An investigation of a partial Dutch disease in Botswana

Abstract Diversification of resource-driven economies has proven to be a very stubborn problem. Even relatively successful countries struggle to achieve a structural change towards manufacturing and high-tech industries. Botswana has been often cited as one of the few countries that escaped the resource curse and performed well in terms of economic growth. However, a significant share of the domestic output and most of the exports are still coming from the mining sector. In this paper we present the spending effect as a possible explanation for the lack of economic diversity. Using a nonlinear autoregressive distributed lag model, we investigate the cointegration of the diamond price index and the pula exchange rate against the currencies of Botswana's main trading partners on monthly time-series data. Our analysis is based on a recent dataset that covers the period from the introduction of the crawling band exchange regime until 2018. The results highlight a partial Dutch disease phenomenon related to Botswana's trade union partners: Namibia and South Africa.


Introduction
Many countries with significant natural wealth suffer from what is labelled as the resource curse (Sachs and Warner, 2001). Despite the long and intense academic debate concerning the effects of resource endowment on economic and social development, results are still controversial (Papyrakis, 2016;Szalai, 2018). Botswana is often considered as one of the few success stories as it has maintained a relatively steady economic growth and political stability since the beginning of diamond mining, for almost five decades. Although the country avoided most negative effects of its booming natural resource sector, it has clearly failed to diversify from diamond extraction and achieve a structural change towards higher added value activities. Around 20% of the GDP and almost 85% of the exports are generated by the diamond industry, while the country ranks relatively low in terms of economic complexity and struggles with significant unemployment (see Table 1). Put differently, Botswana is currently being "trapped by diamonds" (Hillbom and Bolt, 2018). Since easily accessible reserves will have been depleted in a few years, dependence causes serious concerns (Grynberg et al., 2015).
A cross-country study by Harding and Venables (2016) found that every dollar of resource revenue decreases non-resource exports by 74 cents and increases imports by 23 cents. According to the Dutch disease theory (Corden and Neary, 1982), the booming resource sector crowds-out manufacturing, drives the domestic price level upwards, and catalyses the Balassa-Samuelson effect. That is, the change in the domestic prices is followed by an equal real appreciation of the currency which causes a significant and persistent misalignment of the exchange rate.
Previous studies on the misalignment of the Botswana pula (BWP) have the common shortcoming of using the real effective exchange rate to determine if the currency is overvalued (Del� echat and Gaertner, 2008;Taye, 2012), i.e. if the Dutch disease hits the economy. As the proxy is a weighted average, these results necessarily lack the currency-specific details. In the case of Botswana, the lack of economic diversity follows from a loss of international competitiveness in manufacturing due to a partial Dutch disease. By partial we mean that the effects are limited to specific trade partners and transmission channels.
In order to test our hypothesis (H 1 ), we utilize non-linear autoregressive distributed lag models to separately reveal if there is a long run stochastic co-movement between diamond prices and real exchange rates of Botswana's main trade partners. Our results suggest that diamond exports cause significant competitive disadvantage in manufacturing within the South African Customs Union (SACU) by driving the real appreciation of the BWP against the South African rand (ZAR) and the Namibian dollar (NAD). 2 Moreover, the specific tariff revenue sharing formula within the SACU probably further amplifies the forces limiting non-resource exports.
Additionally we test if the long run effects are asymmetric (H 2 ), that is, a positive price shock of unity causes the BWP to appreciate more than a negative shock of the same amplitude would cause it to depreciate. This asymmetry could further amplify the partial Dutch disease effect even when resource price movements are less volatile. However, based on our calculations, asymmetries are not present in Botswana's trade relations.
This paper is organized into five sections. Section 2 gives an overview on both the theoretical background and the country-specific issues, while section 3 discusses the statistical properties of the selected time series. In section 4 and 5 we present our methodology and evaluate the results. Section 6 concludes.

Overview
The classic Dutch disease theory concerns a three-sector small open economy and suggests that a resource boom causes both direct and indirect de-industrialization (Corden and Neary, 1982;Yokoyama, 1989). As the value of the marginal product in resource extraction rises, workers start to flow from other sectors in hope of a higher real wage. This resource movement effect causes direct de-industrialization as manufacturing output falls due to lower employment. Increased wages provide more disposable income and boost the demand for domestic services along the income-consumption curve. The excess demand drives flexible prices upwards and raises the value of the marginal product in the non-tradable sector. Consequently, more workers move away from manufacturing which leads to a further decline in output. This secondary impact, called the spending effect, causes indirect de-industrialization in the long run. Put differently, the Dutch disease develops as a resource bonanza: it crowds-out manufacturing by altering the relative returns on various economic activities.
Another consequence of the spending effect is the appreciation of the domestic currency that is likely to further impede manufacturing by causing a loss in international competitiveness. This "chronic exchange rate overvaluation" (Bresser-Pereira, 2013, p. 372) was verified by numerous studies under several different assumptions and model specifications. Beyond the original framework, Krugman (1987) adapted increasing returns to scale and showed that resource windfalls cause further specialization towards extraction, while Sachs and Warner (1995) developed an endogenous growth model where the relative price of non-tradables grows as excess revenue drives the domestic demand. According to the Balassa-Samuelson theorem, the appreciation of the currency ensues, providing theoretical support for our analysis. In a recent paper Iacono (2018) extends the framework by incorporating the concepts of both learning-by-doing and absorption constraints, but draws the same conclusion: resource booms cause currency overvaluation. On the other hand, some models suggest that appreciation is not inevitable. Torvik (2002) finds that if learning spillovers are present in the non-tradable sector as well, then the exchange rate is likely to depreciate in the long run because of a relative increase in sector productivity. Similarly, van der Ploeg and Venables (2013) argue that appreciation only happens in the short run as adjustments take time due to the intersectoral immobility of capital. However, all theoretical models agree that if the currency appreciates, the loss in international competitiveness eventually leads to indirect de-industrialization.
The idea that the spending effect creates disadvantage in manufacturing is also supported by a growing body of empirical evidences. Mironov and Petronevich (2015) used a cointegration model to examine the structural development of the Russian economy and reported that a 1% increase in resource export revenues had caused the real effective exchange rate to appreciate by 0.2%. In accordance with the theoretical findings, they also emphasize the negative impact on manufacturing. A similar study about Australia by Koitsiwe and Adachi (2015) reveals a causal link between the increasing share of mining, the appreciation of the currency, and eventually the decrease in manufacturing output. Based on a vector autoregressive model, they conclude that 23.6% of the exchange rate fluctuation is explained by mining booms while 24.8% of the variation in manufacturing is explained by the exchange rate. In the case of Canada, Beine et al. (2012) find that currency misalignment due to the Dutch disease induced a loss of 33-39% in manufacturing employment between 2002 and 2007. However, they also note that the Canadian dollar depreciated shortly after as oil prices dropped sharply during 2009.
To contribute to the mounting body of literature, we deliberately chose a country that is highly dependent on a single resource but escapes most negative effects of resource dependence. In terms of value and volume, Botswana is the second most important diamond-producer in the world, providing around 15% of the world supply (Kimberley Process, 2017). Unlike most Sub-Saharan African economies struggling with several aspects of the resource curse, diamonds have brought about a growth miracle in Botswana. They transformed the poor, agrarian, rural, uneducated, foreign-aid dependent country (Samatar, 1999;Gwebu, 2012;Maipose, 2008) into an upper-middle income economy in the course of a generation (Hillbom, 2008). Corruption rates and societal tensions are relatively low whereas state capacity and developmental institutions are sound and strong (Robinson et al., 2003;Leith, 1999;Robinson and Parsons, 2006;Mulinge, 2008;Acemoglu and Robinson, 2010).
However, persistently high unemployment rates coupled with high production costs and competitiveness problems as well as the low level of export diversification (Jefferis, 2019, see Table 1) imply deep-seated structural transformation problems that might be related to the presence of the Dutch disease.
Tswana decision-makers used a multitude of micro-and macro-level policies from the beginning of the mining boom to counter competitiveness-related diversification problems (MITI, 2016, interview). On the macro-level, prudent exchange rate management impeded significant and permanent nominal exchange rate appreciation. 2 Please note that the NAD is pegged to the ZAR at a fixed nominal rate of 1:1.
Government spending was kept below government revenue growth, surpluses were accrued, and assets were invested abroad through a sovereign wealth fund (Pegg, 2010). The government also promoted tradable sectors by establishing various subsidy schemes for arable agriculture, employment grants for agricultural industries, manufacturing and small-scale mining, and by conceiving coherent industrial policies in 1984, 1998respectively (MITI, 2014. Above all, the government "earmarked" the manufacturing sector "through which the twin goals of economic diversification and employment creation" (Sekwati, 2010, p. 13) would be realized.
In some respects, government policies were successful. GDP and sources of growth have become more diversified (Jefferis, 2014). The relocation of the diamond sorting and aggregation centre of the De Beers group 3 from London to Gaborone in 2012 is considered as a great leap forward in local value creation (Kufa, 2016, interview). Indeed, the new processing capabilities have led to an expansion in external trade with other diamond producers (see Namibia in Table 2) and integrated Botswana one step further in the value chain.
The manufacturing sector has expanded parallel to the growing GDP but failed to grow its share within the GDP. It has not become a driver of diversification and "has consistently failed to perform to expectations" (Sekwati, 2010, p. 9). Exports are heavily dominated by diamonds (see Table 1).
Diversification-enhancing policies have only had limited success so far. This is concerning because the depletion of diamonds is forthcoming: a permanent and rapid production decline is expected to take place between 2025 and 2027, possibly putting an end to this African success story (Grynberg et al., 2015). According to a Gaborone-based think-tank, ten years after the depletion of easily accessible stocks Botswana GDP might be "47% below the non-depletion path" (Grynberg et al., 2015, p. 124).
An important reason for the failure of diversification is related to the unbalanced power relations in the region favouring the Republic of South Africa. Southern Africa's most important regional economic alliancethe South African Customs Union (SACU) -binds Botswana, Lesotho, Namibia, Swaziland and South Africa together. Referred to as the "industrial powerhouse" (Martin, 1990, p. 59), South Africa contributes more than 90% of SACU GDP and manufacturing output (McCarthy, 2014), and aggressively pursues its own interests of industrialization. This is reflected in the current revenue-sharing formula (RSF) of SACU: South Africa withholds a fixed share of customs revenue and the remaining customs component is shared according to intra-SACU imports (Grynberg and Motswapong, 2012). As Table 2 shows, Botswana has negligible trade relations with smaller member states and imports mostly from South Africa. The more Botswana imports from other members within the union, the more it harvests from the SACU revenues. As in Botswana these incomes surmount even mineral revenues (see Jefferis, 2014, p. 9), the country is heavily dependent on this arrangement and sticks to it (Grynberg, 2016, interview) although when it comes to diversification and building a viable domestic manufacturing sector, it is clearly counter-productive.
In this paper, we intend to highlight that the failure of diversification is also related to the presence of Dutch disease in Botswana. The description of the main symptoms, such as low rates of manufacturing, overvalued currency, artificially high real wages, and unemployment (Bresser-Pereira, 2013) seem to fit, but the literature is far from consensus. Norberg and Blomstrom (1993) do not find significant evidence to corroborate Dutch disease effects in Botswana. Mogotsi (2002) analyses the 1982-1990 boom period and concludes that Botswana did experience a mild form of Dutch disease that was manifested in the form of real exchange rate appreciation which negatively affected the textile industry. Love (1994) and Molaodi (2004Molaodi ( , 2005 both assert that the country was hit by the Dutch disease, the former based on the contraction of agriculture during the mining boom and the latter on the unbalanced non-tradeable sector growth (construction, commerce) up to 2004. Pegg (2010) contends that Botswana suffers from negative effects related to its natural resource dependence, but not through the traditional channels posited in the Dutch disease theory. According to Poteete and Marroquín (2005, p. 23), Botswana keeps Dutch disease symptoms "in check", therefore a clear diagnosis cannot be established.
Another approach is to identify possible BWP exchange rate misalignment periods. A joint Botswana Institute for Development Policy Analysis and WTO study (BIDPA and WTO, 2005) assesses the BWP real effective exchange rate from the 1970s until 2004 and only finds proof of real appreciation between 2002 and 2004. Iimi (2006Iimi ( , 2007 uses a behavioral equilibrium exchange rate approach and finds that BWP seems to have been undervalued in the late 1980s and overvalued by 5-10 percent from 2002 to 2005. In a more recent analysis Taye (2012) compares effective exchange rate data from 1990 to 2010 against an estimated equilibrium exchange rate over the same period and finds no evidence of long-term misalignment. Taye also concludes that the crawling peg system introduced in 2005 works adequately. Most authors agree that, notwithstanding its spectacular development, the country suffered from BWP exchange rate misalignment or Dutch disease during a certain period of its post-independence economic history, but only in a "mild form".
We have already referred to the crowding-out effect as a consequence of the Dutch disease in terms of manufacturing. In the case of Botswana, we assume that crowding-out is present (Rigler, 2019, interview) and it stems from the spending effect. We rule out the resource movement effect since (i) diamond extraction is a capital-intensive industry and (ii) the unemployment rate is traditionally high in Botswana (see Table 1). Thus, it is very unlikely that any diamond boom would divert further workers from manufacturing. On the other hand, we argue that the spending effect is significant in terms of international competitiveness as the relative increase in the price of non-tradable services would cause an equal appreciation of the domestic currency (Balassa, 1964). In order to test our theory, we examine if diamond world prices are cointegrated with the pairwise BWP exchange rates.

Data and time series properties
In this section we introduce our datasets and determine the time frame of our analysis. Later, we carry out the necessary transformations and run statistical tests to prepare for the modelling phase. Note: Data was collected from WTO's WITS database. All values represent the average percentage import/export from selected SACU member country, compared to total imports/exports of the indicated time interval.

Data sources and time frame
We selected Botswana's ten most considerable trade partners in the diamond market including Canada, China, India, Israel, Namibia, South Africa, Switzerland, the United Arab Emirates, the United Kingdom and the USA. Monthly financial statistics were gathered from the International Monetary Fund's IFS database for all entries. These time series include nominal exchange rates between the BWP and the selected trade partners as well as the consumer price indices of the same countries. Using the currencies of the appropriate pairs from the time series, we have calculated the real exchange rates by the following formula 4 : where e t and E t denote real and nominal exchange rates respectively, whereas CPI t and CPI t * are domestic and trade partner inflation rates at time t as measured by the consumer price index. A rise in e t indicates the appreciation of the BWP and thus a loss of export competitiveness. Our indicator of the monthly price change of diamonds is the Bloomberg's Overall Diamond Index (PLPH0AAI), which weighs different diamond products by several important characteristics (e.g.: shape, carat, refraction, etc.). Although rough price indices would serve our purposes better, the available Rough Diamond Index (PLPHROAI) has a high amount of unsystematic missing observations, therefore inappropriate for statistical modelling.
In the models, p t denotes the real price of the US dollar denominated Overall Diamond Index that is deflated with the US CPI rate which -as usual in empirical studies (Jahan-Parvar and Mohammadi, 2011)represents the world price index. We follow the literature on exchange rate determination in Dutch disease affected countries, and after adjusting for seasonal effects, we use the logarithm of both variables to mitigate potential heteroscedasticity (Olusi and Olagunju, 2005;Yol, 2009).
Although data in general are widely available for all time series (see Table 3), we still encountered two problems. First, the Bank of Botswana, as the monetary authority of the country changed the exchange rate regime during May 2005. The former pegged regime was abandoned for a more liberal crawling band mechanism. Based on Bank of Botswana's statistics, we believe that exchange rate fluctuations caused by the regime change diminished until the beginning of 2006. Second, it is only possible to analyze the statistical relationship be- Our data coverage spans until May 2018, but in some cases the time series are shorter due to temporarily unavailable financial reports. Namibia and India have one and two missing months respectively from the end of the series, while the United Arab Emirates had not announced the exchange rate statistics since the end of 2017. In general, the number of observations are enough to conduct single equation modeling, excluding Canada as a borderline case.

Time series properties
A long-run relationship between stochastic time series -i.e. a partial Dutch disease -can only occur if either (i) both variables are stationary or (ii) although at least one of them is non-stationary, their linear combination satisfies a required mean reverting attribute, thus they are cointegrated (Engle and Granger, 1987). In any other cases the model would be dynamically unstable, and following a shock there would be no equilibrium level to which the system could revert.
For this reason we first conducted unit root tests and verified if both time series are stationary. Table 4 reports the results of the augmented Dickey Fuller -and Phillips-Perron tests. The main difference is that the latter is more robust to heteroscedasticity and serial correlation, while it also performs better in border-line cases (Phillips and Perron, 1986). Both procedures were calculated with a drift term, and with an additional trend to account for potential world level inflation effects in the real exchange rate series. In all ADF cases the optimal lag length criterion was determined by AIC, which uniformly suggested a lag structure of one. Contrary to the findings of the ADF tests, the PP procedure identified an optimal lag structure of four, except for the case of Canada where the length was three, possibly because of the shorter sample period. All calculated test statistics are concordantly unable to reject the  null hypothesis of non-stationarity at a 5% significance level, therefore our results drive out the possibility of long-run relationship in levels. If a partial Dutch disease exists in specific trade relations, only cointegration tests can reveal their presence. Currently popular cointegration testing procedures originated from Johansen (1991), but his method is based on multiple equation models. Equation system techniques consume a high amount of degree of freedom and are not suitable for our analysis. Technically the Johansen procedure with two integrated time series can only result in a maximum of one cointegrating relation. Therefore, we have only tested our time series against the null hypothesis of no cointegration (i.e. H 0 : r ¼ 0). Rejecting the null indicates a common linear stochastic trend for the variables in question. Pesaran and Shin (1999) suggested a single equation autoregressive distributed lag approach to cointegration testing when the time series are moderately short. The so called bounds testing approach follows a different method. It includes the long-run variables in the model, and evaluates their joint significance by a modified F-test. The critical values of the Pesaran approach are tabulated by the authors in their seminal paper (Pesaran et al., 2001) and expanded to non-linear models by Shin et al. (2014). The test separates the possible values of the F-statistic into three intervals. A test result smaller than the lower bound critical value indicates that the underlying time series are both I(0) and there is no long run common stochastic trend. On the contrary, results exceeding the upper bound indicate two I(1) cointegrated processes. Between the bounds the test is inconclusive and a potential mixed cointegration has to be evaluated based on the unit root tests.
To identify possible cointegrating relationships in practice, we have calculated Johansen's trace and eigenvalue tests (Johansen, 1991) for linear cointegrating relationships. For a possible non-linear hidden cointegration the Pesaran and Shin (1999) bounds test was utilized. The significance level was 10% in both cases. Being aware of the sensitivity of the Johansen procedures in small samples, we have given a decisive role to the Pesaran bounds test in every ambiguous case.
Out of ten trade partners five, namely, Canada, Switzerland, the United Arab Emirates, the United Kingdom and the United States of America, showed no signs of cointegration based on the bounds tests (see Table 5). Among these, the United Arab Emirates and the United Kingdom might be cointegrated according to both Johansen procedures. The confusing outcome is not uncommon as Shin et al. (2014) recalls that these tests often result in contradictory outputs.
China, India, and Israel fall in the inconclusive range, but based on the fact that the ADF and the PP tests were not able to reject the null of non-stationarity, we assume that there is no cointegration among the time series in the separate models.
There are two series where the bounds test denotes cointegration: Namibia and South Africa. Namibia's test outcomes give an F-test value of 11.86, far beyond the 3.23 asymptotic upper bound. The trace and eigenvalue test results are also in accordance with this finding. On the other hand, South Africa barely passes the upper bound with a 3.24 Fstatistic, while the results of the Johansen procedures contradict this.

Methodology
According to Jahan-Parvar and Mohammadi (2011), identifying the Dutch disease through the spending effect essentially requires a stable and long-run cointegrating relationship between particular real exchange rates and the diamond price index. In section 3 we showed that diamond prices are pairwise cointegrated with the exchange rates of Botswana's two main trade partners: Namibia and South Africa.
Besides the cointegrating relations, we would also like to examine the asymmetric nature of positive and negative logp t price shocks separately on the related loge t real exchange rates. Therefore, following Shin et al. (2014) we deconstruct the I(1) explanatory time series into partial sum processes the following way: After the transformation, we can estimate the asymmetric long-run relationship: but trends with similar slopes in the time series could lead to spurious regression and biased parameter estimates (Granger and Newbold, 1974;Hendry, 1980). To consistently estimate simultaneous short -and long-run effects, an error correction model has to be specified, where the presence of a long-run stochastic trend is separated from the short-run dynamics. Originally Engle and Granger (1987) extracted the stationary ξ t error term from the symmetric version of equation (4): and included it in their model as an explanatory variable. By this specification, the term ξ t represents deviations from the long-run stochastic equilibrium of the variables and it accounts for the missing effect. Alternatively Shin et al. (2014) estimate a nonlinear autoregressive distributed lag type error correction model with the same result in two steps: where equation (6) is estimated directly, and the original long-run parameters of equation (5) can be calculated from the model's output as follows: β þ ¼ À θ þ /ρ and β À ¼ À θ À /ρ. In equation (7), ρ represents the coefficient of the error correction term. If ρ < 0 and significant, then the model is dynamically stable and represents a causal relationship.
Additionally, ρ measures the adjustment speed. The value of the estimated parameter indicates the percentage of correction to a shock of unity within one period. In equations (6) and (7), all coefficients of the differenced variables -i.e. γ j ; φ þ = À j -represent the short-run effects of the variables in question, which in terms of our analysis, are of secondary importance.
The goals of this estimation process are the following: 1) Estimate ρ, the coefficient of the error correction term to identify if the Dutch disease is present with a trade partner. 2) Determine the speed of adjustment based on the value of ρ. 3) Estimate the long run coefficients β þ and β À , to verify if the relation is asymmetric. Through the remainder of the paper, we estimate the error correction models between real exchange rates selected in section 3 and diamond prices following the introduced methodology.

Estimation results
The results from section 3 indicate that error correction models should be specified in the cases of Namibia and South Africa. Notwithstanding the fact that the obtained statistical results for South Africa contain some ambiguity, we turn to the modelling of these two countries.
After the scrutiny of the time series properties we have decided to include intercepts and trends into both of our models. Table 6 contains all results of the error correction specifications while the long run asymmetric parameter estimations are also represented.
The value of ρ indicates the presence of cointegration and measures the error correction speed. In both models the values are negative and significant on a 1% significance level, which means that our hypothesis H 1 is supported by the model. Botswana suffers from a partial Dutch disease in trade relations with Namibia and South Africa. Based on Table 2, between 2013 and 2016, approximately 74% of Botswana's imports suffered from constant competitiveness loss.
Moreover, the value of ρ for Namibia is significantly larger, indicating a much stronger attraction of the common stochastic trend. This means that 28% of the deviation from the stochastic equilibrium is corrected in each month, whenever Namibia experiences exchange rate misalignment. The same parameter value for South Africa is significantly less: 12%. The outcome clearly suggests that unsystematic diamond price shocks cause long-lasting misalignment of the ZAR, eventuating competitiveness loss in all non-resource sectors of Botswana.
The β asymmetric long run parameters are calculated using the estimated values of the error correction term and the θ values from the error correction model. Namibia's long run parameters seemingly show a huge difference. A value of 0.52 means that for a 1% appreciation in the exchange rate of BWP, the price index of diamonds has to increase by 1.94%. Although it is a credible rate of response to a price shock, it is suggestive that the β À value is not statistically different from zero even at a 10% significance level. This would mean that there is no chance of exchange rate correction in the long run, which seems an extreme assumption since it would lead to irreversible repercussions in trade relations. Fortunately, turning to model diagnostics, we can observe that the long run symmetry test -with the null hypotheses of long run parameter symmetry -cannot be rejected in our models on any conventional significance level. The p-values are 0.21 and 0.62, respectively. By these results we reject our second hypothesis H 2 , which states that potential long run asymmetries intensify the already deteriorating expectations.
We also run further diagnostic tests to verify our models. The model fit of the Namibia specification is approximately average, while the South Africa model has only a very limited adjusted R 2 of 7.9%. Furthermore, based on the LM test there is no sign of serial correlation in either of the models, but the estimation for South Africa suffers from ARCH effects.
We consider our results as evidence for the cointegration and the consequent Dutch disease phenomena in the aforementioned trade relations, but this effect is symmetric in its long run parameters. The model for Namibia can be considered as a well-defined approximation of the original data generating process, while the model for South Africa is less practicable.

Conclusion
Botswana, an African success story of resource-based development, is surprisingly unsuccessful in diversifying its economy away from diamonds. In this paper, we suggest that diamond price fluctuations from 2006 to 2018 caused real exchange rate appreciation in certain trade relations. Subsequent tradables sector competitiveness problems vis-avis particular trade partners probably played an important role in the failure of recent diversification efforts.
Using separate nonlinear autoregressive distributed lag models, we tested whether there is a long run stochastic co-movement between diamond prices and the real exchange rates of Botswana's main trade partners. We also tested whether long-run effects are asymmetric. We have found no evidence of asymmetry but out of Botswana's ten most important diamond trading partners, we revealed significant real exchange rate effects in the cases of Namibia and South Africa. Since these two countries account for 74% of the overall value of Botswana's imports, and all three countries are members of the SACU, the consequences are far-reaching. Taken the massive reserves in labor force (See Table 1), de-industrialization due to the resource movement effect is likely to be insignificant. Thus, Botswana suffers from a partial Dutch diseaselimited in scope to the spending effect and specific to particular trade relations. The 2012 relocation of De Beers' diamond sorting and aggregation centre to Botswana's capital was a milestone in raising local added value but it also reshuffled diamond commerce and resulted in unintended exchange rate pressure with regards to key trading partners. Our empirical results draw decision-makers' attention to these pressures that must be contemplated when formulating the country's development strategy.
We are convinced that the identified partial Dutch disease is an important factor in the failure of diversification policies. As diamond price booms increase disposable income and boost the demand for domestic services, the change in the price ratio translates to an equal

Note:
The error correction panel shows the model estimations of equation (6), while the asymmetric long run parameters were acquired from equation (4). Both equations were estimated with the R implementation of Zaghdoudi (2018). Among model diagnostics X 2 SC denotes the LM test for serial correlation, and X 2 H the test for residual heteroscedasticity. W LR denotes the long run symmetry test of the long run parameters. Stars indicate parameter significance in the usual manner: *p < 0.10, **p < 0.05, ***p < 0.01. change in the real exchange rate and causes the BWP to appreciate against the NAD and the ZAR. The domestic manufacturing sector is losing international competitiveness and fails to achieve economies of scale. Appreciation-related competitiveness problems are exacerbated by the peculiar revenue-sharing formula of the SACU, rewarding imports from South Africa. These influences make the development of domestic manufacturing in Botswana more difficult. Our methodology, the country-level separation of the exchange rate effects differs from previous efforts to reveal the presence of the Dutch disease in Botswana. Former evidence was less robust as most authors investigated the spending effect on a global level as measured by the real effective exchange rate. We, on the other hand, focused on specific trade relations of importance and concluded that the Dutch disease evolves on a regional level. Beyond the specific case of Botswana, our results suggest that analyzing pairwise foreign exchange effects helps to resolve some controversies in the resource curse literature. Some resource exporter countries have a partial experience of the Dutch disease because the negative impact is limited to specific transmission channels and/or trade relations. We presume that extending research on these partial effects would improve the robustness of the empirical evidence, provide a better understanding of the problem, and yield more efficient policy proposals.
As pointed out at the beginning of this paper, Botswana is considered one of the developmental success stories of the continent. However, due to the depletion of easily accessible diamond reserves in the near future, the sustainability of the country's forty-year development comes into question. Our intention was to contribute to a deeper understanding of the evolution of Botswana economy and arm decision-makers with thorough empirical results when weighing the future directions of economic diversification policies.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.