Aggregation Issues of Foreign Direct Investment Estimation in an Interdependent World

• Author: Ayça Tekin‐Koru,Helen Naughton,Pehr‐Johan Norbäck
• Published date: 01 December 2016
• Date: 01 December 2016
• DOI: http://doi.org/10.1111/twec.12388

Aggregation Issues of Foreign Direct

Investment Estimation in an

Interdependent World

Helen Naughton

1

, Pehr-Johan Norb€

ack

2

and Ayça Tekin-Koru

3

1

Department of Economics, University of Montana, Missoula, MT, USA,

2

Research Institute of

International Economics (IFN), Stockholm, Sweden, and

3

Department of Business Administration, TED

University, Ankara, Turkey

1. INTRODUCTION

FROM David Ricardo to the present day, economics literature has explained the existence

of international trade using differences in technology, factor endowments, tastes or some

combination of these. This two-century long journey produced many theories resulting in reg-

ularities that apply to reasonably homogeneous groups of products, firms, industries or even

countries. However, there exists considerable heterogeneity at every level. Different entities

behave differently across different boundaries, space and time. In terms of empirical research,

this means recognising aggregation problems and perhaps offering solutions.

Heterogeneity across multinational corporations (MNCs) conducting foreign direct invest-

ment (FDI) is very extensive as well,

1

and there is no easy or obvious fix to dealing with

aggregation problems in FDI. However, the massive increase in the availability of data on

individual firms over sequential time periods coupled with the rapid rise in computing power

makes it possible to implement models that are necessary for understanding the impact of

FDI heterogeneity in aggregation.

This paper attempts to understand FDI heterogeneity and offers useful insights about aggre-

gation issues in FDI estimations by carrying out an empirical analysis with rich, affiliate-level

data on sales activities of Swedish MNCs around the globe in manufacturing sectors from

1965 to 1998.

Historically, the empirical FDI literature has used data combined over FDI types (horizon-

tal, vertical, complex) often aggregated to country level and limited to a subset of the world

mostly due to data unavailability. Running regressions on aggregate data is appropriate if one

is interested in the aggregate model. However, most of the FDI theory is written for the firms

choosing affiliate-level activity. In this case, the researcher faces the widely recognised prob-

lem of ecological fallacy, which occurs when results based on aggregate, grouped data are

applied to the individual entities that form the groups being studied (Openshaw, 1984). The

modifiable areal unit problem (MAUP) is perhaps one of the most well-known forms of eco-

logical fallacy (Harris, 2006). MAUP occurs when results change depending on the definition

of modifiable and arbitrary areal data units.

MAUP is composed of the scale problem and the zoning problem (Fotheringham

and Wong, 1991). The scale problem arises when the same data set is aggregated at different

scales in a system where these units are modifiable. The textbook example from the

1

See Melitz (2003), Helpman et al. (2004), Greenaway and Kneller (2007), Melitz and Ottaviano

(2008), and Yeaple (2009) for a review of firm heterogeneity in FDI.

©2016 John Wiley & Sons Ltd

2046

The World Economy (2016)

doi: 10.1111/twec.12388

The World Economy

geography literature is the successive aggregation of individuals into postcodes, neighbour-

hoods, regions and districts. In our case, the scale problem refers to the variation in results as

units of observation (affiliates) are aggregated into fewer and larger units (firms, industries,

countries) for analysis. The zoning problem arises because multitude of zones can be formed

and used at any given scale by the researcher with different results being obtained by simply

changing the boundaries of the zones. The zoning problem may not directly apply to our case

because affiliates, in the empirical FDI literature, are not aggregated into arbitrary entities by

researchers. Most often than not, they are aggregated to firms, industries and countries. There-

fore, in this paper, we examine the prevalence only of the scale problem in the FDI estima-

tions.

Using spatial econometrics and data on Swedish MNCs’ affiliate-level sales activity, we

first evaluate evidence of different motivations for FDI. As proxy for complex FDI theory, we

use third-country sales; as proxy for horizontal FDI, we use host–country sales; and as proxy

for vertical FDI, we use exports back to Sweden. Affiliate, firm and country-level analyses

then offer links to previous literature and determine the consequences of using aggregated

data.

Our results indicate that the multilayered nature of aggregation in FDI matters for empiri-

cal analysis. Affiliates are nested within a hierarchy and FDI is determined by both the char-

acteristics of the affiliate and the level of nesting. For host-country and third-country affiliate

sales, which dominate affiliate sales, we find a negative spatial lag that supports the export-

platform theory. For affiliate exports back to Sweden, we find a positive spatial lag providing

evidence of agglomeration of vertical production activities.

In this paper, inspired by a widely recognised aggregation problem in the geography litera-

ture we hope to increase our understanding of FDI heterogeneity and offer useful insights

about aggregation issues in FDI estimations. The potential contributions of this paper to the

economics literature are twofold.

First, we provide evidence of a severe scale problem in the FDI estimations when micro-

data are sequentially aggregated from affiliate level to firm and country level. Bria nt et al.

(2010) have shown that scale effect jeopardises the estimations of gravity equations more than

that of wage equations in their paper where they investigate the magnitude of the distortions

arising from MAUP using French zones. Taking it up to the international level and using glo-

bal network of Swedish MNCs, at the affiliate level, we estimate negative statistically signifi-

cant spatial spillovers that decrease in magnitude when data are aggregated to firm level and

finally disappear or flip to positive signs in country-level analyses.

Second, we contribute to a new line of FDI literature, which tries to uncover mechanisms

through which multiproduct multinationals behave in global markets (Baldwin and Ottaviano,

2001; Eckel and Neary, 2010; Yeaple, 2013; Eckel et al., 2015). In particular, we expand the

literature on multiproduct firms to understand the mechanisms through which the aggregation

bias in FDI estimations emerge, and present a simple multiproduct multinational firm model

of export-platform FDI and show that aggregation of individual units into larger units in the

estimations hides the effects of intrafirm competition which causes variation in results from

one scale to the other.

More specifically, we find evidence for aggregation bias for local sales and exports to third

markets, which are proxies for horizontal FDI and export-platform FDI. For exports back to

Sweden, which can be viewed as proxy for vertical FDI, aggregation bias is less apparent.

However, since local sales and exports to third countries dominate MNEs’ host country sales,

©2016 John Wiley & Sons Ltd

AGGREGATION ISSUES OF FDI ESTIMATION 2047