In developing countries, where food expenditure makes up a large proportion of household consumption, these price fluctuations have led to a proliferation of policies to control or stabilize food prices. Temporary export restrictions have been particularly widespread, with at least 33 countries using some form of export restriction during the 2007 – 2008 food price spike and its aftermath, including 4 of the top 5 rice producers and 7 of the top 13 wheat producers . This chapter focuses on the most common and severe of such restrictions: the short-term export ban. The literature on export restrictions has focused on understanding why countries implement them and the role they play in exacerbating international price spikes. Theoretically, export restrictions introduce welfare-reducing price distortions, with local farmers losing more than local consumers gain from lower domestic prices . Governments likely implement export restrictions because they put more weight on consumers’ interests than those of producers, are more concerned about negative deviations from the status quo than positive ones, or seek to avoid extreme events .
Gouel and Jean have also shown that export restrictions can be part of an optimal dynamic food price stabilization policy when consumers are risk averse and insurance markets are incomplete. Regardless of their domestic rationale, the welfare effects on other countries appear to be unambiguously negative: by cutting off supply to the world market during times of high prices,stacking pots export restrictions magnify international price fluctuations and have been criticized for representing a beggar-thy-neighbor approach to trade . This chapter provides new empirical evidence from East and Southern Africa that export bans do not always have the effects that governments think they do. Unlike other parts of the world where export restrictions were one-time policies implemented during the 2007 – 2008 food price spike , export bans in East and Southern Africa are regularly used to respond to high international prices or domestic production shortfalls of maize, the main staple grain in the region. I use monthly, market-level maize price data from 49 large hub markets in 12 countries over a 10-year period during which 5 of these countries implemented 13 distinct export bans on maize. I document a surprising and robust empirical result: export bans in this region do not have a statistically significant effect on the gaps in prices between pairs of affected cross-border markets. I compare my empirical results to results from simulations using the estimated dynamic monthly model of grain storage and trade in sub-Saharan Africa from the previous chapter, which includes nearly all of the same markets and cross-border trade routes.
The model predicts a large and statistically significant increase in the gaps in prices between the affected cross-border markets due to the 13 export bans, even when traders are able to anticipate the bans with perfect foresight. The absence of an effect on the price gaps in the data matches a model simulation in which the export bans are not implemented. However, prices in both implementing and trading partner countries during export bans are significantly higher in the data than in the model simulation with no implementation. Information collected from market participants in sub-Saharan Africa indicates that export bans are imperfectly enforced, with informal local traders as well as some formal traders who are able to secure export permits through back-door channels able to continue trading during bans. These alternative trade channels may be subject to capacity constraints, but these constraints appear to only bind at the very end of bans. However, the unpredictable, ad hoc nature of the bans and their enforcement appears to destabilize markets on both sides of the border. In addition to prices that are higher than they would have been without a ban, price volatility is also significantly higher in the implementing country.
Taken together, my results suggest that export bans in East and Southern Africa do not have their intended effects of stabilizing or lowering domestic prices or insulating them from high international prices and have unintended destabilizing consequences instead. Policymakers in the region should therefore re-evaluate their use even when they appear justified on political economy grounds. My results are also a cautionary note for studies that have used model-based simulations to estimate the effects of export restrictions , as these effects are likely different in practice if the export restrictions in question are imperfectly enforced.My primary dataset consists of a panel of monthly maize price data from large hub markets in East and Southern Africa assembled by the Famine Early Warning System Network and covering the 10-year period from January 2002 to December 2011. Using local newspaper archives and FEWS NET monitoring reports, I identified the starting and ending dates of 13 short-term export bans implemented by 5 countries during this period, ranging in duration from 4 to 54 months . I then selected the major markets on either side of the affected international borders from the FEWS NET database and identified the pairs of cross-border markets directly linked by transportation infrastructure. With competitive trade, any price change caused by an export ban should be detectable at these directly-linked cross-border markets, with markets further away from the border experiencing equivalent price changes if they are trading with the directly-linked markets and no price change otherwise. The resulting dataset includes 49 markets and 40 cross-border market pairs . This includes an additional 6 markets in areas not covered by the FEWS NET database in western Tanzania, eastern Malawi, and northern Mozambique, which I added to my dataset using price data from the Ministries of Agriculture and of Industry, Trade,strawberry gutter system and Marketing in these countries.The median market town has a population of 178,000, and the median market pair is separated by a road distance of 345 kilometers. All prices are expressed in US dollars per kilogram using monthly exchange rates provided by FEWS NET. The mean maize price across all markets and all periods is $0.274/kg. The price data is not complete as data collection began in some markets after January 2002 and there are a few missing observations throughout. The median price series has 102 of 120 possible observations, and 40 of the 49 markets have at least 6 years of data. Of 5,880 possible price observations, 1,435 are missing. I will show that my results are robust to restricting the panel to a more balanced subset.
My main empirical specification estimates the effects of export bans on the price gaps between pairs of cross-border markets instead. Export bans are unlikely to be endogenous to price gaps, since the events that trigger them are unlikely to affect the costs of trade between the cross-border market pairs. In the following section, I confirm with model simulations that in the absence of any bans price gaps would not have been higher or lower during periods in which bans were in fact implemented than in periods in which they were not. In theory, export bans work by increasing the costs of trade between cross-border market pairs . The spatial price analysis literature has thought carefully about the relation between price gaps, which are observable, and total trade costs, which are typically unobservable . Under competitive trade, the price gap between a pair of markets is equivalent to the total trade costs between those markets if trade is occurring, which Baulch and others have called “regime 1.” If trade is not occurring, the markets are in a segmented equilibrium , and the price gap between them is a lower bound on the total trade costs. The price gap may also temporarily exceed the trade costs if the markets are in disequilibrium following a shock . For markets with relatively low trade costs and consistent import-export relationships, restricting attention to regimes 1 and 3 would be appropriate. However, given the high trade costs in the agricultural sector in sub-Saharan Africa estimated in the previous chapter and the fact that maize is produced locally in all of the markets in my dataset, it is important to account for the possibility of regime 2 segmented equilibria in which export bans would have no effect because they are not binding. If I include these “no-trade” observations in my estimation, the resulting estimate of my parameter of interest β is a valid measure of the effects of export bans in a reduced-form sense conditional on market conditions at the time of ban implementation but is a down wardly-biased estimate of the effects of export bans conditional on the ban actually binding and preventing trade that would otherwise have occurred. Recent empirical evidence in contexts where regime 2 observations can be identified confirms this downward bias when all observations are included . In the regressions that follow, I experiment with different ways of identifying and excluding potential regime 2 no-trade observations and compare my subsequent results to my baseline reduced-form result using all observations. Column 1 of table 2.2 shows results from the specification given in equation 2.1 with all observations. The mean of the dependent variable is $0.0853/kg. The point estimate for the effect of export bans on this gap is less than three-thousandths of a US cent or less than three-hundredths of a percent of the mean price gap and is not statistically significantly different from zero at any confidence level. I calculate standard errors directly because of the complicated nature of potential correlation between the residuals in my dataset.
The standard approach with panel data would be to cluster at the market pair level to allow for correlation of residuals for a given market pair in different time periods. However, as is clear from the map in figure 2.1, the market pair structure also has features of a dyadic regression, with a single market often being a member of multiple market pairs. To deal with this additional source of correlation, I extend the approach of Fafchamps and Gubert for calculating consistent standard errors in cross-sectional dyadic regressions, allowing for correlation of residuals between any observations sharing at least one common market while continuing to assume that residuals are independent across observations with no common markets. The standard errors calculated using this dyadic approach are very close to those obtained by clustering at the market pair level . Using the standard errors from column 1 and the mean maize price of $0.274/kg, I can reject an alternate hypothesis that export bans have an effect at least as large as that of a 5% export tax at an 8% significance level. A 5% export tax is at the low end of short-term trade policy responses to commodity market price fluctuations — temporary export taxes of 25–40% are not uncommon . Of course, such taxes may not translate into empirical price differences, so the benchmark used here should be interpreted as the theoretical effect of a permanent 5% export tax. The specification in equation 2.1 implicitly assumes that no other variables besides export bans systematically affect price gaps over time. In columns 3 and 4 of table 2.2, I introduce additional covariates to capture some of this potential temporal variation. Recent results from Dillon and Barrett highlight the importance of fuel prices for maize trade in East and Southern Africa. I construct monthly retail diesel price series in US dollars per liter at the national level for my 12 countries of interest by using biennial observations from the International Fuel Prices project of GTZ to compute markups over the Dubai Fateh crude oil index and filling in gaps between GTZ observations using markups inferred by linear interpolation. In column 3, I add a term interacting these fuel prices in the origin market with the distance to the destination market as well as a set of indicator variables for major infrastructure projects affecting particular cross-border links compiled from government ministries and local newspaper archives. The point estimate for the diesel-distance coefficient corresponds to the expected cost of a 10- metric ton truck consuming 23 liters per 100 kilometers , although it is not statistically significant at conventional levels. In column 4, I include quarterly time fixed effects and a time trend instead. In both of these new specifications, the coefficient estimate on export bans is negative and not statistically different from zero.