Conditional technology spillovers from foreign direct investment: evidence from Indian manufacturing industries

Abstract

This paper examines the productivity effect of technology spillovers via vertical linkages through foreign direct investment (FDI) in India. An analysis of firm-level panel data from the Indian manufacturing sector from 2000–2001 to 2007–2008 employing the semi-parametric method of Levinsohn–Petrin to correct the endogeneity bias in productivity estimation shows productivity improvements in domestic firms due to vertical technology spillovers are only through backward linkages from FDI. The study confirms that firms in high-technology industries benefit more from vertical technology spillovers from foreign firms compared to firms in low-technology industries. It also shows that minority-owned foreign firms are more prone to technology spillovers than majority-owned foreign firms. Nonetheless, domestic firms in high-technology industries are able to access both horizontal as well as vertical technology spillovers from majority-owned foreign firms. The paper concludes that technology spillovers from FDI are not spontaneous, but are constrained by the technological ability of domestic firms and the ownership structure of foreign firms.

Appendices

Appendix 1: Capital stock estimation

For the estimation of capital stock we have employed the perpetual inventory method of Srivastava (1996). Balakrishnan et al. (2000) have also used this method to estimate capital stock for their study. The method goes in the following manner

$$K_{t + 1} = K_{t} + I_{t + 1}$$

$$K_{t} = K_{t - 1} + I_{t - 1}$$

$$K_{t - 2} = K_{t} - I_{t} - I_{t - 1}$$

and so on.

Here, $K_{t}$ is the capital stock at replacement cost in the base year t and $I_{t}$ is the investment in the year t. The year 2004–2005 is taken as the base year because we have the maximum number of observations for this year. We have hence converted the reported $GFA_{2004 - 05}^{h}$ into $GFA_{2004 - 05}^{r}$ using a revaluation factor computed using the procedure given in Srivastava (1996) (h and r denote historical costs and replacement costs respectively).

1.1 Capital stock at replacement cost in the base year

Given the available data, there is no perfect way to estimate capital stock at replacement cost in the base year and any method used is an approximation. Our method of estimation is based on the following assumptions.

• (a) No firm has any capital stock in the base year (2004–2005) of a vintage earlier than 1985-86. The year 1985–1986 itself is chosen because the life of machinery is assumed to be 20 years, as noted in the report of the Census of Machine Tools (1986) of the Central Machine Tool Institute Bangalore (National Accounts Statistics: Sources and Methods’ New Delhi: Central Statistical Organization, 1989). For firms incorporated before 1985-86 it is assumed that the earliest vintage capital in their capital mix dates back to the year of incorporation. Clearly, as stated by Srivastava (1996) the year of incorporation and the vintage of the oldest capital in the firm’s asset mix may not coincide for some firms, but the assumption is made for want of a better alternative.

• (b) The price of capital changes at a constant rate $\pi = \frac{{P_{t} }}{{P_{t - 1} }} - 1$ from 1985–1986 or the year of incorporation up to 2004–2005 (base year). The values for $\pi$ are obtained from a series of price deflators constructed from CSO’s data on gross fixed capital formation of manufacturing published in various issues of National Accounts Statistics (NAS) of India.

• (c) Investment changes at a constant rate $g = \frac{{I_{t} }}{{I_{t - 1} }} - 1$. Here the rate of growth of gross fixed capital formation in manufacturing at 1993-94 prices is assumed to apply to all firms. Again different average annual growth rates are obtained for firms established after 1985–1986.

Using the values of $\pi$ and g, we arrive at a revaluation factor (${\text{R}}^{\text{G }}$) to arrive at measure of capital stock at replacement cost for the base year. Let us denote $GFA_{t}^{h}$ and $GFA_{t}^{r}$ as gross fixed asset at historical costs and replacement cost respectively and $I_{t}$ is the real investment at time t. The $GFA_{t}^{h}$ can be defined as

$$GFA_{t}^{h} = P_{t} I_{t} + P_{t - 1} I_{t - 1} + P_{t - 2} I_{t - 2} + \ldots$$

which can be rewritten as

$$GFA_{t}^{h} = {\mathbf{P}}_{t} {\mathbf{I}}_{t} \left( {\frac{{\left( {1 + g} \right)\left( {1 + \pi } \right)}}{{\left( { 1+ {\text{g}}} \right)\left( {1 + \pi } \right) - 1}}} \right)$$

and similarly $GFA_{t}^{r}$ can be written as

$$GFA_{t}^{r} = P_{t} I_{t} + P_{t} I_{t - 1} + P_{t} I_{t - 2} + \ldots$$

which can be rewritten as

$$GFA_{t}^{r} = {\mathbf{P}}_{t} {\text{I}}_{t} \left( {\frac{{\left( {1 + g} \right)}}{g}} \right)$$

Now if the earliest vintage of capital dates back infinitely, the revaluation factor $R^{G}$, defined as the ratio of the $GFA_{t}^{r}$ to the $GFA_{t}^{h}$, will be

$${\mathbf{R}}^{G} = \frac{(1 + g)(1 + \pi ) - 1}{g(1 + \pi )}$$

(The superscripts r and h denote values at replacement and historical cost respectively). In this study we have assumed that the capital stock has finite economic life. Now the revaluation factor becomes

$${\mathbf{R}}^{G} = \frac{{[(1 + g)^{\tau + 1} - 1](1 + \pi )^{\tau } [(1 + g)(1 + \pi ) - 1]}}{{g\{ [(1 + g)(1 + \pi )]^{\tau + 1} - 1\} }}$$

where $\tau$ is the life of the machine.

Now we can scale up the GFA in the base year by the revaluation factor to obtain an estimate of the capital stock at replacement costs at the current prices. We then deflate the capital stock at replacement cost by the wholesale price index for machinery and machine tools (with base 1993–1994 = 100) to arrive at a measure of the capital stock in real terms for the base year.

It is to note that we have used gross fixed asset rather than net fixed asset. Dennison (1967) argues that a true measure of capital stock falls somewhere between the gross and the net stock of capital, hence advocates the use of a weighted average of the two with higher weight for the gross as the true value is expected to be closer to it. This is however not feasible empirically as the reliable data on accounting and economic rate of depreciation are not available in India.

Appendix 2: Methodology for industry × industry coefficient matrix

For our study, we need to construct an $industry \times industry$ coefficient matrix using the “Input–Output Transaction Table of India” of 2003–2004, published by the Central Statistical Organization (CSO). The Input–Output Transaction Table consists of two matrices—the absorption matrix (commodity–industry), and the make matrix (industry–commodity). The former records the value of purchases of commodities by industries and the latter records the value of commodities produced by industries. There are two basic assumptions, which combine both information in the make and absorption matrices to estimate a “pure” table of $industry \times industry$ or $commodity \times commodity$ (Input–Output Tables and Analysis 1973). They are generally referred to as the commodity technology and industry technology assumptions. The former assumes that a commodity has the same input structure in whichever industry it is produced. The industry technology assumption, on the other hand, assumes that all commodities produced by an industry are produced with the same input structure and thus commodities will have different input structures depending on the industry in which they are produced.

The following briefly gives the methodology in mathematical terms for constructing “pure” tables. The basic data available from industry input and output tabulations satisfy the following relationships:

$${\text{Input}}\;{\text{relations}}:q_{j} = \sum\limits_{K} {X_{jk} } + f_{j}$$

$${\text{Output}}\;{\text{relations}}:q_{j} = \sum\limits_{i} {M_{ij} }$$

$$g_{i} = \sum\limits_{j} {M_{ij} }$$

where $q_{j}$ = total output of the jth commodity group, $g_{i}$ = total output (of all products and by-products) of the ith industry group, $f_{j}$ = final demand of the jth commodity, $X_{jk}$ = output of the jth commodity used as input in the kth sector (industry group), $M_{ij}$ = output of the jth commodity produced by the ith industry group. The above symbols without subscript refer to the corresponding vectors.

We can put all the mathematical expressions of the input–output relationships explained above into a simplified accounting framework (see Table 4).

Table 4 Input–Output Relationships

Given the industry technology assumption, the industry × industry coefficient matrix can be constructed using the above accounting data. Symbolically, it is defined as follows.

$$E = DB$$

where E is the $industry \times industry$ coefficient matrix, D is the market share matrix, the columns of which show proportions in which various industries produce the total output of a particular commodity. Symbolically, it is as D = M (q)$^{ - 1}$, and B is the $commodity \times industry$ coefficient matrix, defined as B = X (g)$^{ - 1}$. Here q is the diagonal matrix with diagonal elements as the elements of vector q and similarly g is the diagonal matrix with diagonal elements as the elements of vector g. For constructing the $industry \times industry$ coefficient matrix, we first have to aggregate the input–output transaction table for the manufacturing sector to two-digit level. Then we construct $industry \times industry$ coefficient matrix using the make and absorption matrices.

Appendix 3

See Tables 5, 6 and 7

Table 5 Summary Statistics of Variables (Domestic Firms Only)

Table 6 Variance Inflation Factor (VIF) for Eq. (2)

Table 7 Variance Inflation Factor (VIF) for Eq. (3)

Appendix 4

See Tables 8 and 9

Table 8 Concordance between NIC-1998 and Input-Output Transaction Table (IOTT) 2003–2004

Table 9 Classification of manufacturing industries by technology