The Dynamic Effect of the Real Effective Exchange Rate
(REER) on Agricultural Trade
Author: Jasper Gerard Gussenhoven* Supervisor: prof.dr. L.J.R. (Bert) Scholtens
Second assessor: dr. B.A. Boonstra
University of Groningen and University of Uppsala
Abstract
This paper examines the impact of real effective exchange rates on agricultural import and export volumes of twelve developing countries in Asia and Latin America. It uses first-difference stationary series to avoid spurious regressions and examine the dynamic relationship. Covering a sample of annual data from 1973 - 2010, we mainly reject hypotheses of the impact of REER on agricultural trade volumes. After imposing a one year lag on each explanatory variable, we do not find support for the existence of an adjustment lag with respect to trade volumes.
I. Introduction
As people from industrial nations seem to take food for granted (e.g. the poorest five percent of the US spend only sixteen percent of their income on food), for inhabitants of developing countries this is certainly not the case. Indian families for example spend around half of their budget on food, whilst for Nigerian families this number even reaches 73 percent (The New
York Times, ‘The world food crisis', 2008). While world food prices during 2007 and 2008
already exceeded those of any prior generation, prices in 2011 even passed the 2008 peak. These skyrocketing prices were mainly due to rising demand in developing countries and volatile weather, which worsened situations in especially developing countries leading to mass hunger and political instability (The New York Times, ‘The food crisis’, 2011).
When the supply of one country comes short, an excess of demand has to be met by the supply of other countries. Supply and demand affect international prices and prices in turn affect supply and demand. Exchange rates act as a medium in this respect, converting foreign prices into domestic currency and vice versa. However, exchange rates do not respond only to demand and supply of agriculture, but depend on all international capital flows (Orden, 2000). Vice versa, exchange rates and exchange rate volatility might affect and diverge agricultural prices from levels that ensure equilibrium of international demand and supply of agricultural products. In this study we are interested in the latter effect. Due to miserable situations in developing countries and the primary concern for food of their inhabitants we assess what the impact is of the real exchange rate on particularly agricultural trade in developing countries. 1
Previous research of the effect of exchange rates on trade, which is mainly based on theoretical foundations of the J-curve and the Marshall Lerner (ML)-condition, yielded mainly varying results. In general, empirical literature demonstrates that the degree of economic development among countries does not seem to affect results regarding the effect of exchange rates on trade, while aggregation issues (sectoral aggregation and the ‘aggregation bias’ stated by Bahmani-Oskooee and Goswami, 2004) are pointed out as causes for insignificant results. With respect to agricultural trade, empirical literature (Doroodian, Jung and Boyd, 1999; Huchet-Bourdon and Korinek, 2011) for the most part demonstrates that the J-curve effect is more pronounced in agriculture than in other sectors due to specific delivery lags, production lags, pre-determined contracts and the ease to change suppliers due to the homogeneity of agricultural products. Previous empirical findings (e.g. Arize, Osang and Slottje, 2008; McKenzie and Brooks, 1997; Tenreyro, 2004) of the effect of exchange rate volatility on trade are mixed. Overall, volatile exchange rates tend to slightly dampen agricultural trade, mainly due to specificities in agriculture which give rise to risk aversion.
1
Assuming absence of reverse causality (section 4.1.1), this paper empirically examines the impact of real exchange rates on agricultural trade volumes of specifically twelve developing countries divided over Asia and Latin America. More specifically, it uses the real effective exchange rate (REER), which is a combined index of a country’s weighted real exchange rates to its trading partners. Although other factors like the productivity of farmers, rainfall and trade barriers are likely to affect supply and demand for agricultural trade, it is impossible to account for each factor. Following previous literature (Bahmani-Oskooee 1986, Doroodian et al. 1999; Fidan, 2006) and theory of Van Marrewijk (2007, p. 491) we include ratios of agricultural import and export prices to world prices (henceforth: relative import and export prices) as control variables and impose one year lags on each explanatory variable to account for the expected ‘adjustment lag’ with respect to trade volumes. ADF unit root tests mainly demonstrate nonstationarity of the variables in their level. To avoid spurious regressions we therefore convert series to stationary I(1) series. Taking first differences of the log-linearized variables implies we are dealing with a dynamic relationship.
Using the Ordinary Least Squares (OLS) - method, we find that results differ from our expectations. Effects of the real exchange rate on trade volumes are weak where half of the coefficients even carry opposite signs. Some import elasticities from the exchange rate even carry significant perverse effects. However, some findings provide reasons to suspect the relation to hold particularly for the import model. Exchange rate estimates of Chile and Venezuela and the lagged estimate of Colombia point in the right direction and are highly significant. Findings of exports do not confirm any hypothesis. Moreover, findings of import models are more in line with our expectations and have higher explanatory power than export models. More than anything else, relative import and export prices seem to affect trade volumes. Out of the twenty four estimates, sixteen are significant with elasticities between -0.25 and -0.98. Opposed to our expectations based on arguments for the adjustment lag and results of Bahmani-Oskooee (1986), findings do not demonstrate significant results for the effect of lagged relative import and export prices on trade volumes.
This paper proceeds as follows. Section II contains a discussion of the empirical literature. Section III outlines the models with accompanying hypotheses. Section IV describes the methodology and data used in this study. In section V, we present the empirical results. Finally, section VI concludes.
II. Literature Review
Much research has been devoted on the impact of exchange rates on imports and exports. Prices and both volumes and values of trade seem to be affected by exchange rates and their volatility (Bahmani-Oskooee and Ardalani, 2006; Smith, 2009; Arize, Osang and Slottje, 2000). Section 2.1 summarizes empirical literature of the effect of exchange rates on trade whereas section 2.2 elaborates on previous research of the effect of exchange rate volatility on trade.
2.1 Impact of Exchange Rates on Trade
Most of the recent literature examines the impact of the real exchange rate on trade values rather than trade quantities. These authors2 use an alternative approach than ours, where they study the relationship between value of exports of a country (inpayments) and value of imports of a country (outpayments) and the real exchange rate. Although we cannot derive from their results the direction and magnitude of the effect of exchange rates on trade quantities, these studies do provide important insights for our study. Moreover, it is the effect on trade quantities which underlie the final effect on trade values. Studies that investigate the effect of exchange rates on trade values are mostly based on theoretical foundations of both the J-curve and the Marshall-Lerner condition.
2.1.1 Trade Balance and the J-curve
The J-curve effect occurs when a depreciation of the exchange rate first leads to a deterioration of the trade balance which later on is followed by an improvement of the trade balance. At the moment a currency depreciates, orders/contracts are fixed as they are mostly pre-determined (medium-term contracts). The devaluation therefore causes a price effect in the first place. Initially, during the ‘contract period’, imports (i.e. import quantity x import price) rise due to higher import prices caused by the devaluation. The opposite happens with exports. The value of exports decline due to the price effect and the result is a deterioration of the trade balance (= exports – imports) in the short-run. The substitution away in case of a currency depreciation from imported to exported goods due to new real prices requires time to adjust, and we refer to the latter with an ‘adjustment lag’. Consumers need time to
substitute between different goods and adjust their consumption patterns, and producers need time to install new plants, attract capital, hire workers, build new distribution channels, etc. Over time, higher import prices reduce quantities of import while lower export prices increase quantities of export. The result according to the theory (Van Marrewijk, 2007, p. 491) is an improvement of the trade balance in the long-run. If one would plot time on the horizontal axis and the trade balance on the vertical axis the graph resembles a curve over time in the form of the letter J, hence the J-curve effect.
Many authors have studied the effect of real exchange rates on national trade values, hereby mainly investigating the impact of a depreciation (or appreciation) on the long-run net effect of the aggregate trade balance. Empirical research with respect to this topic however, has yielded varying results, where the general economic development of a country does not seem to play a specific role: 3.4
Rose (1990) for instance studies the effect of a real depreciation on the trade balance for thirty developing countries and only Thailand’s trade balance showed to be significantly improved following depreciation. Similarly, Upadyaya and Dhakal (1997) only find an improvement for Mexico, while they demonstrate that Cyprus, Greece and Colombia even carry significant perverse estimates. Baharumshah (2001) and Stucka (2004) provide on the other hand favorable results among developing countries. Baharumshah assesses bilateral trade of both Thailand and Malaysia versus both Japan and US and demonstrates that the trade balance of both developing countries improve in the long-run. Short-run patterns could not be detected so none of the countries seem to experience a J-curve effect. The effect of depreciation in Croatia is comprehensively investigated in a study by Stucka (2004). Using two types of real exchange rates (CPI and PPI based) he demonstrates that a 1% devaluation leads on average to a 0.94 percent (CPI based) and a 1.3 percent (PPI based) improvement of the trade balance. Moreover, he finds evidence for the J-curve effect.
Results of studies with respect to developed countries vary as well. Bahmani-Oskooee and Brooks (1999) study trade between the US and six developed European countries and find significant improvements over the long-run (except for the UK). On average a devaluation of 1% led to an improvement of 0.9% of bilateral trade balances. Proof for short-run effects to support the J-curve effect is missing in their study. Complementary to Bahmani-Oskooee and Brooks (1999), is the study of Bahmani-Oskooee and Ratha (2004). While 50% of US’ trade constitutes trade with developed countries examined by Bahmani-Oskooee and Brooks (1999), the other approximately 50% constitute trade with developing countries. Bahmani-Oskooee and Ratha examine therefore the effect of exchange rates on trade between the US and 13 developing countries. They obtain mixed results. For Chile, Korea, Mexico, Malaysia
3
See appendix table A1 for an overview of these relevant papers.
4
and South-Africa a bilateral devaluation of the US dollar indeed led to an improvement of the trade balance but for trade with other countries any significant result remained absent. Again, short-run effects could not be detected to support any J-curve (except for Malaysia). Bahmani-Oskooee and Goswami (2004) examine the trade balance of Japan to her nine largest developed trading partners and show that exports are not affected by a devaluation in the long-run but imports from the US, the UK, Italy and Germany are, on the contrary, highly sensitive with an average import elasticity of -1.82. This is contrary to empirical literature that show exports to be more sensitive to exchange rates than imports (e.g. Bahmani-Oskooee and Ardalani, 2006; Huchet-Bourdon and Korinek, 2011). For an explanation they refer to Haynes et al. (1986) who argue that Japanese exporters focus on preventing a decline in their export share by adjusting prices and profit margins at the moment the exchange rate moves in a – for them - harmful way. Related is feedback from exporters provided by Smith (2009) who suggest that “many exporters will absorb lower current profitability in the expectation that the exporting environment will eventually improve. This reflects the tendency of exporters to maintain their position rather than exiting.” Bahmani-Oskooee and Ardalani (2006) demonstrate in their study the exact opposite where long-run imports do not seem to be significantly affected by the exchange rate while exports are. Their results furthermore imply that 10% depreciation leads to an improvement of the trade balance of 7.9%. Once more, evidence in favor of an improvement of the trade balance in the long-run is mixed.
A major shortcoming of Rose (1990), Upadyaya and Dhakal (1997), Stucka (2004), and Bahmani-Oskooee and Ardalani (2006) - according to Bahmani-Oskooee and Goswami (2004) - is the employment of aggregate data; both aggregate trade balances and aggregate exchange rates like REER. The so-called ‘aggregation bias’ arises when data of multiple trading partners are combined in one elasticity. Bahmani-Oskooee and Goswami (2004) explain the issue by stating that: “The problem (with aggregate data) is that a significant price elasticity with one trading partner could be more than offset by an insignificant elasticity with another partner yielding an insignificant trade elasticity.” Avoiding any problems associated with aggregate data they continue, “…necessitates estimating trade elasticities on a bilateral basis.” However, since bilateral import and export prices are not available, an examination of the effect of exchange rates on trade volumes is impossible using bilateral data, because one cannot control for price changes.
2.1.2 Sectoral Disaggregation
Another aggregation problem could arise from different effects between sectors or industries. When data is aggregated like in aforementioned studies, results – following a similar reasoning as the previous paragraph - can be misleading when effects differ among industries. Avoiding such a problem requires disaggregation of industry or sector. Appendix table A2 provides an overview of papers discussed below regarding disaggregation of sectors.
sector but do support it for agriculture. They show that the initial deterioration of the agricultural trade balance lasts two quarters, before it starts improving. For an explanation they point out that the initial deterioration of the trade balance may be especially pronounced among agricultural products due to both longer delivery and production lags. A delivery lag comprises the time between the orders and the actual delivery from the seller to the buyer. Doroodian et al. show that the ratio of inventories to production output is three times higher with manufacturing than agriculture. Therefore, manufactured goods have a shorter delivery lag because a rise in demand is easier met with substantial inventories. A production lag comprises the time from the moment production has started until it is finished. The existence of the production lag in agriculture is well known and addressed among others by Buse and Bromley (1975). Many crops require four to six months to produce, while manufactured goods are not subject to such long production lags. It is not surprising therefore that, due to this lag and the associated risk, the contract market in agriculture is substantial. Doroodian et al. additionally provide a second explanation for lacking support for the J-curve in previous studies. They argue that because trade from industrialized countries (which was examined more often than developing countries prior to 1999) constitutes a higher manufacturing to agriculture ratio, especially studies that aggregate sectors for developed countries fail to find evidence in favor of the J-curve because manufacturing goods does not seem to follow such a pattern. When we review recent studies that do cover developing countries we cannot confirm their second explanation due to varying results.
Similarly, Huchet-Bourdon and Korinek (2011) also investigate the effect of changing exchange rates on both agriculture and manufacturing.5 Their geographical span is very wide covering China, the US and the Euro area. Similar to Doroodian et al. (1999), their findings suggest that agricultural trade in the long-run is more affected by exchange rates than manufacturing trade is. Huchet-Bourdon and Korinek explain the difference by referring to the degree of product homogeneity among the sectors. A higher degree of homogeneity among agricultural products makes it easier to substitute suppliers of both exports and imports. A devaluation of a country’s currency for example will, due to the ease of changing suppliers, lead to extra demand for exports and reduced demand for suppliers of import (e.g. suppliers of import might be replaced by domestic suppliers). This will probably not hold for the short-run where medium-term contracts – which are particularly common in agriculture due to lags with associated risks (Doroodian et al., 1999) - are fixed. When the contract period ends, not only quantities will be adjusted, but suppliers will be replaced as well. Therefore, the greater ease to change suppliers accentuates the improvement of the trade balance in agriculture. Short-run findings of Huchet-Bourdon and Korinek (2011) are however mixed in both sectors and do not support any J-curve effect.
Shane, Roe and Somwaru (2008) take disaggregation a step further as they examine the effect of the exchange rate on US agricultural exports. They divide agriculture into twelve
5
commodities and use commodity-specific REER’s. Eight out of the twelve commodities carry a significant negative estimate and interestingly, they find different effects between bulk and high value commodities. High value near consumer-ready products (such as red meats, poultry, fresh fruits and vegetables) carry low trade elasticities (on average -0.57) from exchange rates and tend to be exported more to high income countries, whereas bulk commodities (e.g. wheat, rice, corn) are more sensitive (on average -0.95) to exchange rates and tend to be exported to lower income countries. Aggregate data suggest furthermore that a 1% appreciation of the dollar led to a 0.51% decrease of total agricultural exports.
A similar extent of disaggregation is applied by Bahmani-Oskooee and Ardalani (2006). Bahmani-Oskooee and Ardalani disaggregate US’ trade into 66 industries where they demonstrate that half of the industries carry significant negative export elasticities from exchange rates. Among these were corn, meat and preparations, rice and wheat with elasticities of respectively -1.82, -1.38, -1.6 and -1.15, suggesting again that agricultural exports are sensitive to the real exchange rate. Nevertheless for imports only 13 out of the 66 industries were found to be significantly sensitive for the exchange rate of which ‘meat and preparations’ was the only agriculture related category. Finally, aggregated exports were significantly sensitive for the real exchange rate with an elasticity of -0.79 while aggregate imports did not prove to be affected. For an explanation of lower and insignificant import sensitivities they argue, as Haynes et al. (1986), that foreign exporters squeeze their profit margin in order to maintain market share of the US market. Bahmani-Oskooee and Ardalani (2006) furthermore separate industries in durable and non-durable goods and in small and large industries but neither of the two separations seemed to obtain any significant result for neither imports nor exports.
2.1.3 Trade volumes and the Marshall-Lerner condition6
In the last section we discussed empirical literature on the effect of exchange rates on values of (sectoral) trade, and the J-curve. However, an improvement of the trade balance resulting from devaluation depends on the underlying so-called Marshall-Lerner condition (henceforth ML-condition). This condition states that an improvement of the trade balance only takes place if the sum of export and import demand elasticities (in absolute value) exceeds unity (Van Marrewijk, 2007, p. 489). When the value of exports is elastic to exchange rates, the increase of export quantities is proportionately more than the decline in price following currency depreciation, such that the total value of exports will increase. Similarly, when the value of imports is elastic to exchange rates, total value of imports will decrease as a result of currency depreciation. However, goods for both imports and exports do not need to be elastic for an improvement of the trade balance to take place. Rather, the condition is met when the increase in income from exports exceeds the decrease in expenditures on imports. If for example imports and exports are inelastic with elasticities of 0.7 and -0.7 respectively, then
the sum in absolute value is 1.4 which exceeds unity. A depreciation of a country’s currency results in an improvement of the trade balance.
Bahmani-Oskooee and Niroomad (1998) examine both import and export price elasticities of 30 countries in order to test whether the ML-condition holds. Their findings suggest that for almost all countries (developed and developing) the condition is satisfied, concluding that a devaluation could indeed improve the trade balance. Mahmud, Ullah and Yucel (2004) find however opposite results using a non-parametric approach. Their study examines ML-condition for six developed countries but exclusively find favorable results for Norway. Only when sub-sample periods are used, the condition seems to hold. Their results furthermore suggest that the condition is more likely to be satisfied under fixed exchange rate regime. Rose (1991) finally neither finds favorable results for ML-condition where he uses both parametric and non-parametric techniques considering five OECD-countries, being Canada, Germany, Japan, UK and the US.
The simplest formulation of aggregate import and exports equations, employed both by Bahmani-Oskooee and Niroomad (1998) and Mahmud et al. (2004), include domestic real income and the ratio of import prices to domestic prices as explanatory variables where they assume a degree of substitutability between imported and domestic goods. Similarly, they include foreign real income and the ratio of export prices to world export prices for an aggregate export function. Since the purpose of Bahmani-Oskooee (1986), Fidan (2006) and Smith (2009) is to assess the effect of exchange rates on trade flows, they all add an exchange rate variable.7 Bahmani-Oskooee (1986) estimate both import and export demand functions for a sample of seven developing countries, namely; Brazil, Greece, India, Israel, Korea, South Africa and Thailand. While import elasticities are positive for the majority of countries, his results indicate that relative import prices and exchange rates do not significantly affect import volumes. Likewise, export volumes are not sensitive to relative export prices. Income elasticities in the export model are, opposed to those with imports, relatively low (except for India and South Africa) and exchange rate coefficients are mixed but mostly insignificant. Expected negative signs are obtained for Israel, South Africa and Greece, where only the latter exceeds unity. As a second objective, Bahmani-Oskooee (1986) assesses the speed with which trade flows respond to changes in both prices and exchange rates. For the long-run most coefficients had the expected signs, where lags of prices tend to stretch out somewhat longer than exchange rate lags, over a range from two up to eight quarters. In line with Bahmani-Oskooee’s second finding, Artus and Knight (1984) demonstrate that one year after devaluation the ML-condition is met for each country except for Denmark, being Austria, Canada, Japan and The Netherlands.
While Bahmani-Oskooee (1986) examines aggregate import and export demand functions, Fidan (2006) and Smith (2009) investigate trade volumes on a sectoral level where they avoid the problem of sectoral aggregation as previously discussed. Smith (2009) investigates the
effect of exchange rates on solely exports of New Zealand, a small open economy. His findings suggest that the measured elasticity of New Zealand’s aggregate export volumes is significant but quite small; if the real exchange rate appreciates by 10 per cent, export volumes decline by 1.4% after eighteen months (which is the lag at which the effect has it peak). Low domestic export prices furthermore, seem to lower aggregate export volumes by the same magnitude and lag. For an explanation Smith points out that most of New Zealand’s exporters are supply-constrained, and again that contracts are pre-arranged. More interesting however, are results regarding sectors of export. Both magnitude and lags with which the real exchange rate affects volumes seem to vary per sector in New Zealand. While service export volumes (e.g. tourism) respond with the same lag to exchange rates as aggregate volumes, manufacturing volumes appear to respond with a lag of twelve to fifteen months and agriculture with only a year. The latter is most remarkable since we would expect agriculture to respond with a relatively longer lag following aforementioned arguments of Doroodian et al. (1999). Additional findings of Smith (2009), in contrast what we have seen with Doroodian et al. and Huchet-Bourdon and Korinek (2011),8 suggest that non-primary export volumes (e.g. service sector) tend to be more sensitive to exchange rates than primary exports (e.g. agriculture sector).Service volumes in particular appear to be three times more sensitive to real exchange rates than aggregate export volumes.
Fidan (2006) studies the effect of the real exchange rate on Turkish agricultural trade volumes. For export he uses a similar model as Bahmani-Oskooee (1986) where he replaces aggregate exports and prices by agricultural exports and prices. For imports however, Fidan substitutes the ratio of import prices to domestic prices by the ratio of import prices to world import prices. His import model therefore ignores any substitution possibility between domestic and imported products. On the other hand, he does control for import prices that differ relative to world import prices of agriculture. His results suggest that exchange rates impact both agricultural imports and exports (elasticities of respectively 1.21 and -0.05), where he demonstrates furthermore that long-run effects are relatively greater than short-run effects.
In summary, we conclude that the effect of exchange rates on trade volumes and values is still ambiguous, and that empirical support for both the ML-condition and the J-curve is mixed. Throughout literature, a central element in this matter is the aggregation of sectors, where exchange rates tend to affect agriculture more than other sectors. Section III and IV empirically examine this relation for developing countries only. Before doing so however, we elaborate in section 2.2 on empirical literature of the effect of exchange rate volatility on trade, which carries its own importance, especially for agriculture.
2.2 Exchange Rate Volatility
This section contains insights and previous empirical findings of the effect of exchange rate volatility (henceforth ERV) on trade, which seems to be more pronounced in agriculture. Although we do not empirically examine ERV since it is beyond the scope of this paper to do so, previous research provides some interesting insights especially in agriculture and more importantly, our empirical findings are potentially biased by the effect. Until now, only Huchet-Bourdon and Korinek (2011) empirically addressed this problem by employing both exchange rate changes and exchange rate volatility in one model.
Most of all, empirical results with respect to the matter are mixed. Broda and Romalis (2004) and Arize et al. (2000, 2004) found in their studies for respectively both less developed and developed countries that exchange rate volatility (henceforth ERV) induces trading firms to shift from trade to non-traded goods, which reduces overall trade. Other studies have found that ERV may actually stimulate trade (Dellas and Zillberfarb, 1993; Franke, 1991; McKenzie and Brooks, 1997). Moreover, Kroner and Lastrapes (1993) find both positive and negative effects of ERV on trade, whereas Gagnon (1993), Tenreyro (2004) and Aristotelous (2001) find insignificant links between ERV and the volume of trade. Among these contradicting studies is the debate of the use of short-run vs. long-run data. A common argument against use of short-run volatility is that exchange rate risk can readily be hedged against in the short-run. Cho, Sheldon and McCorriston (2002) and Obstfeld (1995) among others show that long-run ERV indeed seems to have more significant impact on trade volumes than short-run exchange rate fluctuations have, following the latter reasoning. Huchet-Bourdon and Korinek (2011) mention however that “although hedging mechanisms are available, they are probably somewhat prohibitive for some particularly small and medium-sized enterprises, which may have less long-term visibility of their foreign exchange needs.” Moreover, Vianne and the Vries (1992) and Obstfeld and Rogoff (1998) argue that when risk-averse firms attempt to hedge against future exchange rate movements, they will impose a risk premium in terms of a mark-up to cover the costs. The result is higher prices which in turn reduces demand.
2.2.1 Specificities in Agriculture
degree of firms’ risk aversion and hence the effect of ERV on trade depends on - next to firm specific elements - sectoral characteristics listed below (Huchet-Bourdon and Korinek, 2011):
Price volatility;
Many production decisions are made prior to actual product sales. A high degree of price volatility therefore brings uncertainty about the price producers will receive. ERV can further affect price differences and so uncertainty.
Homogeneity of products:
If a commodity has a high degree of homogeneity, there is a greater ease to change suppliers. Hence, high ERV increases risk for sectors more with relative homogeneous products.
Trade barriers:
“Ad valorem tariffs (i.e. tariffs as a percentage of the value of the imported good) magnify the effect of international price changes since they are based on the imported price of the good, whereas specific tariffs (i.e. tariffs expressed in value per ton of merchandise) have a dampening effect.” Sectors with ad valorem tariffs could therefore be more risk-averse to ERV.
Wang and Barett (2007) and Klein (1990) name the following additional varying elements among sectors which they argue to likely result in different volatility effects on export volumes:
Level of competition:
If competition is high, firms cannot exert influence on their price. If they do, they will be substituted (see homogeneity of products). Hence they have a higher risk aversion.
Nature and currency of contracting:
Sectors with longer terms contracts are less sensitive for ERV as their contracts are set for a period of time.
Hedging instruments:
The higher the availability of hedging practices, the lower the risk aversion towards ERV as changes of exchange rates can be hedged against (in the short-run).
Storability of products:
ERV can be more damaging for suppliers of products with low storability. There is little room to adjust to altered real prices resulting from changed exchange rates.
shows that in five out of nine categories of goods (i.e. ‘Fuels and lubricants’, ‘Oils and fats’, ‘Chemicals’, ‘Manufactured goods’, ‘Machinery and transport’) ERV of all countries positively affect the value of US exports. Remarkable however is that only exports of ‘Food and Animal products’ prove to be significantly negative affected.
Partly in line with Klein’s (1990) results are those of Wang and Barett (2007). They also assess the effect of monthly ERV on export volumes of different sectors, where they examine bilateral trade between Taiwan and the US from 1989-1999. Wang and Barett, like Klein, demonstrate that monthly ERV negatively affects agricultural trade flows, but their results differ since ERV did not significantly affect trade in any other sector.9 For an explanation they argue that specific agricultural elements might play a substantial role. A high level of competitiveness, flexible pricing on relatively short-term contracts, a high degree of product homogeneity, and the relatively low storability of agricultural products are pointed out as possible explanations. Finally they mention that agriculture in particularly Taiwan – which receives little support from the government – might be affected more because it is based on small-sized firms operating in a low-margin industry.
Finally, Cho et al. (2002) also compare the effect of exchange rate volatility on trade between agriculture and other sectors (machinery, chemicals and other manufacturing) and provide similar results. Once more they prove that real exchange rate volatility had had a significant negative effect on agricultural trade between the US and the G-10 countries over the period 1974-1995 while total trade or trade in any other sector did not show to be significantly affected by medium- to long-run volatility in real exchange rates. Kandilov (2008) extends the analysis of Cho et al. (2002), but contents after he first replicates results of Cho et al. that the negative effect for agriculture disappears when he adds more developed countries. Apart from this result, he does find a small negative volatility effect on agricultural trade for developing countries.
Summing up, we conclude that – in line with most literature – the effect of ERV on trade is ambiguous, although the effect seems to be more pronounced for agriculture in the negative. Keeping that in mind, we depart from the effect of ERV for now and empirically attempt to answer our main research question: What is the effect of real exchange rates on trade
volumes?
9
III. Models and Hypotheses
3.1 Other FactorsBefore we continue with the empirical analysis it might be intuitive to consider, apart from included factors in empirical models, other determinants of trade volumes. Although most factors - directly or indirectly – are mentioned before, others are skipped because they are not employed in empirical studies. Figure 1 (Smith, 2009) shows the main determinants of export volumes. Investigation of the impact of exchange rates on agricultural volumes of both import and export requires an empirical model where ideally each determinant is controlled for. Due to both data and statistical constraints, this is not feasible. Figure 1 shows for example that demand influences for export consists among other factors of preferences (under ‘foreign demand’) and productivity (under ‘competitiveness’). Both variables are very difficult, if not impossible, to include in a regression.
Figure 1. Factors affecting export volumes
3.2 The Models
Since the objective is to examine the effect of the real exchange rate on agricultural trade volumes, we start with employment of the real effective exchange rate, REi,t, as an
explanatory variable.10 Despite the inability to include each influencing factor, it is possible to include some additional factors which are expected to impact trade volumes. In previous literature, world and domestic income have proven to positively affect export and import volumes respectively. Previous models and therefore our models include income as a control variable. Rather, considering an adjustment for inflation, it requires the use of real income. Following Bahmani-Oskooee (1986), a second control variable for import volumes would ideally be the ratio of import prices to domestic prices to include an allowance for substitution between domestic and imported goods. Due to data unavailability, we replace
this variable – imitating Fidan (2006) - by the ratio of import price to world price.11 An increase of this ratio would imply that agricultural import prices (relative to agricultural world prices) have gotten more expensive which therefore hurt demand for import, and vice versa for a decrease of the ratio. For export we employ the ratio of export price to world price as a second control variable (Bahmani-Oskooee, 1986; Fidan, 2006). A higher export price (to world price) would hurt demand for exports, whereas a lower export price would stimulate demand for exports.
Based on arguments of Doroodian et al. (1999), Van Marrewijk (2007, p. 491) and Smith (2009) concerning agricultural production lags, delivery lags, physical supply-constraints and fixed contracts, it is reasonable to expect trade volumes not to adjust instantaneously. Empirical findings in literature (section 2.1.3) mostly confirm such an adjustment lag, where most significant effects were found after lags of about one year. Lamb (2000)12 approaches this issue by adding one year lagged variables on both the real exchange rate and prices, and Bahmani-Oskooee (1986) imposes a distributed lag structure on each independent variable. Therefore, we also employ one year lags on REER, income and relative prices. In summary, we formulate a country i’s agricultural import and export volumes in the following equations respectively:
MVi,t = a + b REi,t + c REi,t-1 +d Ydi,t + e Ydi,t-1 + f (PM/Pw)i,t + g (PM/Pw)i,t-1 + ε (1)
XVi,t = h + i REi,t + j REi,t-1 + k Ywt + l Ywt-1 + m (PX/Pw)i,t + n (PX/Pw)i,t-1 + µ (2)
In equation (1), MVi,t is country i’s quantity of imports, which is assumed to depend on the
real value of its currency, REi,t, the lagged real value of its currency, REi,t-1, domestic income,
Ydi,t, the lagged value of domestic income, Ydi,t-1, the ratio of import prices to world prices,
(PM/Pw)i,t, and the lagged ratio of import prices to world prices, (PM/Pw)i,t-1. In (2), XVi,t is
country i’s quantity of exports, which is assumed to depend on the real and lagged real value of its country’s currency, REi,t and REi,t-1, real and lagged world income, Ywt and Ywt-1 and
the ratio and lagged ratio of export prices to world prices, (PX/Pw)i,t and (PX/Pw)i,t-1.
Summing up, it is expected that signs of b, c, d, e, k and l will be positive and those of f, g, i, j,
m and n will be negative. The parameters a and h are intercepts and all information that is not
included in the independent variables is summarized in the error terms ε and µ.
11 Fidan (2006) actually uses ‘world import price’ in the denominator.
3.3 Hypotheses
In order to test whether REER has a significant impact on agricultural trade volumes, we conduct the following hypotheses for the import and export model respectively:
H0: REER does not affect agricultural import volumes b = 0
H1: REER positively affects agricultural import volumes b > 0
H0: REER does not affect agricultural export volumes i = 0
H1: REER negatively affects agricultural export volumes i < 0
Additionally, incorporating an adjustment lag as previously described, we test whether the lagged REER significantly impacts agricultural trade volumes:
H0: Lagged REER does not affect agricultural import volumes c = 0
H1: Lagged REER positively affects agricultural import volumes c > 0
H0: Lagged REER does not affect agricultural export volumes j = 0
H1: Lagged REER negatively affects agricultural export volumes j < 0
It is tested whether we can reject the null hypothesis by performing an ordinary least squares regression of the econometric models (3) and (4). On a 5% significance level we test whether we reject the null hypothesis and fail to reject the alternative hypotheses in favor of their signs. If this is the case, we conclude REER significantly impacts trade volumes of agricultural products.
IV. Methodology and Data
4.1 MethodologyEstimating coefficients of equations (1) and (2) necessitates a regression analysis. In this section we describe implementation of the Ordinary Least Squares-method (OLS) and its accompanying underlying assumptions. Prior to testing, it is crucial for the estimations to ensure stationarity among the data series. The use of nonstationary data may lead to spurious results. Section 4.2.1 applies Augmented Dickey-Fuller (ADF) unit root tests for stationarity of the variables. Except for export volumes of Peru, each series could be made stationary taking first differences.
Because the nonstationary series are transformed to stationary after first differencing, estimation methods require modelling in first differences to avoid spurious regressions. In equations (3) and (4), the ‘Δ’ before each variable signifies the first differences (e.g. Δ ln Ywt
= ln Ywt - ln Ywt-1). In addition, dependent and independent variables are transformed by the
Bahmani-Oskooee and Goswami, 2004; Stucka, 2004; Bahmani-Oskooee and Ardalani, 2006). In the formulas this can be seen by an added ‘ln’ before each variable. Estimations of equations (3) and (4) measure therefore effects of relative changes of independent variables on relative changes of the dependent variable. An additional nice feature of this “log-log” model - which is also called the constant elasticity model (Hill et al., 2007, p. 87) - is that if the assumptions of the regression hold, each parameter b - n (except for h) is the elasticity of the trade volumes with respect to that particular factor for all points on the regression line. After log-linearization and transformation of the models into first differences, equations (1) and (2) – similar to models of Fidan (2006) - can be written as:
Δ ln MVi,t = a + b Δ ln REi,t + c Δ ln REi,t-1 +d Δ ln Ydi,t +e Δ ln Ydi,t-1 + f Δ ln (PM/Pw)i,t
+ g Δ ln (PM/Pw)i,t-1 + ε (3)
Δ ln XVi,t = h + i Δ ln REi,t + j Δ ln REi,t-1 + k Δ ln Ywt + l Δ ln Ywt-1 + m Δ ln (PX/Pw)i,t
+ n Δ ln (PX/Pw)i,t-1 + µ (4)
Another approach is to conduct separate models of which two models contain non-lagged explanatory variables, while two other models contain only non-lagged explanatory variables. Regression results of these alternative models do not vary substantially with results of equation (3) and (4). For simplicity we therefore solely present results of equation (3) and (4) in section V.
4.1.1 Underlying assumptions
We refer to equation (3) and (4) as multiple regression models because there are three explanatory variables included rather than one, as is the case with a simple regression model. If the assumptions of the multiple regression models hold, then the least squares estimators are the best linear unbiased estimators (BLUE) of the parameters (Hill et al., 2007, p.115). Appendix F describes and tests these assumptions. Overall, test results indicate that only a few assumptions are violated. Heteroskedasticity (which is present in three models) is treated by applying White heteroskedasticity-consistent standard errors. Non-normal distributed errors are found only for the import model of Chile and we detect autocorrelation for the import models of Colombia and Malaysia.
4.2 Data
This study assesses the effect of exchange rates on agricultural trade of developing countries. To obtain credible results we enlarge the country sample to a total of twelve. Additionally, for comparison of effects between two continents we include six countries of both Latin America and Asia. Brazil, Chile, Colombia, Ecuador, Peru and Venezuela represent Latin America whereas India, Indonesia, Malaysia, Pakistan, Philippines and Thailand represent Asia.
“When the United States abandoned the Bretton Woods agreement on relative fixity of exchange rates in 1973, a new era of international capital mobility was launched, and the rules of the game for macroeconomic interdependence among nations were altered” (Orden, 2002). Therefore, this study and many others analyze the effect of exchange rates from 1973 onwards. Aiming for a large possible and recent dataset we include data up to and including 2010. Due to data unavailability of quarterly (and monthly) incomes and prices, we use annual data. In total, 38 observations are in place which is reduced to 37 after first differencing.
Each dependent and independent variable in this study constitutes an index. Since the indexed variables initially contained different base-years, we re-indexed series according to the base year 1973 (i.e. 1973=100). While the latter does not alter regression results (since they are based on relative changes), it does simplify comparison among the variables and countries. Table 1 contains descriptive statistics of our data. Correlation tables can be found in appendix C and the relationship between changes in REER and changes in agricultural import and export volumes is demonstrated in appendix E for each country separately. Appendix B includes definitions and sources of the data.
Perhaps reflecting a globalization pattern for developing countries, agricultural trade volumes increased substantially over the last three decades. Table 1 shows that the average total growth over 38 years of agricultural import and export quantities is respectively 643% and 542%. Growth in Asian countries exceeded growth of Latin American countries both in import and export volumes. One outlier was present where we accounted for: Export volumes of Chile have grown according to original numbers in total by 6000% between 1973 and 2010. When we take a closer look, we find that over the first two years (1973-1975) these volumes increased by 350%, which is an unrealistic growth in the absence of any external events. However, a great event was in place. In May 1973, the socialist president Salvador Allende was overthrown in the so-called ‘Chilean coup d’état’. Prior to the coup, attempts of Allende to re-structure the nation’s economy led to soaring inflation and food shortages (BBC
News, ‘1973: President overthrown in Chilean coup’, n.d.). A prolonged strike and economic
Table 1. Descriptive statistics, 1973=100
Import volume REER Relative import price
Mean Median Std. dev. Min. Max. Mean Median Std. dev. Min. Max. Mean Median Std. dev. Min. Max.
Latin America 834,7a 148,3 100,1 Brazil 236,7 234,7 102,2 66,7 397,2 89,7 85,2 26,0 57,4 175,8 93,3 91,0 25,8 57 162,4 Chileb 202,5 150,0 140,0 40,7 592,6 129,5 125,1 22,4 98,9 176,6 83,4 77,6 20,3 55,9 154,4 Colombia 518,9 315,0 408,3 70,0 1340,0 89,5 81,9 20,0 63,6 135,0 79,5 62,1 32,9 46,8 173,7 Ecuador 492,1 335,0 357,2 100,0 1350,0 100,8 101,6 24,4 59,9 143,8 95,2 88,2 30,4 45,2 172,3 Peru 518,4 483,3 260,5 100,0 1116,7 143,5 166,4 39,6 76,8 204,4 69,0 61,5 20,6 47,4 131,7 Venezuela 236,1 237,2 71,7 100,0 417,9 103,7 103,5 32,7 59,4 216,0 102,2 92,7 42,8 53,7 231,1 Asia 1050 83,8 125,2 India 131,8 107,3 87,3 24,4 380,5 58,9 44,9 20,7 39,3 100,0 149,1 135,8 42,4 98,4 277,2 Indonesia 265,2 193,2 170,6 59,1 613,6 68,2 57,8 23,2 34,1 112,6 103,9 90,0 32,9 68,1 219,1 Malaysia 363,2 325,0 216,9 92,9 778,6 77,1 73,3 13,1 60,1 107,3 87,6 74,7 27,0 59 153,8 Pakistan 203,8 206,5 107,5 54,8 454,8 89,3 79,6 18,5 70,0 129,1 171,3 149,1 61,7 95,1 311,4 Philippines 348,8 287,5 224,1 93,8 287,5 111,5 110,1 12,0 88,8 135,7 89,0 79,8 27,4 60 179,9 Thailand 1152,0 812,5 1053,4 75,0 3625,0 100,8 96,9 16,9 70,2 136,7 69,7 54,8 33,2 36,7 178,2
Export volume Income Relative export price
Mean Median Std. dev. Min. Max. Mean Median Std. dev. Min. Max. Mean Median Std. dev. Min. Max.
World 127,9 125,3 21,0 98,0 164,8 World 164,0 Latin America 572,3 177,2 111,6 Brazil 362,1 219,2 273,5 100,0 1015,4 134,7 134,2 17,1 100,0 179,3 62,2 53,8 22,8 32,6 105,6 Chileb 598,8 511,1 398,6 100,0 1355,6 174,6 159,3 69,1 88,0 300,0 118,3 115,7 30,1 68,2 212,6 Colombia 282,6 28,1 114,0 100,0 452,2 139,6 140,3 25,1 100,0 191,5 69,4 52,8 37,6 28,7 179,5 Ecuador 205,5 180,4 119,0 64,3 421,4 124,1 121,2 13,2 100,0 157,1 135,6 109,6 51,8 78,4 266,2 Peru 82,8 64,3 64,5 17,5 269,8 101,3 100,3 16,0 75,5 147,5 189,5 169,3 58,7 100 370,1 Venezuela 202,3 127,9 174,9 32,8 627,9 85,1 82,8 9,9 63,3 104,5 147,1 130,3 73,6 61,1 367 Asia 652,8 337,5 118,2 India 333,3 226,7 261,9 93,3 1060,0 179,2 151,9 77,2 98,8 377,2 86,6 75,7 35,7 41,8 168,6 Indonesia 381,4 270,8 316,0 100,0 1175,0 235,0 238,0 92,2 100,0 418,4 106,4 99,2 32,2 57,7 192,6 Malaysia 406,5 357,7 244,3 100,0 892,3 223,0 208,2 87,6 100,0 377,9 70,1 58,5 29,2 31,3 132,9 Pakistan 197,0 162,1 108,6 69,0 489,7 157,1 161,4 38,2 100,0 228,9 116,7 106,7 49,8 52,7 269,9 Philippines 112,8 109,5 18,7 81,1 148,6 115,4 112,3 13,5 98,4 151,7 102,2 95,3 30,7 59,6 205,7 Thailand 329,5 337,0 102,7 100,0 508,7 263,7 268,5 120,7 100,0 471,1 113,0 102,9 35,7 72,7 206,4 a Bold numbers following a continent represent the average end index of the variable in that continent
b Chile contains a reduced sample of 36 years (1975-2010), 1975=100
REER’s do not tend to follow one evident pattern. Half of the countries carry a mean below 100, while the other half carries a mean above 100. Over the total time-horizon however, real exchange rates appreciated in Latin America on average by 48.3% while real exchange rates in Asia depreciated by 16.2%. A probable cause for currency depreciations experienced in Asian countries is the financial crisis which struck most Southeast Asian countries in 1996 (The Economist, ‘Ten years on’, 2007). Appendix E demonstrates depreciations of the real exchange rate during that period for Indonesia, Malaysia, Philippines and Thailand.
Real income increased over the sample for each country except Venezuela. While income of Latin American counties increased on average by 77.2%, this number reaches 237.5% for Asian countries. Furthermore, real world income (based on constant local currencies) increased by 64% between 1973 and 2010.
equal for Latin American countries, relative import prices of Asian countries increased by 25.2% on average. Relative agricultural export prices (PX/Pw) increased on average by 11.6% and 18.2% for Latin America and Asia respectively.
While we cannot solve the matter, it should be noted that our examined data is not homogeneous. Both countries and continents operate (partially) with different agricultural products. Imported and exported products therefore differ between countries and as a result the composition of trade volumes and import and export prices vary as well. This problem is called heterogeneity and statistical results are most likely influenced by it.
4.2.1 Stationarity
To avoid spurious regressions, we test whether series are stationary or nonstationary. We therefore apply ADF unit root tests for all series. Results can be found in appendix tables D1 and D2 and provide strong evidence of nonstationarity among the majority of series. The unit root tests for the series in first differences show that each series, apart from export volumes of Peru, is stationary in their first differences. Because the rest of the variables of Peru are stationary after first differencing, we assume export volumes of Peru to be first-difference stationary as well. We therefore conclude that all series contain a single unit root, implying that the variables are nonstationary in their level.
V. Results
Estimation results are shown in tables 2 and 3. Overall, where asterisks represent significance of coefficients to some degree, we find the majority of the estimates to be insignificant. Most noteworthy are estimates of the real exchange rate. Columns 1 and 2 in both tables indicate that trade volumes are not strongly affected by the real exchange rate. Regarding non-lagged exchange rate coefficients in Asia, we do not find any significant estimate which supports the hypothesis of the effect of the real exchange rate. Most coefficients (7/12) carry opposite signs of which the import elasticity to exchange rates of Pakistan (-1.04) is significant. Likewise, non-lagged exchange rates in Latin America carry for half of the countries wrong signs, where for Brazil and Peru these numbers are significant (elasticities of 0.57 and 0.38 respectively). Our results match therefore findings of Rose (1990) and Upadhyaya and Dhakal (1997) who investigated most of the countries in our sample and found mainly insignificant results. Nevertheless, some results point in favour of our hypotheses. From column 1 in table 3 we gather that import elasticities from exchange rates for Chile and Venezuela are positive and highly significant with coefficients of 1.60 and 0.62 respectively.13 These findings imply that in these countries the real exchange rate positively
13
affects agricultural export volumes within one year. For Chile, Bahmani-Oskooee and Ratha (2004) found similar results.
Of course, when we consider previous arguments regarding adjustment lags, these results should not come as a surprise. However, lagged exchange rates neither demonstrate supporting results with respect to our hypotheses. Again, seven out of the twelve elasticities in Asia carry opposite (although insignificant) signs, while for Latin American countries this is the case for five out of the twelve coefficients. Overall, lagged REER’s in import models are more according to expectations than those of export models. While eight elasticities of the import model carry the right sign - opposed to three for the export model – three import elasticities prove to have an expected and significant sign: The Philippines and Thailand carry, on a 10% significance level, coefficients of 0.52 and 0.65 respectively, and the import sensitivity of Colombia reaches 1.28 and is highly significant. Higher sensitivities with imports than exports are not uncommon (section 2.1.1), where authors - for explanations - mainly point to the behaviour of exporters to maintain market share by absorbing losses from exchange rates in their profit margins. Overall, we conclude that findings of column 1 and 2 in both tables do not support the J-curve effect. Because most studies could not detect J-curve effects, these results are not surprising.
Results of domestic real income are more intuitive. While an increase in real income is expected to result in more purchasing power and hence more imports, we expect (also based on previous research) to find positive signs in the third and fourth column of table 2. For Latin American countries this is mostly the case. All income coefficients are positive, while estimates for Ecuador and Venezuela of 2.40 and 1.06 respectively are additionally significant. Estimates of lagged income carry expected signs for four out of the six countries, although none is significant. In Asia half of the estimates carry positive signs. India in particular stands out for Asia. While its coefficient of income is significant and very high (3.72), its lagged coefficient of income is significant and very negative (-4.67). Furthermore, the lagged import elasticity from income of Pakistan is 4.00 and significant on a 10% significance level, indicating that an increase in real income of 1% increases agricultural import quantities by 4% the next year.
Table 2. Coefficient estimates of import quantities. Numbers inside parentheses are the t-ratios. The sign above each variable represents its expected sign.
Country + + + + - -
RE RE(-1) Y Y(-1) PM/PW PMP/PW(-1) Adj R2 F-statistic Latin America Brazil -0.17 0.03 2.80 -2.23 -0.46 0.41 0.15 2.03* (-0.67) (0.12) (2.39)** (-1.33) (-1.50) (1.81) Chile 1.60 -0.13 0.37 1.43 -0.98 -0.26 0.51 6.99*** (3.54)*** (-0.32) (0.40) (1.58) (-4.80)*** (-1.27) Colombia 0.09 1.28 1.66 0.11 -0.73 0.13 0.34 3.96*** (0.22) (2.96)*** (0.96) (0.06) (-3.18)*** (0.69) Ecuador -0.17 0.22 2.40 0.40 -0.39 -0.12 0.15 2.05* (-0.63) (0.79) (2.04)* (0.34) (-2.46)** (-0.77) Peru -0.37 0.14 0.23 0.56 0.22 -0.11 0.00 1.03 (-1.46) (0.53) (0.35) (0.87) (0.94) (-0.45) Venezuela 0.62 0.11 1.06 -0.40 -0.06 -0.20 0.32 3.80*** (3.75)*** (0.61) (2.23)** (-0.83) (-0.35) (-1.30) Asia India 0.88 -0.34 3.72 -4.67 -0.96 0.05 0.47 6.19*** (1.20) (-0.45) (2.15)** (-2.75)** (-4.04)*** (0.20) Indonesia 0.23 0.22 0.70 -0.84 -0.57 -0.06 0.34 4.05*** (1.35) (1.02) (0.76) (-0.96) (-4.06)*** (-0.43) Malaysia 0.12 0.03 0.65 -0.25 -0.17 -0.10 0.28 1.85 (0.57) (0.12) (1.52) (-0.56) (-1.49) (-1.02) Pakistan -1.04 -0.20 -1.21 4.00 -0.45 0.41 0.50 6.80*** (-2.33)** (-0.36) (-0.59) (1.96)* (-2.57)** (2.79)*** Philippines -0.12 0.52 -0.40 1.26 -0.20 0.04 0.06 1.35 (-0.43) (1.99)* (-0.47) (1.44) (-1.31) (0.26) Thailand -0.02 0.65 -0.06 0.86 -0.66 -0.18 0.44 5.54*** (-0.09) (1.90)* (-0.07) (1.19) (-4.86)*** (-1.36) Note: *Significant at the 10% level.
Table 3. Coefficient estimates of export quantities. Numbers inside parentheses are the t-ratios. The sign above each variable represents its expected sign.
Country - - + + - -
RE RE(-1) Y* Y*(-1) PX/PW PX/PW(-1) Adj R^2 F-statistic Latin America Brazil 0.57 0.30 -3.98 1.79 -0.64 0.04 0.16 2.11* (2.40)** (1.63) (-1.89)* (1.03) (-2.53)** (0.85) Chile -0.04 0.08 -0.70 1.99 -0.25 -0.16 0.20 2.40* -0.26 0.54 -0.74 (2.06)** (-2.51)** (-2.26)** Colombia 0.07 -0.10 -2.35 2.00 -0.12 -0.03 0.00 0.99 0.29 -0.42 (-1.80)* (1.46) (-1.35) (-0.30) Ecuador -0.01 0.05 -1.16 4.76 -0.07 0.10 0.11 1.70 (-0.06) (0.32) (-0.77) (3.04)*** (-0.53) (0.75) Peru 0.38 -0.04 -2.45 2.14 -0.69 -0.03 0.14 1.94 (1.94)* (-0.13) (-0.86) (0.78) (-1.85)* (-0.14) Venezuela -1.1 0.05 -3.31 6.17 -0.39 -0.03 0.26 3.02** (-2.96) (0.10) (-0.88) (2.17)** (-1.86)* (-0.18) Asia India 0.10 0.68 -3.00 3.04 -0.82 0.08 0.57 8.86*** (0.25) (1.84) (-1.55) (1.52) (-6.74)*** (0.62) Indonesia 0.06 0.18 1.54 1.59 -0.32 0.06 0.12 1.81 (0.48) (1.42) (1.19) (1.24) (-2.80)*** (0.54) Malaysia 0.12 -0.09 0.62 -0.15 -0.20 0.06 0.13 1.85 (0.74) (-0.61) (0.88) (-0.20) (-3.43)*** (0.95) Pakistan 0.60 0.29 -0.63 1.49 -0.27 -0.05 -0.05 0.70 (1.02) (0.52) (-0.19) (0.42) (-1.66) (-0.31) Philippines -0.07 0.12 -0.38 1.46 -0.56 -0.03 0.26 3.10** (-0.29) (0.50) (-0.24) (0.94) (-4.22)*** (-0.25) Thailand -0.08 0.19 0.8 1.89 -0.57 0.38 0.31 3.67*** (-0.31) (0.76) (0.54) (1.26) (-3.53)*** (2.39)** Note: *Significant at the 10% level.
Contrary to findings of other variables, the majority of both lagged and non-lagged relative price coefficients point significantly in the right direction. In both continents and for both models signs are correct (except for the import model of Peru) and for sixteen out of the twenty four estimations the coefficients – which do not exceed unity - are significant. These results therefore indicate the strong impact of relative export and import prices on agricultural trade. Conversely, from column 6 in both tables it can be seen that the lagged effect of relative prices is not strong at all. Only thirteen estimates carry the correct sign of which only Chilean relative export price is significant but low (-0.16). Moreover, relative import prices of Pakistan and relative export prices of Thailand carry significant signs in the wrong direction with estimates of 0.41 and 0.38 respectively. In summary, agricultural trade volumes evidently depend on relative prices while there is no support for the effect of lagged relative import and export prices. These results are exactly in contrast with results of Bahmani-Oskooee (1986) who finds trade volumes to respond to relative prices mostly between four to eight quarters. However, his sample covered the years 1973-1980. Over the last three decades technology within agriculture has made great improvements (Barrett, Carter and Timmer, 2010). In addition, developing countries developed substantially and world markets got more integrated. These factors have potentially contributed to the ability of exporters and importers to respond faster to price changes.
In conclusion, results indicate that an adjustment lag of one year – which we expected to result for both import and export quantities irrespective of the explanatory variable – lacks support for the vast majority of coefficients. In the long-run a depreciation or appreciation may still affect the agricultural trade balance, but a clear effect after one year is not evident. Since an adjustment lag was particularly present in studies of Bahmani-Oskooee (1986) and Doroodian et al. (1999), an adjustment lag may be dated considering agricultural and world developments over the last three decades. Second, we conclude that for both immediate and lagged real exchange rates we reject the hypotheses (section 3.3) apart from a few exceptions. The aggregation bias stated by Bahmani-Oskooee and Goswami (2004) is most likely the cause for insignificant exchange rate effects. Thirdly, findings of import models point more in the direction of our expectations than those of export models. For an explanation we content (like Haynes et al., 1986 and Smith, 2009) that exporters absorb changes in exchange rates in their profit margin to maintain market share. Moreover, the goodness-of-fit (R2) indicate that the import models have a higher explanatory power than export models and F-statistics for general significance (last column of tables 2 and 3) are also higher for imports. Finally results indicate that - more than any other variable - relative prices significantly affect trade volumes within one year, indicating that imports and exports of agricultural products clearly react to price changes. This result therefore might represent the highly competitive market characteristic of agriculture due to its homogeneous products.
VI. Conclusion
In assessing the impact of exchange rates on trade, researchers have followed different paths. Most authors examine the effect of exchange rates on trade balances rather than trade quantities, and base their study mainly on theories of the Marshall-Lerner condition and the J-curve effect. However, previous results are far from conclusive and aggregation issues are mainly pointed out as fundamental underlying problems. On the one hand aggregation of sectors can lead to insignificant results due to different adjustment speeds of trade volumes among sectors, while on the other hand a country’s aggregate exchange rates and trade balances to its collective trading partners can cause biased results (Bahmani-Oskooee and Goswami, 2004). While authors furthermore examine either developing countries or developed countries, results in that respect do not seem to vary.
In this paper we empirically examined the dynamic effect of exchange rates on agricultural trade volumes. Prior to testing, we conducted ADF-tests for stationarity of the series and by taking first differences we ascertained the use of stationary I(1) variables to avoid spurious regressions. On account of theory of the J-curve effect, arguments of Doroodian et al. (1999) and Van Marrewijk (2007, p. 489), and results of empirical papers (Doroodian et al., 1999; Smith, 2009; Huchet-Bourdon and Korinek, 2011) we imposed a one year lag on each explanatory variable to account for an adjustment lag with respect to trade volumes. Real income and relative import and export prices were employed as control variables following Bahmani-Oskooee (1986) and Fidan (2006). After we ensured most of the assumptions of the regression to hold, we conducted a regression analysis utilizing the Ordinary Least Squares (OLS) – method.
Findings overall did not meet our expectations. Support for adjustment lags (based on coefficients of lagged variables) could not be detected and therefore support for any J-curve effect is lacking. With respect to real exchange rates we mainly reject the hypotheses since half of the trade elasticities from exchange rates carry opposite signs. Nevertheless, import elasticities of Chile and Venezuela and the lagged import elasticity of Colombia were in favour of hypotheses and highly significant, indicating the relation might still hold for particularly import models. Moreover, coefficients of import models were more in line with our expectations and had more explanatory power than export models. Findings furthermore indicate that relative import and export prices significantly affect trade volumes, while their lagged counterparts do not. These results are however in exact contrast with earlier findings of Bahmani-Oskooee (1986) who covered a time-span up to 1980. Considering developments over the last three decades, his results and the adjustment lag as such might be dated.
necessitates estimations of trade elasticities on a bilateral basis. Since bilateral import and export prices are not available, an examination of trade volumes is not possible on a bilateral basis.
REFERENCES
Aristotelous, K. (2001), “Exchange-rate Volatility, Exchange-rate Regime, and Trade Volume: Evidence from the UK-US Export Function (1889-1999),” Economics
Letters 72:1, 87-94.
Arize, A.C., T. Osang, and D.J. Slottje (2008), “Exchange Rate Volatility in Latin America and Its Impact on Foreign Trade,” International Review of Economics and Finance 17:1, 33-44.
Bahmani-Oskooee, M. (1986), “Determinants of International Trade Flows,” Journal of
Development Economics: The Case of Developing Countries 20:1, 107-123.
Bahmani-Oskooee, M. and Z. Ardalani (2006), “Exchange Rate Sensitivity of U.S. Trade Flows: Evidence from Industry Data,” Southern Economic Journal 72:3, 542-559.
Bahmani-Oskooee, M. and T.J. Brooks (1999), “Bilateral J-Curve Between US and Her Trading Partners,” Weltwirtschaftliches Archives 135, 156-165.
Bahmani-Oskooee, M. and G.G. Goswami (2004), “Exchange Rate Sensitivity of Japan’s Bilateral Trade Flows,” Japan and the World Economy 16:1, 1-15.
Bahmani-Oskooee, M. and F. Niroomand (1998), “Long-Run Price Elasticities and the Marshall-Lerner Condition Revisited,” Economics Letters 61:1, 101-109.
Bahmani-Oskooee, M. and A. Ratha (2004), “Dynamics of the U.S. Trade with Developing Countries,” Journal of Developing Areas 37:2, 1-11.
Barett, C.B., M.R. Carter, and C.P. Timmer (2010), “A Century-Long Perspective on Agricultural Development,” American Journal of Agricultural Economics 92:2, 447-468.
BBC News (n.d.), “1973: President Overthrown in Chile Coup” [Online], Available: http://news.bbc.co.uk/onthisday/hi/dates/stories/september/11/newsid_3199000/3199155.stm (Accessed 5th April 2013).
Bini-Smaghi, L. (1991), “Exchange Rate Variability and Trade: Why Is It So Difficult to Find Any Empirical Relationship,” Applied Economics 23:5, 927-936.
Broda C. and J. Romalis (2004), “Identifying the Relationship between Trade and Exchange Rate Volatility,” NBER Working Paper 20:6, 79-110.
