The Gravity Model which was introduced by Jan Tinbergen in 1962 (Tinbergen,1962), is a

foundational framework that predicts and analyzes trade flows between entities, typically
countries, using principles inspired by Newton's law of gravity. This model posits that trade
flows are positively correlated with the economic mass of the countries involved, typically
measured by GDP, and inversely related to the geographic distance between them. Tinbergen's
pioneering work laid the groundwork for its widespread adoption in empirical research, aiming
to quantify the effects of economic relationships on trade. Subsequent developments by
Bergstrand (1985,1989) further refined the model by incorporating the gravity equation with
concepts of monopolistic competition to elucidate bilateral trade patterns. It has been defined as
the “workhorse” of international trade and its ability to accurately estimate bilateral trade flows
has made it one of the most stable empirical relationships in economics (Leamer & Levinsohn,
1995). These enhancements have made the Gravity Model instrumental in understanding the
dynamics of international trade, evaluating the influence of trade policies, and examining
processes of economic integration, highlighting its enduring relevance and utility in the field of
economics.

Jan Tinbergen (Tinbergen,1962) applied an analogy with Newton's law of universal gravitation
to analyze the bilateral aggregate trade patterns between two countries i and j, describing that the
trade volume between them is directly proportional to the gross national products of the countries
and inversely proportional to the distance between them,

Ei E j
X ij = β (1)
Dij

Where:

 X ij is the total trade between country i and j,

 Ei ∧E j are the GDPs of country i and j, respectively,
 Dij denotes the distance between the two countries,
 β is a coefficient that reflects the degree to which trade values decrease due to distance.

However, despite the correlation between trade and the physical law of gravity, a contradiction
arises that no set of parameters can make equation (1) accurately represent a randomly observed
data set. Based on equation (1), the gravity model in trade has been developed to identify
coefficients for the variables of volume and bilateral distance derived from data to depict their
relationship. This allows for statistical inference on the relationship between trade flows and the
variables of volume and distance. The so-called "gravity equation" in international trade has been
demonstrated to be remarkably stable over time and across various country samples and
methodologies, standing as one of the most solid and stable empirical laws in economics.
Therefore, the gravity equation is formulated as follows:
a1 a2
X ij =a0 Ei E j D ij ε ij
a3
(2)
With a 1 , a2 , a3 are unknown parameters

Tinbergen (1962) presented the model in a log-log form, making the parameters elasticities of
trade flow with respect to explanatory variables. For equation (2), it is assumed that neighboring
countries experience stronger trade than what the distance variable alone might predict;
adjacency is indicated by a dummy variable N ij , taking a value of 1 for countries sharing a land
border. Additionally, the equation incorporates political factors: a dummy variable V ij indicates
preferential trade treatment under bilateral or multilateral agreements. The use of dummy
variables to assess the impact of Preferential Trade Agreements (PTA) has been a common
practice in research. More recently, an alternative approach specifying the proportion of
preferential treatment guaranteed by new agreements has been utilized, thus including a variable
V ij (i.i.d) in the equation.

ln X ij=a0 +a 1 ln Ei + a2 ln E j +a 3 ln Dij +a 4 N ij + a5 V ij +ε ij (3)

Over the past decades, the gravity model has always been applied and developed, incorporating
new theoretical frameworks into the model and giving rise to various schools of thought. The
first school of the gravity model was created under the assumption of "perfect competition".
Anderson J. (1979) posited the Constant Elasticity of Substitution (CES), where each country
produces and sells goods on the international market that are distinct from goods produced in any
other country. Goods are purchased from many sources because they are valued differently by
the end consumers. Another version, mathematically equivalent to the gravity model proposed by
Eaton & Kortum (2002), is based on homogenous goods from the demand side, known as the
"Iceberg Trade Costs" model.

The existence of bilateral trade flows with a value of zero holds significant implications for the
gravity equation because, in Newton's formula, the force of gravity can be very small but never
zero. Even though zeros may result from erroneous reporting and measurement, especially from
smaller and poorer nations, these observed zeros carry valuable information that needs to be
leveraged for effective estimation. Indeed, if zero observations stem from firms' decisions not to
sell goods in specific markets (or an inability to do so), the reality that trade between some
country pairs is actually zero could signal a problem with sampling methods (Chaney, 2008;
Helpman, Melitz, & Rubinstein, 2008). Thus, there is a necessity for appropriate econometric
techniques that can extract more information from the data, particularly regarding the role of
distance and other factors influencing the extent of global trade.

According to Anderson J. E. in "The Gravity Model" (2011), the gravity model stands out in its
capacity to analytically represent and manage economic interactions in a world comprising
multiple countries. This uniqueness stems from the model's modularity: the spatial distribution of
goods or factors is governed by gravitational forces, given that the scale of economic activities at
each location is accounted for. This modular characteristic facilitates segmentation at any desired
scale and allows for deductions about trade costs without relying on any specific production
models or the comprehensive market structure in a general equilibrium context.

Gravity model structures can be achieved by imposing two essential requirements, as identified
by Anderson & Van Wincoop (2004). The first requirement is that the transformation of total
demand must be consistent across countries, and the second is the application of the Constant
Elasticity of Substitution (CES). In essence, CES employs the principle of homotheticity
(ensuring that relative demand is solely a function of relative prices) and also separates
preferences. As mentioned, products are defined by location since goods are differentiated by
their place of origin: this segmentation structure is referred to as the "Armington assumption"
(Armington, 1969).

Building on this, the starting point for Anderson and Van Wincoop (2003) is a CES utility
function. If Xij represents the consumption of goods in region j imported from region i, then
consumers in region j must maximize the following utility function:
σ / ( σ −1)

(∑ β i X ij )
1/ σ ( σ −1 ) / σ
(4)
i

Subject to budget constraints

∑ pij X ij=E j (5)

i

With the given parameters:

 σ is the CES elasticity of substitution,

 β is a parameter with a positive distribution,
 Ej is the nominal income of the population in region j, and
 Pij is the price of goods from region i sold to consumers in region j.

To find the expenditure share for goods from region i by consumers in region j that satisfies the
condition of maximizing the utility function under budget constraints, we need to use both
equations (4) and (5). Specifically, the expenditure share can be calculated as follows:

( )
1−σ
X ij β i pi t ij
= (6)
Ej Pj

With pi representing the factory gate price of goods, and tij >1 as the trade cost coefficient
between the starting point i and the destination j. The parameters βi for goods shipped from i may
be either exogenous or, in the context of monopolistic competition applications, proportional to
the variety of products supplied by firms from country i. The CES index is determined using the
following formula:
 1/ ( 1−σ )
P j=(∑ ( βi pi t ij )1−σ ) (7)

It should be noted that the gravity equation is based on an expenditure function. This accounts
for two main factors. First, the Gross Domestic Product (GDP) of the importing country enters
the gravity equation (as Ej) because it reflects the income effect in the expenditure function.
Second, the distance between two countries is included in the gravity equation because it
represents the cost of bilateral trade, which is passed through the price of goods and thus
decreases bilateral trade, all else being equal. The most important takeaway from the
mathematical formulation above is that the expenditure function depends on relative prices rather
than absolute prices. This allows for the inclusion of payment adjustments in the competition of
firms in market j through the price index Pj. Therefore, equation (4) informs us that omitting the
price index Pj of the importing country from the original gravity equation described in equation
(3) leads to a misrepresentation of the function. It should also be noted that excluding dynamic
factors (lagged variables) may pose a problem.

Although the equation has omitted time suffixes for the sake of simplification, it should be noted
that Pj is a time-varying variable, and thus it will not be properly controlled if used as a fixed
variable over time, unless the researcher is computing with cross-sectional data (Benedictis &
Taglioni, 2011). After pointing out why the GDP of the importing country and the distance
between the two countries are included in the gravity equation, it is necessary to further explain
why the GDP of the exporting country should also be included. The proof by Anderson and
Wincoop is based on Armington's assumption about trade, which is that goods are differentiated
by country of origin. In other words, each country produces a unique product, so if there is a
change, the prices will reflect that. This implies that countries with larger GDPs export more of
their products to all other countries, as their goods are relatively cheaper. This means their goods
must be relatively cheap if they wish to sell all of their produced output under the assumption of
full employment of labor. Thus, countries with large GDPs must have relatively low prices to
sell all of their products (market clearing condition). To determine the price pi that will clear the
market, we sum the sales revenue of country i across all markets, including that country's own
market, and set it equal to the overall output. This can be written as follows:

E
Ei =∑ X ij = pi ∑ ( βi tij )1−σ P1−σj
1−σ
(8)
j j j

Solving the above equation for p1−σ

i , we get:

1−σ Ei
pi = (9)
Ωi

Ej
With Ωi=∑ ( β i t ij )
1−σ
1−σ (10)
j Pj
Where, Ωi represents the average market demand of all importers - measured in terms of trade
costs. It has been referred to in various ways in studies, such as Market Potential (Head &
Mayer, 2004; Helpman, Melitz, & Rubinstein, 2008), Market Openness (Anderson & Wincoop,
2003), etc. Substituting equation (10) into equation (6) yields a basic but correctly formed
gravity equation:

X ij 1−σ Ei
=( β i t ij ) 1−σ (11)
Ej Pi Ω i

Taking the logarithm of both sides, we get:

log X ịj =log ( E j ) +log ( Ei ) −log ( Ω i ) + ( 1−σ ) [ log ( β i ) +log ( t ij ) −log ( Pi ) ]

Thus, the GDP of the country of origin enters the gravity equation because large economies
provide goods that are either competitively priced or diverse in variety, or both. The equation
also indicates that the market potential of the exporting country is a critical issue, and the
difference between (11) and (6) is increasing due to the more apparent asymmetry between
countries (Benedictis & Taglioni, 2011).

Anderson and van Wincoop (2003) assume that Ωi=P 1−σ i under three important assumptions.
First, they assume that trade costs are symmetrically bilateral across all country pairs. However,
this assumption is inherently violated in the case of preferential trade agreements. Second, they
posit that trade is balanced, meaning Xij=Xji, which is also an assumption often violated in
reality. Lastly, they assume that there is only one period of data. If these three conditions are
1−σ
satisfied, the two variables Ωi and Pi can be empirically controlled by fixed effects that are
invariant over time per country. Equation (11) represents the general form of the gravity model.
For simplicity, the model can be rewritten in the following form from equation (11):

log X ịj =log ( E j ) +log ( Ei ) + ( 1−σ ) [ log ( β i ) +log ( t ij )−log ( Pi ) ]−log ( Ωi )

log X ịj =log ( E j ) +log ( Ei ) + ( 1−σ ) log ( t ij ) + (1−σ ) [ log ( β i )−log ( Pi ) ]−log ( Ωi )

GDPj GDPi ε ij

By adding the intercept and regression coefficients, we get:

log X ij =c+ b1 log GDP i+ b2 log GDP j +b 3 log t ij + ε ij (*)

Where:

 Xij is the export from country i to country j,

 E is the GDP of the country,
  t ij is the trade barrier between the two countries, often represented by the distance
between the two countries,
 ε ijis the random error.

The term "gravity" originates from the nonlinear form of equation (*) resembling Newton's law
of gravity: exports are directly proportional to the "mass" (GDP) of the exporting and importing
countries and inversely proportional to the distance between them (not the square of the distance,
unlike in physics). In other words, the gravity model suggests that we expect larger country pairs
to trade more, but countries farther apart are expected to trade less, possibly due to higher
transportation costs between them (Shepherd, 2013).