EssLait final conference ISTAT Examples of empirical applications and findings based on the ESSLait Micro Moments Database Martin Falk Rome, 2013 October 17 EssLait final conference ISTAT
Examples of possible research questions: motivation ESSLait Micro Moments Database data has a great potential for economic modeling studies Examples of possible research questions: Drivers of productivity Determinants of exports Labour demand and technological innovations/Labour demand and ICT usage/ecommerce usage Skill biased technological change Determinants of technological innovations/ICT and E-commerce adoption/Knowledge production function Impact of public R&D/innovation funding Limitations: FILTER question in the CIS data (affecting innovation input variables) and EC data Release of the EU KLEMS industry data was a great success (also WIOD database) e. g. Michaels, Natraj and Van Reenen (2013) RESTAT
Disadvantages of using aggregate data motivation Advantages of using aggregate data (industry level X firm size data ) sample attrition and missing values are less of a problem modeling of dynamics (persistence and lagged effects) Inter-industry and inter-country spillover effects (innovation and /or ICT activities of the neighbouring country) availability and access Disadvantages of using aggregate data causal effects are difficult to identify aggregation bias (e.g. large degree of within industry/firm size heterogeneity) Variables based on filter questions are difficult to use (Innovation funding, R&D activities) => work with the sample of innovative firms
Determinants of exports (export share/foreign market presence) outline Determinants of exports (export share/foreign market presence) Role of firm size, ICT, innovations and skills Employment effects of technological innovations Are the impacts of the determinants different for small firms and for low and high growth industries/firm size? Role of lagged effects Impact of broadband usage on labour demand across quartiles Impact of technological innovations and ICT/e-commerce on the demand for skilled and unskilled workers Data: PS-CIS-EC (no annual variation for CIS data =>averaged over two years ) Sums: PS-CIS-EC by industry (ConsG, Distr, Elecom, FinBu, IntmdG, InvesG) and by four size classes Quartiles of growth rates: PS-CIS-EC by industry (ConsG, Distr, Elecom, FinBu, IntmdG, InvesG)
MOB INPS INPD EXPMKT e b a + = HKPCT + time effects ikt ijkt ijk 4 3 2 exporting Exporting and technological innovations t=2002,2004, 2006, 2008 and 2010; j=1,...,6 ConsG, Distr, Elecom, FinBu, IntmdG, InvesG; K=1,..4 size classes; i=1,..,12 countries EXPMKT= foreign market presence or NX_NQ export to output ratio INPD % of firms with product innovations (alternatively new market products; turnover share due to new market products) INPS: % of firms with process innovations HKPCT: tertiary graduates share in % MOB : % of firms with mobile access to internet Time effects Estimation methods: FE, OLS long differences, dynamic panel data m. ikt ijkt ijk MOB INPS INPD EXPMKT e b a + = 4 3 2 1 HKPCT + time effects
exporting Theoretical background Innovative firms are more likely to export Likelihood of exporting increase with firm size Firms with higher productivity and skills are more likely to export and exhibit a larger export share (self selection into exporting) Causality between innovation and exports can go in both ways (=> dynamic panel data models) Findings Export share increase with firm size Large firms account for the largest proportion of exports (more pronounced in services) Innovation activities and exporting are significantly positively related Positive impact of human capital on the export share Export share is increasing over time
Export share by firm size class and broad industry groups in % exporting Export share by firm size class and broad industry groups in % 20 35 17 32 22 25 41 27 42 54 33 51 66 67 40 45 62 10 30 50 60 70 80 intermediate goods Investment Consumer Electrical machinery communication services total manuf. Excluding post & telecommunication - 19 49 249 > 250
Share of exports in total exports by firm size exporting Share of exports in total exports by firm size Note: Ireland is excluded
Percentage of firms that sells goods or services abroad exporting Percentage of firms that sells goods or services abroad 10 20 30 40 50 60 70 80 90 100 IntmdG InvesG Elecom ConsG Distr FinBu - 19 49 249 > 250
exporting Correlation between export share and new market products (manufacturing) corr: 0.46 p-value: 0.00 (# of obs 702)
exporting Correlation between foreign market presence and new market products (man + ser) corr: 0.70 p-value: 0.00
exporting Fixed effects model of the impact of innovations on foreign market presence (manufacturing + services) (i) (ii) (iii) (iv) % of firms with Coef. t producti nnovations 0.43 *** 3.81 new market products 0.33 5.50 turnover due to market products 0.45 3.74 0.29 3.30 process innovations 0.21 1.60 0.24 3.03 0.53 6.19 0.38 4.23 organisational innovations -0.01 -0.18 0.06 0.78 0.08 1.08 mobile access to internet 0.15 4.20 0.11 3.14 0.13 3.52 0.25 5.23 tertiary graduates share 0.12 0.83 time effects yes constant 0.30 13.29 0.36 14.99 0.31 12.14 0.35 9.43 R-sq: within 0.47 0.27 Number of obs 745 703 576 Number of groups 214 190 144 *** *** *** *** Time effects are included (2004, 2006, 2008 and 2010) with 2002 as the reference group. Cluster adjusted standard errors across country industry and size class
exporting Fixed effects model of the impact of innovations on the export share (manufacturing + services excl IE) coef t product innovations 0.05 0.62 new market products 0.17 *** 3.06 process innovations -0.09 -1.05 -0.11 -1.45 tertiary graduates share 2.93 0.13 ** 2.35 period_03_04 ref (01-02) 0.02 * 1.71 2.15 period_05_06 0.04 3.08 3.23 period_07_08 3.29 3.26 period_09_10 4.17 3.46 constant 0.22 7.00 7.71 Number of obs 765 Number of groups 192 Number of countries 8 Time effects are included (2004, 2006, 2008 and 2010) with 2002 as the reference group. Cluster adjusted standard errors across country industry and size class
ORGIN INPS INPD PQ E PAY NQ e b a + = ) / ln(( ln( ln time effects + Labour demand Key question: Impact of technological innovations on labour demand Labour demand equation: i=1,..,12 countries, j=1,.,6 industries, k=1,..4, size classes, t =2002, 2003-2004, 2005-2006, 2007-2008 and 2009-2010 E: employment (full time equivalents) NQ/PQ: output deflated by the output price index (PAY/E)/PQ: wage bill per employee deflated by the output price index INPD, INPS, ORGIN: product, process & organizational innovations Time effects Estimation: FE model, long difference OLS, quantile reg. ;dynamic panel data methods ikt ijkt k ijk ORGIN INPS INPD PQ E PAY NQ e b a + = 5 4 3 2 1 ) / ln(( ln( ln time effects +
Labour demand Correlation between employment change and change in the percentage of firms with new market products current impact lagged impact 1 1 employment change t/t-2 in % employment change t/t-2 in % -1 -1 -2 -2 -1 -.5 .5 -1 -.5 .5 change in new market prod. t/t-2 in p. p change in new market prod. t-2/t-4 in p. p
Labour demand Static fixed effects model Wages and output shows the expected sign Positive impact of product and process innovations New market products and organisational changes not significant Larger impact of product innovations among small firms Higher employment losses among small firms during the economic and financial crises (ceteris paribus) Dynamic model with lagged effects of technological innovations Lagged new market products (t-2) are now significant at the 5 percent level => takes time to realize its full impact Quantile regressions of the determinants of labour demand Impact of product innovations decreases along the distribution of the conditional employment growth rate
Labour demand Fixed effects model of the impact of technological innovations in labour demand (man + ser) Coef. t log output c.p 0.37 *** 10.25 9.71 log real wages c.p. -0.58 -5.26 -0.60 -5.27 new products 0.51 3.43 new market products 0.06 0.64 process innovations 0.19 1.59 0.48 2.88 organisational change -0.10 -0.73 0.00 0.03 year_2003_2004 -0.01 -1.21 -1.12 year_2005_2006 -0.02 -0.82 -0.55 year_2007_2008 0.01 0.17 0.22 year_2009_2010 -0.04 -1.10 -0.03 -0.86 constant 6.33 10.11 6.44 9.83 R-sq: within 0.55 0.54 Number of obs 1043 Number of groups 262
Labour demand Fixed effects model of the impact of technological innovations in labour demand 10-19 employees (man + ser) ------------ Coef. t log output c.p 0.44 *** 7.50 0.47 6.82 log real wages -0.66 -3.78 -0.71 -3.69 new products c.p 0.76 ** 2.53 new market products -0.16 -0.55 process innovations 0.42 1.20 1.16 2.99 organisational change -0.37 0.04 0.08 year_2003_2004 -0.08 -2.94 -2.95 year_2005_2006 -0.01 -0.29 -0.18 year_2007_2008 -0.05 -1.18 -1.02 year_2009_2010 -0.09 -2.29 -0.06 -1.42 constant 5.03 5.29 4.88 4.54 R-sq: within 0.66 0.65 Number of obs 258 Number of groups 65
Labour demand Quantile regression estimates of the determinants of labour demand (measured in first differences) (man + ser) Coefficient of product innovations becomes insignificant at higher quantiles
Labour demand System GMM estimates of the of the impact of technological innovations on labour demand (man + ser) coef t coef t log employment t-2 0.51 *** 4.87 0.21 * 1.85 log real wages c.p t -0.86 *** -3.98 -0.94 *** -5.06 Log output c.p t 0.46 *** 6.26 0.67 *** 7.77 new market products t 0.18 0.92 new market products t-2 0.52 ** 2.31 process innovations t 0.93 *** 2.91 process innovations t-2 0.82 *** 3.25 year_2005_2006 -0.04 -0.93 -0.09 *** -2.84 year_2007_2008 0.00 -0.05 -0.06 ** -1.99 year_2009_2010 -0.07 -1.59 -0.04 -1.20 constant 0.67 0.76 0.71 0.73 R-sq 0.96 0.96 Number of obs 860 860 Number of groups 261 261
Labour demand using the Moments database and quartiles Moments database using quantiles q1 q2 q3 q4 Positive impact of broadband penetration on employment change across quartiles U-shaped relationship between broadband penetration and change in labour demand Coefficient of wages is very volatile across quartiles (?)
Q1 Q2 Q3 Q4 coef. t coef. t coef. t coef. t 0.07 *** 2.60 -0.21 *** Labour demand Determinants of employment change: Robust regression estimates based on first differences for four quartiles (based on changes) Q1 Q2 Q3 Q4 coef. t coef. t coef. t coef. t D.lnWmean 0.07 *** 2.60 -0.21 *** -4.40 -0.19 *** -3.79 0.17 ** 2.46 D.lnQmean 0.28 *** 9.95 0.34 *** 16.05 0.32 *** 13.01 0.38 *** 9.93 lbroadpct 0.04 ** 2.03 0.01 0.86 0.02 ** 2.41 0.09 ** 2.30 constant -0.09 *** -9.44 -0.02 *** -5.37 0.02 *** 2.94 0.00 0.07 # of obs 463 403 408 354 Source: moments database
Demand for skilled workers Key question: Impact of technological innovations on the demand for skills (skill biased tech. change or skill biased organisation change) Chennells & Van Reenen 2002; Caroli & Van Reenen 2001) Share equation: i=1,..,8 countries, j=1,.,6 industries, k=1,..4, size classes, t =2002, 2003-2004, 2005-2006, 2007-2008 and 2009-2010 HKPCT: share of workers with tertiary degree ISCED 5 +6 NQ/PQ: output deflated by the output price index (PAY/E)/PQ: wage bill per employee deflated by the output price index INPD, INPS, ORGIN: product, process & organizational innovations WEB : having a website extensions: separate estimates for different size classes ijkt k ijk ORGIN INPS INPD PQ E PAY NQ HKPCT e b a + = 5 4 3 2 1 ) / ln(( ln( b WEB + ijkt 6k
Demand for skilled workers Findings Positive and significant impact of product innovations However, the effect is only significant for small firms Positive and significant impact of process innovations in large firms For the sample of small firms: Having a website increase the relative demand for skilled workers Increase in the share of skilled workers in large and medium sized firms over time but not in small firms
Demand for skilled workers
Demand for skilled workers Fixed effects model of the share of workers with a tertiary degree total 10>=emp<20 emp>250 coef t log output c.p 0.02 ** 2.10 1.13 -0.01 -0.93 log real wages c.p 0.00 0.13 0.12 0.03 0.45 new market products 0.11 *** 3.56 0.32 6.85 1.16 process innovations 0.04 1.52 -0.08 0.08 2.47 organisational change -0.02 -0.89 -0.14 -2.71 -0.07 -1.49 website 0.06 2.34 0.10 2.33 -0.05 -1.16 period_03_04 (ref 01_02) 0.01 3.50 3.11 period_05_06 3.88 0.21 4.92 period_07_08 5.32 0.97 4.41 period_09_10 5.36 0.95 3.97 constant -0.30 -2.14 -0.22 -1.50 0.33 1.29 R2 within 0.39 0.69 0.37 Number of obs 720 180 Number of groups 192 48
Conclusions Impact of technological innovations on exports, labour and skills Large degree of heterogeneity in the determinants across firm size (firm size and low/fast growing industries) =>future work: differences in the determinants across age classes etc. micro-aggregated dataset of linked PS-EC-IS has a great potential for applied research and as an input for policy oriented work Advantages of the data Min 8 years long panel dimension (matched PS-CIS-EC) Almost complete coverage of industries (incl. services) but construction, retail trade not available
Conclusions preferred level of disaggregation Disaggregation by country and sector and size class Disaggregation by country, sector and age class (two categories) Less important : Disaggregation by innovation Further comments/suggestions EUK industry classification for services is too broad (FinBu & Distr) Focus on variables that cover a 10 years period Limited coverage of innovation variables in the moments (quartile) data base Definition of DISTR: no data on retail trade in the CIS Staistical break in the Human capital variable in NL, export ratio in IE in services is >10 Link with industry level R&D data