Report on the Impact of Regulation on Innovation

Lin Beibei, 2201212280

October 23, 2022

1 Introduction

This is a report on the paper titled The Impact of Regulation on Innovation[Aghion et al., 2021].
Existing models often do not consider how regulation affects innovation, as technology is seen as
an exogenous factor in the model. But regulation also matters when companies choose to invest in
innovation. In countries with strict labor regulations, slower growth may be due to red tape hindering
companies from innovating. Intuitively, companies might invest less in R&D because taxes can make
growth expensive if companies cross regulatory thresholds.
This paper formalizes this intuition using Klette and Kortum’s model[Klette and Kortum, 2004],
and conducts an empirical study using panel data of French firms from 1994 to 2007. The model
provides many predictions about the relationship between innovation and firm size. The main research
approaches and corresponding results are summarized as follows:

1. First, the research conducts a non-parametric analysis of how innovation varies with firm size.
The proportion of innovative firms just below the regulatory threshold falls sharply, an “innova-
tion valley” that suggests regulation has had a chilling effect on the desire for growth.

2. Analyze the heterogeneous responses of firms of different sizes to exogenous demand shocks.
Overall, positive market size shocks significantly boosted innovation activity. However, for firms
with sizes just below the regulatory threshold, firms’ innovative responses to shocks decline
sharply, and an innovation valley also exists.

3. Quantitatively estimate the impact of regulation on aggregate innovation and welfare. Baseline
estimates show a reduction of about 5.8% in aggregate innovation and at least a 2.3% reduction
in welfare levels.

2 Theory

Based on Klette and Kortum’s model of growth and firm dynamics, the study introduces size-contingent
regulation. Then solve the model with entry and exit conditions to explore the relationship between
innovation and employment. Finally, the study investigates the impact of demand shocks on firm
innovation activities.

1
2.1 Klette-Kortum model (2 period)

β t log(Ct ) with budget constraint

P
Suppose individuals have a logarithmic utility function U = t>0
wt + (1 + rt )at = at+1 + Ct . This implies the Euler equation β(1 + rt ) = 1 + gt . On the balanced growth
path, final output y and consumption Ct grow at a constant rate g, so that the Euler equation can
1+g
be expressed as β = 1+r , where β is the discount factor and r is the interest rate at the steady-state
level.
R1
For each period, final output y is produced according to production function ln y = 0
ln(yj )dj.
Intermediate firm owner j produces intermediates monopolistically, with total quantity yj = Aj lj
where Aj is the labor productivity and lj is the labor employed for product line j.
Intermediate firm i innovate based on its existing knowledge stock ni and its spending on R&D
sector Ri . The function for innovation flow rate Zi is in the form of Cobb-Douglas:

  η1
Ri 1− η1
Zi = ni
ζy

where η is a concavity parameter and ζ is a scale parameter. Define the intensity of intermediate
firm i as zi ≡ Zi /ni This implies the innovation cost of R&D sector: C(zi , ni ) = ζni ziη y.
If intermediate firm j 0 innovation goes well, the productivity will increase from Aj 0 to Aj 0 γ. Since
any intermediate firm j is a monopolist with marginal cost w/Aj , it would set its price equal to the
marginal cost of the previous innovator pj = γw/Aj . The equilibrium output and profit for product
line j are:

 
Aj y 1
yi = , Πj = 1− y
γw γ

The demand for the labor force in each line is y/(γw). Aggregate all ni lines so we can get the
labor employed in firm i:

Z
y yni ni
li = = =
j∈Ni wγ wγ ωγ

where ω ≡ w/y and the firm’s employment is positive correlated with the number of product lines
ni .

2.2 Regulatory threshold and innovation

If the labor force is greater than ¯l, a tax on profit is imposed to model the regulation. According to
equation li = ni /(ωγ), denote n̄ = ¯lωγ, if ni > n̄, a tax on profit at rate τ must be incurred by firm i.
For the firm to start with n = n̄ − 1 product lines, it selects its optimal innovation intensity z to
maximize its expected net present value:

max = {π(n)(1 + β) − ζz η + βz[(n + 1)π(n + 1) − nπ(n)] + βx[(n − 1)π(n − 1) − nπ(n)]}

z≥0

If n < n̄, π(n) = (1 − 1/γ), otherwise, π(n) = (1 − 1/γ)(1 − τ ). Solve the optimal value for
innovation intensity:

2
  1
 η−1
 β(γ−1)
 , if n ≤ n̄ − 1
 γζη

 1
 η−1
z(n) = β(γ−1)(1−τ n̄)
 γζη , if n = n̄ − 1

   1
 β(γ−1)(1−τ ) η−1 , if n ≥ n̄ − 1

γζη

We can see that innovation intensity z(n) decreases as firm size n increases.

2.3 Regulatory threshold and firm size distribution

Model the distribution of firm size at the steady state. Denote the share of firms with n lines as µ(n).
According to the steady-state condition, the number of exiting firms equals that of entering firms,
which is µ(1)x = ze , where x is the rate of creative destruction and ze is the innovation intensity of
entering firms.
When n > 1, the steady state condition can be expressed as:

nµ(n)(z(n) + x) = µ(n − 1)z(n − 1)(n − 1) + µ(n − 1)x(n − 1)

P∞ P∞
Close conditions: a) n=1 µ(n) = 1 ; b) x = ze + n=1 µ(n)nz(n)

2.4 Qualitative analysis

Innovation keeps increasing linearly as the employment size grows. However, only when the employ-
ment size approaches the regulatory threshold from the left-hand side, the innovation meets a sharp
valley.
From Figure 3(a), we can see no innovation valley after removing the regulation (τ = 0). Also, the
level of innovation without regulation on the right-hand side of the regulation threshold is higher than
that with regulation.
From Figure 3(b), we can see that the regulated economy generates more firms whose employment
populations are just below the regulation threshold. The number of large firms decreases because firms
are reluctant to pay taxes, so they are unwilling to employ more people.

3
2.5 Dynamic analysis: effect of demand shocks

Denote an exogenous shock on product j as ˜j , it shifts the quantity of products from yj to yj (1 + ˜j ).
Since firm i does not know the shock will occur on which product, it expects a shock i = ˜j /ni in
1
each product’s demand. Multiplying (1 + i ) η+1 to the optimal value of innovation intensity, the value
of Z becomes:

 1
 η−1
βπ(n) 1
Z(n, ) = ωγl(n)(1 + ) η+1
ζη

 1
 η−1
βπ(n) h 1
i
∆Z(n, ) ≡ Z(n, ) − Z(n, 0) = ωγl(n) (1 + ) η+1 − 1
ζη

Therefore, the demand shock has the slightest effect on the firms whose labor size is just below the
regulatory threshold.

3 Empirical study

We use the panel data of firms in France from 1994 to 2007, which consists of 182,348 firms and 1.66
million observations. The data include employment, sales, patents, innovation and manufacturing.

3.1 Static analysis: non-parametric cross-sectional analysis

• Linear relationship: The relationship between the share of innovative firms and the firm size is
almost linear, which is consistent with the optimal value of innovation intensity. In other words,
larger firms are more likely to have more patents.

• Innovation valley: Just below the 50 employees threshold, there exists a innovation valley
where the fraction of innovative firms drops.

• Slope changes: On the right-hand side of the threshold, the slope becomes flatter as the firm
size grows.

4
3.2 Dynamic analysis: response to exogenous demand shocks

To investigate firms’ response to market size shocks, we consider the following regression:

˜ i,t = b1 li,t−2
∆Y ∗ ∗
+ b2 [∆Si,t−2 × P (log(li,t−2 ))] + b3 [∆Si,t−2 × li,t−2 ] + φP (log(li,t−2 )) + ψs(i,t),t + i,t

where Yi,t measures innovation based on patents, ∆Si,t−2 is an exogenous demand shock ψs(i,t),t
∗
is a set of industry-year dummies. li,t is a binary variable that takes the value of 1 when firm i is just
below the regulatory threshold. If b3 is negative, it implies a discouraging effect of the regulation on
innovation.

• Column (1) implies when the market demand increases by 10%, the patents which measure the
innovation activities increase approximately by 1.1%.

5
 • Column (2) includes the interaction between demand shock and labor size. The coefficient of the
interaction is positive and significant, which implies larger firms respond more to demand shocks
than smaller ones.

• Column (4) includes the interaction between demand shock and the dummy variable, which
recognizes firms just below the regulatory threshold. The coefficient of the interaction is negative
and significant, which implies firms just below the regulatory threshold are reluctant to respond
to demand shocks.

Figure 6 shows the marginal effect of demand shocks on innovation. There is a sharp drop in the
marginal effect of demand shocks just below the regulatory threshold. Further away from the threshold
on two sides, the marginal effect of demand shocks grows greater as firms become larger.

4 Equilibrium effect of regulation

4.1 Effect on aggregate innovation

First, quantify important parameters of the model.

Plugging in these quantitative estimates of the key parameters implies a loss of aggregate innovation
of about 5.8% percent compared to the no-regulation benchmark.

6
4.2 Effect on welfare

The market clearing condition states that the final output must equal to the sum of consumption Ct
and R&D expenditure. The expenditure for producing an innovation intensity Z = nz is ζnz η units
of the final good. Since there are µ(i) firms of size i, R&D expenditure R ≡ i≥1 ζµ(i)iz(i)η , so that
P

the final output namely Yt = Ct + RYt .

The difference in utility between an economy with regulation and without regulation can be calcu-
lated by:

     
1 + g(τ ) β 1 − R(τ ) 1 Y0 (τ ) 1
∆U ≡ U (τ ) − U (0) = log + log + log
1 + g(0) (1 − β)2 1 − R(0) 1−β Y0 (0) 1−β

1. Since the growth g is lower with regulation (g(τ ) < g(0)), the first term is negative, which is a
welfare loss due to regulation.

2. Since R&D expenditure R is lower with regulation (R(τ ) < R(0)), the second term is positive.
The welfare gain is due to the increasing consumption, which is the residual of output.

Plug in the estimated value β = 0.96, we can compute that the welfare is 2.3% lower in the regulated
economy.

5 Conclusion and remarks

Based on Klette and Kortum’s model, the study introduces a size-contingent regulation effect and
applies it to actual firm data in France. Non-parametric analysis suggests the existence of an ”innova-
tion valley,” implying that regulation has a chilling impact on firms’ desire for growth, which are just
below the regulatory threshold. In addition, dynamic analysis is performed to examine the response
of firms of different sizes to market demand shocks. Furthermore, quantitative estimates suggest that
regulation results in a 5.8% loss in aggregate innovation and a 2.3% loss in welfare.
Overall, the effects of labor regulation appear to be negative for both innovation and welfare levels.
From my perspective, labor regulation should have some benefits. For example, it can promote the
efficiency of employees and innovative activities, as managers tend to achieve the production goals
with limited labor so the firm won’t cross the regulatory threshold. It would be better if the model
considered the dual effects of labor regulation on innovation.

References

[Aghion et al., 2021] Aghion, P., Bergeaud, A., and Van Reenen, J. (2021). The impact of regulation
on innovation. Technical report, National Bureau of Economic Research.

[Klette and Kortum, 2004] Klette, T. J. and Kortum, S. (2004). Innovating firms and aggregate inno-
vation. Journal of political economy, 112(5):986–1018.