. (2017) find that international diversification is associated with higher PEAD but document that this finding is confined to the period prior to SFAS 131 (that is, before 1998). In untabulated findings, we examine the time-series of the slope coefficients on the SUE*Conglo interaction term as well as on the interaction of SUE with alternative measures of firm complexity for a structural break around 1998 but do not find any evidence of this. In fact, the average slope on SUE*Conglo is at least 50% greater in the post-1998 period, though the difference lacks statistical significance largely due to sample size restrictions, since the comparison involves averages from two periods of 50–60 quarters each. This finding further suggests that the impact of organizational complexity on PEAD is distinct from the impact of geographic complexity on PEAD and that our results are not explained by the impact of geographic complexity on PEAD. This comports with our results in Table 2, as we find that GeoMulti does not affect the information environment in our sample period.

Single-segment firms are matched to a conglomerate of the same size and with the same two-digit SIC code of the largest segment.

One can also notice that Carhart and six-factor CARs are uniformly more positive than Daniel et al. (1997) CARs, to the extent that for size-industry matches six-factor CARs are positive in all SUE quintiles. This positive bias in the Carhart and six-factor CARs is likely introduced by the presence of the size effect, which is more efficiently removed by Daniel et al. (1997) adjustment.

Briefly, conglomerates are high-uncertainty firms, because, relative to similar single-segment firms, they are covered by fewer analysts, are ignored by institutions and other informed investors (see Table 2), and the high uncertainty coupled with short-sale constraints creates overpricing. As Miller (1977) suggests, short-sale constraints limit the ability of pessimists to impact the prices, and the price becomes equal to the average valuation of optimists, which increases in disagreement.

Size-Loss-Amihud matching matches single-segment firms to a respective conglomerate with the same value of the Loss dummy, with similar size (picking a single-segment firm that is the closest to the respective conglomerate in terms of market capitalization), and with a similar Amihud (2002) price impact measure (between 70 and 130% of the Amihud measure of the single-segment firm).

This version of PEAD is close to a short-term version of earnings momentum.

In untabulated results, we perform subsample analysis of returns to PEAD strategies and find no change in PEAD of conglomerates from 1980s to later years and an economically significant decline of equal-weighted PEAD of single-segment firms.

Another reason why the crash is stronger in value-weighted returns is that the market beta of the winners-minus-losers PEAD strategy for conglomerates is -0.25 in value-weighted returns and -0.06 in equal-weighted returns, helping the strategy in the falling market of 2008 but hurting it in the growing market of 2009. The PEAD strategy for single-segment firms has a slightly positive market beta.

As in the work of DellaVigna and Pollet (2009), firms outside of the top and bottom SUE deciles are excluded from this analysis; the analysis is effectively the analysis of the 10–1 SUE hedge decile return spread in returns.

This evidence is consistent with what is depicted in Fig. 1, which graphs CARs of extreme SUE quintile portfolios in event time. Figure 1 finds that there is minor difference between CARs of conglomerates and matching single-segment firms in the first few days after the announcement, and the gap between CARs starts to emerge around day five.

Conglomerates are on average larger, less volatile, and more transparent, and as such they are expected to have lower limits to arbitrage. Further cementing this idea, we find, in untabulated results, that, according to several measures of liquidity, including the Gibbs measure (Hasbrouck, 2009), the Roll (1984) measure, the effective spread estimate of Corwin and Schultz (2012), the Amihud (2002) measure, and the frequency of no-trade days from Lesmond et al. (1999), conglomerates on average are significantly more liquid than single-segment firms. Results are available upon request.

We argue that changes to the unobserved characteristic are associated with organizational structure, i.e., when the unobserved characteristic exceeds a certain threshold, the firm becomes a conglomerate. Conglomeration is not the cause of the change in this unobserved characteristic but rather the change in the unobserved characteristic itself leads to conglomeration. There could be a different omitted variable, separate from the one we consider, such that it can increase in response to conglomeration and then subside. If such an alternative omitted variable is also associated with higher PEAD, then PEAD would be stronger for new conglomerates. We argue in this paper that this potential alternative omitted variable is organizational complexity. Nevertheless, we acknowledge that there could be more alternative omitted variables that could behave similar to organizational complexity but are fundamentally different. While acknowledging that such alternative omitted variables may offer different explanations of the association between organizational complexity and PEAD, we suggest that it is almost impossible to control for all such alternative scenarios. In conclusion, we do not claim to solve all omitted variables problems.

Since the number of new conglomerates is low, in Panel A of Table 6 we do not control for # Analysts. Requiring that new conglomerates have non-missing analyst coverage data leaves us, in some years that have little M&A/conglomeration activity, with new conglomerates numbering in low double-digits and even in single digits.

The estimates of PEAD would be roughly twice in magnitude for both single-segment firms and existing conglomerates if we instead use the difference between the 97.5^th and 2.5^th percentiles of SUE.

SDC data includes both public and private firms. We include acquisitions of both public and private targets as potential ways of adding a new segment through mergers and acquisitions.

Rajan, Servaes, and Zingales (2000) measure the diversity of investment opportunities among the segments of a conglomerate as they study how this diversity affects internal capital allocations.

Momentum is not interacted with SUE. As in Panel A, we have to exclude # Analysts from the set of controls, since requiring nonmissing # Analysts would have left us with too few observations for the required analyses.

The average for log(1 + HTSD) is 0.783 among conglomerates. The 90^th percentile value of log(1 + HTSD) is 2.448, while the 10^th percentile value for log(1 + HTSD) is 0.0046.

In untabulated results, we find that the larger drift experienced by organizationally more complicated firms is not confined to the first month of the quarter. Results are available upon request.

Strictly speaking, the correct way to estimate industry momentum would be to compute industry returns using all firms in the industry, including conglomerates. We tried that and found a slight change in the slope of "PCRet" for single-segment firms defined this way, which suggests that the average return to all single-segment firms in an industry is a good enough proxy for the true industry return.

In Table 8, we do not control for # Analysts, as requiring nonmissing variables of analyst coverage would significantly reduce the sample in some years in several columns of Table 8.

Earnings Persistence (EP) is the firm-specific time-varying autocorrelation between two adjacent quarterly seasonally differenced earnings (SDE), where the autocorrelation is estimated in a two-step procedure using 14 persistence-related firm characteristics each quarter, following Chen (2013).

We got the data from Feng Li’s website, for which we are grateful.

Future research may attempt to decompose FOG into innate business-complexity and managerial obfuscation components, following Bushee et al. (2017), and analyze the impact of these components on PEAD separately.

Unlike Zhang (2008), however, our interaction term is statistically insignificant. We attribute this difference mainly to methodology. When we use the panel regressions of Zhang (2008), instead of Fama–MacBeth-style (1973) regressions, the interaction term becomes significant.

Results in column (7) suggest that, for an average single-segment firm, the hedge return to buying the highest SUE decile and selling the lowest SUE decile is 2.72%, while for a similar conglomerate the hedge return for the same trading strategy would be 3.95%. The difference is 1.23% for the three months after the earnings announcement and is tradable.