Similar results were obtained using monthly and annual fixed effects with and without a time trend as well as including all variables from both columns 3 and 4. In column 5 of table 2.2, I exclude two outliers: the pairs involving Juba, South Sudan and Hargeisa, Somaliland, which have the highest and most volatile prices of the markets in my dataset. Their exclusion does not affect my results but does reduce the standard error on my export ban coefficient estimate. This enables me to reject my alternate hypothesis that the effect of export bans is as large as the theoretical effect of a 5% export tax at a 3% significance level. In column 6, I explore whether the unbalancedness of the panel is affecting my results by excluding all observations before January 2006, reducing my dataset from ten years to six. With this adjustment, of the 3,528 possible price observations in my new panel, only 171 are missing, as opposed to 19% in my original panel. My basic result that export bans do not have a statistically significant effect on the price gaps between pairs of affected cross-border markets remains unchanged. In a further set of robustness checks not presented here, I interacted implementing country indicator variables with the export ban indicator variable to look at potential heterogeneous effects. Again, none of the coefficients were statistically different from zero, hydroponic fodder system indicating that none of the countries’ export bans had a statistically significant effect on the price gaps between pairs of affected cross-border markets.
I next consider the possibility that regime 2, segmented equilibrium observations are biasing my coefficient estimate towards zero. Suppose that export bans do increase price gaps significantly for market pair-periods where they prevent trade from occurring but that I am not detecting this increase because I am including many other observations of segmented equilibria where trade would not occur with or without an export ban. In this case, if I were to progressively drop increasing numbers of these regime 2 observations from my dataset my coefficient estimate should increase. In figure 2.2, I experiment with two different ways of identifying and excluding potential regime 2 observations. Teravaninthorn and Raballand present data on transport prices for several major African transport corridors that range from $0.07 to $0.13 per metric tonkilometer. In the previous chapter, I find that total trade costs are roughly double these baseline freight rates and are higher off of major corridors, with a median trade cost of $0.29/t-km. Among the cross-border market pairs considered in this chapter, the maximum per-distance trade cost estimated in the previous chapter is $0.70/t-km, and the maximum absolute trade cost is $0.20/kg. Since trade costs are un upper bound on price gaps in regime 2, I proceed by dropping all observations in my dataset with price gaps below aprogressively increasing threshold. In the left panel of figure 2.2, I use the per-distance price gap and progressively drop observations from 0 up to $0.70/t-km. In the right panel, I use the absolute price gap and progressively drop observations from 0 up to $0.20/kg. At the maximum threshold, only 354 and 297 observations remain in the dataset. In both cases, the point estimate stays statistically insignificant and very close to zero, and there is no sign of an upward trend as I drop increasing numbers of potential regime 2 observations.
I conclude that my failure to detect an effect of export bans on cross-border price gaps is not due to the presence of regime 2 observations. In this section, I run simulations using the estimated dynamic monthly model of grain storage and trade in sub-Saharan Africa from the previous chapter to help understand the surprising empirical result from the previous section. The model consists of a representative consumer and a representative competitive trader in each of 230 large hub markets covering all 42 countries of continental sub-Saharan Africa. The model includes maize and five other major staple grains, and its demand, storage cost, and trade cost parameters were estimated using data from May 2003 – April 2013. Given monthly production, world prices, demand, storage costs, and trade costs, traders decide each month how much of each grain to sell locally, to put into storage in each of the 230 locations, and to trade along each of 413 overland bilateral transportation links as well as with the world market through 30 major ports. I adjust the model to start and stop a year earlier so as to match the time frame of the empirical exercise in the previous section and run three simulations. In the first simulation, Iassume that export bans are not implemented so that trade is possible between cross-border market pairs during every month at the constant, pair-specific trade costs from the original estimated model. In the second simulation, I assume that the 13 export bans from table 2.1 are implemented and perfectly enforced and that traders are na¨ıve, so the imposition and lifting of the bans takes traders by surprise. This means that prior to bans storage and trade decisions are made assuming that trade will always be possible at the constant, pair specific trade costs, and during bans these decisions are made assuming that trade will never again be possible between the affected pairs .
In the third simulation, I assume that the bans are implemented and perfectly enforced but that traders have perfect foresight about the imposition and lifting of bans. This gives them the possibility of exporting prematurely before bans are imposed and storing for future exports during bans. Realistically, trader behavior is likely somewhere in between the second and third simulations,fodder system given that precise information about future discretionary government actions is not available but that some anticipation is certainly possible. After solving month by month for the full continent-wide equilibrium for each of the three simulations, I extract the price series for the 47 markets and 33 market pairs corresponding most closely to the 49 markets and 40 market pairs from the dataset used in the previous section. Figure 2.3 highlights these market pairs against the backdrop of the other markets and transportation links in the continent-wide model. I then run the same regression fromequation 2.1 using the price gaps for these market pairs from each of the model simulations . The results reported here do not change significantly when all affected cross border pairs from the model are included. The results in table 2.3 are helpful for distinguishing between different explanations for my finding in the previous section that export bans do not have a statistically significant effect on the price gaps between cross-border markets. One possible explanation is that export bans do increase price gaps but are implemented during periods of abnormally small price gaps, which prevents me from detecting the effect. I can rule out this explanation using the first simulation, which shows that in the absence of export bans, the difference between the cross-border price gaps during periods when export bans were and were not actually implemented would not have been significantly different from zero. A second explanation is that export bans are not binding or that the trade flows they do prevent are so small that the bans do not have a significant effect on price gaps. I can rule out this explanation using the second simulation, which shows a very large effect of export bans on cross-border price gaps when traders do not anticipate ban imposition and lifting. The size of the effect is over four times larger than the average price gap of $0.0853/kg. A third possible explanation is that since maize is storable and bans are temporary, traders are able to limit the actual effects of export bans when they can anticipate their imposition and lifting. The third simulation shows that perfect foresight would enable traders to cut the effect of export bans on cross-border price gaps nearly in half, but the effect is still large and statistically significant at the 1% level. Interviews with traders in the region confirm that high storage costs make it costly for them to hold on to stocks while waiting for a ban to be lifted. Comparing my results from the model simulation without export ban implementation in the first column of table 2.3 with my results using the actual price data in table 2.2, it is clear that I cannot reject a hypothesis that export bans are simply not enforced.
To shed additional light on this hypothesis, I run additional regressions looking at the effects of export bans on prices in both the origin markets in the export ban implementing countries and the destination markets in the trading partner countries.A concern with the specifications in equations 2.2 and 2.3 discussed previously is that export bans are likely endogenous to prices as they are ostensibly implemented during periods of high prices. Results in table 2.4 using the price series from the simulation with no bans confirm that in the absence of export bans, prices in both origin and destination markets would have been 2 US cents/kg higher during periods when bans were in fact in place than in periods when they were not. However, in the data, prices in both origin and destination markets are 6 US cents/kg higher during export ban periods, and the difference with the noban simulation is statistically significant. This suggests that export bans are in fact having some effect on market outcomes, although the empirical effects of export bans in the data are still very different from those in the model simulations in which the bans are fully enforced with either na¨ıve traders or traders with perfect foresight. In both of these simulations, the price response to bans is more consistent with theory, with destination market prices increasing substantially and origin prices falling or remaining statistically unchanged . The results from this section suggest that the lack of an effect of export bans on the price gaps between pairs of cross-border markets is consistent with the bans not being enforced and is not consistent with the bans being perfectly enforced, even if traders can anticipate the bans. However, the fact that prices in both origin and destination markets are higher during export bans than they would have been in the absence of bans suggests that the bans are affecting markets somehow. In the following section, I present information collected from market participants in the region about ban enforcement that helps explain these findings. As part of the research for this chapter, I obtained information about ban enforcement from market participants in Malawi, Tanzania, and Zambia, which together are responsible for 10 of the 13 export bans in my dataset. I conducted interviews with formal and informal private traders of all sizes, trader associations, farmers, government officials, and market observers including FEWS NET and the World Food Programme in these countries. I also visited six border points in the region during export bans. The consensus among market participants is that export bans are implemented but imperfectly enforced. The formal export of maize requires an export permit, typically issued by the Ministry of Agriculture. When an export ban is imposed, permits are no longer issued. The bans shut down most of the formal maize trade, but maize continues to cross borders. Some formal traders are able to obtain export permits during bans through back-door channels. Informal traders, who may not be eligible for or choose to obtain export permits even during non-ban periods, are also able to continue moving maize across borders during bans. At official border points, informal traders often use bicycles, which are not regulated, to move maize between trucks on either side of the border. Informal traders also use unofficial border crossings along the long, porous land borders between countries in the region. The total volume of maize that can be transported across affected borders during bans may be subject to capacity constraints. At and around official border points, market participants report that enforcement is positively correlated with volume, with border officials generally tolerating low volumes of informal trade during bans but initiating patrols and crackdowns when volumes increase.