Innovation and Prices
Jordi Jaumandreu∗
Shuheng Lin†
Boston University
Boston University
April 5, 2014
(Preliminary and incomplete)
Abstract
This paper investigates the impact of process and product innovations on the prices
set by firms. It draws on the prices reported by a sample of manufacturing firms over
an extended period of time. We assume prices are set with a markup on (short run)
marginal cost and hence they reflect variations in production efficiency, both across
firms and over time. Productivity, in turn, shifts with firms’ process and product innovations. Preliminary GMM estimates of the model show that, after controlling for
cyclical margins, productivity pushes down prices at a very sensible average pace of
about 1.5% a year. An increase in unobserved productivity decreases prices in relationship to the observed part of marginal cost. Firms that introduce new process, product
or both tend to have higher productivity increases and thus even lower prices, but this
is not a general rule. The refinement of the model in several directions is expected to
give new insights on these findings.
∗
†
Dep. of Economics, 270 Bay State Road, Boston, MA01522, USA. E-mail: jordij@bu.edu.
Dep. of Economics, 270 Bay State Road, Boston, MA02215, USA. E-mail: slin619@bu.edu.
1
1. Introduction.
This paper investigates the impact of process and product innovations on the output
prices set by firms. It uses prices reported over an extended period of time by a sample
of manufacturing firms. We assume prices are set with a markup on (short run) marginal
cost and hence they have to reflect variations in efficiency of production, both across firms
and over time. Productivity, in turn, shifts with the introduction of process and product
innovations by firms.
Models almost universally assume firms set prices with reference to marginal cost, although also suppose that optimal markups can widely change with market structure, firm
behavior and other factors.1 Marginal cost, both in the short and long run, depend on firm
level productivity that can evolve according to its innovation activities and, in particular,
the introduction of process and product innovations.2 This level of productivity is unobserved. Process innovations, aimed at reducing cost, are expected to shift the marginal cost
function downwards. However, the effect of product innovation (new or improved good) on
unobserved productivity is more ambiguous. Product innovations modify the product mix
and as a result shift the marginal cost functions in unexpected ways.3 As a simple example,
when “learning by doing” is important, firms may lack the experience in producing the new
good and unobserved productivity is likely to go down.
There are many complexities in the relationship between innovation and prices, and we
discuss now four crucial ones we hope to address. The first two are related to the difficulties
in measuring marginal cost, and the other two concern the assessment of its transmission
to prices. First, as remarked above, marginal cost includes the unobserved component that
we know with the name of individual or firm-level idiosyncratic productivity. We should
1
Hall and Hicht (1939) is a departure from this view, in what Ellison (2006) characterizes as an early
contribution to behavioral industrial organization.
2
Peters, Roberts, Vuong and Fryges (2013) model firms’ marginal cost as depending on process and
product innovation in the way that we are going to adopt in this paper. Aw, Roberts and Xu (2011) and
Doraszelsky and Jaumandreu (2013) are papers in which endogenous efficiency depends on R&D expenditure.
3
In practice, firms produce a range of products and their cost function depends on a given product mix.
2
model this component and its dependence on innovation. Second, marginal cost may have
a second component typically unobservable that are input adjustment costs.
4
Third,
prices may be set in relation to a particular marginal cost (short-run, long-run, expected)
and only be updated at time intervals. In fact, the possibility of price rigidity with its
consequence of a discontinuous time pattern of markup adjustments has been the object of
exploration in the literature.5 This may create a selection problem in estimation because
the errors corresponding to observed adjustments are likely to become correlated with the
included variables, particularly innovation. The last complexity concerns the modelling of
the markup. This is crucial since innovation may lower the elasticity of demand, in which
case prices are affected as the result of the optimal adjustment of margins.
At this stage we are going to abstract from all these complexities but the first. In our
baseline model, firms set prices with a markup over marginal cost. The markup is modelled
to only have a cyclical feature, and the marginal cost has an unobserved component that
evolves according to an endogenous Markov process impacted by the firm-level innovations.
Our plan is to extend the model to encompass the testing (and may be modelling) of each
one of the other aspects: input adjustment costs, non-continuous price adjustments, and
endogenous markup.
There are at least three reasons for studying the impact of innovation on prices. First
of all, how technological change via firm-level innovation transmits to prices is relevant for
understanding the working of many economic forces. For example, prices play a crucial
role in the allocation of resources and growth. When a process innovation induces a price
decrease, market demand for the good is expected to increase. As a result, the demand
for inputs required for production is expected to increase by more than what is needed
to compensate for any input displacement provoked by technology. However, the magni4
5
For a recent assessment on input adjustment costs see Hall (2006).
Sheshinski and Weiss (1983) provided a formal microeconomic model of price adjustment in the presence
of adjustment costs, and Cecchetti (1986) was the first to assess adjustments empirically. Klenow and Malin
(2010) review recent evidence on firm-level price sluggishness. An alternative explanation and modelling
for price rigidity is collusion (see Athey and Bagwell, 2008) but we are going to exclude this possibility by
assuming monopolistic competition.
3
tude and how complete, rapid and heterogeneous is this mechanism is something to assess
empirically.
Second, the quantities used in productivity analysis are, almost without exception, obtained by deflating firm revenue by a price index. When an industry-wide index is used, it
misses out what is left in the difference between the firm-level and the general indices and
may get biased estimates.6 However, are all the problems solved with firm-level price indices? To illustrate this is not necessarily the case, let us consider for simplicity the scenario
of constant returns to scale.7 Under CRTS, revenue deflated by a firm-level price index only
reflects productivity that has been passed onto prices. Hence, only by supposing firm-level
prices fully reflect the variations in productivity we can conclude that these variations are
present in the quantity index obtained when dividing revenue by price. It is thus crucial to
explore whether prices set by firms exhibit this behavior. If not, firm-level indices are not
sufficient to eliminate biases in the measurement of productivity.
Third, how well prices reflect productivity improvements is important for the measurement of welfare. On the one hand, if prices accurately reflect technical efficiency improvements then they can be used to measure increases in consumer surplus coming from innovation. However, it is much more difficult to measure the effect of product improvements
that directly affect consumer utility. There is an ongoing literature about how the methods
applied by Statistical Offices with respect to new goods may impact this measurement.8
One relevant dimension of this problem is just to understand how firms price product improvements in practice.
To estimate our model we use the prices reported over an extended period of time by a
6
This problem was firstly pointed out by Klette and Griliches (1996). Later, it has been at least addressed
by Foster, Haltiwanger and Syverson (2008), De Loecker (2011) and Katayama, Lu and Tybout (2011), who
propose different ways to deal with it.
7
Let the production function be  =    where  stands for idiosyncratic productivity and  shows
CRTS. Cost is  = ()− , where  represents prices, and marginal cost   = ()− Under a
constant markup, revenue is  =   =

()−  
−1
=

()
−1
so it does not depend on productivity.
If we divide  by  all productivity is brought up to the result by the price index.
8
See, for example, Pakes (2003).
4
sample of Spanish manufacturing firms. We use data on ten (unbalanced panel) industry
samples, which in total amount to more than 2,300 manufacturing firms and 17,000 observations, corresponding to the period 1990-2006. We have firm-level price indices for each
firm, constructed from the reported yearly output price increases in the main market, and
we observe the moment of the process and product innovations introduced by firms as well
as other relevant information.
Preliminary GMM estimates of the model show that innovation effectively reduces prices.
Controlling for cyclical margins, we get plausible estimates of the marginal cost function
parameters (that are also the parameters of the underlying production function). Unobserved productivity, estimated as an endogenous Markov process reducing marginal cost,
has an impact on prices, decreasing them in relationship to the observed part of marginal
cost. Firms that introduce new process and/or product have higher productivity increases
and thus lower prices, but this is not a general rule. The refinement of the model in several
directions is expected to give new insights on these findings.
The paper is organized as follows. In Section 2, we explain with detail the model and
discuss its identification. In Section 3, we briefly report the parametric specification that
we are currently applying in estimation and the method of estimation. Section 4 briefly
explains the data and Section 5 presents preliminary results. Section 6 are a few concluding
remarks.
2. Model and identification.
We assume that firm  operates in a monopolistically competitive market and sets the
price with a markup on (short-run) marginal cost. By monopolistic competition we understand that each firm faces a downward-sloping demand for its product, makes no profit in
the long run equilibrium, and a price change by one firm has only a negligible effect on the
demand of any other firm (Tirole, 1989). How large is demand for each product may differ
by exogenous reasons and the demand for the product of a firm may change over time by the
effect of exogenous and endogenous shifters such as well as the state of the market. A firm
5
knows its cost function, which includes among its arguments its specific level of efficiency.
The firm looks at the state of its demand and decides simultaneously the price and
quantity that maximize profits, determining the quantity of inputs to be employed. In our
first approximation we are going to suppose that the demand elasticity with respect to price,
and hence the markups, only change according to a firm specific indicator. This gives, by
no means, an idyosincratic variation to this elasticity. So price is
 =
 
 exp( )
  − 1
(1)
where  is an error orthogonal to all information available when the firm takes the decisions.
We comment more on this error below.
Let us first specify marginal cost. Assume that the firm production function is
 =  (     ) exp(  )
where   represents the Hicks neutral level of efficiency of firm , that we will simply call
productivity. We assume  and  variable factors and take  as given. Variable cost
minimization implies the cost function
 = (        exp(  ))
where  and  stand for the prices of the variable inputs  and  respectively
Assuming that short-run returns to scale (returns given capital) are not constant, marginal
cost is
  =


(        exp(  )) exp(−  )
=

(  exp(  ))
On the other hand, by Shephard’s lemma, optimal materials choice conditional on output
is
 =

=  (        exp(  ))
 
Inverting the latest equation for   exp(  ) and using the resulting expression to replace
  exp( ) in marginal cost, we get
 = (        ) exp(− ) = ( ) exp(−  )
6
(2)
where  = {       } is a vector of observable variables.
Under the above conditions, price can be written as
 =
 
 ( ) exp(−  +  )
  − 1
or, in logs, as
 = ln
 
+ ( ) −   +  
  − 1
This equation shows that we expect prices to structurally evolve according to productivity
and that we should be able to pick up this relationship as long as we control adequately by
markups and marginal cost.
We want to estimate endogenous productivity, or productivity that is (at least partly) the
result of the innovation effort done by firms (Doraszelski and Jaumandreu, 2013). Assuming,
as in all the literature subsequent to Olley and Pakes (1996), that   follows a first order
Markov process, we model productivity as depending on past productivity and the shifts
induced by the introduction of process and product innovations (as in Peters, Roberts,
Vuong and Fryges, 2013). That is,
 = ( −1     ) +   
where −1 and −1 are measures of the process and product innovations introduced
by firm ,  is an unknown function aimed at picking up both the path dependence of productivity as well as the unspecified impact of the innovations, and    a random innovation
orthogonal to all the arguments of (·) Our structural model linking prices to innovation
is hence
 = ln
 
+ ( ) − ( −1     ) −   +  
  − 1
(3)
The model cannot be estimated as it is because it includes unobservable lagged productivity −1 interrelated with the observables  and   In addition,  −1 is likely to
be correlated with most of the explanatory variables. In order to estimate the model, we
will employ an Olley and Pakes (1996) type of procedure, replacing the unobservable  −1
by an expression in terms of observables.
7
The choice of which expression to use is, however, not trivial. The inclusion of the disturbance  implies that we only know the price decided by the firm up to this error:
 = ∗ exp( ) By means of this disturbance we want to allow both for possible measurement errors and arbitrary factors, unexpected at the time of setting the price, that may
induce a different observed price. As in the structural estimation of production functions,
a natural choice for the expression to be used would be an inverted demand relationship.
For example, profit maximization implies the demand for materials
 = (   ∗ (1 −
1
1
)  ∗ (1 −
)   )
 
 
But, as this equation makes clear, demands for labor and materials include price and specified in terms of observed price will include the error   Having an unobservable in the
expression to use in the (·) function, even if it is uncorrelated with everything, implies a
problem for the identification of the model.9
If we assume that we observe   however, we have more than one solution.10 A first
possibility is inverting the (lagged) conditional demand for materials, that is
(−1  −1  −1  −1 )
 −1 = ln −1 − ln −1

Another possibility is directly inverting the production function
 −1 = ln −1 − ln  (−1  −1  −1 )
Let us write in general  −1 = (−1 ), noticing that the set −1 changes with the
different alternatives. The estimable model can then be written as
9
Scholars estimating production functions tend to avoid this complication by assuming that all prices are
the same for all firms and can be replaced by dummies. In this case prices disappear from the demand for
materials, that can be written  =  (    ) See, for example, Levinsohn and Petrin (2003) or De
Loecker (2011).
10
We can suppose, for example, that the price at which the choices were done is not observed, and we
observe, instead, the actual price that has been finally adopted to effectively sell the planned and produced
quantity. Revenue and price are then affected by the same  error, but quantity obtained by deflating revenue
at the current price gives the correct quantity. This matches well with the dominant idea that price is the
most flexible instrument in the hands of the firms.
8
 = ln
 
+ ( ) − ((−1 )    ) −   +  
  − 1
(4)
Until now we have developed the model with a minimum of functional form or distributional assumptions. Is the model identified at this level of generality? To answer this
question notice first that the set  contains only one endogenous variable, the input
quantity   and the set −1 none  is the only variable that is determined once
the innovation   is known. All the rest of the variables are likely to be correlated with
productivity but, once the predictable part of productivity has been specified in terms of
observables, it only remains unobserved the value of the innovation    This innovation is
only revealed after the values of these variables have been chosen. This is true for the innovation variables because they are the (partly random) outcome of past R&D investments.
And this is true for the input prices under the assumption that they are determined in
competitive markets and hence given for the firm.
The discussion about identification can start by asking ourselves if we have instruments
for   The answer is yes, but the available instruments depend on the alternative −1
sets. If we are replacing  −1 by the conditional demand for materials, a natural instrument for  is −1  Variable −1 is excluded from the inverted conditional demand,
so it doesn’t appear in the equation, and according to the model is correlated with −1
and hence   If we are using the production function, two natural instruments are the
lagged input prices −1 and −1  With this kind of substitution there are no input
prices in the expression that replaces  −1 and lagged input prices are, however, related
to   Since  is correlated to −1 , it is also correlated with the lagged prices that
determine −1  The model is hence identified in very general conditions. In what follows
we use a particular parametric specification that makes the model particularly simple and
identification straightforward.
9
3. Estimation.
We consider the Cobb-Douglas production function
 =  0 +    +    +    +   
so the short-run marginal cost function in terms of materials is
 =  −    + (1 −   − ) + (1 −   ) +    −   
where  = − 0 − ln(  +   ) −   ln   +   ln   
In the empirical application we are going to use the in-homogeneous Markov process
  =   + ( −1     ) +    So, from the pricing equation, the Markov process
assumption, and the use of the lagged inverted production function to substitute for  −1 ,
we have the estimating equation
 
−   −    + (1 −   − ) + (1 −   ) +   
  − 1
−(−1 −   −1 −    −   −1     ) −   +  
 =  + ln
(5)
where we are going to specify (·) as a complete polynomial of order three in its arguments.
Using −1 as a shorthand for −1 −   −1 −    −   −1  we may write (·) in the
following way11
(−1  −1  −1 ) =  1 −1 +  2 2−1 +  3 3−1 +  4  +  5 
+ 6 −1 ·  +  7 2−1 · 
+ 8 −1 ·  +  9 2−1 · 
+ 10  ·  +  11 −1 ·  ·  
To estimate the markup, we are going to use expression
ln
11
 
1 + exp( )
= ln
= ln(1 + exp( )) −  
  − 1
exp( )
We implicitly collapse the constant of the unknown function in the constant of the equation.
10
which restricts elasticity to be greater than one while allows for its cyclical fluctuation
according to the firm-level state of the market indicator  or market dynamism.
We are going to use a nonlinear GMM estimator. Writing the error of the equation as a
function of all parameters  () our moments have the form
[( ) ()] = [( )(−  +  )]
where ( ) is a vector of functions of the exogenous variables.
4. Data.
We estimate the model with data on ten (unbalanced panel) industry samples, which in
total amount to more than 2,300 manufacturing firms and 17,000 observations, corresponding to the period 1990-2006. All variables come from the survey ESEE (Encuesta Sobre
Estrategias Empresariales), a firm-level panel survey of Spanish manufacturing starting in
1990. At the beginning of this survey, firms with fewer than 200 workers were sampled
randomly by industry and size strata, retaining 5%, while firms with more than 200 workers were all requested to participate, and the positive answers represented more or less a
self-selected 70%. To preserve representation, samples of newly created firms were added
to the initial sample every subsequent year. At the same time, there are exits from the
sample, coming from both death and attrition.
The survey then provides a random sample of Spanish manufacturing with the largest
firms oversampled. Information on the firms include, in addition to the usual output and
input quantity measures, the firm-level variations for the price of the output and the price
of the inputs, the introduction of technological (process and product) innovations, and the
indicator of the state of the market market dynamism.
Tables 1 and 2 report the size of the samples and some descriptive statistics. Table 1
is focused on prices. The evolution of prices over the whole period is moderate, but heterogeneous enough to motivate the exploration of the role of innovation in their evolution.
An OLS regression of the growth of prices on the dummies of process and product innovation, controlling by time dummies, turns out to be hardly informative: almost no effect
11
is statistically significant. Table 2 is focused on innovation. The introduction of process
and product innovations is more heterogeneous. Firms tend to introduce more process
than product innovations, at the approximate paces of one innovation every three and four
years respectively. Innovation is, as expected, especially important in the more technology
intensive sectors 3, 4, 5 and 6.
5. Preliminary results.
We have carried out estimates of equation (5) with very sensible results that allow us
a first look at the impacts of innovation on prices. In what follows we first present the
estimates and then derive the implications for the productivity growth and price variations
due to the impact of innovation.
We estimate equation (5) by nonlinear GMM using the following set ( ) of instruments:
constant, time dummies (15), lagged input prices (price of materials and wage), a polynomial
of order three in lagged capital and labor, lagged materials, the two dummies of lagged
innovation, and the interaction of each of these dummies, as well as their product, with
each one of the lagged inputs. In industries 1,7,8 and 10 we also include the square and
the cube of materials.12 And, in industry 8, contemporaneous capital.13 This gives a total
of 36 instruments, sometimes enlarged up to 39. We have to estimate 4 parameters that
enter nonlinearly:          the parameters of the marginal cost/production functions, and
, the parameter on the variable market dynamism in the modeling of the markups. We
have other 27 parameters that enter linearly (constant, 15 time dummies and 11 polynomial
coefficients), for whose estimation we apply the procedure of “concentrating out.” This gives
us a minimum of 5 overidentifying restrictions that allow to test for the specification.
The results of the preliminary estimates are summarized in Table 3. To reach these re12
We find useful to avoid raising materials to the powers 2 and 3 in the rest of industries. Materials is
a variable likely to be subject to errors of measurement whose role might be exacerbated when raised to
powers.
13
Contemporaneous capital is a legitimate instrument, that we try to avoid because of the likely effects
the errors in its measurement.
12
sults, the modelling of the markup by means of the variable market dynamism was crucial.
According to the signs, five industries tend to have procyclical margins and other five countercyclical. We omit the estimates from the table because the coefficient is only precisely
estimated in one industry. However, the presence of the variable is very important for the
plausibility of the rest of estimates. This points to the presence of a high heterogeneity, only
partially picked up by our current specification. We should also emphasize that we carry
out the regressions with the structural modelling of the margins (demand elasticity changes
with the state of the market), but the variable works also quite well when it is introduced
linearly. This suggests that some cases could also be picking up (or be counterbalanced by)
adjustment costs. We leave this for future research.
The parameter estimates of the marginal cost/production functions, reported in columns
(1) to (3), are quite sensible, although they contain a few minor flaws. The coefficients on
capital are small and imprecisely estimated in industries 6, 7 and 8. And the materials
coefficients are too small in industries 4 and 10. When some coefficient is underestimated,
seems to be related to some overestimation of the rest of the coefficients. In all industries,
returns to scale are very close to constant. The overidentifying restrictions tests (columns
(4) and (5)) strongly support the specification in all industries at very high probability
values.
Despite the preliminary characteristics of the parameter estimation, we have tentatively
estimated the distribution of   and subjected its values to some descriptive exercises. We
first compute the individual values of productivity growth rates, given by ∆  =  − −1 .
Then we split the industry samples into observations corresponding to the introduction of
a process innovation only, observations corresponding to the introduction of a product
innovation only, observations corresponding to the introduction of both kind of innovations
and, finally, observations corresponding to moments without any kind of innovation. Finally,
we calculate weighted averages of the rates of growth, using as weights the firms sales
shares lagged two periods and replicating the observations for the small firms (less than 200
workers) as indicated by the known starting representativeness in the survey of each kind
13
of firm.14 The results are reported in columns (6) to (9) of the table.
To read properly the innovation effects recall that we are modelling the unobserved part
of marginal cost. We call this part unobserved productivity. Productivity reduces marginal
cost and, under the assumption currently embedded in the model that this reduction is
perfectly passed on prices, reduces prices. A positive impact of innovation on productivity
implies a decrease in prices, and a negative impact of innovation on productivity implies an
increase of prices. Of course these decreases or increases are in addition to the price variation
induced by the observed part of marginal cost. In fact, we are estimating productivity as
what remains to be explained after deducing the part corresponding to observed marginal
costs (and margins). The own prices are giving us then the indication about what happens with productivity. Importantly, the model is estimated in levels, but its implications
scrutinized in first differences.
First, and importantly, the estimations show that productivity pushes down prices at a
very sensible pace of about 1.5% a year. Both process and product innovations increase
efficiency and push down prices. The same do the process and product innovations that are
implemented together. Only in a few cases (5 out of 30) innovations imply price increases
and there is only one case of significant magnitude (product innovation in industry 6). The
variation in productivity in the absence of innovations also decreases prices. If one excludes
the few negative averages, the observations without innovation convey a price reduction of
about 1.3% a year, the observations with either a process or a product innovation of about
1.8% and the observations with both of about 1.9%.
According to what we have said before, there is not a particular relationship to expect
between the increases in productivity when there is no innovation and when there is product
innovation. Hence, the efficiency increases in columns (8) and (9) can be greater or lower
than the increases reported in column (6). In fact, price reductions when there is product
innovation are less intense that when there is no innovation in four industries (in one case
there is even a price increase). And price reductions when there is process and product
14
Since the proportions were approximately 5% and 70% this implies replicating the smallest firms 14
times.
14
innovation are less intense than when there is no innovation in five cases (two cases with
a slight price increase). More surprisingly, price reductions induced by process innovation
seem to be more intense than the price reductions in the absence of innovation in only half
of the cases. We have no straightforward explanation for this, a finding whose robustness
should be checked with the refinement of the model.
6. Concluding remarks
We have started an investigation on the impact of process and product innovations on the
prices set by firms, using the prices reported over an extended period of time by a sample
of manufacturing firms. Our baseline model is simple: firme set prices over short run
marginal costs, markups are only allowed to show idyosincratic cyclical variation, marginal
cost has an unobserved component that evolves according to an endogenous Markov process
impacted by the firm-level innovations and is passed on its integrity to prices. It is crucial
to explore whether prices set by firms effectively exhibit this behavior. If not, idiosyncratic
productivity variation would tend not to be timely passed on to the quantity indices used
in the measurement of productivity and would bias the estimates. Our plan is to extend
the model to encompass the testing (and probably modelling) of each one of the complex
aspects that still remain: input adjustment costs, non-continuous price adjustments, and
endogenous markup variations. But, importantly, the preliminary estimations show that
productivity pushes down prices at a very sensible average pace of about 1.5% a year.
Both process and product innovations, as well as process and product innovations that are
implemented together, increase efficiency and push down prices. A series of questions remain
over the relative magnitude of the impacts.The refinement of the model in the mentioned
directions is expected to give new insights on our findings.
15
References
Athey, S. and Bagwell, K. (2008), "Collusion With Persistent Cost Shocks," Econometrica,
76, 493-540
Cecchetti, S. G. (1986), "The Frequency of Price Adjustment : A study of the Newsstand
Prices of Magazines," Journal of Econometrics, 31, 255-274
De Loecker, J. (2011), "Product Differentiation, Multiproduct Firms, and Estimating the
Impact of Trade Liberalization on Productivity", Econometrica, 79, 1407-1451
Doraszelski, U. and J Jaumandreu (2013), "R&D and Productivity: Estimating Endogenous Productivity," Review of Economic Studies, 80. 1338-1383.
Ellison, G. (2006), "Bounded Rationality in Industrial Organization," mimeo, MIT.
Foster, L., Haltiwanger, J. and Syverson, C. (2008), "Reallocation, Firm Turnover, and
Efficiency: Selection on Productivity or Profitability," American Economic Review,
98, 394-425.
Hall, R. E. (2006), "Measuring Factor Adjustment Costs," The Quarterly Journal of Economics, 119, 899-927.
Hall, R.L. and Hicht, C.J. (1939), "Price Theory and Business Behavior," Oxford Economic
Papers, 12-45.
Katayama, H., Lu, S. and Tybout, J. R. (2009), "Firm-level Productivity Studies: Illusions
and a Solution," International Journal of Industrial Organization, 27, 403-413.
Klenow, P. J. and Malin, B. A. (2010), "Microeconomic Evidence on Price-Setting", Working paper no. 15826, NBER, Cambridge.
Klette, T. J. and Griliches, Z. (1996), "The Inconsistency of Common Scale Estimators
When Output Prices Are Unobserved and Endogenous", Journal of Applied Econometrics, 11, 343-61.
16
Levinsohn, J. and Petrin, A. (2003), "Estimating Production Functions Using Inputs to
Control for Unobservables", Review of Economic Studies, 70, 317-341.
Olley, S. and A. Pakes (1996), "The Dynamics of Productivity in the Telecommunications
Equipment Industry," Econometrica, 64, 1263-1298.
Peters, B., M. Roberts, V.A. Vuong and H. Fryges (2013), "Estimating Dynamic R&D
Demand: An Analysis of Cost and Long-Run Benefits," mimeo.
Sheshinski and Weiss (1983), "Optimal pricing policy under stochastic inflation", Review
of Economic Studies, 50, 513-529.
Tirole, J. (1989), The Theory of Industrial Organization, MIT Press.
17
Industry
Table 1: Sample size and price descriptive statistics, 1991-2006
Sample size
Price indices
OLS of price growth on innovation (time dums. incl.)
Obs. Firms
Log (s.d.) % Growth (s.d.)
Constant (s.d.) Process (s.d.) Product (s.d.) Standard error
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
1. Metal and metal products
2365
313
0.106
(0.185)
0.017
(0.052)
0.008
0.006
-0.004
0.003
0.003
0.003
0.050
2. Non-metallic minerals
1270
163
0.066
(0.199)
0.012
(0.058)
-0.003
0.011
-0.002
0.004
0.002
0.004
0.057
3. Chemical products
2168
299
0.045
(0.197)
0.008
(0.055)
0.014
0.005
-0.003
0.002
0.001
0.003
0.053
4. Agric. and ind. machinery
1411
178
0.122
(0.150)
0.015
(0.026)
0.023
0.004
-0.002
0.002
0.003
0.002
0.026
5. Electrical goods
1505
209
0.051
(0.186)
0.008
(0.046)
0.015
0.007
-0.004
0.003
-0.004
0.004
0.045
6. Transport equipment
1206
161
0.053
(0.136)
0.008
(0.031)
0.011
0.010
-0.007
0.002
0.005
0.002
0.030
7. Food, drink and tobacco
2455
327
0.160
(0.188)
0.021
(0.054)
0.037
0.005
-0.001
0.003
0.002
0.003
0.053
8. Textile, leather and shoes
2368
335
0.119
(0.171)
0.015
(0.042)
0.015
0.005
-0.001
0.002
0.002
0.003
0.042
9. Timber and furniture
1445
207
0.128
(0.141)
0.020
(0.031)
0.022
0.005
0.001
0.002
0.010
0.002
0.030
10. Paper and printing products
1414
183
0.124
(0.232)
0.017
(0.074)
0.018
0.009
-0.007
0.004
0.005
0.004
0.069
Industry
Table 2: Descriptive statistics on the introduction of innovations, 1991-2006
Proportion of obs.
Obs. with process
Obs. with product
Firms with process
Proc. (s.d.) Prod. (s.d.)
Stable (%) Occas (%).
Stable (%) Occas (%).
Stable (%) Occas. (%)
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Firms with product
Stable (%) Occas.(%)
(9)
(10)
1. Metal and metal products
0.373
(0.484)
0.184
(0.387)
151
(6.4)
732
(31.0)
66
(2.8)
369
(15.6)
27
(8.6)
196
(62.6)
12
(3.8)
126
(40.3)
2. Non-metallic minerals
0.265
(0.442)
0.172
(0.378)
31
(2.4)
306
(24.1)
18
(1.4)
201
(15.8)
7
(4.3)
96
(58.9)
4
(2.5)
72
(44.2)
3. Chemical products
0.403
(0.490)
0.345
(0.476)
121
(5.6)
752
(34.7)
152
(7.0)
597
(27.5)
29
(9.7)
197
(65.9)
32
(10.7)
175
(58.5)
4. Agric. and ind. machinery
0.332
(0.471)
0.354
(0.478)
85
(6.0)
384
(27.2)
79
(5.6)
420
(29.8)
17
(9.6)
96
(53.9)
14
(7.9)
89
(50.0)
5. Electrical goods
0.375
(0.484)
0.365
(0.481)
68
(4.5)
496
(33.0)
164
(10.9)
385
(25.6)
16
(7.7)
131
(62.7)
33
(15.8)
94
(45.0)
6. Transport equipment
0.464
(0.499)
0.313
(0.464)
149
(12.4)
411
(34.1)
105
(8.7)
273
(22.6)
32
(19.9)
97
(60.2)
21
(13.0)
82
(50.9)
7. Food, drink and tobacco
0.305
(0.461)
0.223
(0.416)
148
(6.0)
602
(24.5)
141
(5.7)
407
(16.6)
31
(9.5)
182
(55.7)
28
(8.6)
149
(45.6)
8. Textile, leather and shoes
0.242
(0.482)
0.230
(0.421)
71
(3.0)
502
(16.5)
122
(5.2)
422
(17.8)
17
(5.1)
158
(47.2)
21
(6.3)
117
(34.9)
9. Timber and furniture
0.285
(0.451)
0.257
(0.437)
71
(4.9)
341
(23.6)
41
(2.8)
331
(22.9)
13
(6.3)
110
(53.1)
11
(5.3)
96
(46.4)
10. Paper and printing products
0.293
(0.455)
0.141
(0.348)
37
(2.6)
378
(26.7)
17
(1.2)
182
(12.9)
7
(3.8)
118
(64.5)
5
(2.7)
60
(32.8)
Industry
Table 3: GMM estimation results and average  growth
Overidentifying
Coefficients (std. err.)
restrictions test
Productivity increase/Price reduction
Capital Labor Materials
2 ( ) p val.
No inn. Proc. only Prod. only Both
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
1. Metals and metal products
0.206
(0.048)
0.355
(0.034)
0.461
(0.059)
4.980
(8)
0.760
0.020
0.028
0.026
0.022
2. Non-metallic minerals
0.109
(0.065)
0.360
(0.032)
0.526
(0.083)
3.390
(6)
0.759
0.025
0.017
0.008
0.011
3. Chemical products
0.247
(0.051)
0.350
(0.045)
0.407
(0.055)
0.539
(6)
0.997
0.016
0.023
0.025
0.021
4. Agric. and ind. Machinery
0.374
(0.172)
0.319
(0.048)
0.277
(0.180)
2.260
(6)
0.855
0.014
0.010
0.028
-0.005
5. Electrical and electronic products
0.194
(0.055)
0.347
(0.036)
0.485
(0.045)
2.656
(6)
0.851
0.018
0.029
0.014
0.044
6. Transport equipment
0.044
(0.065)
0.218
(0.013)
0.806
(0.073)
4.858
(6)
0.562
0.017
0.013
-0.034
0.014
7. Food, drink and tobacco
0.038
(0.032)
0.306
(0.053)
0.704
(0.059)
5.289
(8)
0.726
0.006
-0.005
0.023
0.021
8. Textile, leather and shoes
0.043
(0.023)
0.359
(0.045)
0.614
(0.054)
5.065
(9)
0.829
0.000
0.008
0.024
0.009
9. Timber and furniture
0.111
(0.090)
0.262
(0.021)
0.569
(0.077)
5.855
(6)
0.990
0.001
0.017
0.009
-0.004
10. Paper and printing products
0.332
(0.043)
0.319
(0.017)
0.262
(0.041)
7.545
(8)
0.479
0.013
-0.002
0.001
0.009