The key question that a researcher is faced with when asked to assess the effects of a given policy
measure is deciding which methodological approach is best suited to answer the question given
existing constraints. At this stage, dialogue between researchers and policy stakeholders is crucial
as, depending on the circumstances, researchers may help policy-makers to determine relevant
questions and to guide the choice of appropriate methodologies.
The choice of a methodology is not necessarily straightforward. It involves choosing between
descriptive statistics and modelling approaches, between econometric estimation and simulation,
between ex ante and ex post approaches, between partial and general equilibrium. Ex ante simulation
involves projecting the effects of a policy change onto a set of economic variables of interest, while
ex post approaches use historical data to conduct an analysis of the effects of past trade policy. The
ex ante approach is typically used to answer “what if” questions. Ex-post approaches, however, can
also answer “what if” questions under the assumption that past relations continue to be relevant.
Indeed, this assumption underlies approaches that use estimated parameters for simulation. Partial
equilibrium analysis focuses on one or multiple specific markets or products, ignoring the link
between factor incomes and expenditures, while general equilibrium explicitly accounts for all the
links between sectors of an economy – households, firms, governments and the rest of the world.
In econometric models, parameter values are estimated using statistical techniques and they come
with confidence intervals. In simulation models, behavioural parameters are typically drawn from a
variety of sources, while other parameters are chosen so that the model is able to reproduce exactly
the data of a reference year (calibration).
In principle, the question should dictate the choice of a methodology. For example, computable
general equilibrium (CGE) seems to be the most appropriate methodology for an ex ante
assessment of the effect of proposals tabled as part of multilateral market access negotiations. In
reality, however, the choice is subject to various constraints. First, methodologies differ significantly
with regard to the time and resources they require. Typically, building a CGE model takes a long
time and requires a considerable amount of data. Running regressions require sufficient time series
or cross sections of data, while the calibration of a partial equilibrium model only requires data for
one year. There are, however, relatively important sunk costs and thus large economies of scale
and/or scope. Once a CGE has been constructed, it can be used to answer various questions
without much additional cost. More generally, familiarity with certain methodologies or institutional
constraints could dictate the use of certain approaches.
Methodologies can also be combined to answer a given question. In most cases, it is sound advice
to start with descriptive statistics, which, besides paving the way for more sophisticated analysis,
often go a long way towards answering questions that one might have on the effects of trade
INTRODUCTION
9
policies. Similarly, when assessing the distributional effects of trade policy, it can be useful to
combine approaches. The effect of changes in tariffs on prices is estimated econometrically, while
the effect of the price changes on household incomes is simulated.
Different methodologies or simply different assumptions may lead to conflicting results. This is not
a problem as long as differences can be traced back to their causes. The difficulty, however, is that
policy-makers do not like conflicting results. This leads us to another important point, which is the
importance of the packaging of results. Presenting and explaining results in a clear and articulate
way, avoiding jargon as much as possible, is at least as important as obtaining those results. It is
also crucial to spell out clearly the assumptions underlying the approach used and to explain how
they affect the results.