A previous version of this paper circulated with the title “Food prices and the multiplier effect of export policy”. We would like to thank Robert Staiger and two anonymous referees for in-depth comments. We are also grateful to Willy Alfaro, Simon Evenett, Gabriel Felbermayr, Caroline Freund, Robert Gulotty, Lee Ann Jackson, Nuno Limao, Marcelo Olarreaga, Gianluca Orefice, Roberta Piermartini, Frederic Robert-Nicoud and seminar participants at Johns Hopkins (SAIS), WTO, ETSG, LUISS, University of Munich, University of Frankfurt, and the IMF/WB/WTO Trade Workshop for helpful discussions. Joelle Latina provided excellent research assistance. Remaining errors are our responsibility. Michele Ruta gratefully acknowledges hospitality at the International Trade Department of CES-ifo during the early stages of this research project. Disclaimer: The opinions expressed in this paper should be attributed to the authors. They are not meant to represent the positions or opinions of the IMF or the WTO and their Members.

A partial list includes Anderson and Martin (2011), Chaffour (2008), Bouet and Laborde (2012), Hochman et al. (2010), Headey (2011).

See also the discussion in Ivanic et al. (2011).

Freund and Ozden (2008) and Tovar (2009) have introduced loss aversion in a model of trade policy and find that this behavioral extension explains several features of the observed pattern of trade protectionism. Classic works on reference-dependent utility include Kahneman and Tversky (1979), Tversky and Kahneman (1991), and Koszegi and Rabin (2006). Moreover, Tversky and Kahneman (1981), Samuelson and Zeckhauser (1988), and Camerer (1995), among others, provide experimental evidence supporting this preference structure.

As confirmed by the evidence provided in this paper and in the studies discussed in the Introduction, governments do indeed actively use trade policy in food sectors. An important question, not addressed in this paper, is why governments use more distorsive policies, such as trade policy, rather than more efficient tools, such as domestic measures. One reason may have to do with the availability of appropriate domestic policies, particularly in developing countries. Limao and Tovar (2011) provide an alternative explanation based on a political economy argument.

To be clear, this model introduces a different feature relative to Freund and Ozden (2008) and Tovar (2009). These authors abstract from the effect that changes in prices have on consumer surplus and government revenue and focus instead on the direct effect that these price changes have on the income of factor owners. This is a reasonable assumption in their models where there are many consumption goods, none of which is supposed to represent a large share of total consumption. The structure presented here is instead better suited to capture the fact that the poor, particularly in developing countries, spend an important portion of their income on food. For example, the poorest decile of the population in Nigeria, Vietnam and Indonesia spend respectively 70, 75, and 50 per cent of their income on food (Ivanic et al., 2011).

The proofs of this and the following statements, which form the basis of Proposition 1 below, are provided in footnote as they are generalizations of results established in previous literature. The proofs of our novel results (Propositions 2, 3 and 4) are in appendix. When H(·) = 0, total welfare reduces to the standard form: G(p) = W(p) = (1 — α) + π(p) + CS(p) + GR(p). The optimal domestic price is determined by the first order condition ∂W/∂p = (p* — p) [y^ʹ — d^ʹ] = 0, which is satisfied for p = p*.

In this case, the FOC is given by $$\frac{\partial G}{\partial p}=W^{\prime }+H^{\prime }=(p*-p)x^{\prime }+(1-\alpha )[-y+(p*-p)x^{\prime }]\cdot h^{\prime }=0.$$ The proof of this statement follows the same steps as in footnote 7. Specifically, for an international price of food $p*=\overline{p}+\epsilon$, with ε small enough, $W^{\prime }(\overline{p}+\epsilon )=0$, while $H^{\prime }(\overline{p}+\epsilon )>0$, which implies $G^{\prime }(\overline{p}+\epsilon )>0$. In this case, the optimal domestic price of food is a corner solution, which equals the reservation price for workers $\overline{p}$. As £ increases, the loss aversion effect weakens, while $W^{\prime }\left(\overline{p}\right)$ becomes more negative. There is a critical level of the world price $p*={\overline{p}}^c>\overline{p}$, above which the optimal domestic price becomes an interior solution.

Anecdotal evidence confirms that export policy in food sectors is often designed to stabilize domestic prices and avoid losses for specific groups (see Piermartini, 2004). For instance, Papua New Guinea had in place an export tax/subsidy rate for cocoa, coffee, copra, and palm oil equal to one-half the difference between a reference price (calculated as the average of the world price in the previous ten years) and the actual price of the year.

The proofs of these statements are omitted as they follow directly from Freund and Ozden (2008).

We only sketch the argument here. Define the political welfare of government as J = Ω + H, where H is defined as in Subsection 2.2, while Ω = bπ + CS + GR, with b > 1 representing the political bias. This can be interpreted as the reduced-form of a two stage lobbying game, as in Grossman and Helpman (1994). Denote by t (p*) the politically optimal subsidy/tariff when the domestic price is intermediate, that is, when $p=p*+t(p*)\in [\underset{\_}{p},\overline{p}]$. This domestic price is the one for which G^ʹ = Ω^ʹ = 0. When -as a result of an increase in the world price p*- the domestic price grows slightly above the upper threshold, say when $p=\overline{p}+\epsilon$, we have that G^ʹ = Ω^ʹ + H^ʹ > 0, as Ω^ʹ = 0 but H^ʹ > 0. For a range of ε small enough, the solution to this maximization problem is a corner solution in which the government utilizes trade policy to keep the domestic price constant at $p=\overline{p}$. This policy consists of gradually reducing the subsidy/tariff t(p*) as p* increases. When p* grows higher than $\overline{p}$, the optimal policy becomes an export tax / import subsidy.

The structure of the trading countries is assumed identical. What makes a fraction of them become food exporters, rather than importers, may for instance be that they are endowed with a greater amount of land, L.

Countries may also differ in their mode of intervention. For instance, developed countries generally have a larger set of policy instruments and are, therefore, less likely to use second-best trade policy to fulill their welfare objectives.

Export subsidies and import tariffs, which instead depress the world price of food, can be discussed in a similar way.

The assumed symmetric structure of the world economy ensures that the policy of compensating protectionism is identical across countries belonging to regions E, I, that it, t (e) = t (i) = t ∀e ϵ E, i ϵ I. Moreover, function p^*(t) is depicted as a linear function for illustrative purposes only (see Example 1).

The proof of the uniqueness of this equilibrium is a straightforward geometric implication of the two following general properties of the model: (i) dp^*/dt ϵ (0, 1), (ii) dt/dp^* = 1 along the regions of compensating protectionism.

The logic of the result also applies to export subsidies (import tariffs). Low prices induce governments to offer subsidies (tariffs) to compensate producers. However, as all exporters (importers) face similar incentives and enact the export promotion (import restriction) policy at the same time, the effect is to increase the export supply (reduce the import demand) of food in world markets and further depress prices. The multiplier effect of trade policy determines, in this case, an equilibrium of high export subsidies and tariffs and low food prices relative to free trade.

In the limit case where E, I = [0, 1], along the regions of compensating protectionism it is dp^*/dt = 1, and the multiplier is infinite: θ → ∞. This result is also proven in the Proof of Proposition 2.

The proof of this proposition is slightly more complicated than the one of Proposition 1, the main reason being that, differently from the small country case, here the terms of trade effect breaks the symmetry between importing and exporting countries (as ${\hat{t}}_E>0,{\hat{t}}_I<0$). As a result, the regions of compensating protectionism do not exactly coincide between countries in E and those in I. In the proof of this proposition, we identify the interval for which the policy of full compensation is optimal for both sets of countries simultaneously.

Again, in the limit case where E = N and I = M, the multiplier is infinite along the regions of compensating protectionism. Details are in the Proof of Proposition 4.

Despite the strategic complementarity, the uniqueness of equilibrium in this policy game is ensured by the fact that the slope of the best-response functions is always strictly lower than 1.

An important question is whether the multiplier effect implies a need for a trade agreement. In the present model, (i) if all countries are small, there is no need for coordination, as the unilaterally optimal policy is also globally optimal; (ii) if countries are large, coordination allows trading countries to internalize the terms-of-trade effect and thus enhances their welfare, as in the standard literature on trade agreements (Bagwell and Staiger, 2002).

The maximization of (10) gives rise to an export tax $\stackrel{⌣}{t}$, which is strictly higher than $\hat{t}$. This allows us to find the value of the free trade world price such that the domestic price resulting from the equilibrium policy be equal to $\overline{p}$, that is to say, that value of ${\overline{p}}_{ft}^c$ such that $p*(\stackrel{⌣}{t},{\overline{p}}_{ft}^c)-\stackrel{⌣}{t}=\overline{p}$. To find the explicit value for ${\overline{p}}_{ft}^c$ as a function of all the parameters of the model however, we would further need to specify a functional form for both h (·) and for the utility function u (·).

A similar exercise could, in principle, be done with export promotion and import restriction for periods where international food prices were historically low and presented a downward trend, such as in the second half of the 1980s.

Data on the state of trade measures are available at http://www.globaltradealert.org/site-statistics

Unfortunately, only in very few instances countries report the size of the measures implemented. With the available information, we are therefore only able to perform a qualitative analysis of the likelihood of the utilization of such measures, rather than a quantification of the determinants of their magnitude.

This figure is likely to underestimate the effective number of trade policy measures that have been implemented. In addition to the lack of information on subsidies discussed above, the reason is that trade measures recorded in each month often include more than a single policy measure. Our data set, however, does not allow us to precisely discern this information.

The data are publicly available on the web at http://precip.gsfc.nasa.gov/

ElectionGuide database is provided by the International Foundation for Electoral Systems (IFES) and is available online at http://www.electionguide.org/

In other words, assuming that a certain country implements a policy measure at time t up to time t + 5, the variable will be equal to 0 before time t, equal to 1 between t and t + 5, and will become 0 again for t > t + 5.

Reference prices have also been calculated using five and one year moving averages and the historical averages between 1990 and 2000 and 1990 and 2006. The results are in line with the ones presented in this section.

To avoid that the size of an exporter/importer of a certain product in the international markets is polluted by trade policy measures adopted by countries during periods of crisis, average export and import shares are calculated using data from 2000–2005. In addition, for countries implementing both import and export policy measures in a certain product and period, we use the maximum share between imports and exports as weights.

In the logit model country fixed effects are excluded from the regression. This is done to avoid that the control group of countries that have never implemented a trade measure over the time period considered is excluded from the regression due to perfect identification.

To save space, in the table we present the estimation results excluding country fixed effects. Results, available under request, still hold.

The table presents the estimation results for the linear probability model. Results for the logit model are qualitatively similar.

Alternative instruments have been created considering the top 5 and the top 10 producers of a product. The main results still hold and do not change significantly.

This instrument is computed as the total count of top producers for which the amount of rain in a certain month has been either above the 75th percentile or below the 25th percentile compared to its historical mean.

The relationship between elections and trade policy could be different from the one described by standard lobbying models for autocratic countries. However, based on the democracy scores from the Polity dataset, none of the large importers and exporters experiencing elections between 2008 and 2011 is an autocratic regime (with a democracy score <6 of the Polity IV index).

Note that here the number of observations for the baseline regression are smaller than in Table 4 due to the lack of elections data for all the countries in the original sample.

Our discussion is in line with Miguel et al. (2004) that highlight the problem of mismeasured explanatory variables, such as per capita GDP, in an instrumental variables approach.

Note that here, differently from previous subsections, the global trade policy utilization variable includes all countries in the sample. Formally, the variable is defined as follows: $GTP{U}_{k,t}={\mathrm{\Sigma }}_j(\frac{{exp}_{j,k,t-1}}{WldTrad{e}_{k,t-1}}exp̲re{s}_{j,k,t-1}+\frac{{imp}_{j,k,t-1}}{WldTrad{e}_{k,t-1}}imp̲re{d}_{j,k,t-1})$

The proof for the multiplier effect of export subsidies / import tariffs, when $p^{*}\in ({\underset{\_}{p}}^c,\underset{\_}{p})$, is totally analogous and is thus omitted.

Fact (1) can be proven as follows. Policy ${\hat{\text{t}}}_I$ is the one for which ${W^{\prime }}_I({\hat{\text{t}}}_I,{\text{t}}_{\text{E}})=0$. Moreover, for any ${\text{t}}_I>{\hat{\text{t}}}_I({\text{t}}_I<{\hat{\text{t}}}_I)$, it is ${W^{\prime }}_I({\hat{\text{t}}}_I,{\text{t}}_{\text{E}})<0\left({W^{\prime }}_I({\hat{\text{t}}}_I,{\text{t}}_{\text{E}})>0\right)$. Policy ${\stackrel{⌣}{t}}_I$ is instead the one such that $W_I^{\prime }({\stackrel{⌣}{t}}_I,{\text{t}}_{\text{E}})+H_I^1({\stackrel{⌣}{t}}_I,{\text{t}}_{\text{E}})=0$. Given that, where it is defined, ${H^{\prime }}_I(\cdot )>0$, it must be $W_I^{\prime }({\stackrel{⌣}{t}}_I,{\text{t}}_{\text{E}})<0$, which only occurs for policy values strictly higher than ${\hat{\text{t}}}_I$. To prove Fact (2), first notice that ${dp}_{\omega }/d{t}_{\omega }=d{p}^{*}/d{t}_{\omega }-1$. Then, it suffices to prove that dp^*/dt_ω is lower than 1 ∀_ω ϵ E ∪ I, which is done in the proof of Proposition 4 below. Fact (3) is true by definition of p^* and $p_{ft}^{*}$.