《(绩效的驱动力是行业因素还是企业特定因素.docx》由会员分享，可在线阅读，更多相关《(绩效的驱动力是行业因素还是企业特定因素.docx（18页珍藏版）》请在三一文库上搜索。
1、.Analysis of English Academic Paper:IS PERFORMANCE DRIVEN BY INDUSTRY OR FIRM-SPECIFIC FACTORS? A RESPONSE TO HAWAWINI, SUBRAMANIAN, AND VERDIN(绩效的驱动力是行业因素还是企业特定因素？对HAWAWINI、SUBRAMANIAN和VERDIN的文章的回应) A paper from the Strategic Management Journal Strat.Mgmt.J.,26:1075-1081(2005) Number: 061620 Name:
2、陈燕 Tele: 13770613178 Email: cendy-803 Blog: Major: Technical Economics and Management ;- 18 -This is the original paper with some marks that I made: (*As the tables (table1,2) have no use for my analysis, in order to beautify my edition, I have adjusted the location of two tables. )ABSTRACTHawawini,
3、 Subramanian, and Verdin (2003) examined the relative impact of industry-vs. firm-level factors shaping firm performance. They demonstrated that variance in firm performance attributable to industry-level factors increases, while variance attributable to firm-level factors decreases when exceptional
4、ly higher-and lower-performing outlier firms in each industry are excluded. They concluded that previous research underestimated the relative impact of industry-level factors for average firms that make up the bulk of an industry. We take issue with their methods used to identify and exclude outlier
5、s as well as their conclusions drawn from such analyses. Rather than excluding true outlier firms, we argue that they incorporated an artificial restriction of within-industry sample variance that almost deterministically led to lower firm and higher industry variance component estimates. We demonst
6、rate this point with a comparable sample of data to which we apply progressively greater restrictions on within-industry sample variance leading to similar results. Finally, we show that exclusion of firms from a data sample based on commonly understood standards of outlier identification leads to l
7、ittle change in industry and firm variance component estimates compared to full-sample estimates. INTRODUCTION1 Recently, Hawawini, Subramanian, and Verdin (2003) asked whether findings from past research on determinants of variability in firm performance (e.g., Rumelt, 1991; McGahan and Porter, 199
8、7) are generalizable across all firms in an industry or whether they are, instead, driven by a few exceptional, outlier firms. These results are important since they shed light on a fundamental issue in strategy research: the relative importance of industry vs. firm factors on firm profitability. In
9、 this commentary, we discuss concerns regarding methods used and conclusions drawn with their examination of this issue. 2 Hawawini et al. began their inquiry by partitioning variance in firm performance using a sample of 5620 annual observations for 562 large corporations from 1987 to 1996. Their i
10、nitial results were consistent with prior studies. With both accounting measures and two economic measures not previously analyzed in this literature (i.e., economic profit and total market value), their results indicated that variance components (VC) related to firm effects were largest with indust
11、ry effects of secondary importance. Consistent with prior studies, their initial VC results suggested that intra-industry strategy perspectives had the greatest explanatory power in any analysis of performance differences among firms.3 BHawawini et al. then questioned this conclusion, A arguing that
12、 these findings may be driven by a sub-set of firms described as exceptional firms (e.g., pp. 1, 2), value leaders and losers (e.g., 10, 12, 13, 14), and quite often as outliers (e.g., 2, 10, 12, 13, 14). To test this proposition, Hawawini et al. repeated their VC analysis, but this time with a modi
13、fied sample excluding the top and bottom two performing firms within each industry. This analysis yielded a new set of VC estimates with different strategy research implications. Stable industry and firm effects became roughly equivalent. Hawawini et al. concluded that previous studies underestimate
14、d the importance of industry-level factors in shaping firm performance. A few outlier firms in each industry skewed VC estimates in favor of firm effects. 4 We disagree. VC results using the modified sample excluding outlier firms potentially misinform scholarship on this fundamental strategy questi
15、on. As we show below, their methods were inconsistent with the commonly understood conceptual definition, identification, and methodological treatment of outlier firms in management research. We empirically demonstrate that this artificial restriction of within-industry variance leads, by constructi
16、on, to lower firm and higher industry VC estimates. CONCEPTUAL ANALYSIS Outlier identification and exclusion 5 Our criticism focuses primarily on the criteria Hawawini et al. used to select their modified sample. In establishing sample selection boundaries, methodologists commonly cite two justifica
17、tions for restricting sample ranges (Cohen et al., 2003). BFirst, researchers may argue for restricting the sample range if initial sample analyses reveal the presence of outliers that skew results and their interpretation. Second, researchers may argue for restricting the sample range if the potent
18、ial range of data to be sampled extends outside the intended range of interest in the current study. 6 At first glance, it seems that Hawawini et al. opted for the first justification for sample restriction. BThey argued for excluding outlier observations from the full sample so that they could focu
19、s on the vast majority of the firms that are not notable leaders or losers in their industry (p. 1). The key question becomes whether they appropriately identified outliers. AOutliers are one or more atypical data points that do not fit with the rest of the data (Cohen et al., 2003: 390), even thoug
20、h those data points were sampled from the population of interest. They are extreme rarities in most samples. Texts dealing with outlier identification typically note heuristic guides of at least 3.0 standard deviations (S.D.) away from the mean as appropriate for preliminary identification in large
21、samples (Cohen et al., 2003).3 BResearchers may consider outlier exclusion from the sample followed by reanalysis of the remaining data (Chatterjee and Wiseman, 1983). Hawawini et al.s definition of outliers is much broader than this. BIf the observations within an industry are roughly normally dist
22、ributed, standard rules for outlier identification (3 S.D. rule) would identify approximately 1 percent of the observations as outliers. In removing the top and bottom two firms per industry, Hawawini et al. reduced their sample from 5620 to 3420 observations, a reduction of 39.1 percent. Exclusion
23、of nearly 40 percent of the observations as outliers certainly violates commonly understood heuristics for identifying outliers in empirical research.Sample range restriction and VC analyses 7 If their outlier exclusion cannot be justified, what about the other common justification? Were Hawawini et
24、 al.s actions in creating their modified sample an appropriate restriction in range for their VC analyses? Sample range issues derive from understanding how an empirical research issue implicates a population to be studied. Researchers restrict the range of observations drawn from a population for a
25、ny number of reasons, but the focal purpose of any sample restriction should be to ensure that Athe most relevant subjects for a given topic of inquiry are included, while subjects outside the range of the research topic are excluded. For example, an empirical researcher examining issues related to
26、management succession in entrepreneurial firms may decide to restrict the size of firms sampled. B Failure to impose the range restriction could result in a sample with observations that go beyond the population of interest and could distort observed results. 8 However, the range restriction employe
27、d in this study was not appropriate for a number of reasons. First, Hawawini et al. applied a restriction on the variance in firm performance, the primary criterion of interest in this study. Thus, they were restricting the range of the criterion they were trying to explain. Second, and more importa
28、ntly, A the range restriction was tied directly to one of the right-hand side terms they sought to estimate: the VC estimate for intra-industry firm effects. Since they restricted the range for only one of the dimensions for which they were decomposing variance, they directly influenced the results.
29、 This rendered firm effects less substantial relative to the others in the model. Also, because of the nested nature of their VC modelAfirm effects are nested within industry effectsthe range restriction on firm variance was indirectly tied to the industry component. As a result, the range restricti
30、on on intra-industry variance in firm performance almost certainly raised the relative value of the VC estimate for stable industry effects. BThe only question is how much it would raise industry effects. With a 40 percent range restriction, the answer was, not surprisingly, a lot. As we will show i
31、n our analyses, Athe degree to which a researcher can alter the relative importance of such VC estimates is directly related to the degree to which he or she restricts the range on any one dimension. EMPIRICAL ANALYSES 9 We illustrate this point with our own set of VC estimates using a comparable sa
32、mple. To start, we follow other previous VC studies (e.g., McGahan and Porter, 1997) by drawing our sample of firm accounting returns from the Compustat Industry Segment database. Our sample includes all U.S.-based corporations in the database from 1987 to 1996 that have at least $100 million in sal
33、es and assets from non-financial and non-governmental industries. Our resulting database comprises 19,926 annual business unit observations from 2686 businesses operating in 84 industries (defined at the 3-digit SIC level).The dependent variable in our model is, thus, business unit (firm) return on
34、assets (ROA).RESULTS AND CONCLUSION Results when increasing intra-industry range restrictions 10 Our full sample VC estimation (Table 1, column 1) produces results very comparable to Hawawini et al.s full sample estimates. Our analysis attributes 9.1 percent of variance in firm ROA to stable industr
35、y effects and 4.0 percent to unstable industry effects, compared to 8.1 percent and 3.1 percent respectively in Hawawini et al. Our firm effects are 43.8 percent, while their firm effects were 35.8 percent. BOur firm effects may be higher since our sample is based on individual business unit data. H
36、awawini et al. used corporate-level data with individual business units consolidated into a single firm. 11 We re-estimate the model after excluding the top and bottom one then three and finally six firms from each industry, for a total reduction of 6 percent, 18 percent, and 38 percent in sample si
37、ze, respectively. We defined these extreme firms based on their average ROA within each industry from 1987 to 1996. Exclusion of the top and bottom six (12 total) eliminates 38 per-cent of the firms in our base samplecomparable to the 39 percent reduction in Hawawini et al. As expected, stable firm
38、effects decrease from 43.8 percent with no intra-industry range restrictions to 37.7 percent then 29.8 percent and finally 22.9 percent of total variance with the greatest intra-industry range restriction. AStable firm effects are nearly halved (43.8% to 22.9%) with the range restriction. This resul
39、t is comparable to the 53 per-cent decline in firm effects reported in Hawawini et al., a result we expected since we replicated their inappropriate sample selection methodology with this analysis. 12 ASince firms are nested within industries, an exclusion of extreme observations within industries a
40、lso has a profound impact on the higher-level industry effects. The increasing intra-industry range restriction increases stable (unstable) industry effects from 9.1 percent (4.0%) to 14.7 percent (5.2%) for combined industry effects of 19.9 percent in the most restricted sample. For us, combined in
41、dustry effects are slightly lower than stable firm effects of 22.9 percent, compared to slightly higher combined industry effects (20.1% industry vs. 16.7% firm effect) in Hawawini et al. AHowever, Bone would be mistaken to conclude that our findings endorse their conclusions. With these analyses, w
42、e intended to show that the changes in firm vs. industry effects seen in Hawawini et al. are simply a consequence of their sample selection methodology. Thus, they concluded that this shift indicates that past research underestimated the impact of industry factors on the great body of unexceptional
43、firms. In contrast, we conclude that it is simply an artifact of a range restriction of intra-industry sample variance. With increasing restrictions, we can obtain ever-higher industry effects at the expense of firm effects. Results when increasing inter-industry range restrictions 13 To highlight t
44、his point, we conduct an additional set of VC analyses where we remove outlier industries to demonstrate that a range restriction on any one dimension will have significant, but predictable, results. Table 2 presents results from VC estimation based on this new inter-(rather than intra-) industry ra
45、nge restriction approach. We re-estimate the model after excluding the top-and bottom-performing two industries (5% of all industries), seven industries (17%) and 16 industries (38%) from the base sample. We define these top (bottom) industry performers based on their aver-age ROA over the study per
46、iod. 14 As expected, stable industry effects decrease from 9.1 percent to 6.5 percent then 4.9 percent and finally 1.5 percent of total variance with the greatest inter-industry range restriction. This decrease in industry effects is much more dramatic than the modest halving of firm effects with a
47、similar intra-industry range restriction. Industry effects compared to firm effects are much more sensitive to range restrictions. Hawawini et al. concluded that prior research understated industry effects for firms that were not exceptional performers. BBased on VC results in Table 2, we could conc
48、lude that prior research overstated industry effects for firms in industries that were not exceptional in performance. However, such conclusions are inappropriate. These results derive from the same sort of artificial sample range restrictions we criticized earlier, only this time the range restrictions are at the industry level. We note one ke