Heterogeneity in Coffee Price Trends: A Prebisch-Singer Analysis

Heterogeneity in Coffee Price Trends: A

Prebisch-Singer Analysis

Master’s Thesis

MSc Economic Development & Globalization

M. Fauaad A. Dar – S3255166

m.f.a.dar@student.rug.nl

Supervisor: Prof. Dr. Jakob de Haan

Co-assessor: Prof. Dr. Erik Dietzenbacher

2

Abstract

This study investigates the validity of the Prebisch-Singer hypothesis with regards to export coffee prices for each producing country. Using annual data for the

period 1965 – 2018, it is found that not all producing countries follow the same price trend; even within the ones that do adhere to the hypothesis, there too exists a spread across the negative trends. The follow-up examination finds evidence that diurnal

temperature, mean temperature and share in global production can explain some of the variation in behaviour of the coffee price series.

3

Table of Contents

3. Data and Methodology ... 10

4. Estimation Results... 13

5. Conclusion ... 17

References ... 19

4

List of Abbreviations

ADF : Augmented Dickey-Fuller

BN : Brazilian Natural

CRUTS : Climatic Research Unit Time Series

CM : Colombian Mild

DS : Difference Stationary

GYCPI : Grilli and Yang Commodity Price Index

ICO : International Coffee Organisation

MUV : Manufactures Unit Value

OM : Other Mild

PP : Phillips-Perron

PS : Prebisch-Singer

RO : Robusta

5

List of Tables

Table 1: Trend estimates and unit root test results ... 14

Table 2: Determinants of the stationarity of deflated price differentials ... 16

6

1. Introduction

Introduced in Prebisch (1950) and Singer (1950), the Prebisch-Singer (PS)

hypothesis claims that in the long run, the price of primary commodity goods should decline relative to the price of manufactures and for an aggregated commodity index, they found that to be true. They argued that in addition to the low income elasticity

of demand for primary commodities, producers of manufactures were the main beneficiaries of increasing productivity levels in the sector through higher incomes,

while consumers of primary commodities benefitted from the increasing productivity levels through lower prices; a win-win for the manufactures exporting, commodity

importing developed countries. Proceeding from that arose a vast literature applying increasingly more sophisticated methodology to test for the existence of negative

trends, with most recent work shifting towards disaggregating the index into individual commodities to account for divergent behaviour; what these studies find is that different commodities display disparate trends.1 _{(Cuddington, 1992; Ghoshray,}

2011; Grilli & Yang, 1988; Harvey et al., 2010; Kellard & Wohar, 2006; Winkelried, 2018)

Of the various commodities that generally make up the common index used in PS

hypothesis literature, coffee is amongst the most valuable; in the preceding few decades, it has largely been second only to oil with regards to trade volume (Ponte,

2002). Most of the world’s coffee production takes place in developing countries and as such, coffee plays an important role for them as a source of foreign exchange and economic development (Addison et al., 2016). It stands as a principal export for many

and for the likes of Burundi, Ethiopia, Rwanda and Uganda, it accounts for more than 50% of export earnings (Cashin et al., 2000).2 _{Consequentially, a secular decline in}

relative coffee prices can have significant adverse effects on the terms of trade of

1 _{As a caveat, Singer (1991) argues that the PS hypothesis was never meant to apply to individual} commodities.

7

producing countries; Bidarkota & Crucini (2000) find that more than half the annual variation in terms of trade of developing countries can be explained by variation in

the prices of their principal commodity exports. Recent times have seen the coffee industry in consuming markets such as the United States, the European Union and Japan expand at breakneck speeds, with an abundance of differentiated products

being innovated and ever-increasing revenues. Meanwhile producing countries face a coffee price crisis that has had severe social, economic and environmental effects

which could lead to a potentially long-lasting impact on the sustainable development of the local coffee industry (International Coffee Organisation [ICO], 2019).

It follows that there has been extensive analysis done on identifying trends within coffee pricing, which brings with it its own set of empirical issues. The literature has

largely focused on identifying if the generating process for the relative price is trend stationary or if the series is better characterised as a unit root process. In case of the latter, an OLS estimation of an equation in levels would lead to spurious estimates of

a linear trend, finding false evidence for its existence. As such, testing for unit roots remains essential in order to avoid distortionary effects on long run behaviour

inferences (Winkelreid, 2018). Concerning coffee, most studies have been unable to reject the unit root null hypothesis, with some finding support for the PS hypothesis in the form of a negative drift. The standard practice within the literature has been to

use the Grilli & Yang (1988) dataset (in some cases an updated version of it), which uses one single price indicator that is averaged across countries;3 _{however, coffee is a}

distinctly heterogenous product and that translates into large variations in pricing. These price differentials are largely a result of country of origin and quality premiums,

which arise due to differences in chemical composition of the coffee itself stemming from environmental factors and post-harvest processing practices (Otero et al., 2018).

8

This thesis aims to investigate the trends and differentials in coffee pricing, whether the export prices across countries display a degree of heterogeneity and if so, could

that heterogeneity be explained by climatic and market factors? Given the importance of coffee as a factor in economic growth to many a developing nation, detecting patterns in price trends might lead to learning opportunities for any laggards with

regards to increasing the value of their product. The rest of the study is organised as follows. Section 2 provides a brief overview of key literature regarding the PS

hypothesis and differentiation within coffee. Section 3 describes the data used for this study along with the methodology applied. Section 4 summarises the main empirical

results and section 5 offers a conclusion.

2. Literature Review

This review presents summaries of the essential literature concerning the PS

hypothesis, coffee pricing differentials, and determinants of the characteristics of coffee.

To begin with, the PS hypothesis was independently developed by Prebisch (1950) and Singer (1950), where they both reached similar conclusions regarding the rising

gap between developed and developing economies; this gave birth to a whole segment of discussion within development economics, with the debate largely

concerned with the validity of their theory, the choice of data and empirical evidence of their claims (Cuddington et al, 2002). Both Prebisch and Singer took the inverse of Britain’s net barter terms of trade between the 1876 and 1947 as a proxy for the relative

price of primary commodities to manufactures, which drew heavy criticism from Spraos (1980), amongst others, for being inadequate representation for the purpose.

Spraos (1980) and Sapsford (1985) attempted to develop a replacement dataset, but it was not until Grilli & Yang (1988) introduced the Grilli and Yang Commodity Price Index (GYCPI) that this debate was settled. The GYCPI consisted of 24 internationally

9

dataset within the literature, regularly being updated to maintain relevancy. With the data quality issues being put to rest, the focus mostly shifted to answering the question

of whether the commodity price series are better characterised as trend stationary (TS) or difference stationary (DS). Prebisch (1950) and Singer (1950) assumed that the former was the case, and for a while authors largely followed their lead, with the likes

of Ardeni & Wright (1992), Helg (1991), Grilli & Yang (1988) and Sapsford (1985) estimating the TS model; Bleaney & Greenaway (1993), Cuddington (1992) and

Cuddington & Urzua (1989) were amongst the first to find evidence for the existence of unit roots and advocate for the application of the DS model. To date there have been

no conclusive findings to give weight to either side of the discussion and as such, it remains necessary to carry out unit root tests when investigating the PS hypothesis.

Despite all this, the PS literature is mainly concerned with aggregate prices, although the practice has now shifted from using the aggregated GYCPI to individually testing the commodities that it is made up of; for a commodity as distinctly differentiated as

coffee, that leaves room for potentially severe aggregation bias (Cuddington & Wei, 1992). Recent work has shown that both sensory characteristics and reputation

variables have a significant effect on the valuation of the coffee, leading to non-negligible differences between prices (Donnet et al., 2008; Teuber & Hermann, 2012). This effect is not only constrained to differences in levels, but it also has an impact on

the dynamic behaviour of the prices, with more similar coffees showcasing a more stable long-term relationship with regards to co-movement (Otero et al., 2018). Up

until only recently, the literature surrounding cointegration of coffee price series largely ignored the possibility of such effects and ICO composite indicator price data

for the 4 main types of coffee traded on the bulk commodity market; the composite data takes the prices for a particular type across all countries that produce it and

10

adjustment in addition to the VEC model used to find the cointegrating vectors within the 4 types. Ghoshray (2009) estimates the relationships between the 4 types using an

approach of cointegration with threshold adjustment, allowing for asymmetric deviations, whereas Ghoshray (2010) finds support for cointegration between pairs using a non-linear model that allows for ESTAR adjustment. Otero et al. (2018) go a

step further and use a sample of 8 different sub-varieties (one each from Brazil, Colombia, Guatemala, Mexico, Peru, El Salvador, Uganda and Indonesia) for their

testing; they conclude that relative prices of similar coffee types are more likely to be stationary, smaller in magnitude, quicker to revert to equilibrium. Considering all of

this, inferences regarding the PS hypothesis derived from a single aggregate price for coffee could potentially be lacking in valuable nuance.

Coffee is primarily differentiated based on sensory characteristics, most of which can be attributed to certain biochemical indicators; these indicators, in turn, can be used to distinguish between various varieties of coffee (Alves et al., 2009; Carrera et al.,

1998). The chemical composition of the coffee is chiefly influenced by the climactic factors of the region, with agronomical and post-harvest practices playing a significant

secondary role (Alves et al., 2009; Avelino et al., 2005; Bote & Vos, 2017; Yadessa et al., 2020). For similar varieties of coffee, Oberthur et al. (2011) find statistically significant differences in biochemical and sensorial characteristics, with a clearly expressed

spatial structure that is related to the environmental data.

3. Data and Methodology

The database consists of annual coffee prices, the MUV Index, recorded observations of climatic factors and the country share in global production, spanning

the period 1965 – 2018. The annual coffee prices are provided by the ICO under their prices to growers table, which consists of the average prices paid to individual producers in a member exporting country (46 in total); in cases where farm level price

11

varieties, those being Colombian Mild (CM), Other Mild (OM), Brazilian Natural (BN) and Robusta. With a number of countries not specialising in producing one single

variety, the total number of price series adds up to 54, of which 10 are dropped due to a large amount of missing data.

Published by the World Bank, the MUV index is the de-facto deflator within the PS

hypothesis literature (Pfaffenzeller, 2007). The index is a weighted average of the prices of manufactured goods exported from the G-5 economies to developing

countries, which fits naturally with the PS hypothesis as it concerns the fall in the relative price of primary commodities exported by developing countries with regards

to manufactured goods imported from developed economies (Cuddington et al., 2002).

The data for the climatic factors is extracted from the Climatic Research Unit gridded

Time Series (CRUTS) dataset (Harris et al., 2020). Based off the findings of Oberthur et al. (2011), the indicators used in this study are mean temperature (in degrees

Celsius), diurnal temperature range (in degrees Celsius) and precipitation rate (in mm/month), averaged across the sample period for each country. Each datapoint is

the spatial average for the particular country for the given time period, where the spatial average is calculated using area-weighted means.

The global production data is also obtained from the ICO, which lists the total production by exporting countries. These figures are not separated by coffee type, so for countries which produce multiple types, the share of each in the country’s total

production is unknown. The production data is averaged across the sample period for each country, then presented as the overall fraction of global production.

To begin with, the price trends shall be computed by estimating a simple linear trend model for each producing country:

12

where 𝑝𝑖,𝑡 is the ratio of coffee price to the MUV for coffee i in the year t. 4 The coefficient 𝛽 of the time index 𝑇𝑖 is the linear trend, indicating whether the PS

hypothesis holds for the coffee price series in question (𝛽 < 0) or if it goes against the grain and displays an improvement over the sample period (𝛽 > 0). Additionally, unit root tests, namely the Augmented Dickey-Fuller (ADF) and Phillips-Perron (PP) tests,

shall also be conducted as robustness check. A failure to reject the null hypothesis of these tests would indicate that the price series in question is best characterised by a

unit root process rather than holding to a linear trend, which would be a rejection of the PS hypothesis.

The second part of the modelling strategy pertains to investigating the root causes of the heterogeneity of the price series via a pairwise examination of the series à la Otero

et al. (2018); this requires the construction of a dataset composed of differentials for each variable. The price differential, for example, shall be defined as 𝑝𝑖𝑗,𝑡 = 𝑝𝑖,𝑡 − 𝑝𝑗,𝑡

where 𝑝𝑖,𝑡 and 𝑝𝑗,𝑡 are the annual deflated prices for series i and j for the year t, i = 1,

…, N – 1, j = i + 1, …, N and N is the total number of price series in the sample set. This

shall be repeated for the three climatic factors and the share in global production, resulting in a total number of (N(N-1)/2) differential series for all five variables.

Proceeding from there, the following models shall be estimated via OLS:

𝐴𝐷𝐹_𝑖𝑗 = 𝛼 + 𝛽|𝐸̅_𝑖𝑗| + 𝛾|𝑄̅_𝑖𝑗| + 𝜀_𝑖𝑗 (2)

|𝑝̅_𝑖𝑗| = 𝛼 + 𝛽|𝐸̅_𝑖𝑗| + 𝛾|𝑄̅_𝑖𝑗| + 𝜀_𝑖𝑗 (3)

𝐴𝐷𝐹𝑖𝑗 is the ADF test statistic for 𝑝𝑖𝑗, used here as a measure of the strength of the

integration of the two price series, with a lower value indicating lower integration. |𝑝̅_𝑖𝑗|, the average magnitude of the stationary price differentials, is a measure of the quality premium for coffees that are co-integrated. |𝐸̅𝑖𝑗| represents the mean absolute differential of the climatic factors that are used as a proxy for the characteristics of the

coffee; the higher this value, the more different the coffees, which would lead to the

13

expectation of a lower quality premium and stronger integration. |𝑄̅𝑖𝑗| is the mean absolute differential of the share of global production, used to account for market factors; the greater this differential, the higher the expected integration, since a large

discrepancy implies that relative to each other, one of the coffees is the price setter and the other the price taker.

Preliminary analysis of the price series reveals that while differentials exists, they exhibit similar behaviour over time, following the same pattern of movement during

the Brazilian frosts of 1975 and 1994, the end of the International Coffee Agreement in 1989 and the flooding of the global coffee market by Vietnam around the turn of the century, for instance.5 _{However, some producers do appear to buck the trend at}

different points in time, suggesting that while there is a degree of market fundamentals influencing the price, not all are affected by it to the same extent.

Generally, the Arabica coffees (CM, OM, BN) have demanded a higher price and displayed greater instability as compared to Robusta over the period in question.

4. Estimation Results

Table 1 summarises the results of the estimation of model (1) and the unit root tests. Based off the PS hypothesis, all deflated price series should have displayed a

negative trend, which appears not to be the case; there are both series with positive linear trends and series where no significant linear trend is detected. Moreover, even

within the negatively trending series, there appears to be a non-negligible amount of variation. This leads credence to the concept of heterogeneity within global coffee

pricing and shows that the PS hypothesis does not hold for all individual coffee price series. However, with regards to geographical location, coffee variety or if the particular country is producing single or multiple varieties, no pattern is to be found,

with the trend distribution seemingly appearing to be random. To account for spurious regressions, ADF and PP tests with a trend term are carried out on the

14 Table 1 Country – Coffee Type Linear Trend Augmented Dickey-Fuller Phillips– Perron Country – Coffee Type Linear Trend Augmented Dickey-Fuller Phillips– Perron Colombia – CM -0.001 -3.186* -3.334* Brazil – BN 0.002 -3.873*** -3.960*** Kenya – CM -0.034*** -3.122 -2.712 Ethiopia – BN -0.009*** -4.708*** -4.721*** Tanzania – CM -0.045*** -3.681** -2.733 Brazil – RO 0.011** -3.053 -2.587 Burundi – OM -0.013*** -2.478 -1.926 Cameroon – RO -0.014*** -3.848*** -2.699 Costa Rica – OM -0.015*** -2.628 -2.796 Central African Republic – RO -0.007*** -3.263* -2.543 Cuba – OM -0.027*** -2.166 -1.789 Congo – RO -0.004* -2.371 -2.239 Dominican Republic – OM -0.007 -3.158* -3.129* Côte d'Ivoire – RO -0.009*** -2.690 -2.249 Ecuador – OM -0.004 -3.317* -2.890 Ecuador – RO -0.008 -3.639** -2.899 El Salvador – OM -0.017*** -3.656** -3.724** Gabon – RO -0.002 -2.719 -2.367 Guatemala – OM -0.010*** -3.366* -3.086 Guinea – RO -0.039*** -2.035 -2.088 Haiti – OM -0.040*** -3.191* -3.236* India – RO 0.004 -3.260* -3.238* Honduras – OM -0.018*** -3.479** -3.571** Indonesia – RO -0.030*** -3.989*** -3.296* India – OM -0.004 -2.550 -2.582 Madagascar – RO -0.009*** -3.438** -3.288* Jamaica – OM 0.028*** -3.352* -3.427** Nigeria – RO -0.029*** -2.513 -2.447

Malawi – OM -0.008 -2.652 -2.096 Papua New

Guinea – RO -0.022*** -2.738 -2.785 Mexico – OM -0.031*** -3.180* -3.299* Philippines – RO -0.027*** -2.864 -2.955 Panama – OM -0.018** -2.324 -1.990 Tanzania – RO -0.017** -3.200* -3.152* Papua New Guinea – OM -0.020*** -3.686** -3.779** Thailand – RO -0.026*** -2.676 -2.544 Peru – OM -0.035*** -1.586 -2.285 Togo – RO -0.001 -3.692* -3.085

Rwanda – OM -0.026*** -4.210*** -2.247 Trinidad &

Tobago – RO -0.049*** -2.923 -2.101 Uganda – OM 0.011*** -4.008*** -4.117*** Uganda – RO 0.008*** -3.041 -2.688

15

deflated price series; the results show that with the former, about half of the series are characterised by a unit root process, while with the PP test this number increases to

2/3rd_{. Once again, this signifies the heterogeneous behaviour of the individual price}

series; contrary to the PS hypothesis, not all deflated price series are stationary around a linear trend, and of the ones that are, not all follow a negative trend.

The estimation results of model (2) are presented in Table 2. The estimated coefficients of mean absolute temperature differential and mean absolute share in global

production positive and negative, respectively, and statistically different from 0. As expected, the greater the magnitude of the mean temperature differential, the lower

the integration of the two coffees, while the greater the magnitude of the mean share in global production differential, the greater the integration of the two coffees. The

magnitude of the differentials of mean diurnal temperature and mean precipitation are found to be statistically insignificant and significant but negligible, respectively; it appears to be that of 3 climatic factors, only mean temperature has a tangible effect on

the likelihood of stationarity.

Table 3 presents the results of estimating model (3), with 602 of the 946 differential

series dropped from the sample due lack of evidence against non-stationarity. Of the three climatic factors and the market factor, only the mean absolute diurnal

temperatures are found to have a significant effect on the magnitude of stationary deflated price differentials. With the expected positive coefficient, the result implies that the greater the value of the differential in diurnal temperature, the greater the

value of the price differential between the two coffees. In other words, diurnal temperature has a positive effect on the characteristics of the coffee that results in it

16

Table 2

Table 3

* p<0.10, ** p<0.05, *** p<0.01 Standard errors in parentheses

rss 769.176 771.800 767.967 763.237 BIC 2509.439 2512.661 2507.951 2515.811 adj. R-sq 0.014 0.011 0.016 0.020 N 946 946 946 946 (0.051) (0.054) (0.052) (0.071) _cons -2.776*** -2.742*** -2.596*** -2.714*** (0.339) (0.337) (0.337) (0.339) prod -1.024*** -1.087*** -1.164*** -1.068*** (0.000) (0.000) precip -0.000** -0.000** (0.021) (0.021) ditemp 0.025 0.027 (0.017) (0.017) temp 0.036** 0.031* t t t t (1) (2) (3) (4) Determinants of the stationarity of deflated price differentials.

* p<0.10, ** p<0.05, *** p<0.01 Standard errors in parentheses

17

5. Conclusion

This study investigates the existence and determinants of heterogeneity within the global coffee market. Using unit root tests and estimating the linear trends in MUV

deflated coffee prices, it finds that not all producing countries follow the same price trend; even within the ones that do adhere to the PS hypothesis, there too exists a

spread across the negative trends. The follow-up examination employs a pairwise cross-section approach to test for the determinants of the variation in stationarity and price and finds evidence that temperature, diurnal temperature and share in global

production can explain some of the variation in behaviour of the coffee price series.

The implications these results have are mainly twofold; one for the PS hypothesis

literature at large and another for producing countries in particular. With regards to the former, any future research into the global commodity price trends must take into

account the significant heterogeneity that exists within international coffee prices, lest they fall victim to aggregation bias. Studies that use one single price to represent the

entire industry might not offer much in the way of specific policy recommendations or learning opportunities for individual producing countries. As for the second point, the importance of climatic factors in determining the characteristics of the coffee

produced, and hence its valuation, means that climate change is a particularly grave issue, particularly for economies that are heavily reliant on exporting coffee as means

of foreign exchange; implementing climate policies is doubly important in their case. Moreover, producing countries may also benefit from investing in studies that

identify the most favourable climate factors and map out where these are found within their borders; in doing so, they can then focus their efforts on those regions in particular, which might also lead to denominations of origin within the global market.

This study does come with its own set of limitations. The frequency of the data is relatively low; using annual data as opposed to daily or monthly data introduces

18

over when employing annual frequency. Moreover, the climate data used is the spatial average of the country as a whole, as opposed to only the coffee growing regions;

doing so means that microclimates are not accounted for, where the coffee growing region might only be a relatively small area with its own unique climatic factors. A country’s coffee growing regions might also be spread over a number of

microclimates, such as in the case of Costa Rica, which results in coffees of vastly different characteristics and valuations – such nuances are lost through aggregation.

There also remains scope to study the different grades and varieties of coffees produced within a country, as the dataset used in this paper only reports the coffees

as divided into the 4 main types. Doing so might indicate whether some varieties are more prone to trending negatively relative to manufactured goods than others, and in such cases, it might be beneficial for the producer to look into planting a better

19

References

Addison, T., Ghoshray, A., & Stamatogiannis, M. P. (2016). Agricultural commodity price shocks and their effect on growth in Sub‐Saharan Africa. Journal of agricultural

economics, 67(1), 47-61.

Alves, R. C., Casal, S., Alves, M. R., & Oliveira, M. B. (2009). Discrimination between arabica and robusta coffee species on the basis of their tocopherol profiles. Food

chemistry, 114(1), 295-299.

Ardeni, P. G., & Wright, B. (1992). The Prebisch-Singer hypothesis: A reappraisal independent of stationarity hypotheses. The Economic Journal, 102(413), 803-812. Avelino, J., Barboza, B., Araya, J. C., Fonseca, C., Davrieux, F., Guyot, B., & Cilas, C.

(2005). Effects of slope exposure, altitude and yield on coffee quality in two altitude terroirs of Costa Rica, Orosi and Santa María de Dota. Journal of the Science of Food and

Agriculture, 85(11), 1869-1876.

Bidarkota, P., & Crucini, M. J. (2000). Commodity prices and the terms of trade. Review of

International Economics, 8(4), 647-666.

Bleaney, M., & Greenaway, D. (1993). Long-run trends in the relative price of primary commodities and in the terms of trade of developing countries. Oxford Economic

Papers, 349-363.

Bote, A. D., & Vos, J. (2017). Tree management and environmental conditions affect coffee (Coffea arabica L.) bean quality. NJAS-Wageningen Journal of Life Sciences, 83, 39-46. Carrera, F., León-Camacho, M., Pablos, F., & González, A. G. (1998). Authentication of

green coffee varieties according to their sterolic profile. Analytica chimica acta, 370(2-3), 131-139.

20

Cuddington, J. T. (1992). Long-run trends in 26 primary commodity prices. Journal of

Development Economics, 39(2), 207–227. doi: 10.1016/0304-3878(92)90037-a

Cuddington, J. T., Ludema, R., & Jayasuriya, S. A. (2002). Prebisch-singer redux (No. 1506-2016-130803).

Cuddington, J. T., & Urzúa, C. M. (1989). Trends and cycles in the net barter terms of trade: a new approach. The Economic Journal, 99(396), 426-442.

Cuddington, J. T., & Wei, H. (1992). An empirical analysis of real commodity price trends: Aggregation, model selection and implications. Estudios Economicos, 159-179.

Donnet, M. L., Weatherspoon, D. D., & Hoehn, J. P. (2008). Price determinants in top‐ quality e‐auctioned specialty coffees. Agricultural Economics, 38(3), 267-276.

Ghoshray, A. (2009). On price dynamics for different qualities of coffee. Review of Market

Integration, 1(1), 103-118.

Ghoshray, A. (2010). The extent of the world coffee market. Bulletin of Economic Research, 62(1), 97-107.

Ghoshray, A. (2011). A reexamination of trends in primary commodity prices. Journal of

Development Economics, 95(2), 242–251. doi: 10.1016/j.jdeveco.2010.04.001

Grilli, E. R., & Yang, M. C. (1988). Primary Commodity Prices, Manufactured Goods Prices, and the Terms of Trade of Developing Countries: What the Long Run Shows. The World Bank Economic Review, 2(1), 1–47. doi: 10.1093/wber/2.1.1

Harris, I., Osborn, T. J., Jones, P., & Lister, D. (2020). Version 4 of the CRU TS monthly high-resolution gridded multivariate climate dataset. Scientific data, 7(1), 1-18.

21

Helg, R. (1991). A note on the stationarity of the primary commodities relative price index. Economics Letters, 36(1), 55-60.

International Coffee Organisation (ICO) (2019). Survey on the impact of low coffee prices on exporting countries.

http://www.ico.org/documents/cy2018-19/Restricted/icc-124-4e-impact-low-prices.pdf.

Kellard, N., & Wohar, M. E. (2006). On the prevalence of trends in primary commodity

prices. Journal of Development Economics, 79(1), 146–167. doi:

10.1016/j.jdeveco.2004.12.004

Oberthür, T., Läderach, P., Posada, H., Fisher, M. J., Samper, L. F., Illera, J., ... & Usma, H. (2011). Regional relationships between inherent coffee quality and growing environment for denomination of origin labels in Nariño and Cauca, Colombia. Food

policy, 36(6), 783-794.

Otero, J., & Milas, C. (2001). Modelling the spot prices of various coffee types. Economic

Modelling, 18(4), 625-641.

Otero, J., Argüello, R., Oviedo, J. D., & Ramírez, M. (2018). Explaining coffee price differentials in terms of chemical markers: Evidence from a pairwise approach.

Economic Modelling, 72, 190-201.

Pfaffenzeller, S., Newbold, P., & Rayner, A. (2007). A Short Note on Updating the Grilli and Yang Commodity Price Index. The World Bank Economic Review, 21(1), 151–163. doi: 10.1093/wber/lhl013

Ponte, S. (2002). The latte revolution'? Regulation, markets and consumption in the global coffee chain. World development, 30(7), 1099-1122.

22

Sapsford, D. (1985). The statistical debate on the net barter terms of trade between primary commodities and manufactures: a comment and some additional evidence.

The Economic Journal, 95(379), 781-788.

Singer, H. W. (1950). The Distribution of Gains between Investing and Borrowing Countries. The American Economic Review, 40(2).

Singer, H. W. (1991). Terms of Trade: New Wine and New Bottles? Development Policy

Review, 9(4), 339–352. doi: 10.1111/j.1467-7679.1991.tb00192.x

Spraos, J. (1980). The statistical debate on the net barter terms of trade between primary commodities and manufactures. The Economic Journal, 90(357), 107-128.

Teuber, R., & Herrmann, R. (2012). Towards a differentiated modeling of origin effects in hedonic analysis: An application to auction prices of specialty coffee. Food Policy, 37(6), 732-740.

Vogelvang, E. (1992). Hypotheses testing concerning relationships between spot prices of various types of coffee. Journal of Applied Econometrics, 7(2), 191-201.

Winkelried, D. (2018). Unit roots, flexible trends, and the Prebisch-Singer

hypothesis. Journal of Development Economics, 132, 1–17. doi:

10.1016/j.jdeveco.2017.11.005

23

Appendix

Table A.1: Descriptive statistics

24

Table A.2: Characteristics of the coffees under consideration, averaged over the study period.

Origin Type Price

25

Figure A.1

27

28

29