Bases de datos difusas

First, the equation 4.2 with the peers defined by vector priorsM_lmt is estimated. The

results are collected in Table 4.7. Note, that in this Table and all afterwards the spillover variable is divided by thousand for easier readability.

In column (1) only the spillover variables are included. The results imply a positive correlation between importing a specific machine and the number of past importers. The relationship is significant for municipality peers and also for those in the micro- region and the county. The value of municipality spillovers suggest that an additional firm having already imported machine m increases import probability for firm i by 0.08 percentage points. Compared to the average propensity of importing machine is about

1 percent,18_{our results mean an 8 percent increase in the probability of machine import}

in a given year. The number of peers not in the same micro-region but county have a

somewhat higher effect.19

In columns (2) to (4) I add additional control variables as column (1) estimates can be biased due to missing firm- and location-specific variables. First, firm level lagged controls such as size, foreign ownership, past trade experience dummies and productivity are added. All controls are significant and are of expected signs. Foreign owned firms, larger firms and more productive firms are more likely to adopt foreign machinery via imports. Firms that have trade experience, exporters and importers are also more likely to import capital items. In column (3) controls for observable characteristics of the firms’ immediate environment are added: number of firms and the log of local employment. Results suggest that the firms in larger cities and in larger labor markets

are more likely to import.20 _{These finding are in line with descriptive statistics of}

subsection 4.3.3. The variable expressing the number of firms is insignificant. In column (4) variables characterizing firms financial situation are added. Results suggest that

18_{The probability of import is this low because in this section we consider all firms in the examined sectors to be a}

possible importers, not only those who actually are going to import.

19_{This, at the first sight surprising result, comes from the special position of Budapest. It holds one fifth of the}

population and a correspondingly large share of firms. However, it is a city, a microresion and a county in itself which means that for all these firms the second spillover variable is by definition zero. When Budapest firms are excluded (Table D.2 in the Appendix) the main results still stand and peers in farther away locations, in other micro-regions have a smaller or an equal sized effect.

20_{These allow to compare the spillover results. One additional peer has one-third of the effect of being an exporter,}

CEU

eTD

Collection

Table 4.7: Machine import spillover estimation

Dep. var: import dummy [1] [2] [3] [4]

num. of prior importers of the same machine

same NUTS4 0.822*** 0.675*** 0.897*** 0.859***

[0.0615] [0.0270] [0.0387] [0.0416]

NUTS3, other NUTS4 2.336*** 2.181*** 2.066*** 2.038***

[0.247] [0.198] [0.130] [0.130] dummy: exporter 0.00367*** 0.00363*** 0.00365*** [0.000296] [0.000298] [0.000311] dummy: importer 0.00533*** 0.00543*** 0.00517*** [0.000365] [0.000286] [0.000314] size (logs) 0.00263*** 0.00250*** 0.00114*** [0.000291] [0.000245] [0.000190]

dummy: foreign own. 0.0111*** 0.0109*** 0.00987***

[0.00112] [0.00109] [0.00116]

TFP (logs) 0.000316 0.000392** 0.000486**

[0.000227] [0.000171] [0.000201]

local employment (logs) 0.00263*** 0.00251***

[0.000456] [0.000448] local # of firms 0.00001 0.00006 [0.000143] [0.000133] firm age -0.00032*** [4.62e-05] return on equity 0.00286*** [0.000393] debt on assests 0.0001 [9.84e-05] depreciation rate 0.00285*** [0.000336]

constant: yes yes yes yes

dummy: year yes yes yes yes

Observations 1714005 1278853 1278853 1267975

R-squared 0.003 0.009 0.01 0.01

*** p < 0.01, ** p < 0.05, * p < 0.1. Moulton corrected s.e. in parentheses

Each columns contain results from three separate linear probability regressions as defined in eq. 4.2. Spillover variables are divided by 1000, all other control variables are lagged by one year.

younger, more profitable and firms with higher capital replacement are more likely to import capital goods. These are in line with expectation. All in all, I find that none of the added control variables change the initial results in column (1).

To see, how the results depend on the number of past years taken into account when defining the spillover variables, I replicate the last column of Table 4.7 with various definitions of N. I find that as N increases, so do the coefficient estimates gets smaller. The significance, sign and interpretation do not change. See Table D.3 in the Appendix.

Testing the results on machine import spillovers against alternate hypotheses

The findings so far can be explained by alternative hypotheses. Table 4.8 offers re- gression results testing if these hypotheses can also explain our findings. Columns (1) and (2) look into the effects on local time-invariant unobservables, while the next two columns of the table look into the how including controls for local business cycles affect

CEU

eTD

Collection

our findings. In other words, columns (1) and (2) look at equation 4.3 type regressions, while columns (3) and (4) looks into equation 4.4.

Table 4.8: Machine spillover LSDV regressions: testing alternate hypotheses

dep. Var: import dummy [1] [2] [3] [4]

num. of prior importers of the same machine in

same NUTS4 0.934*** 0.939*** 0.939*** 0.987***

[0.0371] [0.0320] [0.0394] [0.0221]

same NUTS3, other NUTS4 2.187*** 2.446*** 2.217*** 2.530***

[0.132] [0.133] [0.139] [0.145]

dummy: year yes yes yes yes

dummy: nuts4 yes yes yes

dummy: nuts4 × sector yes

dummy: nuts4 × year yes

dummy: firm × year yes

Controls: yes yes yes yes

Observations 1278777 1278777 1278777 1278777

R-squared 0.011 0.015 0.021 0.003

*** p < 0.01, ** p < 0.05, * p < 0.1, Moulton corrected s.e. in parentheses

Controls: size, foreign ownership, TFP, local agglomeration, number of firms in NUTS5, all controls are lagged by one year. Each column shows results from separate regressions. The first two regressions test for the effect of time invariant unobservables, while the next two test for the effects of local business cycles.

Column (1) shows that including location fixed effects do not change our basic inference. In column (2) location × sector fixed effects are included to control for the effect of unobserved local benefits for certain industries. To control for local business cycles, in column (3) location × year fixed effects, while in column (4) firm × year fixed effects are included. Note that as firm dimension defines both location and sector dimensions, column (4) implicitly controls for these cross-terms as well. Results do not change the previous findings.

In addition, to the alternative hypotheses I also check for the possibility that a firm moves into a location where it expects that adopting a specific machine will be easier. In order to control for this possibility I re-run the last regression of Table 4.7 on subsample of firms established before the year our sample starts. This avoids self-selection into a location that, in the post-transition era, is abundant of future importers of m. Table D.4 collects regressions on subsamples that contain firms started business before 1992, 1990 or 1988. I find positive significant correlation between firms’ capital import propensity and the presence of past importers for pre-transition firms too.