View
215
Download
1
Tags:
Embed Size (px)
Citation preview
1
Why Demand Uncertainty Curbs Investment: Evidence
from a Panel of Italian Manufacturing Firms
Maria Elena Bontempi (University of Ferrara)
Roberto Golinelli (University of Bologna)
Giuseppe Parigi (Bank of Italy)
Workshop:
MODELLING AND INFERENCE IN MICROECONOMICS
Bologna, March 5, 2007
2
The theoretical literature
For a risk-averse firm with CRS technology
PERFECT COMPETITION
The effect of uncertainty on
investment decisions is
negative
LABOUR FLEXIBILITY
IMPERFECT COMPETITION
+IRREVERSIBILITY
Sign and dimension of the effect depend on the relevance of different
hypotheses
positive (or zero)
3
The empirical literature
In general empirical studies concentrate on the sign of the investment-uncertainty relationship
Guiso e Parigi (1999) is the first paper where the role of irreversibility and of the degree of competition is analysed
Main limitation is due to a lack of relevant data
The result is that uncertainty has a negative effect on investment decisionsThe implications of
adopting different theoretical frameworks are very rarely analysed
Data about uncertainty, expected demand, investment plans and other indicators are needed...
4
Aim of the paper
To analyse the investment-uncertainty relationship in the light of alternative hypotheses, by using a panel of Italian manufacturing firms.
The availability of a panel allows to: reduce the risk of biased estimates due to the
omission of unobservable variables varying among firms and almost fixed over time (i.e. risk aversion); varying over time and almost fixed among firms (i.e. macroeconomic shocks)
perform a deeper analysis of the effects of the hypotheses on the degree of irreversibility, of firm’s market power and of the flexibility of labour input
assess the time dynamics of the investment-uncertainty relationship and the stability of parameters in different subsamples
5
Data
Bank of Italy’s Survey on Investment of Manufacturing firms (SIM) is the main data source: an unbalanced panel of about 17,000 observations over 1996-2004. We use only the subsample of firms with more than 50 employees, replying to the question about demand uncertainty (about 8,000 observations). The source of data about capital stock; cash-flow, price-cost margins is Company Accounting Data Service (CADS, Centrale dei Bilanci).
The merge of SIM-CADS firms leads to a loss of observations: our basic sample (unbalanced sample) has about 7,500 observations.
6
The theoretical modelMain problem: a closed-form solution of the general model for investments with uncertainty does not exist.
A solution would be to impose restrictions about: adjustment costs specification; returns to scale; demand elasticity to prices.
BUT this is exactly what we want to examine!
Idea (Guiso-Parigi, 1999): if investments are in some way irreversible, the level of demand that drives investment is related to uncertainty, hence:
investments elasticity to demand is negatively related to uncertainty.
7
The empirical model
tIit+1 investment plans in year t for t+1
Iit realised investments in t
tYit+1 demand predicted in t for t+1
u(tYit+1) uncertainty on future demand, obtained by using the expected growth rate of demand (tgit+1) reported by SIM respondents:
u(tgit+1)Yit = (tgmaxit+1 – tgmin
it+1)Yit
itit
it
it
it
itt
it
ittti
it
itt ZK
I
K
Yu
K
Ya
K
I
4
13
12
11
1 1
ai , λt panel fixed effects by firm (i) and by year (t)
The uncertainty effect depends on the sign of 2
8
Cross-section estimates
1 estimate is always significantly positive; while 2 estimate is negative.
Both effects decrease over time: 2 is not significant in the last two years of the sample.
Preliminary estimates over repeated cross-sections: one estimation set for each year by imposing ai = a and λt = 0
-.6
-.4
-.2
.0
.2 .0.1.2.3.4.5.6
95 96 97 98 99 00 01 02 03 04 05
Elasticity toexpected demand
Uncertainty parameter
9
Main finding with cross-sections
If we had had data only for one year in the first half of the sample, before 2000, we would had estimated a negatively significant uncertainty effect, as in Guiso-Parigi;
BUT …
… if we had had data only for one year in the second half of the sample, after 2002, we would had estimated a close to zero (and not significant) uncertainty effect.
In order to interpret such conflicting results, the PANEL approach may be
useful
10
Panel and sub-samples
The pooling of cross-sections may be assumed:
(1) for the whole sample
(2) in sub-samples of firms selected on the basis of:
(2a) high/low labour flexibility (the turnover is above-below the sector median in t)
(2b) high/low market power (the PCMit isabove-below the sector median
in t)
(2c) high/low reversibility, the indicator REVi=1 if operate at least two times in the second-hand market
11
Panel estimates
0.10130.1827
0.14010.2443
4.642.35
11411000
52892353
0.0001(0.0390)
0.0615(0.1139)
-0.0206(0.0148)
-0.0368(0.0143)
0.0181(0.0034)
0.0452(0.0307)
(3)(2)
highlow
Reversibility
0.11280.1900
0.13960.1974
2.682.54
13221400
35493555
-0.0898(0.0911)
0.0591(0.0822)
-0.0493(0.0215)
-0.0157(0.0085)
0.0226(0.0060)
0.0422(0.0256)
(5)(4)
highlow
Market power
0.19330.1089
0.19120.1300
2.422.62
15601477
37753867
0.1087(0.0486)
0.0199(0.0426)
-0.0151(0.0105)
-0.0673(0.0241)
0.0445(0.0251)
0.0188(0.0076)
(7)(6)
highlow
Labour flexibility Total sample (1996-2004)
Regressors (1)
Expected Demand
0.0301 (0.0125)
Uncertainty
-0.0318 (0.0139)
Realised Invest.
0.0284 (0.0513)
Num. Obs. 7642
Firms 2141
Avg. Time 3.57
RMSE 0.1807
R2 0.1284
12
Labour flexibility and PCM interaction
Labour flexibility
High Low
Market power Market power
Total sample
All cases
High Low
All cases
High LowRegressors (1) (2) (3) (4) (5) (6) (7)
0.0301 0.0445 0.0161 0.0776 0.0188 0.0719 0.0086Expecteddemand (0.0125) (0.0251) (0.0063) (0.0486) (0.0076) (0.0170) (0.0058)
-0.0318 -0.0151 -0.0536 -0.0215 -0.0673 -0.0707 -0.0283Uncertainty(0.0139) (0.0105) (0.0262) (0.0095) (0.0241) (0.0345) (0.0272)
0.0284 0.1087 -0.0467 0.1223 0.0199 -0.0252 0.0239Realisedinvestments (0.0513) (0.0486) (0.0910) (0.0567) (0.0426) (0.0705) (0.0300)
NT 7642 3775 1678 1820 3867 1871 1735
N 2141 1560 847 957 1477 884 865
T 3.57 2.42 1.98 1.90 2.62 2.12 2.01
RMSE 0.1807 0.1912 0.1104 0.2208 0.1300 0.1186 0.0947
R2 0.1284 0.1933 0.1285 0.3314 0.1089 0.2106 0.0709
13
Time-constancy of panel estimates
The panel estimates of the three parameters of interest are constant at 5% in alternative sample splits in 1999, 2000, 2001, 2002 e 2003; using the sup-Wald statistic of Andrews (1993).
Behind this overall constancy there are systematic and significant shifts of the uncertainty effect.
-.15
-.10
-.05
.00
.05
95 96 97 98 99 00 01 02 03 04 05
Plot of the 2t deterministic shifts:
9 t t2
2
ˆˆ
14
Theoretical explanation
Risk aversion reduction
Empirical assessment
Not identified (nested in the individual effects)
Increase in capital reversibility
Our indicators suggest a reduction
Decrease in firms’ market power
.04
.06
.08
.10
.12
.14
95 96 97 98 99 00 01 02 03 04 05
3rd Quartile
1st Quartile
Median
MeanOther: labour flexibility and openness …
Determinants of time-fluctuations
15
Modelling uncertainty evolution
The time evolution of 2 parameter can be modelled by a linear function with :
where Xit is a vector of variables (i.e. PCM) interacting, through 2 parameter, with the uncertainty effects on planned investments. The general model is:
141
122
111 1
3
ititZ
itK
itI
itK
itYtu
itX
itK
itYtti
itK
itIt
itit X22,2
16
Significance of the interactions
Xit = PCMit PCMit PCMit and OPst PCMit and WTit
Parameters:
1 0.0302 0.0305 0.0300 0.0305(0.0134) (0.0137) (0.0132) (0.0136)
2 0.0050
(0.0124)
3 0.0093 0.0091 0.0098 0.0087(0.0536) (0.0537) (0.0533) (0.0535)
2 -0.3405 -0.3037 -0.5088 -0.4504
(0.1493) (0.1109) (0.2059) (0.1825)
2 0.0175 0.0259(0.0138) (0.0248)
RMSE 0.1790 0.1790 0.1787 0.1788R2 0.1351 0.1349 0.1382 0.1371
17
PCM interaction with… a
lph
a_2t
est
imat
es
PCM-.05 0 .05 .1 .2 .3
-.2
-.1
-.032
0
.1
Year1996 1997 1998 1999 2000 2001 2002 2003 2004
-.05
-.04
-.032
-.02
-.01
alp
ha_
2t e
stim
ates
PCM-.05 0 .05 .1 .2 .3
-.2
-.1
-.032
0
.1
Year1996 1997 1998 1999 2000 2001 2002 2003 2004
-.05
-.04
-.032
-.02
-.01
Openness
stitit OPPCM 22,2 ˆˆˆ
Turnover
stitit WTPCM 22,2 ˆˆˆ
year
PCM
18
Main findings - 1
Overall, the finding about a negative effect of the uncertainty on investments is confirmed by our results. However, such effect weakens with less market power, more competitiveness and more labour flexibility.The economic environment in Italy has greatly changed: Euro adoption Stronger competition of new industrialised countries More flexible labour marketTogether, these changes lead to a higher elasticity of investment plans to expected
demand
19
Main findings - 2
GDP Growth and Propensity to I nvest
0,07
0,08
0,09
0,1
0,11
0,12
0,13
1980 1986 1992 1998 2004
-2
-1
0
1
2
3
4
5
ΔGDP
I/GDP
the propensity to invest (non residential investments on GDP) …
In fact, since the beginning of this century, it seems that the minimum level of expected demand necessary to trigger investment has lowered. At macroeconomic level …
… during the last cycle did not fall, as it did in the previous two recessions.