«The Effectiveness of Industrial Policy in Developing Countries: Causal Evidence from Ethiopian Manufacturing Firms Tewodros Makonnen Gebrewolde, ...»
Given our sample is relatively large, and contains the universe of ˜ ˆ manufacturing ﬁrms, it is reasonable to expect β ∆ = Λ − β 0 if there were no ﬁrm entry. Thus β ∆ 0 (conversely, β ∆ 0) implies entering ﬁrms are more (less) productive than existing ﬁrms. Standard 8 Ifthe productivities did not vary over time then the instrument relevance assumption would also be violated.
errors are obtained via the bootstrap. Column 2 of Table 4 presents the results and shows that the productivity impact of ﬁrm entry to be negative and signiﬁcant as suggested by the theory. To see this, note that ∆ ∆ ∆ βτ 1 + βτ 2 + βτ 3 = −2.53, which is signiﬁcant at all conventional levels.
Interestingly, ﬁrms entering only due to the geographical treatment are more productive on average than other new ﬁrms. Whether, this reﬂects a positive eﬀect of the policy or some other factor is unclear. Notably, a calculation of the the average impact of the geographical treatment across both treated and untreated sectors shows it to be equal to −0.75, suggesting an overall negative eﬀect of the treatment. Note, that whilst productivity has fallen, output has increased. In the long-run, the presence of additional low-productivity ﬁrms may eventually impede growth, but the associated increase in output may be important in the short-run. However, the cost estimates presented in Section 7 suggest that this output increase has come at a substantial ﬁscal cost.
Indirect Increases in Productivity Due to Spillovers
As described in Section 2, one margin on which IP might improve productivity is through spill-over eﬀects. That is, if there are more or larger companies operating in an industry there may be more innovations to emulate, downwards pressure on input prices, for instance, thus leading to faster TFP growth. This is particularly true at Ethiopia’s current level of development where production techniques often lag signiﬁcantly behind those used in richer countries but are improving rapidly. Thus, the importation of new techniques and innovation may both be expected to be common. The framework in Section 2 suggests that if φ 3 then ∂A 0 and the policy would thus raise the productivity of both treated ∂τ and untreated ﬁrms approximately equally, meaning our DDD strategy would identify no eﬀect.
To estimate the extent of agglomeration externalities we consider the eﬀects of the presence of treated ﬁrms on nearby ﬁrms in untreated sectors. We do this by exploiting the rich geographic detail of the data and contrast productivity in untreated sectors in those Zones with few treated ﬁrms to those with more. We generate a geographic proximity variable for each non-treated ﬁrm measured by the number of treated ﬁrms in the same Zone, Nzt. For ease of interpretation Nzt is standardized, and thus the coeﬃcient κ describes the percentage impact on productivity of a standard deviation increase.We include Zone and year ﬁxed eﬀects, ψt and φz, and a vector of controls Xit as in (6). Thus κ is identiﬁed using the within variation of ﬁrm productivities as the number of treated ﬁrms with their zone varies. Our regression model is
yit = κNzt + βXit + φz + ψt + µi + (11) it We ﬁnd, as Column 1 of Table 4 reveals, that there is evidence for a small but positive eﬀect on existing ﬁrms in untreated sectors.
Speciﬁcally, a 1 standard deviation increase in the number of treated ﬁrms increases the productivities of ﬁrms in other sectors by 0.15%.
However, when we also consider entering ﬁrms in Column 2, this eﬀect is smaller and less precise.9 Thus, whilst there is some evidence of agglomeration externalities for existing ﬁrms, there is no evidence that these would be suﬃcient to oﬀset the deleterious eﬀects of the policy on average productivity. There are several reasons why spill-over eﬀects may not be larger.10 Ferracci et al. (2014) ﬁnd evidence, in the rather diﬀerent context of the French labour market, for the opposite mechanism – that large-scale training programmes in a given local labour market may lead to crowding-out eﬀects. Here, an equivalent explanation might focus on the limited ability of local markets to absorb additional production. Alternatively, it could be that the low density of manufacturing in Ethiopia itself limits the scale of agglomeration eﬀects. Evidence for these explanations would not alter our ﬁnding of the policies ineﬀectiveness, but might suggest that a similar policy might work better in a more industrialised country.
9 These ﬁndings are robust to a variety of alternative speciﬁcations of (11).
10 Blonigen (2015) considers the eﬀect of IP targeting steel-producers on users of steel and ﬁnds a negative eﬀect on their export competitiveness. Here, however, the treated sectors are by design those that use largely bulky agricultural products, producing products for domestic consumption so we should not necessarily expect this form of negative spillover.
Decreases in Productivity Due to Diversiﬁcation
One important way in which ﬁrms grow is through diversiﬁcation (Berry, 1971). In Table 4 the coeﬃcient on diversiﬁcation is consistently positive, that is more diversiﬁcation is associated with lower productivity. However, one reading of the model is that tax-breaks will lead to additional diversiﬁcation, and that this will lower productivity within existing ﬁrms. The model in Section 2 does not describe multi-product ﬁrms speciﬁcally, but note that instead of a continuum of potential ﬁrms, we can imagine the speciﬁcation describing one ﬁrm potentially producing a continuum of individual products.
Then, our expectation is that the IP will have induced ﬁrms to diversify, and thus that this is one way in which the policy led to lower average productivity. Column 3 reports the results of estimating a similar speciﬁcation as in (6) except now we move our diversiﬁcation measure to the LHS. We ﬁnd that overall eﬀect of the policy is negative, that is it increased diversiﬁcation. Speciﬁcally, τ2 is signiﬁcant at the 10% level and τ3 is insigniﬁcant but relatively precisely estimated.
Testing the joint signiﬁcance of τ we are able to reject the null of no overall eﬀect at all levels.
6.1 Eﬀects on Capital We have now seen that the policy was unsuccessful in encouraging productivity growth. We also seen that this is because as predicted by the theory, the new ﬁrms were less productive, and there were insuﬃcient spillovers to oﬀset this. We now consider the key mechanism by which ﬁrms were to be aﬀected – cheaper capital. One might be contented, as governments often are in rich countries, with a policy that was at least successful in increasing capital levels and employment rates.
We now see that the policy was also unsuccessful when judged on these criteria. Whilst, the provision of tax-breaks and subsidised loans did indeed increase capital levels, we ﬁnd that this increased capital was normally used for investments other than new machinery necessary for greater or more eﬃcient production, but rather in buildings or vehicles.
Furthermore, we show that this can be understood as a hedge against inﬂation and changes in market conditions given rampant inﬂation and a dynamic but challenging business environment. We then show, that as suggested by the theory, the lack of investment in productive assets limited employment growth due to the policy.
Direct Increases in Capital Due to Subsidies
Both our intuition, and Section 2 suggest that treated ﬁrms should increase investment as the policy lowers the cost of capital. Column 4 of Table 5 reports the results of again estimating (6); but, now with ﬁrms’ total book-capital on the left-hand side. The results suggest that ﬁrms in the treated sectors increased their capital levels, and that those treated by both arms of the policy did by slightly more;
but, the geographical component of the treatment was associated with lower than average capital accumulation. This latter ﬁnding suggests that owners of ﬁrms preferred to take additional proﬁts rather than reinvest. This might explain the results in Column 2 of Table 4 – that the tax-break discouraged capital accumulation. This suggests that the subsidised loan programme that was a large part of the sectoral treatment was more eﬀective at increasing capital levels than the taxbreaks. Testing the overall eﬀect of the policy we can rule out that the policy did not increase capital levels, and thus on this basis may be judged as successful.
Increases in Capital are Not Invested in Machinery
Column 5 reports that despite the increases in Capital there were no overall eﬀects on the Marginal Product of Capital; this is surprising as we would expect that a large increase in the capital stock should be reﬂected in a decrease, other things equal, in the marginal product.11 Column 6 reports estimates with the ratio of machinery to overall capital on the left hand side and documents that the sectoral treatment, led to a decrease in this ratio. This implies that new investments occasioned by the policy were in other forms of capital such as buildings and vehicles.
11 This results also suggests that the policy is not encouraging growth by reallocating capital. If it were we would expect a large positive and signiﬁcant coeﬃcient here.
We do not adjust for quantities sold of these products to avoid potential endogeneity bias due to responses in production decisions due to changes in prices or vice-versa. We then estimate the following
12 As discussed by Lencho (2008), Ethiopian Law does provide a Bankruptcy procedure; but, the law has rarely been applied since 1960, and most lawyers are unfamiliar with it.
ln(machinery) = βln(bookcapital) + γln(T oT ) + µi + (13) it.
The results are reported in Column 7. In line with our hypothesis we ﬁnd that the ratio of capital in machines, etc., to total book capital is higher when the ‘terms of trade’ of a particular ﬁrm are higher. This highlights the challenges in designing successful IP – this behaviour is the upshot of several interrelated features of the particular context.
Firstly, the high-growth high-inﬂation environment means that ﬁrms will seek to avoid holding cash whilst being willing to incur debt.
Second, entrepreneurs will be more risk-averse due to the lack of eﬀective bankruptcy protection. Finally, the absence of a well-developed ﬁnancial services sector means that ﬁrms are unable to diversify, through acquisition, for example; thus, we get the accumulation of unproductive capital. However, these three factors are not unique to Ethiopia and neither, therefore, are the diﬃculties they suggest in the encouragement of investment.
6.2 Eﬀects on Employment The ﬁnal outcome variable we consider is employment. The theoretical framework discussed above suggests that the ﬁrm-level eﬀects of the IP on employment will depend on the relative magnitudes of the substitution and scale eﬀects. Column 8 of Table 5 shows that there was no overall eﬀect of the policy on employment. Again, we observe a negative eﬀect of the geographical treatment, whether this reﬂects the failure of the tax-breaks to lead to additional capital accumulation is unclear. But, the positive and signiﬁcant coeﬃcient on (log) Total Book Capital suggests that this may be the case.
7 The Cost of the Policy Rigorous policy evaluation techniques are by now routinely applied to assessing the eﬀectiveness of diﬀerent forms of aid at both a macroeconomic level, and also at the level of individual policies. Many development agencies and charities are committed to funding projects only based on evidence that they represent value for money. This suggests that IP should be evaluated on a similar beneﬁt–cost basis.
Given that we ﬁnd little evidence of any positive eﬀects of the policy, we could assume the policy had no beneﬁts and focus on its costs. Instead, more conservatively, we prefer to assume the policy had the maximum plausible impact – the maximum of the 99% conﬁdence interval of each of τ1, τ2, τ3. Thus, we evaluate the policy on the premise, that contrary to our results, it achieved an 83% increase in TFP. We also take into account the increase in the tax base due to additional entry of ﬁrms due to the policy. We do this by comparing the number of ﬁrms that entered in treated sectors to untreated sectors and use the diﬀerence as the number of ﬁrms caused by the policy. Again conservatively, we assume that all of the additional new ﬁrms in treated sectors are because of the policy.
Following the the arguments in Section 2, and the results in the previous section, we assume that the least productive entrants are those induced by the policy. Thus, following the notation in Section 2, the proﬁt of ﬁrm i is Πi. Denote the set of existing ﬁrms as X and the set of additional
entering ﬁrms as E beneﬁts in year t, Bt are given by:
where T1 is the tax rate for ﬁrms treated by the policy and T0 is the taxrate without it. We take a similarly conservative approach to the costs of the policy. We focus only on the loss of tax-revenue although this focus will understate the cost of the policy substantially as it ignores the costs of concessionary loans and the investment in sector speciﬁc training and technology transfer programmes. In particular, the costs of the loans will be substantial, given real interest rates were far below zero. We ignore both of these other costs as the cost of the loans will depend on future delinquency rates as well as future inﬂation, and there is no data on the costs of training and technology transfer. Costs are given by the
loss of tax revenues on existing ﬁrms:
Figure 3 plots the lost tax receipts due to the policy – the blue line – and the additional tax due to TFP growth and ﬁrm entry – the red line –by year. The cost ranges from $39.4 Million (358 Million Birr) to over $121 Million (1100 Million Birr). Put diﬀerently, the average cost over the period was 0.5% of GDP or 5% of total Government spending.