Download presentation
Presentation is loading. Please wait.
Published byTobias Booth Modified over 9 years ago
1
Dynamic Text Based Industry Classifications and Endogenous Product Differentiation By Gerard Hoberg University of Maryland and Gordon Phillips University of Maryland and NBER Research presented to IFN, Stockholm
2
Motivation: Relatedness and Competition: How close and to whom? If I merge, which partner? How might the industry evolve over time? 2 R1R1 R2R2 R3R3 R4R4 R5R5 R6R6 R9R9 R7R7 R8R8 R 10 T R1R1 R2R2 R3R3 R4R4 R5R5 R6R6 R9R9 R7R7 R8R8 T Very Close Competition? Incentives to change competition? R10 in same industry? Somewhat Close More Synergies?
3
Motivation - 1 Endogenous Barriers to Entry: ( Shaked and Sutton (1987), Sutton (1991), Siem (2006), Nevo (2000, 2006)) Firms advertise/conduct R&D/introduce new products in order to create future barriers to entry through product differentiation Economies of Scope and the Boundaries of the firm (Panzar and Willig – 1981) Which firms can combine successfully? Firms with close potential rivals, price more competitively. What areas are related to each other in product market space? Why do profits increase for some mergers? Increased cost efficiency? economies of scale? Market power? Or are asset complementarities important especially for new product introduction? 3
4
Motivation - 2 Competition can affect merger success and motivation, profitability, and successful product introduction. We develop new industry groupings & new measures of industry competition. Old measures based on fixed industry classifications do not have much explanatory power. “Network” groupings. Industry Classifications are used everywhere. Asset pricing/ corporate finance benchmarks. Existing classifications in many cases do not “perform” that well. Existing SIC classifications have “Zero-One” fixed measures of groupings that rarely change. What we need is a new measure of “relatedness” that captures both within and across industry classifications. 4
5
Our contributions: Part of a 2 paper series Both papers rely on the following central ideas and methods: Economic Idea: Relatedness of products are fundamental to industries and notion of competition (Hotelling, Lancaster) Shaked Sutten –Product Differentiation is endogeneous and thus industries change over time. Compute degree of asset complementarities and similarity of every firm with each other -all pairs – both within and across industries: (5,000*5,000/2) X 9 years. New automated methodology to read 47,609+ firm 10-Ks, and extract product descriptions. Web crawling based in PERL, SEC Edgar website. APL based text parsing similarity matrix algorithms extract and process product descriptions for each 10-K. 5
6
Our contributions: Part of a 2 paper series Paper 1: Develop new measures of firm relatedness and new industry classifications. Jointly test importance of competition and endogenous product differentiation. Test theories of the endogeneous product market competition/ product differentiation (Shaked and Sutton (1987), Sutton (1991), Nevo (2000, 2001), Seim (2006). Paper 2: Examine merger likelihood and outcomes. Test the importance of asset complementarities to merger synergies and new product introduction. (Robinson 6
7
Real Data: Merger of Symantec (anti-virus) and Veritas (internet security ) 7 Conclude: Example of similar but different. Merger permits new products (different enough), but similar enough to permit integration. Very different WITHIN the same industry. Variable Industry groupings do not impose transitivity across firms – similar to Networks
8
General Dynamics (372) – Antheon (737)
9
Real Data: Merger of Disney and Pixar 9 Conclude: SIC codes miss the point, example of similar but different.
10
Related literature - 1 Endogeneous product market competition (Shaked and Sutton (1987), Sutton (1991)), economies of scale Panzar and Willig (1981). Changes in competition and industry formation should be analyzed jointly. Feasible with continuous similarity measure. Industries are best classified with VIC Network methods vs. Fixed Leontief classifications with transitive properties. 10
11
Related literature - 2 Why are we interested in relatedness? In the context of mergers we would like to distinguish between: (1.) Market power (Eckbo, Baker and Breshnahan(1985), Nevo (2000 RJE, Econometrica) (2.) Vertical Mergers (Fan and Goyal (2006), (3.) Economies of scale, Cost cutting. Or (4.) Synergies from Asset Complementarities (Berry and Waldfogel (2001, QJE), Rhodes-Kropf and Robinson (2008)). Relatedness: Merger literature empirically use SIC codes with 0-1 measures. [Kaplan and Weisbach (1992), Healy, Palepu and Ruback (1992), Andrade, Mitchell and Stafford (2001), Maksimovic, Phillips, and Prabhala (2008).] Open question: How related are firms within industries and across industries??? 11
12
Hypotheses about Industry Competition Key Industrial Organization Predictions: H1: More concentration, more profitability (Lack of strong link in many previous studies). H2: Limit pricing: Firms with “close” potential rivals price more competitively and thus have lower profits. H3: Endogenous Barriers to Entry: Firms actively engage in mechanisms to increase their product differentiation and reduce future product market competition. Need accurate measures of “closeness” and product market differentiation 12
13
Hypotheses about Merger Likelihood Key Industrial Organization Prescription: Prediction of Baker and Breshahan (1985), Nevo (2005) and others: Optimal merger partner for firm i is firm j (with rival k) when: High Own Cross Price Elasticity of Demand and Low Cross price elasticity of demand with Rivals: H1: Asset Complementarity: Firms are more likely to merge (and get better ex post merger outcomes) with other firms whose assets have high complementarity with their assets. H2: Competition and Differentiation from Rivals: Acquirers in competitive product markets should be more likely to choose targets that help them to increase product differentiation relative to their nearest ex-ante rivals. 13
14
Hypotheses about Ex Post Outcomes Profitability of new products: Profit function for new products: prob(success) *(p n – c n )*q n H3: Differentiation from rivals: Acquirers outcomes better with targets that differentiate products from rivals, higher price cost margin, (p n – c n ). H4: Synergy/Asset Complementarity: Outcomes better when T closer to A: (1.) higher prob(n) above, and (2.) more cost synergies from managerial skill: [(Cs a – Cs t )<0], where Cs i for acquirer, target. H5: H3, H4 stronger when – Unique products (patents) protect target technology and give potential for new product introduction. 14
15
Sample: 10-K population of firms All 10-Ks on SEC Edgar that have a valid link to COMPUSTAT tax number. Hand correct when tax numbers change. Must have a valid CRSP permno. Prior to matching with COMPUSTAT/CRSP, 49,000+ 10-Ks. After cleaning, 47,607 10-Ks from 1997 to 2005 (almost 5,000 /year). We use 10-Ks from 1996 only to compute starting values of lagged variables. Overall, we get 95% of the eligible COMPUSTAT/CRSP sample. Firms are excluded if they do not have a valid tax ID link. Coverage from 1997 to 2005 nearly uniform at 95%. 15
17
Document Similarity Take all words used in universe of 10-Ks in product description each year (87,385 in 1997). Exclude words (3027 of them in 1997) appearing in more than 5% of all 10-Ks. Form boolean vectors for each firm in each year (1=word used, 0=not used). Normalize to unit length. Dot products => pairwise product similarity. 17
18
Document Similarity Doc 1: “They sell cabinet products.” Doc 2: “They operate in the cabinet industry.” Step 1) Drop words "they", "the", "and", "in" (common words). Step 2) 5 elements: "sell" "operates", "cabinet", "products", "industry" P 1 = (1,0,1,1,0) P 2 = (0,1,1,0,1) Step 3) Normalize vector to have unit length of 1: V 1 = (.577,0,.577,.577,0) V 2 = (0,.577,.577,0,.577) Step 4) Compute document similarity V 1 V 2 =.33333 This dot product has a natural geometric interpretation: Document similarity is bounded between (0,1) 18
19
Geometric interpretation Suppose θ is the angle between a and b as shown in the image below with 0<= θ <= : Then: If orthogonal, Cos(θ) = 0, and firms are unrelated.
20
20 Conclude: Mergers are (1) far more similar than random firms, (2) heterogeneous in degree of similarity, and (3) still very highly similar even when in different SIC-2. Similarity Distrib. Range (0,100)
21
Why not just use SIC codes? Mergers in 2005 in different SIC-2 Conclude: SIC codes are informative but do not fully describe similarity nor product market competition. 21
22
Examples: T+A shared words Conclude: common words indeed related to product offerings. 22
23
Text Product Based Industry Measures of Competition First fix industry groups. Industry groups defined by maximizing within group similarity. From groups compute: Similarity Concentration Index: Total Summed Similarity: 3.Average Similarity index: 4.Sales 10K based Herfindahl: 5.Sales 10K based C4 6.High Potential Entry Indicator 7.Firm level: Similarity with respect to “10 nearest” neighbors. 23
24
T5: Reality Check: Document Similarity “The Profitability of Differentiated Products” 24 Conclude: Most basic I/O theoretical prediction: product differentiation is profitable. Huge significance, equal in importance to value/growth variables.
25
Future Product Differentiation and Advertising/R&D Dependent variable: change in differentiation 25 Conclude: Firms invest and advertise to generate ex-post product differentiation and hence ex-post profitability.
26
T2: New Industry Classifications 26
27
Industry Classifications Adjusted RSQ of variable on industry “dummies” 27 Conclude: Industry definitions constructed from 10Ks are better and more flexible than SIC/NAICS (see companion paper). For merger paper: We use 10-K based measures b/c they better explain competitiveness and offer flexibility. Flexibility in firm location measurement is pivotal in examining mergers. Dependent VariableSIC3NAICS4 10-K based (constrain) 10-K based (generalize) Operating Inc/Sales 28.3% 28.5% 33.1% 38.9% Advertising/Sales4.5%6.6% 7.3% 9.4% Market Beta 29.2% 30.2% 36.5% 45.5%
28
T3: New Industry Classifications 28 Regress Firm characteristic on Industry Dummies/Averages
29
T7: 10K Based Competition and Profitability 29 Conclude: New Industry Definitions work well in explaining profitability.
30
T8: Reality Check: Normal SIC codes 30 Conclude: SIC codes and NAICs codes don’t perform very well.
31
T9: Sutton: Endogenous Competition 31 Conclude: Our new competition measures pick up incentives to differentiate yourself – endogenous competition.
32
Conclusions: New Product Based Industries Text-based analysis of product descriptions produces improved measures of: (1) Industry competition (2) Relatedness between firms both within and across industries. (3) These new measures allow tests of theories of economies of scope and endogenous barriers to entry, and tests of merger pair relatedness Competition and product differentiation. We can use these new industries to examine many finance related questions as well. 32
33
Hypotheses about Merger Likelihood Key Industrial Organization Prescription: Prediction of Baker and Breshahan (1985), Nevo (2005) and others: Optimal merger partner for firm i is firm j (with rival k) when: High Own Cross Price Elasticity of Demand and Low Cross price elasticity of demand with Rivals: H1: Asset Complementarity: Firms are more likely to merge with other firms whose assets have high complementarity with their assets. H2: Competition and Differentiation from Rivals: Acquirers in competitive product markets should be more likely to choose targets that help them to increase product differentiation. H2b: Firms with complementary assets are more likely to introduce new products post merger to increase diff. 33
34
Database of Restructuring Transactions SDC Platinum. We consider mergers and acquisition of assets transactions. Target and acquirer must also both have a valid link to the machine readable firms database. Final sample of 5,643 restructuring transactions from 1995 to 2005. 34
35
Text Measures of Complementarities and Competition 1.Asset Complementarity (Own similarity): Pairwise similarity b/t target and acquirer using text similarity. 2.Similarity between T and T’s closest rivals (ranked in terms of text similarity). Intensity of Target product market competition. 3.Similarity between A and A’s closest rivals. Intensity of Acquirer product market competition. 4.Similarity between T and A’s closest rivals. Comparing to above, permits computation of how much the acquirer’s product market competition. 5.Number or % of words in prod description having word root “patent” or “Trademark” A more direct measure of unique assets / potential for new products. 35
36
Nested Logit with spreading sorts – all 5000 firms
37
T8: Nested Logit Conclude: Product similarity is most important determinant of pairings. In competitive industries, also dissimilarity to rivals
38
T9: Announcement Returns (1)Combined firm returns larger when acquirer in comp. product market and when target is more unique. (2)Especially large when target is dissimilar to acquirer’s near rivals and when pairwise similarity is larger. (3)Results also larger when patent-proxy for unique assets is higher. 38
39
Table 10: Long-term Real Outcomes 39 Conclude: acquirers in competitive product markets experience higher profitability and sales growth when similar and gain in differentiation. Results stronger as horizon is lengthened.
40
Table 11: Synergies Growth in Product Descriptions Conclude: Acquirer product market competitiveness very related to product desc. growth. Support for post-merger real gains being related to synergies and unique assets. 40
41
Table 12: Economic Magnitude (Returns+Profitability) Conclude: Economic impact on announcement returns modest, stronger on fundamentals, especially sales growth and growth in product descriptions. 41
42
Merger paper conclusions “Synergies and competition matter” Merger pair similarity – while high - is quite heterogeneous ** Best mergers with higher ex post cash flows and new product introductions are ones (1) with similar acquirer and target (2) with targets that are further away from A’s nearest rivals (3) that have unique, hard to replicate assets (patents) that make potential new products. “Similar but Different”. 42
Similar presentations
© 2025 SlidePlayer.com Inc.
All rights reserved.