Risk Sharing and Industrial Sp ecialization:
Regional and International Evidence
Sebnem Kalemli-Ozcan
Brown University
Bent E. Srensen
Brown University
Oved Yosha
Tel Aviv University
April 1999
Abstract
We provide empirical evidence that risk sharing enhances specialization in production.
To the b est of our knowledge, this well-established and imp ortant theoretical prop osition
has not b een tested b efore. Our empirical pro cedure is summarized as follows. First, we
construct a measure of sp ecialization in pro duction, and calculate an index of specializa-
tion for each of the European Community (EC) and non-EC OECD countries, U.S. states,
Canadian provinces, Japanese prefectures, Latin American countries, and regions of Italy,
Spain, and the United Kingdom. Then, we estimate the degree of capital market integra-
tion (a measure of risk sharing) within each of these groups of regions: the EC countries,
the non-EC OECD countries, the United States, Canada, Japan, Italy, Spain, and the
United Kingdom (and rely on another author's estimate for Latin America). Finally,we
perform a regression of the specialization index on the degree of risk sharing, controlling
for relevant economic variables. We nd a p ositive and signicant relation b etween the
degree of sp ecialization of individual members of a group of countries, provinces, states,
or prefectures, and the amount of risk that is shared within the group. We perform
regressions using variables such as shareholder rights and the size of the nancial sector
(relative to GDP) as instruments for the amountof inter-regional risk sharing. These
regressions conrm that risk sharing|facilitated byafavorable legal environment and a
developed nancial system|is a direct causal determinant of industrial sp ecialization.
JEL Classication: F15, F2, F36, F43
We thank Charles Kroll, Elena Krop, and in particular, Lauren Sella for research assistance as well
as Jacques Melitz, Frederic Zumer, and Maria Luengo-Prado for help with the Italian and Spanish data,
respectively. Elise Brezis, Jess Gaspar, Elhanan Helpman, Pravin Krishna, Eric van Wincoop, and participants
at seminars at Albany, Arizona State, Bank of Israel, Bar Ilan, Birkbeck, Brown, CEMFI, DELTA, ECARE,
Haifa, Kansas CityFederal Reserve, LSE, New York Federal Reserve, Ro chester, Royal Holloway, Rutgers,
Tel Aviv, Toulouse, Vanderbilt, at the 1999 North American Winter Meetings of the Econometric So ciety,
and in a conference on Risk Sharing and Economic Vulnerabili ty at the Joint Center for Poverty Research,
University of Chicago and Northwestern University, provided helpful comments and suggestions.
1 Intro duction
That countries gain from specialization is one of the few widely accepted economic tenets.
Gains from sp ecialization may arise from technological dierences (Ricardo), factor endow-
ments (Heckscher-Ohlin), or from increasing returns to scale.
1
These theories have tradition-
ally been formulated in non-stochastic environments. In the presence of production risk and
in the absence of markets for insuring this risk, countries that sp ecialize in the pro duction
of a small number of goods may suer a loss in economic welfare due to the high variance of
Gross Domestic Product (GDP). These countries may therefore choose not to specialize, as
noted by Brainard and Coop er (1968), Kemp and Liviatan (1973), and Run (1974).
Insurance of production risk may take many forms, from explicit insurance against ad-
verse outcomes (typically natural disasters) to forward markets where commo dities are sold
at a xed price for future delivery. The main mechanism for spreading risk among regions
and countries is, however, diversication of ownership, achieved via capital markets. If inter-
regional and international capital markets are well integrated, regions and countries, b eing
insured against idiosyncratic shocks, can aord to specialize more thereby exploiting com-
parative advantage further, whether such advantage is due to technology, factor endowments,
or economies of scale. Indeed, from the analysis in Helpman and Razin (1978a, 1978b) it
follows that country specialization will b e higher when there is international trade in b oth
securities and goo ds.
2
Helpman and Razin's analysis covers both the Ricardian case (without
insurance, countries may not fully sp ecialize in the go od they can pro duce at low unit cost),
and the Heckscher-Ohlin case (without insurance, countries will sp ecialize less in the go od
that is intensive in the factor in which they are relatively abundant), but do es not address
the case of trade driven by increasing returns to scale. A simple example provided in the
next section illustrates how their prop osition applies to this case as well.
Insurance induced specialization has potentially non-trivial consequences for economic
growth. Greenwoo d and Jovanovic (1990), Saint-Paul (1992), Obstfeld (1994a), Acemoglu
and Zilib otti (1997), and Feeney (1997) have written theoretical mo dels where capital market
integration induces higher specialization which in turn stimulates output and growth. In
Obstfeld (1994a) the basic premise is that countries choose the investment mix in risky (high
1
See, e.g., Krugman (1979) and Helpman (1981, 1984).
2
Further work on this topic include papers by Anderson (1981), Grossman and Razin (1984, 1985), Helpman
(1988), and Feeney (1994).
1
return) pro jects and safe (low return) pro jects. International asset trade allows countries
to hold a diversied portfolio encouraging them to shift investmenttowards high return
pro jects.
3
In the model prop osed by Saint-Paul (1992), the basic trade-o is between the
gains from sp ecialization due to comparative advantage in production and a lower variance
of output, while Feeney (1997) develops the idea that in the presence of learning by doing in
production, an increase in specialization entails higher growth during a transition p eriod.
To the best of our knowledge, no evidence has b een brought to bear on this issue. Hufbauer
and Chilas (1974) and Krugman (1991) demonstrated that U.S. states are more specialized
than OECD countries, and interpreted this observation as evidence that barriers to trade are
greater across countries than across U.S. states, but neither performed a systematic empirical
study of the determinants of regional specialization patterns|a task we undertake here.
Researchers (see King and Levine (1993), Levine and Zervos (1998), and Ra jan and Zin-
gales (1998) for prominent examples) have found a p ositive correlation b etween \nancial
depth" and the rate of economic growth. They have not, however, provided direct evidence
regarding the mechanism through which nancial intermediation and capital market integra-
tion promote growth. One such mechanism is higher sp ecialization in production facilitated
by b etter spreading of pro duction risk which, in turn, is more easily achieved where nancial
markets are developed and reliable. Our empirical results provide strong supp ort for this
view.
The theoretical models described ab ove p oint to the following empirical strategy: For
various groups of regions or countries (e.g., U.S. states, Japanese prefectures, Europ ean
Community (EC) countries), calculate a measure of the degree of insurance among members
of the group, and for each member compute an index of industrial specialization in production.
Then, to test the common empirical prediction of the above theories, check whether a high
degree of insurance (risk sharing) within a group of regions or countries is associated with
high sp ecialization in production of the group members, when other potential determinants
of industrial sp ecialization are controlled for. Next, nd variables which are exogenous to the
degree of specialization but likely to be correlated with the extent of observed inter-regional
risk sharing, and perform instrumental variables regression in order to determine the direction
3
A similar trade-o is modeled in Acemoglu and Zilibotti (1997) who stress the fact that developing coun-
tries have less diversication opportunities in pro duction and therefore tend to sp ecialize in safe technologies.
In Greenwood and Jovanovic (1990), nancial intermediaries po ol risks and help achieve higher and safer
returns on investment.
2
of causality.
Much of the variation in our sample is due to dierences in risk sharing and sp ecialization
across groups of regions that constitute countries (U.S. states, Canadian provinces, and so
forth), but the basic logic of our approach is b est illustrated by the striking dierence in
patterns of risk sharing and sp ecialization in groups of regions within countries versus groups
of countries.
It is, bynow,awell established empirical regularity that there is little risk sharing be-
tween countries; see French and Poterba (1991) and Tesar and Werner (1995) who document
the \home bias" puzzle, Backus, Keho e, and Kydland (1992) who compare cross-country
GDP correlations and consumption correlations, and Srensen and Yosha (1998) and Ar-
reaza (1998) who carry out cross-country variance decompositions of sho cks to GDP for
EC/OECD and Latin American countries respectively. All these studies p oint to negligible
risk sharing through cross-country ownership of assets.
In contrast, as shown by Asdrubali, Srensen, and Yosha (1996), there is substantial risk
sharing among states within the United States. Using the same metho dology, Alberola and
Asdrubali (1998) and Dedola, Usai, and Vannini (1998) found substantial risk sharing among
regions within Italy, Spain, and the United Kingdom. In this pap er we further document con-
siderable risk sharing among provinces within Canada and among prefectures within Japan.
4
If risk sharing is important for sp ecialization one would expect U.S. states, regions of Spain,
Italy, and the United Kingdom, Canadian provinces, and Japanese prefectures to be more
specialized than EC, OECD, or Latin American countries. Our empirical work conrms this
hypothesis.
The nding survives when we p erform a more detailed regression analysis with the ap-
proximately 170 regions and countries in our sample, controlling for characteristics suchas
population density and per capita GDP, and using various measures of specialization and
various sub-samples (e.g., omitting U.S. states or Latin American countries from the sam-
ple).
Finally, and p erhaps most imp ortant, the p ositive relation between risk sharing and sp e-
cialization also survives when we p erform the regressions only with regions within countries
(i.e., eliminating groups of countries from the sample). This conrms that the dierences in
4
Other work on inter-regional risk sharing includes Crucini (1999), Athanasoulis and van Wincoop (1998),
del Negro (1998), and Hess and Shin (1998).
3
specialization patterns are not (entirely) driven by dierences in the amount and composition
of, or barriers to, international versus intranational trade. It also suggests that the results are
not driven by higher factor (physical capital or labor) mobility across regions within countries
than across countries.
Exogeneityofinter-regional risk sharing through cross-holding of pro ductive assets|the
central explanatory variable of specialization in our empirical sp ecication|is an issue of
importance. In all the theoretical models cited above, insurance among economic agents
is regarded as exogenous. To justify this assumption, one mayinvoke the exogeneityof
institutions that allow agents to insure. Decisions regarding the choice of product lines,
entry to an industry,orinvestment in new technologies are taken by rm managers and
shareholders and byentrepreneurs who take insurance opportunities as given. Even if these
institutions change over time, the change is slow and does not aect the decision to sp ecialize
in production.
Although this view has merit, it is wise to allow for a dierent scenario, where the degree
of inter-regional cross-holding of productive assets
is
aected|at least to some extent|by
the degree of specialization in pro duction. One might imagine, for example, a federation
with geographic or demographic characteristics that render high regional specialization par-
ticularly attractive. Institutions that facilitate risk sharing may then develop in response to
the need for insurance raised by the sp ecialized regional production structure. In order to
obtain estimates that are robust to such \reverse causality," we use instrumental variables
estimation. Fortunately, instruments are available that are highly unlikely to b e aected by
reverse causality. In particular, we use quantitative indicators of the strength of the part of
the \legal environment," for instance protection of investor rights, that is likely to havean
impact on the amount of cross-regional ownership. In the corp orate nance literature it has
been do cumented that the legal tradition|as it impacts on, shareholder rights for example|
is a strong determinant of national stock market capitalization (the premier institution for
nationwide risk sharing). We use the legal environment indicators suggested byLaPorta et
al. (1997, 1998) as instruments. We further use the share of the nancial sector in GDP|a
direct indicator of \nancial depth"|as an alternative instrument. Our results are not sen-
sitive to what instruments we use and are similar to the OLS results, which indicates that,
to a rst approximation, risk sharing is exogenous for sp ecialization and causality runs from
risk sharing to specialization.
Constructing measures of industrial sp ecialization in pro duction is standard. By contrast,
4
