The Market Valuation of Innovation: The Case of Indian Manufacturing Sunil Kanwar Delhi School of Economics Bronwyn H. Hall UC Berkeley, NBER, IFS, and NIESR 1. Motivation Innovation prime motive force behind economic growth Firms spend large amounts of scare resources on innovative activities Desirable to know whether financial markets value innovation-intensive firms differentially Persuasive evidence that developed country stock markets value innovative activity by firms Can we expect the same in less developed economies? Major reason for incredulity is the fact that predominant share of intellectual capital is generated in a handful of developed economies A few developing countries do some innovation (Bogliacino et al. 2012) – process patents, utility models, smaller innovations. Often for imitation and diffusion, but generate profits and hence market value Hence, stock market’s valuation of innovation also relevant in developing country context: • Are more innovative firms valued more highly than less innovative ones? • Is market valuation responsive to the quality of innovation spending? • Is market value responsive to market risk? • Does the market value-innovation relation vary across industries; and if so, how? We explore such issues in the context of manufacturing industries in India. 2. Prior literature Very informative, but mostly pertains to developed countries: Griliches (1981); Bloom and Van Reenen (2002); Hall, Jaffe and Trajtenberg (2005); Greenhalgh and Rogers (2006); Griffiths and Webster (2006); Hall and Oriani (2006); exception Chadha and Oriani (2010) Our study broad-bases the available evidence by providing further evidence on a developing country, namely India. Number of distinguishing features, including the context - mostly no product patents; few process patents, limited to certain industries; utility models never an option Far from obvious that stock market would value such innovation as does occur 3. The Model (1) π = π(πΎπ + π½πΎπΎ + πΎπΎππΌ + πΏπ)π π ln πΎπ ππ‘ πΎπΎ = π ln(πΎπ )ππ‘ + π ln 1 + π½ πΎπ ππ‘ 4. Sample and Variables Firm-level data for Indian manufacturing sector (‘Prowess‘; CMIE) Sample: 380 firms, 3494 observations, over 2001-2010, covering 22 industries (mostly 2-digit, some 3-digit levels): Auto ancillaries, automobiles, cement, chemicals, (other) construction material, (other) consumer goods, domestic appliances, drugs and pharmaceuticals, electrical machinery, electronics, food and agro-products, gems and jewellery, glass and glassware, leather and leather products, metals, non-electrical machinery, paper and paper products, personal care, petroleum, plastics and plastic products, rubber and rubber products, and textiles and textile products. Variables: • Market Value (V): equity + debt • Physical capital (Kp): net fixed assets • Knowledge capital (KK): capitalized value of R&D expenditure; perpetual inventory method (15% depreciation rate) π²π²π = π − π½ π²π² π−π + πΉπ«π • Other intangible capital (KOI): capitalized value of advertising expenditure; perpetual inventory method (30% depreciation rate) • DUM (adv = 0); invariably insignificant • Quality of capital (S): real profit aftertax, purged of knowledge capital and other intangible capital (and time dummies) Table 1 Sample statistics (3,494 observations on 380 firms, 2001-2010) Variable π½ π²π· π²π² π²π· π²′π² π²π· π²πΆπ° π²π· πΊ π²π· π²π· (M rupees) D (π²πΆπ° = 0) Mean Median Standard Deviation Minimum Maximum Share Variance Within†† 4.36 0.12 0.17 0.13 0.00 1140.7† 42.4% 3.23 0.05 0.06 0.00 –0.03 1110.8 3.43 0.20 0.32 0.42 0.31 1.71 0.16 0.00 0.00 0.00 –1.94 2.30 19.82 2.72 5.39 7.38 2.02 1,500,007 0.265 0.159 0.181 0.078 0.427 0.050 0.052 Correlation Matrix π²′π² π²π· π²πΆπ° π²π· πΊ π²π· π₯π§ π²π· 1 0.004 1 π₯π§β‘ ( π½ π²π· ) π²π² π²π· π₯π§β‘ ( π½ π²π· ) π²π² π²π· π²′π² π²π· π²πΆπ° π²π· πΊ π²π· π₯π§ π²π· 1 0.330 0.338 0.302 0.391 –0.024 1 0.906 0.112 –0.004 –0.131 1 0.077 –0.140 –0.045 1 –0.001 –0.039 Definitions: π½ = Market value = Equity + Book Debt π²π· = Net fixed assets π²π² = Knowledge capital at 15% depreciation π²′π² = Knowledge capital at 30% depreciation π²πΆπ° = Advertising capital at 30% depreciation πΊ = Quality of capital = Profit surprise † Geometric mean †† Within-firm variance as a proportion of total variance (controlling for overall year means) Table 2 Nonlinear Regressions Dependent Variable: ln (π½ π²π· ) (1) NLLS (2) NLLS (3) NLLS (4) NLLS (5) NLLS, lag RHS (6) NLIV 2.275*** (0.389) [0.164] *** (0.018) 2.009*** (0.375) [0.140] *** (0.018) 0.988*** (0.224) [0.058] *** (0.009) –0.028 (0.057) ln π²π· 0.020 (0.015) 0.020 (0.015) 1.790*** (0.330) [0.134] *** (0.018) 0.817*** (0.183) [0.052] *** (0.008) –0.037 (0.053) 0.508*** (0.103) 0.012 (0.014) 1.473*** (0.336) [0.114] *** (0.019) 0.974*** (0.191) [0.059] *** (0.008) –0.004 (0.056) 0.464*** (0.101) 0.015 (0.015) 1.661*** (0.324) [0.126] *** (0.018) 0.815*** (0.185) [0.051] *** (0.008) –0.031 (0.055) 0.527*** (0.095) 0.013 (0.014) 1.764*** (0.329) [0.137] *** (0.018) 0.640*** (0.145) [0.044] *** (0.008) –0.083 (0.053) 0.709*** (0.031) 0.012 (0.016) Industry d.v. Year FEs πΉπ Standard Error Panel D-W Observations Firms No Yes 0.199 0.608 No Yes 0.267 0.582 No Yes 0.318 0.561 Yes Yes 0.383 0.536 No Yes 0.286 0.571 No Yes 0.270 0.579 0.266 3494 380 0.285 3494 380 0.316 3494 380 0.345 3494 380 0.360 3114 380 0.346 3114 380 Regressor π²π² π²π· π²πΆπ° π²π· D (π²πΆπ° = π) πΊ π²π· Note: Robust standard errors clustered on firm in parentheses below each coefficient Elasticity at the means in square brackets, with its standard error below it In column (5), all right hand side (RHS) variables are lagged one year In column (6), the instruments are the right hand side variables lagged one year *** ** , and * denote significance at the 1%, 5% and 10% levels, for a two-tail test Table 3 Linear Regressions Dependent Variable: ln (π½ π²π· ) (1) OLS (2) OLS (3) OLS (4) OLS (5) OLS, lag RHS (6) IV 1.025*** (0.136) [0.128] *** (0.017) 0.939*** (0.129) [0.117] *** (0.016) 0.368*** (0.062) [0.049] *** (0.008) –0.079 (0.051) ln π²π· 0.006 (0.015) 0.007 (0.015) 0.943*** (0.116) [0.117] *** (0.014) 0.368*** (0.051) [0.049] *** (0.007) –0.079 (0.047) 0.704*** (0.076) 0.007 (0.013) 0.790*** (0.118) [0.098] *** (0.015) 0.393*** (0.055) [0.053] *** (0.007) –0.054 (0.047) 0.633*** (0.071) 0.010 (0.015) 0.912*** (0.118) [0.114] *** (0.015) 0.392*** (0.055) [0.053] *** (0.007) –0.076 (0.049) 0.686*** (0.074) 0.009 (0.014) 0.964*** (0.120) [0.118] *** (0.015) 0.385*** (0.053) [0.051] *** (0.007) 0.039 (0.050) 0.500*** (0.100) 0.011 (0.014) Industry d.v. Year Fes πΉπ Standard Error Panel D-W Observations Firms No Yes No Yes No Yes Yes Yes No Yes No Yes 0.177 0.616 0.238 0.593 0.339 0.552 0.396 0.530 0.301 0.565 0.318 0.559 0.265 0.282 0.364 0.385 0.413 0.335 3494 380 3494 380 3494 380 3494 380 3114 380 3114 380 Regressor π²π² π²π· π²πΆπ° π²π· D (π²πΆπ° = π) πΊ π²π· Note: Robust standard errors clustered on firm in parentheses Elasticity at the means in square brackets, with its standard error below it In column (5), all right hand side variables are lagged one year In column (6), the instruments are the right hand side variables lagged one year *** ** , and * denote significance at the 1%, 5% and 10% levels, for a two-tail test Table 4 Regressions with Firm Effects Dependent Variable: ln (π½ π²π· ) Regressor (1) OLS with industry fixed effects (2) OLS with random firm effects (3) OLS with firm fixed effects (4) OLS with firm fixed effects (5) GMM-SYS With lag 2+ instruments 0.428*** (0.140) 0.250*** (0.064) 0.352*** (0.051) –0.158*** (0.042) 0.484*** (0.023) 0.315*** (0.087) 0.192*** (0.054) 0.239*** (0.045) –0.182*** (0.032) 0.706*** (0.036) 0.302*** (0.071) 0.146*** (0.028) 0.251*** (0.056) –0.005 (0.014) 1.026*** (0.221) 0.495*** (0.091) Yes 3096 379 Lagged dep. Var. π²π² π²π· π²πΆπ° π²π· πΊ π²π· ln π²π· 0.785*** (0.117) 0.413*** (0.054) 0.631*** (0.071) 0.011 (0.015) 0.688*** (0.117) 0.353*** (0.048) 0.428*** (0.053) –0.047*** (0.018) Yes 3494 380 0.395 0.530 0.566 Yes 3494 380 0.372 0.347 0.602 Yes 3494 380 0.381 0.321 0.737 0.609*** (0.172) 0.372*** (0.107) Yes 3114 380 0.522 0.271 0.662 69.9*** 29.0*** 30.1*** 1.8*** LR coef: π²π² π²π· LR coef: π²πΆπ° π²π· Year Fes Observations Firms πΉπ Std. Err. Within Share of variance across firms AR(1) t-test Hansen test (df) AR(1) test (p-value) AR(2) test (p-value) 255.1** (206) –10.7*** (0.000) 2.0** (0.050) Note: Robust standard errors clustered on firm in parentheses. Hausman test for correlated effects: πππ = 137.0 (π-value = 0.000). Instruments in col. (5) are lags 2 and earlier (level and differenced) of the dependent and independent variables. *** ** , and * denote significance at the 1%, 5% and 10% levels, for a two-tail test. Table B1 GMM-SYS regressions Dependent Variable: ln (π½ π²π· ) Regressor (1) (2) GMM-SYS with lag 2+ instruments GMM-SYS with lag 3+ instruments (3) (4) Estimation Method GMM-SYS GMM-SYS with lag 3/4 with lag 2+ instruments instruments Lagged dep. var. π²π² π²π· π²πΆπ° π²π· πΊ π²π· ln π²π· 0.991*** (0.174) 0.336*** (0.055) 0.793*** (0.115) 0.002 (0.035) 0.711*** (0.130) 0.287*** (0.073) 0.802*** (0.149) –0.024 (0.035) 0.668*** (0.144) 0.287*** (0.079) 0.821*** (0.153) 0.018 (0.035) LR coef: π²π² π²π· LR coef: π²πΆπ° π²π· Observations Firms Hansen test (df) AR(1) test (p-value) AR(2) test (p-value) 3494 380 279.3*** (216) –6.7*** (0.000) –1.0 (0.328) 3494 380 224.1*** (184) –6.9*** (0.000) –0.9 (0.357) 3494 380 165.1*** (96) –6.8*** (0.000) –0.9 (0.365) (5) (6) GMM-SYS with lag 3+ instruments GMM-SYS with lag 3/4 instruments 0.706*** (0.036) 0.302*** (0.071) 0.146*** (0.028) 0.251*** (0.056) –0.005 (0.014) 0.694*** (0.039) 0.326*** (0.094) 0.169*** (0.039) 0.181*** (0.073) –0.011 (0.016) 0.677*** (0.045) 0.238*** (0.110) 0.165*** (0.039) 0.203*** (0.085) 0.001 (0.019) 1.026*** (0.221) 0.495*** (0.091) 1.067*** (0.286) 0.553*** (0.133) 0.735*** (0.312) 0.510*** (0.125) 3096 379 255.1*** (206) –10.7*** (0.000) 2.0** (0.050) 3096 379 220.2*** (170) –10.4*** (0.000) 1.9* (0.065) 3096 379 155.1*** (95) –9.9*** (0.000) 1.9* (0.065) Note: Robust standard errors in parentheses Instruments are lags (level and differenced) of dependent and independent variables – In columns (1) and (4) they include lag 2 and earlier values, In columns (2) and (5) lag 3 and earlier values, and In columns (3) and (6) lags 3 and 4 only. *** ** , and * denote significance at the 1%, 5% and 10% levels, for a two-tail test Economic Significance: Elasticity w.r.t Knowledge Capital: 0.14 (Hall-Oriani 2006) France: 24% Germany: 22% Italy: 18% US: 42% UK: 24% Semi-elasticity w.r.t Knowledge Capital: 1.75 France: 0.66 Germany: 0.56 Italy: 0.94 US: 0.80 UK: 1.92 Evidence of under-investment Sectoral Heterogeneity (Pavitt 1984) 1: supplier dominated - leather, textiles & textile products, rubber, gems & jewellery 2: production intensive (scale intensive) – automobiles, cement, (other) construction material, (other) consumer goods, domestic appliances, food & agro-products, glass & glassware, metals & metal products, personal care, paper & paper products 3: production intensive (specialised suppliers) - automobile ancillaries, non-electrical machinery 4: science-based - chemicals, drugs & pharmaceuticals, electrical machinery, electronics, petroleum products, & plastic products Table 5 Nonlinear Regressions by Pavitt Sector Dependent Variable: ln (π½ π²π· ) (1) Regressor Supplierdominated (3) Pavitt Sector ScaleSpecializedintensive supplier π²π² π²π· 4.24 (3.45) [0.102] (0.063) 2.74*** (0.89) [0.097] *** (0.022) 0.41 (0.46) 0.08 (0.06) 1.80*** (0.65) [0.093] *** (0.025) 0.73*** (0.18) [0.077] *** (0.013) 0.50*** (0.18) 0.03 (0.02) 1.28*** (0.38) [0.152] *** (0.034) 0.83 (0.56) [0.030] (0.019) 0.48*** (0.13) –0.10*** (0.04) 1.73*** (0.50) [0.155] *** (0.034) 1.51** (0.64) [0.055] ** (0.016) 0.48*** (0.18) 0.03 (0.02) Yes 0.450 0.464 0.350 316 32 Yes 0.352 0.549 0.329 1,235 134 Yes 0.357 0.541 0.343 690 78 Yes 0.289 0.581 0.294 1,253 136 π²πΆπ° π²π· πΊ π²π· ln π²π· Year Fes πΉπ Std. Error Panel D-W Observations Firms (2) (4) Sciencebased Notes: Robust standard errors clustered on firm in parentheses Elasticity at means in square brackets, with standard error below it *** ** , and * denote significance at 1%, 5% and 10% levels, for twotail test Table A1 Observations by Industry and Pavitt sector Pavitt sector Industry Obs . Firms (i) supplier-dominated (i) supplier-dominated (i) supplier-dominated (i) supplier-dominated (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (ii) scale-intensive (iii) specialized supplier (iii) specialized supplier (iv) science-based (iv) science-based (iv) science-based (iv) science-based (iv) science-based (iv) science-based Total Gems and jewellery Leather products Rubber products Textiles & textile products Domestic appliances Automobiles Cement Food & agri. products Glass and glassware Metals & metal products Other consumer goods Other construction products Paper & paper products Personal care Automobile ancillaries Nonelectrical machinery Chemicals Electrical machinery Electronics Petroleum products Drugs & pharmaceuticals Plastic products 7 30 20 259 60 101 140 352 25 217 30 171 129 10 419 271 600 129 68 64 268 124 3494 1 3 2 26 7 12 14 39 3 22 3 18 13 3 43 35 62 15 8 7 31 13 380 Mean R&D growth 0.23% 3.21% 0.91% 1.42% 1.29% 1.59% 1.33% 0.89% –1.97% 0.51% –2.80% 0.62% 1.79% –2.92% 1.58% 2.59% 0.79% 2.36% 1.39% –0.36% 2.72% 1.21% 1.32% Mean ADV growth 4.81% 0.60% –0.88% –1.08% 4.11% 0.96% 0.78% 0.51% 7.93% 1.03% –0.17% 2.71% 0.08% 1.46% 1.18% 1.94% 0.08% 2.32% 0.83% 2.58% 1.76% 0.09% 1.00% Table B2 GARCH Model for log(Sales) Parameter (1) (2) (3) (4) (5) Parameters of Equation 8(a) π·π 0.999*** (0.002) 1.000*** (0.002) 1.002*** (0.001) 1.002*** (0.001) 1.001*** (0.001) Parameters of Equation 8(c) πΆπ πΆπ π π π π –3.070*** (0.190) –0.064*** (0.026) 0.904*** (0.348) –0.040 (0.049) –2.980*** (0.150) –0.078*** (0.021) 0.636*** (0.075) πΈπ –5.030*** (0.480) –0.321*** (0.072) –0.050*** (0.003) –4.600*** (0.580) –0.384*** (0.086) –0.049*** (0.003) 1.063*** (0.008) 1.029*** (0.034) 0.005 (0.004) 1.075*** (0.008) In eq (8a) No 2752 1172.7 In eq (8a) No 2752 1173.2 In eq (8a) In eq (8c) 2752 1216.5 πΈπ Year FEs Industry FEs Observations Log-likelihood In eq (8a) No 2752 466.9 In eq (8a) No 2752 466.5 –0.235*** (0.076) –0.056*** (0.003) Equations (8a)-(8c) in the text are reproduced below for convenience: πππ = ππ + π·π ππ,π−π + πΊππ πΊππ ~ π±(π, πππ ) πππ = ππ±π© ππ + πΆπ πππ + (π π + π π πππ )(πΊπ,π−π )π + (πΈπ + πΈπ πππ )ππ,π−π where π is log(sales), π is log(π²π· ), π is the industry to which the πππ firm belongs, ππ are the year dummies, and ππ are the industry dummies. Table 6 Market Value Regressions Allowing for Uncertainty Dependent Variable: ln (π½ π²π· ) Regressor (1) (2) (3) (4) π²π² π²π· 0.959*** (0.110) 0.380*** (0.050) 0.945*** (0.110) 0.374*** (0.050) 5.790** (2.640) 0.925*** (0.120) 0.378*** (0.050) –5.540 (8.310) 203.6 (144.0) 1.227*** (0.220) 0.376*** (0.050) 8.300*** (3.180) π²πΆπ° π²π· π† ππ 0.727*** (0.080) 0.009 (0.014) 0.716*** (0.078) 0.013 (0.014) 0.713*** (0.078) 0.012 (0.014) –13.240 (8.390) 0.716*** (0.078) 0.014 (0.014) Yes 0.329 0.553 0.351 3114 380 Yes 0.335 0.551 0.329 3114 380 Yes 0.337 0.550 0.345 3114 380 Yes 0.337 0.550 0.294 3114 380 π x (π²π² π²π· ) πΊ π²π· ln π²π· Year Fes πΉπ Std. Error Panel D-W Observations Firms Notes: OLS regressions. Robust standard errors clustered on firm in parentheses † Industry sales variance estimated as shown in Appendix B, Table B3. *** ** , and * denote significance at 1%, 5% and 10% levels, for two-tail test Conclusions Where most firms do not patent, or have utility models, we find that: • Stock market places greater value on more innovative firms, ceteris paribus • Rate of return appears to be larger than that in developed countries, excepting UK • • • • Depreciation rate too high? Probably not Firms underinvest in R&D. Probably R&D-intensive firms valued more for option value of R&D programmes Market value-innovation relation appears to vary between supplier-dominated & other industry groups, but few firms in former group, & differences insignificant.