Trade and Investment under Policy Uncertainty:

Firm Evidence from Portugal�s European Integration

Kyle Handley� Nuno Limãoy

Prepared for the 5th conference of Banco de Portugal on:

�Desenvolvimento Económico Português no Espaço Europeu�

May 14, 2010

ABSTRACT: Portugal experienced profound economic changes in the period immediately following

its accession to the European Community (EC). In this paper we focus on one speci�c dimension of

these changes� international trade. There are several important reasons to do so. First, in small

open economies such as Portugal, trade is central not only to consumers but also to �rms that rely

on international markets to purchase inputs and sell their goods. These �rms employ a large fraction

of Portuguese workers and so trade has a direct impact on them as well. Second, trade increased

dramatically during this period, both in real value and in terms of the fraction of �rms involved and

their employment. Third, as we will see, this trade expansion was strongly oriented towards the EC

partners, indicating that it was caused by the accession. Finally, the export expansion upon accession

was characterized by considerable entry of Portuguese �rms into EC markets even though in some cases

there were few or no changes in applied trade barriers such as tari¤s. We will provide novel evidence that

an important reason for this was that EC membership substantially lowered trade policy uncertainty

faced by Portuguese exporting �rms.

�Email: handley@econ.umd.edu. University of Maryland.yCorreponding author. Department of Economics, University of Maryland, College Park, 20742, USA. Email: li-

mao@econ.umd.edu. Other current a¢ liations: Yale University, NBER and CEPR. The Bank of Portugal Research Group,the Portuguese Census (INE) and Mario Centeno provided valuable help with the �rm level data. I thank Yale University�sLeitner Program, which hosted me during part of this project and the UMD Economics Department for funding to digitizedata. I am grateful to Stephanie Aaronson, for helpful discussions. The usual disclaimer applies.

1 Introduction

Portugal experienced profound economic changes in the period immediately following its accession to the

European Community (EC). In this paper we focus on one speci�c dimension of these changes� international

trade. There are several important reasons to do so. First, in small open economies such as Portugal, trade

is central not only to consumers but also to �rms that rely on international markets to purchase inputs

and sell their goods.1 These �rms employ a large fraction of Portuguese workers and so trade has a direct

impact on them as well. Second, trade increased dramatically during this period, both in real value and in

terms of the fraction of �rms involved and their employment.2 Third, as we will see, this trade expansion

was strongly oriented towards the EC partners, indicating that it was caused by the accession. Finally,

the export expansion upon accession was characterized by considerable entry of Portuguese �rms into EC

markets even though in some cases there were few or no changes in applied trade barriers such as tari¤s.

We will provide novel evidence that an important reason for this was that EC membership substantially

lowered trade policy uncertainty faced by Portuguese exporting �rms.

Firms face considerable uncertainty about future conditions a¤ecting their costs, demand and pro�tabil-

ity. This uncertainty arises both from purely economic shocks, e.g. changes in productivity, or tastes, as

well as policy shocks, e.g. macro-policies a¤ecting aggregate demand and production costs or reforms of

taxes and regulations a¤ecting goods, investment or pro�ts. The role of future conditions is particularly

important when the �rm is deciding to make investments that are costly and irreversible such as adopting

a technology, producing a new good or selling in a new market. In these cases �rms may opt to wait for

current conditions to be su¢ ciently good or for uncertainty about future conditions to be su¢ ciently low

before they invest.

Our focus on the impact of trade policy uncertainty on �rm decisions is motivated by the Portuguese

experience. However, there are broader reasons why this source of uncertainty is an interesting one for

investment and entry decisions. First, trade policies and their outcomes for �rms and trade are observable.

Thus we can construct easy-to-interpret measures of uncertainty and test their implications. These results

in turn can inform trade and tax policy more broadly. Second, while average tari¤s are no longer high in

some countries, trade protection remains important once other trade policies are accounted for; moreover

1 In 1986, imports of investment goods and intermediates were equal to 20% of Portugal�s GDP value.2Between 1985 and 1992 real exports grew by 90% and imports by about 300% (Authors�calculations based on data from

Pinheiro et al, 1997). The fraction of �rms involved in trade went from 22% in 1986 to 26% in 1992 and employment in �rmsthat trade increased by about 200,000 (Authors�s calculation from merged information of Quadros de Pessoal and InternationalTrade statistics available from INE.)

1

the application of trade policy generates considerable uncertainty, as we describe below. Third, there is

growing evidence that �rms must incur substantial �xed costs before starting to trade (c.f. Roberts and

Tybout, 1997). The interaction of these costs with uncertainty about the policy a �rm will face in a new

market may lead it to delay or eliminate entry investments and not trade. To capture these e¤ects, we use a

dynamic model to derive the impacts of current and future trade policy on �rm-level investment and export

decisions. We then test the predictions of the model by combining new �rm-level data with detailed policy

data on Portugal�s accession to the European Community (EC) in 1986.

Despite the potential importance of uncertainty in �rm decisions to invest and trade, there is scant

research on this question. As we discuss in section 2, the work that addresses this issue has focussed on

exchange rate risk. However, evidence has been mixed, perhaps because exchange rate shocks are short-lived.

So another interesting motivation for considering the impact of uncertainty in policy, and particularly trade

policy, is that these shocks can be large and persistent. One possible reason why trade policy uncertainty

and its impacts are not more studied and better understood is perhaps the perception that it is not very

volatile; after all changes in its most common instrument� the tari¤� are legislated at most on a yearly

basis. Therefore, in section 3, we provide multiple examples of sources of uncertainty in the multilateral

trading system in general.

One increasingly important source of trade policy uncertainty that we will examine in detail, is prefer-

ential trade agreements (PTAs), whereby countries eliminate protection relative to a subset of partners. As

of February 2010, there were 271 such agreements in force� a dramatic increase since 1990� and 462 have

so far been noti�ed to the World Trade Organization (WTO). As we note in section 3, there are multiple

reasons for these agreements and active research on what their real value is. One of their implications is

that trade policy is increasingly country speci�c so, in any given year, an exporter is now more likely to face

prices that are either higher (e.g. a Portuguese �rm after receiving EC preferences) or lower (if an Eastern

European country accedes to the EC). While on average countries may currently face more uncertainty due

to rising numbers of PTAs, those that secure preferences can lower the uncertainty they face in exporting

to that market.

In section 3.2 we argue that an important impact of the Portuguese accession to the EC in 1986 was that

it removed uncertainty about pre-existing preferences. Despite the fact that Portugal already enjoyed free

trade in industrial goods with most EC countries and preferences in the Spanish market we �nd that there

was still a substantial increase in exports and that 75% of that growth was generated by new �rm entry, only

2

25% by expansion in exports per �rm. These e¤ects persist even after we account for income and exchange

rate e¤ects, which suggests they may be driven by changes in expectations about future trade policies. Using

a dynamic model, we show that reductions in uncertainty increase entry and trade. We also show that an

agreement that secures existing preferences is valuable for the �rm and we derive a structural equation that

relates the entry decision to applied policy and a measure of future policy uncertainty. We follow the model

in constructing a speci�c empirical measure of uncertainty at the industry level: the percent loss in pro�ts

due to a negative tari¤ shock, which the Portuguese exporters to a particular market (Spain and EC-10)

may have feared before the agreement. Using this measure and �rm level data for Portuguese exporters in

the 1980�s we �nd evidence for the entry predictions of the model. Using the structure of the model we can

also estimate that on average the Portuguese exporters placed a 25% probability on the possibility of losing

preferences to those markets before the agreement and zero after. We also �nd that the accession lead to

investment and entry into EC markets that was considerably higher than what would have occurred if the

reduction in current tari¤s had not been credible, i.e. if the uncertainty about future tari¤ changes had not

been eliminated.

In the �nal section we discuss additional policy implications of these results for Portugal and how they

can be extended to analyze foreign direct investment and exchange rate volatility for example. While we

think that the Portuguese case is particularly interesting and important, we argue that our approach and

results have broader implications. For example, currently there are many developing countries that receive

unilateral preferences. One common concern raised by these countries is that the uncertain nature of these

preferences, which are subject to periodic renewal and eligibility conditions, considerably lowers their value.

Our results suggest that securing those preferences via permanent PTAs, as sought by some countries could

generate considerable investment, entry and trade.3 We show that quantifying the value of PTAs depends

on whether the current policy change is credible, i.e. whether the agreement also removes uncertainty about

future policy reversals. In this sense our results have broader implications for evaluating the investment and

market-entry e¤ects of other types of trade or tax policy reforms.4

The paper has the following structure. In section 2 we review the existing literature. In section 3

we provide some background on uncertainty in the world trading system and then focus on stylized facts

regarding the historical impact of preferential agreements on Portuguese trade. We also present preliminary

3Our results can also help explain recent aggregate evidence about the extremely large impact of some PTAs on trade �owsand the mixed results for others (Baier, et al. 2007).

4For a theoretical model of the role of domestic tax uncertainty on investment see Rodrik (1991).

3

evidence that� despite the already low trade barriers� the 1986 EC accession caused Portuguese export

growth and that this growth was largely driven by investment and entry of new �rms. This preliminary

evidence motivates and informs the theory in section 4, which models the impact of policy uncertainty on

�rm-level investment and export entry decisions. In section 5 we test the entry predictions from the model

using detailed �rm and policy data for the case of Portuguese accession to the EC and quantify the relative

impacts of the main policy e¤ects: reduction in applied tari¤s vs. uncertainty. In section 6 we summarize

the key results, discuss their policy implications and future research. We include detailed derivations of

some theory and estimation results in the appendices.

2 Literature Review

While the main motivation for the paper is to analyze the impact of European integration on Portuguese

�rms� investment and export decisions, our approach contributes more broadly to existing research on

related topics. Thus the following section provides some context to better understand our contributions in

this paper.

To examine the impact of policy uncertainty we focus on a dynamic model of �rm investment and entry

decisions. If entry costs are sunk and partially or wholly irreversible, a prospective �rm must consider

the time path of other variables that a¤ect pro�tability. Dixit (1989) shows that uncertainty about future

prices creates an option value of waiting to enter so �rms will delay investments in entry or exit until more

information about the state of the world is received. In this setting, entry and exit rates depend on the

variance of shocks, their persistence and the size of sunk costs. Baldwin and Krugman (1989) extend these

theoretical insights in a model of entry and exit under uncertainty about exchange rate processes. They

focus on homogenous �rms and show there is a possibility for �beachhead e¤ects�: after a �rm receives a

positive shock and paid the sunk cost of entry into exports it will not immediately reverse its investment in

the event the initial shock is reversed. Thus even temporary shocks can have lasting e¤ects.

There is considerable evidence that �rms are heterogenous, a fact that is particularly important in

the context of international trade. Starting with Bernard and Jensen (1995) an extensive literature has

developed, which documents the fact that exporters tend to be larger, more e¢ cient and produce higher

quality products than non-exporters.5 Moreover, there is evidence of self-selection into exporting: i.e.

5We can also verify this directly in our data for Portugal in the period we are interested: in 1987 the median number ofemployees for all exporting �rms (with at least one employee) was 28, which is 7 times larger than the median number for all

4

that the larger, more productive �rms are the ones that can overcome �xed costs and export. A large

number of recent models incorporate �rm heterogeneity and show it has important theoretical and empirical

implications for trade (c.f. Melitz, 2003, and Bernard et al. 2003). Particularly important from our

perspective is the fact that in this type of model the extensive margin may dominate the response of trade

�ows to reductions in trade barriers (as argued by Chaney, 2008) and that the failure to control for �rm

heterogeneity in gravity models results in an upward bias to aggregate estimates of trade frictions (Helpman

et al., 2008). Therefore we will focus on a dynamic model of entry into exports where �rms have heterogenous

productivity.

While there is increasing evidence for the importance of sunk costs in export-market entry (c.f. Roberts

and Tybout, 1997), the analysis of uncertainty remains largely con�ned to the impact of exchange rate

volatility, about which evidence remains mixed. Baldwin (1988) uses aggregate data and �nds that large

exchange rate shocks in the 1980s may have led to �beachhead e¤ects�. But given the aggregate nature of

the data he is unable to rule out alternative explanations for the �ndings. Das et al. (2007) �nd that sunk

costs are quantitatively important in explaining export participation of marginal �rms in Colombia and

use a structural model to show that subsidies to sunk costs could raise entry substantially but �nd limited

evidence that exchange rate volatility a¤ects entry and exit. Campa (2004) �nds evidence of sunk costs

of entry for Spanish �rms but smaller than anticipated e¤ects of exchange rate volatility. More broadly,

studies of the impact of exchange rate volatility on aggregate trade �ows �nd that e¤ect is negative but

�fairly small and is by no means robust�. (IMF, 2004, p.6)

The impact of trade and tax policy uncertainty when there are sunk costs of investment, has received far

less attention. The di¢ culty is that most policy processes are not readily adapted to a standard stochastic

process and major regime changes may be �rare events�.6 This does not mean however that such �rare

events� are irrelevant for investment decisions, as recently emphasized in a di¤erent context by Barro

(2006). Even if feared reversals to disastrous trade protection or threatened trade wars never materialize,

the small possibility of these worst case scenario outcomes can have measurable economic e¤ects. The scant

work on this area is largely theoretical, for example Rodrik (1991) develops a model of capital investment

when �rms believe an investment tax credit reform may be reversed in the future. If the probability or cost

private �rms in the economy.6This leads Hassett and Metcalf (1999) to model the application and removal of an investment tax credit as poisson jump

process. They �nd such a model is more consistent with observed �rm behavior when the price process for investment is alreadysubject to uncertainty.

5

of a policy reversal is high, a reform to promote investment may produce exactly the opposite outcome.7

Empirically, Aizenman and Marion (1993) show that low persistence of monetary and �scal aggregates has

negative e¤ects on investment and growth in cross-country regressions.

There is an ongoing empirical debate regarding the value of bilateral and multilateral trade agreements.

The impact of European Community (EC) membership on trade �ows, which we examine here for the case

of Portugal, is certainly no exception. Early work on the trade e¤ects of PTAs delivered mixed results, e.g.

Frankel (1997) reports small and sometimes negative e¤ects of EC membership on bilateral trade between

members in the 1960s and 1970s but positive ones in the 1980s and 1990s. Small trade e¤ects have also been

found by several other ex-post econometric studies of other PTAs; they seem puzzling given the ex-ante zeal

of policy makers for entering such agreements. Baier and Bergstrand (2007) however provide a solution to

this puzzle; by using panel data to control for potential selection into PTAs, they �nd that these agreements

increase trade by as much at 100% in some cross-country estimates.8

In addition to self-selection there are other potential explanations for the recent �ndings of the large trade

impacts of PTAs. They may be due to competitive reallocation and productivity enhancing investments

induced by trade liberalization (Constantini and Melitz, 2008; Chaney, 2005; Tre�er, 2004). Alternatively,

PTAs may imply permanent reductions in trade frictions so future shocks to macro variables may have larger

e¤ects on expected pro�ts and this can generate entry as argued by Ruhl (2008). The latter motive is related

to the one we explore but we model the policy uncertainty channel and estimate its impact econometrically.

Much less is known about how and why trade grows following PTAs. However, some recent research

examines the e¤ect of trade policy uncertainty at the product level. Francois and Martin (2004) demonstrate

how tari¤volatility can have negative welfare implications in the presence of risk aversion and therefore WTO

tari¤ bindings may be valuable in truncating the distribution of feasible tari¤s. Most empirical work remains

at the aggregate cross-country level and does not examine the details of PTA policy changes. One exception

is Evenett et al. (2004) who examine whether MFN tari¤s are more secure than the ad-hoc preferential

duties of the Generalized System of Preferences (GSP). They �nd inconclusive evidence for Bulgaria and

Ecuador�s exports to developed countries. The other exception is Handley (2010) who models trade policy

uncertainty in a heterogeneous �rms�setting. He tests the predictions of the model for trade �ows using

7Johnson et al. (1997) show that reform credibility is essential to inducing �rms to switch to costly but more productivetechnology.

8Applied general equilibrium models often grossly under predict the response of trade �ows to the tari¤ reductions in PTAs,a challenge documented by Kehoe (2005) for the North American Free Trade Agreement. PTAs may also be valued if marketagents prefer policy stability as hypothesized by Mans�eld and Reinhart (2008) who also provide aggregate evidence that PTAsreduce trade variability and increase exports.

6

product data for exports to Australia. He measures uncertainty faced by those exporters as the gap between

Australian applied tari¤s and bindings and �nds that this policy uncertainty lowers both the level and the

responsiveness of exports to Australia to unbound applied tari¤ reductions. To our knowledge there are no

tests of the impact of trade policy uncertainty on �rms�investment and entry decisions into export markets.

3 Trade Policy Uncertainty and Portugal�s European Integration

The main purpose of this section is to provide some stylized facts and a preliminary aggregate analysis of

Portugal�s European trade integration. In order to better understand how this integration may have reduced

the uncertainty Portuguese exporters faced in Europe before 1986 we �rst describe some basic features of the

world trading system and highlight several sources of trade policy uncertainty. We then provide background

information on Portugal�s preferential agreements with various European countries; evidence that the EC

accession in 1986 generated considerable export growth towards those partners and that it was mostly driven

by entry of new �rms into those markets (rather than higher sales per �rm). The evidence is consistent

with an uncertainty-reducing role of EC accession but possibly also with other explanations. Therefore to

identify the role of trade policy uncertainty we provide a detailed theory in section 4 that generates speci�c

predictions, which we test in section 5.

3.1 Trade Policy Uncertainty in the World Trading System

As we note in the Introduction there are good reasons to be concerned about trade policy uncertainty and

yet very little research on its sources and impacts. This may partly be due to the fact that trade policy

is perceived not to be very volatile; after all statutory tari¤ rates are legislated at most on a yearly basis.

However, applied trade policy can be more volatile than what is suggested by focusing on statutory tari¤

rates since they are by no means the only type of protection. Limão and Tovar (2009) employ the estimates

in Kee et al. (2009) and note that the trade restrictiveness index for the typical country in the world is

equivalent to a uniform tari¤ of 14% , but this jumps to 27% when non-tari¤ barriers are included. Several

of these NTBs are not strictly (if at all) regulated by the WTO and even the ones that are can be used by

countries, sometimes on a temporary basis and for speci�c goods. But even temporary measures can remain

in place for months or years.9

9For example, in June 2001 the US started an investigation that eventually lead to the steel safeguards of about 30% inMarch of 2002. These duties remained in place for almost 20 months and were only removed after a negative ruling from the

7

The interaction between the ability to use unregulated policy instruments and macroeconomic or political

shocks can generate considerable uncertainty. For example, there was widespread fear that the recent

economic downturn would result in a substantial increase in protectionism. This included the possibility

of anti-dumping measures; increases in developing country tari¤s from their applied level to the maximum

allowed under international agreements; and the use of government procurement measures such as the �buy-

American�provision attached to the US stimulus bill. While the worst fears of a trade war were not realized,

the outcome was considered a real possibility and that uncertainty alone can a¤ect investment and exporting

decisions, as we will show.

Turning to more permanent sources of trade policy uncertainty a number of examples stand out: First,

concerns with product quality and safety raise the possibility that certain products may be completely

banned from a market, e.g. genetically modi�ed foods in the EU; second, the US threat of import duties

to counter Chinese currency �manipulation�; third, the possibility of using �environmental�duties at the

border to o¤set di¤erences in carbon emissions in production. Again we stress that even if these policies

remain only a remote possibility, the fact that if they materialize they would be signi�cant and possibly

permanent can have important impacts in current investment and export decisions. It is conceivable that

these e¤ects could be larger than temporary exchange rate movements that can be hedged against.

One measure of governments�concern with this source of policy uncertainty is their attempts to negotiate

trade agreements. In fact, one of the central reasons for the foundation of the GATT was the desire to avoid

the disastrous tari¤wars in the 1930�s, which shutdown many markets to exporters. To this day the GATT�s

successor, the WTO lists as one of its functions and foundation of the trading system: "Predictability through

bindings and transparency [to] promote investment and allow(s) consumers to fully enjoy the bene�ts of

competition."10 and we will see that these channels will be central in our model.

However, multilateral agreements are themselves uncertain in terms of timing, negotiation outcomes and

implementation. Successive rounds of trade negotiations have repeatedly failed and later been resurrected.

For example, an aborted attempt was made to start the Uruguay Round in 1982 and negotiations only

restarted in 1986. After the UR, attempts to start a new round failed at Seattle in 1999. Moreover, each

successive round has taken longer to complete than the previous� the UR took over 7 years to complete,

twice as long as expected and the Doha Round was launched in 2001 and nine years later it is still unresolved.

WTO. Foreign exporters of steel were not compensated for this loss. More generally, Grinols and Perrelli (2006) report thatthe typical U.S. dispute under the WTO lasts about 18 months with a large standard deviation of about 10 months. Anotherexample of NTBs include anti-dumping duties, which can be punitive.10Accessed April 28 2010 at <http://www.wto.org/english/thewto_e/whatis_e/tif_e/fact2_e.htm>

8

Even when an agreement is successfully concluded the implementation takes some time, disputes arise and

not all policies are covered.

Moreover, multilateral agreements do not regulate all types of trade policy. This can generate uncertainty

in periods of crisis, as discussed above, but also in quieter times. To see why note that currently two �rms

exporting a similar product to the same market may face very di¤erent policy barriers. While the tari¤s

that countries negotiate multilaterally must be available to all WTO members, this so called Most-Favoured-

Nation (MFN) tari¤ is in practice often the policy faced by the �least-favoured-nation�. The reason is the

myriad of preference schemes available. These include not only the standard PTAs but also unilateral

preferences the US, EU and several other developed countries extend to developing nations, e.g. through

the Generalized System of Preferences (GSP). These preferences generate uncertainty for the �least-favoured-

nations�whose �rms don�t know if they will face more competition from �rms that receive preferences and

also become less certain of any future multilateral tari¤ reductions11 .

Unilateral preference schemes, such as GSP, are also extremely uncertain for the recipients themselves.

These preferences are often conditional not only on trade but also non-trade related criteria that can and

have triggered non-renewal for speci�c countries.12 This is one reason why recipients of such unilateral

preferences try to negotiate more permanent arrangements with developed countries even if that requires

the former to open up their markets. Examples include Peru�s and Colombia�s FTAs with the US.13

As just noted, preferences tend to be more secure when they are part of a formal and reciprocal preferen-

tial trade agreement (PTAs). There are currently hundreds of such arrangements re�ecting both trade and

non-trade motives (Limão, 2007). Potential trade bene�ts include guaranteeing access to speci�c markets to

secure (i) pre-existing unilateral preferences (US-Colombia, and as we will argue Portugal�s EC accession),

(ii) insure against some forms of protection in that country (e.g. U.S. PTA partners were exempt from the

steel safeguards) or (iii) in case a trade war breaks out in the rest of the world (Perroni and Whalley, 1994).

But, even the best laid plans to move forward on regional and bilateral arrangements are fraught with un-

11Limao (2006) and Karacaovali and Limao (2008) �nd that preferences provided by the US and EU respectively causedthem to maintain relatively higher multilateral tari¤s against the rest of the world in the UR.12During the period 1993-2008, the United States allowed the GSP to expire seven times for periods lasting from two to

fourteen months (Jones, 2008).13These countries bene�ted from preferences into the US market along with other Andean countries through the ATPA.

Given their temporary nature they sought FTAs to make them permanent. A recent USITC report describes the issues for theremaining ATPA members

"The probable future e¤ ects of ATPA are likely to be minimal, as investor uncertainty over ATPA renewaland concerns about the impact of recently negotiated U.S. bilateral FTAs with Colombia and Peru have dampenedregional interest in investment to produce ATPA-eligible exports, particularly in Bolivia and Ecuador. (p. ix)"(USITC, 2008)

9

certainty. Plans for an FTA of the Americas began in the 1990s and have been abandoned. FTAs between

the US and Korea as well as Colombia were signed but never rati�ed by the US after years of negotiations.

Similar issues have a¤ected accessions to the European Market: the United Kingdom was initially vetoed

for membership in the 1960s, but later joined the club in 1972; Turkey has been in negotiations for over

20 years; and Portugal�s road to full membership was also long and fraught with uncertainty, as we now

describe.

3.2 Portugal�s European Trade Integration

Portugal�s market access to its European partners in the 1970s and early 80s displayed many of the same

characteristics associated with uncertainty outlined above. Prior to joining the European Community (EC),

Portugal was a founding member of the European Free Trade Area (EFTA), which was signed in 1960. By

the late 1960s, EFTA had achieved free trade in industrial products. When the UK and Denmark left EFTA

in 1972 to joint the EC, the remaining EFTA countries (including Portugal) signed bilateral agreements with

the EC that implemented free trade in industrial products by 1977.14

Portugal�s trade with neighboring Spain remained highly restricted until the EFTA-Spain agreement of

1980. This agreement began a partial liberalization of Spain�s tari¤s against the EFTA countries. In the

�rst phase from 1980-1983, a three tiered system of reductions on industrial products would reduce tari¤s

by 25% to 60% with EFTA partners. Portugal was granted even greater reductions of up to 80%.15

A second phase of reductions over a period of indeterminate length was to commence in 1984. The

EFTA-Spain agreement contained no de�nite timetable or scheduled reductions for the second phase. This

so-called �dynamic clause" was possibly incompatible with the criteria of Article XXIV of the GATT al-

lowing preferential trade agreements. It was uncertain at the time if and when further liberalization would

commence. In a working party report to the GATT secretariat, one member noted that the EFTA-Spain

agreement that also governed Spain preferences to Portugal

�provided only an expectation that at some point in time the duties and other regulations

of commerce would be eliminated but no speci�c provisions existed in this respect. There was a

14The schedules appear the GATT submission "Agreement between the European Communities and Portugal",L/3781/Add.1, December 29, 1972.15Details of the reductions can be found in the text of the �Agreement Between the EFTA Countries and Spain," signed May

26, 1979 and entering into force on May 1, 1980. Annex P contains the timetable and list products with tari¤ reduction forSpain and Portugal. GATT noti�cations indicate that these scheduled reductions were implemented as planned (�AgreementBetween the EFTA Countries and Spain, Information Furnished by Parties to the Agreement" L/5465, March 8, 1983).

10

great di¤erence between an expectation and a speci�c plan and schedule�.16

By 1984 both Spain and Portugal were in protracted negotiations for accession to the EC. Noti�cations

to the GATT show that the preferential reductions in place by 1983 were simply extended and then renewed

multiple times by an oversight committee.17 The Articles of Accession required another round of tari¤

reductions between the Portugal, Spain and the EC-10 countries and harmonization with the EC Common

Customs Tari¤ (CCT). The accession entered into force on March 1, 1986. Protocol 3 of the Acts of

Accession required Spain to fully liberalize industrial tari¤s against Portugal immediately in accordance with

the preferences already granted by the existing EC-10 countries. Spain�s agricultural tari¤s were reduced

by 12.5% per year, with respect to Portugal and the EC-10 with free trade by 1993 in most products. Some

non-tari¤ measures and quantitative restrictions would by fully or partially liberalized by 1996. Both Spain

and Portugal would implement the external CCT immediately on products with tari¤s that were 15% higher

or lower than the CCT. For tari¤s outside this range, the CCT would be phased in by 1993. The EC-10

countries phased in full liberalization by 1992 of agricultural tari¤s against Portugal at 14.3% per year.

Before modelling and estimating the impact of uncertainty it is useful to examine the broader impacts

of these preferences on Portugal�s trade and investment in exporting. In the recent past Portugal has been

a fairly open economy; in 2006 its imports and export to GDP ratios were respectively 39 and 31%. But

that was not always so. During the 1950�s and 60�s the overall goods trade/GDP ratio only averaged about

30% going above 40% only in the 1970�s and 50% in the 1980�s.18

The historical impact of European preferential agreements on Portugal�s aggregate trade/GDP ratio is

sometimes clear, e.g. imports/GDP rose rapidly upon EC accession, but not always. What seems clear

is that these agreements had a strong e¤ect on the trade orientation towards preferential partners. The

trade share with EFTA countries increased from about 20% in 1960 to 30% in 1973, as shown in Figure 1.19

The �gure also reveals that the termination of agreements is important. The exit of Denmark and the UK

(which accounted for half of Portugal�s trade with EFTA) to join the EC in 1973 initiated a rapid decline

in Portugal�s trade share with these countries.

16�AGREEMENT BETWEEN THE EFTA COUNTRIES AND SPAIN, Report of the Working Party," L/5405, October 24,1980, p.317Agreement Between the EFTA Countries and Spain, Information Furnished by Parties to the Agreement" L/5886, October

31, 1985.18The 2006 ratio is from Bank of Portugal online statistics. The historical ratios for trade in goods are calculated from

current price data in Pinheiro et al (1997).19Source of the trade data: IMF Direction of Trade Statistics.

11

0.08

0.30

1951 2008year

1960: EFTA signed; 1972: EFTA­EC agree FTA; UK/DK exit EFTA

Portuguese Trade Shares with original EFTA countries

Figure 1

0.40

0.50

0.60

0.70

0.80

1951 2008year

72: EC industrial preferences agreed 85: Accession signed; 93: transition complete

Portuguese Trade Shares with EC­10 & Spain

Figure 2

Figure 2 shows the re-orientation of Portugal�s trade with its EC preferential partners starting in 1985.

The share with the EC-11 goes up from 52% in 1985 to 72% in 1992. If we excluded Spain we would still �nd

that the trade share with EC-10 went from 47 to 57% over that period. We also see that after the transition

period was complete (around 1993) the trade share �attens and eventually starts to fall; this latter fall is

driven by trade with the EC-10 since trade with Spain continued growing and currently stands at just below

30%. The other interesting point to note above is that the initial preferential agreement between the EC

and Portugal (agreed in 1972, fully implemented by 1977) and Spain and Portugal (early 1980�s) left their

trade share nearly unchanged at about 50% between 1972 and 1985. We can detect more of an e¤ect during

this period if we focus on Portuguese export shares alone, which go from 50% to 62% in this 13 year period.

But export growth is faster after the 1986 accession and the EC share in Portugal exports goes up to 73%

in only 7 years.

12

The strong increase in trade shares with the EC partners after 1985 was not merely a switch away from

exporting to other markets. There is strong evidence of trade creation given that total real exports in 1993

were almost twice as high as in 1985 (Pinheiro et al., 1997). Starting in 1981 we have access to data from

the Portuguese census (INE) that, to our knowledge, has never been analyzed in this period: �rm level

international trade by Portuguese �rms. This allows us to examine whether the source of the growth in

trade is related to new �rms entering the preferential markets. For this channel to be important there should

be a noticeable uptick in entry. That is what we �nd for the overall sample of Portuguese exporters. In

1987, for example, the gross entry rate was about 42%, i.e. that was the growth in the number of new

exporting �rms with positive values to any destination relative to 1986. This is considerably more than the

gross entry rate in 1985, which was 35%. The di¤erences in net growth rates of the number of �rms were

not as pronounced because the higher gross entry in 1987 was somewhat o¤set by higher gross exit rates

(28% in 1987 vs. 22% in 1985).

To determine if net entry is di¤erentially larger for preferential markets we contrast it to the growth in

the number of �rms exporting to large non-preferential markets such as the U.S. As the dotted line in Figure

3 shows there was positive and rather substantial net entry into the US between 1981 and 1985 but almost

none between 1985 and 1992. In contrast, the number of new �rms exporting to Germany (dashed line)

grew by 65 log points between 1985 and 1992.20 Entry into the Spanish market was even more pronounced,

over 150 log points in the 1985-1992 period with an apparent upward break in the trend around 1985.21

20Other important Portuguese preferential markets such as the UK displayed a similar trend to GErmany, as did France butthe latter exhibing faster growth post-1985.21Our analysis stops in 1992 for two reasons. First, as discussed above this was the end of the initial period accession.

Second, there was a major change in the data collection procedures in 1993 due to the removal of physical customs barrierswithin the EC. The new system, Intrastat, is based on self-reporting and has minimum export value thresholds, both of whichimply that the number of �rms in the data in 1993 exhibits a discrete fall that a¤ects only EC partners.

13

­.50

.51

1.5

1980 1985 1990 1995

Spain Germany US

Portugal's  Export Firm Entry Growth 1981­1992

Figure 3

We get an even clearer sense of the relative importance of this channel in real export growth if we

decompose the latter into growth of number of �rms and exports/�rm. As �gure 4 shows for the case of

the US real export growth was driven by both the intensive and extensive margin prior to 1985. After that

period the extensive margin contributed little, as seen above, and the reduction in real exports was largely

driven by reductions in real sales per �rm.

The picture for Germany is considerably di¤erent with �rm entry accounting for almost all the growth

in exports between 1985 and 1988. For Spain the �gures are even more striking with the extensive margin

driving most of the large growth, almost 200 log points, in real exports between 1985 and 1992.

­1­.5

0

1980 1985 1990 1995ano

# Firms Real Exports/Firm Real Exports

Portugal's Real Exports Growth Margins to US

Figure 4

14

­1­.5

0.5

11980 1985 1990 1995

ano

# Firms Real Exports/Firm Real Exports

Portugal's Real Exports Growth Margins to Germany

Figure 5

­10

12

1980 1985 1990 1995ano

# Firms Real Exports/Firm Real Exports

Portugal's Real Exports Growth Margins to Spain

Figure 6

The EC accession was not the only notable economic event Portugal experienced in the 1980�s. Earlier, in

August 1983, Portugal completed an agreement with the IMF to help it resolve a balance of payment crisis.

The nominal Portuguese exchange rate continued to depreciate against the major European currencies until

1990 but starting in 1995 it experienced some appreciation relative to the US dollar.22 To account for this

and other e¤ects, e.g. in incomes and prices, we can estimate an aggregate gravity equation for Portuguese

exports. This is by now a standard tool (Baltagi et al., 2003; Anderson and van Wincoop, 2003). We

include country e¤ects to account for time invariant di¤erences in exports between Portugal and each of

its partners (distance, colonial ties, etc.) and year e¤ects to control for Portuguese productivity, nominal

export price or exchange rate changes. We also allow the time e¤ects to di¤er between advanced economies

and others to control for any di¤erences in the composition of exports. Moreover, we control for bilateral

22The aggregate real exchange rate did not exhibit large changes between 1980-1981 according to the IMF IFS statistics.

15

nominal exchange rates, price de�ators in the import country and their real GDP. By interacting an EC

accession time dummy (=1 for 1986-1992) with the member country dummies (Spain and EC-10) we can

then test if Portuguese exports to these preferential markets grew di¤erentially relative to other advanced

economies.23

We �nd an increase of about 24 log points towards the EC-10 in the post-accession period that cannot be

accounted by the standard determinants. That increase is about 5 times larger for Spain. Given our interest

in uncertainty and the role of investment and entry we also go beyond the standard gravity estimation and

use the (ln) number of �rms as a dependent variable (our model will provide a formal justi�cation for doing

so). Those results in the second column of table 1, con�rm what we saw in the �gures above: that the

growth in trade was largely driven by new entry. By also running a similar regression for exports/�rm we

can clearly see this growth decomposition from the accession slightly more than three quarters was accounted

for by �rm entry both for the EC-10 and Spain. If we include a similar variable for the US as a falsi�cation

test we see no signi�cant increase in exports and if anything the contribution of entry was negative (and

statistically insigni�cant) once we control for other factors.

Given the importance of exchange rate volatility in Portugal and the prominence of this channel in

discussions of trade and uncertainty we also extended the speci�cations to include it. This did not a¤ect

any of the results we discussed. Moreover, the elasticity is fairly small, both for aggregate exports (-0.096)

and for �rm entry (-0.07). This is consistent with previous studies that �nd con�icting and typically small

e¤ects of exchange rate volatility on aggregate trade �ows.24 To get a sense of the magnitudes, a two

standard deviation reduction in this variable (1.74) leads only to a 17 log point increase in exports and 12

in number of �rms, which is also a small fraction of the standard deviation in either one of these variables.

Moreover, the volatility within individual nearly all EC-10 countries and Spain over this period was even

lower than this. The volatility relative to the Peseta fell only by 0.7 between the height of the crisis in 1982

and 1992 and only by 0.25 relative to the German Mark. So this was clearly not important in generating

the export and �rm entry boom we observe.25

In sum, there is strong evidence of an increase in aggregate trade due to accession and that it is driven

by �rm entry even after we control for standard aggregate determinants. Given that Portuguese exporters

23 In addition to the EC-10 and Spain these include EFTA countries, US, Canada and Japan.24For a recent review of the academic literature see the IMF report at http://www.imf.org/external/np/res/exrate/2004/eng/051904.pdf.

The measure we use is the one the report cites as the preferred one: log(standard deviation of monthly exchange rate changes).25 It is still possible that the entry into the euro had stronger e¤ects both because it eliminated the volatility completely and

possibly more permanently than any earlier changes. Our approach can be extended to analyze this interesting question.

16

already enjoyed some trade preferences in Spain and zero or close to zero tari¤s in the EC-10 this evidence

seems puzzling. The model in the next section provides a potential explanation: the agreement removed

policy uncertainty faced by exporters, which we subsequently test.

4 Theory

We now model the impact of policy uncertainty on �rm-level investment and export entry decisions.

4.1 Demand

The utility function of the representative consumer, U , is identical across countries and de�ned over a

numeraire good, denoted by 0, which is homogenous and freely traded on world markets, and a subutility

index de�ned over di¤erentiated goods Q with constant expenditure share �

U = Q�q1��0 (1)

We consider a CES aggregator over a continuum of di¤erentiated goods, indexed by v and with mass . For

simplicity of exposition we focus on a simple symmetric structure with common elasticity across varieties,

� = 1= (1� �) > 1.26

Q =

�Zv2

(qv)�dv

�1=�(2)

Each country has a mass of identical consumers and aggregate income equal to Y . Consumers in country i

face prices piv so their optimal demand for each di¤erentiated product v and the associate price index Pi

are given by the standard expressions

qiv =�YiPi

�pivPi

���(3)

Pi =

�Zv2

(piv)1��

dv

�1=(1��)(4)

The consumer price, piv, includes any trade costs. We focus on ad valorem import tari¤s and note that

they are generally not �rm speci�c but rather product or industry speci�c, and denote the tari¤ factor that

26We can show that most theoretical and empirical results can be easily extended to a more general multi-sector structurethat allows for di¤erent elasticities of substitution within each sector and across sectors, e.g. if Q is a cobb douglas aggregatoracross H sectors , each representing a distinct CES aggregate.

17

i sets on the set of the group of products V by �iV � 1 , so free trade is represented by �iV = 1. Therefore

producers of any v 2 V receive piv=�iV where �iV will be unity if the good is produced and sold in i (i.e. we

assume no domestic sales taxes). We now determine these prices.

4.2 Supply and Pricing

We �rst determine the optimal price and operating pro�ts for each monopolistically competitive �rm condi-

tional on supplying a market; in the following section we examine its initial decision of whether to invest to

enter that market. We assume constant marginal costs of production, cv, but allow them to be heterogenous

across �rms. The parameter 1=cv can be interpreted either as labor productivity or the productivity of

an input bundle, so the �rms variable cost of producing a unit in the exporting country e is simply wecv.

Since our analysis will be from the perspective of a given �rm in a particular exporting country we drop the

exporter country subscript.

In a deterministic setting it is well known that the monopolist would then choose prices (or quantities)

to maximize operating pro�ts in each period, leading to the standard mark-up rule

�iv = (piv=�iV � wcv) qiv (5)

piv =wcv��iv (6)

The consumer price in country i , piv, re�ects the price received by the producer in e � the markup over

cost wcv� � augmented by the ad valorem tari¤ if the good is imported.

Under uncertainty we need to be more clear about the timing of the �rm�s production and pricing

decisions. If we were focusing on uncertainty surrounding a variable with high frequency variation it may

be reasonable to consider production, and possibly pricing, decisions undertaken prior to the realization of

the state of foreign demand. But since we are focusing on trade policy, which changes at low frequency, we

don�t think that is the most relevant friction to focus on in analyzing the impact of uncertainty. Therefore

we allow the �rm to make all its production and pricing decisions after the policy and thus demand are

known, only its investment decision will be made under uncertainty. This production �exibility has two basic

implications. First, the pricing decision is exactly the same as above. Second, we are making the �rms less

averse to policy risk, e.g. to variability in tari¤s, after they enter the market since they can optimally adjust

to shocks and their operating pro�ts are convex in the policy. To clearly see the last point we substitute the

18

optimal price into demand to calculate revenue received by the producer

pivqiv=�iV = (�iV )��c1��v �Yi

�w

Pi�

�1��(7)

We can see that, al. else equal, the export values for a �rm that has entered a market are directly a¤ected

only by the current applied policy, there is no direct e¤ect of uncertainty. This occurs because we allow

production to occur after the uncertainty is resolved. Therefore the direct impact of uncertainty on individual

�rms in our model will arise via the investment/entry margin rather than the intensive margin.27

Substituting revenues into the pro�t expression and simplifying we obtain

�iv = (�iV )��c1��v Ai (8)

where Ai = (1� �)�Yi�w

Pi�

�1��:

where Ai summarizes domestic cost, w, and foreign demand conditions.28

4.3 Firm Value, Investment and Entry Setup

We focus on how foreign trade policy uncertainty a¤ects the decision to enter export markets. Therefore, we

assume there are no �xed costs to enter (or produce) in the domestic market (as in Helpman et al. (2008)).

As such, for each group of products V there exists a mass of �rms in the exporting country equal to NV ;

all of which produce for their home market but only a subset of them will export to any given market. As

we noted above, these �rms are heterogeneous only in terms of their productivity, which has a cumulative

distribution function GV (1=c) that is strictly increasing.

To serve a foreign market a �rm must �rst make a �xed cost investment that is sunk. As noted in section

2 there is considerable evidence that these investments can be large when it comes to serving foreign markets.27This is not to say that uncertainty will be irrelevant for the intensive margin. As we wil see it will imply that only the more

e¢ cient �rms in an industry export and as we can see from the sales equation these will have higher sales. So the model willgenerate higher average sales per �rm in industries with more uncertainty. There may also be an indirect e¤ect of uncertaintyin our framework. Namely, if there is a su¢ ciently large reduction in uncertainty and this leads to a large increase in thenumber of �rms in this market then the price index will fall and thus lower sales for existing exporters, who must now competewith additional �rms. While in a two country setting lower trade policy uncertainty is likely to reduce the importer�s priceindex via this entry e¤ect, the net e¤ect is less clear in a multi-country world where some countries face uncertainty reductions,e.g. preferential partners, whereas the others potentially face increases. More generally, uncertainty would a¤ect the intensivemargin even for �rms that have decided to export if investments in capacity are required after entry.28We are ignoring exchange rates but these can be incorporated and would simply entail rede�ning A to include a multiplica-

tive e¤ect ex�i . Since this variable does not vary across product it will not have a �rst order in our empirical results and thuswe do not include it here either. Future work may consider interactions in uncertainty processes between tari¤s and exchangerates and try to estimate those second order e¤ects.

19

To understand the basic e¤ect of these costs consider �rst a deterministic environment where pro�ts are

constant. A �rm considering entering a new export market invests and enters if the net present discounted

value of its pro�ts exceeds the investment cost of entry K, i.e.

�iv1� � � KiV (9)

We allow this investment to be destination market speci�c and possibly industry speci�c in that �rms

producing v 2 V all face the same cost, but this cost may di¤er for another set of varieties. The discount

factor � combines a �true� discount rate R and an exogenous �exit� probability, �, such that � = (1 �

�)=(1 + R). This de�nes a zero pro�t cuto¤ for unit costs as a function of the tari¤, cD (�iV ) for �rms

considering exporting a product v 2 V to country i

cD (�iV ) =

�(�iV )

��Ai(1� �)KiV

�1=(��1)(10)

Clearly tari¤ reductions induce entry since they increase demand and thus allow the �xed cost investment

to be covered even by �rms that are less productive. The elasticity of the cuto¤ to a once-and-for-all change

in � is �d ln cD=d ln � = ���1 . It is also clear that the cuto¤ is common to all �rms that face a similar tari¤

and �xed cost, so for v 2 V all �rms with cv < cDiV (�iV ) enter.29

As we discuss in section 3 there are several potential sources of trade policy uncertainty that exporters

face. Moreover, potential exporters can optimally choose not just whether to invest but when to do so.

Therefore ongoing policy uncertainty generates an option value of waiting, which can have important e¤ects

for investment, which we now examine. The analysis below applies for each �rm in an export country e that

is considering the decision to invest to enter in market i and sell some good v so we drop these subscripts

for simplicity.

Formally, the �rm�s decision to enter an export market is modeled as an optimal stopping problem.30

Firms can be divided into exporters and non-exporters. The value of being an exporter is denoted by �e

and such a �rm exits only when hit by a �death�shock since it has no other �xed costs after it enters.31

29The cuto¤ elasticity with respect to tari¤s exceeds unity because the tari¤ is not paid by the exporter, so pro�t increasesmore rapidly in the tari¤ than in the cost, as seen in (8).30Formally, our approach is similar to the one Baldwin and Krugman (1989). There are some key di¤erences. First, they

focus on exchange rates whereas we analyze trade policy, which as we describe below has a di¤erent stochastic process and ismore permanent than exchange rates. Second, they focus on homogenous �rms whereas we incorporate �rm heterogeneity,which allows us to analyze the a¤ect of policy uncertainty both between and within industries that already have some exportparticipation.31While the assumption of no per period �xed costs of exporting may seem extreme, Das et al. (2007) �nd these per period

20

Non-exporters will enter a foreign market only when the expected value of exporting, �e, net of the sunk

entry costs, K, exceeds the option value of waiting, �w. The value of this option in our model arises from the

fact that in the following period conditions may improve and so the �rm may be better o¤ in waiting until

that occurs and then entering. Given this, the investment and entry decision rule for each �rm, identi�ed

by its unit cost requirement c, can be de�ned as a function of the threshold tari¤ �� that makes that �rm

indi¤erent between entry and waiting.

�e(�� ; c)�K = �w(�� ; c) (11)

Therefore, any tari¤ �t � �� (c) triggers entry into the export market by any �rm with cost c. To determine

this cuto¤ and the impact of changes in policy uncertainty we now describe the policy process and de�ne

these value functions.

Trade policy is a¤ected by several factors: economic, political, unilateral, bilateral and multilateral. From

the perspective of most individual �rms, the trade policy uncertainty in foreign markets may plausibly be

taken as given.32 Therefore we do not explicitly model the source of these shocks but simply posit that they

can arise due to changes in political pressure by interest groups, by the initiation, conclusion or breakdown

of trade agreements, by macroeconomic shocks, etc. So trade policy is summarized by a random variable

with two components: the uncertainty over the timing of policy changes and the magnitude of those changes

when they arrive.

More speci�cally, we model policy shocks as a Poisson process with arrival rate .33 We will generally

think of these as aggregate shocks (e.g. a new agreement, arrival of a new government with di¤erent policy

preferences, etc.). When a shock arrives, a policy maker reconsiders the current policy and sets a new one

denoted � 0. Even though the outcome of policy changes is unknown ex-ante, �rms can form expectations over

future policies. We assume they form rational expectations based on the value of and a probability measure

of di¤erent tari¤ outcomes de�ned in the policy space of � 0. These potential outcomes are summarized by

the distribution function H(� 0) with support � 0 2 [1; �H ] , where �H is the worst case scenario. These

distributions are common knowledge to all �rms in a given industry V but they can di¤er across industries

in order to captures the possibility that after a shock, e.g. a trade agreement, some products will experience

�xed costs are negligible, on average, across all sectors analyzed in their structural model of Colombian exporters.32An interesting topic for future research is to examine endogenous choice of uncertainty levels by countries and the role of

industry lobbies.33Similar Poisson arrival processes for policy shocks are used by Rodrik (1991), Aizenman and Marion (1993) and Hassett

and Metcalf (1999).

21

larger policy changes than others.

This policy process implies a long-run mean policy, denoted by E(� 0), which is determined by the

distribution H (� 0) and independent of . Therefore a government that announces a current policy equal

to �t will not have any impact on the long-run expected policy unless they can also convince the producers

that this policy is now permanent, i.e. that = 0, or that the distribution of all future policies has

somehow changed. What we will show is that the resulting lack of credibility, or possibility of reversal of

the current policy to the long-run mean, captured by , reduces the value of current tari¤ reductions and

thus their e¤ectiveness in generating investment in foreign markets. Therefore our goal is to contrast the

e¤ect of changes in current policies at di¤erent with the e¤ect of changing itself where we will argue

that certain agreements work precisely because they successfully reduce the current tari¤ and make that

reduction credible, i.e. they also lower or eliminate .

Since largely captures the probability that current policy will change we will simply refer to it as policy

uncertainty.34 We can also see that it relates directly to another measure of uncertainty: the variability of

next period�s policy. To see this we can express next period�s mean and variance as follows

E(�t+1) = (1� )�t + E(� 0)

V ar(�t+1) = (1� ) (�t � E(� 0))2 + V ar (� 0)

We can see that (i) if = 0 there is no policy variance, (ii) if the current policy is equal to the long-run

mean then the elasticity of the variance with respect to is exactly unity and has no e¤ect on tomorrow�s

mean tari¤, i.e. in this case shocks translate exactly into variance shocks. Finally, we note that it is

conceivable that a shock arrives, but policy remains largely unchanged (or completely unchanged if H is a

discrete CDF). This is consistent with our de�nition of policy shocks as events that raise the possibility of

a new tari¤ regime, but do not necessarily result in tari¤ changes (e.g. negotiations begin on an agreement,

but subsequently fail so no policy changes).

34Technically, 1� is the long-run auto-correlation of tari¤ policy.

22

4.4 Value of Credible vs. �Incredible�Policies

The expected value of starting to export at time t conditional on having observed �t is

�e(�t) = �(�t) + �[(1� )�e(�t)| {z }No Shock

+ Et�e(�0)| {z }

Shock

]: (12)

With probability 1 � , there is no shock to the policy and the �rm continues to the next period with the

same continuation value �e(�t). With probability , a policy shock arrives and the tari¤ changes. The

ex-ante expected value of exporting following a policy shock to a (potentially) new tari¤ � 0 is

Et�e(�0) = Et�(�

0) + �[(1� )Et�e(� 0) + Et�e(� 0)] (13)

In period t, the unconditional expected value of exporting next period after a policy shock arrival is Et�e(� 0)

in (13). We assume that the distribution of future tari¤s, H(� 0), is time invariant, which implies that whether

a policy shock arrives in the next period or ten periods from then, the ex-ante expected value of the tari¤

draw and pro�ts remain the same. This considerably simpli�es the problem and allows us to solve for

Et�e(�0) = Et�(�

0)= (1� �). Note however that the conditional means of the tari¤ and value of exporting,

�e(�t) , still vary over time since they depend on the current tari¤.

The second part of the �rm�s problem is the value of waiting

�w(�t) = 0 + �[(1� )�w(�t)| {z }No Shock

+ (1�H(��))�w(�t)| {z }Shock Above Trigger

+ H(��)(Et�e (�0j� 0 � ��)�K)| {z }

Shock Below Trigger

] (14)

In the current period a �rm that does not enter into an export market receives zero pro�ts from it. In the

following period the �rms continuation payo¤ is still �w(�t) if either no policy shock arrives (the �rst term)

or it arrives but the tari¤ is still above the trigger (the second term). If a policy shock arrives next period,

it will be below �� with some probability H(��) and the �rm will then �nd it optimal to pay K and transition

to the exporting state. The expected value of exporting after a policy shock that generates a � � �� is

Et�e (�0j� 0 � ��) = Et� (� 0j� 0 � ��) + � [(1� )Et�e (� 0j� 0 � ��) + Et�e(� 0)] (15)

This equation is structurally the same as (12), but it is evaluated ex-ante at the expected value of exporting

to a �rm that delays entry until a more favorable policy shock arrives. If a �rm waits to enter, it must

23

be the case that �t > �� . Expected pro�ts at the time of entry are greater than pro�ts today such that

Et� (�0j� 0 � ��) > �(�t). Inevitably, a policy shock will eventually occur and the value of exporting after a

delay will transition to the unconditional expected value of exporting given by (13).

The set of four equations (12),(13),(14), (15) is linear in four unknown quantities �e(�t), Et�e(� 0),

�w(�t), Et�e (� 0j� 0 � ��), which we solve to obtain the value functions in each alternative state of exporting

for a �rm that has a threshold tari¤ ��(c)35

�e(�t; c) =�(�t)

1� �(1� )+�

1� �Et�(�

0)

1� �(1� ) (16)

�w(�t; c) =� H(��(c))

1� � (1� H (��))

�Et�(�

0j� 0 � ��(c))1� �(1� ) +

�

1� �Et�(�

0)

1� �(1� ) �K�

(17)

The interpretation of �e(�t) is straightforward: after investment, the value of exporting conditional on �t

equals the discounted value of expected pro�ts. If were 0 this would correspond simply to the deterministic

value �(�t)= (1� �). But with a probability > 0 the policy will change and the per period expected pro�ts

after that are Et�(� 0). The value for a �rm that is not exporting at that tari¤, �w(�t), is zero if there would

be no tari¤ change, but with some probability H(��) the tari¤ will be below the trigger and so the �rm

will incur the �xed cost K and export. The gross value of exporting is captured by the remaining terms in

curly brackets, which are similar to those in �e except we must use Et�(� 0j� 0 ���) instead of �(�t).

We can now ask what is the value for an exporter of alternative policy changes. Consider �rst a situation

where governments announce that the current policy is free trade, �t = 1. We will call this a credible policy

change or agreement if the exporters expect it to remain in place, i.e. if = 0. We will call it an �incredible�

agreement otherwise, i.e. if it is expected to be revised with probability > 0. The �rst basic point is that

the credible agreement is more valuable for the exporter since the tari¤ reduction is permanent, that is

� @

@�t�e(�t; c; = 0) = �

@

@�t

� (�t)

1� � > �@

@�t

� (�t)

1� �(1� ) = �@

@�t�e(�t; c; > 0) (18)

The second, and closely related point, is that even if the initial agreement is �incredible�so pre > 0, and

it has been in place for some time there may still be considerable value to making it credible, i.e. post = 0.

In these cases the primary impact of a formal agreement may simply be to eliminate uncertainty, which

35See the appendix for details on this derivation.

24

results in the following change in the value of exporting

�e(�t = 1; c; post = 0)��e(�t = 1; c; pre > 0) =� (1)� Et�(� 0)

1� �� pre

1� �(1� pre)> 0 (19)

This value may well capture why the recipients of multiple preferential tari¤ programs around the world

expend considerable resources in making those preferences permanent through formal PTAs. Examples in-

clude GSP preferences provided by most developed countries as well as European and US special preferences

to developing countries. This policy uncertainty also describes the EC-10 and Spain�s preferences to Por-

tuguese exports prior to 1986. Therefore, the change in value above may well capture one of the important

channels by which entry into the EC bene�ted Portuguese exporters. To determine if uncertainty reduction

was an important factor we now examine the predictions of the model for investment and entry into foreign

markets, which we will then empirically estimate.

4.5 Policy Impacts on Investment and Entry

Using (17), (16) and the expression in (11) we can determine the threshold tari¤ that would leave any

given �rm with costs c indi¤erent between starting to export or waiting. From an empirical perspective it

will be more useful to recast this in a di¤erent way and ask what �rms will invest and enter at any given

current tari¤. We have assumed that �rms can be ranked by their productivity (the inverse of unit costs

1=c) according to a CDF that is monotone increasing. Therefore for any current tari¤ �t we can determine

a cuto¤ cost cUt that satis�es ���cUt�= �t.

A �rm with costs equal to cUt is indi¤erent between investing today and starting to export or waiting.

As will be clear that will also be true this period for all �rms with lower costs if they had not yet started

to export. Despite the apparent complexity, there is a closed form expression for cUt in terms of the current

tari¤. To provide some intuition we solve it in two steps. First, we set the di¤erence between �e and �w

equal to entry costs and simplifying terms we obtain

K =�(�t; c

Ut )

1� �(1� ) +�

1� �E�(�; cUt )

1� �(1� ) +�

1� �H(�t)[�(�t; c

Ut )� E�(� j� � �t; cUt )]1� �(1� )

Entry requires that the �xed cost not exceed the sum of the three terms on the RHS. The �rst term is

discounted pro�ts at the current tari¤. If the model were deterministic, the �rm would discount by 1 � �

25

and the next two terms would disappear. The second term is the present value, following a shock, of the

contribution to pro�ts at the ex-ante expected tari¤. The third term is non-positive: it is the present value

of the expected loss of entering today, given that the next policy change is at or below the tari¤ entry trigger.

In the second step, we combine this expression with the operating pro�t function in (8) to solve directly

for cUt

cUt =

�A

K

����t

1� �(1� ) +� E(���)

(1� �)[1� �(1� )] +� H(�t)[�

��t � E(���j� � �t)]

(1� �)[1� �(1� )]

�� 1��1

(20)

The cuto¤ from the deterministic model can be recovered here by setting = 0 in (20). Otherwise, the

cuto¤ condition depends on the shape of the tari¤ distribution, the frequency of policy shock arrivals and

expectations about future tari¤s.

To compare this cuto¤ to the deterministic and examine the impact of alternative policy experiments

it is useful to factor out the applied tari¤ and rewrite cUt as a product of the deterministic cuto¤ and an

uncertainty term.

cUt =

=Utz }| {�1� � + � �(�t)1� � + �

� 1��1

=cDtz }| {�A���t

K(1� �)

� 1��1

(21)

To show that uncertainty in this model generates a lower cuto¤, i.e. requires �rms to be more e¢ cient to

enter, than a deterministic tari¤ at the level �t we must show that the uncertainty component Ut � 1. This

in turn requires that �(�t) � 1 as is clear from the equation above. We show this in the appendix and

further simplify the expression to obtain the equation below. We can interpret �(�t)�1 as the proportional

reduction in operating pro�ts that is expected to occur if we start at the trigger tari¤ �t and a policy shock

occurs and (with probability (1�H(�t))) it generates a tari¤ above that trigger level.

�(�t) ��E(���) +H(�t)[�

��t � E(���j� � �t)]

�=���t (22)

�(�t)� 1 = � (1�H(�t))���t � E(���j� � �t)

���t� 0

Moreover, this inequality is always strict except when the trigger is exactly at the maximum of the tari¤

distribution in which case the two cuto¤s coincide. Note also that even though the policy shock can trigger a

lower or higher tari¤, it is only the latter that a¤ects the decision.36 In sum, the model predicts that policy

36This is an example of the �bad news� principle �rst identi�ed by Bernanke (1983) and is due to the fact that good news

26

uncertainty increases the hurdle for �rms to invest and enter into new markets relative to the deterministic

case. This occurs despite the convexity of pro�ts in tari¤s. This result along with the fact that at = 0

we obtain the deterministic cuto¤ implies that increases in uncertainty lower the cuto¤ under the option

approach at any initial tari¤s below the maximum. As an intermediate step to deriving the estimation

equation it is useful to record here the the semi-elasticity of the cuto¤ with respect to

d ln cUtd

j�t =�

1� �(1� )1� �

1� � (1� �(�t))� (�t)� 1� � 1 � 0

which is negative given �(�t) � 1 .

Consider now the impact of applied tari¤s on the cuto¤. In the absence of uncertainty that elasticity

is simply � ���1 , as shown for the deterministic case. It is simple to see that is also the limit value for

d ln cUtd ln �t

j � > 0. Since most work, theoretical and empirical ignores the uncertainty component we will take

that as our null hypothesis, = 0, and test if this uncertainty parameter has any �rst order e¤ects . To

do so we employ a �rst order, log linear Taylor approximation to cUt ( t; �t) around = 0 and the original

applied policy values (�0). We provide the derivation in the appendix. The general form for any period t is

ln(cUt )j�t=�0; =0 = t�

�

1� �

��(�0)� 1� � 1 � �

� � 1 ln �t +1

� � 1 lnAt

Kt(1� �)+ rt (23)

where rt captures second and higher order terms of the approximation. This shows that increasing uncer-

tainty has a �rst order e¤ect and reduces the cuto¤ even if we are initially at = 0 (i.e. in the deterministic

case). This is true for any trigger value of the tari¤ and strictly so if that trigger it is below the maximum

tari¤. It also holds for cases when the current applied tari¤s are zero, which stresses the point that even

�rms that currently, and possibly for some time, have faced zero tari¤s may not enter if there is some chance

that policy will be reversed in the future. We also see that increasing applied tari¤s around = 0 changes

the cuto¤ by � ���1 , the deterministic elasticity. However, we note that in general, the impact of applied

tari¤s on the cuto¤ is dampened by the presence of uncertainty. We provide the exact expression in the

appendix, but the intuition should be clear from equation (18) in the last section: a reduction in current

tari¤s will not lead to as much entry if it may be reversed in the future. This implies that in the presence

of considerable uncertainty, e.g. prior to an agreement, the coe¢ cient on the applied tari¤ from estimating

the equation above will be biased towards zero.

has a symmetric e¤ect on payo¤s whether the �rm is already in or not (since it can enter) whereas bad news will only a¤ectthose that are already in.

27

In sum, we have shown that one potential value of an agreement with a country that already applies

low policy barriers is to remove uncertainty about those policies. We have also shown that under the real

option approach to investment such an agreement will generate entry and identi�ed one potential way to

measure the degree of that uncertainty, �(�0)�1. The �nal point that we note is that even though the real

option approach is somewhat more complicated than the the standard net present value (NPV) calculation

it has two distinct advantages. First, it clearly captures the behavior of �rm investment more closely since

they clearly have the option to wait and time their investments. Second, some qualitative and quantitative

results are di¤erent under the two approaches. For example, in the appendix we show that (i) the cuto¤

under the option approach is always lower than under the NPV approach which implies more entry under

the latter; (ii) the cuto¤ under the NPV approach can be higher or lower than the deterministic and thus

reductions in uncertainty can lead to less incentive for entry under the NPV approach, which is the opposite

e¤ect of uncertainty from what we �nd using the real option approach.

5 Evidence

5.1 Empirical Approach

While we do not directly observe whether �rms have costs above or below the cuto¤we do observe the number

of �rms and their export status at the country-product level. We could then examine the probability of

individual �rms exporting. However, our model focuses on variation in policies over time and across products

and the cuto¤s we derived are common across some sets of �rms. In particular producers of a variety v

exporting to i will all face a tari¤ that does not discriminate by �rms, but rather by product classi�cation,

denoted V , and therefore the same critical cuto¤ cUTiV . Therefore we examine the fraction of exporters in

an �industry" V to each country pair.

In the model, all potential exporters of a given good in industry V that have productivity above the

threshold we determined (or equivalently with a cost below cUtiV ) will invest and export good v to i. If

that productivity follows a Pareto distribution G (:) with shape k and minimum productivity 1=cV then the

model predicts that at least a fraction

G�cUtiV

�=

�cUtiVcV

�k(24)

of domestic producers of any good in industry V will actually export to that market.

28

The empirical counterpart to the fraction of �rms exporting at time t to i in an industry V is simply the

observed number of exporters relative to the potential number, ntiVntV. The relationship with the theoretical

measure is

lnntiVntV

= lnG(cUtiV ) + utiV (25)

where utiV is a random disturbance term due to measurement error. The term can also capture the potential

for �legacy��rms. If demand or cost conditions in earlier periods had been considerably more favorable this

would generate a legacy of �rms that may survive until period t even though they have costs above cUtiV .

A su¢ cient condition to rule out legacy �rms is that cUtiV � maxfcUTiV 8T < tg, i.e. if current conditions

are better than in the past. In this case, G(cUtiV ) captures the fraction of �rms in the market. In the case

of Portugal in the mid-80�s exporting conditions were improving, as is clear from the observed high entry

rates into EC countries. Therefore, we do not think legacy �rms are particularly important for our analysis.

Nonetheless, in the appendix we argue that our approach and results are robust to certain instances where

legacy �rms are present.

We now model the uncertainty variable and describe how we measure it.

In general, to construct �(�t) precisely we require a speci�c probability distribution H. Therefore, we

consider a discrete state space for tari¤s that is tractable and covers the main cases that are present in our

data. After a policy shock exporters consider three potential tari¤ values, low , medium or high.

�t = �s, Pr(�s) = ps for each s 2 fl;m; hg

We take �l=1 so it captures the many industrial goods that Portugal exported to the EC free of ad valorem

tari¤s both after the accession and before it. We will model the high tari¤ as either the bound tari¤ max-

imum or, if unbound, the rate applied to GATT/WTO members that did not receive any preferences. This

may somewhat underestimate the degree of uncertainty in these goods but seems a reasonable approximation

of what the Portuguese exporters may have placed the highest probability on as a worst case scenario. The

medium tari¤ represents an intermediate level that captures the transitional preferences that were mostly

a feature of Spanish policy towards Portugal. It is important to stress that the latter were transitional and

could not remain for long since they are not GATT legal, as we discuss in section 3.2. Therefore although

we did observe �medium" tari¤s in that period of the mid 80�s, the Portuguese exporters likely placed a

probability close to zero (i.e. pm � 0) on these tari¤s remaining in any future policy shock. Either an

29

agreement would be signed and tari¤s would transition to the low state or negotiations would fail and no

preferences would remain.

Some of the terms in �(�0) depend on the initial state, e.g. if we start at �l=1 then there is no option

value of waiting since no improvement is forthcoming on the applied tari¤. Despite this, in the appendix we

show that whether the tari¤ was initially at the high or medium states we can use the derivation in (22) to

derive write

�(�0iV )� 1 = �ph����0iV � �

��hiV

���0iV

�(26)

This is the percentage pro�t reduction of a shock that moves tari¤s from �0 to the worst case scenario,

which happens with probability ph. The same term applies to cases when the initial tari¤ is low and pm

is negligible:37 Alternatively, if we consider only a two state world, s = h; l the expression above applies to

tari¤s with either history.

We model the uncertainty parameter ti by assuming that prior to an agreement there is a common

probability of policy reversal, pre, and that after an agreement with a country i (or set of countries) such

as the entry into the EU then EUit = 1 and the probability is now post. So we use

ti = pre (1� EUit) + postEUit (27)

We will test if post < pre and subsequently also if post = 0.

Our basic estimation equation can then be obtained by substituting (27) and (26) into the cuto¤ ex-

pression (22); substitute this into the share equation in (25) and use the distribution assumption in (24) to

obtain for each t; i; V

lnntiV = k

"� ( pre (1� EUit) + postEUit)

�

1� �ph����0iV � �

��hiV

�=���0iV

� � 1 � �

� � 1 ln �tiV

#(28)

+ k

�1

� � 1 lnAti

KiV (1� �)+ riV t � ln cV

�+ lnntV + utiV

We note a couple of relevant points for the estimation. First, it is not obvious how to measure the potential

producers in an industry, moreover for some of our results it will be useful to know what happens to the

37 In the appendix we show that if pm were large then there would be an additional term where the high probability andtari¤ are replaced by the medium ones. Since, (a) there is no obvious empirical counterpart for the medium term, (b) it wouldbe highly correlated with the high value and (c) we have good reasons to believe pm � 0 given these were transitional tari¤sthat could not be sustained under GATT rules, we ignore this extra term.

30

number, not just the fraction of exporters, so we move the lnntV term to the RHS and control for it via

industry by time e¤ects. Second, there are three assumptions that we use in the baseline estimation to

identify the e¤ect of uncertainty: (i) the shape parameter k is common across V (but we allow the other

parameter cV to be �exible); (ii) conditional on a policy shock for the policy in a particular importer we

assume that producers share a common ph among them and across importers. However, they have market

speci�c information about the impact of that worst case scenario on pro�ts (captured by the tari¤uncertainty

measure); (iii) the elasticities of substitution are similar across sectors. In the robustness section we will

discuss the impact of relaxing some of these assumptions. Given these we can write the estimation equation

in terms of variables and parameters as follows

lnntiV = (b pre (1� EUit) + b postEUit)����0iV � �

��hiV

�=���0iV

� � 1 +b� ln �tiV+ati+aiV+atV+~utiV for each t; i; V

(29)

where b = � pTh �k1�� captures the impact of uncertainty on entry. The coe¢ cient on the applied tari¤ is

b� = � k���1 . The ax terms represent country-year, country-industry and industry-time e¤ects that absorb

among other things, the demand and cost conditions in Ait, the investment cost KiV , the productivity

heterogeneity across industries cV as well as other terms that vary at the x level and were previously

included in the remainder term riV t and uiV t with the remaining part of the disturbance that varies at the

iV t level included in ~utiV .

Since the central question for the estimation is whether the agreement reduced uncertainty, i.e. post <

pre , we focus on testing if b post � b pre > 0 by estimating (29) in di¤erences taking a period after the

agreement was implemented and one before it.

�t lnntiV = (b post � b pre)EUi����TiV � �

��hiV

�=���TiV

� � 1 + b��t ln �tiV + ai + aV + ~uiV for each i; V (30)

In sum, we are interested in understanding the impact of the agreement on investment and entry. To

isolate the impact from trade policy we explore the variation across industries and countries. To isolate

the uncertainty e¤ect we control for changes in applied tari¤s and ask if there was larger entry into those

industries that were previously subject to more uncertainty. We also control for industry e¤ects to account

for, among other things, the fact that these industries could have been expanding to all markets for other

reasons The importer e¤ects address any shocks speci�c to those markets.

31

5.2 Data and Implementation

To estimate (30) we collect detailed data on trade policy for Spain and the original EC-10 countries before

and after the agreement, as described in more detail in the data Appendix. So the uncertainty measure

varies not only across industries but also across members of the agreement. For some industries the policy

data are recorded at a �ne level of disaggregation, so they could potentially be matched to 6-digit NIMEXE

classi�cations for the trade data, which includes over 5000 products. However, we argue this is not the

correct level of disaggregation to test the model for a few reasons. First, the model suggests that we de�ne

industries according to a set of characteristics (such as productivity distribution) that is common across a

set of �rms, which is clearly broader than the 6-digit level. Second, even though the policy is recorded at

a �ne product level, most of the variation in the policy occurs across industries, rather than within them.

For example, about 80% of the variation in applied tari¤s faced by Portugal in exporting to the EC 10

before the agreement is accounted for by di¤erences across 2-digit industries (of which there are 99). For

the main uncertainty variable, 75% of the variation is across 2-digit industries. Those fractions are lower

for Spain but still more than half of the variation is accounted for by cross-industry di¤erences. Third,

even in 2-digit industries where there is some variation in tari¤s, an exporter�s perception of the worst case

scenario is likely to be broader than what is implied by the worst case for a single 6-digit good. This is

clear if he exports more than one good in an industry. But is relevant even if he exports a single 6-digit

good. To see why note that goods can face tari¤ changes simply because they are reclassi�ed. For example,

there were product reclassi�cations in 1983, 1988, 1992, 1996, 2002. Between 1987 and 1988 for example the

classi�cation system changed dramatically with the introduction of the harmonized system. However, the

top classi�cation, the so-called 2-digit or chapters are actually quite similar even across these two systems.

Therefore it seems more reasonable that an exporter of a good in a 2-digit industry consider the uncertainty

for a typical good in that industry rather than only considering the uncertainty for a particular 6-digit

product.38 Finally, if we were to run the model at the 6-digit level there would be a large number of 0�s.

Since our estimation equation is in logs we would eventually have to drop those categories, which could be

those where uncertainty was most important. 39

38Thus another advantage of de�ning the relevant industry at the 2-digit level is that it will allow future work to examinethe uncertainty impacts even after the product classi�cation changes in 1988.39We could try to address this using a selection model to explain the participation at the 6-digit level. To the extent that

non-participation is driven by policy uncertainty at the 6-digit product level that would be an interesting exercise. However, aswe just argued, the policy uncertainty will mostly vary at a higher level and so it would be unlikely to explain a lot of variationin participation within industries. Moreover, there will be a number of reasons outside the model why no �rms export in aparticular 6-digit category. So we are skeptical that this type of selection correction would address the sample bias that resultsfrom droping the 0�s.

32

To construct the uncertainty measure we �rst take �hi for a product to be the ad valorem conventional

GATT tari¤ that country i (EC-10 or Spain) had before the agreement. If that tari¤ was not bound in the

GATT then we use the autonomous ad valorem tari¤ that i applied. We take �0i to be the tari¤ that i

actually applied to Portuguese exports in that product before the agreement, where we employ data on the

set of preferences that these countries provided to Portugal, as described in section 3.2. We then construct

the uncertainty measure in (26) using elasticity values that are consistent with the data for these countries

(� between 2 and 4). In the robustness section we provide supporting evidence for this choice of elasticity

and show the results are robust to alternative values. We then aggregate this measure and the applied tari¤

to the 2digit industry level.

The tari¤s that Portuguese �rms exporting to Spain faced in the years 1985 and 1987 appear in the table

below. The typical industry in Portugal enjoyed preferential tari¤ that was nearly 50% below the tari¤

levied on the rest of the world. Figure 7 also shows that this di¤erence is not driven by any one set of goods

but occurs along the full distribution of tari¤s, which is shifted towards zero for Portuguese exporters (blue

line) relative to the rest of the world (red). If Portugal were to lose these preferences, the typical exporter

would see his pro�ts reduced by over 16% per annum. With respect to the EC-10, the table shows Portugal

enjoyed lower preferential tari¤s by 1985 but the proportional loss in pro�ts was nearly as high as in Spain

at 15%. The magnitude of tari¤ reductions in 1987 is small since tari¤s in industrial products were already

zero prior to accession.

02

46

810

Den

sity

0 .2 .4 .6Log Tariff

Spain EFTA Preference 1985 Spain Tariff on RoW 1985

Spain: Preferences for Portugal vs. RoW

Figure 7

33

5.3 Baseline Estimates

Table 3 provides estimates of the parameters in (30). We �nd that �rm entry is negatively a¤ected by

applied tari¤s, as predicted by the theory. Moreover, entry was strongest in the industries that initially

faced more uncertainty. In particular, b post � b pre > 0, which according to our model implies that the

agreement reduced uncertainty, i.e. post < pre, and this lead Portuguese exporters to enter the EC and

Spanish markets.

One potential concern with the results in column 1 is that ad valorem tari¤s were only one part of the

protection faced by Portuguese exporters. If applied protection that used other instruments fell by more in

those industries where there was higher uncertainty this would bias the estimates.40 Therefore in column

2 we control for changes in speci�c tari¤s and in column 3 for changes in �non-tari¤ barriers�Both have

the predicted negative sign but they are insigni�cant. Neither a¤ects the baseline results for uncertainty

and applied ad valorem tari¤s.41 The results are robust to other changes, which we discuss in detail in the

following section.

The theory has implicitly assumed single product �rms. However, it can be easily re-interpreted as

applying to a �rm�s decision to invest in order to introduce a new product into a country. That cost may be

present even if the �rm already sells another good in that market. If most �rms are single product this should

not a¤ect our results but if the trade expansion had been driven by multi-product �rms were introducing

new products then the results in table 3 would not fully re�ect the impact of uncertainty reduction. Given

that we can observe in the data that the typical Portuguese exporter sells only two types of product (at

6-digit Nimexe) both in 1985 and 1987 we do not expect this to a¤ect our results (the average number is also

approximately unchanged at about 6). Nonetheless we can test this directly by re-estimating (30) using the

number of varieties (i.e. total number of product �rm combinations in an industry). The results in Table 4

con�rm the baseline estimates both in terms of magnitude and signi�cance.

40To construct these measures we use information in the tari¤ schedules on whether a product line was subject to speci�ctari¤s, special import authorization or other conditions that were not translated into an advalorem tari¤. As is standard inthis type of literature we construct a coverage ratio measure: fraction of products in industry V that are subject to a particularmeasure (e.g. speci�c tari¤, or other NTB) and took the di¤erence before and after the agreement.41Note that the policy measures vary across industry and for Spain vs. the EC-10 but not within the EC-10. To address

this we also re-estimated the standard errors allowing for clustering with arbitrary correlation across EC-10 countries withineach industry , and similarly for spain. This makes no di¤erence to the statistical signi�cance reported in the table. The sameapplies to subsequent results in tables 4 (varieties) and 5 (falsi�cation)

34

5.4 Quanti�cation and Additional Evidence

We now provide some quanti�cation of the impact of the alternative policy dimensions� applied tari¤s and

uncertainty� on both entry and the value of exporting.

One way to compare the relative impact of the policies is to ask how much variation in entry they

explain. For the full sample we �nd that a one standard deviation reduction in applied tari¤s leads to a 0.14

standard deviation increase in entry whereas for uncertainty that e¤ect is 0.4, which is almost 3 times larger

(0.4/0.14). If we focus on the EC-10 we �nd that uncertainty is 5.8 times more important than applied

tari¤s since there was little variation in the latter.

Since the applied tari¤s went to zero and, as we will argue below, so did uncertainty, we can also ask what

the estimates imply for the magnitude of the removal of the applied tari¤s and uncertainty. The elimination

of applied tari¤s only generated about 4% growth in entry overall, 2% into EC-10 (their mean reduction was

only 0.7 p.p.) and 20% for Spain (mean reduction of about 7 p.p). The elimination of uncertainty on the

other hand generated a 31% growth in entry overall, similar in both Spain and the EC-10. So the overall

growth explained by both these e¤ects is about 35% and it roughly matches that of the sample (33%).42

We can also quantify the impact of policy on exporting pro�ts. It is simple to do so for applied tari¤s,

since the elasticity of operating pro�ts with respect to � is simply �� and the same is true for the value

of exporting around = 0. So if after the agreement the EC and Spain were to raise tari¤s back up to

pre-levels and we assumed there was no uncertainty then the percent reduction in pro�ts of Portuguese

exporters would be simply ��� ln � ~4.2% if � = 3 and using the sample mean tari¤ change of 1.4%. The

impact is larger for exports to Spain (~20%) given their higher initial tari¤s.43

To compare the �gure above with the impact of uncertainty, we �rst recall our interpretation of����0V � ���hV

�=���0V

as the percentage pro�t reduction from a shock that moves tari¤s from �0 to the worst case scenario. This

variable ranges from zero to 44% and, as the summary statistics in table 2 indicate, its average is about

16% (similar for the EC-10 and Spain). For the many industries where the EC was already providing full

preferences to Portugal the mean is 19%. Once we accumulate these values over time we can understand

why exporters may be reluctant to pay a sunk �xed cost to invest that would have to amortized over a long

period of time.44

42Naturally, there are other factors a¤ecting the sample value, some of which accounted by the country and industry e¤ectsand others unexplained by the model.43This assumes � = 3 and provides intermediate values to the alternative cases � = 2; 4. While this estimate is clearly

sensitive to the � value, the relative magnitude when we compare to uncertainty is less so.44We think these may be underestimates for some industries of the true loss in pro�t if a bad shock hits for the following

35

To quantify the uncertainty reducing impact of the agreement on the value of exporting we now use

our baseline estimates. The counterfactual we are interested in mirrors the one described above for the

applied tari¤. Namely, we ask what is the percent change in the value of exporting if we were to reverse

the uncertainty e¤ect of the agreement but maintain current tari¤s at free trade. To do so we divide the

di¤erence in the exporting value function derived in (19) by �e(�t = 1; post = 0) to obtain

��e(� = 1; post = 0)��e(� = 1; pre > 0)�e(� = 1; post = 0)

= � � pre1� �(1� pre)

� (1)� E�(� 0)� (1)

(31)

� � � preph1� �(1� pre)

�1� ���hV

�where the last line uses the de�nition of operating pro�t and the approximation is exact in a two-state world

(since in that case E�(� 0) = (1� ph)� (1)� ph�(�h)).

We now show that our estimates measure � ( pre � post) ph and we can then use data for the discount

factor, �, and tari¤s to obtain�1� ���hV

�, which was about 19% on average for the many industries where

the EC provided full preferences to Portugal before the agreement.

Using the structure of the model we can provide an estimate of the change in the probability of a

policy shock that generates the worst case scenario. We estimate b post � b pre = ( pre � post) phk �1��

and b� = � k���1 so we can calculate ( pre � post) ph =

���1

1���

�b post�b pre

�b�

�. We use � = 3, the value

used to construct the uncertainty measure and a discount factor � = 0:9. We obtain that the change in

probability was about 0.24 or 24 percentage points. First, we note that it is striking that the result is within

the admissible range [0,1] since nothing in the estimation ensures that would be the case. Second, this seems

to be a signi�cant but not unreasonably large e¤ect.45

We can use this to quantify (31). Since the latter assumes post = 0 (and we will provide evidence

below that supports this) we now have that preph=0.24 and so pre and ph are each bounded to be at

least 0.24. Therefore we obtain that the value of exporting if the uncertainty e¤ect of the agreement were

reversed is at least 4.1% (= 0:91�0:9(1�1)0:24 � 0:19, when ph = 0:24 and = 1) but may be as high as 13%

(= 0:91�0:9(1�0:24)0:24�0:19, if ph = 1 and = 0:24). Given our interpretation of as the arrival rate of shocks

it seems extremely unlikely that exporters expected a large shock that leads to all policies being reviewed

reason. To obtain these values we must take �hiV to be an observed tari¤ that is imposed on the rest of the world. Whilethis captures a situation where Portugal loses its preferences it is an underestimate for an even worse case scenario where theimporter raises temporary protection via other trade policies, e.g. anti-dumping.45Recall that � = (1� �)=(1+R) so our assumption is equivalent to alternative reasonable combinations of these parameters

such as a real interest rate R = 0:03 (average for Portugal in 1983-1995 period) and � = 0:08. For a reasonable range ofalternative discount factors (� 2 (0:85; 0:95) we continue to obtain reasonable values ranging from 12-39 percentage points.

36

every year so an intermediate value of = 0:5 seems more reasonable and implies an e¤ect equal to 7: 5%

( 0:91�0:9(1�0:5)0:24 � 0:19). This indicates that the value of the agreement for Portuguese exporters that was

generated by the uncertainty reduction alone is almost twice as large as the value generated by applied tari¤

changes.

If post is zero then our estimate of 0.24 above also captures exactly preph, i.e. the level e¤ect of

uncertainty before the agreement. If that is the case we should �nd a negative impact of uncertainty on

the level equation prior to the agreement and that coe¢ cient should be similar in magnitude to the one

estimated in di¤erences. To test this we need an additional identi�cation assumption. Namely, to identify

the b pre level we can use

lnn0iV = b pre

����0iV � �

��hiV

�=���0iV

� � 1 + b� ln �0iV + ai + aV + ~u0iV for each i; V (32)

The key di¤erence is that we can no longer include industry by importer e¤ects (we can and do include

industry e¤ects though). In terms of the structural model this amounts to a restriction that lnKiV be

additively separable into an importer and industry component and a random disturbance uncorrelated with

tari¤s. Column 1 in table 5 estimates this and indicates that b pre = �4 which is nearly identical to our

estimate for � (b post � b pre). The results for varieties (column 3) have a similar implication. This suggests

that the impact of uncertainty on entry we estimate in the baseline is coming from the elimination of that

uncertainty.46 We also obtain more direct evidence of this when we run the equation above but pool all

years 1983-1987 and allow the uncertainty e¤ect to be di¤erent post agreement, we then �nd that b post = 0

and b pre < 0 and signi�cant. So the interpretation of change e¤ect estimated in the baseline as simply

capturing �b pre seems correct.

Table 5 also indicates the importance of using �rm level data to detect these e¤ects and provides some

additional evidence for the model. It runs a similar regression for the period prior to the agreement using

total exports (column 5). It �nds no signi�cant e¤ect of the uncertainty measure. Column 2 indicates why

this is. We can decompose total exports into an extensive margin (e.g. �rms or variety) and an intensive

one (average sales/�rm or variety). When we do so the e¤ect of any given variable x on the (ln) of total

46The results are also consistent with another prediction from the theory: if there is uncertainty in the period before theagreement then the tari¤ impact on entry is attenuated and if this e¤ect is large enough it will bias down the �rst order e¤ect ofthe tari¤ that is estimated prior to the agreement. This can explain why the magnitude of the tari¤ impact in this estimationis lower and less precise and indicates one reason why estimates of the impact of policy changes that focus on the applied policyalone and are done prior to the agreement can undertate their true impact.

37

sales to a destination can be decomposed into the sum of the e¤ects on each of these margins

@

@xln�v2V pivqiv=�iV =

@

@xlnntiV +

@

@xln�v2V pivqiv=�iV

ntiV

The theory predicts that uncertainty depresses entry, as we have veri�ed, but also that the impact should

not be the same for the intensive margin. It predicts that we should observe higher sales per �rm in industries

where uncertainty is higher. The intuition is as follows: in industries with higher uncertainty only the more

productive �rms enter and sell, as the cuto¤ equations showed. Moreover, as we can see from (7), sales are

higher for more productive �rms. That selection e¤ect is visible in the positive impact of uncertainty on the

intensive margin in columns 2 and 4 of Table 5. However, it is not statistically signi�cant, perhaps because

it is attenuated at this level of aggregation.

The �gure below also provides direct evidence that new exporters in 1987 were smaller than continuing

ones, as the theory would predict. It would also be interesting for future work to examine whether the size

distribution of this cohort of new exporters in 1987 evolves towards that of previously existing exporters.

We expect the two to become closer but the interesting question is how fast and through what mechanism:

one possibility is selection: there is a larger share of less productive (and smaller) �rms that are thus more

likely to exit. Other alternatives include growth of the entrants either by learning or by overcoming �nancial

constraints.47

0.0

5.1

.15

Den

sity

0 5 10 15 20Log Export Value (euro)

new exporters continuing exporterskernel = epanechnikov, bandwidth = .45

Distribution of Continuing vs. New Exporters (1987)

Figure 8

47Cabral and Mata (2003) provide evidence for the �nancial constraints explanation but their focus is on employment sizeof all Portuguese manufacturing �rms rather than sales size for exporters.

38

5.5 Robustness

We now discuss some robustness tests of the baseline results.

Column 3 of Table 3 adds the change in the standard deviation of the tari¤ faced by Portugal in each

industry, i.e. �(stdev ln �tiv) where v 2 V . There are two possible motivations for this control. First, one

may argue that our model is misspeci�ed and for some reason the exporters care not only about the mean

of the applied tari¤ in an industry but also its dispersion, particularly since we are aggregating �rms up to

the industry level. To the extent that our uncertainty variable includes some nonlinear transformation of

the applied tari¤ it may be capturing some of that potential e¤ect. The second argument would be that our

measure of uncertainty is incorrect and that the perhaps more intuitive measure of change in uncertainty is

�(stdev ln �tiv). We �nd a positive e¤ect of this variable, which may be consistent with the last argument.

However, that e¤ect is insigni�cant and does not change the value or signi�cance of the theoretically based

uncertainty measure.48

We now provide some supporting evidence for our use of common elasticities and investigate if the results

are sensitive to it. There are two assumptions: �rst, the typical elasticity within industry V is similar to the

typical elasticity in another industry. Below we provide some direct evidence based on estimated elasticities

that support this assumption. Second, the elasticity of substitution across industries is similar to the typical

elasticity within them. We do not have estimates for cross industry elasticities to fully justify this second

assumption and thus we examine directly whether the results are robust to it.

The elasticity of substitution across industries is possibly lower than within industries. Our model can

be extended to accommodate this. In particular, if we assume that the subutility index Q in (2) is a Cobb-

Douglas aggregator with shares �V =� then the elasticity of substitution across industries is unity (so smaller

than �) and the key di¤erence for our model would simply be that that the price index is now PiV , which

is de�ned only over the varieties in each industry V . Therefore, we should rewrite the A term as follows

lnAitV = ln(1� �)�V Yit�

wtPitV �

�1��

Our baseline estimation is in di¤erences and we can show that a number of components that this alter-

native speci�cation of demand introduces are di¤erenced out. To see this clearly suppose we can rewrite the

48We also �nd that the applied tari¤ e¤ect is slightly higher and more precise, suggesting perhaps that exporters care aboutboth dimensions of applied policy.

39

price index as a product of four terms, PitV = PitPiV PtV pitV , which re�ect respectively variation that is

only country-time, country-industry or industry-year speci�c and the last term, pitV , which can vary along

all three dimensions. If we consider changes in lnAitV we then have

�t lnAitV = (� � 1)�t ln pitV +h(� � 1)�t (lnPitPiV PtV ) + �t ln

�Yit (wt)

1���i

The key thing to note is that in terms of our di¤erenced estimation equation (30) the industry and

country e¤ects will still continue to capture all the variation in the costs and demand (Yit (wt)1��) and also

a substantial part of the variation in the price index, namely �t (lnPitPiV PtV ). We are left with the residual

variation in the price index, �t ln pitV . This is only an issue for our estimates to the extent that it may be

correlated with the policy measures. Recall that these price indices re�ect the prices of all varieties sold in

those industries in country i. Therefore it will be dominated by the domestic varieties and imports from

countries other than Portugal since Portugal is a relatively small exporter. Therefore we do not think that its

expansion into their markets had a substantial direct e¤ect on those price indices �t ln pitV . However, there

may be omitted variable bias if a third factor a¤ected these indices and was correlated with the changes in

policy faced by Portugal. The most obvious candidate would be if the EC-10 or Spain were simultaneously

reducing their tari¤s on the rest of the world and those reductions were correlated with the policy changes

they were implementing for Portugal. This was not the case for the EC-10 external tari¤ in the period

we consider. However, Spain was reducing its external tari¤s on the rest of the world (to converge to the

European Common tari¤) and these reductions were correlated to the ones faced by Portugal. Therefore

we use changes in tari¤s to the rest of the world to proxy for �t ln pitV .

The results that control for industry and country speci�c price index changes are presented in columns 5

and 6 of Table 3. We �nd a positive relationship between the price index and entry. This is as predicted by

the theory: a decrease in the price index in an export market makes Portuguese exporters less competitive

and thus lowers entry. However, this e¤ect is insigni�cant whether we use log changes (column 5) or add a

quadratic term to account for the non-linearity of the price index in tari¤s (column 6). More importantly,

controlling for these e¤ects does not change the baseline results regarding uncertainty or the applied tari¤

e¤ects. The same is true if in addition to these price index terms we also include all the other applied policy

controls in columns 2,3 and 4. Since these controls were individually and jointly insigni�cant we generally

focus on the baseline results without them.

40

We now examine our assumption that the typical elasticity within industries is similar across 2-digit

categories. Thus far the estimates assume a � = 3�a value based on our calculations using the sub-sample

of estimates from Broda, Limão and Weinstein (2008) for Spain and the other EC-10 countries (except for

Greece, Belgium and Ireland, which were not in their sample). The median for these countries over all

industries is 3.4 and the mean is 4.5. Since they estimate the elasticity at a more disaggregated level (hs-4)

than what we use (roughly hs-2), it is possible their estimates are upper bounds on the 2-digit elasticities.

To test if our results are sensitive to this we re-estimated the baseline results in Table 3 using � = 2; 4 and

found no signi�cant changes (results available on request).

We can also provide evidence for one of our simplifying assumptions in the model and baseline estimation:

similar � across countries and industries. While this elasticity is not constant within several 2-digit categories,

it turns out not to vary that much across those broad industries. For example, if we take the estimates of

� at the hs-4 level for Spain we �nd that only 10% of its variation occurs across 2-digit industries. There

is also not considerable dispersion across countries: the median elasticity across all hs-4 categories ranges

only from 2.8 in Spain to 3.9 in Austria. Moreover, they are highly positively correlated across countries.49

There is also not a lot of dispersion in the typical elasticity across 2-digit industries in these countries. As

we noted the overall median is 3.4 and, in 90 out of the 93 industries for which we have data, the median

(over European countries) of �V is between 2.2 and 4.8, only 3 industries have higher values: 5.5-6.3. Given

these estimates are subject to measurement error it is unlikely that there would be signi�cant statistical

di¤erences between most of them. Nonetheless we also re-estimated the baseline speci�cation dropping the

three industries with higher elasticities (18, 47 and 87) and veri�ed the results are unchanged.

Finally, we note that the variation that does exist across industries is not in any obvious group, e.g.

industries 1-14 (which contain basic agricultural products) has a median elasticity of 4, which is only

somewhat higher than overall sample. One potential concern with the agricultural products is that they are

subject to non-tari¤barriers and they are a non-negligible fraction of industries and thus of the sample (about

22%). So they could bias our results if these NTBs were correlated with our uncertainty measure before the

agreement and were removed after. One way to address this is to control for NTBs directly. We did this in

table 3 and veri�ed the results did not change. One may also object to applying a monopolistic competition

framework to these types of goods and argue that they should be dropped altogether. We are agnostic

49For example, if we take the parameter on applied tari¤s that we assume to be constant, �iV = (�iV � 1), for each industryV in Spain and regress it on the median value for that industry across the EC-10 countries we obtain a coe¢ cient of 1.2 witha s.e. of about 0.2.

41

about this but nevertheless when we do drop agricultural goods we continue to �nd that uncertainty has an

e¤ect that is qualitatively and quantitatively similar to the baseline case (for number of �rms of varieties).

However, the applied tari¤ coe¢ cient is now less than half in magnitude and statistically insigni�cant. This

is not surprising since the tari¤ reductions by the EC mostly occurred in those agricultural products so

the products that remain in the sample were ones where Portugal was already receiving signi�cant tari¤

concessions. This again stresses the message that the uncertainty reduction was a key motive for entry.

6 Conclusions and Policy Implications

We �nd evidence that Portugal�s trade integration in 1986 generated a signi�cant reduction in the uncertainty

its exporters faced in EC countries. By combining the insights from a dynamic model with detailed trade

policy we can compute a trade uncertainty measure that indicates that on average Portuguese exporters

stood to lose about 16% of exporting pro�ts in the event they lost their preferences in the EC-10 or Spanish

markets. Combining this with �rm entry data we estimate that exporters thought such an event had a real

probability of occurring before 1986 (25%) but not after. This generated considerable investment and �rm

entry into Spain and the EC-10, more so than the applied tari¤ changes themselves. We also showed that

�rm entry was a key margin of growth during this period, accounting for about 75% of Portuguese real

export growth to these countries.

The theoretical and empirical results have interesting implications for Portugal and the world trading

system and for future research, some of which we now highlight.

Accession to the EC lowered average trade barriers and uncertainty surrounding them for foreign ex-

porters to Portugal as well. This is obvious for the EC exporters but is also true for the rest of the world

because Portugal�s high tari¤s had to be reduced and harmonized with the EC. Protection in Portugal had

been both high and variable, in addition to tari¤s that were about 17-20% in the typical industry, there

were also several non-tari¤ barriers in the early 1980�s aimed at constraining imports to address the external

de�cit. Therefore, the EC accession should have lead to a marked improvement in market access for foreign

exporters to Portugal. The aggregate data is consistent with this prediction since it shows that real imports

increased almost three times between 1985 and 1992. The growth was quite high for consumption, invest-

ment and intermediate goods, which indicates that both consumers and �rms may have bene�ted from it.

Further work is required to determine the role of trade policy but the data suggest that this import boom

42

was not simply an income e¤ect. First, imports grew much faster than GDP. Second, the nominal prices in

escudos for total imports and intermediates remained approximately unchanged between 1985-1992 despite

the nominal depreciations, which would tend to make imports more expensive. This downward pressure on

prices suggests entry of new foreign exporters, which is consistent with a reduction in trade barriers and

uncertainty. New trade theory and recent estimates highlight the important role of new intermediate goods

in increasing �rm level productivity and so in future work it would be interesting to test if this was one

outcome of Portugal�s EC accession and whether it worked via reductions in uncertainty or applied tari¤s.

Another interesting issue to examine is the impact of the agreement on foreign direct investment (FDI),

which involves substantial sunk costs and is subject to much policy uncertainty. While it is known that FDI

in Portugal increased since the accession, we would like to know the exact channels. The model highlights

how reduction in policy uncertainty can attract foreign investors to Portugal. The basic mechanism of the

model suggests this occurs because they can now produce in Portugal, possibly taking advantage of lower

wages, and export to the EC without worrying about policy uncertainty in those markets. An extended

version of the model would provide an added bene�t for FDI to locate in Portugal: the reduced uncertainty in

Portuguese barriers now allows imports of a wider variety of intermediates more cheaply, as argued above.50

Our framework can also be extended to analyze the interaction of trade with other sources of uncertainty,

such as exchange rates. We did not �nd much evidence of the impact of exchange rate volatility in the 1981-

1992 period. However, such e¤ects may have been stronger during the 1990�s due stress put on the European

Exchange Rate Mechanism and the lead up to adoption of the Euro. One may also examine if the trade

policy e¤ects we �nd are stronger relative to countries with lower exchange rate volatility. These second

order e¤ects can be estimated by extending our approach to include interactions of measures of uncertainty

in trade policy and the exchange rate.

Our results have policy implications for Portugal and the world trade system more broadly. First, as

we describe in 3, many countries receive unilateral preferential tari¤s that are subject to the discretion and

uncertainty of policymaking. These programs share some of the characteristics that Portugal�s preferences

with the EC-10 and Spain did before 1986. Thus our results provide one reason why these programs are

not always successful in promoting trade and investment and how this may change if those preferences are

secured through formal PTAs. If these countries seek secure access to the European market Portugal can

bene�t from cheaper, more diverse imports. However, this may also present a challenge for Portuguese

50These �gures are based on authors�calculations from data in Pinheiro et al (1997).

43

exporters that will then have to compete with a wave of entrants. This challenge is even more pronounced

when we consider the growing number of PTAs being concluded by countries where Portugal does not enjoy

preferential treatment such as the US, which in 2005 still accounts for about 10% of Portuguese exports.

Emerging markets such as China, Brazil and Angola are also becoming an increasingly important des-

tination for Portuguese exports (5% in 2005) and their trade policy tends to be uncertain for a variety of

reasons. Our results suggest that agreements that secure access to these and other large emerging markets

(e.g. India) can be as important or more than extracting reductions in their current trade policy barriers.

This insight is particularly important for an ongoing disagreement in the context of the WTO where some

developing countries, such as India, want the tari¤ reductions to apply only to maximum allowable tari¤s (or

bindings). Many developed countries seem to place little value on binding reductions if these remain above

applied rates (as they do in several developing countries) and thus want to focus on reduction to current

applied tari¤s. Our results suggest that reductions in both types of protection are the most e¤ective but

they also show that reducing the maximum and leaving the applied unchanged will also generate investment

and trade by reducing uncertainty.

The last point underscores one important bene�t of the WTO that has not yet been quanti�ed: the

value of tari¤ bindings. This bene�t of WTO membership was probably an important motive for China�s

accession and a possible explanation for some of its subsequent export growth. Prior to 2002, Chinese

exporters were never certain of the amount of protection they would face in markets such as the U.S. where

tari¤s on Chinese exports were subject to annual Congressional review. The role of the WTO in ensuring

predictability in trade policy will remain an important factor in coming years since we anticipate a number

of potential sources of trade restrictions. These include rising food prices; concerns with product safety;

import duties to counter Chinese currency �manipulation�; and �environmental�duties at the border to

o¤set di¤erences in carbon emissions in production.

In conclusion, our results highlight why and how much trade policy uncertainty a¤ects investment and

entry into new markets. While the importance of policy credibility is by now recognized, it is generally

di¢ cult to measure its impacts. To the extent that our approach and results do just that they may be of

broader interest to economists and policy makers interested in evaluating the impact of other policy reforms

on �rm-decisions.

44

A Theory Appendix

A.1 Derivation of Value Functions

Deriving the full set of value functions is a basic application of linear algebra. The solutions to the set ofequations in (12),(13),(14), and (15) is

�e(�t) =�(�t)[1� �(1� Et[�(� 0)])][1� �(1� )](1� �)

Et�e(�0) =

Et[�(�0)]

1� �

�w(�t) = � H(�1)(1� �)Et[�(� 0)j� 0 < �� ]� � Et[�(� 0)]� (1� �)[1� �(1� )]Ke

[1� �(1� )][1� �(1� H(��))]

Et�e (�0j� 0 � ��) = � Et[�(�

0)]� Et[�(� 0)j� 0 < �� ](1� �)(1� � + � )(1� �)

A.2 Pro�t loss term

1. �(�t) � 1We denote the maximum tari¤ by �h.

�(�t) =�E(���) +H(�t)[�

��t � E(���j� � �t)]

�=���t

=

�Z �h

1

���dH(�) +H(�t)���t �

Z �t

1

���dH(�)

�=���t

=

�Z �h

�t

���dH(�) +H(�t)���t

�=���t

=�(1�H(�t))E(���j� � �t) +H(�t)���t

�=���t

Then to show that �(�t) � 1 we take the di¤erence D of the numerator and denominator in the �nal lineabove

D =�(1�H(�t))E(���j� � �t) +H(�t)���t

�� ���t

= (1�H(�t))[E(���j� � �t)� ���t ]

� 0

The inequality follows because ���t is always greater than E(���j� > �t). When the current tari¤ is at themaximum of the support of H(�) such that �t = �h, then the di¤erence in brackets and the term (1�H(�t))are both zero.

2. �(�t) � 1 is increasing in the current tari¤As the tari¤ increases, the di¤erence between the current state and any shock to a higher tari¤ is reduced.

45

We will show �rst that d�t

d�t> 0

@�(�t)

@�t= [����t h(�t) + h(�t)�

��t � �H(�t)����1t ]=���t + [(1�H(�t))E(���j� � �t) +H(�t)���t ](����1)

= [��H(�t)��1t ] + ����1[(1�H(�t))E(���j� � �t) +H(�t)���t ]

= ����1[�H(�t)���t + (1�H(�t))E(���j� � �t) +H(�t)���t ]

= ����1[(1�H(�t))E(���j� � �t)]= �[(1�H(�t))E(���j� � �t)]=�1��

� 0

In semi-elasticity terms, this becomes

@�(�t)

@ ln �t= �[(1�H(�t))E(���j� � �t)]=���

� 0

This implies that as the current tari¤ �t increases, the proportional gap between the deterministic and

uncertain cuto¤ narrows. We can see that that if �t = �hthe derivative goes to zero. Thend ln cUtd ln �t

= ���1

and the elasticity of the cuto¤ under uncertainty evaluated at the tari¤ maximum equals the elasticity atthe deterministic cuto¤.

A.3 Cost cuto¤ results

1. The elasticity of the cuto¤ threshold is reduced under uncertainty: d ln cUt

d ln �t� d ln cDt

d ln �t

Using the expression for cUt from the text, we log di¤erentiate and derive this result in several steps

d ln cUtd ln �t

=d ln cDtd ln �t

+d lnUtd ln �t

= � �

� � 1 +d lnUtd ln �t

= � �

� � 1 +1

� � 1

��

(1� � + � �)d�td ln �t

�= � �

� � 1 +1

� � 1

��

(1� � + � �)

��[(1�H(�t))E(���j� � �t)]

���

��= � �

� � 1

�1�

��

(1� � + � �)

�[(1�H(�t))E(���j� � �t)]

���

���< 0

The elasticity of the cuto¤ with respect the entry trigger is negative under uncertainty. In addition, theterm in brackets is less than or equal to one. Entry is less responsive to tari¤ changes under uncertainty.The two exceptions (limiting cases) are when = 0 (i.e. tari¤s are deterministic) or when �t is already atthe maximum of the tari¤ distribution.

2. The cuto¤ is decreasing in the arrival rate of shocks : d ln cUt

d < 0

46

In the empirical work we focus on the semi-elasticity and we derive this comparative static here

d ln cUtd

=d lnUtd

=1

� � 1

�d

d ln(1� � (1� �))� d

d ln (1� �(1� ))

�=

�

� � 11� �

(1� �(1� )) ((1� � (1� �)) (�� 1)

We thus have

sign

�d ln cUtd

�= sign

��� 1

((1� � (1� �))

�< 0

which is negative since we showed above that �� 1 < 0 when > 0.Using the derivative, we can write the �rst order e¤ect around = 0 used in the estimation

d ln cUtd

���� =0

=

��

� � 11� �

(1� �) ((1� �)

�(�� 1) (33)

=�

1� �(�� 1)(� � 1) (34)

3. First-order Cuto¤ ApproximationWe take a �rst-order Taylor series approximation of cuto¤ condition in logarithms around �t = �0 and t = 0.

ln cU ( t; �t) = ln�cDt � Ut

�= ln cD (ln �0; t = 0) + lnU (ln �0; t = 0)+

+ (ln �t � ln �0)@

@ ln �ln cDt j(�0; =0) + (ln �t � ln �0)

@

@ ln �lnUtj(�0; =0)

+ ( t � 0)@

@ ln cDt j(�0; =0) + ( t � 0)

@

@ lnUtj(�0; =0) + rt

Evaluating this approximation at t = 0 and �t = �0 and using the equation (34) from above, we �nd

ln cU ( t = 0; �t = �0) = ln cD (�0; 0) + t

@

@ lnU j(�0; =0) + rt

= ln cDt (�0; 0) + 0

��

1� �

��(�0)� 1� � 1 + rt

=1

� � 1

��� ln �0 + ln

A

K(1� �)

�+ 0

��

1� �

��(�0)� 1� � 1 + rt

4. Real Option vs. NPV Cuto¤sIn section 4 of the text we note that(i) the cuto¤ under the option approach is always lower than under the NPV approachTo see this note that in the absence of the option to wait the last term in (20) drops out and we obtain thestandard NPV cuto¤, denoted cMt . Since the last term in (20) is non-positive the option cuto¤ is lower, i.e.cUt � cMt , which implies less entry than under the standard NPV case.(ii) the cuto¤ under the NPV approach can be higher or lower than the deterministic and thus reductionsin uncertainty can lead to less incentive for entry under the NPV approach.If the deterministic tari¤ were such that ���t = E(���) then these two cuto¤s coincide (as can be seen

47

if we combine the �rst two terms of (20) to obtain cMt =hAE(���)K(1��)

i 1��1

= cDt ). But if instead we hold

the current tari¤ at its long-run mean, i.e. �t = E(� 0), then the convexity of pro�ts in tari¤s implies thatthe Marshallian cuto¤ is higher than the deterministic cuto¤. To see this note that if �t = E (� 0) then(�t)

��= (E (� 0))

�� � E(���) (Jensen�s inequality for � > 1) so cDt � cMt at the long run mean of the tari¤distribution. This implies that if we actually eliminate uncertainty while holding the current tari¤s equal atthe mean in the deterministic case then there would be less incentive for entry, which is the opposite e¤ectof uncertainty from what we �nd using the real option approach.

B Data and Estimation Appendix

B.1 Data

Pre-accession policy dataThe earliest trade data for Portugal is from 1981 and the closest full EC trade policy schedule before thenis for 1980 (OJ L 342, 31.12.1979, p. 1�382 ). This, and the fact that EC applied tari¤s to Portugal inindustrial goods were the ones set in the 1977 agreement, and thus remained in place until 1985, lead us toinitially digitize and use the 1980 schedule.51 The 1980 schedule already re�ects some of the EC multilateraltari¤ bindings negotiated in the Tokyo Round. However, some of these bindings, which we use to constructour uncertainty measure, continued to be reduced over a period of time.52 Therefore, the actual realizationof the worst case scenario for Portuguese exporters between 1981-1985 may have entailed a lower tari¤ thanthat implied by the 1980 binding. Even for those goods the 1980 binding may be the appropriate one forthe model if for example the exporters do not immediately update their beliefs about the tari¤ distribution.The only trade policy schedule for Spain closest to the pre-accession year was 1984 (in fact it was theonly year we found a full schedule for in the 1980�s). This schedule contains Spain�s preferences relative toPortugal and the EEC as well as its policy relative to the rest of the world. The documentation we foundimplies that Spain�s preferential tari¤s for Portugal remained unchanged between 1984 and 1985 becausethe EFTA-Spain agreement that regulated these had reached a phase requiring additional negotiations ofindeterminate length.Post-accession policy dataTo construct the tari¤ pro�le faced by Portugal immediately after the agreement we applied the concessionsschedule in the Articles of Accession, Protocol 3 for Spain (O¢ cial Journal L 302 , 15/11/1985 P. 0410)and Article 243 for the EC (O¢ cial Journal L 302 , 15/11/1985 P. 0094). These imply staged reductions of12.5% per year for Spain and 14.2% for EC-10 with some variation across goods.

B.2 Estimation details

1. Empirical Implementation in Discrete CaseTo construct the empirical measure of �(�t) we consider a discrete probability distribution for tari¤s. Wethen ask, given that a policy shock above the current trigger �t arrives, what is expected value of theproportional loss in pro�ts? This quantity is summarized neatly by the term �(�t)� 1. In the tables below,we compute �(�t)� 1 for a two- and three-state tari¤ process relevant to our empirical implementation.

Two State Tari¤ Distribution: High, LowInitial State (�T = �s) Probability (ps) �(�T = �s)� 1

�h ph 0�l 1� ph �ph

����l � ���h

�=���l

In the two state case, any �rm with an entry trigger �t � �hwould enter when the tari¤ is in the high state.The likelihood of a shock to trade policy leading to a worse outcome is zero. As was the case with a general51While ultimately our baseline results only use data for 1985 and 1987 in order to isolate the e¤ect of the agreement in 1986

we also planned and ran robustness tests that include earlier years.52"Implementation of MTN concessions: Note by the secretariat, revision" TAR/W/8/Rev.3, October 15, 1981

48

continuous distribution, the cuto¤s in the deterministic and uncertain model will coincide. In the low state,�(�t)� 1 is nonzero and less than unity. In the estimations, we construct the observable counterpart to the�(�t) � 1 from tari¤ data and assumptions on �. We then use this measure to estimate the unobservableparameters and ph via regressions.

Three State Tari¤ Distribution: High, Medium and LowInitial State (�T = �s) Probability (ps) �(�T = �s)� 1

�h ph 0�m pm �ph

����m � ���h

�=���m

�l 1� pm � ph ��(pm + ph)

����l � E(���j� > �l)

��=���l

= �P

s=m;h[ps(���l � ���s )]=���l

The three state distribution is slightly more involved, but makes it clear how to generalize to many discretestates. We argue in the empirical section that Portugal had �medium" preferential tari¤s with respect toSpain by 1983 of a tenuous and inde�nite nature due to the EFTA-Spain agreement. If pm �! 0, then wesee that the measures in the second and third row coincide with our empirical implementation for the ECand Spain.

2. Legacy �rmsThe true fraction of �rms in a market at time t is n�tiV

n�tVand it is related to the model�s distribution by

n�tiVn�tV

�= G(cUtiV ) if c

UtiV � max cUt�niV all n

� G(cUtiV ) if cUtiV < max cUt�niV

so if cUtiV < max cUt�niV we can write

n�tiVn�tV

= G(cUtiV )�tiV

where �tiV =�1 + (1� �t�n)

G(cUt�niV )�G(cUtiV )

G(cUtiV )

�and (1� �t�n) is the survival probability until time t of

�rms that were present at the maximum cuto¤ period, say it is t� n . Using the distribution we then get

G(cUtiV )�tiV = G(cUtiV )

1 + (1� �t�n)

�cUt�niV

�k � �cUtiV �k�cUtiV

�k!!

So if we consider changes in the demand or cost conditions: foreign income, domestic wages, other timevariation not product speci�c, then the policy variables and anything that is product speci�c and not timevarying cancels out in � term and we get

�tiVG(cUtiV ) = G(c

UtiV )

1 + (1� �t�n)

[At�ni]

k��1 � [Ati]

k��1

[Ati]k

��1

!!

The observed fraction is equal to n�tiVn�tV

etiv where etiv is some random disturbance with a log normal distrib-ution centered around 0 (e.g. measurement error) and

lnntiVntV

= lnG(cUtiV ) +�utiV

(ln�ti + ln etiv)

We can then interpret the error term in the text as utiV = ln etiv if cUtiV � max cUt�niV or ln�ti + ln etivotherwise. Since we control for country-time e¤ects this potential source of legacy does not bias our estimates.

3. Structural interpretation of parameters

49

Comparing (28) to (29) we see thatb t = � t �

1��kph

ati+aiV +atV =k

��1 lnAti+�� k��1 lnKiV (1� �)� k ln cV

�+lnntV +(uti + rti + uiV + riV + utV + rtV )

where the last term in brackets simply accounts for the fact that some of the variation in the utiV + riV tterm can be broken down into an importer*year, importer*industry and industry*time e¤ect (one of theways to address the legacy �rms issues, as described in appendix)~utiV � utiV + riV t � (uti + rti + uiV + riV + utV + rtV )

50

References

Anderson, J. E. and van Wincoop, E. (2003). Gravity with Gravitas: A Solution to the Border Puzzle.

American Economic Review, 93(1):170–192.

Baier, S. L. and Bergstrand, J. H. (2007). Do free trade agreements actually increase members’ international

trade? Journal of International Economics, 71(1):72–95.

Baier, S. L., Bergstrand, J. H., and Vidal, E. (2007). Free Trade Agreements In the Americas: Are the Trade

Effects Larger than Anticipated? The World Economy, 30(9):1347–1377.

Baldwin, R. (1988). Hyteresis in import prices: The beachhead effect. American Economic Review,

78(4):773–85.

Baldwin, R. and Krugman, P. (1989). Persistent trade effects of large exchange rate shocks. The Quarterly

Journal of Economics, 104(4):635–54.

Baltagi, B. H., Egger, P., and Pfaffermayr, M. (2003). A generalized design for bilateral trade flow models.

Economics Letters, 80(3):391–397.

Barro, R. J. (2006). Rare Disasters and Asset Markets in the Twentieth Century. Quarterly Journal of

Economics, 121(3):823–866.

Bernanke, B. S. (1983). Irreversibility, Uncertainty, and Cyclical Investment. The Quarterly Journal of

Economics, 98(1):85–106.

Bernard, A. B. and Bradford Jensen, J. (1999). Exceptional exporter performance: cause, effect, or both?

Journal of International Economics, 47(1):1–25.

Bernard, A. B., Eaton, J., Jensen, J. B., and Kortum, S. (2003). Plants and productivity in international

trade. American Economic Review, 93(4):1268–1290.

Bernard, A. B. and Jensen, J. B. (1995). Exporters, Jobs, and Wages in US Manufacturing: 1976-87.

Brookings Papers on Economic Activity: Microeconomics, pages 61–112.

Bernard, A. B. and Jensen, J. B. (2004). Why Some Firms Export. The Review of Economics and Statistics,

86(2):561–569.

Bernard, A. B., Jensen, J. B., Redding, S. J., and Schott, P. K. (2007). Firms in international trade. Journal

of Economic Perspectives, 21(3):105–130.

Bloom, N. (2007). Uncertainty and the Dynamics of R&D. American Economic Review, 97(2):250–255.

Bloom, N., Bond, S., and Reenen, J. V. (2007). Uncertainty and investment dynamics. Review of Economic

Studies, 74(2):391–415.

Broda, C., Limao, N., and Weinstein, D. E. (2008). Optimal tariffs and market power: The evidence.

American Economic Review, 98(5):2032–65.

Cabral, L. M. B. and Mata, J. (2003). On the Evolution of the Firm Size Distribution: Facts and Theory.

American Economic Review, 93(4):1075–1090.

Campa, J. M. (2004). Exchange rates and trade: How important is hysteresis in trade? European Economic

Review, 48(3):527–548.

Chaney, T. (2005). Productivity overshooting: The dynamic impact of trade opening with heterogeneous

firms. mimeo, University of Chicago.

Chaney, T. (2008). Distorted Gravity: The Intensive and Extensive Margins of International Trade. American

Economic Review, 98(4):1707–21.

Constantini, J. and Melitz, M. (2008). The dynamics of firm-level adjustment to trade liberalization. In

Helpman, E., Marin, D., and Verdier, T., editors, The Organization of Firms in a Global Economy,

chapter 4, pages 107–141. Harvard University Press.

Cooper, R. W. and Haltiwanger, J. C. (2006). On the nature of capital adjustment costs. Review of Economic

Studies, 73(3):611–633.

Das, S., Roberts, M. J., and Tybout, J. R. (2007). Market entry costs, producer heterogeneity, and export

dynamics. Econometrica, 75(3):837–873.

Dixit, A. K. (1989). Entry and exit decisions under uncertainty. Journal of Political Economy, 97(3):620–38.

Evenett, S., Gage, J., and Kennett, M. (2004). WTO Membership and Market Access: Evidence from the

Accessions of Bulgaria and Ecuador. Staff report, Universitat St. Gallen.

Francois, J. F. and Martin, W. (2004). Commercial policy variability, bindings, and market access. European

Economic Review, 48(3):665–679.

Frankel, J. A. (1997). Regional Trading Blocs. Institute for International Economics, Washington, DC.

Grinols, E. L. and Perrelli, R. (2006). The WTO Impact on International Trade Disputes: An Event History

Analysis. The Review of Economics and Statistics, 88(4):613–624.

Handley, K. (2010). Entry and Exit Under Trade Policy Uncertainty: Theory and Evidence. Working paper,

University of Maryland.

Hassett, K. A. and Metcalf, G. E. (1999). Investment with Uncertain Tax Policy: Does Random Tax Policy

Discourage Investment? Economic Journal, 109(457):372–93.

Helpman, E., Melitz, M., and Rubinstein, Y. (2008). Estimating trade flows: Trading partners and trading

volumes. The Quarterly Journal of Economics, 123(2):441–487.

Hillberry, R. (2009). Review of international experience: ex-post studies of other PTAs and implications

for PTA design. In Jayasuriya, S., MacLaren, D., and Magee, G., editors, Negotiating a Preferential

Trading Agreements: Issues, constraints and Practical Options, chapter 2, pages 12–34. Edward Elgar,

Northhampton, MA.

IMF (2004). Exchange Rate Volatility and Trade Flows - Some New Evidence. Technical report, International

Monetary Fund.

Johnson, S., Kouvelis, P., and Sinha, V. (1997). On reform intensity under uncertainty. Journal of Compar-

ative Economics, 25(3):297–321.

Jones, V. C. (2008). Generalized System of Preferences: Background and Renewal Debate. CRS Report for

Congress RL33663, Congressional Research Service.

Karacaovali, B. and Limao, N. (2008). The clash of liberalizations: Preferential vs. multilateral trade

liberalization in the european union. Journal of International Economics, 74(2):299–327.

Kee, H. L., Nicita, A., and Olarreaga, M. (2009). Estimating Trade Restrictiveness Indices. Economic

Journal, 119(534):172–199.

Kehoe, T. J. (2005). An evaluation of the performance of applied general equilibrium models of the impact

of NAFTA. In Kehoe, T. J., Srinivasan, T. N., and Whalley, J., editors, Frontiers in Applied General

Equilibrium Modeling, chapter 13, pages 341–377. Cambridge University Press, Cambridge, UK.

Limao, N. (2007). Are preferential trade agreements with non-trade objectives a stumbling block for multi-

lateral liberalization? Review of Economic Studies, 74(3):821–855.

Limao, N. and Tovar, P. (2009). Policy choice: Theory and evidence from commitment via international

trade agreements. NBER Working Papers 14655, National Bureau of Economic Research, Inc.

Mansfield, E. D. and Reinhardt, E. (2008). International institutions and the volatility of international trade.

International Organization, 62(04):621–652.

Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate industry productivity.

Econometrica, 71(6):1695–1725.

Pinheiro, M. et al. (1997). Series Longas para a Economia Portuguesa pos II Guerra Mundial, Volume I -

Series Estatısticas. Banco de Portugal, Lisboa.

Roberts, M. J. and Tybout, J. R. (1997). The decision to export in colombia: An empirical model of entry

with sunk costs. American Economic Review, 87(4):545–64.

Rodrik, D. (1991). Policy uncertainty and private investment in developing countries. Journal of Development

Economics, 36(2):229–242.

Ruhl, K. J. (2008). The International Elasticity Puzzle. Working paper, NYU Stern.

Trefler, D. (2004). The Long and Short of the Canada-U. S. Free Trade Agreement. American Economic

Review, 94(4):870–895.

U.S. International Trade Commission (2008). Andean Trade Preference Act:Impact on U.S. Industries and

Consumers and on Drug Crop Eradication and Crop Substitution, 2007. Technical Report Publication

4037, United States International Trade Commission.

1 2 3Dependent variable (ln): Exports Number of Firms Exports/firm

EC-10*Post_86 0.239*** 0.182*** 0.0573[0.0595] [0.0460] [0.0766]

Spain*Post_86 1.231*** 0.965*** 0.266**[0.130] [0.0932] [0.116]

US*Post_86 -0.103 -0.152 0.049[0.132] [0.0952] [0.0924]

Real Importer GDP (ln) 1.208*** 0.628*** 0.580***[0.259] [0.117] [0.224]

Importer Price Index (ln) 0.165** 0.0501 0.115**[0.0668] [0.0365] [0.0546]

Exchange Rate (ln) 0.163** -0.0188 0.182***[0.0670] [0.0341] [0.0554]

Observations 1590 1590 1590Adj R2 0.912 0.967 0.682

Margins of Growth DecompositionEC-10 1 0.76 0.24Spain 1 0.78 0.22

Notes:Includes dummies for country, year and year*advanced country. Robust standard errors in brackets. *** p<0.01, ** p<0.05Sample: Aggregate values to each country of destination where data is available

Table 1: Portuguese Export Growth Margins 1981-1992

EC-10 Spain Total

Pre Tariff (Portugal) 2.39 7.80 3.03(5.37) (5.14) (5.62)

Pre Tariff (GATT) 7.95 14.1 8.67(4.20) (7.75) (5.14)

Post Tariff (Portugal) 1.79 1.33 1.74(3.96) (3.51) (3.91)

Tariff Change (Port) -0.658 -6.56 -1.39(1.44) (4.78) (2.90)

Change in No. Firms 24.7 91.1 32.9(48.7) (62.55) (55.1)

15.6 16.3 15.6(11.2) (9.77) (11.0)

Observations (in levels) 682 92 781Notes:

Proportion of Profits Lost if Preference Reversed

Means of variables in percentage points. Standard deviations in parentheses. Tariffs are natural logs of 1 plus ad-valorem tariff aggregated to the industry level evaluated in 1985 (pre-accession) and 1987 (post-accession). Profit loss computed for an elasticity of subsitution of σ=3. We normalize the loss measure in regressions dividing it by by σ-1.

Table 2: Tariff Profile faced by Portuguese Exporters

1 2 3 4 5 6Dependent variable (ln):

Uncertainty Measure 3.952** 3.857** 3.975** 3.653** 4.223** 3.848**[1.716] [1.759] [1.731] [1.810] [1.783] [1.887]

Applied Tariff Change (ln) -2.719** -2.783** -2.655** -3.385*** -3.117** -2.847*[1.182] [1.177] [1.165] [1.131] [1.242] [1.604]

NTM Share Change (ln) -0.159[0.258]

Specific Tariff Share Change -0.434[1.162]

Applied Tariff SD Change 2.912[3.994]

Price Index Proxy Change (ln) 1.5 0.453[2.243] [3.730]

SD of Price Index Proxy Change 3.479[10.56]

Observations 731 731 731 731 731 731R-squared 0.48 0.48 0.48 0.48 0.48 0.48

Change in probability of policy reversal -0.24 -0.24 -0.25 -0.18 -0.23 -0.23Notes:All specifications include country and industry effects Robust standard errors in brackets. *** p<0.01, ** p<0.05, * p<0.1Sample: Spain and EC 10 countries, 1987-1985Parameters: For uncertainty measure and computing probability of reversal, σ=3, β=0.90

Table 3: Firm entry growth into EC-10 and Spain (by industry)

Change in Number of Firms

1 2 3 4 5 6Dependent variable (ln):

Uncertainty Measure 4.399** 4.301** 4.431** 4.351** 4.752** 4.415**[1.844] [1.884] [1.858] [1.914] [1.929] [2.151]

Applied Tariff Change (ln) -3.006** -3.072** -2.919** -3.113** -3.520*** -3.279*[1.301] [1.305] [1.288] [1.333] [1.328] [1.724]

NTM Share Change (ln) -0.166[0.269]

Specific Tariff Share Change -0.579[1.042]

Applied Tariff SD Change 0.468[4.243]

Price Index Proxy Change (ln) 1.946 1.006[2.276] [3.923]

SD of Price Index Proxy Change 3.121[11.13]

Observations 731 731 731 731 731 731Adj R2 0.379 0.379 0.379 0.378 0.378 0.378

-0.24 -0.24 -0.25 -0.23 -0.23 -0.23Notes:Robust standard errors in brackets. *** p<0.01, ** p<0.05, * p<0.1All specifications include country and industry effects Sample: Spain and EC 10 countries, 1987-1985Parameters: For uncertainty measure and computing probability of reversal, σ=3, β=0.90

Table 4: Firm-product growth into EC-10 and Spain (by industry)

Change in probability of policy reversal

Change in Number of Varieties (Firm*Product)

1 2 3 4 5Dependent variable (ln): Number of

FirmsExports per

FirmNumber of Varieties

Exports per Variety Exports

Uncertainty Measure -4.064** 2.435 -4.711** 3.082 -1.629[1.829] [4.841] [2.299] [4.394] [6.190]

Applied Tariff (ln) -1.426 -3.688 -1.331 -3.783 -5.113[1.952] [2.760] [2.412] [2.461] [4.351]

Observations 781 781 781 781 781Adj R2 0.848 0.499 0.842 0.486 0.677Notes:All specifications include country and industry effects Robust standard errors in brackets. *** p<0.01, ** p<0.05, * p<0.1Sample: Spain and EC 10 countries, 1985Parameters: For uncertainty measure, σ=3

Table 5: Pre-Agreement (1985) Intensive and Extensive Margins of Firms and Firm-Varieties

1 2 3Dependent variable (ln):

Uncertainty Measure 2.721** 3.952** 4.823**[1.209] [1.716] [2.051]

Applied Tariff Change -2.686** -2.719** -2.751**[1.184] [1.182] [1.180]

Observations 731 731 731Adj R2 0.389 0.389 0.39

σ 2 3 4Change in Probability of Policy Reversal -0.23 -0.24 -0.26Notes:All specifications include country and industry effects Robust standard errors in brackets. *** p<0.01, ** p<0.05, * p<0.1Sample: Spain and EC 10 countries, 1987-1985Parameters: For uncertainty measure and computing probability of reversal, σ=2, 3, 4 as indicated; β=0.90

Table 6: Firm entry growth into EC-10 and Spain (Robustness across σ)

Change in the Number of Firms