The Diversification Discount Puzzle: Evidence for a Transaction-Cost Resolution

Abstract

The literature on the corporate diversification discount and the relative efficiency of internal versus external capital markets provides mixed results. We argue that transaction-cost economics is useful in understanding this puzzle. According to transaction-cost economics, diversified firms should outperform single segment firms in industries with higher external transaction costs (e.g., emergent industries) and under-perform in industries with low external transaction costs and high agency and other internal costs (e.g., some mature industries). This paper provides evidence supporting these contentions.

1. Introduction

Are diversified firms valued more or less than their stand-alone counterparts? While some studies show a significant diversification discount, others contend that there is no such discount and that prior studies suffer from measurement problems, with the discount disappearing when errors are corrected. This study proposes an alternative explanation to the puzzle. We argue that the finding of a diversification discount depends on segmenting firms based on the balance between transaction costs in internal versus external capital markets. Capital for investments can be allocated internally by diversified firms or can be allocated to undiversified “pure play” firms by external capital markets. We contend that diversification can be value enhancing if external capital markets face large transaction costs relative to internal costs of allocating capital and vice versa.

Empirically, we report that in emergent industries, where companies face relatively higher external market transaction costs, diversified firms perform better compared to single segment firms. Further, underperforming units of diversified firms are more likely to be shut down than underperforming single segment firms in these industries. In contrast, in mature industries, where companies face relatively high internal and low external transaction costs, diversified firms perform worse than single segment firms. The findings are consistent with the transaction-cost theory of firm boundaries.

Our findings imply that future studies of the diversification discount should segment firms based on the balance between internal and external transaction costs. Otherwise, firms with a positive diversification effect will offset firms with a negative effect.

2. Diversification, transaction costs and firm boundaries

2.1. The diversification value controversy

Many studies examine the effect of diversification on firm value, finding a diversification discount (e.g., Lang and Stulz, 1994; Berger and Ofek, 1995; and Burch and Nanda, 2003). Two strands of research try to explain this discount. One strand tries to identify the costs and benefits of internal versus external markets. Some authors note the managerial incentive problems in the efficient allocation of capital. For example, Holmstrom and Costa (1986) and Jensen (1993) explain that due to information costs and imperfect managerial incentive alignment, optimal internal control systems in firms are imperfect and can lead to varying degrees of internal capital misallocation. Ozbas (2005) notes that, in diversified firms, managers can have incentives to exaggerate the payoffs of their projects to obtain funding, reducing allocational efficiency. Other studies argue that agency and other costs of internal capital allocation lead to resource misallocation and subsidization of poor investments in internal markets, resulting in a diversification discount (e.g., Lang and Stulz, 1994; Rajan, Servaes and Zingales, 2000).

Other studies identify advantages related to information and control rights of internal markets over external markets. For example, segment managers are less likely to be able to hide embarrassing facts from their supervisors as easily as they can from outside shareholders, and internal capital markets can better keep sensitive data away from competitors. Matsusaka and Nanda (2002) develop a model that shows that using internal markets to allocate capital avoids the deadweight costs of using external capital markets, but also raises the costs of managerial overinvestment. This tradeoff determines whether there is a diversification discount or premium. These authors further show that differences in control rights between internal and external providers of capital can also lead to diversification discounts or premia. This is because managers of a diversified firm can terminate nonperforming projects more efficiently than managers of single segment firms. Gertner, Scharfstein and Stein (1994) make a similar argument in favor of internal markets. Diversified firms can also be more valuable because their segments can share valuable nontradable resources and because of economies of scope and market power (Teece, 1982; Tirole, 1995; Williamson, 1998).

Based on reported standard deviations of the diversification discount or premium, prior empirical work suggests that at least a third of diversified firms trade at a premium compared to their stand-alone counterparts (e.g., Lang and Stulz, 1994; Berger and Ofek, 1995; Rajan, Servaes and Zingales, 2000). Indeed, this literature concludes that diversification can create value and that internal capital markets can be efficient (e.g., Schipper and Thompson, 1983; Matsusaka, 1993; Servaes, 1996; Hubbard and Palia, 1999; Khanna and Tice, 2001; Maksimovic and Phillips, 2002; Villalonga, 2004).

The second strand of research asks whether it is diversification that causes the discount, or poor performance that causes firms to diversify. For example, Campa and Kedia (2002) and Villalonga (2004) suggest that the diversification discount disappears when corrections for selection bias are applied. On the other hand, Singh, Mathur and Gleason (2004) find that investors do not avoid diversified firms, suggesting that diversification does not hurt firm value.

This paper belongs to the first strand of research. We use transaction-cost theory to resolve the controversy regarding the positive or negative effect of diversification.

2.2. Transaction costs and firm boundaries

Transaction-cost economics focuses on the notion that when the transaction costs of market exchange are high, it can be less costly to coordinate production through a formal organization than through a market, i.e., it can be cheaper to use hierarchy to reduce contracting problems and costs (Williamson, 1998). The transaction costs of market exchange arise chiefly from efforts to reduce the uncertainties in contractual relationships (Hart, 1995). Such transaction costs depend on factors including uncertainty, frequency and asset specificity. These factors contribute to the costs of market exchange, such as search costs, costs of developing and delineating choices and options, costs of designing, negotiating and enforcing market-exchange contracts, and costs of activities that facilitate exchange.

According to transaction-cost economics (Coase, 1988; Williamson, 2000), firms should internalize operations when external markets are relatively inefficient, i.e., when a firm faces higher transaction costs in external compared to internal markets. In other words, a firm is likely to find it optimal to internalize operations in businesses that face external market transaction costs that are higher than agency and other internal transaction costs. Thus, according to transaction-cost economics, the balance between internal and external transaction costs determines the boundaries of a firm (Hart, 1995; Holmstrom and Roberts, 1998).

Transaction costs can be high for two reasons: difficulty in designing the optimal contract because of information asymmetry and difficulty in implementing the contracts because of a lack of control rights. Therefore, internalization of independent organizations is likely to be particularly beneficial in industries where there is a severe problem of information asymmetry and where the exercise of control rights in resource shifting is especially important. Emergent high-tech industries fit these descriptions.

Numerous studies report that firms with much research and development (R&D) tend to have high information asymmetry (e.g., Aboody and Lev, 2000). The information asymmetry stems from the high degree of uncertainty in payoffs from R&D (Mansfield, Romeo, Villani, Wagner and Husic, 1977; Harhoff, Narin, Scherer and Vopel, 1999) and because many R&D assets are firm-specific and hard-to-transfer tangible and intangible assets. Disclosure about such assets may be limited to keep competitors from inferring valuable technical knowledge. Further, many intangible assets are unrecorded in corporate accounts (Lev and Sougiannis, 1996; Boone and Raman, 2001). Therefore, firms in emergent high-tech industries are difficult for outsiders to evaluate and monitor.

On the other hand, many projects in high-tech industries are still at the early, exploratory stage. Some projects will be successful, but most will fail (e.g., Mansfield, Romeo, Villani, Wagner and Husic, 1977; Harhoff, Narin, Scherer and Vopel, 1999). Therefore, firms with high R&D are associated with higher managerial discretion (Himmelberg, Hubbard and Palia, 1999) and it is thus important that the owners have the control rights to quickly redeploy poorly performing assets. It is difficult for outside shareholders to exercise this control right effectively. Like other industries, outside ownership is generally widely distributed, which makes coordinated effort difficult to achieve. In addition, high-tech firms tend to be closely controlled by company insiders. Lorsch, Zelleke and Pick (2001) argue that this structure is deeply flawed because insiders have strong incentives to maximize short run returns and to engage in cash-out events, not to sacrifice their own jobs by terminating unpromising projects and returning cash to outside investors. In contrast, it is much easier for the central management in an internal market to terminate projects quickly, because the central management generally has the entire control rights over a subsidiary.

In summary, high-tech industries likely suffer from more severe problems of information asymmetry and external markets find these firms more difficult to monitor. Further, external markets lack the ability to shift resources quickly between operating units effectively, which the internal markets can do much better.

In contrast, an internal market is likely to be particularly costly in cases where (1) there is low information asymmetry and (2) there is little need to shift resources between divisions, so the value of the real option to avoid deadweight external financing costs is low, and the agency costs of an internal market become dominant (Matsusaka and Nanda, 2002). Firms in mature industries have high free cash flows and fewer investment opportunities, little or no R&D and little or no need for external financing. Diversification should be costly and value destroying in these industries.

In summary, we contend that internal capital markets can be more efficient in emergent high-tech industries; in such firms, we expect better operating performance across multiple segments and a diversification premium. We argue that the opposite is likely for firms in mature industries where we expect better operating performance for single segment firms and a diversification discount.

3. Research design and data

Prior studies of the relative efficiency of internal versus external markets include a wide range of industries, confounding the effect of diversification. This study avoids the problem by focusing on companies in emergent high-tech industries and in mature low-tech industries.

3.1. Data and sample selection

The study period is 1995–2001, as detailed segment data needed for this study are unavailable for earlier periods. This period includes an extraordinary boom in technology investments and provides a relatively large sample of high-tech single segment firms and diversified firm segments. During the late 1990s, equity markets attached unprecedented valuations, especially to high-tech companies. High-tech start-up firms obtained funding in spite of operating losses and severe information asymmetries. Staged financing and strict scrutiny by venture capitalists were often replaced by lump-sum IPO financing. It is possible that in this environment, external capital markets were not very efficient. This high-tech boom ended in 1999, so we examine whether our results differ between the pre- and post-1999 periods.

Firms in the emergent high-tech sample set are required to have R&D expense greater than 8% of assets, higher than median investment opportunities (as reflected in market-to-book ratios) and higher than median needs for external financing.¹ We require the mature sample to have no R&D, low investment opportunities (market-to-book ratios less than one) and no need for external financing. These criteria result in eight emergent high-tech industries and 17 mature low-tech industries.

The industries and their characteristics are summarized in Table 1. Among the emergent industries, R&D expenses range from 8.23% to 22.65% of assets. Among the emergent industries, external financing ranges from 4.42% to 74.52% of assets (all positive) and from –0.70% to –6.97% (all negative) for the mature industries. The market to book ratios range from 1.58 to 2.97 for the emergent industries and for the mature industries they range between 0.69 and 0.97. Clearly, while there are more firms in the emergent industries group, the two groups of industries have very different characteristics.²

Table 1.  Characteristics of emergent and mature industries, 1995–2001 Emergent industries have R&D expenses above 8% of total assets, and market-to-book ratio and external financing needs above the medians for all industries combined. Mature industries have zero R&D, median market-to-book ratios below one and negative financing needs. Need for external financing is defined as the ratio of Compustat (data 113 – data 109 + data 128 – data 107 + data 129 + data 127 + data 236 – data 123 – data 124 – data 125 – data 126 – data 106 – data 123) to data 6. Total assets are in 1995 constant dollars (millions). During 1995–2001, the median market-to-book ratio is 1.23 and the median deficit is 2.6%, for all industries combined. During this period, 45% of firm years do not report R&D expenses. For those firms that report R&D, the median ratio of R&D (data 46) to assets (data 6) is 4.9%. Panel C is based on I/B/E/S data.

NAICS industry	N	R&D as percent of assets	Operating and investment deficit	Market-to-book ratio	Total assets
Panel A: Emergent industries
3,254: Pharmaceutical and medicine manufacturing	3,201	22.65%	50.97%	2.97	39.76
3,341: Computer and equipment manufacturing	1,775	12.70%	11.70%	1.74	52.56
3,342: Communications equipment manufacturing	1,573	11.20%	4.42%	1.60	58.87
3,345: Instrument manufacturing	2,223	10.92%	3.23%	1.58	38.10
5,112: Software publishers	3,584	17.28%	23.54%	2.33	53.69
5,141: Information services	1,396	8.23%	74.52%	2.07	51.68
5,415: Computer systems design and services	1,945	9.62%	9.36%	1.67	42.30
5,417: Research and development services	402	20.14%	47.43%	2.62	24.08
Panel B: Mature industries
2,212: Natural gas distribution	452	0	−1.44%	0.85	861.72
2,332: Residential building construction	325	0	−6.97%	0.82	154.65
3,152: Cut and sew apparel manufacturing	475	0	−5.79%	0.96	143.67
3,219: Millwork	155	0	−6.63%	0.97	201.84
4,211: Motor vehicle and parts wholesalers	120	0	−2.77%	0.92	87.28
4,217: Hardware, and plumbing and heating equipment and supplies wholesalers	127	0	−0.88%	0.79	99.83
4,218: Machinery and equipment wholesalers	239	0	−2.94%	0.85	132.59
4,229: Misc. nondurable goods wholesalers	108	0	−2.80%	0.94	45.34
4,431: Electronics and appliance stores	139	0	−1.22%	0.83	213.48
4,451: Grocery stores	369	0	−0.70%	0.92	742.22
4,511: Sporting goods, hobby and musical stores	154	0	−2.47%	0.69	155.00
4,521: Department stores	141	0	−3.00%	0.79	1,514.39
4,529: Other general merchandise stores	217	0	−3.76%	0.92	645.80
4,811: Scheduled air transportation	282	0	−0.83%	0.82	672.90
4,821: Rail transportation	151	0	−2.85%	0.76	1,209.27
4,831: Sea, coastal and lakes transportation	154	0	−1.48%	0.81	484.75
4,841: General freight trucking	287	0	−4.45%	0.83	157.06

Panel C: Measures of information asymmetry
	Emergent industry	Mature industry	Difference	T-stat
**indicates statistical significance at the 0.05 level.
Mean number of analyst coverage	1.63	2.11	−0.48**	−9.08
(S.E.)	(2.25)	(2.48)
Mean earnings forecast error	27.03	18.98	8.05**	6.26
(S.E.)	(66.67)	(58.29)
Mean earnings forecast dispersion	15.60	9.77	5.83**	8.82
(S.E.)	(38.35)	(28.74)

Higher R&D expenses make firms in emergent industries more risky (e.g., Chan, Lakonishok and Sougiannis, 2001; Brown and Kapadia, 2007). To assess the role of information asymmetry in the relative balance between the benefits and costs of internal markets, we examine three measures of information asymmetry: mean analyst coverage, earnings forecast error and forecast dispersion based on data from the I/B/E/S database.

For analyst following, we use the mean number of analysts making one-year-ahead earnings forecasts in any month of the year for each firm-calendar-year. Analyst earnings forecast error is defined as the absolute value of the difference between mean earnings forecasts and actual earnings, divided by the absolute value of actual earnings. Dispersion of analyst earnings forecasts is defined as the standard deviation of earnings forecasts scaled by the absolute value of the mean earnings forecast. These measures are widely used in the literature as measures of firms' information environment (e.g., Krishnaswami and Subramaniam, 1999; Kang and Liu, 2008). Panel C of Table 1 shows that emergent high-tech firms are indeed more subject to the problem of information asymmetry: they have lower analyst coverage, higher earnings forecast error and higher earnings forecast dispersion.³

We select all firms that operate in the eight emergent and 17 mature industries from the Segment NAICS file. We separate these firms into stand-alone and diversified firms. We classify a firm as diversified if it has one or more segments operating outside the eight high-tech industries or outside the 17 low-tech industries. Otherwise, firms in these eight and 17 industries, respectively, are regarded as single segment firms. However, if a firm operates in more than one high-tech industry in this study, for example, both software publishing and business support services, we classify it as a stand-alone. There are few firms operating in more than one high-tech segment; excluding these firms from the analysis does not change the results. Further, although in principle a firm could have a segment in both an emergent and a mature industry, no such conglomerate exists in our sample.

Segment data are available from the Business Information File, which contains seven years of information for operating segments required under SFAS 131 (issued in June 1997 and effective for fiscal years beginning after December 15, 1997). SFAS 131 requires a public corporation to report financial information about its operating segments. An operating segment is defined as “an enterprise about which separate financial information is available that is evaluated regularly by the chief operating decision maker in deciding how to allocate resources and in assessing performance.”⁴

SFAS 131 also requires “reconciliation of total segment revenues, total segment profit or loss, total segment assets and other amounts disclosed for segments to corresponding amounts in the enterprise's general-purpose financial statements.” As the data in our sample are reported in accordance with SFAS 131, we expect no discrepancy between the segment and firm levels. For 1995 and 1996 data, while we omit observations where the aggregate segment data deviates from the firm-level data by more than 10%, if there is a smaller discrepancy between the segment and firm levels, we follow the convention of previous studies and allocate any discrepancy to each segment in proportion to total assets.

3.2. Statistical analysis

3.2.1. Summary statistics

Because diversified firm segments and single segment firms differ in size, we adopt a matching method. For each segment-year observation, we select the single segment firm with the same four-digit NAICS code that has the closest total assets in the same year. The distributions of many variables differ from the normal distribution and means for accounting ratios are like to be distorted by outliers, so we use the Wilcoxon–Mann–Whitney test. As neither market values nor replacement costs are available for segments, we cannot calculate or use Tobin's Q. Therefore, we follow Berger and Ofek (1995) and use accounting variables reported at the segment level, or at the firm level for single segment firms. Only five accounting variables are reported at the segment level: total assets, sales, capital expenditure, operating profits and depreciation.

Differences in operating efficiency and profitability are assessed using differences in asset turnover ratios and operating profitability (profit margin on sales and return on assets [ROA]). An advantage of operating efficiency versus profitability is that diversified firms could intentionally allocate more fixed costs or overhead to low-tech segments so that high-tech segments appear artificially more profitable. Since sales and assets are not as easy to misallocate across segments, results using asset turnover ratios are less likely to be contaminated. Finally, we compare the observed market-to-book ratios for diversified firms with imputed market-to-book ratios for the same firms. We apply the same matching method to the segments of these diversified firms operating in other industries to find their stand-alone matches. To calculate the imputed market-to-book ratio for a diversified firm, for each segment, we use the market-to-book ratio of a single segment firm in the same four-digit NAICS industry with the closet total assets. The resulting imputed diversified firm market-to-book ratio is the average of the segment market-to-book ratios weighted by segment total assets.

3.3.2. Regression analysis

Regressions are estimated using operating efficiency and profitability as dependent variables. The independent variables include a dummy variable equal to one if the observation is a segment of a diversified firm, and firm size, lagged investment ratio and depreciation. To account for industry differences, the industry median of the dependent variable is also used as an independent variable.

Some variables are characterized by outliers or have skewed distributions. To reduce distortions caused by such characteristics, we use quantile regression (Kroenker and Bassett, 1978). Quantile regression minimizes the absolute deviations instead of the sum of squared residuals in traditional OLS regression and the regression results are relatively robust to departures from normality.

Define q as the quantile to be estimated; the median is q= 0.5. For each observation i, let r_i be the residual

• 

Define the multiplier h_i

• 

The quantity being minimized with respect to β_i is To cope with heterogeneity, we use a bootstrapped estimator included in Stata to estimate standard errors.

3.3. Sample characteristics

Table 2 reports descriptive statistics of diversified firm segments and single segment firms. Diversified firm segments are compared to their stand-alone peers. Distributions of diversified firm segments and single segment firms in high-tech industries are in Panel A, while those in the mature industries are in Panel B. In total, there are 3,568 and 8,488 diversified firm segments and single segment firms in high-tech industries, and 1,825 diversified firm segments and 1,697 single segment firms in the mature industries. Panel A shows that there are differences in the distribution of the sample by industry.

Table 2.  Diversified firm segments and single segment firms in emergent and mature industries Emergent industries have R&D expenses above 8% of total assets and market-to-book ratio and external financing needs above the medians for all industries combined. Mature industries have zero R&D, median market-to-book ratios below one and negative financing needs. Each diversified firm has one or more segments operating outside the mature or emergent industries. The single segment firm sample consists of firms matched to diversified firm segments; for each segment-year observation, we use the single segment firm with the same four-digit NAICS code that has the closest total assets in the same year. During the sample period of 1995–2001, the median market-to-book ratio is 1.23 and the median deficit is 2.6%, for all industries combined. During this period, 45% of firm years do not report R&D expenses. For firms that report R&D, the median ratio of R&D (data 46) to assets (data 6) is 4.9%.

NAICS	Diversified firm segments		Single segment firms
	Number	Percentage	Number	Percentage
Panel A: Distribution of diversified firm segments and single segment firms in emergent industries
3,254	401	11.24%	1,853	21.83%
3,341	342	9.59%	1,053	12.41%
3,342	485	13.59%	868	10.23%
3,345	719	20.15%	1,175	13.84%
5,112	431	12.08%	1,860	21.91%
5,141	396	11.10%	543	6.40%
5,415	604	16.93%	896	10.56%
5,417	190	5.33%	240	2.83%
Total	3,568	100%	8,488	100.00%
Panel B: Distribution of diversified firm segments and single segment firms in mature industries
2,212	403	22.08%	73	4.30%
2,332	161	8.82%	95	5.60%
3,152	113	6.19%	268	15.79%
3,219	128	7.01%	56	3.30%
4,211	64	3.51%	46	2.71%
4,217	57	3.12%	57	3.36%
4,218	164	8.99%	111	6.54%
4,229	76	4.16%	49	2.89%
4,431	26	1.42%	75	4.42%
4,451	86	4.71%	158	9.31%
4,511	36	1.97%	85	5.01%
4,521	39	2.14%	63	3.71%
4,529	86	4.71%	114	6.72%
4,811	35	1.92%	172	10.14%
4,821	76	4.16%	47	2.77%
4,831	90	4.93%	72	4.24%
4,841	185	10.14%	156	9.19%
Total	1,825	100.00%	1,697	100.00%

	Total assets ($ million)		Total sales ($ million)
	Diversified firm segments	Single segment firms	Diversified firm segments	Single segment firms
Panel C: Total assets and sales of diversified firm segments and single segment firms in emergent industries
Median	26.90	37.61	23.79	21.53
Mean	924.99	424.46	935.78	370.45
Panel D: Total assets and sales of diversified firm segments and single segment firms in mature industries
Median	275.58	295.40	295.40	369.63
Mean	1,370.35	1,587.06	2,013.34	2,164.15

Panels C and D report the summary statistics of total assets and sales for diversified firm segments and single segment firms in emergent and mature industries, respectively. For the emergent industries, single segment firms have larger median total assets but diversified firm segments have larger mean total assets. Diversified firm segments also have higher sales than single segment firms. Among mature industries, single segment firms are larger in both sales and assets. All the size distributions are highly skewed.

4. Results

4.1. Informational efficiency

For the emergent industries, Table 3 reports that diversified firm segments have an unambiguously higher median asset turnover ratio compared to their single segment counterparts, and the difference is highly statistically significant. Further, segments of diversified firms in emergent high-tech industries are more profitable than their single segment counterparts. In contrast, as shown in Table 4 for the mature industries, segments of diversified firms generally have lower asset turnover, lower profit margin and lower ROA compared to their single segment counterparts. The results support the transaction-cost hypothesis.

Table 3.  Comparison of efficiency and profitability of diversified firm segments and single segment firms in emergent industries Emergent industries have R&D expenses above 8% of total assets and market-to-book ratio and external financing needs above the medians for all industries combined. Mature industries have zero R&D, median market-to-book ratios below one and negative financing needs. Each diversified firm has one or more segments operating outside the mature or emergent industries. The single segment firm sample consists of firms matched to diversified firm segments; for each segment-year observation, we use the single segment firm with the same four-digit NAICS code that has the closest total assets in the same year. We omit firm observations where the aggregate segment data deviates from the firm-level data by more than 10%. We allocate any discrepancy between the firm and segment levels to each segment in proportion to segment total assets.

	Median asset turnover			Median profit margin			Median ROA
	Diversified firm segment	Single segment firms	Difference	Diversified firm segment	Single segment firms	Difference	Diversified firm segment	Single segment firms	Difference
**indicates statistical significance at the 0.05 level.
Panel A: By size quartile
1st quartile	1.68	1.09	0.59	−17.47%	−103.73%	86.26%	−24.90%	−100.40%	75.51%
2nd quartile	1.24	0.82	0.42	−2.10%	−38.21%	36.11%	−2.33%	−29.00%	26.67%
3rd quartile	0.99	0.76	0.23	5.08%	−2.80%	7.88%	5.36%	−2.53%	7.89%
4th quartile	0.98	0.85	0.13	8.55%	11.19%	−2.64%	8.29%	10.56%	−2.28%
Panel B: By year
1995	1.19	0.98	0.21	5.42%	5.16%	0.26%	6.32%	6.16%	0.16%
1996	1.10	0.99	0.11	4.59%	1.85%	2.74%	4.51%	1.38%	3.14%
1997	1.07	0.93	0.14	4.84%	−0.29%	5.13%	5.50%	−0.36%	5.86%
1998	1.12	0.98	0.14	3.72%	−6.73%	10.44%	4.10%	−7.35%	11.45%
1999	1.19	0.82	0.37	2.98%	−11.93%	14.90%	2.96%	−9.89%	12.84%
2000	1.13	0.75	0.38	−0.55%	−32.27%	31.71%	−0.46%	−18.93%	18.47%
2001	1.10	0.72	0.38	1.71%	−25.69%	27.40%	1.85%	−19.03%	20.88%
Overall	1.13	0.85	0.28	3.05%	−11.53%	14.58%	3.11%	−10.37%	13.47%
Mann–Whitney p-value			<0.0001**			<0.0001**			<0.0001**

Table 4.  Comparison of efficiency and profitability of diversified firm segments and single segment firms in mature industries Mature industries have zero R&D, median market-to-book ratios below one and negative financing needs. Each diversified firm has one or more segments operating outside the mature or emergent industries. The single segment firm sample consists of firms matched to diversified firm segments; for each segment-year observation, we use the single segment firm with the same four-digit NAICS code that has the closest total assets in the same year. We omit firm observations where the aggregate segment data deviates from the firm-level data by more than 10%. If there is a discrepancy between data at segment and firm levels, we allocate the discrepancy to each segment in proportion to segment total assets.

	Median asset turnover			Median profit margin			Median ROA
	Segment	Single segment firms	Difference	Diversified firm segment	Single segment firms	Difference	Diversified firm segment	Single segment firms	Difference
**indicates statistical significance at the 0.05 level.
Panel A: By size quartile
1st quartile	1.61	1.68	−0.07	2.80%	4.02%	−1.23%	5.54%	8.62%	−3.09%
2nd quartile	1.42	1.32	0.11	5.29%	6.96%	−1.67%	7.09%	7.01%	0.08%
3rd quartile	0.96	1.02	−0.06	8.06%	9.26%	−1.20%	7.31%	9.39%	−2.08%
4th quartile	0.97	0.91	0.06	8.25%	9.18%	−0.92%	8.08%	8.46%	−0.38%
Panel B: By year
1995	1.20	1.20	−0.01	6.51%	8.19%	−1.68%	7.68%	8.03%	−0.35%
1996	1.12	1.16	−0.04	6.38%	8.12%	−1.73%	8.23%	8.52%	−0.30%
1997	1.15	1.09	0.06	6.56%	7.45%	−0.89%	7.80%	8.69%	−0.89%
1998	1.24	1.38	−0.14	6.68%	8.25%	−1.57%	7.07%	8.32%	−1.26%
1999	1.26	1.25	0.00	6.45%	7.34%	−0.89%	7.26%	8.93%	−1.67%
2000	1.34	1.25	0.09	5.44%	7.03%	−1.59%	7.17%	8.23%	−1.06%
2001	1.34	1.23	0.11	3.39%	6.77%	−3.38%	6.09%	7.68%	−1.59%
Overall	1.26	1.25	0.01	6.11%	7.31%	−1.20%	7.29%	8.43%	−1.14%
Mann–Whitney p-value			0.2918			<0.0001**			0.0002**

Maksimovic and Phillips (2002) find differences in behavior between peripheral and main segments. To check whether our results are contaminated by differences in relative sizes of segments in diversified firms, we measure relative segment position by asset weight, defined as segment total assets as a proportion of total firm size. We break the sample into four quartiles; the smallest asset-weight quartile is composed of peripheral segments, while the largest asset-weight quartile consists of main segments. We compare the same efficiency and profitability ratios across different asset-weight quartiles in Table 5. The results show that, across different asset-weight quartiles, diversified firm segments are more efficient and more profitable than single segment firms in the emergent industries and less profitable than single segment firms in the mature industries. Therefore, we conclude that our results are not driven by different behavior between peripheral and main segments.

Table 5.  Efficiency and profitability of diversified firm segments and matched single segment firms by segment importance Emergent industries have R&D expenses above 8% of total assets and market-to-book ratio and external financing needs above the medians for all industries combined. Mature industries have zero R&D, median market-to-book ratios below one and negative financing needs. Each diversified firm has one or more segments operating outside the mature or emergent industries. The single segment firm sample consists of firms matched to diversified firm segments; for each segment-year observation, we use the single segment firm with the same four-digit NAICS code that has the closest total assets in the same year. Asset weight measures relative segment importance and is defined as segment total assets divided by total firm assets. We omit firm observations where the aggregate segment data deviates from the firm-level data by more than 10%. We allocate any discrepancy between the firm and segment levels to each segment in proportion to total assets. Wilcoxon–Mann–Whitney test p-values are in parentheses.

	Emergent industries			Mature industries
	Segment	Single segment firms	Difference	Segment	Single segment firms	Difference
** and *indicate statistical significance at the 0.05 and 0.10 level, respectively.
Panel A: Efficiency (asset turnover ratios)
Smallest asset weight quartile	1.27	1.31	−0.04	1.98	1.85	0.14
			(0.66)			(0.20)
2nd quartile	1.14	0.81	0.33	1.78	2.25	−0.48
			(<0.01)**			(<0.01)**
3rd quartile	0.91	0.78	0.14	1.34	1.33	0.01
			(<0.01)**			(0.41)
Largest asset weight quartile	0.90	0.82	0.08	1.02	0.99	0.02
			(0.04)**			(0.69)
Panel B: Profitability (profit margin − profit as a percent of sales)
Smallest asset weight quartile	−19.89%	−59.79%	39.90%	0.88%	3.10%	−2.23%
			(<0.01)**			(0.09)
2nd quartile	−4.12%	−40.77%	36.64%	3.10%	2.64%	0.45%
			(<0.01)**			(0.87)
3rd quartile	4.68%	−4.69%	9.36%	4.92%	6.68%	−1.75%
			(<0.01)**			(<0.01)**
Largest asset weight quartile	8.65%	8.71%	−0.06%	7.87%	8.98%	−1.10%
			(0.60)			(<0.01)**
Panel C: Profitability (ROA)
Smallest asset weight quartile	−32.82%	−76.36%	43.54%	2.34%	10.15%	−7.82%
			(<0.01)**			(0.68)
2nd quartile	−6.78%	−32.30%	25.53%	5.92%	8.23%	−2.31%
			(<0.01)**			(0.51)
3rd quartile	3.11%	−4.19%	7.30%	6.69%	7.57%	−0.89%
			(<0.01)**			(0.06)*
Largest asset weight quartile	6.66%	9.11%	−2.46%	7.80%	8.53%	−0.73%
			(0.17)			(<0.01)**

4.2. Determinants of performance

Table 6 presents the quantile regression with the segment dummy (diversified firm segments as one and single segment firms as zero). Panel A shows that after controlling for other relevant variables, diversified firm segments have significantly higher asset turnover ratios in the emergent industries, but for mature industries the results are mostly not significant. Panels B and C show that both profit margins and ROA are significantly lower for diversified firm segments in the mature industries and significantly higher for diversified firm segments in the emergent industries. These contrasting results support our transaction-cost–based line of reasoning.

Table 6.  Performance determinants: quartile regressions by subperiod Emergent industries have R&D expenses above 8% of total assets and market-to-book ratio and external financing needs above the medians for all industries combined. Mature industries have zero R&D, median market-to-book ratios below one and negative financing needs. Each diversified firm has one or more segments operating outside the mature or emergent industries. The single segment firm sample consists of firms matched to diversified firm segments; for each segment-year observation, we use the single segment firm with the same four-digit NAICS code that has the closest total assets in the same year. We omit firm observations where the aggregate segment data deviates from the firm-level data by more than 10%. We allocate any discrepancy between the firm and segment levels to each segment in proportion to total assets. The segment dummy is equal to one for a diversified firm segment.

Panel A: Efficiency (asset turnover ratios)
Variables	Emergent industries				Mature industries
	1995–1997	1998–1999	2000–2001	1995–2001	1995–1997	1998–1999	2000–2001	1995–2001
Segment (dummy)	0.22	0.24	0.33	0.30	0.05	0.05	0.01	0.05
	(0.04)**	(0.04)**	(0.03)**	(0.02)**	(0.03)*	(0.04)	(0.02)	(0.02)**
Size (log sales)	0.02	0.03	0.05	0.04	0.03	−0.24%	0.02	0.02
	(0.72%)**	(0.71%)**	(0.42%)**	(0.27%)**	(0.01)**	(0.01)	(0.01)**	(0.38%)**
Industry median	0.88	1.07	0.75	0.91	1.03	1.01	0.97	1.02
	(0.09)**	(0.08)**	(0.05)**	(0.03)**	(0.02)**	(0.03)**	(0.02)**	(0.01)**
Lagged	−0.08	−0.02	−0.51%	−0.70%	−0.09	−0.21	−0.72	−0.10
Investment ratio	(0.03)**	(0.10)	(0.03%)**	(0.02%)**	(0.01)**	(0.16)	(0.12)**	(0.01)**
Lagged	2.31	3.12	0.48	0.71	−0.93	−0.84	2.55	−0.04
Depreciation	(0.23)**	(0.25)**	(0.03)**	(0.02)**	(0.50)	(0.39)**	(0.24)**	(0.19)
Intercept	−0.11	−0.40	−0.15	−0.22	−0.19	0.03	−0.12	−0.14
	(0.09)	(0.08)**	(0.05)**	(0.03)**	(0.05)**	(0.08)	(0.04)**	(0.03)**
Pseudo R^2	0.09	0.09	0.11	0.10	0.34	0.30	0.26	0.29

Panel B: Profitability (profit margin – profit as a percent of sales)
Variables	Emergent industries				Mature industries
	1995–1997	1998–1999	2000–2001	1995–2001	1995–1997	1998–1999	2000–2001	1995–2001
Segment (dummy)	0.03	0.08	0.13	0.09	−0.83%	−0.67%	−0.18%	−0.59%
	(0.01)**	(0.02)**	(0.02)**	(0.01)**	(0.40%)**	(0.36%)*	(0.08%)**	(0.25%)**
Size (log sales)	0.05	0.06	0.12	0.07	0.50%	0.40%	0.68%	0.56
	(0.23%)**	(0.37%)**	(0.33%)**	(0.22%)**	(0.10%)**	(0.09%)**	(0.02%)**	(0.06%)**
Industry median	0.19	0.50	1.15	0.80	1.00	1.00	0.95	1.01
	(0.03)**	(0.04)**	(0.03)**	(0.02)**	(0.05)**	(0.04)**	(0.01)**	(0.03)**
Lagged	0.06	−0.02	−0.20	−0.19	0.28%	0.55%	0.05	0.27%
Investment ratio	(0.01)**	(0.05)	(0.01%)**	(0.02%)**	(0.20%)	(1.48)	(0.40%)**	(0.22%)**
Lagged	−1.34	−1.67	−0.34	−0.99	0.02	−0.23	−0.21	−0.20
Depreciation	(0.06)**	(0.13)**	(0.01)**	(0.02)**	(0.07)	(0.04)**	(0.01)**	(0.03)**
Intercept	−0.13	−0.27	−0.70	−0.36	−0.02	−0.82%	−0.04	−0.02
	(0.01)**	(0.02)**	(0.02)**	(0.01)**	(0.01)**	(0.66%)	(0.14%)**	(0.046%)**
Pseudo R^2	0.02	0.00	0.00	0.00	0.23	0.20	0.05	0.09

Panel C: Profitability (ROA)
Variables	Emergent industries				Mature industries
	1995–1997	1998–1999	2000–2001	1995–2001	1995–1997	1998–1999	2000–2001	1995–2001
** and *indicate statistical significance at the 0.05 and 0.10 level, respectively.
Segment (dummy)	0.05	0.08	0.12	0.09	−0.46%	−0.99%	−1.34%	−0.81%
	(0.02)**	(0.01)**	(0.02)**	(0.01)**	(0.34%)	(0.28%)**	(0.37%)**	(0.23%)**
Size (log sales)	0.05	0.06	0.06	0.05	0.86%	0.69%	0.74%	0.77%
	(0.28%)**	(0.20%)**	(0.29%)**	(0.18%)**	(0.09%)**	(0.07%)**	(0.08%)**	(0.06%)**
Industry median	0.18	0.14	0.49	0.38	1.49	0.77	0.69	0.95
	(0.08)**	(0.05)**	(0.07)**	(0.04)**	(0.12)**	(0.10)**	(0.12)**	(0.08)**
Lagged	0.04	−0.12	−0.19%	−0.21%	0.43%	0.41%	0.07	0.35%
Investment ratio	(0.01)**	(0.03)**	(0.01%)**	(0.01%)**	(0.17%)**	(1.12)	(0.02)**	(0.20%)**
Lagged	−2.42	−2.75	−1.98	−1.98	−0.02	−0.31	−0.63	−0.39
Depreciation	(0.09)**	(0.07)**	(0.67%)**	(0.59%)**	(0.06)	(0.03)**	(0.03)**	(0.03)**
Intercept	−0.12	−0.17	−0.27	−0.22	−0.08	−0.01	0.47%	−0.02
	(0.02)**	(0.01)**	(0.02)**	(0.01)**	(0.01)**	(0.01)	(0.11%)**	(0.01)**
Pseudo R^2	0.14	0.16	0.28	0.23	0.08	0.04	0.03	0.04

An alternative explanation for the above finding can be differences in life-cycle stages between diversified firm segments and single segment firms in high-tech industries. Borghesi, Houston and Naranjo (2007) argue that firm age explains a significant proportion of the diversification discount. In other words, single segment firms may be younger than diversified firm segments, so it might not be surprising that diversified firm segments have better operating performance. To address this concern, we examine year-by-year increases in diversified firm segments and single segment firms. We find that in the high-tech industry the proportion of new entrants that are single segment firms is not higher than the proportion of new diversified firm segments. Because of the accounting rule change, the number of diversified firm segments in 1998 is not comparable to that before 1998. From 1998 to 2000 (the peak of the high-tech bubble), there is an increase of 57% in the number of new high-tech segments, from 520 to 818. In contrast, there is an increase of 36% in the number of single segment firms, from 1,166 to 1,591. Thus, there are more new entrants to high-tech among diversified firm segments. Therefore, the alternative explanation is not likely to hold.

4.3. Likelihood of project termination

We now examine whether a lack of control rights can explain why internal markets outperform external markets in emergent industries. Emergent high-tech firms generally raise large amounts of equity from IPOs or SEOs that is then spent over the next few years. During this period, outside investors cannot withdraw financing from the firms even though it might later be found that the firms have chosen bad projects. This is highlighted by the “deathwatch” of many dot.coms after March 2000, when analysts were estimating when these dot.coms would go bankrupt given their monthly “burn rate” and existing financial resources.⁵ On the contrary, high-tech diversified firm segments can only obtain staged financing from their head offices and are evaluated on a regular basis. The head offices have the power to quickly shut the segment down if conditions deteriorate. As a result, it is worthwhile to inspect the survival rate for loss-making diversified firm segments and single segment firms.

Our focus here is not the actual survival rate of a loss-making diversified firm segment or a single segment firm. Instead, our goal is to investigate whether there is a difference in survival rate for loss-making diversified firm segments and single segment firms. We use the semi-parametric Cox proportional hazard model (Cox, 1972) for this purpose. This model assumes a linear parametric form for the effects of the explanatory variables, but allows an unspecified form for the underlying survival functions. The model is widely used in the analysis of censored survival data to explain the effect of explanatory variables on survival times. It is

• 

where H(.) is the hazard function, which describes the rate at which a loss-making operating unit deceases after duration t_(i), given that they last at least until t_(i). H₀(t_(i)) is the underlying survival function that is unspecified. x is a dummy variable equal to one for diversified firm segments. We test whether β is significantly different from zero. A β not equal to zero suggests that diversified firm segments and single segment firms are equally likely to survive. A positive β implies that loss-making diversified firm segments face lower survival rates than single segment firms.

We include only loss-making diversified firm segments and single segment firms when carrying out this test. There are 556 loss-making diversified firm segments and 1,237 loss-making single segment firms. The proportional hazard model shows that the β estimate is significantly different from zero (, standard error = 0.08177, p-value < 0.0001). This implies that the survival probability is significantly lower for diversified firm segments. These findings support the idea that internal capital markets have a control-rights advantage as nonperforming projects can be terminated relatively quickly by corporate management in a diversified firm as compared to nonperforming single segment firms.⁶

The lower survival rate in internal capital markets could also mean that diversified firm segments with profit potential are randomly or wrongly eliminated. If so, the percentage of loss-making diversified firm segments that later become profitable should be lower than that of loss-making single segment firms. Instead, we find that 57 diversified firm segments (10.25%) turn profitable, compared with 110 single segment firms (8.89%); and 23 (4.13%) diversified firm segments improve their ROA to more than 20% by the end of the sample period compared with only nine (0.73%) single segment firms. The result is consistent with the idea that internal capital markets eliminate segments with poor futures but retain those with promising futures.

4.4. Value of diversification

Table 7 reports observed versus imputed market valuations for diversified firms. This table shows that observed valuations for diversified firms in mature industries are significantly lower than their imputed valuations, indicating a significant diversification discount. In contrast, imputed valuations are significantly higher than observed valuations for diversified firms in emergent industries, indicating a diversification premium. This result supports the transaction-cost hypothesis that internal markets should be more efficient in the high-tech industries but less efficient in the mature industries.

Table 7.  Observed versus imputed valuations for diversified firms Emergent industries have R&D expenses above 8% of total assets and market-to-book ratio and external financing needs above the medians for all industries combined. Mature industries have zero R&D, median market-to-book ratios below one and negative financing needs. Each diversified firm has one or more segments operating outside the mature or emergent industries. We omit firm observations where the aggregate segment data deviates from the firm-level data by more than 10%. We allocate any discrepancy between the firm and segment levels to each segment in proportion to total assets. The imputed market-to-book ratio for each segment is that of the matching single segment firm; the imputed ratio of the diversified firm as a whole is the mean of the segment imputed ratios weighted by segment total assets.

	Diversified firms in mature industries			Diversified firms in emergent industries
	Median observed market-to-book ratio	Median imputed market-to-book ratio	Median diversification discount	Median observed market-to-book ratio	Median imputed market-to-book ratio	Median diversification premium
1995	0.83	0.94	−11.85%	1.15	1.04	10.99%
1996	0.83	0.94	−12.17%	1.17	1.07	9.43%
1997	0.92	1.07	−13.97%	1.34	1.12	19.77%
1998	0.83	0.98	−14.76%	1.28	1.07	19.54%
1999	0.76	0.93	−17.64%	1.50	1.44	4.73%
2000	0.77	0.85	−9.79%	1.09	1.07	1.67%
2001	0.78	0.88	−11.45%	1.19	0.98	22.27%
Overall	0.81	0.93	−12.41%	1.24	1.10	12.68%
Wilcoxon–Mann–Whitney p-value			<0.0001			<0.0001

6. Conclusions

In this paper, we apply transaction-cost economics to argue that diversification is likely to be value enhancing when external capital markets are relatively inefficient (as in the case of emergent high-tech industries) and vice versa. We examine this contention using the performance of firms in eight emergent high-tech industries and 17 mature industries. We show that diversified firm segments outperform their single segment counterparts in high-tech industries. Further, underperforming units of diversified firms are more likely to be shut down than underperforming single segment firms in these high-tech industries. In contrast, in mature industries, where companies face relatively high internal costs and low external transaction costs, diversified firms under-perform single segment firms.

These results support our contention that the value impact of diversification depends on the balance between internal and external transaction costs. These findings help clarify the reasons for the mixed evidence on the diversification discount found in prior literature. Future research on firm diversification should separate firms based on characteristics that reflect internal versus external transaction costs to avoid confounding and offsetting diversification effects.

Footnotes

• 1

  We gauge the need for external financing using the gap between internal funds from operations and investment activities and the amount needed to finance capital expenditure, dividend payout, and other business activities. In the annual Compustat database, financing deficit is the ratio of (data 113 − data 109 + data 128 − data 107 + data 129 + data 127 + data 236 − data 123 − data 124 − data 125 − data 126 − data 106 − data 123) to data 6.

• 2

  To control for the possibility that our findings are driven by different industry characteristics, rather than information asymmetry and control rights, we examine the sample of just the manufacturing firms in both emergent and mature industries, and the results are similar.

• 3

  It is difficult to measure agency costs because agency problems are multi-dimensional. The corporate governance literature shows that one governance measure (or a few) cannot capture the overall agency problem, and that although some governance mechanisms are complementary, some are supplementary (Gillan, Hartzell and Starks, 2006). For this reason, we do not include governance measures in this study.

• 4

  This requirement of SFAS 131 is particularly important for this study. Since fiscal years that ended after December 15, 1977, SFAS 14 requires only that some data be reported for segments that exceed 10% of total sales (with no requirement for overall reconciliation or other details).

• 5

  See, for example, http://archive.salon.com/tech/log/2000/06/06/deadpool and http://www.downside.com/deathwatch.html.

• 6

  The median first-year ROA is –37.44% for loss-making diversified firm segments and –53.88% for single segment firms. Upon exit time or censor time, the median ROA is –33.22% for diversified firm segments and –43.77% for single segment firms.