Determinants of FDI Flows to Developing Asia

This section undertakes an empirical investigation of some of the possible determinants of FDI flows to developing Asia from the OECD and the rest of the region over the period 1990–2005 using an augmented gravity model framework.

A. The Model

Our aim is to develop a relatively parsimonious model that includes specific bilateral variables as well as selected host country policy variables. In view of this, we followed the basic gravity-type framework, which argues that market size and distance are important determinants in the choice of the location of source countries for direct investments.³

The basic specification of our estimated model is outlined below:

1n(FDI_ijt) = β₀ + β₁1n(GDP_it) + β₂1n(GDP_jt) + β₃LANG_ij + β₄1n(DIST_ij) + β₅ X_ijt + α_j + δ_t + V_ijt

where FDI_ijt is the real FDI flow from source country (i) to host country (j) in time (t); GDP_it and GDP_jt are real GDPs in US dollars for the source country (i) and the host country (j) in time (t); LANG is a binary variable equal to 1 if the source and host countries have a common official language; DIST_ij is the geographical distance between the host and source countries; X_ijt is a vector of control variables influencing FDI outflows; α_j denotes the unobservable type of source country effects (source country dummies are used); α_t denotes the unobservable time effects (year dummies are used); and V_ijt is a nuisance term. Our baseline gravity model is augmented with measures of trade openness and financial openness of the host country as well as bilateral imports between the two countries.

We assumed the coefficients of the real GDP of the source and destination countries to be both positive, as they proxy for important masses in gravity models. A destination country that has a large market tends to attract more FDI. The coefficient of the source country size could either be negative or positive. While large real GDP indicates greater aggregate income and/or more companies, and therefore higher ability to invest abroad, small real GDP in the source implies limited market size and consequent desire by companies to expand their wings overseas in order to gain market share. The sign for common language ought to be positive, while the sign for distance from the source to the host country should be negative, as greater distance between countries makes a foreign operation more difficult and expensive to supervise and might therefore discourage FDI.⁴

We also added a measure for bilateral trade (i.e. imports). The idea here is that a source country could either import from the host country or choose to establish a production base there in order to sell directly to their home market. Alternatively, insofar as companies in the home market are losing market share to “cheap” imports, they may choose to relocate overseas in order to counter this competition. There may be issues of reverse causality between FDI and imports, so we lagged imports by one period. In addition to bilateral trade, in general, the more open the economy is to trade and capital flows, the more likely it is to attract FDI.⁵ We used total trade to GDP ratio as a measure of trade openess and used a normalized Chinn-Ito index (see Chinn and Ito 2007) as a proxy for financial openness.⁶ Clearly, there may be a number of other host country determinants of FDI (for instance, see Hattari and Rajan 2008). However, since our aim is to focus on the basic gravity model, and in particular, to compare the difference between OECD and developing Asian sources of FDI, we have kept the regressions fairly economical. We also included time (year) dummies to account for global changes in FDI trends, and also controlled for other source country effects to account for unobservable or omitted factors.

B. Data and Methodology

Table A1 [ PDF 24.6KB | 1 page ] and Table A2 [ PDF 24.6KB | 1 page ] summarize the data sources that are used. The FDI data are based on the UNCTAD FDI/TNC and EIU's World Investment Service databases in millions of US dollars. We deflated the FDI data by using the 1996 US consumer price index (CPI) for urban consumers. Data for real GDP and real GDP per capita are taken from the World Bank's World Development Indicators database. Imports data from the source countries to the host countries are taken from the International Monetary Fund's (IMF) Direction of Trade and Statistics (DOTS) database (although the data are limited to merchandise trade). We also deflated our export data using the 1996 US CPI for urban consumers. Data on distance and common official language are taken from the CEPII (http://www.cepii.fr/).

Our sample is based on a balanced panel dataset of annual data on 187 source-host country pairs, which consist of 24 source countries and 12 host countries between 1990– 2005 (Table 3 [ PDF 28.5KB | 1 page ]). The dataset contains a large number of missing variables for bilateral FDI (roughly 40% of the total observations) and a small number of disinvestment figures shown in the data as negative (188 observations). Excluding missing and negative observations, our panel consists of around 1,600 observations. In all of our estimations, we dealt with the issue of censored data using the Tobit model, a commonly used approach to dealing with censored data (for instance see Stein and Daude 2006; and Loungani, Mody, and Razin 2002).⁷ We followed di Giovanni (2005) by computing a Tobit model using the two-step procedure: first, a probit model is estimated based on whether a deal is observed to be conditional or not on the same right-hand variables as in equation (1), and the inverse Mills' ratio is constructed from the predicted values of the model. Second, a regression is run to estimate equation (1) including the inverse Mills ratio as a regressor.⁸

Download this Paper [ PDF 110.6KB| 17 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.