A. Universal gravity model

We rely on a trade model that falls within the universal gravity framework outlined by Allen et al. (Reference Allen, Arkolakis and Takahashi2020). This model enables us to analyze the trade effects of changes in religious composition, assuming that these changes affect bilateral trade costs. Similar cultural factors have also been treated as determinants of trade costs in the gravity literature (Melitz, Reference Melitz2008; Melitz and Toubal, Reference Melitz and Toubal2014, Reference Melitz and Toubal2019). In the model, we assume that there are $N$ countries, indexed by subscripts $i$ or $j$. Shipping wine from country $i$ to country $j$ involves incurring iceberg trade costs ${\tau _{ij}} \geq 1$. There is roundabout production, as in Eaton and Kortum (Reference Eaton and Kortum2002), that combines immobile labor, ${L_i}$, with intermediate input, ${M_i}$, in a Cobb-Douglas function with constant returns to scale. Thus, total production, ${Q_i}$, is given by:

(1)\begin{equation}{Q_i} = {\left( {{A_i}{L_i}} \right)^\zeta }M_i^{1 - \zeta },\end{equation}

where ${A_i} \gt 0$ is labor productivity, and $\zeta \in \left( {0,1} \right]$ is the share of labor in costs. ${M_i}$ is a constant elasticity of substitution (CES) aggregate of the differentiated varieties in all locations, and is assumed to be the same bundle of goods as those entering final consumption, so that the price index, ${P_i}$, for intermediates, is the price index taken over all goods.

Under perfect competition, the price of output in country i is given by:

(2)\begin{equation}{p_i} = \bar \kappa {\left( {\frac{{{w_i}}}{{{A_i}}}} \right)^\zeta }P_i^{1 - \zeta },\end{equation}

where $\bar \kappa $ is a constant and ${w_i}$ is the local wage rate. The value of total output is given by ${Y_i} \equiv {p_i}{Q_i}$. Since we assume that labor is the only factor of production and profits are zero, all output is distributed to workers, resulting in the identity ${Y_i} = {p_i}{Q_i} = {w_i}{L_i}$. This means that ${Y_i}{\text{ }}$ is simultaneously a measure of a country's value of production and total income.

The price of wine from country $i$ to country $j$ is expressed as:

(3)\begin{equation}{p_{ij}} = {p_i}{\tau _{ij}}.\end{equation}

The quantity of wine reaching country $j$ after subtracting iceberg trade costs, is denoted by ${q_{ij}}$. The expenditure on wine from country $i$ in country $j$ is given by:

(4)\begin{equation}{X_{ij}} = {p_{ij}}{q_{ij}}.\end{equation}

In equilibrium, markets clear, which means that prices and quantities adjust so that total output at every origin country equals aggregate demand from all locations, accounting for iceberg trade costs, so that ${Q_i} = \mathop \sum \limits_{j = 1}^N {q_{ij}}{\tau _{ij}}$.

Each country $i$ has a representative consumer who values wine from different countries based on CES preferences, as in Armington (Reference Armington1969), Anderson (Reference Anderson1979), and Anderson and Van Wincoop (Reference Anderson and Van Wincoop2003). Wine is differentiated by country of origin, following the Armington (Reference Armington1969) assumption. The consumer supplies labor inelastically and earns all income from it. The consumer's optimization problem leads to a demand equation for wine from country $i$ in country $j,\;{X_{ij}}$, which is given by:

(5)\begin{equation}{X_{ij}} = \frac{{{{\left( {{p_i}{\tau _{ij}}} \right)}^{ - \theta }}}}{{P_j^{ - \theta }}}{E_j},\end{equation}

where $\textstyle\theta$ is the trade elasticity with respect to trade costs, which is a parameter governing how trade flows react to changes in trade costs.Footnote ¹ $P_j^{ - \theta } = \mathop \sum \limits_{k = 1}^N {\left( {{p_k}{\tau _{kj}}} \right)^{ - \theta }}$ is the importer multilateral resistance term and ${E_j} = \mathop \sum \limits_i {X_{ij}}$ is total expenditure.

We assume that trade deficits are exogenous. Expenditure is a multiple of output value, such that ${E_i} = {\Xi }{\xi _i}{Y_i},$ where $\Xi = \displaystyle{\sum_i}Y_i/\displaystyle{\sum_i}{\xi_i}Y_i$ is an endogenous scalar that ensures that the sum of all trade deficits equals zero. ${\xi _i}$ is the exogenous expenditure-output ratio. Finally, we set the numeraire $\mathop \sum \limits_i {Y_i} = \bar Y \gt 0$.

B. Partial trade effects

We operationalize the universal gravity model to estimate how changes in religiosity and the religious composition of trading partners affect bilateral trade flows by assuming that bilateral trade costs, ${\tau _{ij}}$, evolve in time according to:

(6)\begin{equation}{\tau _{ij,t}} = \exp \left( {\ln \left( {{{\text{t}}_{ij,t}} + 1} \right) + {\beta _2}PT{A_{ij,t}} + {\beta _3}CST{U_{ij,t}} + {\text{RE}}{{\text{L}}_{ij,t}}{\gamma } + Inte{r_{ij,t}} + {\lambda _{ij}}} \right),\end{equation}

where ${{\text{t}}_{ij,t}}$ is the applied tariff that country $j{\text{ }}$collects from country $i$ in year $t$. $PT{A_{ij,t}}$ is an indicator that $i$ and $j$ have a preferential trade agreement (PTA) in year $t$, while $CST{U_{ij,t}}$ is an indicator for when both countries are part of the same customs union in that year. $Inte{r_{ij,t}}$ is a set of time-varying international border dummies, which help capture globalization trends (Bergstrand et al., Reference Bergstrand, Larch and Yotov2015). ${\text{ }}{\lambda _{ij}}{\text{ }}$is a country pair fixed-effect that captures unobserved time-invariant components of trade costs. ${\text{RE}}{{\text{L}}_{ij,t}}$ is a row vector that contains variables related to the religious similarities of the exporter and importer, and ${\gamma }$ is the column vector of estimated partial trade effects of these variables on wine trade flows. The elements in ${\text{RE}}{{\text{L}}_{ij,t}}$ are as follows.

First, we define cross-country differences in religiosity to analyze the impact of the differences in the strength of religious belief. Religiosity is measured as in Doan and Mai (Reference Doan and Mai2025), who use three questions from the WVS to construct three measures: the share of respondents affiliated with a religion, the average reported importance of religion, and the average frequency of religious service attendance. $Religiosit{y_{ij,t}}$ is the simple average of these three measures. We then compute the absolute difference in religiosity between exporter and importer, $\Delta Rel_{ij,t}$, as:

(7)\begin{equation}{\Delta }Re{l_{ij,t}} = \left| {Religiosit{y_{i,t}} - Religiosit{y_{j,t}}} \right|.\end{equation}

Second, we examine how shared religious composition between exporter and importer influences wine trade. Specifically, we calculate the share of respondents in the WVS who report belonging to one of nine major religious groups $r$: Roman Catholic, Protestant, Orthodox (e.g., Russian or Greek), Jewish, Muslim, Hindu, Buddhist, Other Christian (e.g., Evangelical, Pentecostal), and Others. Then, we calculate the common share of respondents belonging to each religious denomination as:

(8)\begin{equation}ShareRel_{ij,t}^r = \min \left( {share_{i,t}^r,{\text{ }}share_{j,t}^r} \right),\end{equation}

where $share_{i,t}^re$ is the share of respondents who reported being of the $r$ denomination in year $t$. We set $ShareRel_{ij,t}^r = 0$ for $i = j$ to isolate the partial effects on international trade flows (Larch et al., Reference Larch, Monteiro, Piermartini and Yotov2025a).

We estimate trade flows, ${X_{ij,t}}$, using the Poisson Pseudo Maximum Likelihood (Poisson PML) estimator proposed by Santos Silva and Tenreyro (Reference Santos Silva and Tenreyro2006). This estimator is widely used in the gravity literature because it addresses heteroskedasticity and can handle zero trade flows (Yotov et al., Reference Yotov, Piermartini and Larch2016). Wine trade flows are specified as:

(9)\begin{align}{X_{ij,t}} &= \exp \left( {\alpha _0} + {\alpha _1}\ln \left( {{{\text{t}}_{ij,t}} + 1} \right) + {\alpha _2}PT{A_{ij,t}} + {\alpha _3}CST{U_{ij,t}} \right.\cr &\left. \quad +\, {\text{RE}}{{\text{L}}_{ij,t}}{\delta } + In{t_{ij,t}} + {\mu _{j,t}} + {\mu _{i,t}} + {\mu _{ij}} \right) + {\varepsilon _{ij,t}},\end{align}

where ${\alpha _0}$ is an intercept term, ${\alpha _1} = - \theta $ is the trade elasticity with respect to trade costs, and ${\alpha _2} = - \theta {\beta _2}$ and ${\alpha _2} = - \theta {\beta _2}$ are the trade-level effects with respect to $PT{A_{ij,t}}$ and $CST{U_{ij,t}}$, respectively. $In{t_{ij,t}} = - \theta Inte{r_{ij,t}}$ are international border-year fixed effects that account for globalization effects. ${\mu _{j,t}}$ is an importer-time fixed effect to account for inward (importer) multilateral resistances. ${\mu _{i,t}}$ is an exporter-time fixed effect to account for outward (exporter) multilateral resistances.Footnote ² These fixed effects also absorb income-related variation, which may otherwise confound the estimated impact of religion (Bryan et al., Reference Bryan, Choi and Karlan2021). ${\mu _{ij}} = - \theta {\lambda _{ij}}$ are country-pair fixed effects to account for unobserved time-invariant trade costs. ${\varepsilon _{ij,t}}$ is an additive error term, which we cluster at the importer-exporter level to control for any remaining country-pair correlations. The column vector ${\delta } = - \theta {\gamma }$ contains trade elasticities with respect to ${\text{RE}}{{\text{L}}_{ij,t}}$.

To better understand how changes in shared religious composition affect trade costs, we first estimate Equation (9) using $ShareRel_{ij,t}^r$ for all nine major religious groups. The reference category in this regression is the common share of respondents who report not belonging to any religion. In this first step, we identify which religious group has the largest effect on ${X_{ij,t}}$. We then run a second regression using only the common share of that specific religion, treating the combined share of all other religions as the base group. This approach addresses potential multicollinearity concerns.

C. GE trade effects

Trade elasticities contained in ${\delta }$ capture the direct effect of changes in religious composition on trade flows, holding prices, income, and expenditures constant. However, changes in trade costs driven by religious composition can also affect these variables. GE effects account for these adjustments, allowing us to capture the full impact of religious composition on trade flows. We estimate GE effects using the “hat algebra” notation introduced by Dekle et al. (Reference Dekle, Eaton and Kortum2007). For each variable, its hat version is the ratio of the value of the variable in a counterfactual equilibrium relative to the value in a baseline equilibrium. In this way, if $x$ is the value in the baseline observed equilibrium, $x'$ is the value in the counterfactual equilibrium, so that $\hat x = x'/x$. Changes in religious composition are leading to counterfactual trade costs, $\tau _{ij,t}^{'}$, so that:

(10)\begin{equation}\hat \tau _{ij,t}^{ - \theta } = \frac{{\tau _{ij,t}^{' - \theta }}}{{\tau _{ij,t}^{ - \theta }}} = \frac{{\exp \left( {{\text{REL}}{'_{ij,t}}{\delta }} \right)}}{{\exp \left( {{\text{RE}}{{\text{L}}_{ij,t}}{\delta }} \right)}} = {\text{exp}}\left( {\left( {{\text{REL}}{'_{ij,t}} - {\text{RE}}{{\text{L}}_{ij,t}}} \right){\delta }} \right).\end{equation}

We examine two scenarios, with the baseline set to the last year in our sample, 2023. In Scenario 1, all global wine tariffs are set to zero.Footnote ³ This scenario provides a benchmark for trade cost reductions. In Scenario 2, all trading partners share the same religion, which is associated with the highest increase in bilateral wine trade. This scenario captures trade diversions if countries' religious composition changes. While these scenarios are extreme, they illustrate the potential magnitude of religion's impact on trade costs.

The trade model is reduced to a system of equations for ${\hat p_i}$, ${\hat P_i}$ and ${\hat \Xi }$ as described in Campos et al. (Reference Campos, Reggio and Timini2024). Once this system is solved, after defining $\psi = \left( {1 - \zeta } \right)/\zeta $ and ${c_i} = {A_i}{L_i}$, the comparative statics are calculated as:

(11)\begin{equation}{{{\mathbf{GE}}{\text{ }}{\mathbf{Impact}}{\text{ }}{\mathbf{on}}{\text{ }}{\mathbf{Trade}}{\text{ }}{\mathbf{Flows}}:{\text{ }{\widehat X}}}_{ij}} = \widehat \tau _{ij}^{ - \theta }\widehat p_i^{ - \theta }\widehat P_j^\theta {\widehat E_j}\end{equation}

\begin{equation*}{\mathbf{GE}}{\text{ }}{\mathbf{Impact}}{\text{ }}{\mathbf{on}}{\text{ }}{\mathbf{Total}}{\text{ }}{\mathbf{Output}}{\text{ }}{\mathbf{Value}}:\;{\widehat Y_i} = {\widehat p_i}{\hat Q_i} = {\widehat c_i}\frac{{\widehat p_i^{1 + \psi }}}{{\widehat P_i^\psi }}\end{equation*}

\begin{equation*}{\mathbf{GE}}{\text{ }}{\mathbf{Impact}}{\text{ }}{\mathbf{on}}{\text{ }}{\mathbf{Welfare}}:\;{\widehat W_i} = {\hat \Xi }{\widehat \xi _i}\frac{{{{\widehat w}_i}}}{{{{\widehat P}_i}}} = {\hat \Xi }{\hat \xi _i}\frac{{{{\widehat c}_i}}}{{{{\widehat L}_i}}}{\left( {\frac{{{{\hat p}_i}}}{{{{\widehat P}_i}}}} \right)^{1 + \psi }}\end{equation*}

We calculate these changes using the iterative algorithm developed by Campos et al. (Reference Campos, Reggio and Timini2024). It builds on Baier et al. (Reference Baier, Yotov and Zylkin2019) to solve for GE price adjustments.

D. Data

We obtain bilateral trade data from an updated version of the International Trade and Production Database for Simulation developed by Borchert et al. (Reference Borchert, Larch, Shikher and Yotov2024). The original database covers 140 industries, with data from 1988 to 2019. Using the same methodology, we extend the database for the wine sector through 2023. An important feature of the database is that it includes domestic trade flows, which are essential for GE analysis. We impute missing domestic trade data following the approach in Borchert et al. (Reference Borchert, Larch, Shikher and Yotov2024). Still, these imputed values are used only in the GE simulations, not in elasticity estimation (Larch et al., Reference Larch, Shikher and Yotov2025b). Country-level religion data is obtained from the WVS (Haerpfer et al., Reference Haerpfer, Inglehart, Moreno, Welzel, Kizilova, Diez-Medrano, Lagos, Norris, Ponarin and Puranen2022). This is a global project that has conducted nationally representative surveys in more than 100 countries annually since 1981. The WVS has also been used in related work, such as Melitz and Toubal (Reference Melitz and Toubal2019). We keep responses to the following WVS questions: A006 (Importance in life: Religion), A065 (Member: belongs to religious organization), F025 (Religious denominations: major groups), and F028 (How often do you attend religious services). For questions with scaled or binary responses, we calculate the simple average of non-missing answers.Footnote ⁴ For F028, we calculate the share of respondents who identify with each major religion.Footnote ⁵

Industry-average tariff rates, in ad valorem equivalent terms, are calculated using data from the WITS TRAINS database (UNCTAD, 2024).Footnote ⁶ Data on PTAs and customs unions is from the USITC Dynamic Gravity Dataset and the World Trade Organization (2024) RTA Database.Footnote ⁷ Similar to Campos et al. (Reference Campos, Estefania-Flores, Furceri and Timini2023), we estimate the supply elasticity for the wine industry using the share of intermediate inputs in total gross output, as reported in the EUKLEMS & INTANProd database by Bontadini et al. (Reference Bontadini, Corrado, Haskel, Iommi and Jona-Lasinio2023).Footnote ⁸ We calculate input shares by country, industry, and year, and restrict the sample to values between the 10th and 90th percentiles for each industry-year combination. Using this trimmed data, we compute an average supply elasticity of 2.81 for the wine industry, which we use for the GE analysis.Footnote ⁹ After merging all datasets, our final dataset covers 102 countries from 1988 to 2023.

We report the descriptive statistics in Table 1. The average annual bilateral export flow is $2.11 million, with a maximum of $2.78 billion. The average ad-valorem equivalent wine tariff is 31.6%. However, a few countries, such as Egypt, imposed extremely high tariffs, in some cases exceeding 2,000%. On average, 27% of importer-exporter pairs share a PTA, and 4% belong to the same customs union. The average difference in religiosity is 0.8, with the average declining between 1988 and 2023. The highest average common shares across countries are found for Roman Catholics, Muslims, and Protestants. However, all three groups show a decline over the sample period.

Table 1.

Descriptive statistics

Note. The table shows the descriptive statistics. We exclude domestic flows from the calculations. Δ1988/2023 is the difference in the mean value between 1988 and 2023. $\Delta Rel$ represents differences in religiosity. PTA indicates whether the trading partners have a preferential trade agreement, while CSTU shows whether they are part of the same customs union. Ad-valorem tariffs and shares of common religion are reported in percentages. In the regression analysis, we use the shares, $ShareRel_{ij,t}^r,$ without multiplying them by 100.